Wealth Concentration in a Developing Economy Paris andFrance 1807ndash1994

By THOMAS PIKETTY GILLES POSTEL-VINAY AND JEAN-LAURENT ROSENTHAL

Using large samples of estate tax returns we construct new series on wealthconcentration in Paris and France from 1807 to 1994 Inequality increased until1914 because industrial and financial estates grew dramatically Then adverseshocks rather than a Kuznets-type process led to a massive decline in inequalityThe very high wealth concentration prior to 1914 benefited retired individuals livingoff capital income (rentiers) rather than entrepreneurs The very rich were in theirseventies and eighties whereas they had been in their fifties a half century earlierand would be so again after World War II Our results shed new light on ongoingdebates about wealth inequality and growth (JEL H20 J14 N20)

This article presents new series on wealthconcentration in Paris and France from 1807 to1994 It thus extends the series presented inThomas Piketty (2001 2003) by a full centuryand our new series are the first homogene-ous series of wealth inequality to cover a spanof time sufficient to fully evaluate SimonKuznetsrsquos hypothesis (1955) about the rise andfall of inequality as economies develop Whileother scholars have put together measures ofwealth inequality over time they have either

done so for much shorter periods or splicedtogether disparate sources Our series were con-structed by collecting the population of individ-ual estate tax returns in the Paris archives forvarious years between 1807 and 1902 and link-ing them to previously published tabulations bysize of estate for various years between 1902and 1994

Our general motivation for building such se-ries is the study of the two-way interactionbetween development and distribution Morespecifically one of our primary goals is to betterunderstand the decline in income and wealthinequality that occurred during the first half ofthe twentieth century in todayrsquos developedcountries Recent research on France suggeststhat this decline was for the most part an acci-dental phenomenon associated with the collapseof capital incomes1 rather than a spontaneoustwo-sector Kuznets-type process2 In particu-lar the only reason why top income shares

Piketty Paris-Jourdan Science Economiques (UMRCNRS-EHESS-ENS-ENPC) 48 boulevard Jourdan 75014Paris France (e-mail pikettyensfr) Postel-Vinay Insti-tut National de la Recherche Agronomique and Ecole desHautes Etudes en Sciences Sociales 48 boulevard Jourdan75014 Paris France (e-mail gpvensfr) Rosenthal De-partment of Economics UCLA Los Angeles CA 90095-1477 (e-mail rosenthaleconuclaedu) We thank theArchives de Paris and the Direction NationaledrsquoIntervention Domaniales du Ministere des Finances fortheir assistance in data collection the Guggenheim Foun-dation the McArthur Foundation the Collins Fund atUCLA the Institut National de la Recherche Agronomiqueand the Ecole des Hautes Etudes en Sciences Sociales forfinancial support and Lara Marchi Madeleine Roux andMarie Carmen Smyrnelis for help in data collection Specialthanks to Alena Lapatniova and Maria Chichtchenkova forextraordinary research assistance We also thank AnthonyAtkinson Peter Lindert Emmanuel Saez KennethSokoloff and the participants of the All-UC group in Eco-nomic History Grouprsquos conference on the New History ofInequality for their comments All our series are availableonline at httpwwwccpruclaeduResearchProjectWebsitesRosenthalRosenthalaspx A longer version of the paper isavailable as a CEPR Discussion Paper (Piketty et al 2004)

1 See Piketty (2003) For similar series covering theUnited States see Piketty and Emmanuel Saez (2003) andfor the United Kingdom see Anthony B Atkinson (2005)Similar top income series covering most of the twentiethcentury have now been constructed for about 20 countries(see Atkinson and Piketty forthcoming)

2 According to Kuznetsrsquos influential hypothesis(Kuznets 1955) income inequality should have declinedspontaneously in advanced capitalist countries as moreand more workers joined the high-paying sectors of theeconomy

236

dropped between 1914 and 1945 is that topcapital incomes fell whereas top wage sharesremained approximately constant (see Figure1) The wealth of the very rich was massivelyreduced by shocks in the first half of the twen-tieth centurymdashthese included war inflation andthe Great Depression The very rich have neverfully rebuilt their estates probably because ofthe dynamic effects of progressive estate andincome taxation on capital accumulation andpre-tax income inequality A central limitationof these top income and wage shares series isthat they begin latemdashjust before World War IThere is no systematic data source on incomesbefore then because the modern progressive in-come tax was not created until around 1913 inmost countries3 Although these series stronglysuggest that the 1914ndash1945 shocks played thekey role one cannot fully exclude the possiblyof a pre-existing Kuznets-type downward trendin inequality prior to World War I Constructing

wealth concentration series covering both thenineteenth and the twentieth centuries allows usto put the 1914ndash1945 period into a broaderhistorical perspective

A second and equally important goal is tounderstand the origins of the high levels ofinequality that we know prevailed on the eve ofWorld War I One can consider two extremehypotheses The first would suggest that thesehigh levels were longstandingmdashthe result of thepolitical structures of societies where the pri-mary form of wealth was land The second isthat capitalism and in particular the intercon-nection between financial development and in-dustrial growth created new forms of wealthwhose distribution was radically unequal Wethus aim to measure both the level of inequalitythat prevailed prior to the onset of industrializa-tion and the changes that modernization broughtforth Luckily for us the 1850s form a conve-nient turning point since industrialization accel-erated under the Second Empire (1852ndash1870)and the stock market boomed (Maurice Levi-Leboyer and Francois Bourguignon 1985)

Finally French historical sources on wealthdistribution are perhaps the richest in the world

3 The modern income tax was introduced in 1909 in theUnited Kingdom in 1913 in the United States and in 1914in France

FIGURE 1 THE FALL OF TOP CAPITAL INCOMES IN FRANCE 1913ndash1998

Source Piketty (2003) (computations based on income tax returns)

237VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

and ideal to investigate long-term changes ininequality As early as 1791 the French Na-tional Assembly introduced a universal estatetax which has remained in force since thenThis estate tax was universal because it appliedat any level of wealth and for nearly all types ofproperty (both real and estate)4 Furthermorethe successors of all decedents with positivewealth were required to file a return The estatetax was made progressive in 1902 (it wasstrictly proportional from 1791 to 1902) whichprompted the French tax administration to startcompiling summary tabulations of all individualestate tax returns5 These tabulations provideinformation about the number and value of es-tates in given wealth ranges No such tabula-tions were compiled prior to 1902 Howeverthe tax authorities transcribed individual returnsin registers that have been preserved We usedthese registers to collect large samples of indi-vidual returns between 1807 and 1902 We thenconstructed homogeneous estimates of wealthconcentration in Paris and France from 1807 to1994 (see below for more details on the data andmethodology)

Other scholars have attempted to use thesesources to examine the evolution of inequalityin France and in Paris In particular AdelineDaumard (1973) led a research group that ex-amined a few cross sections of estate returns(1821 1847 and 1911) in a small number ofcities in France Although the data collectedwere extraordinarily detailed the intervals be-tween samples were too long to uncover theevolution of inequality prior to World War IAnother ongoing project follows the descen-dants of all couples marrying in France between1800 and 1830 and whose family name startedwith the letters ldquoTRArdquo up to 1940 While thisapproach yields critical information about theintergenerational transmission of wealth withinthe broad population the sample size is toosmall to study the very wealthy In fact theTRA survey contains too few observations to

deliver reliable estimates above the ninety-fifthpercentile of the distribution (which is unfortu-nate because this is where most of the wealthlies)6

In other countries direct and homogeneousevidence on the evolution of wealth inequalityis scarce For instance the United Kingdom didnot see a universal estate tax before 1894 andthe United States waited until 1916 As a resulthomogeneous wealth concentration series basedupon estate tax returns can cover only the twen-tieth century in those two countries7 Prior toestablishment of estate taxes scholars relied onother sources in particular probate records Theinformation provided by probate records how-ever is neither as rich nor as systematic as thatcontained in estate tax returns (in particularprobate records were purely voluntary and alltypes of property were not covered)8 Conse-quently it is very difficult to compare the eigh-teenth- and nineteenth-century probate-basedestimates to the fiscal-based twentieth-centuryestimates Nevertheless they all suggest thatwealth concentration rose during the nineteenthcentury and dropped during the first half of thetwentieth century In contrast there is little ev-idence as to the course of inequality in the latenineteenth century (see eg the survey byLindert 2000) Had it started to decline as

4 The one glaring exception was government bondsthese were exempted until 1850

5 Prior to 1902 the tax on estates that devolved tochildren was a flat 1 percent In 1902 when the tax becameprogressive the top marginal rate was 5 percent by themid-1930s it was 35 percent it remains today at 40 percent(see Piketty 2001 Appendix J)

6 The TRA survey can be used for other purposes how-ever For instance Bourdieu et al (2003) use the survey tomeasure the evolution of the fraction of poor decedents (iedecedents with zero or near-zero wealth) and they find thatthis fraction had been increasing in nineteenth-centuryFrance (see below)

7 The standard references are Atkinson and Alan JHarrison (1978) for the United Kingdom and Robert JLampman (1962) for the United States Atkinson and Har-rison use estate tax return tabulations covering the 1923ndash1972 period to compute top wealth share series (thetabulations compiled by the UK tax administration overthe 1894ndash1914 period are less rich and do not allow for thesame computations as the post-1923 tables) Lampman usesestate tax return tabulations covering the 1922ndash1956 periodto compute top wealth share series (these series have beenupdated by various authors including Wojciech Kopczukand Saez 2004) See Peter H Lindert (2000) for a recentsurvey

8 In particular real estate was not probated in the UnitedKingdom before 1898 (realty and personalty were alsotreated differently in US probate records) For estimates ofwealth concentration in the United Kingdom based on eigh-teenth- and nineteenth-century probate records see Lindert(1986) For corresponding estimates for Colonial Americasee Alice H Jones (1977)

238 THE AMERICAN ECONOMIC REVIEW MARCH 2006

Kutznets would have thought Did it stabilizeDid it keep increasing until World War I OurFrench series allow us to cast new light on thiscentral issue because they are homogeneousover the 1807ndash1994 period

Our main conclusions are the followingFirst wealth concentration in Paris and inFrance increased up to World War I with anacceleration (rather than a stabilization) of thetrend at the end of the period The bulk of therise in inequality actually took place during the1860ndash1913 period This was largely driven bythe growth of large industrial and financial es-tates and coincided with the decline of aristo-cratic fortunes During the first half of thenineteenth century the share of aristocrats intop estates actually rose Next the decline inwealth concentration observed after World WarI appears to have been driven by the 1914ndash1945shocks rather than by a two-sector Kuznets-type process The decline in inequality was notdue to a reduction in the gap between Paris andthe provinces since it occurred both in Parisand in the rest of France Finally and perhapsmost importantly the very high levels of wealthconcentration observed at the eve of World WarI seem to have been associated with retiredindividuals who had lived off capital income(henceforth rentiers) rather than with active en-trepreneurs In particular the age-wealth profileof decedents is markedly steeper between 1900and 1913 than in other periods Top wealthholders were very old at the turn of the lastcentury (in their seventies and eighties)whereas they are usually in their fifties in otherperiods both at the beginning of the nineteenthcentury and at the end of the twentieth centuryAlthough our data do not allow us to evaluatethe inefficiency of wealth concentration di-rectly these results shed new light on the on-going debate about inequality and growth Thatis to the extent that credit constraints wereimportant in 1900 France (which we cannotprove directly with our data) our findings aboutthe changing age profile of wealth suggest thathigh wealth concentration might have been as-sociated with lower growth9

The paper is organized as follows Section Idescribes our data sources and outlines ourmethodology Section II presents our estimatesof wealth concentration and composition atdeath in Paris Section III discusses how theestimates from nineteenth-century Paris can beextended to the rest of France and presents pre-liminary results for wealth concentration at deathin France from 1807 to 1994 Section IV showshow our data on wealth and age at death can beused to estimate series on wealth concentrationamong the living using the estate multipliermethod Section V examines age-wealth profilesand discusses the efficiency implications of highwealth concentration Section VI concludes

I Data Sources

All of our estimates are based upon estate taxreturns As noted above the estate tax wascreated in 1791 and it became a progressive taxin 1902 Since then the tax administration hasperiodically compiled tables indicating thenumber of decedents and the value of theirestate for a large number of estate bracketsThese tables were already used by Piketty(2001 2003) and they are available over the1902ndash1994 period10 They were compiled andpublished by departement (middle-level admin-istrative jurisdictions there are about 90 ofthem in France including Paris)11 These tablescan be used to study the evolution of wealth

9 One way to test directly for the efficiency impact ofhigh wealth concentration would be to look at investmentpatterns across wealth fractiles and age groups (ie the

extent to which older wealth holders invest their wealth inlow-yield assets) The sources we use lend themselves toprecisely this kind of investigation and we intend to con-tinue this practice in further research

10 These tabulations were published in the official statis-tical publications of the French Finance Ministry (for exactreferences and page numbers see Piketty 2001 AppendixJ) The basic national tabulation indicating the number ofdecedents and amount of their estate for a large number ofestate brackets is available for the following years 1902ndash1913 (except 1906 and 1908) 1925ndash1960 (except 1928 and1934) 1962 and 1964 The French tax administrationstopped compiling such tables in 1964 but micro-files in-cluding large national samples of estate tax returns areavailable for 1984 and 1994 (in the present paper we useonly the 1994 micro-file)

11 Tables by estate brackets are available at the departe-ment level for the following years 1902ndash1913 (except 1906and 1908) and 1925ndash1958 (except 1928 and 1934) for otheryears tables by estate brackets are available only at thenational level In addition national tables broken down byestate brackets and age of decedents are available for years

239VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

concentration both in France and in Paris duringthe twentieth century using standard Pareto in-terpolation techniques

Prior to 1902 the tax administration pub-lished only the aggregate amount of wealth re-ported on estate tax returns broken down byreal (land and buildings) and personal (furni-ture businesses stocks bonds etc) assets12

Studying concentration thus required collectingour own samples of individual returns Collect-ing information on every individual return fromevery departement for a given year was impos-sible It would have required going to the ar-chives of each departement acquiring access tothe registers of each tax bureau (the lowest-level tax jurisdiction) and dealing with hun-dreds of thousands of declarations each yearWe therefore had to devise a sampling strategyOne option was to select randomly (eg on thebasis of birth dates or family names) a nation-ally representative sample of decedents for var-ious years during the nineteenth century Thatsample would need to be extremely large how-ever to include enough large estates (Given

that wealth is extremely concentrated it is crit-ical to observe many of the very wealthy)

Therefore we decided to pursue a completelydifferent strategy and collected data for all de-cedents in Paris for selected years (1807 18171827 1837 1847 1857 1867 1877 1887 and1902) We chose Paris because a disproportion-ate share of the very rich lived there As one cansee from Table 1 around 1810 the annual num-ber of decedents (20 years old and over) in Pariswas about 12000 (25 percent of the Frenchtotal) that figure nearly tripled during the nine-teenth century to about 35000 by 1900 (65percent of the French total) However only 30percent of decedents in Paris had an estate dur-ing the nineteenth century (about half as manyas in the rest of France) so we needed only tocollect detailed information on 3500 decedentsor so per year at the beginning of the nineteenthcentury and 10000 or so decedents per year atthe end (see Table 1) Although Paris had moredecedents with zero wealth than the rest of thecountry the average estate was about 45 timeslarger in Paris than elsewhere in France duringthe nineteenth century13 It is particularly strik-ing to notice that this ratio actually increased

1943ndash1954 The 1994 micro-file also allows us to breakdown the data by departement and age

12 These published aggregates were computed by theadministration on the basis of tax receipts

13 Average estates as well as top estate fractiles arealways defined in this paper over the set of all decedentsaged 20 and older including those with zero wealth

TABLE 1mdashESTATE TAX RETURNS IN PARIS 1807ndash1994mdashSUMMARY STATISTICS

N decedents20-yr N estate 0

N estate 0(percent

N deced 20)

N deced 20-yr(percent

ParisFrance)

Total estate(percent

ParisFrance)

Average estate(Ratio Parisrest

of France)

1807 11622 3647 314 25 82 3561817 11925 3287 276 25 84 3561827 14151 3877 274 28 94 3561837 16902 4922 291 31 98 3421847 18169 4814 265 33 115 3861857 19248 6048 314 36 143 4511867 26844 7370 275 46 168 4161877 28777 8245 287 51 186 4221887 34411 9815 285 59 201 4011902 36366 9830 270 65 260 5051913 35677 11927 334 65 266 5231929 35842 14495 404 58 250 5421938 30274 16013 529 53 173 3761947 24955 14090 565 55 150 3071956 27940 16053 575 55 159 3241994 18553 12528 675 36 97 286

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed series and sources)

240 THE AMERICAN ECONOMIC REVIEW MARCH 2006

over time in spite of the fact that Paris nearlytripled in population14 On the eve of WorldWar I the estates of Paris decedents made upover 26 percent of the French total (see Table1 and Figure 2)

We designed our data collection to take ad-vantage of the work of the estate tax adminis-tration For every person who either died inParis or might have taxable wealth in any of oneof Parisrsquos nine bureaux the administrationopened an entry in a first set of volumes (thetables des successions et absences henceforthTSA) Later the entry was completed eitherwhen estate taxes were paid or when the admin-istration became satisfied that the individual hadleft no wealth behind The entries include nameoccupation residence marital status age andfor individuals with wealth information aboutheirs and the date at which the declaration was

filed Up to 1870 the TSA also include a sum-mary of the individualrsquos estate broken downinto personal wealth and real estate Hence thecross sections up to 1867 rely heavily on theTSA After 1870 the administration no longerrecorded wealth information in the TSA butonly whether returns had been filed for theindividual For 1877 1887 and 1902 westarted with the TSA and for each individual forwhom a return had been filed we collected thefirst three letters of the last name gender ageday and month of death and the date(s) atwhich returns had been filed We then opened asecond set of registers (the registres de muta-tions par deces henceforth RMD) where a com-plete description of the estates is transcribedand the information not gleaned in the TSA wasappended to the first set of entries

Yet these data gave information by tax returnnot by individual A decedentrsquos heirs could filemultiple returns either because they amendedtheir original declaration or before 1902 be-cause they paid taxes in multiple bureaux In-deed prior to 1902 estate taxes on real estatewere paid in the bureau of the asset rather thanthat of the residence of the decedent In an eraof strictly proportional taxation such dispersed

14 Note that there is a discontinuity in the growth of Parisduring the nineteenth century as new districts (arrondisse-ments) previously registered in the suburb were integratedinto the City of Paris in 1860 The results reported here donot make any correction for this discontinuity which ex-plains the discontinuity observed in some of the figuresaround 1860

FIGURE 2 THE PARIS SHARE IN FRENCH ESTATES AT DEATH 1807ndash1994

Source Authorsrsquo computations based on estate tax returns (see Table 1)

241VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

payment of taxes reduced administrative costsbecause information about real estate values didnot have to be centralized Naturally when theestate tax became progressive in 1902 returnshad to be unified Because TSA entries providelinks to the different declarations in the samebureau reassembling these declarations waseasy reassembling returns from different bu-reaux was another matter because there are nolinks across bureaux To reassemble fully allindividual returns would have required us tocollect detailed information on every decedentin Paris But we did not need to do so becausethe very high levels of inequality in Paris cameto our assistance By collecting nominal infor-mation on the top 10 percent of returns we wereable to attribute successfully 92 percent of mov-able assets and 97 percent of all real assets tospecific individuals15 Given the high variety offirst names and last names as well as detailedinformation on residence the likelihood offalsely positive matches is very low The re-maining returns were treated as individualsthus biasing downward our inequality estimates

Our 1902 Paris sample is fully consistentwith the table compiled by the tax administra-tion for the same year for Paris Therefore wecan link up our 1807ndash1902 Paris files with the1902ndash1994 Paris tables to construct homoge-neous 1807ndash1994 series for inequality in ParisThe more difficult task is to use the Paris data toinfer changes in wealth concentration for Francefrom 1807 to 1902 To do so we must estimatehow the relative importance of Paris in each topestate class evolved over the nineteenth centuryTo achieve this goal we used other estate sur-veys16 as well as nonestate fiscal sources (seeSection III below) The other difficult part is theconstruction of estimates for wealth concentra-tion among the living from estate tax datawhich we do using the estate multiplier method

and mortality data by age group (see Section IVbelow)

When using tax data it is also important tokeep in mind that tax evasion and manipulationcan potentially bias the results There are how-ever good reasons to believe that this is not toomuch of a problem here First estate tax rateswere extremely modest until World War I (lessthan 2 percent) which implies that the incen-tives for tax evasion were small In contrastpenalties for evasion were stiff Moreover theadministration made every effort to keep upwith changing composition of assets and totrack down individuals with some wealthAmong other things financial institutions andpublic utilities were required to notify the ad-ministration when accounts changed owners Asa result it was not easy to dissimulate thewealth of a decedent (either real estate or finan-cial assets in a publicly traded firm) and inher-itors had a strong incentive to register theirproperty in order to benefit from state protec-tion This suggests that the nineteenth-centurydata collected in the Paris archives is probably ofvery high quality Tax evasion is potentially amore serious issue for the twentieth century whentax rates become substantial Although top estatetax rates have rarely exceeded 20 to 30 percent fordirect transmissions in France (the top rate hasbeen 40 percent since 1984 its highest level ever)it is obvious that incentives for tax evasion haveincreased over time However several indepen-dent data sources suggest that the trends observedduring the twentieth century are robust and are notdue to the rise of tax evasion17

II Wealth Concentration at Death in Paris1807ndash1994

Figure 3 shows the evolution of wealth con-centration at death in Paris from 1807 to 1994Given that the bottom two thirds of the distri-

15 To check our procedure for 1817 1827 1877 and 1887we also assembled all declarations that matched on the firstthree letters of last name gender day of death and age theestimates of inequality are slightly higher but trivially so

16 In addition to the TRA survey (which gives a reliablepicture of the national distribution up to the ninety-fifthpercentile) we should mention the study by Daumard(1973) which relied on samples of estate tax returns col-lected in five French cities at the beginning and at the end ofthe nineteenth century (we shall come back to this importantstudy below)

17 See Bourdieu et al (2004) Furthermore the twentieth-century decline in wealth concentration observed in estatetax returns is qualitatively and quantitatively consistent withthe decline in capital income concentration observed inincome tax returns (and the latter appears to be robust inparticular it holds after scaling up tax-return capital incomeusing national accounts aggregates) This is also consistentwith several other data sources on wealth concentration andtop fortunes (especially equity ownership data) See Piketty(2001 2003) for a detailed discussion

242 THE AMERICAN ECONOMIC REVIEW MARCH 2006

bution own no wealth and the richest decileaccounts for at least 95 percent of the value ofall assets during the nineteenth century (seeTable 2) we focus on the top 1 percent Therichest 1 percent of (adult decedents) Parisiansappears to have held a stable and very highfraction of all assets during the first half of thenineteenth century (around 50 to 55 percent oftotal wealth) The 1817 spike was short-livedand was due not to a large increase in the size oftop estates but rather to a large decline inmodest estates (which apparently suffered themost from the Napoleonic Wars)18 Inequalityin Paris increased substantially after 1867 withthe top-1-percent share of wealth at death

climbing from about 52 percent to over 72 per-cent in 1913 World War I and the ensuingshocks then prompted an abrupt decline Thetop-1-percent share dropped by 34 percentagepoints between 1913 and 1947 and by about 10

18 Other spikes in the top-1-percent share are due for themost part to the volatility of the very top estates (thetop-01-percent share and mostly the top-001-percentshare) Note that with about 20000 decedents per year inParis the top-01-percent fractile includes only 20 dece-dents and the top 001 percent only 2 decedents so that theestimates for these fractiles are unstable They depend onthe identity of very wealthy individuals who happened todie in a specific year The figures reported in Table 2 are theraw figures with no adjustment whatsoever for this topwealth volatility Note however that the 1867ndash1913 up-ward trend is highly significant and does not rely on a smallnumber of very top wealth holders

FIGURE 3 WEALTH CONCENTRATION AT DEATH IN PARIS 1807ndash1994

Source Authorsrsquo computations based on estate tax returns (see Table 2)

TABLE 2mdashWEALTH CONCENTRATION AT DEATH IN PARIS1807ndash1994

Top-10-percentestate share

Top-1-percentestate share

Top-01-percentestate share

1807 960 512 1791817 976 573 2281827 973 495 1481837 977 501 1481847 983 558 2131857 969 510 1341867 971 530 1631877 969 589 2461887 971 554 2011902 991 648 2611913 996 721 3281929 949 631 2641938 904 536 2411947 767 381 1481956 750 346 1171994 669 237 65

Source Authorsrsquo computations using estate tax returns (seePiketty et al 2004 Table A2 for detailed series andsources)

243VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

percentage points between 1956 and 1994 Con-verting these wealth-at-death concentration es-timates into wealth-of-the-living concentrationestimates leaves this general picture unchanged(see Section IV below)

Who were the individuals who enjoyed sucha substantial increase in their relative wealthbetween 1867 and 1913 For the most part theirfortunes derived from large industrial and finan-cial estates As Figure 4 illustrates the share ofpersonal (non-real) estate has always been aU-shaped function of wealth This reflects thewell-known fact that real estate is a middle-class asset The poor are too poor to own land orbuildings what little they have is in furniturecash or other moveable items In contrast therich hold most of their wealth in stocks andbonds What is more interesting is that duringthe nineteenth century the relative importanceof personal wealth in Parisian estates also fol-lowed a U-shaped curve over time This wasespecially true for the very wealthy (see Figures4 and 5) where real assets became more andmore important from 1807 to 1837 Real estatethen entered a relative decline after 1837 andaccelerated after 1867

The ebb and flow of the relative importanceof real estate was linked to Parisrsquos recoveryfrom the French Revolution Prior to the Revo-lution the peripheral parts of the city had beena maze of convents monasteries and educa-tional institutions all belonging to the CatholicChurch When the wealth of the Church wasnationalized these real estate assets wereabruptly put on the private market creating aglut of buildings and low prices As the cityrsquospopulation expanded building and land valuesrecovered and the relative importance of realestate grew before being overshadowed by thefinancial boom of the last part of the century(Michel Lescure 1982)

The share of aristocratic decedents among thevery rich follows an inverted-U-shaped curveover the nineteenth century (see Figure6)19 That is nobles became more and morenumerous in top wealth fractiles from 1807 until1847 then the trend reversed and their impor-

19 We take a very broad view of aristocrats they includethe Old Regime nobility the members of the elite who weregiven titles by Napoleon (1801ndash1814) and the Bourbons(1815ndash1830)

FIGURE 4 WEALTH COMPOSITION AT DEATH IN PARIS 1807ndash1902(Share of personal (non-real) estate in total estate)

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives

244 THE AMERICAN ECONOMIC REVIEW MARCH 2006

FIGURE 5 WEALTH COMPOSITION AT DEATH IN PARIS AND FRANCE 1807ndash1902(Share of personal (non-real) estate in total estate)

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives and national aggregateestate statistics compiled by the French tax administration

FIGURE 6 ARISTOCRATIC ESTATES AT DEATH IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives

245VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

tance declined steadily To be sure aristocratsremain overrepresented throughout the periodincluding in 1902 (about 13 percent of nobles inthe top 1 percent of estates over 25 percent inthe top 01 percent versus less than 1 percent inthe population as a whole) The inverted-U pat-tern is yet another of the Revolutionrsquos legaciesIn 1807 when we first observe it aristocraticwealth was at a temporary nadir On the onehand the nobility was impoverished by theRevolutionrsquos inflation and by the sharp declineof the value of Parisian real estate On the otherhand part of the Old Regime nobility was inexile and thus if they died we do not observetheir moveable wealth Aristocrats were able torecoup part of their losses during the first half ofthe nineteenth century Napoleon providedsome assistance by conferring titles of nobilityon his chief military officers and endowingthem with wealth Later the Restoration gov-ernment (1815ndash1830) compensated individualswho fled abroad during the Revolution for thelosses they suffered when their property wasconfiscated The government distributed nearlyone billion francs in the famous milliard desemigres (Andre Gain 1929) The beneficiariesof Napoleonrsquos and the Restorationrsquos largess ap-

pear among the very rich until mid-centuryPresumably such redistribution did not contrib-ute to accelerate French industrialization

III From Paris to France

We can use the Paris data to construct wealthconcentration at death estimates for all ofFrance from 1807 to 1902 To do so we need toknow the evolution of the share of Paris estatesin top estates Between 1902 and 1994 avail-able data (broken down by departement) showsthat the evolution of top estate shares in Francewas parallel to that of top estate shares in ParisWealth inequality is always lower for the coun-try as a whole but the trends are similar (seeFigure 7) It is also striking to note that Parisrsquosshare of the top 1 percent of French estates hasremained fairly stable over the twentieth cen-tury (it fluctuates between 20 percent and 25percent with no trend) even though Parisrsquosshare of all decedents has been dwindling overtime reflecting the population decline of thecapital (see Table 3) In 1902 Paris decedentswere four times more likely to belong to thenational top 1 percent of estates than averagedecedents (26665 41) in 1994 Paris de-

FIGURE 7 WEALTH CONCENTRATION AT DEATH IN PARIS AND FRANCE 1807ndash1994

Source Authorsrsquo computations based on estate tax returns (see Tables 2 and 4)

246 THE AMERICAN ECONOMIC REVIEW MARCH 2006

cedents were seven times more likely to belongto the national top 1 percent of estates thanaverage decedents (25236 70) If anythingthe geographic concentration of fortunes waslarger at the end of the twentieth century than atthe beginning The decline of wealth concentra-tion that followed World War I was not due toredistribution between Paris and the provinces

How did the fraction of Paris estates in topestates evolve over the course of the nineteenthcentury Our estimates rely on a simple andvery conservative assumption from 1807 to1902 Parisrsquos share of estates in the top percen-tile increased at the same rate as Parisrsquos share ofFrench adult deaths More precisely let us de-note FPt(w) the cumulative distribution functionfor wealth-at-death in Paris in year t Ft(w) thecorresponding distribution for France nPt thetotal number of adult deaths in Paris in year tand nt the corresponding number for all ofFrance The ninetieth-percentile threshold P90Ptis defined by FPt(P90Pt) 09 the ninety-ninth-percentile threshold P99Pt is defined byFPt(P99Pt) 099 etc and similarly for theFrench thresholds P90t P99t etc We observeFPt(w) nPt and nt throughout the 1807ndash1994period but we do not observe Ft(w) until 1902(before this date we observe only national ag-

gregate average wealth wt Wtnt) To con-struct our benchmark estimates we assume thatthe shares s99t s995t s999t and s9999t of Parisestates in the national top 1 percent 05 percent01 percent and 001 percent of the nationalwealth-at-death distribution increased at thesame rate as nPtnt during the 1807ndash1902 period(see Table 3) Using this approximation and ourParis samples of individual tax returns we com-pute the threshold wealth levels for the toppercentiles of the national wealth distribution(eg P99 P995 P999 and P9999)20 Wealso calculate the average wealth levels for therelevant wealth classes (eg P99ndash995 P995ndash999 P999ndash9999 and P9999ndash100) usingPareto interpolation techniques These are thenweighted by the number of individuals inFrance in that wealth class in order to compute

20 For instance the number of decedents (aged 20 yearsand older) in France was 583976 in 1887 (see Piketty et al2004 Appendix Table A1) so that the top 1 percent of theestate distribution at death consists of the top 5840 estatesIf the share of Paris among French top-1-percent estates was241 percent in 1887 (see Table 3) then the national P99threshold for 1887 corresponds to the top 1410 Parisianestates (0241 5840 1410) (the national P99 thresholdfor 1887 reported in Piketty et al 2004 Appendix TableA3 was computed using this formula)

TABLE 3mdashTHE FRACTION OF PARIS ESTATES IN TOP ESTATES AT DEATH IN FRANCE 1807ndash1994

(1) Fraction of Parisdecedents in all

decedents 20-yr

(2) Fraction of Parisestates in top-10-

percent estates

(3) Fraction of Parisestates in top-1-percent estates

(4) Fraction of Parisestates in top-01-

percent estates

1807 25 101 2051817 25 103 2101827 28 116 2371837 31 126 2561847 36 146 2971857 36 146 2971867 49 199 4041877 51 211 4281887 59 241 4911902 65 75 266 5411913 65 75 255 5231929 58 83 239 5301938 53 74 216 4211947 55 110 198 3521956 55 128 223 3501994 36 89 252 352

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources) No datasource exists to compute columns (3)ndash(4) prior to 1902 and the numbers reported on this table for years 1807ndash1887 werecomputed assuming that the columns (3)ndash(4) followed the same trend as column (1) over the 1807ndash1902 period (see textSection III)

247VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

the average wealth levels for top fractiles (P99ndash100 P995ndash100 P999ndash100 and P9999ndash100)Lower thresholds of the national wealth distri-bution (P90 and P95) were computed using thenational TRA survey21 and the P90ndash95 andP95ndash99 intermediate wealth levels were alsocomputed using Pareto interpolation techniques

The national top estate shares estimates re-ported in Table 4 were computed using thismethodology They suggest that wealth concen-tration (as measured by the top-1-percent estateshare) rose throughout the nineteenth century inFrance during both the 1807ndash1867 and 1867ndash1902 periods although less sharply than in Parisduring the latter period (see Figure 7) Theseestimates are conservative in the sense that it isalmost certain that they underestimate the riseof wealth concentration that took place inFrance during the nineteenth century First weknow that the bulk of population growth in Parisduring the nineteenth century was due to theannexation of suburbs in 1860 and to populationgrowth in these peripheral arrondissements Be-

cause the outskirts of the city were poor theannexation added few top estates Thus there islittle doubt that Parisrsquos share of top estates inFrance actually increased less than its share of thetotal population This hypothesis is confirmed bynineteenth-century housing tax tabulations show-ing that the fraction of Paris taxpayers in thenational top 1 percent of taxpayers was substan-tially larger than 10 percent at the beginning of thenineteenth century22 Giving Paris a larger (andmore realistic) share of top estates in 1807 wouldboth reduce the share of wealth of the top 1percent in France at that date and lead to morerapid rise in inequality over time

Next and most importantly other estate sur-veys are consistent with the view that wealth in-equality was growing The important study byAdeline Daumard (1973) which relied on samplesof estate tax returns collected in Paris Lyon Tou-louse Lille and Bordeaux found that wealth con-centration increased in each of these five citiesduring the nineteenth century23 The TRA surveyalthough it is ill-suited for the study of top estatesis also consistent with our view Wealth dispersionwas on the rise in nineteenth-century France ac-cording to the TRA survey both in the sense thatthe fraction of decedents with positive estates de-clined over time (in spite of the sharp increase inthe value of the average estate) and that ratios suchas the P90P50 increased24 We also comparedour benchmark national P99 series extrapolatedfrom our Paris samples and the national P99series computed using the TRA survey We foundthat both series display the same overall upwardtrend in concentration (which is reassuring regard-ing the general validity of our Paris-France extrap-olation technique) except that the growth ofinequality from 1807 to 1902 in the TRA series is

21 See Bourdieu et al (2003) for full details about theTRA survey The P90 and P95 thresholds reported onPiketty et al (2004 Appendix Table A3) were computedusing ten-year moving averages around the target years inorder to make sure that each estimate was based on asufficient number of observations

22 These tabulations were published in the same FinanceMinistry official publications as the estate tabulations Wechose not to use them in our formal computations becausethe tax base of the housing tax (namely the rental value ofthe real estate property where the household lives) is onlyloosely connected to the estate tax base (in particular onecannot rule out the possibility that the housing tax baseoverrepresents Paris-based taxpayers)

23 Unfortunately Daumardrsquos samples are not available inmachine-readable format she has only two or three years ofdata for each city and she did not try to compute homog-enous inequality indicators (top fractiles shares etc) withher data Thus although her results and our work are con-sistent they cannot be compared directly

24 See Bourdieu et al (2004)

TABLE 4mdashWEALTH CONCENTRATION AT DEATH IN FRANCE1807ndash1994

Top-10-percentestate share

Top-1-percentestate share

Top-01-percentestate share

1807 791 434 1631817 810 445 1811827 824 452 1631837 796 438 1471847 816 479 1841857 829 495 1741867 810 480 1741877 838 471 2011887 839 487 1921902 839 516 2311913 863 549 2601929 820 502 2471938 776 420 1991947 699 299 1101956 694 304 1101994 610 213 63

Source Authorsrsquo computations using estate tax returns (seePiketty et al 2004 Table A3 for detailed series)

248 THE AMERICAN ECONOMIC REVIEW MARCH 2006

more severe than in our series (see Figure 8) Thisagain suggests that the latter provide a conserva-tive lower bound for the upward trend in wealthconcentration In any case the finding of a largeincrease in wealth inequality in nineteenth-centuryFrance (and up until World War I) appears to berobust25

As was mentioned earlier there exists nocomparable continuous data source covering thenineteenth and twentieth centuries in othercountries which makes it difficult to put ourFrench long-run series in international perspec-tive We note however that existing series forthe United States and the United Kingdom areconsistent with our findings on France26 Re-garding levels existing evidence suggests thatduring the nineteenth and most of the twentiethcenturies France was in an intermediate posi-tion in terms of wealth concentration in be-tween the United States (more equal) and theUnited Kingdom (the most unequal)27 These

25 Note that this continuous rise in wealth inequalitydoes not necessarily imply that a parallel rise occurredregarding income inequality Given that there exists nomicro source on incomes prior to the creation of the incometax in 1914 it is very difficult (if not impossible) to properlyaddress this issue Christian Morrisson and Wayne Snyder(2000) have attempted to link income inequality estimatesbased upon Old-Regime fiscal sources (pre-1789) withmodern income-tax-based twentieth-century estimates andthey have argued that income inequality might have startedto decline during the later part of the nineteenth century andon the eve of World War I (see also Morrisson 2000)Although our data do not allow us to rule out such apossibility we note that their nineteenth-century personaldistribution estimates are based on fragile macroeconomicdata on functional distribution and are not homogenous totheir eighteenth- and twentieth-century estimates Given theevidence that we provide on wealth inequality any signif-icant decline in aggregate income inequality would have tobe associated with severe compression of the wage distri-bution There is little research on this issue however

26 Lee Soltow and Jan L Van Zanden (1998) also find adecline in inequality in the twentieth century in the Neth-erlands Their data are consistent with a rise in inequality inthe nineteenth century but they have no direct evidenceabout its extent

27 According to our series the top-1-percent wealthshare in France rose from around 45 percent in 1800 toabout 55 percent around World War I and then fell to about20 percent by the end of the twentieth century (see Figure7) wealth concentration among the living appears to besomewhat larger (see Section V below) According to theseries pieced together by Lindert (2000 pp 181ndash82 and186) the UK top-1-percent wealth share rose from about

FIGURE 8 ESTIMATES OF THE P99 THRESHOLD FOR THE FRENCH DISTRIBUTION OF ESTATES AT DEATH EXTRAPOLATION

FROM PARIS SAMPLES VERSUS ESTIMATES FROM TRA SAMPLES (CURRENT FRENCH FRANCS)

Source Authorsrsquo computations based on estate tax returns (Paris samples and TRA samples)

249VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

differences in inequality largely hark back to dif-ferential concentration of landownership En-glandrsquos land was extremely narrowly held whilethe United States was most egalitarian The impactof access to real estate assets can also be seenwhen we contrast Paris (where it was extremelyconcentrated) with the provinces

Trends in inequality over time may be easierto compare because biases in source materialmay have a significant effect on levels ratherthan trends The historical pattern is similar ineach of the three economies In particular thereis evidence that wealth concentration increasedduring the nineteenth century in both Anglo-Saxon countries and declined during the twen-tieth century with a turning point around WorldWar I28 Neither exhibits patterns consistentwith a Kuznet process

IV From the Wealth of the Decedents to theWealth of the Living

The estimates reported thus far refer to in-equality among decedents as described in thetax returns filed by their heirs The evolution ofthe distribution of wealth among the livingmight however have followed a different pat-tern In order to convert wealth-at-death con-centration estimates into wealth-of-the-livingconcentration estimates it is standard to use theldquoestate multiplierrdquo method29 It consists ofweighting each observation of an estate at death

by the inverse of the mortality rate for this agegroup That is if the mortality rate for ages 20to 24 was 068 percent in Paris in 1902 theneach decedent aged 20 to 24 represented about147 living individuals of the same age (100068 147) Similarly if the mortality ratefor ages above 80 was 2143 percent in Paris in1902 then each decedent in that group repre-sented about 47 living individuals in the sameage group (102143 47) Applying thismethod requires mortality tables (these are eas-ily available) and estate tabulations brokendown by estate size and age at death (these arescarcer) Fortunately the cityrsquos statistical bu-reau published annual death-by-age totals theFrench censuses report the age distribution forthe capital every five years and we collectedage at death from the estate declarations Thesedata allowed us to compute estimates of wealthconcentration among the living over the 1807ndash1902 period using various assumptions aboutthe wealth profiles of mortality rates

The base population for the living is the set ofall individuals aged 20 and over living in Parisin year t which we denote pt The number ofliving individuals in age bracket a is denoted pta(a 20ndash24 25ndash29 30ndash34 75ndash79 80 andover) and the number of decedents in agebracket a is denoted nta The mortality rate forage bracket a is given by mta ntapta Webegin with a uniform-mortality benchmarkThese estimates are based on the simplifyingassumption that these mortality rates dependsolely on age and are the same for all wealthgroups (and in particular are the same forzero-wealth and positive-wealth individuals)We can then weigh each decedent with positiveestate and age a collected in the Paris archivesin year t by ptanta This allows us to computethe number of living Parisians with positivewealth in year t and also (by differentiatingwith pt) the number of living Parisians with zerowealth at year t which is used to weight zero-estate observations We then use our weighteddatasets to compute top estate fractiles amongthe living in Paris

The main conclusion is that the living expe-rienced the same upward trend in wealth con-centration as the decedents (see Figure 9) Wefind that inequality was significantly higheramong the living than among decedents be-cause survivors were on average younger than

55 percent in 1800 to 70 percent around World War I thenfell to about 20 percent in the 1990s The US top-1-percentwealth share rose from about 15 to 20 percent in 1800 toabout 40 percent around World War I then fell to about 30percent in the 1990s (and as low as 20 to 25 percentaccording to the more recent estimates from Kopczuk andSaez 2004) Wealth concentration is now larger in theUnited States than in European countries but the reversewas true during the nineteenth century up until World WarII (It is only since the 1950sndash1970s period that US wealthconcentration has been somewhat larger)

28 See Lindert (2000 pp 181ndash82 and 188)29 This method was widely used in England and France

in the late nineteenth and early twentieth centuries to com-pute the stock of national wealth on the basis of the flow ofwealth transmitted at death Standard references that use thistechnique to estimate the wealth distribution of the livingfrom estate tax data tabulated by estate size and age at deathinclude Atkinson and Harrison (1978) and Lampman(1962) For a more recent application of this technique tothe United States see Kopczuk and Saez (2004)

250 THE AMERICAN ECONOMIC REVIEW MARCH 2006

those who died and the young were on aver-age poorer In particular the estate multipliermethod leads to lower average weights forpositive-wealth decedents than for zero-wealthdecedents (the former are on average older andtherefore ldquorepresentrdquo a smaller number of livingindividuals) As a result the fraction of positive-wealth individuals is even smaller among theliving than among decedents Hence our bench-mark uniform-mortality estimates of wealthconcentration among the living are significantlylarger than corresponding estimates among de-cedents (eg top-1-percent wealth shares areabout 15 percent larger among the living)Changes over time however are similar In-creased life expectancy and declining mortalityrates over the course of the nineteenth centuryhave only a small effect on the trends

In order to make the estate multiplier methodmore reliable one would prefer to take intoaccount differential mortality by wealth Doingso would require having access to mortalityschedules based both on wealth and age at dif-ferent points in time unfortunately these are notavailable We have nonetheless reestimatedwealth of the living based upon the same as-sumption as Kopczuk and Saez (2004) That is

we assumed uniform mortality among the poor(here defined as zero-wealth individuals) andamong the rich (here defined as positive-wealthindividuals) and we assumed that the ratiomtaRmtaP between the mortality rate of the richand the mortality rate of the poor followed aU-shaped age profile from about 85 percent forthe young (ie the rich die 15 percent less oftenthan the poor when they are 20ndash24 or 25ndash29years old) down to about 70 percent for middle-age individuals in their forties to fifties and upto 100 percent for very old individuals in theireighties to nineties30 This profile correspondsto the best available estimates in the literatureand it appears to be relatively stable over timeand across developed countries In the absenceof better data it is the best one can do31 Thebenchmark differential-mortality estimates re-ported on Figure 9 show that although addingdifferential mortality produces different levelsof inequality it does not have much impact onthe upward trend in concentration

30 See Kopczuk and Saez (2004 Table A4)31 See Kopczuk and Saez (2004 Appendix B) for refer-

ences to the US and international literature devoted to theage-wealth profile of mortality rates

FIGURE 9 WEALTH CONCENTRATION AMONG DECEDENTS AND AMONG THE LIVING IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives (see Piketty et al 2004Table A4 for detailed series)

251VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

The resulting differential-mortality inequalityestimates lie between those based on decedentsonly and those using uniform mortality for theliving (see Figure 9) Note that moving fromuniform-mortality to differential-mortality es-tate multiplier methodology can either increaseor decrease inequality Here the reason whysuch a move leads to lower wealth concentra-tion seems to be due to the fact that differentialmortality tends to put higher weights onpositive-wealth decedents (for a given age)thereby increasing the estimated fraction of liv-ing individuals with positive wealth The im-portant point however is that the resulting leveleffects are relatively small in magnitude con-stant in time and dwarfed by the upward timetrend Even if we were to assume an enormousincrease in differential mortality during thenineteenth century in the sense that differentialmortality between the rich and poor was equalto 0 percent of the benchmark differential in1807 and 100 percent of the benchmark differ-ential in 1902 the resulting wealth concentra-tion estimates would still be significantly higherin 1902 than in 1807 (see Figure 9) Yet wehave no reason to believe that differential mor-tality increased to such an extent During thenineteenth century real wages for unskilledworkers rose which would have reduced mor-tality more for the poor than for the rich After1850 public health measures (sanitation watervaccination) were in place Again these wouldhave had a significant effect on the poor and themiddle class who could not privately purchasesuch health-improving services To be sure therich could avail themselves of more medicalservices than the poor or the middle class butthe impact of these services was probably small(bear in mind that neither antibiotics nor car-diovascular interventions were available)

Finally we have applied the estate multipliermethod to available data for 1947 and 1994 Over-all the sharp decline in wealth concentration ob-served during the twentieth century (andespecially between 1914 and 1945) is very robustIf anything the decline appears to be even largerwhen one looks at wealth concentration amongthe living rather than among decedents32

V The Changing Age Profile of Wealth

In the previous sections we focused almostexclusively on aggregate top wealth shares Ourdata however also detail the characteristics ofeach decedent in particular their gender andage The evolution of wealth by gender is ofrelevance for over the past two centuries therehave been massive changes in womenrsquos laborforce participation capacity to manage theirown affairs and life expectancy relative to menThe evolution of wealth by age is of relevancebecause there was a significant increase in adultlife expectancy over the twentieth century andbecause the progressive diffusion of pensionsmay have changed savings motivations More-over age-wealth profiles also inform us aboutthe motives of wealth accumulation and theeconomic impact of high wealth concentration

A first pass at the data considers the genderbreakdown of wealth at death Remarkably inour micro data the share of women in top estatestakes its highest value on the eve of World WarI For instance the womenrsquos share in the top 05percent rose from 35 percent prior to 1850 to 45percent in 1902 only to fall to 40 percent afterWorld War II Strikingly womenrsquos share ofwealth follows almost exactly the pattern ofaggregate inequality Women were relativelyricher when inequality reached its apex inFrance than at any other time Moreover insti-tutional variables seem to have played almostno role in changing the relative wealth ofwomen Unlike in common law countriesFrench law starting with the code civil of 1804required nearly equal treatment of all children inbequests Further research will help us deter-mine to what extent women of great wealthwere heirs or part of economically very success-ful couples

The data also reveal striking changes in the

32 See Piketty et al (2004 Table A4) It is unfortunatelynot possible to construct complete series for wealth concen-

tration among the living for the twentieth century due todata limitations tables broken down by estate brackets andage of decedents are available solely for years 1943ndash1954and at the national level (no table broken down by estatebrackets and age of decedents has ever been compiled at thedepartement level except in 1931 for Seine departementsee E S Danysz 1934) and the 1994 micro sample is notlarge enough to allow for a reliable application of the estatemultiplier method at the Paris level Thus the only wealth-of-the-living concentration estimates we provide for thetwentieth century are national estimates for 1947 and 1994

252 THE AMERICAN ECONOMIC REVIEW MARCH 2006

age profile of wealth between 1807 and 1994(see Table 5) During the nineteenth century aswealth concentration was increasing the veryrich were getting older and older At the begin-ning of the nineteenth century in the aftermathof the French Revolution the richest individualswere those in their fifties they were typically100 percent richer on average than people intheir forties 25 percent richer than those in theirsixties and 40 percent richer than those in theirseventies and eighties By the 1870s howeverthe age-wealth pattern had become stronglymonotonic the richest individuals were the old-est individuals In 1902 people in their sixtiesand seventies bequeathed 150 percent morethan those in their fifties and those in theireighties 300 percent more On the eve of WorldWar I top wealth holders were old and likely tobe retired This pattern breaks some time duringthe 1914ndash1945 period33 In 1947 as well as in1994 we are back to a pattern where the richestindividuals are those in their fifties Overall theperiod of maximal wealth inequality (1860ndash1913) also appears to be a period characterizedby a very specific age profile of wealth and largeconcentration of assets among the elderly

Another way to analyze the changing age-wealth relationship is to look at average age by

top estate fractile34 In 1817 average age wasvirtually the same for the top 10 percent and thetop 1 percent of estates (or even slightly declin-ing) The average-age-per-fractile relationshipbecomes upward sloping during the nineteenthcentury and by 1902 those in the top 1 percentwere almost six years older than those in the top10 percent The relationship is flat in 1947 anddownward-sloping in 1994 Finally one canapply the estate multiplier method (see SectionIV above) and analyze how wealth concentra-tion by age group among the living changedover the course of the nineteenth century Thegeneral population in Paris did not becomeolder during the nineteenth century those aged60 or older made up about 15 percent of thepopulation in 1817 and after 1847 about 10 to11 percent35 The share of total wealth ownedby the elderly rose significantly however aswealth distribution worsened The wealth be-longing to those aged 60 or more rose fromabout 25 to 30 percent of the total at the begin-ning of the nineteenth century to about 40 to 45percent by the end of the century The wealthshare of those aged 70 or older doubled fromless than 10 percent to about 20 percent36

33 Existing evidence on the age-wealth profile for 1931(see Danysz 1934) suggests that the Great Depression andWorld War II (rather than World War I) played the leadingroles in breaking this pattern This is an issue we plan toinvestigate in future research

34 See Piketty et al (2004 Table 6)35 Although life expectancy was increasing which

should have led to large shares of population for oldergroups the city was also growing quickly The large num-ber of immigrants (who were typically in their twenties)increased the relative size of the younger cohorts (seePiketty et al 2004 Table 7 and Figure 11)

36 See Piketty et al (2004 Figure 12)

TABLE 5mdashTHE AGE PROFILE OF WEALTH AT DEATH IN PARIS 1817ndash1994(Average estate left by 50- 59-year-old 100)

20ndash29yr-old

30ndash39yr-old

40ndash49yr-old

50ndash59yr-old

60ndash69yr-old

70ndash79yr-old

80ndash89yr-old

90ndash99yr-old

1817 26 22 28 100 54 59 591827 44 50 53 100 88 87 601837 133 90 107 100 116 123 1101847 87 73 102 100 117 204 1321857 84 77 101 100 104 109 1451867 67 58 136 100 141 125 1541877 66 73 63 100 197 260 4301887 45 33 63 100 152 233 2951902 29 40 80 100 253 272 4011947 31 51 73 100 113 105 105 1091994 11 45 100 87 93 95 68

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources)

253VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

It is perhaps not surprising that inequalitybecame strongly correlated with age in the1860s Those who died at an old age in the1820s and 1830s had lived through the difficultyears of the French Revolution and the disloca-tion of the end of the Napoleonic period Theirability to accumulate wealth had been severelyimpaired Furthermore they were rather lesslikely to inherit much wealth since the Revo-lution wiped out the bond portfolios of theirparents through a prolonged period of high in-flation Those who died from the 1860s to theearly 1910s did not suffer from the adversitiesthat plagued their forebears Instead they en-joyed the fruits of the financial sector expansionthat began in the 1850s After 1947 we seem tohave returned to a situation quite like that of1817 Presumably the capital damages associ-ated with both world wars and the Great De-pression had a strong negative effect on thewealth holdings of older generations The per-sistence of a flat age-wealth profile until 1994 islikely to be associated with two factors First insocieties where income growth is rapid abso-lute wealth accumulation is faster by youngercohorts than by older ones because their in-comes are higher at every age This is an im-portant distinction between the nineteenthcentury and the twentieth century Furthermorehighly progressive rates of income and estatetaxation have probably made it more difficult toaccumulate large fortunes thereby flattening theobserved age-wealth profile

The more interesting (and more difficult)question relates to the possible efficiency im-pact of high wealth concentration and changingage-wealth profiles Although our data do notallow us to address efficiency issues in a rigor-ous way our results allow us to formulate anumber of hypotheses and to shed new light onthe ongoing debate on inequality and growth37

From a theoretical viewpoint whether highwealth concentration can have a negativegrowth impact depends critically on the exis-tence of credit constraints With first-best credit

markets money flows toward the best entrepre-neurs and investment projects irrespective ofthe initial distribution High levels of wealthconcentration can be bad from a social justiceviewpoint but they entail no efficiency lossWhen credit constraints bind however initialwealth matters and high levels of inequality canhurt growth Whether the loss is large or smalldepends on who owns the assets If the rich areefficient investors (they know which projects tofund etc) then wealth concentration may evenbe useful If the rich are retired rentiers how-ever investing their wealth in low-yield assets(or low-ability inheritors) then high wealthconcentration and credit constraints might pre-vent talented but penniless investors from un-dertaking efficient projects thereby entailingnegative growth consequences The data used inthis paper are not ideal to address whether creditconstraints were important in a country likeFrance at the end of the nineteenth century Ourresults suggest however that to the extentcredit constraints were indeed severe highwealth concentration did have a negativegrowth impact In order to investigate this hy-pothesis further one would need to gather moresystematic data on investment strategies andasset returns Preliminary evidence suggeststhat the rich elderly of the 1860ndash1913 perioddid indeed hold a disproportionate fraction oftheir wealth in low-yield assets (such as gov-ernment bonds) An alternative hypothesishowever is that steeper age-wealth profileswere the consequence of the growth of financialmarkets as their children faced fewer creditconstraints parents decided to hold on to moreof their wealth

VI Conclusion

Evidence from wealth at death in Paris and inFrance over the last two centuries reveals threekey patterns First wealth concentration haschanged dramatically over time In 1807 thetop-1-percent share of wealth (40 percent inFrance 50 percent in Paris) was twice as high asit would be in 1994 but substantially less thanin 1913 when it peaked above 55 percent inFrance and 70 percent in Paris Some of thesechanges were due to economic phenomena thathave long been emphasized as creating inequal-ity namely industrialization and financial cen-

37 Thus far this literature has concentrated upon cross-country regressions of inequality on growth a methodologythat raises serious identification problems especially giventhe low quality of available international datasets on in-equality which are neither long-run nor homogeneous (seeeg Atkinson and Andrea Brandolini 2001)

254 THE AMERICAN ECONOMIC REVIEW MARCH 2006

tralization Yet the decline comes largely fromadverse shocks rather than economic conver-gence These changes are of such magnitudethat they are not sensitive to whether one ex-amines wealth at death in Paris or in France orwhether one examines it directly rather thanconverting it to wealth of the living by an estatemultiplier method

Our second key result is that there was asignificant transition during the nineteenth cen-tury from an important role for real estate as aform of wealth to moveable assets as the keyform of wealth for the very rich Similarly theshare of wealth held by aristocrats first rose andthen was eclipsed by that of financiers and in-dustrialists in the second half of the nineteenthcentury Hence mobility within this highly un-equal society might have been quite high Yetthis conjecture is tempered by our third findingthe wealthy were getting older over time andolder relative to less wealthy decedents Suchaging among the very wealthy would have hadnegative consequences for growth if financialmarkets were imperfect This issue requires fur-ther investigation and we hope it will attractfuture research

REFERENCES

Atkinson Anthony B ldquoTop Incomes in the UKover the 20th Centuryrdquo Journal of the RoyalStatistical Society Series A (Statistics in So-ciety) 2005 168(2) pp 325ndash43

Atkinson Anthony B and Brandolini AndrealdquoPromise and Pitfalls in the Use of lsquoSecond-aryrsquo Data-Sets Income Inequality in OECDCountries as a Case Studyrdquo Journal of Eco-nomic Literature 2001 39(3) pp 771ndash99

Atkinson Anthony B and Harrison Alan J Dis-tribution of personal wealth in Britain Cam-bridge Cambridge University Press 1978

Atkinson Anthony B and Piketty Thomas Topincomes over the twentieth century A con-trast between European and English speak-ing countries Oxford Oxford UniversityPress (forthcoming)

Banerjee Abhijit V and Duflo Esther ldquoInequalityand Growth What Can the Data Sayrdquo Journalof Economic Growth 2003 8(3) pp 267ndash99

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoPourquoi la richesse ne

srsquoest-elle pas diffusee avec la croissance Ledegre zero de lrsquoinegalite et son evolution enFrance 1800ndash1940rdquo Histoire et mesure 200318(1ndash2) pp 147ndash98

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoDefense et illustration delrsquoenquete des 3000 familles Lrsquoexemple deson volet patrimonialrdquo Annales de demogra-phie historique 2004 19 pp 19ndash57

Danysz E S ldquoContribution a lrsquoetude des for-tunes privees drsquoapres les declarations de suc-cessionsrdquo Bulletin de la statistique generalede France 1934 XXIV (1) pp 111ndash171

Daumard Adeline Les fortunes francaise auXIXe siecle Enquete sur la repartition et lacomposition des capitaux prives a ParisLyon Lille Bordeaux et Toulouse drsquoapreslrsquoenregistrement des declarations de succes-sion Paris Mouton 1973

Gain Andre La restauration et les biens desemigres 2 vols Nancy Societe drsquoimpressiontypographique 1929

Jones Alice H American colonial wealth Doc-uments and methods 3 vols New York ArnoPress 1977

Kopczuk Wojciech and Saez Emmanuel ldquoTopWealth Shares in the United States 1916ndash2000 Evidence from Estate Tax ReturnsrdquoNational Tax Journal 2004 57(2) pp 445ndash87

Kuznets Simon ldquoEconomic Growth and IncomeInequalityrdquo American Economic Review1955 45(1) pp 1ndash28

Lampman Robert J The share of top wealth-holders in national wealth 1922ndash1956Princeton Princeton University Press 1962

Lescure Michel Les banques lrsquoEtat et lemarche immobilier en France a lrsquoepoquecontemporaine 1820ndash1940 Paris EditionslrsquoEcole des Hautes Etudes en Sciences Socia-les 1982

Levy-Leboyer Maurice and Bourguignon Fran-cois LrsquoEconomie francaise au XIXe siecleAnalyse macroeconomique Paris Eco-nomica 1985 English translation Cam-bridge Cambridge University Press 1990

Lindert Peter H ldquoUnequal English Wealthsince 1670rdquo Journal of Political Economy1986 94(6) pp 1127ndash62

Lindert Peter H ldquoThree Centuries of Inequalityin Britain and Americardquo in Anthony B At-kinson and Francois Bourguignon eds

255VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

and ideal to investigate long-term changes ininequality As early as 1791 the French Na-tional Assembly introduced a universal estatetax which has remained in force since thenThis estate tax was universal because it appliedat any level of wealth and for nearly all types ofproperty (both real and estate)4 Furthermorethe successors of all decedents with positivewealth were required to file a return The estatetax was made progressive in 1902 (it wasstrictly proportional from 1791 to 1902) whichprompted the French tax administration to startcompiling summary tabulations of all individualestate tax returns5 These tabulations provideinformation about the number and value of es-tates in given wealth ranges No such tabula-tions were compiled prior to 1902 Howeverthe tax authorities transcribed individual returnsin registers that have been preserved We usedthese registers to collect large samples of indi-vidual returns between 1807 and 1902 We thenconstructed homogeneous estimates of wealthconcentration in Paris and France from 1807 to1994 (see below for more details on the data andmethodology)

Other scholars have attempted to use thesesources to examine the evolution of inequalityin France and in Paris In particular AdelineDaumard (1973) led a research group that ex-amined a few cross sections of estate returns(1821 1847 and 1911) in a small number ofcities in France Although the data collectedwere extraordinarily detailed the intervals be-tween samples were too long to uncover theevolution of inequality prior to World War IAnother ongoing project follows the descen-dants of all couples marrying in France between1800 and 1830 and whose family name startedwith the letters ldquoTRArdquo up to 1940 While thisapproach yields critical information about theintergenerational transmission of wealth withinthe broad population the sample size is toosmall to study the very wealthy In fact theTRA survey contains too few observations to

deliver reliable estimates above the ninety-fifthpercentile of the distribution (which is unfortu-nate because this is where most of the wealthlies)6

In other countries direct and homogeneousevidence on the evolution of wealth inequalityis scarce For instance the United Kingdom didnot see a universal estate tax before 1894 andthe United States waited until 1916 As a resulthomogeneous wealth concentration series basedupon estate tax returns can cover only the twen-tieth century in those two countries7 Prior toestablishment of estate taxes scholars relied onother sources in particular probate records Theinformation provided by probate records how-ever is neither as rich nor as systematic as thatcontained in estate tax returns (in particularprobate records were purely voluntary and alltypes of property were not covered)8 Conse-quently it is very difficult to compare the eigh-teenth- and nineteenth-century probate-basedestimates to the fiscal-based twentieth-centuryestimates Nevertheless they all suggest thatwealth concentration rose during the nineteenthcentury and dropped during the first half of thetwentieth century In contrast there is little ev-idence as to the course of inequality in the latenineteenth century (see eg the survey byLindert 2000) Had it started to decline as

4 The one glaring exception was government bondsthese were exempted until 1850

5 Prior to 1902 the tax on estates that devolved tochildren was a flat 1 percent In 1902 when the tax becameprogressive the top marginal rate was 5 percent by themid-1930s it was 35 percent it remains today at 40 percent(see Piketty 2001 Appendix J)

6 The TRA survey can be used for other purposes how-ever For instance Bourdieu et al (2003) use the survey tomeasure the evolution of the fraction of poor decedents (iedecedents with zero or near-zero wealth) and they find thatthis fraction had been increasing in nineteenth-centuryFrance (see below)

7 The standard references are Atkinson and Alan JHarrison (1978) for the United Kingdom and Robert JLampman (1962) for the United States Atkinson and Har-rison use estate tax return tabulations covering the 1923ndash1972 period to compute top wealth share series (thetabulations compiled by the UK tax administration overthe 1894ndash1914 period are less rich and do not allow for thesame computations as the post-1923 tables) Lampman usesestate tax return tabulations covering the 1922ndash1956 periodto compute top wealth share series (these series have beenupdated by various authors including Wojciech Kopczukand Saez 2004) See Peter H Lindert (2000) for a recentsurvey

8 In particular real estate was not probated in the UnitedKingdom before 1898 (realty and personalty were alsotreated differently in US probate records) For estimates ofwealth concentration in the United Kingdom based on eigh-teenth- and nineteenth-century probate records see Lindert(1986) For corresponding estimates for Colonial Americasee Alice H Jones (1977)

238 THE AMERICAN ECONOMIC REVIEW MARCH 2006

Kutznets would have thought Did it stabilizeDid it keep increasing until World War I OurFrench series allow us to cast new light on thiscentral issue because they are homogeneousover the 1807ndash1994 period

Our main conclusions are the followingFirst wealth concentration in Paris and inFrance increased up to World War I with anacceleration (rather than a stabilization) of thetrend at the end of the period The bulk of therise in inequality actually took place during the1860ndash1913 period This was largely driven bythe growth of large industrial and financial es-tates and coincided with the decline of aristo-cratic fortunes During the first half of thenineteenth century the share of aristocrats intop estates actually rose Next the decline inwealth concentration observed after World WarI appears to have been driven by the 1914ndash1945shocks rather than by a two-sector Kuznets-type process The decline in inequality was notdue to a reduction in the gap between Paris andthe provinces since it occurred both in Parisand in the rest of France Finally and perhapsmost importantly the very high levels of wealthconcentration observed at the eve of World WarI seem to have been associated with retiredindividuals who had lived off capital income(henceforth rentiers) rather than with active en-trepreneurs In particular the age-wealth profileof decedents is markedly steeper between 1900and 1913 than in other periods Top wealthholders were very old at the turn of the lastcentury (in their seventies and eighties)whereas they are usually in their fifties in otherperiods both at the beginning of the nineteenthcentury and at the end of the twentieth centuryAlthough our data do not allow us to evaluatethe inefficiency of wealth concentration di-rectly these results shed new light on the on-going debate about inequality and growth Thatis to the extent that credit constraints wereimportant in 1900 France (which we cannotprove directly with our data) our findings aboutthe changing age profile of wealth suggest thathigh wealth concentration might have been as-sociated with lower growth9

The paper is organized as follows Section Idescribes our data sources and outlines ourmethodology Section II presents our estimatesof wealth concentration and composition atdeath in Paris Section III discusses how theestimates from nineteenth-century Paris can beextended to the rest of France and presents pre-liminary results for wealth concentration at deathin France from 1807 to 1994 Section IV showshow our data on wealth and age at death can beused to estimate series on wealth concentrationamong the living using the estate multipliermethod Section V examines age-wealth profilesand discusses the efficiency implications of highwealth concentration Section VI concludes

I Data Sources

All of our estimates are based upon estate taxreturns As noted above the estate tax wascreated in 1791 and it became a progressive taxin 1902 Since then the tax administration hasperiodically compiled tables indicating thenumber of decedents and the value of theirestate for a large number of estate bracketsThese tables were already used by Piketty(2001 2003) and they are available over the1902ndash1994 period10 They were compiled andpublished by departement (middle-level admin-istrative jurisdictions there are about 90 ofthem in France including Paris)11 These tablescan be used to study the evolution of wealth

9 One way to test directly for the efficiency impact ofhigh wealth concentration would be to look at investmentpatterns across wealth fractiles and age groups (ie the

extent to which older wealth holders invest their wealth inlow-yield assets) The sources we use lend themselves toprecisely this kind of investigation and we intend to con-tinue this practice in further research

10 These tabulations were published in the official statis-tical publications of the French Finance Ministry (for exactreferences and page numbers see Piketty 2001 AppendixJ) The basic national tabulation indicating the number ofdecedents and amount of their estate for a large number ofestate brackets is available for the following years 1902ndash1913 (except 1906 and 1908) 1925ndash1960 (except 1928 and1934) 1962 and 1964 The French tax administrationstopped compiling such tables in 1964 but micro-files in-cluding large national samples of estate tax returns areavailable for 1984 and 1994 (in the present paper we useonly the 1994 micro-file)

11 Tables by estate brackets are available at the departe-ment level for the following years 1902ndash1913 (except 1906and 1908) and 1925ndash1958 (except 1928 and 1934) for otheryears tables by estate brackets are available only at thenational level In addition national tables broken down byestate brackets and age of decedents are available for years

239VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

concentration both in France and in Paris duringthe twentieth century using standard Pareto in-terpolation techniques

Prior to 1902 the tax administration pub-lished only the aggregate amount of wealth re-ported on estate tax returns broken down byreal (land and buildings) and personal (furni-ture businesses stocks bonds etc) assets12

Studying concentration thus required collectingour own samples of individual returns Collect-ing information on every individual return fromevery departement for a given year was impos-sible It would have required going to the ar-chives of each departement acquiring access tothe registers of each tax bureau (the lowest-level tax jurisdiction) and dealing with hun-dreds of thousands of declarations each yearWe therefore had to devise a sampling strategyOne option was to select randomly (eg on thebasis of birth dates or family names) a nation-ally representative sample of decedents for var-ious years during the nineteenth century Thatsample would need to be extremely large how-ever to include enough large estates (Given

that wealth is extremely concentrated it is crit-ical to observe many of the very wealthy)

Therefore we decided to pursue a completelydifferent strategy and collected data for all de-cedents in Paris for selected years (1807 18171827 1837 1847 1857 1867 1877 1887 and1902) We chose Paris because a disproportion-ate share of the very rich lived there As one cansee from Table 1 around 1810 the annual num-ber of decedents (20 years old and over) in Pariswas about 12000 (25 percent of the Frenchtotal) that figure nearly tripled during the nine-teenth century to about 35000 by 1900 (65percent of the French total) However only 30percent of decedents in Paris had an estate dur-ing the nineteenth century (about half as manyas in the rest of France) so we needed only tocollect detailed information on 3500 decedentsor so per year at the beginning of the nineteenthcentury and 10000 or so decedents per year atthe end (see Table 1) Although Paris had moredecedents with zero wealth than the rest of thecountry the average estate was about 45 timeslarger in Paris than elsewhere in France duringthe nineteenth century13 It is particularly strik-ing to notice that this ratio actually increased

1943ndash1954 The 1994 micro-file also allows us to breakdown the data by departement and age

12 These published aggregates were computed by theadministration on the basis of tax receipts

13 Average estates as well as top estate fractiles arealways defined in this paper over the set of all decedentsaged 20 and older including those with zero wealth

TABLE 1mdashESTATE TAX RETURNS IN PARIS 1807ndash1994mdashSUMMARY STATISTICS

N decedents20-yr N estate 0

N estate 0(percent

N deced 20)

N deced 20-yr(percent

ParisFrance)

Total estate(percent

ParisFrance)

Average estate(Ratio Parisrest

of France)

1807 11622 3647 314 25 82 3561817 11925 3287 276 25 84 3561827 14151 3877 274 28 94 3561837 16902 4922 291 31 98 3421847 18169 4814 265 33 115 3861857 19248 6048 314 36 143 4511867 26844 7370 275 46 168 4161877 28777 8245 287 51 186 4221887 34411 9815 285 59 201 4011902 36366 9830 270 65 260 5051913 35677 11927 334 65 266 5231929 35842 14495 404 58 250 5421938 30274 16013 529 53 173 3761947 24955 14090 565 55 150 3071956 27940 16053 575 55 159 3241994 18553 12528 675 36 97 286

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed series and sources)

240 THE AMERICAN ECONOMIC REVIEW MARCH 2006

over time in spite of the fact that Paris nearlytripled in population14 On the eve of WorldWar I the estates of Paris decedents made upover 26 percent of the French total (see Table1 and Figure 2)

We designed our data collection to take ad-vantage of the work of the estate tax adminis-tration For every person who either died inParis or might have taxable wealth in any of oneof Parisrsquos nine bureaux the administrationopened an entry in a first set of volumes (thetables des successions et absences henceforthTSA) Later the entry was completed eitherwhen estate taxes were paid or when the admin-istration became satisfied that the individual hadleft no wealth behind The entries include nameoccupation residence marital status age andfor individuals with wealth information aboutheirs and the date at which the declaration was

filed Up to 1870 the TSA also include a sum-mary of the individualrsquos estate broken downinto personal wealth and real estate Hence thecross sections up to 1867 rely heavily on theTSA After 1870 the administration no longerrecorded wealth information in the TSA butonly whether returns had been filed for theindividual For 1877 1887 and 1902 westarted with the TSA and for each individual forwhom a return had been filed we collected thefirst three letters of the last name gender ageday and month of death and the date(s) atwhich returns had been filed We then opened asecond set of registers (the registres de muta-tions par deces henceforth RMD) where a com-plete description of the estates is transcribedand the information not gleaned in the TSA wasappended to the first set of entries

Yet these data gave information by tax returnnot by individual A decedentrsquos heirs could filemultiple returns either because they amendedtheir original declaration or before 1902 be-cause they paid taxes in multiple bureaux In-deed prior to 1902 estate taxes on real estatewere paid in the bureau of the asset rather thanthat of the residence of the decedent In an eraof strictly proportional taxation such dispersed

14 Note that there is a discontinuity in the growth of Parisduring the nineteenth century as new districts (arrondisse-ments) previously registered in the suburb were integratedinto the City of Paris in 1860 The results reported here donot make any correction for this discontinuity which ex-plains the discontinuity observed in some of the figuresaround 1860

FIGURE 2 THE PARIS SHARE IN FRENCH ESTATES AT DEATH 1807ndash1994

Source Authorsrsquo computations based on estate tax returns (see Table 1)

241VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

payment of taxes reduced administrative costsbecause information about real estate values didnot have to be centralized Naturally when theestate tax became progressive in 1902 returnshad to be unified Because TSA entries providelinks to the different declarations in the samebureau reassembling these declarations waseasy reassembling returns from different bu-reaux was another matter because there are nolinks across bureaux To reassemble fully allindividual returns would have required us tocollect detailed information on every decedentin Paris But we did not need to do so becausethe very high levels of inequality in Paris cameto our assistance By collecting nominal infor-mation on the top 10 percent of returns we wereable to attribute successfully 92 percent of mov-able assets and 97 percent of all real assets tospecific individuals15 Given the high variety offirst names and last names as well as detailedinformation on residence the likelihood offalsely positive matches is very low The re-maining returns were treated as individualsthus biasing downward our inequality estimates

Our 1902 Paris sample is fully consistentwith the table compiled by the tax administra-tion for the same year for Paris Therefore wecan link up our 1807ndash1902 Paris files with the1902ndash1994 Paris tables to construct homoge-neous 1807ndash1994 series for inequality in ParisThe more difficult task is to use the Paris data toinfer changes in wealth concentration for Francefrom 1807 to 1902 To do so we must estimatehow the relative importance of Paris in each topestate class evolved over the nineteenth centuryTo achieve this goal we used other estate sur-veys16 as well as nonestate fiscal sources (seeSection III below) The other difficult part is theconstruction of estimates for wealth concentra-tion among the living from estate tax datawhich we do using the estate multiplier method

and mortality data by age group (see Section IVbelow)

When using tax data it is also important tokeep in mind that tax evasion and manipulationcan potentially bias the results There are how-ever good reasons to believe that this is not toomuch of a problem here First estate tax rateswere extremely modest until World War I (lessthan 2 percent) which implies that the incen-tives for tax evasion were small In contrastpenalties for evasion were stiff Moreover theadministration made every effort to keep upwith changing composition of assets and totrack down individuals with some wealthAmong other things financial institutions andpublic utilities were required to notify the ad-ministration when accounts changed owners Asa result it was not easy to dissimulate thewealth of a decedent (either real estate or finan-cial assets in a publicly traded firm) and inher-itors had a strong incentive to register theirproperty in order to benefit from state protec-tion This suggests that the nineteenth-centurydata collected in the Paris archives is probably ofvery high quality Tax evasion is potentially amore serious issue for the twentieth century whentax rates become substantial Although top estatetax rates have rarely exceeded 20 to 30 percent fordirect transmissions in France (the top rate hasbeen 40 percent since 1984 its highest level ever)it is obvious that incentives for tax evasion haveincreased over time However several indepen-dent data sources suggest that the trends observedduring the twentieth century are robust and are notdue to the rise of tax evasion17

II Wealth Concentration at Death in Paris1807ndash1994

Figure 3 shows the evolution of wealth con-centration at death in Paris from 1807 to 1994Given that the bottom two thirds of the distri-

15 To check our procedure for 1817 1827 1877 and 1887we also assembled all declarations that matched on the firstthree letters of last name gender day of death and age theestimates of inequality are slightly higher but trivially so

16 In addition to the TRA survey (which gives a reliablepicture of the national distribution up to the ninety-fifthpercentile) we should mention the study by Daumard(1973) which relied on samples of estate tax returns col-lected in five French cities at the beginning and at the end ofthe nineteenth century (we shall come back to this importantstudy below)

17 See Bourdieu et al (2004) Furthermore the twentieth-century decline in wealth concentration observed in estatetax returns is qualitatively and quantitatively consistent withthe decline in capital income concentration observed inincome tax returns (and the latter appears to be robust inparticular it holds after scaling up tax-return capital incomeusing national accounts aggregates) This is also consistentwith several other data sources on wealth concentration andtop fortunes (especially equity ownership data) See Piketty(2001 2003) for a detailed discussion

242 THE AMERICAN ECONOMIC REVIEW MARCH 2006

bution own no wealth and the richest decileaccounts for at least 95 percent of the value ofall assets during the nineteenth century (seeTable 2) we focus on the top 1 percent Therichest 1 percent of (adult decedents) Parisiansappears to have held a stable and very highfraction of all assets during the first half of thenineteenth century (around 50 to 55 percent oftotal wealth) The 1817 spike was short-livedand was due not to a large increase in the size oftop estates but rather to a large decline inmodest estates (which apparently suffered themost from the Napoleonic Wars)18 Inequalityin Paris increased substantially after 1867 withthe top-1-percent share of wealth at death

climbing from about 52 percent to over 72 per-cent in 1913 World War I and the ensuingshocks then prompted an abrupt decline Thetop-1-percent share dropped by 34 percentagepoints between 1913 and 1947 and by about 10

18 Other spikes in the top-1-percent share are due for themost part to the volatility of the very top estates (thetop-01-percent share and mostly the top-001-percentshare) Note that with about 20000 decedents per year inParis the top-01-percent fractile includes only 20 dece-dents and the top 001 percent only 2 decedents so that theestimates for these fractiles are unstable They depend onthe identity of very wealthy individuals who happened todie in a specific year The figures reported in Table 2 are theraw figures with no adjustment whatsoever for this topwealth volatility Note however that the 1867ndash1913 up-ward trend is highly significant and does not rely on a smallnumber of very top wealth holders

FIGURE 3 WEALTH CONCENTRATION AT DEATH IN PARIS 1807ndash1994

Source Authorsrsquo computations based on estate tax returns (see Table 2)

TABLE 2mdashWEALTH CONCENTRATION AT DEATH IN PARIS1807ndash1994

Top-10-percentestate share

Top-1-percentestate share

Top-01-percentestate share

1807 960 512 1791817 976 573 2281827 973 495 1481837 977 501 1481847 983 558 2131857 969 510 1341867 971 530 1631877 969 589 2461887 971 554 2011902 991 648 2611913 996 721 3281929 949 631 2641938 904 536 2411947 767 381 1481956 750 346 1171994 669 237 65

Source Authorsrsquo computations using estate tax returns (seePiketty et al 2004 Table A2 for detailed series andsources)

243VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

percentage points between 1956 and 1994 Con-verting these wealth-at-death concentration es-timates into wealth-of-the-living concentrationestimates leaves this general picture unchanged(see Section IV below)

Who were the individuals who enjoyed sucha substantial increase in their relative wealthbetween 1867 and 1913 For the most part theirfortunes derived from large industrial and finan-cial estates As Figure 4 illustrates the share ofpersonal (non-real) estate has always been aU-shaped function of wealth This reflects thewell-known fact that real estate is a middle-class asset The poor are too poor to own land orbuildings what little they have is in furniturecash or other moveable items In contrast therich hold most of their wealth in stocks andbonds What is more interesting is that duringthe nineteenth century the relative importanceof personal wealth in Parisian estates also fol-lowed a U-shaped curve over time This wasespecially true for the very wealthy (see Figures4 and 5) where real assets became more andmore important from 1807 to 1837 Real estatethen entered a relative decline after 1837 andaccelerated after 1867

The ebb and flow of the relative importanceof real estate was linked to Parisrsquos recoveryfrom the French Revolution Prior to the Revo-lution the peripheral parts of the city had beena maze of convents monasteries and educa-tional institutions all belonging to the CatholicChurch When the wealth of the Church wasnationalized these real estate assets wereabruptly put on the private market creating aglut of buildings and low prices As the cityrsquospopulation expanded building and land valuesrecovered and the relative importance of realestate grew before being overshadowed by thefinancial boom of the last part of the century(Michel Lescure 1982)

The share of aristocratic decedents among thevery rich follows an inverted-U-shaped curveover the nineteenth century (see Figure6)19 That is nobles became more and morenumerous in top wealth fractiles from 1807 until1847 then the trend reversed and their impor-

19 We take a very broad view of aristocrats they includethe Old Regime nobility the members of the elite who weregiven titles by Napoleon (1801ndash1814) and the Bourbons(1815ndash1830)

FIGURE 4 WEALTH COMPOSITION AT DEATH IN PARIS 1807ndash1902(Share of personal (non-real) estate in total estate)

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives

244 THE AMERICAN ECONOMIC REVIEW MARCH 2006

FIGURE 5 WEALTH COMPOSITION AT DEATH IN PARIS AND FRANCE 1807ndash1902(Share of personal (non-real) estate in total estate)

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives and national aggregateestate statistics compiled by the French tax administration

FIGURE 6 ARISTOCRATIC ESTATES AT DEATH IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives

245VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

tance declined steadily To be sure aristocratsremain overrepresented throughout the periodincluding in 1902 (about 13 percent of nobles inthe top 1 percent of estates over 25 percent inthe top 01 percent versus less than 1 percent inthe population as a whole) The inverted-U pat-tern is yet another of the Revolutionrsquos legaciesIn 1807 when we first observe it aristocraticwealth was at a temporary nadir On the onehand the nobility was impoverished by theRevolutionrsquos inflation and by the sharp declineof the value of Parisian real estate On the otherhand part of the Old Regime nobility was inexile and thus if they died we do not observetheir moveable wealth Aristocrats were able torecoup part of their losses during the first half ofthe nineteenth century Napoleon providedsome assistance by conferring titles of nobilityon his chief military officers and endowingthem with wealth Later the Restoration gov-ernment (1815ndash1830) compensated individualswho fled abroad during the Revolution for thelosses they suffered when their property wasconfiscated The government distributed nearlyone billion francs in the famous milliard desemigres (Andre Gain 1929) The beneficiariesof Napoleonrsquos and the Restorationrsquos largess ap-

pear among the very rich until mid-centuryPresumably such redistribution did not contrib-ute to accelerate French industrialization

III From Paris to France

We can use the Paris data to construct wealthconcentration at death estimates for all ofFrance from 1807 to 1902 To do so we need toknow the evolution of the share of Paris estatesin top estates Between 1902 and 1994 avail-able data (broken down by departement) showsthat the evolution of top estate shares in Francewas parallel to that of top estate shares in ParisWealth inequality is always lower for the coun-try as a whole but the trends are similar (seeFigure 7) It is also striking to note that Parisrsquosshare of the top 1 percent of French estates hasremained fairly stable over the twentieth cen-tury (it fluctuates between 20 percent and 25percent with no trend) even though Parisrsquosshare of all decedents has been dwindling overtime reflecting the population decline of thecapital (see Table 3) In 1902 Paris decedentswere four times more likely to belong to thenational top 1 percent of estates than averagedecedents (26665 41) in 1994 Paris de-

FIGURE 7 WEALTH CONCENTRATION AT DEATH IN PARIS AND FRANCE 1807ndash1994

Source Authorsrsquo computations based on estate tax returns (see Tables 2 and 4)

246 THE AMERICAN ECONOMIC REVIEW MARCH 2006

cedents were seven times more likely to belongto the national top 1 percent of estates thanaverage decedents (25236 70) If anythingthe geographic concentration of fortunes waslarger at the end of the twentieth century than atthe beginning The decline of wealth concentra-tion that followed World War I was not due toredistribution between Paris and the provinces

How did the fraction of Paris estates in topestates evolve over the course of the nineteenthcentury Our estimates rely on a simple andvery conservative assumption from 1807 to1902 Parisrsquos share of estates in the top percen-tile increased at the same rate as Parisrsquos share ofFrench adult deaths More precisely let us de-note FPt(w) the cumulative distribution functionfor wealth-at-death in Paris in year t Ft(w) thecorresponding distribution for France nPt thetotal number of adult deaths in Paris in year tand nt the corresponding number for all ofFrance The ninetieth-percentile threshold P90Ptis defined by FPt(P90Pt) 09 the ninety-ninth-percentile threshold P99Pt is defined byFPt(P99Pt) 099 etc and similarly for theFrench thresholds P90t P99t etc We observeFPt(w) nPt and nt throughout the 1807ndash1994period but we do not observe Ft(w) until 1902(before this date we observe only national ag-

gregate average wealth wt Wtnt) To con-struct our benchmark estimates we assume thatthe shares s99t s995t s999t and s9999t of Parisestates in the national top 1 percent 05 percent01 percent and 001 percent of the nationalwealth-at-death distribution increased at thesame rate as nPtnt during the 1807ndash1902 period(see Table 3) Using this approximation and ourParis samples of individual tax returns we com-pute the threshold wealth levels for the toppercentiles of the national wealth distribution(eg P99 P995 P999 and P9999)20 Wealso calculate the average wealth levels for therelevant wealth classes (eg P99ndash995 P995ndash999 P999ndash9999 and P9999ndash100) usingPareto interpolation techniques These are thenweighted by the number of individuals inFrance in that wealth class in order to compute

20 For instance the number of decedents (aged 20 yearsand older) in France was 583976 in 1887 (see Piketty et al2004 Appendix Table A1) so that the top 1 percent of theestate distribution at death consists of the top 5840 estatesIf the share of Paris among French top-1-percent estates was241 percent in 1887 (see Table 3) then the national P99threshold for 1887 corresponds to the top 1410 Parisianestates (0241 5840 1410) (the national P99 thresholdfor 1887 reported in Piketty et al 2004 Appendix TableA3 was computed using this formula)

TABLE 3mdashTHE FRACTION OF PARIS ESTATES IN TOP ESTATES AT DEATH IN FRANCE 1807ndash1994

(1) Fraction of Parisdecedents in all

decedents 20-yr

(2) Fraction of Parisestates in top-10-

percent estates

(3) Fraction of Parisestates in top-1-percent estates

(4) Fraction of Parisestates in top-01-

percent estates

1807 25 101 2051817 25 103 2101827 28 116 2371837 31 126 2561847 36 146 2971857 36 146 2971867 49 199 4041877 51 211 4281887 59 241 4911902 65 75 266 5411913 65 75 255 5231929 58 83 239 5301938 53 74 216 4211947 55 110 198 3521956 55 128 223 3501994 36 89 252 352

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources) No datasource exists to compute columns (3)ndash(4) prior to 1902 and the numbers reported on this table for years 1807ndash1887 werecomputed assuming that the columns (3)ndash(4) followed the same trend as column (1) over the 1807ndash1902 period (see textSection III)

247VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

the average wealth levels for top fractiles (P99ndash100 P995ndash100 P999ndash100 and P9999ndash100)Lower thresholds of the national wealth distri-bution (P90 and P95) were computed using thenational TRA survey21 and the P90ndash95 andP95ndash99 intermediate wealth levels were alsocomputed using Pareto interpolation techniques

The national top estate shares estimates re-ported in Table 4 were computed using thismethodology They suggest that wealth concen-tration (as measured by the top-1-percent estateshare) rose throughout the nineteenth century inFrance during both the 1807ndash1867 and 1867ndash1902 periods although less sharply than in Parisduring the latter period (see Figure 7) Theseestimates are conservative in the sense that it isalmost certain that they underestimate the riseof wealth concentration that took place inFrance during the nineteenth century First weknow that the bulk of population growth in Parisduring the nineteenth century was due to theannexation of suburbs in 1860 and to populationgrowth in these peripheral arrondissements Be-

cause the outskirts of the city were poor theannexation added few top estates Thus there islittle doubt that Parisrsquos share of top estates inFrance actually increased less than its share of thetotal population This hypothesis is confirmed bynineteenth-century housing tax tabulations show-ing that the fraction of Paris taxpayers in thenational top 1 percent of taxpayers was substan-tially larger than 10 percent at the beginning of thenineteenth century22 Giving Paris a larger (andmore realistic) share of top estates in 1807 wouldboth reduce the share of wealth of the top 1percent in France at that date and lead to morerapid rise in inequality over time

Next and most importantly other estate sur-veys are consistent with the view that wealth in-equality was growing The important study byAdeline Daumard (1973) which relied on samplesof estate tax returns collected in Paris Lyon Tou-louse Lille and Bordeaux found that wealth con-centration increased in each of these five citiesduring the nineteenth century23 The TRA surveyalthough it is ill-suited for the study of top estatesis also consistent with our view Wealth dispersionwas on the rise in nineteenth-century France ac-cording to the TRA survey both in the sense thatthe fraction of decedents with positive estates de-clined over time (in spite of the sharp increase inthe value of the average estate) and that ratios suchas the P90P50 increased24 We also comparedour benchmark national P99 series extrapolatedfrom our Paris samples and the national P99series computed using the TRA survey We foundthat both series display the same overall upwardtrend in concentration (which is reassuring regard-ing the general validity of our Paris-France extrap-olation technique) except that the growth ofinequality from 1807 to 1902 in the TRA series is

21 See Bourdieu et al (2003) for full details about theTRA survey The P90 and P95 thresholds reported onPiketty et al (2004 Appendix Table A3) were computedusing ten-year moving averages around the target years inorder to make sure that each estimate was based on asufficient number of observations

22 These tabulations were published in the same FinanceMinistry official publications as the estate tabulations Wechose not to use them in our formal computations becausethe tax base of the housing tax (namely the rental value ofthe real estate property where the household lives) is onlyloosely connected to the estate tax base (in particular onecannot rule out the possibility that the housing tax baseoverrepresents Paris-based taxpayers)

23 Unfortunately Daumardrsquos samples are not available inmachine-readable format she has only two or three years ofdata for each city and she did not try to compute homog-enous inequality indicators (top fractiles shares etc) withher data Thus although her results and our work are con-sistent they cannot be compared directly

24 See Bourdieu et al (2004)

TABLE 4mdashWEALTH CONCENTRATION AT DEATH IN FRANCE1807ndash1994

Top-10-percentestate share

Top-1-percentestate share

Top-01-percentestate share

1807 791 434 1631817 810 445 1811827 824 452 1631837 796 438 1471847 816 479 1841857 829 495 1741867 810 480 1741877 838 471 2011887 839 487 1921902 839 516 2311913 863 549 2601929 820 502 2471938 776 420 1991947 699 299 1101956 694 304 1101994 610 213 63

Source Authorsrsquo computations using estate tax returns (seePiketty et al 2004 Table A3 for detailed series)

248 THE AMERICAN ECONOMIC REVIEW MARCH 2006

more severe than in our series (see Figure 8) Thisagain suggests that the latter provide a conserva-tive lower bound for the upward trend in wealthconcentration In any case the finding of a largeincrease in wealth inequality in nineteenth-centuryFrance (and up until World War I) appears to berobust25

As was mentioned earlier there exists nocomparable continuous data source covering thenineteenth and twentieth centuries in othercountries which makes it difficult to put ourFrench long-run series in international perspec-tive We note however that existing series forthe United States and the United Kingdom areconsistent with our findings on France26 Re-garding levels existing evidence suggests thatduring the nineteenth and most of the twentiethcenturies France was in an intermediate posi-tion in terms of wealth concentration in be-tween the United States (more equal) and theUnited Kingdom (the most unequal)27 These

25 Note that this continuous rise in wealth inequalitydoes not necessarily imply that a parallel rise occurredregarding income inequality Given that there exists nomicro source on incomes prior to the creation of the incometax in 1914 it is very difficult (if not impossible) to properlyaddress this issue Christian Morrisson and Wayne Snyder(2000) have attempted to link income inequality estimatesbased upon Old-Regime fiscal sources (pre-1789) withmodern income-tax-based twentieth-century estimates andthey have argued that income inequality might have startedto decline during the later part of the nineteenth century andon the eve of World War I (see also Morrisson 2000)Although our data do not allow us to rule out such apossibility we note that their nineteenth-century personaldistribution estimates are based on fragile macroeconomicdata on functional distribution and are not homogenous totheir eighteenth- and twentieth-century estimates Given theevidence that we provide on wealth inequality any signif-icant decline in aggregate income inequality would have tobe associated with severe compression of the wage distri-bution There is little research on this issue however

26 Lee Soltow and Jan L Van Zanden (1998) also find adecline in inequality in the twentieth century in the Neth-erlands Their data are consistent with a rise in inequality inthe nineteenth century but they have no direct evidenceabout its extent

27 According to our series the top-1-percent wealthshare in France rose from around 45 percent in 1800 toabout 55 percent around World War I and then fell to about20 percent by the end of the twentieth century (see Figure7) wealth concentration among the living appears to besomewhat larger (see Section V below) According to theseries pieced together by Lindert (2000 pp 181ndash82 and186) the UK top-1-percent wealth share rose from about

FIGURE 8 ESTIMATES OF THE P99 THRESHOLD FOR THE FRENCH DISTRIBUTION OF ESTATES AT DEATH EXTRAPOLATION

FROM PARIS SAMPLES VERSUS ESTIMATES FROM TRA SAMPLES (CURRENT FRENCH FRANCS)

Source Authorsrsquo computations based on estate tax returns (Paris samples and TRA samples)

249VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

differences in inequality largely hark back to dif-ferential concentration of landownership En-glandrsquos land was extremely narrowly held whilethe United States was most egalitarian The impactof access to real estate assets can also be seenwhen we contrast Paris (where it was extremelyconcentrated) with the provinces

Trends in inequality over time may be easierto compare because biases in source materialmay have a significant effect on levels ratherthan trends The historical pattern is similar ineach of the three economies In particular thereis evidence that wealth concentration increasedduring the nineteenth century in both Anglo-Saxon countries and declined during the twen-tieth century with a turning point around WorldWar I28 Neither exhibits patterns consistentwith a Kuznet process

IV From the Wealth of the Decedents to theWealth of the Living

The estimates reported thus far refer to in-equality among decedents as described in thetax returns filed by their heirs The evolution ofthe distribution of wealth among the livingmight however have followed a different pat-tern In order to convert wealth-at-death con-centration estimates into wealth-of-the-livingconcentration estimates it is standard to use theldquoestate multiplierrdquo method29 It consists ofweighting each observation of an estate at death

by the inverse of the mortality rate for this agegroup That is if the mortality rate for ages 20to 24 was 068 percent in Paris in 1902 theneach decedent aged 20 to 24 represented about147 living individuals of the same age (100068 147) Similarly if the mortality ratefor ages above 80 was 2143 percent in Paris in1902 then each decedent in that group repre-sented about 47 living individuals in the sameage group (102143 47) Applying thismethod requires mortality tables (these are eas-ily available) and estate tabulations brokendown by estate size and age at death (these arescarcer) Fortunately the cityrsquos statistical bu-reau published annual death-by-age totals theFrench censuses report the age distribution forthe capital every five years and we collectedage at death from the estate declarations Thesedata allowed us to compute estimates of wealthconcentration among the living over the 1807ndash1902 period using various assumptions aboutthe wealth profiles of mortality rates

The base population for the living is the set ofall individuals aged 20 and over living in Parisin year t which we denote pt The number ofliving individuals in age bracket a is denoted pta(a 20ndash24 25ndash29 30ndash34 75ndash79 80 andover) and the number of decedents in agebracket a is denoted nta The mortality rate forage bracket a is given by mta ntapta Webegin with a uniform-mortality benchmarkThese estimates are based on the simplifyingassumption that these mortality rates dependsolely on age and are the same for all wealthgroups (and in particular are the same forzero-wealth and positive-wealth individuals)We can then weigh each decedent with positiveestate and age a collected in the Paris archivesin year t by ptanta This allows us to computethe number of living Parisians with positivewealth in year t and also (by differentiatingwith pt) the number of living Parisians with zerowealth at year t which is used to weight zero-estate observations We then use our weighteddatasets to compute top estate fractiles amongthe living in Paris

The main conclusion is that the living expe-rienced the same upward trend in wealth con-centration as the decedents (see Figure 9) Wefind that inequality was significantly higheramong the living than among decedents be-cause survivors were on average younger than

55 percent in 1800 to 70 percent around World War I thenfell to about 20 percent in the 1990s The US top-1-percentwealth share rose from about 15 to 20 percent in 1800 toabout 40 percent around World War I then fell to about 30percent in the 1990s (and as low as 20 to 25 percentaccording to the more recent estimates from Kopczuk andSaez 2004) Wealth concentration is now larger in theUnited States than in European countries but the reversewas true during the nineteenth century up until World WarII (It is only since the 1950sndash1970s period that US wealthconcentration has been somewhat larger)

28 See Lindert (2000 pp 181ndash82 and 188)29 This method was widely used in England and France

in the late nineteenth and early twentieth centuries to com-pute the stock of national wealth on the basis of the flow ofwealth transmitted at death Standard references that use thistechnique to estimate the wealth distribution of the livingfrom estate tax data tabulated by estate size and age at deathinclude Atkinson and Harrison (1978) and Lampman(1962) For a more recent application of this technique tothe United States see Kopczuk and Saez (2004)

250 THE AMERICAN ECONOMIC REVIEW MARCH 2006

those who died and the young were on aver-age poorer In particular the estate multipliermethod leads to lower average weights forpositive-wealth decedents than for zero-wealthdecedents (the former are on average older andtherefore ldquorepresentrdquo a smaller number of livingindividuals) As a result the fraction of positive-wealth individuals is even smaller among theliving than among decedents Hence our bench-mark uniform-mortality estimates of wealthconcentration among the living are significantlylarger than corresponding estimates among de-cedents (eg top-1-percent wealth shares areabout 15 percent larger among the living)Changes over time however are similar In-creased life expectancy and declining mortalityrates over the course of the nineteenth centuryhave only a small effect on the trends

In order to make the estate multiplier methodmore reliable one would prefer to take intoaccount differential mortality by wealth Doingso would require having access to mortalityschedules based both on wealth and age at dif-ferent points in time unfortunately these are notavailable We have nonetheless reestimatedwealth of the living based upon the same as-sumption as Kopczuk and Saez (2004) That is

we assumed uniform mortality among the poor(here defined as zero-wealth individuals) andamong the rich (here defined as positive-wealthindividuals) and we assumed that the ratiomtaRmtaP between the mortality rate of the richand the mortality rate of the poor followed aU-shaped age profile from about 85 percent forthe young (ie the rich die 15 percent less oftenthan the poor when they are 20ndash24 or 25ndash29years old) down to about 70 percent for middle-age individuals in their forties to fifties and upto 100 percent for very old individuals in theireighties to nineties30 This profile correspondsto the best available estimates in the literatureand it appears to be relatively stable over timeand across developed countries In the absenceof better data it is the best one can do31 Thebenchmark differential-mortality estimates re-ported on Figure 9 show that although addingdifferential mortality produces different levelsof inequality it does not have much impact onthe upward trend in concentration

30 See Kopczuk and Saez (2004 Table A4)31 See Kopczuk and Saez (2004 Appendix B) for refer-

ences to the US and international literature devoted to theage-wealth profile of mortality rates

FIGURE 9 WEALTH CONCENTRATION AMONG DECEDENTS AND AMONG THE LIVING IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives (see Piketty et al 2004Table A4 for detailed series)

251VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

The resulting differential-mortality inequalityestimates lie between those based on decedentsonly and those using uniform mortality for theliving (see Figure 9) Note that moving fromuniform-mortality to differential-mortality es-tate multiplier methodology can either increaseor decrease inequality Here the reason whysuch a move leads to lower wealth concentra-tion seems to be due to the fact that differentialmortality tends to put higher weights onpositive-wealth decedents (for a given age)thereby increasing the estimated fraction of liv-ing individuals with positive wealth The im-portant point however is that the resulting leveleffects are relatively small in magnitude con-stant in time and dwarfed by the upward timetrend Even if we were to assume an enormousincrease in differential mortality during thenineteenth century in the sense that differentialmortality between the rich and poor was equalto 0 percent of the benchmark differential in1807 and 100 percent of the benchmark differ-ential in 1902 the resulting wealth concentra-tion estimates would still be significantly higherin 1902 than in 1807 (see Figure 9) Yet wehave no reason to believe that differential mor-tality increased to such an extent During thenineteenth century real wages for unskilledworkers rose which would have reduced mor-tality more for the poor than for the rich After1850 public health measures (sanitation watervaccination) were in place Again these wouldhave had a significant effect on the poor and themiddle class who could not privately purchasesuch health-improving services To be sure therich could avail themselves of more medicalservices than the poor or the middle class butthe impact of these services was probably small(bear in mind that neither antibiotics nor car-diovascular interventions were available)

Finally we have applied the estate multipliermethod to available data for 1947 and 1994 Over-all the sharp decline in wealth concentration ob-served during the twentieth century (andespecially between 1914 and 1945) is very robustIf anything the decline appears to be even largerwhen one looks at wealth concentration amongthe living rather than among decedents32

V The Changing Age Profile of Wealth

In the previous sections we focused almostexclusively on aggregate top wealth shares Ourdata however also detail the characteristics ofeach decedent in particular their gender andage The evolution of wealth by gender is ofrelevance for over the past two centuries therehave been massive changes in womenrsquos laborforce participation capacity to manage theirown affairs and life expectancy relative to menThe evolution of wealth by age is of relevancebecause there was a significant increase in adultlife expectancy over the twentieth century andbecause the progressive diffusion of pensionsmay have changed savings motivations More-over age-wealth profiles also inform us aboutthe motives of wealth accumulation and theeconomic impact of high wealth concentration

A first pass at the data considers the genderbreakdown of wealth at death Remarkably inour micro data the share of women in top estatestakes its highest value on the eve of World WarI For instance the womenrsquos share in the top 05percent rose from 35 percent prior to 1850 to 45percent in 1902 only to fall to 40 percent afterWorld War II Strikingly womenrsquos share ofwealth follows almost exactly the pattern ofaggregate inequality Women were relativelyricher when inequality reached its apex inFrance than at any other time Moreover insti-tutional variables seem to have played almostno role in changing the relative wealth ofwomen Unlike in common law countriesFrench law starting with the code civil of 1804required nearly equal treatment of all children inbequests Further research will help us deter-mine to what extent women of great wealthwere heirs or part of economically very success-ful couples

The data also reveal striking changes in the

32 See Piketty et al (2004 Table A4) It is unfortunatelynot possible to construct complete series for wealth concen-

tration among the living for the twentieth century due todata limitations tables broken down by estate brackets andage of decedents are available solely for years 1943ndash1954and at the national level (no table broken down by estatebrackets and age of decedents has ever been compiled at thedepartement level except in 1931 for Seine departementsee E S Danysz 1934) and the 1994 micro sample is notlarge enough to allow for a reliable application of the estatemultiplier method at the Paris level Thus the only wealth-of-the-living concentration estimates we provide for thetwentieth century are national estimates for 1947 and 1994

252 THE AMERICAN ECONOMIC REVIEW MARCH 2006

age profile of wealth between 1807 and 1994(see Table 5) During the nineteenth century aswealth concentration was increasing the veryrich were getting older and older At the begin-ning of the nineteenth century in the aftermathof the French Revolution the richest individualswere those in their fifties they were typically100 percent richer on average than people intheir forties 25 percent richer than those in theirsixties and 40 percent richer than those in theirseventies and eighties By the 1870s howeverthe age-wealth pattern had become stronglymonotonic the richest individuals were the old-est individuals In 1902 people in their sixtiesand seventies bequeathed 150 percent morethan those in their fifties and those in theireighties 300 percent more On the eve of WorldWar I top wealth holders were old and likely tobe retired This pattern breaks some time duringthe 1914ndash1945 period33 In 1947 as well as in1994 we are back to a pattern where the richestindividuals are those in their fifties Overall theperiod of maximal wealth inequality (1860ndash1913) also appears to be a period characterizedby a very specific age profile of wealth and largeconcentration of assets among the elderly

Another way to analyze the changing age-wealth relationship is to look at average age by

top estate fractile34 In 1817 average age wasvirtually the same for the top 10 percent and thetop 1 percent of estates (or even slightly declin-ing) The average-age-per-fractile relationshipbecomes upward sloping during the nineteenthcentury and by 1902 those in the top 1 percentwere almost six years older than those in the top10 percent The relationship is flat in 1947 anddownward-sloping in 1994 Finally one canapply the estate multiplier method (see SectionIV above) and analyze how wealth concentra-tion by age group among the living changedover the course of the nineteenth century Thegeneral population in Paris did not becomeolder during the nineteenth century those aged60 or older made up about 15 percent of thepopulation in 1817 and after 1847 about 10 to11 percent35 The share of total wealth ownedby the elderly rose significantly however aswealth distribution worsened The wealth be-longing to those aged 60 or more rose fromabout 25 to 30 percent of the total at the begin-ning of the nineteenth century to about 40 to 45percent by the end of the century The wealthshare of those aged 70 or older doubled fromless than 10 percent to about 20 percent36

33 Existing evidence on the age-wealth profile for 1931(see Danysz 1934) suggests that the Great Depression andWorld War II (rather than World War I) played the leadingroles in breaking this pattern This is an issue we plan toinvestigate in future research

34 See Piketty et al (2004 Table 6)35 Although life expectancy was increasing which

should have led to large shares of population for oldergroups the city was also growing quickly The large num-ber of immigrants (who were typically in their twenties)increased the relative size of the younger cohorts (seePiketty et al 2004 Table 7 and Figure 11)

36 See Piketty et al (2004 Figure 12)

TABLE 5mdashTHE AGE PROFILE OF WEALTH AT DEATH IN PARIS 1817ndash1994(Average estate left by 50- 59-year-old 100)

20ndash29yr-old

30ndash39yr-old

40ndash49yr-old

50ndash59yr-old

60ndash69yr-old

70ndash79yr-old

80ndash89yr-old

90ndash99yr-old

1817 26 22 28 100 54 59 591827 44 50 53 100 88 87 601837 133 90 107 100 116 123 1101847 87 73 102 100 117 204 1321857 84 77 101 100 104 109 1451867 67 58 136 100 141 125 1541877 66 73 63 100 197 260 4301887 45 33 63 100 152 233 2951902 29 40 80 100 253 272 4011947 31 51 73 100 113 105 105 1091994 11 45 100 87 93 95 68

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources)

253VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

It is perhaps not surprising that inequalitybecame strongly correlated with age in the1860s Those who died at an old age in the1820s and 1830s had lived through the difficultyears of the French Revolution and the disloca-tion of the end of the Napoleonic period Theirability to accumulate wealth had been severelyimpaired Furthermore they were rather lesslikely to inherit much wealth since the Revo-lution wiped out the bond portfolios of theirparents through a prolonged period of high in-flation Those who died from the 1860s to theearly 1910s did not suffer from the adversitiesthat plagued their forebears Instead they en-joyed the fruits of the financial sector expansionthat began in the 1850s After 1947 we seem tohave returned to a situation quite like that of1817 Presumably the capital damages associ-ated with both world wars and the Great De-pression had a strong negative effect on thewealth holdings of older generations The per-sistence of a flat age-wealth profile until 1994 islikely to be associated with two factors First insocieties where income growth is rapid abso-lute wealth accumulation is faster by youngercohorts than by older ones because their in-comes are higher at every age This is an im-portant distinction between the nineteenthcentury and the twentieth century Furthermorehighly progressive rates of income and estatetaxation have probably made it more difficult toaccumulate large fortunes thereby flattening theobserved age-wealth profile

The more interesting (and more difficult)question relates to the possible efficiency im-pact of high wealth concentration and changingage-wealth profiles Although our data do notallow us to address efficiency issues in a rigor-ous way our results allow us to formulate anumber of hypotheses and to shed new light onthe ongoing debate on inequality and growth37

From a theoretical viewpoint whether highwealth concentration can have a negativegrowth impact depends critically on the exis-tence of credit constraints With first-best credit

markets money flows toward the best entrepre-neurs and investment projects irrespective ofthe initial distribution High levels of wealthconcentration can be bad from a social justiceviewpoint but they entail no efficiency lossWhen credit constraints bind however initialwealth matters and high levels of inequality canhurt growth Whether the loss is large or smalldepends on who owns the assets If the rich areefficient investors (they know which projects tofund etc) then wealth concentration may evenbe useful If the rich are retired rentiers how-ever investing their wealth in low-yield assets(or low-ability inheritors) then high wealthconcentration and credit constraints might pre-vent talented but penniless investors from un-dertaking efficient projects thereby entailingnegative growth consequences The data used inthis paper are not ideal to address whether creditconstraints were important in a country likeFrance at the end of the nineteenth century Ourresults suggest however that to the extentcredit constraints were indeed severe highwealth concentration did have a negativegrowth impact In order to investigate this hy-pothesis further one would need to gather moresystematic data on investment strategies andasset returns Preliminary evidence suggeststhat the rich elderly of the 1860ndash1913 perioddid indeed hold a disproportionate fraction oftheir wealth in low-yield assets (such as gov-ernment bonds) An alternative hypothesishowever is that steeper age-wealth profileswere the consequence of the growth of financialmarkets as their children faced fewer creditconstraints parents decided to hold on to moreof their wealth

VI Conclusion

Evidence from wealth at death in Paris and inFrance over the last two centuries reveals threekey patterns First wealth concentration haschanged dramatically over time In 1807 thetop-1-percent share of wealth (40 percent inFrance 50 percent in Paris) was twice as high asit would be in 1994 but substantially less thanin 1913 when it peaked above 55 percent inFrance and 70 percent in Paris Some of thesechanges were due to economic phenomena thathave long been emphasized as creating inequal-ity namely industrialization and financial cen-

37 Thus far this literature has concentrated upon cross-country regressions of inequality on growth a methodologythat raises serious identification problems especially giventhe low quality of available international datasets on in-equality which are neither long-run nor homogeneous (seeeg Atkinson and Andrea Brandolini 2001)

254 THE AMERICAN ECONOMIC REVIEW MARCH 2006

tralization Yet the decline comes largely fromadverse shocks rather than economic conver-gence These changes are of such magnitudethat they are not sensitive to whether one ex-amines wealth at death in Paris or in France orwhether one examines it directly rather thanconverting it to wealth of the living by an estatemultiplier method

Our second key result is that there was asignificant transition during the nineteenth cen-tury from an important role for real estate as aform of wealth to moveable assets as the keyform of wealth for the very rich Similarly theshare of wealth held by aristocrats first rose andthen was eclipsed by that of financiers and in-dustrialists in the second half of the nineteenthcentury Hence mobility within this highly un-equal society might have been quite high Yetthis conjecture is tempered by our third findingthe wealthy were getting older over time andolder relative to less wealthy decedents Suchaging among the very wealthy would have hadnegative consequences for growth if financialmarkets were imperfect This issue requires fur-ther investigation and we hope it will attractfuture research

REFERENCES

Atkinson Anthony B ldquoTop Incomes in the UKover the 20th Centuryrdquo Journal of the RoyalStatistical Society Series A (Statistics in So-ciety) 2005 168(2) pp 325ndash43

Atkinson Anthony B and Brandolini AndrealdquoPromise and Pitfalls in the Use of lsquoSecond-aryrsquo Data-Sets Income Inequality in OECDCountries as a Case Studyrdquo Journal of Eco-nomic Literature 2001 39(3) pp 771ndash99

Atkinson Anthony B and Harrison Alan J Dis-tribution of personal wealth in Britain Cam-bridge Cambridge University Press 1978

Atkinson Anthony B and Piketty Thomas Topincomes over the twentieth century A con-trast between European and English speak-ing countries Oxford Oxford UniversityPress (forthcoming)

Banerjee Abhijit V and Duflo Esther ldquoInequalityand Growth What Can the Data Sayrdquo Journalof Economic Growth 2003 8(3) pp 267ndash99

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoPourquoi la richesse ne

srsquoest-elle pas diffusee avec la croissance Ledegre zero de lrsquoinegalite et son evolution enFrance 1800ndash1940rdquo Histoire et mesure 200318(1ndash2) pp 147ndash98

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoDefense et illustration delrsquoenquete des 3000 familles Lrsquoexemple deson volet patrimonialrdquo Annales de demogra-phie historique 2004 19 pp 19ndash57

Danysz E S ldquoContribution a lrsquoetude des for-tunes privees drsquoapres les declarations de suc-cessionsrdquo Bulletin de la statistique generalede France 1934 XXIV (1) pp 111ndash171

Daumard Adeline Les fortunes francaise auXIXe siecle Enquete sur la repartition et lacomposition des capitaux prives a ParisLyon Lille Bordeaux et Toulouse drsquoapreslrsquoenregistrement des declarations de succes-sion Paris Mouton 1973

Gain Andre La restauration et les biens desemigres 2 vols Nancy Societe drsquoimpressiontypographique 1929

Jones Alice H American colonial wealth Doc-uments and methods 3 vols New York ArnoPress 1977

Kopczuk Wojciech and Saez Emmanuel ldquoTopWealth Shares in the United States 1916ndash2000 Evidence from Estate Tax ReturnsrdquoNational Tax Journal 2004 57(2) pp 445ndash87

Kuznets Simon ldquoEconomic Growth and IncomeInequalityrdquo American Economic Review1955 45(1) pp 1ndash28

Lampman Robert J The share of top wealth-holders in national wealth 1922ndash1956Princeton Princeton University Press 1962

Lescure Michel Les banques lrsquoEtat et lemarche immobilier en France a lrsquoepoquecontemporaine 1820ndash1940 Paris EditionslrsquoEcole des Hautes Etudes en Sciences Socia-les 1982

Levy-Leboyer Maurice and Bourguignon Fran-cois LrsquoEconomie francaise au XIXe siecleAnalyse macroeconomique Paris Eco-nomica 1985 English translation Cam-bridge Cambridge University Press 1990

Lindert Peter H ldquoUnequal English Wealthsince 1670rdquo Journal of Political Economy1986 94(6) pp 1127ndash62

Lindert Peter H ldquoThree Centuries of Inequalityin Britain and Americardquo in Anthony B At-kinson and Francois Bourguignon eds

255VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

Kutznets would have thought Did it stabilizeDid it keep increasing until World War I OurFrench series allow us to cast new light on thiscentral issue because they are homogeneousover the 1807ndash1994 period

Our main conclusions are the followingFirst wealth concentration in Paris and inFrance increased up to World War I with anacceleration (rather than a stabilization) of thetrend at the end of the period The bulk of therise in inequality actually took place during the1860ndash1913 period This was largely driven bythe growth of large industrial and financial es-tates and coincided with the decline of aristo-cratic fortunes During the first half of thenineteenth century the share of aristocrats intop estates actually rose Next the decline inwealth concentration observed after World WarI appears to have been driven by the 1914ndash1945shocks rather than by a two-sector Kuznets-type process The decline in inequality was notdue to a reduction in the gap between Paris andthe provinces since it occurred both in Parisand in the rest of France Finally and perhapsmost importantly the very high levels of wealthconcentration observed at the eve of World WarI seem to have been associated with retiredindividuals who had lived off capital income(henceforth rentiers) rather than with active en-trepreneurs In particular the age-wealth profileof decedents is markedly steeper between 1900and 1913 than in other periods Top wealthholders were very old at the turn of the lastcentury (in their seventies and eighties)whereas they are usually in their fifties in otherperiods both at the beginning of the nineteenthcentury and at the end of the twentieth centuryAlthough our data do not allow us to evaluatethe inefficiency of wealth concentration di-rectly these results shed new light on the on-going debate about inequality and growth Thatis to the extent that credit constraints wereimportant in 1900 France (which we cannotprove directly with our data) our findings aboutthe changing age profile of wealth suggest thathigh wealth concentration might have been as-sociated with lower growth9

The paper is organized as follows Section Idescribes our data sources and outlines ourmethodology Section II presents our estimatesof wealth concentration and composition atdeath in Paris Section III discusses how theestimates from nineteenth-century Paris can beextended to the rest of France and presents pre-liminary results for wealth concentration at deathin France from 1807 to 1994 Section IV showshow our data on wealth and age at death can beused to estimate series on wealth concentrationamong the living using the estate multipliermethod Section V examines age-wealth profilesand discusses the efficiency implications of highwealth concentration Section VI concludes

I Data Sources

All of our estimates are based upon estate taxreturns As noted above the estate tax wascreated in 1791 and it became a progressive taxin 1902 Since then the tax administration hasperiodically compiled tables indicating thenumber of decedents and the value of theirestate for a large number of estate bracketsThese tables were already used by Piketty(2001 2003) and they are available over the1902ndash1994 period10 They were compiled andpublished by departement (middle-level admin-istrative jurisdictions there are about 90 ofthem in France including Paris)11 These tablescan be used to study the evolution of wealth

9 One way to test directly for the efficiency impact ofhigh wealth concentration would be to look at investmentpatterns across wealth fractiles and age groups (ie the

extent to which older wealth holders invest their wealth inlow-yield assets) The sources we use lend themselves toprecisely this kind of investigation and we intend to con-tinue this practice in further research

10 These tabulations were published in the official statis-tical publications of the French Finance Ministry (for exactreferences and page numbers see Piketty 2001 AppendixJ) The basic national tabulation indicating the number ofdecedents and amount of their estate for a large number ofestate brackets is available for the following years 1902ndash1913 (except 1906 and 1908) 1925ndash1960 (except 1928 and1934) 1962 and 1964 The French tax administrationstopped compiling such tables in 1964 but micro-files in-cluding large national samples of estate tax returns areavailable for 1984 and 1994 (in the present paper we useonly the 1994 micro-file)

11 Tables by estate brackets are available at the departe-ment level for the following years 1902ndash1913 (except 1906and 1908) and 1925ndash1958 (except 1928 and 1934) for otheryears tables by estate brackets are available only at thenational level In addition national tables broken down byestate brackets and age of decedents are available for years

239VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

concentration both in France and in Paris duringthe twentieth century using standard Pareto in-terpolation techniques

Prior to 1902 the tax administration pub-lished only the aggregate amount of wealth re-ported on estate tax returns broken down byreal (land and buildings) and personal (furni-ture businesses stocks bonds etc) assets12

Studying concentration thus required collectingour own samples of individual returns Collect-ing information on every individual return fromevery departement for a given year was impos-sible It would have required going to the ar-chives of each departement acquiring access tothe registers of each tax bureau (the lowest-level tax jurisdiction) and dealing with hun-dreds of thousands of declarations each yearWe therefore had to devise a sampling strategyOne option was to select randomly (eg on thebasis of birth dates or family names) a nation-ally representative sample of decedents for var-ious years during the nineteenth century Thatsample would need to be extremely large how-ever to include enough large estates (Given

that wealth is extremely concentrated it is crit-ical to observe many of the very wealthy)

Therefore we decided to pursue a completelydifferent strategy and collected data for all de-cedents in Paris for selected years (1807 18171827 1837 1847 1857 1867 1877 1887 and1902) We chose Paris because a disproportion-ate share of the very rich lived there As one cansee from Table 1 around 1810 the annual num-ber of decedents (20 years old and over) in Pariswas about 12000 (25 percent of the Frenchtotal) that figure nearly tripled during the nine-teenth century to about 35000 by 1900 (65percent of the French total) However only 30percent of decedents in Paris had an estate dur-ing the nineteenth century (about half as manyas in the rest of France) so we needed only tocollect detailed information on 3500 decedentsor so per year at the beginning of the nineteenthcentury and 10000 or so decedents per year atthe end (see Table 1) Although Paris had moredecedents with zero wealth than the rest of thecountry the average estate was about 45 timeslarger in Paris than elsewhere in France duringthe nineteenth century13 It is particularly strik-ing to notice that this ratio actually increased

1943ndash1954 The 1994 micro-file also allows us to breakdown the data by departement and age

12 These published aggregates were computed by theadministration on the basis of tax receipts

13 Average estates as well as top estate fractiles arealways defined in this paper over the set of all decedentsaged 20 and older including those with zero wealth

TABLE 1mdashESTATE TAX RETURNS IN PARIS 1807ndash1994mdashSUMMARY STATISTICS

N decedents20-yr N estate 0

N estate 0(percent

N deced 20)

N deced 20-yr(percent

ParisFrance)

Total estate(percent

ParisFrance)

Average estate(Ratio Parisrest

of France)

1807 11622 3647 314 25 82 3561817 11925 3287 276 25 84 3561827 14151 3877 274 28 94 3561837 16902 4922 291 31 98 3421847 18169 4814 265 33 115 3861857 19248 6048 314 36 143 4511867 26844 7370 275 46 168 4161877 28777 8245 287 51 186 4221887 34411 9815 285 59 201 4011902 36366 9830 270 65 260 5051913 35677 11927 334 65 266 5231929 35842 14495 404 58 250 5421938 30274 16013 529 53 173 3761947 24955 14090 565 55 150 3071956 27940 16053 575 55 159 3241994 18553 12528 675 36 97 286

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed series and sources)

240 THE AMERICAN ECONOMIC REVIEW MARCH 2006

over time in spite of the fact that Paris nearlytripled in population14 On the eve of WorldWar I the estates of Paris decedents made upover 26 percent of the French total (see Table1 and Figure 2)

We designed our data collection to take ad-vantage of the work of the estate tax adminis-tration For every person who either died inParis or might have taxable wealth in any of oneof Parisrsquos nine bureaux the administrationopened an entry in a first set of volumes (thetables des successions et absences henceforthTSA) Later the entry was completed eitherwhen estate taxes were paid or when the admin-istration became satisfied that the individual hadleft no wealth behind The entries include nameoccupation residence marital status age andfor individuals with wealth information aboutheirs and the date at which the declaration was

filed Up to 1870 the TSA also include a sum-mary of the individualrsquos estate broken downinto personal wealth and real estate Hence thecross sections up to 1867 rely heavily on theTSA After 1870 the administration no longerrecorded wealth information in the TSA butonly whether returns had been filed for theindividual For 1877 1887 and 1902 westarted with the TSA and for each individual forwhom a return had been filed we collected thefirst three letters of the last name gender ageday and month of death and the date(s) atwhich returns had been filed We then opened asecond set of registers (the registres de muta-tions par deces henceforth RMD) where a com-plete description of the estates is transcribedand the information not gleaned in the TSA wasappended to the first set of entries

Yet these data gave information by tax returnnot by individual A decedentrsquos heirs could filemultiple returns either because they amendedtheir original declaration or before 1902 be-cause they paid taxes in multiple bureaux In-deed prior to 1902 estate taxes on real estatewere paid in the bureau of the asset rather thanthat of the residence of the decedent In an eraof strictly proportional taxation such dispersed

14 Note that there is a discontinuity in the growth of Parisduring the nineteenth century as new districts (arrondisse-ments) previously registered in the suburb were integratedinto the City of Paris in 1860 The results reported here donot make any correction for this discontinuity which ex-plains the discontinuity observed in some of the figuresaround 1860

FIGURE 2 THE PARIS SHARE IN FRENCH ESTATES AT DEATH 1807ndash1994

Source Authorsrsquo computations based on estate tax returns (see Table 1)

241VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

payment of taxes reduced administrative costsbecause information about real estate values didnot have to be centralized Naturally when theestate tax became progressive in 1902 returnshad to be unified Because TSA entries providelinks to the different declarations in the samebureau reassembling these declarations waseasy reassembling returns from different bu-reaux was another matter because there are nolinks across bureaux To reassemble fully allindividual returns would have required us tocollect detailed information on every decedentin Paris But we did not need to do so becausethe very high levels of inequality in Paris cameto our assistance By collecting nominal infor-mation on the top 10 percent of returns we wereable to attribute successfully 92 percent of mov-able assets and 97 percent of all real assets tospecific individuals15 Given the high variety offirst names and last names as well as detailedinformation on residence the likelihood offalsely positive matches is very low The re-maining returns were treated as individualsthus biasing downward our inequality estimates

Our 1902 Paris sample is fully consistentwith the table compiled by the tax administra-tion for the same year for Paris Therefore wecan link up our 1807ndash1902 Paris files with the1902ndash1994 Paris tables to construct homoge-neous 1807ndash1994 series for inequality in ParisThe more difficult task is to use the Paris data toinfer changes in wealth concentration for Francefrom 1807 to 1902 To do so we must estimatehow the relative importance of Paris in each topestate class evolved over the nineteenth centuryTo achieve this goal we used other estate sur-veys16 as well as nonestate fiscal sources (seeSection III below) The other difficult part is theconstruction of estimates for wealth concentra-tion among the living from estate tax datawhich we do using the estate multiplier method

and mortality data by age group (see Section IVbelow)

When using tax data it is also important tokeep in mind that tax evasion and manipulationcan potentially bias the results There are how-ever good reasons to believe that this is not toomuch of a problem here First estate tax rateswere extremely modest until World War I (lessthan 2 percent) which implies that the incen-tives for tax evasion were small In contrastpenalties for evasion were stiff Moreover theadministration made every effort to keep upwith changing composition of assets and totrack down individuals with some wealthAmong other things financial institutions andpublic utilities were required to notify the ad-ministration when accounts changed owners Asa result it was not easy to dissimulate thewealth of a decedent (either real estate or finan-cial assets in a publicly traded firm) and inher-itors had a strong incentive to register theirproperty in order to benefit from state protec-tion This suggests that the nineteenth-centurydata collected in the Paris archives is probably ofvery high quality Tax evasion is potentially amore serious issue for the twentieth century whentax rates become substantial Although top estatetax rates have rarely exceeded 20 to 30 percent fordirect transmissions in France (the top rate hasbeen 40 percent since 1984 its highest level ever)it is obvious that incentives for tax evasion haveincreased over time However several indepen-dent data sources suggest that the trends observedduring the twentieth century are robust and are notdue to the rise of tax evasion17

II Wealth Concentration at Death in Paris1807ndash1994

Figure 3 shows the evolution of wealth con-centration at death in Paris from 1807 to 1994Given that the bottom two thirds of the distri-

15 To check our procedure for 1817 1827 1877 and 1887we also assembled all declarations that matched on the firstthree letters of last name gender day of death and age theestimates of inequality are slightly higher but trivially so

16 In addition to the TRA survey (which gives a reliablepicture of the national distribution up to the ninety-fifthpercentile) we should mention the study by Daumard(1973) which relied on samples of estate tax returns col-lected in five French cities at the beginning and at the end ofthe nineteenth century (we shall come back to this importantstudy below)

17 See Bourdieu et al (2004) Furthermore the twentieth-century decline in wealth concentration observed in estatetax returns is qualitatively and quantitatively consistent withthe decline in capital income concentration observed inincome tax returns (and the latter appears to be robust inparticular it holds after scaling up tax-return capital incomeusing national accounts aggregates) This is also consistentwith several other data sources on wealth concentration andtop fortunes (especially equity ownership data) See Piketty(2001 2003) for a detailed discussion

242 THE AMERICAN ECONOMIC REVIEW MARCH 2006

bution own no wealth and the richest decileaccounts for at least 95 percent of the value ofall assets during the nineteenth century (seeTable 2) we focus on the top 1 percent Therichest 1 percent of (adult decedents) Parisiansappears to have held a stable and very highfraction of all assets during the first half of thenineteenth century (around 50 to 55 percent oftotal wealth) The 1817 spike was short-livedand was due not to a large increase in the size oftop estates but rather to a large decline inmodest estates (which apparently suffered themost from the Napoleonic Wars)18 Inequalityin Paris increased substantially after 1867 withthe top-1-percent share of wealth at death

climbing from about 52 percent to over 72 per-cent in 1913 World War I and the ensuingshocks then prompted an abrupt decline Thetop-1-percent share dropped by 34 percentagepoints between 1913 and 1947 and by about 10

18 Other spikes in the top-1-percent share are due for themost part to the volatility of the very top estates (thetop-01-percent share and mostly the top-001-percentshare) Note that with about 20000 decedents per year inParis the top-01-percent fractile includes only 20 dece-dents and the top 001 percent only 2 decedents so that theestimates for these fractiles are unstable They depend onthe identity of very wealthy individuals who happened todie in a specific year The figures reported in Table 2 are theraw figures with no adjustment whatsoever for this topwealth volatility Note however that the 1867ndash1913 up-ward trend is highly significant and does not rely on a smallnumber of very top wealth holders

FIGURE 3 WEALTH CONCENTRATION AT DEATH IN PARIS 1807ndash1994

Source Authorsrsquo computations based on estate tax returns (see Table 2)

TABLE 2mdashWEALTH CONCENTRATION AT DEATH IN PARIS1807ndash1994

Top-10-percentestate share

Top-1-percentestate share

Top-01-percentestate share

1807 960 512 1791817 976 573 2281827 973 495 1481837 977 501 1481847 983 558 2131857 969 510 1341867 971 530 1631877 969 589 2461887 971 554 2011902 991 648 2611913 996 721 3281929 949 631 2641938 904 536 2411947 767 381 1481956 750 346 1171994 669 237 65

Source Authorsrsquo computations using estate tax returns (seePiketty et al 2004 Table A2 for detailed series andsources)

243VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

percentage points between 1956 and 1994 Con-verting these wealth-at-death concentration es-timates into wealth-of-the-living concentrationestimates leaves this general picture unchanged(see Section IV below)

Who were the individuals who enjoyed sucha substantial increase in their relative wealthbetween 1867 and 1913 For the most part theirfortunes derived from large industrial and finan-cial estates As Figure 4 illustrates the share ofpersonal (non-real) estate has always been aU-shaped function of wealth This reflects thewell-known fact that real estate is a middle-class asset The poor are too poor to own land orbuildings what little they have is in furniturecash or other moveable items In contrast therich hold most of their wealth in stocks andbonds What is more interesting is that duringthe nineteenth century the relative importanceof personal wealth in Parisian estates also fol-lowed a U-shaped curve over time This wasespecially true for the very wealthy (see Figures4 and 5) where real assets became more andmore important from 1807 to 1837 Real estatethen entered a relative decline after 1837 andaccelerated after 1867

The ebb and flow of the relative importanceof real estate was linked to Parisrsquos recoveryfrom the French Revolution Prior to the Revo-lution the peripheral parts of the city had beena maze of convents monasteries and educa-tional institutions all belonging to the CatholicChurch When the wealth of the Church wasnationalized these real estate assets wereabruptly put on the private market creating aglut of buildings and low prices As the cityrsquospopulation expanded building and land valuesrecovered and the relative importance of realestate grew before being overshadowed by thefinancial boom of the last part of the century(Michel Lescure 1982)

The share of aristocratic decedents among thevery rich follows an inverted-U-shaped curveover the nineteenth century (see Figure6)19 That is nobles became more and morenumerous in top wealth fractiles from 1807 until1847 then the trend reversed and their impor-

19 We take a very broad view of aristocrats they includethe Old Regime nobility the members of the elite who weregiven titles by Napoleon (1801ndash1814) and the Bourbons(1815ndash1830)

FIGURE 4 WEALTH COMPOSITION AT DEATH IN PARIS 1807ndash1902(Share of personal (non-real) estate in total estate)

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives

244 THE AMERICAN ECONOMIC REVIEW MARCH 2006

FIGURE 5 WEALTH COMPOSITION AT DEATH IN PARIS AND FRANCE 1807ndash1902(Share of personal (non-real) estate in total estate)

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives and national aggregateestate statistics compiled by the French tax administration

FIGURE 6 ARISTOCRATIC ESTATES AT DEATH IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives

245VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

tance declined steadily To be sure aristocratsremain overrepresented throughout the periodincluding in 1902 (about 13 percent of nobles inthe top 1 percent of estates over 25 percent inthe top 01 percent versus less than 1 percent inthe population as a whole) The inverted-U pat-tern is yet another of the Revolutionrsquos legaciesIn 1807 when we first observe it aristocraticwealth was at a temporary nadir On the onehand the nobility was impoverished by theRevolutionrsquos inflation and by the sharp declineof the value of Parisian real estate On the otherhand part of the Old Regime nobility was inexile and thus if they died we do not observetheir moveable wealth Aristocrats were able torecoup part of their losses during the first half ofthe nineteenth century Napoleon providedsome assistance by conferring titles of nobilityon his chief military officers and endowingthem with wealth Later the Restoration gov-ernment (1815ndash1830) compensated individualswho fled abroad during the Revolution for thelosses they suffered when their property wasconfiscated The government distributed nearlyone billion francs in the famous milliard desemigres (Andre Gain 1929) The beneficiariesof Napoleonrsquos and the Restorationrsquos largess ap-

pear among the very rich until mid-centuryPresumably such redistribution did not contrib-ute to accelerate French industrialization

III From Paris to France

We can use the Paris data to construct wealthconcentration at death estimates for all ofFrance from 1807 to 1902 To do so we need toknow the evolution of the share of Paris estatesin top estates Between 1902 and 1994 avail-able data (broken down by departement) showsthat the evolution of top estate shares in Francewas parallel to that of top estate shares in ParisWealth inequality is always lower for the coun-try as a whole but the trends are similar (seeFigure 7) It is also striking to note that Parisrsquosshare of the top 1 percent of French estates hasremained fairly stable over the twentieth cen-tury (it fluctuates between 20 percent and 25percent with no trend) even though Parisrsquosshare of all decedents has been dwindling overtime reflecting the population decline of thecapital (see Table 3) In 1902 Paris decedentswere four times more likely to belong to thenational top 1 percent of estates than averagedecedents (26665 41) in 1994 Paris de-

FIGURE 7 WEALTH CONCENTRATION AT DEATH IN PARIS AND FRANCE 1807ndash1994

Source Authorsrsquo computations based on estate tax returns (see Tables 2 and 4)

246 THE AMERICAN ECONOMIC REVIEW MARCH 2006

cedents were seven times more likely to belongto the national top 1 percent of estates thanaverage decedents (25236 70) If anythingthe geographic concentration of fortunes waslarger at the end of the twentieth century than atthe beginning The decline of wealth concentra-tion that followed World War I was not due toredistribution between Paris and the provinces

How did the fraction of Paris estates in topestates evolve over the course of the nineteenthcentury Our estimates rely on a simple andvery conservative assumption from 1807 to1902 Parisrsquos share of estates in the top percen-tile increased at the same rate as Parisrsquos share ofFrench adult deaths More precisely let us de-note FPt(w) the cumulative distribution functionfor wealth-at-death in Paris in year t Ft(w) thecorresponding distribution for France nPt thetotal number of adult deaths in Paris in year tand nt the corresponding number for all ofFrance The ninetieth-percentile threshold P90Ptis defined by FPt(P90Pt) 09 the ninety-ninth-percentile threshold P99Pt is defined byFPt(P99Pt) 099 etc and similarly for theFrench thresholds P90t P99t etc We observeFPt(w) nPt and nt throughout the 1807ndash1994period but we do not observe Ft(w) until 1902(before this date we observe only national ag-

gregate average wealth wt Wtnt) To con-struct our benchmark estimates we assume thatthe shares s99t s995t s999t and s9999t of Parisestates in the national top 1 percent 05 percent01 percent and 001 percent of the nationalwealth-at-death distribution increased at thesame rate as nPtnt during the 1807ndash1902 period(see Table 3) Using this approximation and ourParis samples of individual tax returns we com-pute the threshold wealth levels for the toppercentiles of the national wealth distribution(eg P99 P995 P999 and P9999)20 Wealso calculate the average wealth levels for therelevant wealth classes (eg P99ndash995 P995ndash999 P999ndash9999 and P9999ndash100) usingPareto interpolation techniques These are thenweighted by the number of individuals inFrance in that wealth class in order to compute

20 For instance the number of decedents (aged 20 yearsand older) in France was 583976 in 1887 (see Piketty et al2004 Appendix Table A1) so that the top 1 percent of theestate distribution at death consists of the top 5840 estatesIf the share of Paris among French top-1-percent estates was241 percent in 1887 (see Table 3) then the national P99threshold for 1887 corresponds to the top 1410 Parisianestates (0241 5840 1410) (the national P99 thresholdfor 1887 reported in Piketty et al 2004 Appendix TableA3 was computed using this formula)

TABLE 3mdashTHE FRACTION OF PARIS ESTATES IN TOP ESTATES AT DEATH IN FRANCE 1807ndash1994

(1) Fraction of Parisdecedents in all

decedents 20-yr

(2) Fraction of Parisestates in top-10-

percent estates

(3) Fraction of Parisestates in top-1-percent estates

(4) Fraction of Parisestates in top-01-

percent estates

1807 25 101 2051817 25 103 2101827 28 116 2371837 31 126 2561847 36 146 2971857 36 146 2971867 49 199 4041877 51 211 4281887 59 241 4911902 65 75 266 5411913 65 75 255 5231929 58 83 239 5301938 53 74 216 4211947 55 110 198 3521956 55 128 223 3501994 36 89 252 352

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources) No datasource exists to compute columns (3)ndash(4) prior to 1902 and the numbers reported on this table for years 1807ndash1887 werecomputed assuming that the columns (3)ndash(4) followed the same trend as column (1) over the 1807ndash1902 period (see textSection III)

247VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

the average wealth levels for top fractiles (P99ndash100 P995ndash100 P999ndash100 and P9999ndash100)Lower thresholds of the national wealth distri-bution (P90 and P95) were computed using thenational TRA survey21 and the P90ndash95 andP95ndash99 intermediate wealth levels were alsocomputed using Pareto interpolation techniques

The national top estate shares estimates re-ported in Table 4 were computed using thismethodology They suggest that wealth concen-tration (as measured by the top-1-percent estateshare) rose throughout the nineteenth century inFrance during both the 1807ndash1867 and 1867ndash1902 periods although less sharply than in Parisduring the latter period (see Figure 7) Theseestimates are conservative in the sense that it isalmost certain that they underestimate the riseof wealth concentration that took place inFrance during the nineteenth century First weknow that the bulk of population growth in Parisduring the nineteenth century was due to theannexation of suburbs in 1860 and to populationgrowth in these peripheral arrondissements Be-

cause the outskirts of the city were poor theannexation added few top estates Thus there islittle doubt that Parisrsquos share of top estates inFrance actually increased less than its share of thetotal population This hypothesis is confirmed bynineteenth-century housing tax tabulations show-ing that the fraction of Paris taxpayers in thenational top 1 percent of taxpayers was substan-tially larger than 10 percent at the beginning of thenineteenth century22 Giving Paris a larger (andmore realistic) share of top estates in 1807 wouldboth reduce the share of wealth of the top 1percent in France at that date and lead to morerapid rise in inequality over time

Next and most importantly other estate sur-veys are consistent with the view that wealth in-equality was growing The important study byAdeline Daumard (1973) which relied on samplesof estate tax returns collected in Paris Lyon Tou-louse Lille and Bordeaux found that wealth con-centration increased in each of these five citiesduring the nineteenth century23 The TRA surveyalthough it is ill-suited for the study of top estatesis also consistent with our view Wealth dispersionwas on the rise in nineteenth-century France ac-cording to the TRA survey both in the sense thatthe fraction of decedents with positive estates de-clined over time (in spite of the sharp increase inthe value of the average estate) and that ratios suchas the P90P50 increased24 We also comparedour benchmark national P99 series extrapolatedfrom our Paris samples and the national P99series computed using the TRA survey We foundthat both series display the same overall upwardtrend in concentration (which is reassuring regard-ing the general validity of our Paris-France extrap-olation technique) except that the growth ofinequality from 1807 to 1902 in the TRA series is

21 See Bourdieu et al (2003) for full details about theTRA survey The P90 and P95 thresholds reported onPiketty et al (2004 Appendix Table A3) were computedusing ten-year moving averages around the target years inorder to make sure that each estimate was based on asufficient number of observations

22 These tabulations were published in the same FinanceMinistry official publications as the estate tabulations Wechose not to use them in our formal computations becausethe tax base of the housing tax (namely the rental value ofthe real estate property where the household lives) is onlyloosely connected to the estate tax base (in particular onecannot rule out the possibility that the housing tax baseoverrepresents Paris-based taxpayers)

23 Unfortunately Daumardrsquos samples are not available inmachine-readable format she has only two or three years ofdata for each city and she did not try to compute homog-enous inequality indicators (top fractiles shares etc) withher data Thus although her results and our work are con-sistent they cannot be compared directly

24 See Bourdieu et al (2004)

TABLE 4mdashWEALTH CONCENTRATION AT DEATH IN FRANCE1807ndash1994

Top-10-percentestate share

Top-1-percentestate share

Top-01-percentestate share

1807 791 434 1631817 810 445 1811827 824 452 1631837 796 438 1471847 816 479 1841857 829 495 1741867 810 480 1741877 838 471 2011887 839 487 1921902 839 516 2311913 863 549 2601929 820 502 2471938 776 420 1991947 699 299 1101956 694 304 1101994 610 213 63

Source Authorsrsquo computations using estate tax returns (seePiketty et al 2004 Table A3 for detailed series)

248 THE AMERICAN ECONOMIC REVIEW MARCH 2006

more severe than in our series (see Figure 8) Thisagain suggests that the latter provide a conserva-tive lower bound for the upward trend in wealthconcentration In any case the finding of a largeincrease in wealth inequality in nineteenth-centuryFrance (and up until World War I) appears to berobust25

As was mentioned earlier there exists nocomparable continuous data source covering thenineteenth and twentieth centuries in othercountries which makes it difficult to put ourFrench long-run series in international perspec-tive We note however that existing series forthe United States and the United Kingdom areconsistent with our findings on France26 Re-garding levels existing evidence suggests thatduring the nineteenth and most of the twentiethcenturies France was in an intermediate posi-tion in terms of wealth concentration in be-tween the United States (more equal) and theUnited Kingdom (the most unequal)27 These

25 Note that this continuous rise in wealth inequalitydoes not necessarily imply that a parallel rise occurredregarding income inequality Given that there exists nomicro source on incomes prior to the creation of the incometax in 1914 it is very difficult (if not impossible) to properlyaddress this issue Christian Morrisson and Wayne Snyder(2000) have attempted to link income inequality estimatesbased upon Old-Regime fiscal sources (pre-1789) withmodern income-tax-based twentieth-century estimates andthey have argued that income inequality might have startedto decline during the later part of the nineteenth century andon the eve of World War I (see also Morrisson 2000)Although our data do not allow us to rule out such apossibility we note that their nineteenth-century personaldistribution estimates are based on fragile macroeconomicdata on functional distribution and are not homogenous totheir eighteenth- and twentieth-century estimates Given theevidence that we provide on wealth inequality any signif-icant decline in aggregate income inequality would have tobe associated with severe compression of the wage distri-bution There is little research on this issue however

26 Lee Soltow and Jan L Van Zanden (1998) also find adecline in inequality in the twentieth century in the Neth-erlands Their data are consistent with a rise in inequality inthe nineteenth century but they have no direct evidenceabout its extent

27 According to our series the top-1-percent wealthshare in France rose from around 45 percent in 1800 toabout 55 percent around World War I and then fell to about20 percent by the end of the twentieth century (see Figure7) wealth concentration among the living appears to besomewhat larger (see Section V below) According to theseries pieced together by Lindert (2000 pp 181ndash82 and186) the UK top-1-percent wealth share rose from about

FIGURE 8 ESTIMATES OF THE P99 THRESHOLD FOR THE FRENCH DISTRIBUTION OF ESTATES AT DEATH EXTRAPOLATION

FROM PARIS SAMPLES VERSUS ESTIMATES FROM TRA SAMPLES (CURRENT FRENCH FRANCS)

Source Authorsrsquo computations based on estate tax returns (Paris samples and TRA samples)

249VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

differences in inequality largely hark back to dif-ferential concentration of landownership En-glandrsquos land was extremely narrowly held whilethe United States was most egalitarian The impactof access to real estate assets can also be seenwhen we contrast Paris (where it was extremelyconcentrated) with the provinces

Trends in inequality over time may be easierto compare because biases in source materialmay have a significant effect on levels ratherthan trends The historical pattern is similar ineach of the three economies In particular thereis evidence that wealth concentration increasedduring the nineteenth century in both Anglo-Saxon countries and declined during the twen-tieth century with a turning point around WorldWar I28 Neither exhibits patterns consistentwith a Kuznet process

IV From the Wealth of the Decedents to theWealth of the Living

The estimates reported thus far refer to in-equality among decedents as described in thetax returns filed by their heirs The evolution ofthe distribution of wealth among the livingmight however have followed a different pat-tern In order to convert wealth-at-death con-centration estimates into wealth-of-the-livingconcentration estimates it is standard to use theldquoestate multiplierrdquo method29 It consists ofweighting each observation of an estate at death

by the inverse of the mortality rate for this agegroup That is if the mortality rate for ages 20to 24 was 068 percent in Paris in 1902 theneach decedent aged 20 to 24 represented about147 living individuals of the same age (100068 147) Similarly if the mortality ratefor ages above 80 was 2143 percent in Paris in1902 then each decedent in that group repre-sented about 47 living individuals in the sameage group (102143 47) Applying thismethod requires mortality tables (these are eas-ily available) and estate tabulations brokendown by estate size and age at death (these arescarcer) Fortunately the cityrsquos statistical bu-reau published annual death-by-age totals theFrench censuses report the age distribution forthe capital every five years and we collectedage at death from the estate declarations Thesedata allowed us to compute estimates of wealthconcentration among the living over the 1807ndash1902 period using various assumptions aboutthe wealth profiles of mortality rates

The base population for the living is the set ofall individuals aged 20 and over living in Parisin year t which we denote pt The number ofliving individuals in age bracket a is denoted pta(a 20ndash24 25ndash29 30ndash34 75ndash79 80 andover) and the number of decedents in agebracket a is denoted nta The mortality rate forage bracket a is given by mta ntapta Webegin with a uniform-mortality benchmarkThese estimates are based on the simplifyingassumption that these mortality rates dependsolely on age and are the same for all wealthgroups (and in particular are the same forzero-wealth and positive-wealth individuals)We can then weigh each decedent with positiveestate and age a collected in the Paris archivesin year t by ptanta This allows us to computethe number of living Parisians with positivewealth in year t and also (by differentiatingwith pt) the number of living Parisians with zerowealth at year t which is used to weight zero-estate observations We then use our weighteddatasets to compute top estate fractiles amongthe living in Paris

The main conclusion is that the living expe-rienced the same upward trend in wealth con-centration as the decedents (see Figure 9) Wefind that inequality was significantly higheramong the living than among decedents be-cause survivors were on average younger than

55 percent in 1800 to 70 percent around World War I thenfell to about 20 percent in the 1990s The US top-1-percentwealth share rose from about 15 to 20 percent in 1800 toabout 40 percent around World War I then fell to about 30percent in the 1990s (and as low as 20 to 25 percentaccording to the more recent estimates from Kopczuk andSaez 2004) Wealth concentration is now larger in theUnited States than in European countries but the reversewas true during the nineteenth century up until World WarII (It is only since the 1950sndash1970s period that US wealthconcentration has been somewhat larger)

28 See Lindert (2000 pp 181ndash82 and 188)29 This method was widely used in England and France

in the late nineteenth and early twentieth centuries to com-pute the stock of national wealth on the basis of the flow ofwealth transmitted at death Standard references that use thistechnique to estimate the wealth distribution of the livingfrom estate tax data tabulated by estate size and age at deathinclude Atkinson and Harrison (1978) and Lampman(1962) For a more recent application of this technique tothe United States see Kopczuk and Saez (2004)

250 THE AMERICAN ECONOMIC REVIEW MARCH 2006

those who died and the young were on aver-age poorer In particular the estate multipliermethod leads to lower average weights forpositive-wealth decedents than for zero-wealthdecedents (the former are on average older andtherefore ldquorepresentrdquo a smaller number of livingindividuals) As a result the fraction of positive-wealth individuals is even smaller among theliving than among decedents Hence our bench-mark uniform-mortality estimates of wealthconcentration among the living are significantlylarger than corresponding estimates among de-cedents (eg top-1-percent wealth shares areabout 15 percent larger among the living)Changes over time however are similar In-creased life expectancy and declining mortalityrates over the course of the nineteenth centuryhave only a small effect on the trends

In order to make the estate multiplier methodmore reliable one would prefer to take intoaccount differential mortality by wealth Doingso would require having access to mortalityschedules based both on wealth and age at dif-ferent points in time unfortunately these are notavailable We have nonetheless reestimatedwealth of the living based upon the same as-sumption as Kopczuk and Saez (2004) That is

we assumed uniform mortality among the poor(here defined as zero-wealth individuals) andamong the rich (here defined as positive-wealthindividuals) and we assumed that the ratiomtaRmtaP between the mortality rate of the richand the mortality rate of the poor followed aU-shaped age profile from about 85 percent forthe young (ie the rich die 15 percent less oftenthan the poor when they are 20ndash24 or 25ndash29years old) down to about 70 percent for middle-age individuals in their forties to fifties and upto 100 percent for very old individuals in theireighties to nineties30 This profile correspondsto the best available estimates in the literatureand it appears to be relatively stable over timeand across developed countries In the absenceof better data it is the best one can do31 Thebenchmark differential-mortality estimates re-ported on Figure 9 show that although addingdifferential mortality produces different levelsof inequality it does not have much impact onthe upward trend in concentration

30 See Kopczuk and Saez (2004 Table A4)31 See Kopczuk and Saez (2004 Appendix B) for refer-

ences to the US and international literature devoted to theage-wealth profile of mortality rates

FIGURE 9 WEALTH CONCENTRATION AMONG DECEDENTS AND AMONG THE LIVING IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives (see Piketty et al 2004Table A4 for detailed series)

251VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

The resulting differential-mortality inequalityestimates lie between those based on decedentsonly and those using uniform mortality for theliving (see Figure 9) Note that moving fromuniform-mortality to differential-mortality es-tate multiplier methodology can either increaseor decrease inequality Here the reason whysuch a move leads to lower wealth concentra-tion seems to be due to the fact that differentialmortality tends to put higher weights onpositive-wealth decedents (for a given age)thereby increasing the estimated fraction of liv-ing individuals with positive wealth The im-portant point however is that the resulting leveleffects are relatively small in magnitude con-stant in time and dwarfed by the upward timetrend Even if we were to assume an enormousincrease in differential mortality during thenineteenth century in the sense that differentialmortality between the rich and poor was equalto 0 percent of the benchmark differential in1807 and 100 percent of the benchmark differ-ential in 1902 the resulting wealth concentra-tion estimates would still be significantly higherin 1902 than in 1807 (see Figure 9) Yet wehave no reason to believe that differential mor-tality increased to such an extent During thenineteenth century real wages for unskilledworkers rose which would have reduced mor-tality more for the poor than for the rich After1850 public health measures (sanitation watervaccination) were in place Again these wouldhave had a significant effect on the poor and themiddle class who could not privately purchasesuch health-improving services To be sure therich could avail themselves of more medicalservices than the poor or the middle class butthe impact of these services was probably small(bear in mind that neither antibiotics nor car-diovascular interventions were available)

Finally we have applied the estate multipliermethod to available data for 1947 and 1994 Over-all the sharp decline in wealth concentration ob-served during the twentieth century (andespecially between 1914 and 1945) is very robustIf anything the decline appears to be even largerwhen one looks at wealth concentration amongthe living rather than among decedents32

V The Changing Age Profile of Wealth

In the previous sections we focused almostexclusively on aggregate top wealth shares Ourdata however also detail the characteristics ofeach decedent in particular their gender andage The evolution of wealth by gender is ofrelevance for over the past two centuries therehave been massive changes in womenrsquos laborforce participation capacity to manage theirown affairs and life expectancy relative to menThe evolution of wealth by age is of relevancebecause there was a significant increase in adultlife expectancy over the twentieth century andbecause the progressive diffusion of pensionsmay have changed savings motivations More-over age-wealth profiles also inform us aboutthe motives of wealth accumulation and theeconomic impact of high wealth concentration

A first pass at the data considers the genderbreakdown of wealth at death Remarkably inour micro data the share of women in top estatestakes its highest value on the eve of World WarI For instance the womenrsquos share in the top 05percent rose from 35 percent prior to 1850 to 45percent in 1902 only to fall to 40 percent afterWorld War II Strikingly womenrsquos share ofwealth follows almost exactly the pattern ofaggregate inequality Women were relativelyricher when inequality reached its apex inFrance than at any other time Moreover insti-tutional variables seem to have played almostno role in changing the relative wealth ofwomen Unlike in common law countriesFrench law starting with the code civil of 1804required nearly equal treatment of all children inbequests Further research will help us deter-mine to what extent women of great wealthwere heirs or part of economically very success-ful couples

The data also reveal striking changes in the

32 See Piketty et al (2004 Table A4) It is unfortunatelynot possible to construct complete series for wealth concen-

tration among the living for the twentieth century due todata limitations tables broken down by estate brackets andage of decedents are available solely for years 1943ndash1954and at the national level (no table broken down by estatebrackets and age of decedents has ever been compiled at thedepartement level except in 1931 for Seine departementsee E S Danysz 1934) and the 1994 micro sample is notlarge enough to allow for a reliable application of the estatemultiplier method at the Paris level Thus the only wealth-of-the-living concentration estimates we provide for thetwentieth century are national estimates for 1947 and 1994

252 THE AMERICAN ECONOMIC REVIEW MARCH 2006

age profile of wealth between 1807 and 1994(see Table 5) During the nineteenth century aswealth concentration was increasing the veryrich were getting older and older At the begin-ning of the nineteenth century in the aftermathof the French Revolution the richest individualswere those in their fifties they were typically100 percent richer on average than people intheir forties 25 percent richer than those in theirsixties and 40 percent richer than those in theirseventies and eighties By the 1870s howeverthe age-wealth pattern had become stronglymonotonic the richest individuals were the old-est individuals In 1902 people in their sixtiesand seventies bequeathed 150 percent morethan those in their fifties and those in theireighties 300 percent more On the eve of WorldWar I top wealth holders were old and likely tobe retired This pattern breaks some time duringthe 1914ndash1945 period33 In 1947 as well as in1994 we are back to a pattern where the richestindividuals are those in their fifties Overall theperiod of maximal wealth inequality (1860ndash1913) also appears to be a period characterizedby a very specific age profile of wealth and largeconcentration of assets among the elderly

Another way to analyze the changing age-wealth relationship is to look at average age by

top estate fractile34 In 1817 average age wasvirtually the same for the top 10 percent and thetop 1 percent of estates (or even slightly declin-ing) The average-age-per-fractile relationshipbecomes upward sloping during the nineteenthcentury and by 1902 those in the top 1 percentwere almost six years older than those in the top10 percent The relationship is flat in 1947 anddownward-sloping in 1994 Finally one canapply the estate multiplier method (see SectionIV above) and analyze how wealth concentra-tion by age group among the living changedover the course of the nineteenth century Thegeneral population in Paris did not becomeolder during the nineteenth century those aged60 or older made up about 15 percent of thepopulation in 1817 and after 1847 about 10 to11 percent35 The share of total wealth ownedby the elderly rose significantly however aswealth distribution worsened The wealth be-longing to those aged 60 or more rose fromabout 25 to 30 percent of the total at the begin-ning of the nineteenth century to about 40 to 45percent by the end of the century The wealthshare of those aged 70 or older doubled fromless than 10 percent to about 20 percent36

33 Existing evidence on the age-wealth profile for 1931(see Danysz 1934) suggests that the Great Depression andWorld War II (rather than World War I) played the leadingroles in breaking this pattern This is an issue we plan toinvestigate in future research

34 See Piketty et al (2004 Table 6)35 Although life expectancy was increasing which

should have led to large shares of population for oldergroups the city was also growing quickly The large num-ber of immigrants (who were typically in their twenties)increased the relative size of the younger cohorts (seePiketty et al 2004 Table 7 and Figure 11)

36 See Piketty et al (2004 Figure 12)

TABLE 5mdashTHE AGE PROFILE OF WEALTH AT DEATH IN PARIS 1817ndash1994(Average estate left by 50- 59-year-old 100)

20ndash29yr-old

30ndash39yr-old

40ndash49yr-old

50ndash59yr-old

60ndash69yr-old

70ndash79yr-old

80ndash89yr-old

90ndash99yr-old

1817 26 22 28 100 54 59 591827 44 50 53 100 88 87 601837 133 90 107 100 116 123 1101847 87 73 102 100 117 204 1321857 84 77 101 100 104 109 1451867 67 58 136 100 141 125 1541877 66 73 63 100 197 260 4301887 45 33 63 100 152 233 2951902 29 40 80 100 253 272 4011947 31 51 73 100 113 105 105 1091994 11 45 100 87 93 95 68

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources)

253VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

It is perhaps not surprising that inequalitybecame strongly correlated with age in the1860s Those who died at an old age in the1820s and 1830s had lived through the difficultyears of the French Revolution and the disloca-tion of the end of the Napoleonic period Theirability to accumulate wealth had been severelyimpaired Furthermore they were rather lesslikely to inherit much wealth since the Revo-lution wiped out the bond portfolios of theirparents through a prolonged period of high in-flation Those who died from the 1860s to theearly 1910s did not suffer from the adversitiesthat plagued their forebears Instead they en-joyed the fruits of the financial sector expansionthat began in the 1850s After 1947 we seem tohave returned to a situation quite like that of1817 Presumably the capital damages associ-ated with both world wars and the Great De-pression had a strong negative effect on thewealth holdings of older generations The per-sistence of a flat age-wealth profile until 1994 islikely to be associated with two factors First insocieties where income growth is rapid abso-lute wealth accumulation is faster by youngercohorts than by older ones because their in-comes are higher at every age This is an im-portant distinction between the nineteenthcentury and the twentieth century Furthermorehighly progressive rates of income and estatetaxation have probably made it more difficult toaccumulate large fortunes thereby flattening theobserved age-wealth profile

The more interesting (and more difficult)question relates to the possible efficiency im-pact of high wealth concentration and changingage-wealth profiles Although our data do notallow us to address efficiency issues in a rigor-ous way our results allow us to formulate anumber of hypotheses and to shed new light onthe ongoing debate on inequality and growth37

From a theoretical viewpoint whether highwealth concentration can have a negativegrowth impact depends critically on the exis-tence of credit constraints With first-best credit

markets money flows toward the best entrepre-neurs and investment projects irrespective ofthe initial distribution High levels of wealthconcentration can be bad from a social justiceviewpoint but they entail no efficiency lossWhen credit constraints bind however initialwealth matters and high levels of inequality canhurt growth Whether the loss is large or smalldepends on who owns the assets If the rich areefficient investors (they know which projects tofund etc) then wealth concentration may evenbe useful If the rich are retired rentiers how-ever investing their wealth in low-yield assets(or low-ability inheritors) then high wealthconcentration and credit constraints might pre-vent talented but penniless investors from un-dertaking efficient projects thereby entailingnegative growth consequences The data used inthis paper are not ideal to address whether creditconstraints were important in a country likeFrance at the end of the nineteenth century Ourresults suggest however that to the extentcredit constraints were indeed severe highwealth concentration did have a negativegrowth impact In order to investigate this hy-pothesis further one would need to gather moresystematic data on investment strategies andasset returns Preliminary evidence suggeststhat the rich elderly of the 1860ndash1913 perioddid indeed hold a disproportionate fraction oftheir wealth in low-yield assets (such as gov-ernment bonds) An alternative hypothesishowever is that steeper age-wealth profileswere the consequence of the growth of financialmarkets as their children faced fewer creditconstraints parents decided to hold on to moreof their wealth

VI Conclusion

Evidence from wealth at death in Paris and inFrance over the last two centuries reveals threekey patterns First wealth concentration haschanged dramatically over time In 1807 thetop-1-percent share of wealth (40 percent inFrance 50 percent in Paris) was twice as high asit would be in 1994 but substantially less thanin 1913 when it peaked above 55 percent inFrance and 70 percent in Paris Some of thesechanges were due to economic phenomena thathave long been emphasized as creating inequal-ity namely industrialization and financial cen-

37 Thus far this literature has concentrated upon cross-country regressions of inequality on growth a methodologythat raises serious identification problems especially giventhe low quality of available international datasets on in-equality which are neither long-run nor homogeneous (seeeg Atkinson and Andrea Brandolini 2001)

254 THE AMERICAN ECONOMIC REVIEW MARCH 2006

tralization Yet the decline comes largely fromadverse shocks rather than economic conver-gence These changes are of such magnitudethat they are not sensitive to whether one ex-amines wealth at death in Paris or in France orwhether one examines it directly rather thanconverting it to wealth of the living by an estatemultiplier method

Our second key result is that there was asignificant transition during the nineteenth cen-tury from an important role for real estate as aform of wealth to moveable assets as the keyform of wealth for the very rich Similarly theshare of wealth held by aristocrats first rose andthen was eclipsed by that of financiers and in-dustrialists in the second half of the nineteenthcentury Hence mobility within this highly un-equal society might have been quite high Yetthis conjecture is tempered by our third findingthe wealthy were getting older over time andolder relative to less wealthy decedents Suchaging among the very wealthy would have hadnegative consequences for growth if financialmarkets were imperfect This issue requires fur-ther investigation and we hope it will attractfuture research

REFERENCES

Atkinson Anthony B ldquoTop Incomes in the UKover the 20th Centuryrdquo Journal of the RoyalStatistical Society Series A (Statistics in So-ciety) 2005 168(2) pp 325ndash43

Atkinson Anthony B and Brandolini AndrealdquoPromise and Pitfalls in the Use of lsquoSecond-aryrsquo Data-Sets Income Inequality in OECDCountries as a Case Studyrdquo Journal of Eco-nomic Literature 2001 39(3) pp 771ndash99

Atkinson Anthony B and Harrison Alan J Dis-tribution of personal wealth in Britain Cam-bridge Cambridge University Press 1978

Atkinson Anthony B and Piketty Thomas Topincomes over the twentieth century A con-trast between European and English speak-ing countries Oxford Oxford UniversityPress (forthcoming)

Banerjee Abhijit V and Duflo Esther ldquoInequalityand Growth What Can the Data Sayrdquo Journalof Economic Growth 2003 8(3) pp 267ndash99

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoPourquoi la richesse ne

srsquoest-elle pas diffusee avec la croissance Ledegre zero de lrsquoinegalite et son evolution enFrance 1800ndash1940rdquo Histoire et mesure 200318(1ndash2) pp 147ndash98

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoDefense et illustration delrsquoenquete des 3000 familles Lrsquoexemple deson volet patrimonialrdquo Annales de demogra-phie historique 2004 19 pp 19ndash57

Danysz E S ldquoContribution a lrsquoetude des for-tunes privees drsquoapres les declarations de suc-cessionsrdquo Bulletin de la statistique generalede France 1934 XXIV (1) pp 111ndash171

Daumard Adeline Les fortunes francaise auXIXe siecle Enquete sur la repartition et lacomposition des capitaux prives a ParisLyon Lille Bordeaux et Toulouse drsquoapreslrsquoenregistrement des declarations de succes-sion Paris Mouton 1973

Gain Andre La restauration et les biens desemigres 2 vols Nancy Societe drsquoimpressiontypographique 1929

Jones Alice H American colonial wealth Doc-uments and methods 3 vols New York ArnoPress 1977

Kopczuk Wojciech and Saez Emmanuel ldquoTopWealth Shares in the United States 1916ndash2000 Evidence from Estate Tax ReturnsrdquoNational Tax Journal 2004 57(2) pp 445ndash87

Kuznets Simon ldquoEconomic Growth and IncomeInequalityrdquo American Economic Review1955 45(1) pp 1ndash28

Lampman Robert J The share of top wealth-holders in national wealth 1922ndash1956Princeton Princeton University Press 1962

Lescure Michel Les banques lrsquoEtat et lemarche immobilier en France a lrsquoepoquecontemporaine 1820ndash1940 Paris EditionslrsquoEcole des Hautes Etudes en Sciences Socia-les 1982

Levy-Leboyer Maurice and Bourguignon Fran-cois LrsquoEconomie francaise au XIXe siecleAnalyse macroeconomique Paris Eco-nomica 1985 English translation Cam-bridge Cambridge University Press 1990

Lindert Peter H ldquoUnequal English Wealthsince 1670rdquo Journal of Political Economy1986 94(6) pp 1127ndash62

Lindert Peter H ldquoThree Centuries of Inequalityin Britain and Americardquo in Anthony B At-kinson and Francois Bourguignon eds

255VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

concentration both in France and in Paris duringthe twentieth century using standard Pareto in-terpolation techniques

Prior to 1902 the tax administration pub-lished only the aggregate amount of wealth re-ported on estate tax returns broken down byreal (land and buildings) and personal (furni-ture businesses stocks bonds etc) assets12

Studying concentration thus required collectingour own samples of individual returns Collect-ing information on every individual return fromevery departement for a given year was impos-sible It would have required going to the ar-chives of each departement acquiring access tothe registers of each tax bureau (the lowest-level tax jurisdiction) and dealing with hun-dreds of thousands of declarations each yearWe therefore had to devise a sampling strategyOne option was to select randomly (eg on thebasis of birth dates or family names) a nation-ally representative sample of decedents for var-ious years during the nineteenth century Thatsample would need to be extremely large how-ever to include enough large estates (Given

that wealth is extremely concentrated it is crit-ical to observe many of the very wealthy)

Therefore we decided to pursue a completelydifferent strategy and collected data for all de-cedents in Paris for selected years (1807 18171827 1837 1847 1857 1867 1877 1887 and1902) We chose Paris because a disproportion-ate share of the very rich lived there As one cansee from Table 1 around 1810 the annual num-ber of decedents (20 years old and over) in Pariswas about 12000 (25 percent of the Frenchtotal) that figure nearly tripled during the nine-teenth century to about 35000 by 1900 (65percent of the French total) However only 30percent of decedents in Paris had an estate dur-ing the nineteenth century (about half as manyas in the rest of France) so we needed only tocollect detailed information on 3500 decedentsor so per year at the beginning of the nineteenthcentury and 10000 or so decedents per year atthe end (see Table 1) Although Paris had moredecedents with zero wealth than the rest of thecountry the average estate was about 45 timeslarger in Paris than elsewhere in France duringthe nineteenth century13 It is particularly strik-ing to notice that this ratio actually increased

1943ndash1954 The 1994 micro-file also allows us to breakdown the data by departement and age

12 These published aggregates were computed by theadministration on the basis of tax receipts

13 Average estates as well as top estate fractiles arealways defined in this paper over the set of all decedentsaged 20 and older including those with zero wealth

TABLE 1mdashESTATE TAX RETURNS IN PARIS 1807ndash1994mdashSUMMARY STATISTICS

N decedents20-yr N estate 0

N estate 0(percent

N deced 20)

N deced 20-yr(percent

ParisFrance)

Total estate(percent

ParisFrance)

Average estate(Ratio Parisrest

of France)

1807 11622 3647 314 25 82 3561817 11925 3287 276 25 84 3561827 14151 3877 274 28 94 3561837 16902 4922 291 31 98 3421847 18169 4814 265 33 115 3861857 19248 6048 314 36 143 4511867 26844 7370 275 46 168 4161877 28777 8245 287 51 186 4221887 34411 9815 285 59 201 4011902 36366 9830 270 65 260 5051913 35677 11927 334 65 266 5231929 35842 14495 404 58 250 5421938 30274 16013 529 53 173 3761947 24955 14090 565 55 150 3071956 27940 16053 575 55 159 3241994 18553 12528 675 36 97 286

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed series and sources)

240 THE AMERICAN ECONOMIC REVIEW MARCH 2006

over time in spite of the fact that Paris nearlytripled in population14 On the eve of WorldWar I the estates of Paris decedents made upover 26 percent of the French total (see Table1 and Figure 2)

We designed our data collection to take ad-vantage of the work of the estate tax adminis-tration For every person who either died inParis or might have taxable wealth in any of oneof Parisrsquos nine bureaux the administrationopened an entry in a first set of volumes (thetables des successions et absences henceforthTSA) Later the entry was completed eitherwhen estate taxes were paid or when the admin-istration became satisfied that the individual hadleft no wealth behind The entries include nameoccupation residence marital status age andfor individuals with wealth information aboutheirs and the date at which the declaration was

filed Up to 1870 the TSA also include a sum-mary of the individualrsquos estate broken downinto personal wealth and real estate Hence thecross sections up to 1867 rely heavily on theTSA After 1870 the administration no longerrecorded wealth information in the TSA butonly whether returns had been filed for theindividual For 1877 1887 and 1902 westarted with the TSA and for each individual forwhom a return had been filed we collected thefirst three letters of the last name gender ageday and month of death and the date(s) atwhich returns had been filed We then opened asecond set of registers (the registres de muta-tions par deces henceforth RMD) where a com-plete description of the estates is transcribedand the information not gleaned in the TSA wasappended to the first set of entries

Yet these data gave information by tax returnnot by individual A decedentrsquos heirs could filemultiple returns either because they amendedtheir original declaration or before 1902 be-cause they paid taxes in multiple bureaux In-deed prior to 1902 estate taxes on real estatewere paid in the bureau of the asset rather thanthat of the residence of the decedent In an eraof strictly proportional taxation such dispersed

14 Note that there is a discontinuity in the growth of Parisduring the nineteenth century as new districts (arrondisse-ments) previously registered in the suburb were integratedinto the City of Paris in 1860 The results reported here donot make any correction for this discontinuity which ex-plains the discontinuity observed in some of the figuresaround 1860

FIGURE 2 THE PARIS SHARE IN FRENCH ESTATES AT DEATH 1807ndash1994

Source Authorsrsquo computations based on estate tax returns (see Table 1)

241VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

payment of taxes reduced administrative costsbecause information about real estate values didnot have to be centralized Naturally when theestate tax became progressive in 1902 returnshad to be unified Because TSA entries providelinks to the different declarations in the samebureau reassembling these declarations waseasy reassembling returns from different bu-reaux was another matter because there are nolinks across bureaux To reassemble fully allindividual returns would have required us tocollect detailed information on every decedentin Paris But we did not need to do so becausethe very high levels of inequality in Paris cameto our assistance By collecting nominal infor-mation on the top 10 percent of returns we wereable to attribute successfully 92 percent of mov-able assets and 97 percent of all real assets tospecific individuals15 Given the high variety offirst names and last names as well as detailedinformation on residence the likelihood offalsely positive matches is very low The re-maining returns were treated as individualsthus biasing downward our inequality estimates

Our 1902 Paris sample is fully consistentwith the table compiled by the tax administra-tion for the same year for Paris Therefore wecan link up our 1807ndash1902 Paris files with the1902ndash1994 Paris tables to construct homoge-neous 1807ndash1994 series for inequality in ParisThe more difficult task is to use the Paris data toinfer changes in wealth concentration for Francefrom 1807 to 1902 To do so we must estimatehow the relative importance of Paris in each topestate class evolved over the nineteenth centuryTo achieve this goal we used other estate sur-veys16 as well as nonestate fiscal sources (seeSection III below) The other difficult part is theconstruction of estimates for wealth concentra-tion among the living from estate tax datawhich we do using the estate multiplier method

and mortality data by age group (see Section IVbelow)

When using tax data it is also important tokeep in mind that tax evasion and manipulationcan potentially bias the results There are how-ever good reasons to believe that this is not toomuch of a problem here First estate tax rateswere extremely modest until World War I (lessthan 2 percent) which implies that the incen-tives for tax evasion were small In contrastpenalties for evasion were stiff Moreover theadministration made every effort to keep upwith changing composition of assets and totrack down individuals with some wealthAmong other things financial institutions andpublic utilities were required to notify the ad-ministration when accounts changed owners Asa result it was not easy to dissimulate thewealth of a decedent (either real estate or finan-cial assets in a publicly traded firm) and inher-itors had a strong incentive to register theirproperty in order to benefit from state protec-tion This suggests that the nineteenth-centurydata collected in the Paris archives is probably ofvery high quality Tax evasion is potentially amore serious issue for the twentieth century whentax rates become substantial Although top estatetax rates have rarely exceeded 20 to 30 percent fordirect transmissions in France (the top rate hasbeen 40 percent since 1984 its highest level ever)it is obvious that incentives for tax evasion haveincreased over time However several indepen-dent data sources suggest that the trends observedduring the twentieth century are robust and are notdue to the rise of tax evasion17

II Wealth Concentration at Death in Paris1807ndash1994

Figure 3 shows the evolution of wealth con-centration at death in Paris from 1807 to 1994Given that the bottom two thirds of the distri-

15 To check our procedure for 1817 1827 1877 and 1887we also assembled all declarations that matched on the firstthree letters of last name gender day of death and age theestimates of inequality are slightly higher but trivially so

16 In addition to the TRA survey (which gives a reliablepicture of the national distribution up to the ninety-fifthpercentile) we should mention the study by Daumard(1973) which relied on samples of estate tax returns col-lected in five French cities at the beginning and at the end ofthe nineteenth century (we shall come back to this importantstudy below)

17 See Bourdieu et al (2004) Furthermore the twentieth-century decline in wealth concentration observed in estatetax returns is qualitatively and quantitatively consistent withthe decline in capital income concentration observed inincome tax returns (and the latter appears to be robust inparticular it holds after scaling up tax-return capital incomeusing national accounts aggregates) This is also consistentwith several other data sources on wealth concentration andtop fortunes (especially equity ownership data) See Piketty(2001 2003) for a detailed discussion

242 THE AMERICAN ECONOMIC REVIEW MARCH 2006

bution own no wealth and the richest decileaccounts for at least 95 percent of the value ofall assets during the nineteenth century (seeTable 2) we focus on the top 1 percent Therichest 1 percent of (adult decedents) Parisiansappears to have held a stable and very highfraction of all assets during the first half of thenineteenth century (around 50 to 55 percent oftotal wealth) The 1817 spike was short-livedand was due not to a large increase in the size oftop estates but rather to a large decline inmodest estates (which apparently suffered themost from the Napoleonic Wars)18 Inequalityin Paris increased substantially after 1867 withthe top-1-percent share of wealth at death

climbing from about 52 percent to over 72 per-cent in 1913 World War I and the ensuingshocks then prompted an abrupt decline Thetop-1-percent share dropped by 34 percentagepoints between 1913 and 1947 and by about 10

18 Other spikes in the top-1-percent share are due for themost part to the volatility of the very top estates (thetop-01-percent share and mostly the top-001-percentshare) Note that with about 20000 decedents per year inParis the top-01-percent fractile includes only 20 dece-dents and the top 001 percent only 2 decedents so that theestimates for these fractiles are unstable They depend onthe identity of very wealthy individuals who happened todie in a specific year The figures reported in Table 2 are theraw figures with no adjustment whatsoever for this topwealth volatility Note however that the 1867ndash1913 up-ward trend is highly significant and does not rely on a smallnumber of very top wealth holders

FIGURE 3 WEALTH CONCENTRATION AT DEATH IN PARIS 1807ndash1994

Source Authorsrsquo computations based on estate tax returns (see Table 2)

TABLE 2mdashWEALTH CONCENTRATION AT DEATH IN PARIS1807ndash1994

Top-10-percentestate share

Top-1-percentestate share

Top-01-percentestate share

1807 960 512 1791817 976 573 2281827 973 495 1481837 977 501 1481847 983 558 2131857 969 510 1341867 971 530 1631877 969 589 2461887 971 554 2011902 991 648 2611913 996 721 3281929 949 631 2641938 904 536 2411947 767 381 1481956 750 346 1171994 669 237 65

Source Authorsrsquo computations using estate tax returns (seePiketty et al 2004 Table A2 for detailed series andsources)

243VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

percentage points between 1956 and 1994 Con-verting these wealth-at-death concentration es-timates into wealth-of-the-living concentrationestimates leaves this general picture unchanged(see Section IV below)

Who were the individuals who enjoyed sucha substantial increase in their relative wealthbetween 1867 and 1913 For the most part theirfortunes derived from large industrial and finan-cial estates As Figure 4 illustrates the share ofpersonal (non-real) estate has always been aU-shaped function of wealth This reflects thewell-known fact that real estate is a middle-class asset The poor are too poor to own land orbuildings what little they have is in furniturecash or other moveable items In contrast therich hold most of their wealth in stocks andbonds What is more interesting is that duringthe nineteenth century the relative importanceof personal wealth in Parisian estates also fol-lowed a U-shaped curve over time This wasespecially true for the very wealthy (see Figures4 and 5) where real assets became more andmore important from 1807 to 1837 Real estatethen entered a relative decline after 1837 andaccelerated after 1867

The ebb and flow of the relative importanceof real estate was linked to Parisrsquos recoveryfrom the French Revolution Prior to the Revo-lution the peripheral parts of the city had beena maze of convents monasteries and educa-tional institutions all belonging to the CatholicChurch When the wealth of the Church wasnationalized these real estate assets wereabruptly put on the private market creating aglut of buildings and low prices As the cityrsquospopulation expanded building and land valuesrecovered and the relative importance of realestate grew before being overshadowed by thefinancial boom of the last part of the century(Michel Lescure 1982)

The share of aristocratic decedents among thevery rich follows an inverted-U-shaped curveover the nineteenth century (see Figure6)19 That is nobles became more and morenumerous in top wealth fractiles from 1807 until1847 then the trend reversed and their impor-

19 We take a very broad view of aristocrats they includethe Old Regime nobility the members of the elite who weregiven titles by Napoleon (1801ndash1814) and the Bourbons(1815ndash1830)

FIGURE 4 WEALTH COMPOSITION AT DEATH IN PARIS 1807ndash1902(Share of personal (non-real) estate in total estate)

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives

244 THE AMERICAN ECONOMIC REVIEW MARCH 2006

FIGURE 5 WEALTH COMPOSITION AT DEATH IN PARIS AND FRANCE 1807ndash1902(Share of personal (non-real) estate in total estate)

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives and national aggregateestate statistics compiled by the French tax administration

FIGURE 6 ARISTOCRATIC ESTATES AT DEATH IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives

245VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

tance declined steadily To be sure aristocratsremain overrepresented throughout the periodincluding in 1902 (about 13 percent of nobles inthe top 1 percent of estates over 25 percent inthe top 01 percent versus less than 1 percent inthe population as a whole) The inverted-U pat-tern is yet another of the Revolutionrsquos legaciesIn 1807 when we first observe it aristocraticwealth was at a temporary nadir On the onehand the nobility was impoverished by theRevolutionrsquos inflation and by the sharp declineof the value of Parisian real estate On the otherhand part of the Old Regime nobility was inexile and thus if they died we do not observetheir moveable wealth Aristocrats were able torecoup part of their losses during the first half ofthe nineteenth century Napoleon providedsome assistance by conferring titles of nobilityon his chief military officers and endowingthem with wealth Later the Restoration gov-ernment (1815ndash1830) compensated individualswho fled abroad during the Revolution for thelosses they suffered when their property wasconfiscated The government distributed nearlyone billion francs in the famous milliard desemigres (Andre Gain 1929) The beneficiariesof Napoleonrsquos and the Restorationrsquos largess ap-

pear among the very rich until mid-centuryPresumably such redistribution did not contrib-ute to accelerate French industrialization

III From Paris to France

We can use the Paris data to construct wealthconcentration at death estimates for all ofFrance from 1807 to 1902 To do so we need toknow the evolution of the share of Paris estatesin top estates Between 1902 and 1994 avail-able data (broken down by departement) showsthat the evolution of top estate shares in Francewas parallel to that of top estate shares in ParisWealth inequality is always lower for the coun-try as a whole but the trends are similar (seeFigure 7) It is also striking to note that Parisrsquosshare of the top 1 percent of French estates hasremained fairly stable over the twentieth cen-tury (it fluctuates between 20 percent and 25percent with no trend) even though Parisrsquosshare of all decedents has been dwindling overtime reflecting the population decline of thecapital (see Table 3) In 1902 Paris decedentswere four times more likely to belong to thenational top 1 percent of estates than averagedecedents (26665 41) in 1994 Paris de-

FIGURE 7 WEALTH CONCENTRATION AT DEATH IN PARIS AND FRANCE 1807ndash1994

Source Authorsrsquo computations based on estate tax returns (see Tables 2 and 4)

246 THE AMERICAN ECONOMIC REVIEW MARCH 2006

cedents were seven times more likely to belongto the national top 1 percent of estates thanaverage decedents (25236 70) If anythingthe geographic concentration of fortunes waslarger at the end of the twentieth century than atthe beginning The decline of wealth concentra-tion that followed World War I was not due toredistribution between Paris and the provinces

How did the fraction of Paris estates in topestates evolve over the course of the nineteenthcentury Our estimates rely on a simple andvery conservative assumption from 1807 to1902 Parisrsquos share of estates in the top percen-tile increased at the same rate as Parisrsquos share ofFrench adult deaths More precisely let us de-note FPt(w) the cumulative distribution functionfor wealth-at-death in Paris in year t Ft(w) thecorresponding distribution for France nPt thetotal number of adult deaths in Paris in year tand nt the corresponding number for all ofFrance The ninetieth-percentile threshold P90Ptis defined by FPt(P90Pt) 09 the ninety-ninth-percentile threshold P99Pt is defined byFPt(P99Pt) 099 etc and similarly for theFrench thresholds P90t P99t etc We observeFPt(w) nPt and nt throughout the 1807ndash1994period but we do not observe Ft(w) until 1902(before this date we observe only national ag-

gregate average wealth wt Wtnt) To con-struct our benchmark estimates we assume thatthe shares s99t s995t s999t and s9999t of Parisestates in the national top 1 percent 05 percent01 percent and 001 percent of the nationalwealth-at-death distribution increased at thesame rate as nPtnt during the 1807ndash1902 period(see Table 3) Using this approximation and ourParis samples of individual tax returns we com-pute the threshold wealth levels for the toppercentiles of the national wealth distribution(eg P99 P995 P999 and P9999)20 Wealso calculate the average wealth levels for therelevant wealth classes (eg P99ndash995 P995ndash999 P999ndash9999 and P9999ndash100) usingPareto interpolation techniques These are thenweighted by the number of individuals inFrance in that wealth class in order to compute

20 For instance the number of decedents (aged 20 yearsand older) in France was 583976 in 1887 (see Piketty et al2004 Appendix Table A1) so that the top 1 percent of theestate distribution at death consists of the top 5840 estatesIf the share of Paris among French top-1-percent estates was241 percent in 1887 (see Table 3) then the national P99threshold for 1887 corresponds to the top 1410 Parisianestates (0241 5840 1410) (the national P99 thresholdfor 1887 reported in Piketty et al 2004 Appendix TableA3 was computed using this formula)

TABLE 3mdashTHE FRACTION OF PARIS ESTATES IN TOP ESTATES AT DEATH IN FRANCE 1807ndash1994

(1) Fraction of Parisdecedents in all

decedents 20-yr

(2) Fraction of Parisestates in top-10-

percent estates

(3) Fraction of Parisestates in top-1-percent estates

(4) Fraction of Parisestates in top-01-

percent estates

1807 25 101 2051817 25 103 2101827 28 116 2371837 31 126 2561847 36 146 2971857 36 146 2971867 49 199 4041877 51 211 4281887 59 241 4911902 65 75 266 5411913 65 75 255 5231929 58 83 239 5301938 53 74 216 4211947 55 110 198 3521956 55 128 223 3501994 36 89 252 352

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources) No datasource exists to compute columns (3)ndash(4) prior to 1902 and the numbers reported on this table for years 1807ndash1887 werecomputed assuming that the columns (3)ndash(4) followed the same trend as column (1) over the 1807ndash1902 period (see textSection III)

247VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

the average wealth levels for top fractiles (P99ndash100 P995ndash100 P999ndash100 and P9999ndash100)Lower thresholds of the national wealth distri-bution (P90 and P95) were computed using thenational TRA survey21 and the P90ndash95 andP95ndash99 intermediate wealth levels were alsocomputed using Pareto interpolation techniques

The national top estate shares estimates re-ported in Table 4 were computed using thismethodology They suggest that wealth concen-tration (as measured by the top-1-percent estateshare) rose throughout the nineteenth century inFrance during both the 1807ndash1867 and 1867ndash1902 periods although less sharply than in Parisduring the latter period (see Figure 7) Theseestimates are conservative in the sense that it isalmost certain that they underestimate the riseof wealth concentration that took place inFrance during the nineteenth century First weknow that the bulk of population growth in Parisduring the nineteenth century was due to theannexation of suburbs in 1860 and to populationgrowth in these peripheral arrondissements Be-

cause the outskirts of the city were poor theannexation added few top estates Thus there islittle doubt that Parisrsquos share of top estates inFrance actually increased less than its share of thetotal population This hypothesis is confirmed bynineteenth-century housing tax tabulations show-ing that the fraction of Paris taxpayers in thenational top 1 percent of taxpayers was substan-tially larger than 10 percent at the beginning of thenineteenth century22 Giving Paris a larger (andmore realistic) share of top estates in 1807 wouldboth reduce the share of wealth of the top 1percent in France at that date and lead to morerapid rise in inequality over time

Next and most importantly other estate sur-veys are consistent with the view that wealth in-equality was growing The important study byAdeline Daumard (1973) which relied on samplesof estate tax returns collected in Paris Lyon Tou-louse Lille and Bordeaux found that wealth con-centration increased in each of these five citiesduring the nineteenth century23 The TRA surveyalthough it is ill-suited for the study of top estatesis also consistent with our view Wealth dispersionwas on the rise in nineteenth-century France ac-cording to the TRA survey both in the sense thatthe fraction of decedents with positive estates de-clined over time (in spite of the sharp increase inthe value of the average estate) and that ratios suchas the P90P50 increased24 We also comparedour benchmark national P99 series extrapolatedfrom our Paris samples and the national P99series computed using the TRA survey We foundthat both series display the same overall upwardtrend in concentration (which is reassuring regard-ing the general validity of our Paris-France extrap-olation technique) except that the growth ofinequality from 1807 to 1902 in the TRA series is

21 See Bourdieu et al (2003) for full details about theTRA survey The P90 and P95 thresholds reported onPiketty et al (2004 Appendix Table A3) were computedusing ten-year moving averages around the target years inorder to make sure that each estimate was based on asufficient number of observations

22 These tabulations were published in the same FinanceMinistry official publications as the estate tabulations Wechose not to use them in our formal computations becausethe tax base of the housing tax (namely the rental value ofthe real estate property where the household lives) is onlyloosely connected to the estate tax base (in particular onecannot rule out the possibility that the housing tax baseoverrepresents Paris-based taxpayers)

23 Unfortunately Daumardrsquos samples are not available inmachine-readable format she has only two or three years ofdata for each city and she did not try to compute homog-enous inequality indicators (top fractiles shares etc) withher data Thus although her results and our work are con-sistent they cannot be compared directly

24 See Bourdieu et al (2004)

TABLE 4mdashWEALTH CONCENTRATION AT DEATH IN FRANCE1807ndash1994

Top-10-percentestate share

Top-1-percentestate share

Top-01-percentestate share

1807 791 434 1631817 810 445 1811827 824 452 1631837 796 438 1471847 816 479 1841857 829 495 1741867 810 480 1741877 838 471 2011887 839 487 1921902 839 516 2311913 863 549 2601929 820 502 2471938 776 420 1991947 699 299 1101956 694 304 1101994 610 213 63

Source Authorsrsquo computations using estate tax returns (seePiketty et al 2004 Table A3 for detailed series)

248 THE AMERICAN ECONOMIC REVIEW MARCH 2006

more severe than in our series (see Figure 8) Thisagain suggests that the latter provide a conserva-tive lower bound for the upward trend in wealthconcentration In any case the finding of a largeincrease in wealth inequality in nineteenth-centuryFrance (and up until World War I) appears to berobust25

As was mentioned earlier there exists nocomparable continuous data source covering thenineteenth and twentieth centuries in othercountries which makes it difficult to put ourFrench long-run series in international perspec-tive We note however that existing series forthe United States and the United Kingdom areconsistent with our findings on France26 Re-garding levels existing evidence suggests thatduring the nineteenth and most of the twentiethcenturies France was in an intermediate posi-tion in terms of wealth concentration in be-tween the United States (more equal) and theUnited Kingdom (the most unequal)27 These

25 Note that this continuous rise in wealth inequalitydoes not necessarily imply that a parallel rise occurredregarding income inequality Given that there exists nomicro source on incomes prior to the creation of the incometax in 1914 it is very difficult (if not impossible) to properlyaddress this issue Christian Morrisson and Wayne Snyder(2000) have attempted to link income inequality estimatesbased upon Old-Regime fiscal sources (pre-1789) withmodern income-tax-based twentieth-century estimates andthey have argued that income inequality might have startedto decline during the later part of the nineteenth century andon the eve of World War I (see also Morrisson 2000)Although our data do not allow us to rule out such apossibility we note that their nineteenth-century personaldistribution estimates are based on fragile macroeconomicdata on functional distribution and are not homogenous totheir eighteenth- and twentieth-century estimates Given theevidence that we provide on wealth inequality any signif-icant decline in aggregate income inequality would have tobe associated with severe compression of the wage distri-bution There is little research on this issue however

26 Lee Soltow and Jan L Van Zanden (1998) also find adecline in inequality in the twentieth century in the Neth-erlands Their data are consistent with a rise in inequality inthe nineteenth century but they have no direct evidenceabout its extent

27 According to our series the top-1-percent wealthshare in France rose from around 45 percent in 1800 toabout 55 percent around World War I and then fell to about20 percent by the end of the twentieth century (see Figure7) wealth concentration among the living appears to besomewhat larger (see Section V below) According to theseries pieced together by Lindert (2000 pp 181ndash82 and186) the UK top-1-percent wealth share rose from about

FIGURE 8 ESTIMATES OF THE P99 THRESHOLD FOR THE FRENCH DISTRIBUTION OF ESTATES AT DEATH EXTRAPOLATION

FROM PARIS SAMPLES VERSUS ESTIMATES FROM TRA SAMPLES (CURRENT FRENCH FRANCS)

Source Authorsrsquo computations based on estate tax returns (Paris samples and TRA samples)

249VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

differences in inequality largely hark back to dif-ferential concentration of landownership En-glandrsquos land was extremely narrowly held whilethe United States was most egalitarian The impactof access to real estate assets can also be seenwhen we contrast Paris (where it was extremelyconcentrated) with the provinces

Trends in inequality over time may be easierto compare because biases in source materialmay have a significant effect on levels ratherthan trends The historical pattern is similar ineach of the three economies In particular thereis evidence that wealth concentration increasedduring the nineteenth century in both Anglo-Saxon countries and declined during the twen-tieth century with a turning point around WorldWar I28 Neither exhibits patterns consistentwith a Kuznet process

IV From the Wealth of the Decedents to theWealth of the Living

The estimates reported thus far refer to in-equality among decedents as described in thetax returns filed by their heirs The evolution ofthe distribution of wealth among the livingmight however have followed a different pat-tern In order to convert wealth-at-death con-centration estimates into wealth-of-the-livingconcentration estimates it is standard to use theldquoestate multiplierrdquo method29 It consists ofweighting each observation of an estate at death

by the inverse of the mortality rate for this agegroup That is if the mortality rate for ages 20to 24 was 068 percent in Paris in 1902 theneach decedent aged 20 to 24 represented about147 living individuals of the same age (100068 147) Similarly if the mortality ratefor ages above 80 was 2143 percent in Paris in1902 then each decedent in that group repre-sented about 47 living individuals in the sameage group (102143 47) Applying thismethod requires mortality tables (these are eas-ily available) and estate tabulations brokendown by estate size and age at death (these arescarcer) Fortunately the cityrsquos statistical bu-reau published annual death-by-age totals theFrench censuses report the age distribution forthe capital every five years and we collectedage at death from the estate declarations Thesedata allowed us to compute estimates of wealthconcentration among the living over the 1807ndash1902 period using various assumptions aboutthe wealth profiles of mortality rates

The base population for the living is the set ofall individuals aged 20 and over living in Parisin year t which we denote pt The number ofliving individuals in age bracket a is denoted pta(a 20ndash24 25ndash29 30ndash34 75ndash79 80 andover) and the number of decedents in agebracket a is denoted nta The mortality rate forage bracket a is given by mta ntapta Webegin with a uniform-mortality benchmarkThese estimates are based on the simplifyingassumption that these mortality rates dependsolely on age and are the same for all wealthgroups (and in particular are the same forzero-wealth and positive-wealth individuals)We can then weigh each decedent with positiveestate and age a collected in the Paris archivesin year t by ptanta This allows us to computethe number of living Parisians with positivewealth in year t and also (by differentiatingwith pt) the number of living Parisians with zerowealth at year t which is used to weight zero-estate observations We then use our weighteddatasets to compute top estate fractiles amongthe living in Paris

The main conclusion is that the living expe-rienced the same upward trend in wealth con-centration as the decedents (see Figure 9) Wefind that inequality was significantly higheramong the living than among decedents be-cause survivors were on average younger than

55 percent in 1800 to 70 percent around World War I thenfell to about 20 percent in the 1990s The US top-1-percentwealth share rose from about 15 to 20 percent in 1800 toabout 40 percent around World War I then fell to about 30percent in the 1990s (and as low as 20 to 25 percentaccording to the more recent estimates from Kopczuk andSaez 2004) Wealth concentration is now larger in theUnited States than in European countries but the reversewas true during the nineteenth century up until World WarII (It is only since the 1950sndash1970s period that US wealthconcentration has been somewhat larger)

28 See Lindert (2000 pp 181ndash82 and 188)29 This method was widely used in England and France

in the late nineteenth and early twentieth centuries to com-pute the stock of national wealth on the basis of the flow ofwealth transmitted at death Standard references that use thistechnique to estimate the wealth distribution of the livingfrom estate tax data tabulated by estate size and age at deathinclude Atkinson and Harrison (1978) and Lampman(1962) For a more recent application of this technique tothe United States see Kopczuk and Saez (2004)

250 THE AMERICAN ECONOMIC REVIEW MARCH 2006

those who died and the young were on aver-age poorer In particular the estate multipliermethod leads to lower average weights forpositive-wealth decedents than for zero-wealthdecedents (the former are on average older andtherefore ldquorepresentrdquo a smaller number of livingindividuals) As a result the fraction of positive-wealth individuals is even smaller among theliving than among decedents Hence our bench-mark uniform-mortality estimates of wealthconcentration among the living are significantlylarger than corresponding estimates among de-cedents (eg top-1-percent wealth shares areabout 15 percent larger among the living)Changes over time however are similar In-creased life expectancy and declining mortalityrates over the course of the nineteenth centuryhave only a small effect on the trends

In order to make the estate multiplier methodmore reliable one would prefer to take intoaccount differential mortality by wealth Doingso would require having access to mortalityschedules based both on wealth and age at dif-ferent points in time unfortunately these are notavailable We have nonetheless reestimatedwealth of the living based upon the same as-sumption as Kopczuk and Saez (2004) That is

we assumed uniform mortality among the poor(here defined as zero-wealth individuals) andamong the rich (here defined as positive-wealthindividuals) and we assumed that the ratiomtaRmtaP between the mortality rate of the richand the mortality rate of the poor followed aU-shaped age profile from about 85 percent forthe young (ie the rich die 15 percent less oftenthan the poor when they are 20ndash24 or 25ndash29years old) down to about 70 percent for middle-age individuals in their forties to fifties and upto 100 percent for very old individuals in theireighties to nineties30 This profile correspondsto the best available estimates in the literatureand it appears to be relatively stable over timeand across developed countries In the absenceof better data it is the best one can do31 Thebenchmark differential-mortality estimates re-ported on Figure 9 show that although addingdifferential mortality produces different levelsof inequality it does not have much impact onthe upward trend in concentration

30 See Kopczuk and Saez (2004 Table A4)31 See Kopczuk and Saez (2004 Appendix B) for refer-

ences to the US and international literature devoted to theage-wealth profile of mortality rates

FIGURE 9 WEALTH CONCENTRATION AMONG DECEDENTS AND AMONG THE LIVING IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives (see Piketty et al 2004Table A4 for detailed series)

251VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

The resulting differential-mortality inequalityestimates lie between those based on decedentsonly and those using uniform mortality for theliving (see Figure 9) Note that moving fromuniform-mortality to differential-mortality es-tate multiplier methodology can either increaseor decrease inequality Here the reason whysuch a move leads to lower wealth concentra-tion seems to be due to the fact that differentialmortality tends to put higher weights onpositive-wealth decedents (for a given age)thereby increasing the estimated fraction of liv-ing individuals with positive wealth The im-portant point however is that the resulting leveleffects are relatively small in magnitude con-stant in time and dwarfed by the upward timetrend Even if we were to assume an enormousincrease in differential mortality during thenineteenth century in the sense that differentialmortality between the rich and poor was equalto 0 percent of the benchmark differential in1807 and 100 percent of the benchmark differ-ential in 1902 the resulting wealth concentra-tion estimates would still be significantly higherin 1902 than in 1807 (see Figure 9) Yet wehave no reason to believe that differential mor-tality increased to such an extent During thenineteenth century real wages for unskilledworkers rose which would have reduced mor-tality more for the poor than for the rich After1850 public health measures (sanitation watervaccination) were in place Again these wouldhave had a significant effect on the poor and themiddle class who could not privately purchasesuch health-improving services To be sure therich could avail themselves of more medicalservices than the poor or the middle class butthe impact of these services was probably small(bear in mind that neither antibiotics nor car-diovascular interventions were available)

Finally we have applied the estate multipliermethod to available data for 1947 and 1994 Over-all the sharp decline in wealth concentration ob-served during the twentieth century (andespecially between 1914 and 1945) is very robustIf anything the decline appears to be even largerwhen one looks at wealth concentration amongthe living rather than among decedents32

V The Changing Age Profile of Wealth

In the previous sections we focused almostexclusively on aggregate top wealth shares Ourdata however also detail the characteristics ofeach decedent in particular their gender andage The evolution of wealth by gender is ofrelevance for over the past two centuries therehave been massive changes in womenrsquos laborforce participation capacity to manage theirown affairs and life expectancy relative to menThe evolution of wealth by age is of relevancebecause there was a significant increase in adultlife expectancy over the twentieth century andbecause the progressive diffusion of pensionsmay have changed savings motivations More-over age-wealth profiles also inform us aboutthe motives of wealth accumulation and theeconomic impact of high wealth concentration

A first pass at the data considers the genderbreakdown of wealth at death Remarkably inour micro data the share of women in top estatestakes its highest value on the eve of World WarI For instance the womenrsquos share in the top 05percent rose from 35 percent prior to 1850 to 45percent in 1902 only to fall to 40 percent afterWorld War II Strikingly womenrsquos share ofwealth follows almost exactly the pattern ofaggregate inequality Women were relativelyricher when inequality reached its apex inFrance than at any other time Moreover insti-tutional variables seem to have played almostno role in changing the relative wealth ofwomen Unlike in common law countriesFrench law starting with the code civil of 1804required nearly equal treatment of all children inbequests Further research will help us deter-mine to what extent women of great wealthwere heirs or part of economically very success-ful couples

The data also reveal striking changes in the

32 See Piketty et al (2004 Table A4) It is unfortunatelynot possible to construct complete series for wealth concen-

tration among the living for the twentieth century due todata limitations tables broken down by estate brackets andage of decedents are available solely for years 1943ndash1954and at the national level (no table broken down by estatebrackets and age of decedents has ever been compiled at thedepartement level except in 1931 for Seine departementsee E S Danysz 1934) and the 1994 micro sample is notlarge enough to allow for a reliable application of the estatemultiplier method at the Paris level Thus the only wealth-of-the-living concentration estimates we provide for thetwentieth century are national estimates for 1947 and 1994

252 THE AMERICAN ECONOMIC REVIEW MARCH 2006

age profile of wealth between 1807 and 1994(see Table 5) During the nineteenth century aswealth concentration was increasing the veryrich were getting older and older At the begin-ning of the nineteenth century in the aftermathof the French Revolution the richest individualswere those in their fifties they were typically100 percent richer on average than people intheir forties 25 percent richer than those in theirsixties and 40 percent richer than those in theirseventies and eighties By the 1870s howeverthe age-wealth pattern had become stronglymonotonic the richest individuals were the old-est individuals In 1902 people in their sixtiesand seventies bequeathed 150 percent morethan those in their fifties and those in theireighties 300 percent more On the eve of WorldWar I top wealth holders were old and likely tobe retired This pattern breaks some time duringthe 1914ndash1945 period33 In 1947 as well as in1994 we are back to a pattern where the richestindividuals are those in their fifties Overall theperiod of maximal wealth inequality (1860ndash1913) also appears to be a period characterizedby a very specific age profile of wealth and largeconcentration of assets among the elderly

Another way to analyze the changing age-wealth relationship is to look at average age by

top estate fractile34 In 1817 average age wasvirtually the same for the top 10 percent and thetop 1 percent of estates (or even slightly declin-ing) The average-age-per-fractile relationshipbecomes upward sloping during the nineteenthcentury and by 1902 those in the top 1 percentwere almost six years older than those in the top10 percent The relationship is flat in 1947 anddownward-sloping in 1994 Finally one canapply the estate multiplier method (see SectionIV above) and analyze how wealth concentra-tion by age group among the living changedover the course of the nineteenth century Thegeneral population in Paris did not becomeolder during the nineteenth century those aged60 or older made up about 15 percent of thepopulation in 1817 and after 1847 about 10 to11 percent35 The share of total wealth ownedby the elderly rose significantly however aswealth distribution worsened The wealth be-longing to those aged 60 or more rose fromabout 25 to 30 percent of the total at the begin-ning of the nineteenth century to about 40 to 45percent by the end of the century The wealthshare of those aged 70 or older doubled fromless than 10 percent to about 20 percent36

33 Existing evidence on the age-wealth profile for 1931(see Danysz 1934) suggests that the Great Depression andWorld War II (rather than World War I) played the leadingroles in breaking this pattern This is an issue we plan toinvestigate in future research

34 See Piketty et al (2004 Table 6)35 Although life expectancy was increasing which

should have led to large shares of population for oldergroups the city was also growing quickly The large num-ber of immigrants (who were typically in their twenties)increased the relative size of the younger cohorts (seePiketty et al 2004 Table 7 and Figure 11)

36 See Piketty et al (2004 Figure 12)

TABLE 5mdashTHE AGE PROFILE OF WEALTH AT DEATH IN PARIS 1817ndash1994(Average estate left by 50- 59-year-old 100)

20ndash29yr-old

30ndash39yr-old

40ndash49yr-old

50ndash59yr-old

60ndash69yr-old

70ndash79yr-old

80ndash89yr-old

90ndash99yr-old

1817 26 22 28 100 54 59 591827 44 50 53 100 88 87 601837 133 90 107 100 116 123 1101847 87 73 102 100 117 204 1321857 84 77 101 100 104 109 1451867 67 58 136 100 141 125 1541877 66 73 63 100 197 260 4301887 45 33 63 100 152 233 2951902 29 40 80 100 253 272 4011947 31 51 73 100 113 105 105 1091994 11 45 100 87 93 95 68

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources)

253VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

It is perhaps not surprising that inequalitybecame strongly correlated with age in the1860s Those who died at an old age in the1820s and 1830s had lived through the difficultyears of the French Revolution and the disloca-tion of the end of the Napoleonic period Theirability to accumulate wealth had been severelyimpaired Furthermore they were rather lesslikely to inherit much wealth since the Revo-lution wiped out the bond portfolios of theirparents through a prolonged period of high in-flation Those who died from the 1860s to theearly 1910s did not suffer from the adversitiesthat plagued their forebears Instead they en-joyed the fruits of the financial sector expansionthat began in the 1850s After 1947 we seem tohave returned to a situation quite like that of1817 Presumably the capital damages associ-ated with both world wars and the Great De-pression had a strong negative effect on thewealth holdings of older generations The per-sistence of a flat age-wealth profile until 1994 islikely to be associated with two factors First insocieties where income growth is rapid abso-lute wealth accumulation is faster by youngercohorts than by older ones because their in-comes are higher at every age This is an im-portant distinction between the nineteenthcentury and the twentieth century Furthermorehighly progressive rates of income and estatetaxation have probably made it more difficult toaccumulate large fortunes thereby flattening theobserved age-wealth profile

The more interesting (and more difficult)question relates to the possible efficiency im-pact of high wealth concentration and changingage-wealth profiles Although our data do notallow us to address efficiency issues in a rigor-ous way our results allow us to formulate anumber of hypotheses and to shed new light onthe ongoing debate on inequality and growth37

From a theoretical viewpoint whether highwealth concentration can have a negativegrowth impact depends critically on the exis-tence of credit constraints With first-best credit

markets money flows toward the best entrepre-neurs and investment projects irrespective ofthe initial distribution High levels of wealthconcentration can be bad from a social justiceviewpoint but they entail no efficiency lossWhen credit constraints bind however initialwealth matters and high levels of inequality canhurt growth Whether the loss is large or smalldepends on who owns the assets If the rich areefficient investors (they know which projects tofund etc) then wealth concentration may evenbe useful If the rich are retired rentiers how-ever investing their wealth in low-yield assets(or low-ability inheritors) then high wealthconcentration and credit constraints might pre-vent talented but penniless investors from un-dertaking efficient projects thereby entailingnegative growth consequences The data used inthis paper are not ideal to address whether creditconstraints were important in a country likeFrance at the end of the nineteenth century Ourresults suggest however that to the extentcredit constraints were indeed severe highwealth concentration did have a negativegrowth impact In order to investigate this hy-pothesis further one would need to gather moresystematic data on investment strategies andasset returns Preliminary evidence suggeststhat the rich elderly of the 1860ndash1913 perioddid indeed hold a disproportionate fraction oftheir wealth in low-yield assets (such as gov-ernment bonds) An alternative hypothesishowever is that steeper age-wealth profileswere the consequence of the growth of financialmarkets as their children faced fewer creditconstraints parents decided to hold on to moreof their wealth

VI Conclusion

Evidence from wealth at death in Paris and inFrance over the last two centuries reveals threekey patterns First wealth concentration haschanged dramatically over time In 1807 thetop-1-percent share of wealth (40 percent inFrance 50 percent in Paris) was twice as high asit would be in 1994 but substantially less thanin 1913 when it peaked above 55 percent inFrance and 70 percent in Paris Some of thesechanges were due to economic phenomena thathave long been emphasized as creating inequal-ity namely industrialization and financial cen-

37 Thus far this literature has concentrated upon cross-country regressions of inequality on growth a methodologythat raises serious identification problems especially giventhe low quality of available international datasets on in-equality which are neither long-run nor homogeneous (seeeg Atkinson and Andrea Brandolini 2001)

254 THE AMERICAN ECONOMIC REVIEW MARCH 2006

tralization Yet the decline comes largely fromadverse shocks rather than economic conver-gence These changes are of such magnitudethat they are not sensitive to whether one ex-amines wealth at death in Paris or in France orwhether one examines it directly rather thanconverting it to wealth of the living by an estatemultiplier method

Our second key result is that there was asignificant transition during the nineteenth cen-tury from an important role for real estate as aform of wealth to moveable assets as the keyform of wealth for the very rich Similarly theshare of wealth held by aristocrats first rose andthen was eclipsed by that of financiers and in-dustrialists in the second half of the nineteenthcentury Hence mobility within this highly un-equal society might have been quite high Yetthis conjecture is tempered by our third findingthe wealthy were getting older over time andolder relative to less wealthy decedents Suchaging among the very wealthy would have hadnegative consequences for growth if financialmarkets were imperfect This issue requires fur-ther investigation and we hope it will attractfuture research

REFERENCES

Atkinson Anthony B ldquoTop Incomes in the UKover the 20th Centuryrdquo Journal of the RoyalStatistical Society Series A (Statistics in So-ciety) 2005 168(2) pp 325ndash43

Atkinson Anthony B and Brandolini AndrealdquoPromise and Pitfalls in the Use of lsquoSecond-aryrsquo Data-Sets Income Inequality in OECDCountries as a Case Studyrdquo Journal of Eco-nomic Literature 2001 39(3) pp 771ndash99

Atkinson Anthony B and Harrison Alan J Dis-tribution of personal wealth in Britain Cam-bridge Cambridge University Press 1978

Atkinson Anthony B and Piketty Thomas Topincomes over the twentieth century A con-trast between European and English speak-ing countries Oxford Oxford UniversityPress (forthcoming)

Banerjee Abhijit V and Duflo Esther ldquoInequalityand Growth What Can the Data Sayrdquo Journalof Economic Growth 2003 8(3) pp 267ndash99

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoPourquoi la richesse ne

srsquoest-elle pas diffusee avec la croissance Ledegre zero de lrsquoinegalite et son evolution enFrance 1800ndash1940rdquo Histoire et mesure 200318(1ndash2) pp 147ndash98

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoDefense et illustration delrsquoenquete des 3000 familles Lrsquoexemple deson volet patrimonialrdquo Annales de demogra-phie historique 2004 19 pp 19ndash57

Danysz E S ldquoContribution a lrsquoetude des for-tunes privees drsquoapres les declarations de suc-cessionsrdquo Bulletin de la statistique generalede France 1934 XXIV (1) pp 111ndash171

Daumard Adeline Les fortunes francaise auXIXe siecle Enquete sur la repartition et lacomposition des capitaux prives a ParisLyon Lille Bordeaux et Toulouse drsquoapreslrsquoenregistrement des declarations de succes-sion Paris Mouton 1973

Gain Andre La restauration et les biens desemigres 2 vols Nancy Societe drsquoimpressiontypographique 1929

Jones Alice H American colonial wealth Doc-uments and methods 3 vols New York ArnoPress 1977

Kopczuk Wojciech and Saez Emmanuel ldquoTopWealth Shares in the United States 1916ndash2000 Evidence from Estate Tax ReturnsrdquoNational Tax Journal 2004 57(2) pp 445ndash87

Kuznets Simon ldquoEconomic Growth and IncomeInequalityrdquo American Economic Review1955 45(1) pp 1ndash28

Lampman Robert J The share of top wealth-holders in national wealth 1922ndash1956Princeton Princeton University Press 1962

Lescure Michel Les banques lrsquoEtat et lemarche immobilier en France a lrsquoepoquecontemporaine 1820ndash1940 Paris EditionslrsquoEcole des Hautes Etudes en Sciences Socia-les 1982

Levy-Leboyer Maurice and Bourguignon Fran-cois LrsquoEconomie francaise au XIXe siecleAnalyse macroeconomique Paris Eco-nomica 1985 English translation Cam-bridge Cambridge University Press 1990

Lindert Peter H ldquoUnequal English Wealthsince 1670rdquo Journal of Political Economy1986 94(6) pp 1127ndash62

Lindert Peter H ldquoThree Centuries of Inequalityin Britain and Americardquo in Anthony B At-kinson and Francois Bourguignon eds

255VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

over time in spite of the fact that Paris nearlytripled in population14 On the eve of WorldWar I the estates of Paris decedents made upover 26 percent of the French total (see Table1 and Figure 2)

We designed our data collection to take ad-vantage of the work of the estate tax adminis-tration For every person who either died inParis or might have taxable wealth in any of oneof Parisrsquos nine bureaux the administrationopened an entry in a first set of volumes (thetables des successions et absences henceforthTSA) Later the entry was completed eitherwhen estate taxes were paid or when the admin-istration became satisfied that the individual hadleft no wealth behind The entries include nameoccupation residence marital status age andfor individuals with wealth information aboutheirs and the date at which the declaration was

filed Up to 1870 the TSA also include a sum-mary of the individualrsquos estate broken downinto personal wealth and real estate Hence thecross sections up to 1867 rely heavily on theTSA After 1870 the administration no longerrecorded wealth information in the TSA butonly whether returns had been filed for theindividual For 1877 1887 and 1902 westarted with the TSA and for each individual forwhom a return had been filed we collected thefirst three letters of the last name gender ageday and month of death and the date(s) atwhich returns had been filed We then opened asecond set of registers (the registres de muta-tions par deces henceforth RMD) where a com-plete description of the estates is transcribedand the information not gleaned in the TSA wasappended to the first set of entries

Yet these data gave information by tax returnnot by individual A decedentrsquos heirs could filemultiple returns either because they amendedtheir original declaration or before 1902 be-cause they paid taxes in multiple bureaux In-deed prior to 1902 estate taxes on real estatewere paid in the bureau of the asset rather thanthat of the residence of the decedent In an eraof strictly proportional taxation such dispersed

14 Note that there is a discontinuity in the growth of Parisduring the nineteenth century as new districts (arrondisse-ments) previously registered in the suburb were integratedinto the City of Paris in 1860 The results reported here donot make any correction for this discontinuity which ex-plains the discontinuity observed in some of the figuresaround 1860

FIGURE 2 THE PARIS SHARE IN FRENCH ESTATES AT DEATH 1807ndash1994

Source Authorsrsquo computations based on estate tax returns (see Table 1)

241VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

payment of taxes reduced administrative costsbecause information about real estate values didnot have to be centralized Naturally when theestate tax became progressive in 1902 returnshad to be unified Because TSA entries providelinks to the different declarations in the samebureau reassembling these declarations waseasy reassembling returns from different bu-reaux was another matter because there are nolinks across bureaux To reassemble fully allindividual returns would have required us tocollect detailed information on every decedentin Paris But we did not need to do so becausethe very high levels of inequality in Paris cameto our assistance By collecting nominal infor-mation on the top 10 percent of returns we wereable to attribute successfully 92 percent of mov-able assets and 97 percent of all real assets tospecific individuals15 Given the high variety offirst names and last names as well as detailedinformation on residence the likelihood offalsely positive matches is very low The re-maining returns were treated as individualsthus biasing downward our inequality estimates

Our 1902 Paris sample is fully consistentwith the table compiled by the tax administra-tion for the same year for Paris Therefore wecan link up our 1807ndash1902 Paris files with the1902ndash1994 Paris tables to construct homoge-neous 1807ndash1994 series for inequality in ParisThe more difficult task is to use the Paris data toinfer changes in wealth concentration for Francefrom 1807 to 1902 To do so we must estimatehow the relative importance of Paris in each topestate class evolved over the nineteenth centuryTo achieve this goal we used other estate sur-veys16 as well as nonestate fiscal sources (seeSection III below) The other difficult part is theconstruction of estimates for wealth concentra-tion among the living from estate tax datawhich we do using the estate multiplier method

and mortality data by age group (see Section IVbelow)

When using tax data it is also important tokeep in mind that tax evasion and manipulationcan potentially bias the results There are how-ever good reasons to believe that this is not toomuch of a problem here First estate tax rateswere extremely modest until World War I (lessthan 2 percent) which implies that the incen-tives for tax evasion were small In contrastpenalties for evasion were stiff Moreover theadministration made every effort to keep upwith changing composition of assets and totrack down individuals with some wealthAmong other things financial institutions andpublic utilities were required to notify the ad-ministration when accounts changed owners Asa result it was not easy to dissimulate thewealth of a decedent (either real estate or finan-cial assets in a publicly traded firm) and inher-itors had a strong incentive to register theirproperty in order to benefit from state protec-tion This suggests that the nineteenth-centurydata collected in the Paris archives is probably ofvery high quality Tax evasion is potentially amore serious issue for the twentieth century whentax rates become substantial Although top estatetax rates have rarely exceeded 20 to 30 percent fordirect transmissions in France (the top rate hasbeen 40 percent since 1984 its highest level ever)it is obvious that incentives for tax evasion haveincreased over time However several indepen-dent data sources suggest that the trends observedduring the twentieth century are robust and are notdue to the rise of tax evasion17

II Wealth Concentration at Death in Paris1807ndash1994

Figure 3 shows the evolution of wealth con-centration at death in Paris from 1807 to 1994Given that the bottom two thirds of the distri-

15 To check our procedure for 1817 1827 1877 and 1887we also assembled all declarations that matched on the firstthree letters of last name gender day of death and age theestimates of inequality are slightly higher but trivially so

16 In addition to the TRA survey (which gives a reliablepicture of the national distribution up to the ninety-fifthpercentile) we should mention the study by Daumard(1973) which relied on samples of estate tax returns col-lected in five French cities at the beginning and at the end ofthe nineteenth century (we shall come back to this importantstudy below)

17 See Bourdieu et al (2004) Furthermore the twentieth-century decline in wealth concentration observed in estatetax returns is qualitatively and quantitatively consistent withthe decline in capital income concentration observed inincome tax returns (and the latter appears to be robust inparticular it holds after scaling up tax-return capital incomeusing national accounts aggregates) This is also consistentwith several other data sources on wealth concentration andtop fortunes (especially equity ownership data) See Piketty(2001 2003) for a detailed discussion

242 THE AMERICAN ECONOMIC REVIEW MARCH 2006

bution own no wealth and the richest decileaccounts for at least 95 percent of the value ofall assets during the nineteenth century (seeTable 2) we focus on the top 1 percent Therichest 1 percent of (adult decedents) Parisiansappears to have held a stable and very highfraction of all assets during the first half of thenineteenth century (around 50 to 55 percent oftotal wealth) The 1817 spike was short-livedand was due not to a large increase in the size oftop estates but rather to a large decline inmodest estates (which apparently suffered themost from the Napoleonic Wars)18 Inequalityin Paris increased substantially after 1867 withthe top-1-percent share of wealth at death

climbing from about 52 percent to over 72 per-cent in 1913 World War I and the ensuingshocks then prompted an abrupt decline Thetop-1-percent share dropped by 34 percentagepoints between 1913 and 1947 and by about 10

18 Other spikes in the top-1-percent share are due for themost part to the volatility of the very top estates (thetop-01-percent share and mostly the top-001-percentshare) Note that with about 20000 decedents per year inParis the top-01-percent fractile includes only 20 dece-dents and the top 001 percent only 2 decedents so that theestimates for these fractiles are unstable They depend onthe identity of very wealthy individuals who happened todie in a specific year The figures reported in Table 2 are theraw figures with no adjustment whatsoever for this topwealth volatility Note however that the 1867ndash1913 up-ward trend is highly significant and does not rely on a smallnumber of very top wealth holders

FIGURE 3 WEALTH CONCENTRATION AT DEATH IN PARIS 1807ndash1994

Source Authorsrsquo computations based on estate tax returns (see Table 2)

TABLE 2mdashWEALTH CONCENTRATION AT DEATH IN PARIS1807ndash1994

Top-10-percentestate share

Top-1-percentestate share

Top-01-percentestate share

1807 960 512 1791817 976 573 2281827 973 495 1481837 977 501 1481847 983 558 2131857 969 510 1341867 971 530 1631877 969 589 2461887 971 554 2011902 991 648 2611913 996 721 3281929 949 631 2641938 904 536 2411947 767 381 1481956 750 346 1171994 669 237 65

Source Authorsrsquo computations using estate tax returns (seePiketty et al 2004 Table A2 for detailed series andsources)

243VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

percentage points between 1956 and 1994 Con-verting these wealth-at-death concentration es-timates into wealth-of-the-living concentrationestimates leaves this general picture unchanged(see Section IV below)

Who were the individuals who enjoyed sucha substantial increase in their relative wealthbetween 1867 and 1913 For the most part theirfortunes derived from large industrial and finan-cial estates As Figure 4 illustrates the share ofpersonal (non-real) estate has always been aU-shaped function of wealth This reflects thewell-known fact that real estate is a middle-class asset The poor are too poor to own land orbuildings what little they have is in furniturecash or other moveable items In contrast therich hold most of their wealth in stocks andbonds What is more interesting is that duringthe nineteenth century the relative importanceof personal wealth in Parisian estates also fol-lowed a U-shaped curve over time This wasespecially true for the very wealthy (see Figures4 and 5) where real assets became more andmore important from 1807 to 1837 Real estatethen entered a relative decline after 1837 andaccelerated after 1867

The ebb and flow of the relative importanceof real estate was linked to Parisrsquos recoveryfrom the French Revolution Prior to the Revo-lution the peripheral parts of the city had beena maze of convents monasteries and educa-tional institutions all belonging to the CatholicChurch When the wealth of the Church wasnationalized these real estate assets wereabruptly put on the private market creating aglut of buildings and low prices As the cityrsquospopulation expanded building and land valuesrecovered and the relative importance of realestate grew before being overshadowed by thefinancial boom of the last part of the century(Michel Lescure 1982)

The share of aristocratic decedents among thevery rich follows an inverted-U-shaped curveover the nineteenth century (see Figure6)19 That is nobles became more and morenumerous in top wealth fractiles from 1807 until1847 then the trend reversed and their impor-

19 We take a very broad view of aristocrats they includethe Old Regime nobility the members of the elite who weregiven titles by Napoleon (1801ndash1814) and the Bourbons(1815ndash1830)

FIGURE 4 WEALTH COMPOSITION AT DEATH IN PARIS 1807ndash1902(Share of personal (non-real) estate in total estate)

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives

244 THE AMERICAN ECONOMIC REVIEW MARCH 2006

FIGURE 5 WEALTH COMPOSITION AT DEATH IN PARIS AND FRANCE 1807ndash1902(Share of personal (non-real) estate in total estate)

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives and national aggregateestate statistics compiled by the French tax administration

FIGURE 6 ARISTOCRATIC ESTATES AT DEATH IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives

245VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

tance declined steadily To be sure aristocratsremain overrepresented throughout the periodincluding in 1902 (about 13 percent of nobles inthe top 1 percent of estates over 25 percent inthe top 01 percent versus less than 1 percent inthe population as a whole) The inverted-U pat-tern is yet another of the Revolutionrsquos legaciesIn 1807 when we first observe it aristocraticwealth was at a temporary nadir On the onehand the nobility was impoverished by theRevolutionrsquos inflation and by the sharp declineof the value of Parisian real estate On the otherhand part of the Old Regime nobility was inexile and thus if they died we do not observetheir moveable wealth Aristocrats were able torecoup part of their losses during the first half ofthe nineteenth century Napoleon providedsome assistance by conferring titles of nobilityon his chief military officers and endowingthem with wealth Later the Restoration gov-ernment (1815ndash1830) compensated individualswho fled abroad during the Revolution for thelosses they suffered when their property wasconfiscated The government distributed nearlyone billion francs in the famous milliard desemigres (Andre Gain 1929) The beneficiariesof Napoleonrsquos and the Restorationrsquos largess ap-

pear among the very rich until mid-centuryPresumably such redistribution did not contrib-ute to accelerate French industrialization

III From Paris to France

We can use the Paris data to construct wealthconcentration at death estimates for all ofFrance from 1807 to 1902 To do so we need toknow the evolution of the share of Paris estatesin top estates Between 1902 and 1994 avail-able data (broken down by departement) showsthat the evolution of top estate shares in Francewas parallel to that of top estate shares in ParisWealth inequality is always lower for the coun-try as a whole but the trends are similar (seeFigure 7) It is also striking to note that Parisrsquosshare of the top 1 percent of French estates hasremained fairly stable over the twentieth cen-tury (it fluctuates between 20 percent and 25percent with no trend) even though Parisrsquosshare of all decedents has been dwindling overtime reflecting the population decline of thecapital (see Table 3) In 1902 Paris decedentswere four times more likely to belong to thenational top 1 percent of estates than averagedecedents (26665 41) in 1994 Paris de-

FIGURE 7 WEALTH CONCENTRATION AT DEATH IN PARIS AND FRANCE 1807ndash1994

Source Authorsrsquo computations based on estate tax returns (see Tables 2 and 4)

246 THE AMERICAN ECONOMIC REVIEW MARCH 2006

cedents were seven times more likely to belongto the national top 1 percent of estates thanaverage decedents (25236 70) If anythingthe geographic concentration of fortunes waslarger at the end of the twentieth century than atthe beginning The decline of wealth concentra-tion that followed World War I was not due toredistribution between Paris and the provinces

How did the fraction of Paris estates in topestates evolve over the course of the nineteenthcentury Our estimates rely on a simple andvery conservative assumption from 1807 to1902 Parisrsquos share of estates in the top percen-tile increased at the same rate as Parisrsquos share ofFrench adult deaths More precisely let us de-note FPt(w) the cumulative distribution functionfor wealth-at-death in Paris in year t Ft(w) thecorresponding distribution for France nPt thetotal number of adult deaths in Paris in year tand nt the corresponding number for all ofFrance The ninetieth-percentile threshold P90Ptis defined by FPt(P90Pt) 09 the ninety-ninth-percentile threshold P99Pt is defined byFPt(P99Pt) 099 etc and similarly for theFrench thresholds P90t P99t etc We observeFPt(w) nPt and nt throughout the 1807ndash1994period but we do not observe Ft(w) until 1902(before this date we observe only national ag-

gregate average wealth wt Wtnt) To con-struct our benchmark estimates we assume thatthe shares s99t s995t s999t and s9999t of Parisestates in the national top 1 percent 05 percent01 percent and 001 percent of the nationalwealth-at-death distribution increased at thesame rate as nPtnt during the 1807ndash1902 period(see Table 3) Using this approximation and ourParis samples of individual tax returns we com-pute the threshold wealth levels for the toppercentiles of the national wealth distribution(eg P99 P995 P999 and P9999)20 Wealso calculate the average wealth levels for therelevant wealth classes (eg P99ndash995 P995ndash999 P999ndash9999 and P9999ndash100) usingPareto interpolation techniques These are thenweighted by the number of individuals inFrance in that wealth class in order to compute

20 For instance the number of decedents (aged 20 yearsand older) in France was 583976 in 1887 (see Piketty et al2004 Appendix Table A1) so that the top 1 percent of theestate distribution at death consists of the top 5840 estatesIf the share of Paris among French top-1-percent estates was241 percent in 1887 (see Table 3) then the national P99threshold for 1887 corresponds to the top 1410 Parisianestates (0241 5840 1410) (the national P99 thresholdfor 1887 reported in Piketty et al 2004 Appendix TableA3 was computed using this formula)

TABLE 3mdashTHE FRACTION OF PARIS ESTATES IN TOP ESTATES AT DEATH IN FRANCE 1807ndash1994

(1) Fraction of Parisdecedents in all

decedents 20-yr

(2) Fraction of Parisestates in top-10-

percent estates

(3) Fraction of Parisestates in top-1-percent estates

(4) Fraction of Parisestates in top-01-

percent estates

1807 25 101 2051817 25 103 2101827 28 116 2371837 31 126 2561847 36 146 2971857 36 146 2971867 49 199 4041877 51 211 4281887 59 241 4911902 65 75 266 5411913 65 75 255 5231929 58 83 239 5301938 53 74 216 4211947 55 110 198 3521956 55 128 223 3501994 36 89 252 352

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources) No datasource exists to compute columns (3)ndash(4) prior to 1902 and the numbers reported on this table for years 1807ndash1887 werecomputed assuming that the columns (3)ndash(4) followed the same trend as column (1) over the 1807ndash1902 period (see textSection III)

247VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

the average wealth levels for top fractiles (P99ndash100 P995ndash100 P999ndash100 and P9999ndash100)Lower thresholds of the national wealth distri-bution (P90 and P95) were computed using thenational TRA survey21 and the P90ndash95 andP95ndash99 intermediate wealth levels were alsocomputed using Pareto interpolation techniques

The national top estate shares estimates re-ported in Table 4 were computed using thismethodology They suggest that wealth concen-tration (as measured by the top-1-percent estateshare) rose throughout the nineteenth century inFrance during both the 1807ndash1867 and 1867ndash1902 periods although less sharply than in Parisduring the latter period (see Figure 7) Theseestimates are conservative in the sense that it isalmost certain that they underestimate the riseof wealth concentration that took place inFrance during the nineteenth century First weknow that the bulk of population growth in Parisduring the nineteenth century was due to theannexation of suburbs in 1860 and to populationgrowth in these peripheral arrondissements Be-

cause the outskirts of the city were poor theannexation added few top estates Thus there islittle doubt that Parisrsquos share of top estates inFrance actually increased less than its share of thetotal population This hypothesis is confirmed bynineteenth-century housing tax tabulations show-ing that the fraction of Paris taxpayers in thenational top 1 percent of taxpayers was substan-tially larger than 10 percent at the beginning of thenineteenth century22 Giving Paris a larger (andmore realistic) share of top estates in 1807 wouldboth reduce the share of wealth of the top 1percent in France at that date and lead to morerapid rise in inequality over time

Next and most importantly other estate sur-veys are consistent with the view that wealth in-equality was growing The important study byAdeline Daumard (1973) which relied on samplesof estate tax returns collected in Paris Lyon Tou-louse Lille and Bordeaux found that wealth con-centration increased in each of these five citiesduring the nineteenth century23 The TRA surveyalthough it is ill-suited for the study of top estatesis also consistent with our view Wealth dispersionwas on the rise in nineteenth-century France ac-cording to the TRA survey both in the sense thatthe fraction of decedents with positive estates de-clined over time (in spite of the sharp increase inthe value of the average estate) and that ratios suchas the P90P50 increased24 We also comparedour benchmark national P99 series extrapolatedfrom our Paris samples and the national P99series computed using the TRA survey We foundthat both series display the same overall upwardtrend in concentration (which is reassuring regard-ing the general validity of our Paris-France extrap-olation technique) except that the growth ofinequality from 1807 to 1902 in the TRA series is

21 See Bourdieu et al (2003) for full details about theTRA survey The P90 and P95 thresholds reported onPiketty et al (2004 Appendix Table A3) were computedusing ten-year moving averages around the target years inorder to make sure that each estimate was based on asufficient number of observations

22 These tabulations were published in the same FinanceMinistry official publications as the estate tabulations Wechose not to use them in our formal computations becausethe tax base of the housing tax (namely the rental value ofthe real estate property where the household lives) is onlyloosely connected to the estate tax base (in particular onecannot rule out the possibility that the housing tax baseoverrepresents Paris-based taxpayers)

23 Unfortunately Daumardrsquos samples are not available inmachine-readable format she has only two or three years ofdata for each city and she did not try to compute homog-enous inequality indicators (top fractiles shares etc) withher data Thus although her results and our work are con-sistent they cannot be compared directly

24 See Bourdieu et al (2004)

TABLE 4mdashWEALTH CONCENTRATION AT DEATH IN FRANCE1807ndash1994

Top-10-percentestate share

Top-1-percentestate share

Top-01-percentestate share

1807 791 434 1631817 810 445 1811827 824 452 1631837 796 438 1471847 816 479 1841857 829 495 1741867 810 480 1741877 838 471 2011887 839 487 1921902 839 516 2311913 863 549 2601929 820 502 2471938 776 420 1991947 699 299 1101956 694 304 1101994 610 213 63

Source Authorsrsquo computations using estate tax returns (seePiketty et al 2004 Table A3 for detailed series)

248 THE AMERICAN ECONOMIC REVIEW MARCH 2006

more severe than in our series (see Figure 8) Thisagain suggests that the latter provide a conserva-tive lower bound for the upward trend in wealthconcentration In any case the finding of a largeincrease in wealth inequality in nineteenth-centuryFrance (and up until World War I) appears to berobust25

As was mentioned earlier there exists nocomparable continuous data source covering thenineteenth and twentieth centuries in othercountries which makes it difficult to put ourFrench long-run series in international perspec-tive We note however that existing series forthe United States and the United Kingdom areconsistent with our findings on France26 Re-garding levels existing evidence suggests thatduring the nineteenth and most of the twentiethcenturies France was in an intermediate posi-tion in terms of wealth concentration in be-tween the United States (more equal) and theUnited Kingdom (the most unequal)27 These

25 Note that this continuous rise in wealth inequalitydoes not necessarily imply that a parallel rise occurredregarding income inequality Given that there exists nomicro source on incomes prior to the creation of the incometax in 1914 it is very difficult (if not impossible) to properlyaddress this issue Christian Morrisson and Wayne Snyder(2000) have attempted to link income inequality estimatesbased upon Old-Regime fiscal sources (pre-1789) withmodern income-tax-based twentieth-century estimates andthey have argued that income inequality might have startedto decline during the later part of the nineteenth century andon the eve of World War I (see also Morrisson 2000)Although our data do not allow us to rule out such apossibility we note that their nineteenth-century personaldistribution estimates are based on fragile macroeconomicdata on functional distribution and are not homogenous totheir eighteenth- and twentieth-century estimates Given theevidence that we provide on wealth inequality any signif-icant decline in aggregate income inequality would have tobe associated with severe compression of the wage distri-bution There is little research on this issue however

26 Lee Soltow and Jan L Van Zanden (1998) also find adecline in inequality in the twentieth century in the Neth-erlands Their data are consistent with a rise in inequality inthe nineteenth century but they have no direct evidenceabout its extent

27 According to our series the top-1-percent wealthshare in France rose from around 45 percent in 1800 toabout 55 percent around World War I and then fell to about20 percent by the end of the twentieth century (see Figure7) wealth concentration among the living appears to besomewhat larger (see Section V below) According to theseries pieced together by Lindert (2000 pp 181ndash82 and186) the UK top-1-percent wealth share rose from about

FIGURE 8 ESTIMATES OF THE P99 THRESHOLD FOR THE FRENCH DISTRIBUTION OF ESTATES AT DEATH EXTRAPOLATION

FROM PARIS SAMPLES VERSUS ESTIMATES FROM TRA SAMPLES (CURRENT FRENCH FRANCS)

Source Authorsrsquo computations based on estate tax returns (Paris samples and TRA samples)

249VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

differences in inequality largely hark back to dif-ferential concentration of landownership En-glandrsquos land was extremely narrowly held whilethe United States was most egalitarian The impactof access to real estate assets can also be seenwhen we contrast Paris (where it was extremelyconcentrated) with the provinces

Trends in inequality over time may be easierto compare because biases in source materialmay have a significant effect on levels ratherthan trends The historical pattern is similar ineach of the three economies In particular thereis evidence that wealth concentration increasedduring the nineteenth century in both Anglo-Saxon countries and declined during the twen-tieth century with a turning point around WorldWar I28 Neither exhibits patterns consistentwith a Kuznet process

IV From the Wealth of the Decedents to theWealth of the Living

The estimates reported thus far refer to in-equality among decedents as described in thetax returns filed by their heirs The evolution ofthe distribution of wealth among the livingmight however have followed a different pat-tern In order to convert wealth-at-death con-centration estimates into wealth-of-the-livingconcentration estimates it is standard to use theldquoestate multiplierrdquo method29 It consists ofweighting each observation of an estate at death

by the inverse of the mortality rate for this agegroup That is if the mortality rate for ages 20to 24 was 068 percent in Paris in 1902 theneach decedent aged 20 to 24 represented about147 living individuals of the same age (100068 147) Similarly if the mortality ratefor ages above 80 was 2143 percent in Paris in1902 then each decedent in that group repre-sented about 47 living individuals in the sameage group (102143 47) Applying thismethod requires mortality tables (these are eas-ily available) and estate tabulations brokendown by estate size and age at death (these arescarcer) Fortunately the cityrsquos statistical bu-reau published annual death-by-age totals theFrench censuses report the age distribution forthe capital every five years and we collectedage at death from the estate declarations Thesedata allowed us to compute estimates of wealthconcentration among the living over the 1807ndash1902 period using various assumptions aboutthe wealth profiles of mortality rates

The base population for the living is the set ofall individuals aged 20 and over living in Parisin year t which we denote pt The number ofliving individuals in age bracket a is denoted pta(a 20ndash24 25ndash29 30ndash34 75ndash79 80 andover) and the number of decedents in agebracket a is denoted nta The mortality rate forage bracket a is given by mta ntapta Webegin with a uniform-mortality benchmarkThese estimates are based on the simplifyingassumption that these mortality rates dependsolely on age and are the same for all wealthgroups (and in particular are the same forzero-wealth and positive-wealth individuals)We can then weigh each decedent with positiveestate and age a collected in the Paris archivesin year t by ptanta This allows us to computethe number of living Parisians with positivewealth in year t and also (by differentiatingwith pt) the number of living Parisians with zerowealth at year t which is used to weight zero-estate observations We then use our weighteddatasets to compute top estate fractiles amongthe living in Paris

The main conclusion is that the living expe-rienced the same upward trend in wealth con-centration as the decedents (see Figure 9) Wefind that inequality was significantly higheramong the living than among decedents be-cause survivors were on average younger than

55 percent in 1800 to 70 percent around World War I thenfell to about 20 percent in the 1990s The US top-1-percentwealth share rose from about 15 to 20 percent in 1800 toabout 40 percent around World War I then fell to about 30percent in the 1990s (and as low as 20 to 25 percentaccording to the more recent estimates from Kopczuk andSaez 2004) Wealth concentration is now larger in theUnited States than in European countries but the reversewas true during the nineteenth century up until World WarII (It is only since the 1950sndash1970s period that US wealthconcentration has been somewhat larger)

28 See Lindert (2000 pp 181ndash82 and 188)29 This method was widely used in England and France

in the late nineteenth and early twentieth centuries to com-pute the stock of national wealth on the basis of the flow ofwealth transmitted at death Standard references that use thistechnique to estimate the wealth distribution of the livingfrom estate tax data tabulated by estate size and age at deathinclude Atkinson and Harrison (1978) and Lampman(1962) For a more recent application of this technique tothe United States see Kopczuk and Saez (2004)

250 THE AMERICAN ECONOMIC REVIEW MARCH 2006

those who died and the young were on aver-age poorer In particular the estate multipliermethod leads to lower average weights forpositive-wealth decedents than for zero-wealthdecedents (the former are on average older andtherefore ldquorepresentrdquo a smaller number of livingindividuals) As a result the fraction of positive-wealth individuals is even smaller among theliving than among decedents Hence our bench-mark uniform-mortality estimates of wealthconcentration among the living are significantlylarger than corresponding estimates among de-cedents (eg top-1-percent wealth shares areabout 15 percent larger among the living)Changes over time however are similar In-creased life expectancy and declining mortalityrates over the course of the nineteenth centuryhave only a small effect on the trends

In order to make the estate multiplier methodmore reliable one would prefer to take intoaccount differential mortality by wealth Doingso would require having access to mortalityschedules based both on wealth and age at dif-ferent points in time unfortunately these are notavailable We have nonetheless reestimatedwealth of the living based upon the same as-sumption as Kopczuk and Saez (2004) That is

we assumed uniform mortality among the poor(here defined as zero-wealth individuals) andamong the rich (here defined as positive-wealthindividuals) and we assumed that the ratiomtaRmtaP between the mortality rate of the richand the mortality rate of the poor followed aU-shaped age profile from about 85 percent forthe young (ie the rich die 15 percent less oftenthan the poor when they are 20ndash24 or 25ndash29years old) down to about 70 percent for middle-age individuals in their forties to fifties and upto 100 percent for very old individuals in theireighties to nineties30 This profile correspondsto the best available estimates in the literatureand it appears to be relatively stable over timeand across developed countries In the absenceof better data it is the best one can do31 Thebenchmark differential-mortality estimates re-ported on Figure 9 show that although addingdifferential mortality produces different levelsof inequality it does not have much impact onthe upward trend in concentration

30 See Kopczuk and Saez (2004 Table A4)31 See Kopczuk and Saez (2004 Appendix B) for refer-

ences to the US and international literature devoted to theage-wealth profile of mortality rates

FIGURE 9 WEALTH CONCENTRATION AMONG DECEDENTS AND AMONG THE LIVING IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives (see Piketty et al 2004Table A4 for detailed series)

251VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

The resulting differential-mortality inequalityestimates lie between those based on decedentsonly and those using uniform mortality for theliving (see Figure 9) Note that moving fromuniform-mortality to differential-mortality es-tate multiplier methodology can either increaseor decrease inequality Here the reason whysuch a move leads to lower wealth concentra-tion seems to be due to the fact that differentialmortality tends to put higher weights onpositive-wealth decedents (for a given age)thereby increasing the estimated fraction of liv-ing individuals with positive wealth The im-portant point however is that the resulting leveleffects are relatively small in magnitude con-stant in time and dwarfed by the upward timetrend Even if we were to assume an enormousincrease in differential mortality during thenineteenth century in the sense that differentialmortality between the rich and poor was equalto 0 percent of the benchmark differential in1807 and 100 percent of the benchmark differ-ential in 1902 the resulting wealth concentra-tion estimates would still be significantly higherin 1902 than in 1807 (see Figure 9) Yet wehave no reason to believe that differential mor-tality increased to such an extent During thenineteenth century real wages for unskilledworkers rose which would have reduced mor-tality more for the poor than for the rich After1850 public health measures (sanitation watervaccination) were in place Again these wouldhave had a significant effect on the poor and themiddle class who could not privately purchasesuch health-improving services To be sure therich could avail themselves of more medicalservices than the poor or the middle class butthe impact of these services was probably small(bear in mind that neither antibiotics nor car-diovascular interventions were available)

Finally we have applied the estate multipliermethod to available data for 1947 and 1994 Over-all the sharp decline in wealth concentration ob-served during the twentieth century (andespecially between 1914 and 1945) is very robustIf anything the decline appears to be even largerwhen one looks at wealth concentration amongthe living rather than among decedents32

V The Changing Age Profile of Wealth

In the previous sections we focused almostexclusively on aggregate top wealth shares Ourdata however also detail the characteristics ofeach decedent in particular their gender andage The evolution of wealth by gender is ofrelevance for over the past two centuries therehave been massive changes in womenrsquos laborforce participation capacity to manage theirown affairs and life expectancy relative to menThe evolution of wealth by age is of relevancebecause there was a significant increase in adultlife expectancy over the twentieth century andbecause the progressive diffusion of pensionsmay have changed savings motivations More-over age-wealth profiles also inform us aboutthe motives of wealth accumulation and theeconomic impact of high wealth concentration

A first pass at the data considers the genderbreakdown of wealth at death Remarkably inour micro data the share of women in top estatestakes its highest value on the eve of World WarI For instance the womenrsquos share in the top 05percent rose from 35 percent prior to 1850 to 45percent in 1902 only to fall to 40 percent afterWorld War II Strikingly womenrsquos share ofwealth follows almost exactly the pattern ofaggregate inequality Women were relativelyricher when inequality reached its apex inFrance than at any other time Moreover insti-tutional variables seem to have played almostno role in changing the relative wealth ofwomen Unlike in common law countriesFrench law starting with the code civil of 1804required nearly equal treatment of all children inbequests Further research will help us deter-mine to what extent women of great wealthwere heirs or part of economically very success-ful couples

The data also reveal striking changes in the

32 See Piketty et al (2004 Table A4) It is unfortunatelynot possible to construct complete series for wealth concen-

tration among the living for the twentieth century due todata limitations tables broken down by estate brackets andage of decedents are available solely for years 1943ndash1954and at the national level (no table broken down by estatebrackets and age of decedents has ever been compiled at thedepartement level except in 1931 for Seine departementsee E S Danysz 1934) and the 1994 micro sample is notlarge enough to allow for a reliable application of the estatemultiplier method at the Paris level Thus the only wealth-of-the-living concentration estimates we provide for thetwentieth century are national estimates for 1947 and 1994

252 THE AMERICAN ECONOMIC REVIEW MARCH 2006

age profile of wealth between 1807 and 1994(see Table 5) During the nineteenth century aswealth concentration was increasing the veryrich were getting older and older At the begin-ning of the nineteenth century in the aftermathof the French Revolution the richest individualswere those in their fifties they were typically100 percent richer on average than people intheir forties 25 percent richer than those in theirsixties and 40 percent richer than those in theirseventies and eighties By the 1870s howeverthe age-wealth pattern had become stronglymonotonic the richest individuals were the old-est individuals In 1902 people in their sixtiesand seventies bequeathed 150 percent morethan those in their fifties and those in theireighties 300 percent more On the eve of WorldWar I top wealth holders were old and likely tobe retired This pattern breaks some time duringthe 1914ndash1945 period33 In 1947 as well as in1994 we are back to a pattern where the richestindividuals are those in their fifties Overall theperiod of maximal wealth inequality (1860ndash1913) also appears to be a period characterizedby a very specific age profile of wealth and largeconcentration of assets among the elderly

Another way to analyze the changing age-wealth relationship is to look at average age by

top estate fractile34 In 1817 average age wasvirtually the same for the top 10 percent and thetop 1 percent of estates (or even slightly declin-ing) The average-age-per-fractile relationshipbecomes upward sloping during the nineteenthcentury and by 1902 those in the top 1 percentwere almost six years older than those in the top10 percent The relationship is flat in 1947 anddownward-sloping in 1994 Finally one canapply the estate multiplier method (see SectionIV above) and analyze how wealth concentra-tion by age group among the living changedover the course of the nineteenth century Thegeneral population in Paris did not becomeolder during the nineteenth century those aged60 or older made up about 15 percent of thepopulation in 1817 and after 1847 about 10 to11 percent35 The share of total wealth ownedby the elderly rose significantly however aswealth distribution worsened The wealth be-longing to those aged 60 or more rose fromabout 25 to 30 percent of the total at the begin-ning of the nineteenth century to about 40 to 45percent by the end of the century The wealthshare of those aged 70 or older doubled fromless than 10 percent to about 20 percent36

33 Existing evidence on the age-wealth profile for 1931(see Danysz 1934) suggests that the Great Depression andWorld War II (rather than World War I) played the leadingroles in breaking this pattern This is an issue we plan toinvestigate in future research

34 See Piketty et al (2004 Table 6)35 Although life expectancy was increasing which

should have led to large shares of population for oldergroups the city was also growing quickly The large num-ber of immigrants (who were typically in their twenties)increased the relative size of the younger cohorts (seePiketty et al 2004 Table 7 and Figure 11)

36 See Piketty et al (2004 Figure 12)

TABLE 5mdashTHE AGE PROFILE OF WEALTH AT DEATH IN PARIS 1817ndash1994(Average estate left by 50- 59-year-old 100)

20ndash29yr-old

30ndash39yr-old

40ndash49yr-old

50ndash59yr-old

60ndash69yr-old

70ndash79yr-old

80ndash89yr-old

90ndash99yr-old

1817 26 22 28 100 54 59 591827 44 50 53 100 88 87 601837 133 90 107 100 116 123 1101847 87 73 102 100 117 204 1321857 84 77 101 100 104 109 1451867 67 58 136 100 141 125 1541877 66 73 63 100 197 260 4301887 45 33 63 100 152 233 2951902 29 40 80 100 253 272 4011947 31 51 73 100 113 105 105 1091994 11 45 100 87 93 95 68

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources)

253VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

It is perhaps not surprising that inequalitybecame strongly correlated with age in the1860s Those who died at an old age in the1820s and 1830s had lived through the difficultyears of the French Revolution and the disloca-tion of the end of the Napoleonic period Theirability to accumulate wealth had been severelyimpaired Furthermore they were rather lesslikely to inherit much wealth since the Revo-lution wiped out the bond portfolios of theirparents through a prolonged period of high in-flation Those who died from the 1860s to theearly 1910s did not suffer from the adversitiesthat plagued their forebears Instead they en-joyed the fruits of the financial sector expansionthat began in the 1850s After 1947 we seem tohave returned to a situation quite like that of1817 Presumably the capital damages associ-ated with both world wars and the Great De-pression had a strong negative effect on thewealth holdings of older generations The per-sistence of a flat age-wealth profile until 1994 islikely to be associated with two factors First insocieties where income growth is rapid abso-lute wealth accumulation is faster by youngercohorts than by older ones because their in-comes are higher at every age This is an im-portant distinction between the nineteenthcentury and the twentieth century Furthermorehighly progressive rates of income and estatetaxation have probably made it more difficult toaccumulate large fortunes thereby flattening theobserved age-wealth profile

The more interesting (and more difficult)question relates to the possible efficiency im-pact of high wealth concentration and changingage-wealth profiles Although our data do notallow us to address efficiency issues in a rigor-ous way our results allow us to formulate anumber of hypotheses and to shed new light onthe ongoing debate on inequality and growth37

From a theoretical viewpoint whether highwealth concentration can have a negativegrowth impact depends critically on the exis-tence of credit constraints With first-best credit

markets money flows toward the best entrepre-neurs and investment projects irrespective ofthe initial distribution High levels of wealthconcentration can be bad from a social justiceviewpoint but they entail no efficiency lossWhen credit constraints bind however initialwealth matters and high levels of inequality canhurt growth Whether the loss is large or smalldepends on who owns the assets If the rich areefficient investors (they know which projects tofund etc) then wealth concentration may evenbe useful If the rich are retired rentiers how-ever investing their wealth in low-yield assets(or low-ability inheritors) then high wealthconcentration and credit constraints might pre-vent talented but penniless investors from un-dertaking efficient projects thereby entailingnegative growth consequences The data used inthis paper are not ideal to address whether creditconstraints were important in a country likeFrance at the end of the nineteenth century Ourresults suggest however that to the extentcredit constraints were indeed severe highwealth concentration did have a negativegrowth impact In order to investigate this hy-pothesis further one would need to gather moresystematic data on investment strategies andasset returns Preliminary evidence suggeststhat the rich elderly of the 1860ndash1913 perioddid indeed hold a disproportionate fraction oftheir wealth in low-yield assets (such as gov-ernment bonds) An alternative hypothesishowever is that steeper age-wealth profileswere the consequence of the growth of financialmarkets as their children faced fewer creditconstraints parents decided to hold on to moreof their wealth

VI Conclusion

Evidence from wealth at death in Paris and inFrance over the last two centuries reveals threekey patterns First wealth concentration haschanged dramatically over time In 1807 thetop-1-percent share of wealth (40 percent inFrance 50 percent in Paris) was twice as high asit would be in 1994 but substantially less thanin 1913 when it peaked above 55 percent inFrance and 70 percent in Paris Some of thesechanges were due to economic phenomena thathave long been emphasized as creating inequal-ity namely industrialization and financial cen-

37 Thus far this literature has concentrated upon cross-country regressions of inequality on growth a methodologythat raises serious identification problems especially giventhe low quality of available international datasets on in-equality which are neither long-run nor homogeneous (seeeg Atkinson and Andrea Brandolini 2001)

254 THE AMERICAN ECONOMIC REVIEW MARCH 2006

tralization Yet the decline comes largely fromadverse shocks rather than economic conver-gence These changes are of such magnitudethat they are not sensitive to whether one ex-amines wealth at death in Paris or in France orwhether one examines it directly rather thanconverting it to wealth of the living by an estatemultiplier method

Our second key result is that there was asignificant transition during the nineteenth cen-tury from an important role for real estate as aform of wealth to moveable assets as the keyform of wealth for the very rich Similarly theshare of wealth held by aristocrats first rose andthen was eclipsed by that of financiers and in-dustrialists in the second half of the nineteenthcentury Hence mobility within this highly un-equal society might have been quite high Yetthis conjecture is tempered by our third findingthe wealthy were getting older over time andolder relative to less wealthy decedents Suchaging among the very wealthy would have hadnegative consequences for growth if financialmarkets were imperfect This issue requires fur-ther investigation and we hope it will attractfuture research

REFERENCES

Atkinson Anthony B ldquoTop Incomes in the UKover the 20th Centuryrdquo Journal of the RoyalStatistical Society Series A (Statistics in So-ciety) 2005 168(2) pp 325ndash43

Atkinson Anthony B and Brandolini AndrealdquoPromise and Pitfalls in the Use of lsquoSecond-aryrsquo Data-Sets Income Inequality in OECDCountries as a Case Studyrdquo Journal of Eco-nomic Literature 2001 39(3) pp 771ndash99

Atkinson Anthony B and Harrison Alan J Dis-tribution of personal wealth in Britain Cam-bridge Cambridge University Press 1978

Atkinson Anthony B and Piketty Thomas Topincomes over the twentieth century A con-trast between European and English speak-ing countries Oxford Oxford UniversityPress (forthcoming)

Banerjee Abhijit V and Duflo Esther ldquoInequalityand Growth What Can the Data Sayrdquo Journalof Economic Growth 2003 8(3) pp 267ndash99

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoPourquoi la richesse ne

srsquoest-elle pas diffusee avec la croissance Ledegre zero de lrsquoinegalite et son evolution enFrance 1800ndash1940rdquo Histoire et mesure 200318(1ndash2) pp 147ndash98

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoDefense et illustration delrsquoenquete des 3000 familles Lrsquoexemple deson volet patrimonialrdquo Annales de demogra-phie historique 2004 19 pp 19ndash57

Danysz E S ldquoContribution a lrsquoetude des for-tunes privees drsquoapres les declarations de suc-cessionsrdquo Bulletin de la statistique generalede France 1934 XXIV (1) pp 111ndash171

Daumard Adeline Les fortunes francaise auXIXe siecle Enquete sur la repartition et lacomposition des capitaux prives a ParisLyon Lille Bordeaux et Toulouse drsquoapreslrsquoenregistrement des declarations de succes-sion Paris Mouton 1973

Gain Andre La restauration et les biens desemigres 2 vols Nancy Societe drsquoimpressiontypographique 1929

Jones Alice H American colonial wealth Doc-uments and methods 3 vols New York ArnoPress 1977

Kopczuk Wojciech and Saez Emmanuel ldquoTopWealth Shares in the United States 1916ndash2000 Evidence from Estate Tax ReturnsrdquoNational Tax Journal 2004 57(2) pp 445ndash87

Kuznets Simon ldquoEconomic Growth and IncomeInequalityrdquo American Economic Review1955 45(1) pp 1ndash28

Lampman Robert J The share of top wealth-holders in national wealth 1922ndash1956Princeton Princeton University Press 1962

Lescure Michel Les banques lrsquoEtat et lemarche immobilier en France a lrsquoepoquecontemporaine 1820ndash1940 Paris EditionslrsquoEcole des Hautes Etudes en Sciences Socia-les 1982

Levy-Leboyer Maurice and Bourguignon Fran-cois LrsquoEconomie francaise au XIXe siecleAnalyse macroeconomique Paris Eco-nomica 1985 English translation Cam-bridge Cambridge University Press 1990

Lindert Peter H ldquoUnequal English Wealthsince 1670rdquo Journal of Political Economy1986 94(6) pp 1127ndash62

Lindert Peter H ldquoThree Centuries of Inequalityin Britain and Americardquo in Anthony B At-kinson and Francois Bourguignon eds

255VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

payment of taxes reduced administrative costsbecause information about real estate values didnot have to be centralized Naturally when theestate tax became progressive in 1902 returnshad to be unified Because TSA entries providelinks to the different declarations in the samebureau reassembling these declarations waseasy reassembling returns from different bu-reaux was another matter because there are nolinks across bureaux To reassemble fully allindividual returns would have required us tocollect detailed information on every decedentin Paris But we did not need to do so becausethe very high levels of inequality in Paris cameto our assistance By collecting nominal infor-mation on the top 10 percent of returns we wereable to attribute successfully 92 percent of mov-able assets and 97 percent of all real assets tospecific individuals15 Given the high variety offirst names and last names as well as detailedinformation on residence the likelihood offalsely positive matches is very low The re-maining returns were treated as individualsthus biasing downward our inequality estimates

Our 1902 Paris sample is fully consistentwith the table compiled by the tax administra-tion for the same year for Paris Therefore wecan link up our 1807ndash1902 Paris files with the1902ndash1994 Paris tables to construct homoge-neous 1807ndash1994 series for inequality in ParisThe more difficult task is to use the Paris data toinfer changes in wealth concentration for Francefrom 1807 to 1902 To do so we must estimatehow the relative importance of Paris in each topestate class evolved over the nineteenth centuryTo achieve this goal we used other estate sur-veys16 as well as nonestate fiscal sources (seeSection III below) The other difficult part is theconstruction of estimates for wealth concentra-tion among the living from estate tax datawhich we do using the estate multiplier method

and mortality data by age group (see Section IVbelow)

When using tax data it is also important tokeep in mind that tax evasion and manipulationcan potentially bias the results There are how-ever good reasons to believe that this is not toomuch of a problem here First estate tax rateswere extremely modest until World War I (lessthan 2 percent) which implies that the incen-tives for tax evasion were small In contrastpenalties for evasion were stiff Moreover theadministration made every effort to keep upwith changing composition of assets and totrack down individuals with some wealthAmong other things financial institutions andpublic utilities were required to notify the ad-ministration when accounts changed owners Asa result it was not easy to dissimulate thewealth of a decedent (either real estate or finan-cial assets in a publicly traded firm) and inher-itors had a strong incentive to register theirproperty in order to benefit from state protec-tion This suggests that the nineteenth-centurydata collected in the Paris archives is probably ofvery high quality Tax evasion is potentially amore serious issue for the twentieth century whentax rates become substantial Although top estatetax rates have rarely exceeded 20 to 30 percent fordirect transmissions in France (the top rate hasbeen 40 percent since 1984 its highest level ever)it is obvious that incentives for tax evasion haveincreased over time However several indepen-dent data sources suggest that the trends observedduring the twentieth century are robust and are notdue to the rise of tax evasion17

II Wealth Concentration at Death in Paris1807ndash1994

Figure 3 shows the evolution of wealth con-centration at death in Paris from 1807 to 1994Given that the bottom two thirds of the distri-

15 To check our procedure for 1817 1827 1877 and 1887we also assembled all declarations that matched on the firstthree letters of last name gender day of death and age theestimates of inequality are slightly higher but trivially so

16 In addition to the TRA survey (which gives a reliablepicture of the national distribution up to the ninety-fifthpercentile) we should mention the study by Daumard(1973) which relied on samples of estate tax returns col-lected in five French cities at the beginning and at the end ofthe nineteenth century (we shall come back to this importantstudy below)

17 See Bourdieu et al (2004) Furthermore the twentieth-century decline in wealth concentration observed in estatetax returns is qualitatively and quantitatively consistent withthe decline in capital income concentration observed inincome tax returns (and the latter appears to be robust inparticular it holds after scaling up tax-return capital incomeusing national accounts aggregates) This is also consistentwith several other data sources on wealth concentration andtop fortunes (especially equity ownership data) See Piketty(2001 2003) for a detailed discussion

242 THE AMERICAN ECONOMIC REVIEW MARCH 2006

bution own no wealth and the richest decileaccounts for at least 95 percent of the value ofall assets during the nineteenth century (seeTable 2) we focus on the top 1 percent Therichest 1 percent of (adult decedents) Parisiansappears to have held a stable and very highfraction of all assets during the first half of thenineteenth century (around 50 to 55 percent oftotal wealth) The 1817 spike was short-livedand was due not to a large increase in the size oftop estates but rather to a large decline inmodest estates (which apparently suffered themost from the Napoleonic Wars)18 Inequalityin Paris increased substantially after 1867 withthe top-1-percent share of wealth at death

climbing from about 52 percent to over 72 per-cent in 1913 World War I and the ensuingshocks then prompted an abrupt decline Thetop-1-percent share dropped by 34 percentagepoints between 1913 and 1947 and by about 10

18 Other spikes in the top-1-percent share are due for themost part to the volatility of the very top estates (thetop-01-percent share and mostly the top-001-percentshare) Note that with about 20000 decedents per year inParis the top-01-percent fractile includes only 20 dece-dents and the top 001 percent only 2 decedents so that theestimates for these fractiles are unstable They depend onthe identity of very wealthy individuals who happened todie in a specific year The figures reported in Table 2 are theraw figures with no adjustment whatsoever for this topwealth volatility Note however that the 1867ndash1913 up-ward trend is highly significant and does not rely on a smallnumber of very top wealth holders

FIGURE 3 WEALTH CONCENTRATION AT DEATH IN PARIS 1807ndash1994

Source Authorsrsquo computations based on estate tax returns (see Table 2)

TABLE 2mdashWEALTH CONCENTRATION AT DEATH IN PARIS1807ndash1994

Top-10-percentestate share

Top-1-percentestate share

Top-01-percentestate share

1807 960 512 1791817 976 573 2281827 973 495 1481837 977 501 1481847 983 558 2131857 969 510 1341867 971 530 1631877 969 589 2461887 971 554 2011902 991 648 2611913 996 721 3281929 949 631 2641938 904 536 2411947 767 381 1481956 750 346 1171994 669 237 65

Source Authorsrsquo computations using estate tax returns (seePiketty et al 2004 Table A2 for detailed series andsources)

243VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

percentage points between 1956 and 1994 Con-verting these wealth-at-death concentration es-timates into wealth-of-the-living concentrationestimates leaves this general picture unchanged(see Section IV below)

Who were the individuals who enjoyed sucha substantial increase in their relative wealthbetween 1867 and 1913 For the most part theirfortunes derived from large industrial and finan-cial estates As Figure 4 illustrates the share ofpersonal (non-real) estate has always been aU-shaped function of wealth This reflects thewell-known fact that real estate is a middle-class asset The poor are too poor to own land orbuildings what little they have is in furniturecash or other moveable items In contrast therich hold most of their wealth in stocks andbonds What is more interesting is that duringthe nineteenth century the relative importanceof personal wealth in Parisian estates also fol-lowed a U-shaped curve over time This wasespecially true for the very wealthy (see Figures4 and 5) where real assets became more andmore important from 1807 to 1837 Real estatethen entered a relative decline after 1837 andaccelerated after 1867

The ebb and flow of the relative importanceof real estate was linked to Parisrsquos recoveryfrom the French Revolution Prior to the Revo-lution the peripheral parts of the city had beena maze of convents monasteries and educa-tional institutions all belonging to the CatholicChurch When the wealth of the Church wasnationalized these real estate assets wereabruptly put on the private market creating aglut of buildings and low prices As the cityrsquospopulation expanded building and land valuesrecovered and the relative importance of realestate grew before being overshadowed by thefinancial boom of the last part of the century(Michel Lescure 1982)

The share of aristocratic decedents among thevery rich follows an inverted-U-shaped curveover the nineteenth century (see Figure6)19 That is nobles became more and morenumerous in top wealth fractiles from 1807 until1847 then the trend reversed and their impor-

19 We take a very broad view of aristocrats they includethe Old Regime nobility the members of the elite who weregiven titles by Napoleon (1801ndash1814) and the Bourbons(1815ndash1830)

FIGURE 4 WEALTH COMPOSITION AT DEATH IN PARIS 1807ndash1902(Share of personal (non-real) estate in total estate)

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives

244 THE AMERICAN ECONOMIC REVIEW MARCH 2006

FIGURE 5 WEALTH COMPOSITION AT DEATH IN PARIS AND FRANCE 1807ndash1902(Share of personal (non-real) estate in total estate)

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives and national aggregateestate statistics compiled by the French tax administration

FIGURE 6 ARISTOCRATIC ESTATES AT DEATH IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives

245VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

tance declined steadily To be sure aristocratsremain overrepresented throughout the periodincluding in 1902 (about 13 percent of nobles inthe top 1 percent of estates over 25 percent inthe top 01 percent versus less than 1 percent inthe population as a whole) The inverted-U pat-tern is yet another of the Revolutionrsquos legaciesIn 1807 when we first observe it aristocraticwealth was at a temporary nadir On the onehand the nobility was impoverished by theRevolutionrsquos inflation and by the sharp declineof the value of Parisian real estate On the otherhand part of the Old Regime nobility was inexile and thus if they died we do not observetheir moveable wealth Aristocrats were able torecoup part of their losses during the first half ofthe nineteenth century Napoleon providedsome assistance by conferring titles of nobilityon his chief military officers and endowingthem with wealth Later the Restoration gov-ernment (1815ndash1830) compensated individualswho fled abroad during the Revolution for thelosses they suffered when their property wasconfiscated The government distributed nearlyone billion francs in the famous milliard desemigres (Andre Gain 1929) The beneficiariesof Napoleonrsquos and the Restorationrsquos largess ap-

pear among the very rich until mid-centuryPresumably such redistribution did not contrib-ute to accelerate French industrialization

III From Paris to France

We can use the Paris data to construct wealthconcentration at death estimates for all ofFrance from 1807 to 1902 To do so we need toknow the evolution of the share of Paris estatesin top estates Between 1902 and 1994 avail-able data (broken down by departement) showsthat the evolution of top estate shares in Francewas parallel to that of top estate shares in ParisWealth inequality is always lower for the coun-try as a whole but the trends are similar (seeFigure 7) It is also striking to note that Parisrsquosshare of the top 1 percent of French estates hasremained fairly stable over the twentieth cen-tury (it fluctuates between 20 percent and 25percent with no trend) even though Parisrsquosshare of all decedents has been dwindling overtime reflecting the population decline of thecapital (see Table 3) In 1902 Paris decedentswere four times more likely to belong to thenational top 1 percent of estates than averagedecedents (26665 41) in 1994 Paris de-

FIGURE 7 WEALTH CONCENTRATION AT DEATH IN PARIS AND FRANCE 1807ndash1994

Source Authorsrsquo computations based on estate tax returns (see Tables 2 and 4)

246 THE AMERICAN ECONOMIC REVIEW MARCH 2006

cedents were seven times more likely to belongto the national top 1 percent of estates thanaverage decedents (25236 70) If anythingthe geographic concentration of fortunes waslarger at the end of the twentieth century than atthe beginning The decline of wealth concentra-tion that followed World War I was not due toredistribution between Paris and the provinces

How did the fraction of Paris estates in topestates evolve over the course of the nineteenthcentury Our estimates rely on a simple andvery conservative assumption from 1807 to1902 Parisrsquos share of estates in the top percen-tile increased at the same rate as Parisrsquos share ofFrench adult deaths More precisely let us de-note FPt(w) the cumulative distribution functionfor wealth-at-death in Paris in year t Ft(w) thecorresponding distribution for France nPt thetotal number of adult deaths in Paris in year tand nt the corresponding number for all ofFrance The ninetieth-percentile threshold P90Ptis defined by FPt(P90Pt) 09 the ninety-ninth-percentile threshold P99Pt is defined byFPt(P99Pt) 099 etc and similarly for theFrench thresholds P90t P99t etc We observeFPt(w) nPt and nt throughout the 1807ndash1994period but we do not observe Ft(w) until 1902(before this date we observe only national ag-

gregate average wealth wt Wtnt) To con-struct our benchmark estimates we assume thatthe shares s99t s995t s999t and s9999t of Parisestates in the national top 1 percent 05 percent01 percent and 001 percent of the nationalwealth-at-death distribution increased at thesame rate as nPtnt during the 1807ndash1902 period(see Table 3) Using this approximation and ourParis samples of individual tax returns we com-pute the threshold wealth levels for the toppercentiles of the national wealth distribution(eg P99 P995 P999 and P9999)20 Wealso calculate the average wealth levels for therelevant wealth classes (eg P99ndash995 P995ndash999 P999ndash9999 and P9999ndash100) usingPareto interpolation techniques These are thenweighted by the number of individuals inFrance in that wealth class in order to compute

20 For instance the number of decedents (aged 20 yearsand older) in France was 583976 in 1887 (see Piketty et al2004 Appendix Table A1) so that the top 1 percent of theestate distribution at death consists of the top 5840 estatesIf the share of Paris among French top-1-percent estates was241 percent in 1887 (see Table 3) then the national P99threshold for 1887 corresponds to the top 1410 Parisianestates (0241 5840 1410) (the national P99 thresholdfor 1887 reported in Piketty et al 2004 Appendix TableA3 was computed using this formula)

TABLE 3mdashTHE FRACTION OF PARIS ESTATES IN TOP ESTATES AT DEATH IN FRANCE 1807ndash1994

(1) Fraction of Parisdecedents in all

decedents 20-yr

(2) Fraction of Parisestates in top-10-

percent estates

(3) Fraction of Parisestates in top-1-percent estates

(4) Fraction of Parisestates in top-01-

percent estates

1807 25 101 2051817 25 103 2101827 28 116 2371837 31 126 2561847 36 146 2971857 36 146 2971867 49 199 4041877 51 211 4281887 59 241 4911902 65 75 266 5411913 65 75 255 5231929 58 83 239 5301938 53 74 216 4211947 55 110 198 3521956 55 128 223 3501994 36 89 252 352

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources) No datasource exists to compute columns (3)ndash(4) prior to 1902 and the numbers reported on this table for years 1807ndash1887 werecomputed assuming that the columns (3)ndash(4) followed the same trend as column (1) over the 1807ndash1902 period (see textSection III)

247VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

the average wealth levels for top fractiles (P99ndash100 P995ndash100 P999ndash100 and P9999ndash100)Lower thresholds of the national wealth distri-bution (P90 and P95) were computed using thenational TRA survey21 and the P90ndash95 andP95ndash99 intermediate wealth levels were alsocomputed using Pareto interpolation techniques

The national top estate shares estimates re-ported in Table 4 were computed using thismethodology They suggest that wealth concen-tration (as measured by the top-1-percent estateshare) rose throughout the nineteenth century inFrance during both the 1807ndash1867 and 1867ndash1902 periods although less sharply than in Parisduring the latter period (see Figure 7) Theseestimates are conservative in the sense that it isalmost certain that they underestimate the riseof wealth concentration that took place inFrance during the nineteenth century First weknow that the bulk of population growth in Parisduring the nineteenth century was due to theannexation of suburbs in 1860 and to populationgrowth in these peripheral arrondissements Be-

cause the outskirts of the city were poor theannexation added few top estates Thus there islittle doubt that Parisrsquos share of top estates inFrance actually increased less than its share of thetotal population This hypothesis is confirmed bynineteenth-century housing tax tabulations show-ing that the fraction of Paris taxpayers in thenational top 1 percent of taxpayers was substan-tially larger than 10 percent at the beginning of thenineteenth century22 Giving Paris a larger (andmore realistic) share of top estates in 1807 wouldboth reduce the share of wealth of the top 1percent in France at that date and lead to morerapid rise in inequality over time

Next and most importantly other estate sur-veys are consistent with the view that wealth in-equality was growing The important study byAdeline Daumard (1973) which relied on samplesof estate tax returns collected in Paris Lyon Tou-louse Lille and Bordeaux found that wealth con-centration increased in each of these five citiesduring the nineteenth century23 The TRA surveyalthough it is ill-suited for the study of top estatesis also consistent with our view Wealth dispersionwas on the rise in nineteenth-century France ac-cording to the TRA survey both in the sense thatthe fraction of decedents with positive estates de-clined over time (in spite of the sharp increase inthe value of the average estate) and that ratios suchas the P90P50 increased24 We also comparedour benchmark national P99 series extrapolatedfrom our Paris samples and the national P99series computed using the TRA survey We foundthat both series display the same overall upwardtrend in concentration (which is reassuring regard-ing the general validity of our Paris-France extrap-olation technique) except that the growth ofinequality from 1807 to 1902 in the TRA series is

21 See Bourdieu et al (2003) for full details about theTRA survey The P90 and P95 thresholds reported onPiketty et al (2004 Appendix Table A3) were computedusing ten-year moving averages around the target years inorder to make sure that each estimate was based on asufficient number of observations

22 These tabulations were published in the same FinanceMinistry official publications as the estate tabulations Wechose not to use them in our formal computations becausethe tax base of the housing tax (namely the rental value ofthe real estate property where the household lives) is onlyloosely connected to the estate tax base (in particular onecannot rule out the possibility that the housing tax baseoverrepresents Paris-based taxpayers)

23 Unfortunately Daumardrsquos samples are not available inmachine-readable format she has only two or three years ofdata for each city and she did not try to compute homog-enous inequality indicators (top fractiles shares etc) withher data Thus although her results and our work are con-sistent they cannot be compared directly

24 See Bourdieu et al (2004)

TABLE 4mdashWEALTH CONCENTRATION AT DEATH IN FRANCE1807ndash1994

Top-10-percentestate share

Top-1-percentestate share

Top-01-percentestate share

1807 791 434 1631817 810 445 1811827 824 452 1631837 796 438 1471847 816 479 1841857 829 495 1741867 810 480 1741877 838 471 2011887 839 487 1921902 839 516 2311913 863 549 2601929 820 502 2471938 776 420 1991947 699 299 1101956 694 304 1101994 610 213 63

Source Authorsrsquo computations using estate tax returns (seePiketty et al 2004 Table A3 for detailed series)

248 THE AMERICAN ECONOMIC REVIEW MARCH 2006

more severe than in our series (see Figure 8) Thisagain suggests that the latter provide a conserva-tive lower bound for the upward trend in wealthconcentration In any case the finding of a largeincrease in wealth inequality in nineteenth-centuryFrance (and up until World War I) appears to berobust25

As was mentioned earlier there exists nocomparable continuous data source covering thenineteenth and twentieth centuries in othercountries which makes it difficult to put ourFrench long-run series in international perspec-tive We note however that existing series forthe United States and the United Kingdom areconsistent with our findings on France26 Re-garding levels existing evidence suggests thatduring the nineteenth and most of the twentiethcenturies France was in an intermediate posi-tion in terms of wealth concentration in be-tween the United States (more equal) and theUnited Kingdom (the most unequal)27 These

25 Note that this continuous rise in wealth inequalitydoes not necessarily imply that a parallel rise occurredregarding income inequality Given that there exists nomicro source on incomes prior to the creation of the incometax in 1914 it is very difficult (if not impossible) to properlyaddress this issue Christian Morrisson and Wayne Snyder(2000) have attempted to link income inequality estimatesbased upon Old-Regime fiscal sources (pre-1789) withmodern income-tax-based twentieth-century estimates andthey have argued that income inequality might have startedto decline during the later part of the nineteenth century andon the eve of World War I (see also Morrisson 2000)Although our data do not allow us to rule out such apossibility we note that their nineteenth-century personaldistribution estimates are based on fragile macroeconomicdata on functional distribution and are not homogenous totheir eighteenth- and twentieth-century estimates Given theevidence that we provide on wealth inequality any signif-icant decline in aggregate income inequality would have tobe associated with severe compression of the wage distri-bution There is little research on this issue however

26 Lee Soltow and Jan L Van Zanden (1998) also find adecline in inequality in the twentieth century in the Neth-erlands Their data are consistent with a rise in inequality inthe nineteenth century but they have no direct evidenceabout its extent

27 According to our series the top-1-percent wealthshare in France rose from around 45 percent in 1800 toabout 55 percent around World War I and then fell to about20 percent by the end of the twentieth century (see Figure7) wealth concentration among the living appears to besomewhat larger (see Section V below) According to theseries pieced together by Lindert (2000 pp 181ndash82 and186) the UK top-1-percent wealth share rose from about

FIGURE 8 ESTIMATES OF THE P99 THRESHOLD FOR THE FRENCH DISTRIBUTION OF ESTATES AT DEATH EXTRAPOLATION

FROM PARIS SAMPLES VERSUS ESTIMATES FROM TRA SAMPLES (CURRENT FRENCH FRANCS)

Source Authorsrsquo computations based on estate tax returns (Paris samples and TRA samples)

249VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

differences in inequality largely hark back to dif-ferential concentration of landownership En-glandrsquos land was extremely narrowly held whilethe United States was most egalitarian The impactof access to real estate assets can also be seenwhen we contrast Paris (where it was extremelyconcentrated) with the provinces

Trends in inequality over time may be easierto compare because biases in source materialmay have a significant effect on levels ratherthan trends The historical pattern is similar ineach of the three economies In particular thereis evidence that wealth concentration increasedduring the nineteenth century in both Anglo-Saxon countries and declined during the twen-tieth century with a turning point around WorldWar I28 Neither exhibits patterns consistentwith a Kuznet process

IV From the Wealth of the Decedents to theWealth of the Living

The estimates reported thus far refer to in-equality among decedents as described in thetax returns filed by their heirs The evolution ofthe distribution of wealth among the livingmight however have followed a different pat-tern In order to convert wealth-at-death con-centration estimates into wealth-of-the-livingconcentration estimates it is standard to use theldquoestate multiplierrdquo method29 It consists ofweighting each observation of an estate at death

by the inverse of the mortality rate for this agegroup That is if the mortality rate for ages 20to 24 was 068 percent in Paris in 1902 theneach decedent aged 20 to 24 represented about147 living individuals of the same age (100068 147) Similarly if the mortality ratefor ages above 80 was 2143 percent in Paris in1902 then each decedent in that group repre-sented about 47 living individuals in the sameage group (102143 47) Applying thismethod requires mortality tables (these are eas-ily available) and estate tabulations brokendown by estate size and age at death (these arescarcer) Fortunately the cityrsquos statistical bu-reau published annual death-by-age totals theFrench censuses report the age distribution forthe capital every five years and we collectedage at death from the estate declarations Thesedata allowed us to compute estimates of wealthconcentration among the living over the 1807ndash1902 period using various assumptions aboutthe wealth profiles of mortality rates

The base population for the living is the set ofall individuals aged 20 and over living in Parisin year t which we denote pt The number ofliving individuals in age bracket a is denoted pta(a 20ndash24 25ndash29 30ndash34 75ndash79 80 andover) and the number of decedents in agebracket a is denoted nta The mortality rate forage bracket a is given by mta ntapta Webegin with a uniform-mortality benchmarkThese estimates are based on the simplifyingassumption that these mortality rates dependsolely on age and are the same for all wealthgroups (and in particular are the same forzero-wealth and positive-wealth individuals)We can then weigh each decedent with positiveestate and age a collected in the Paris archivesin year t by ptanta This allows us to computethe number of living Parisians with positivewealth in year t and also (by differentiatingwith pt) the number of living Parisians with zerowealth at year t which is used to weight zero-estate observations We then use our weighteddatasets to compute top estate fractiles amongthe living in Paris

The main conclusion is that the living expe-rienced the same upward trend in wealth con-centration as the decedents (see Figure 9) Wefind that inequality was significantly higheramong the living than among decedents be-cause survivors were on average younger than

55 percent in 1800 to 70 percent around World War I thenfell to about 20 percent in the 1990s The US top-1-percentwealth share rose from about 15 to 20 percent in 1800 toabout 40 percent around World War I then fell to about 30percent in the 1990s (and as low as 20 to 25 percentaccording to the more recent estimates from Kopczuk andSaez 2004) Wealth concentration is now larger in theUnited States than in European countries but the reversewas true during the nineteenth century up until World WarII (It is only since the 1950sndash1970s period that US wealthconcentration has been somewhat larger)

28 See Lindert (2000 pp 181ndash82 and 188)29 This method was widely used in England and France

in the late nineteenth and early twentieth centuries to com-pute the stock of national wealth on the basis of the flow ofwealth transmitted at death Standard references that use thistechnique to estimate the wealth distribution of the livingfrom estate tax data tabulated by estate size and age at deathinclude Atkinson and Harrison (1978) and Lampman(1962) For a more recent application of this technique tothe United States see Kopczuk and Saez (2004)

250 THE AMERICAN ECONOMIC REVIEW MARCH 2006

those who died and the young were on aver-age poorer In particular the estate multipliermethod leads to lower average weights forpositive-wealth decedents than for zero-wealthdecedents (the former are on average older andtherefore ldquorepresentrdquo a smaller number of livingindividuals) As a result the fraction of positive-wealth individuals is even smaller among theliving than among decedents Hence our bench-mark uniform-mortality estimates of wealthconcentration among the living are significantlylarger than corresponding estimates among de-cedents (eg top-1-percent wealth shares areabout 15 percent larger among the living)Changes over time however are similar In-creased life expectancy and declining mortalityrates over the course of the nineteenth centuryhave only a small effect on the trends

In order to make the estate multiplier methodmore reliable one would prefer to take intoaccount differential mortality by wealth Doingso would require having access to mortalityschedules based both on wealth and age at dif-ferent points in time unfortunately these are notavailable We have nonetheless reestimatedwealth of the living based upon the same as-sumption as Kopczuk and Saez (2004) That is

we assumed uniform mortality among the poor(here defined as zero-wealth individuals) andamong the rich (here defined as positive-wealthindividuals) and we assumed that the ratiomtaRmtaP between the mortality rate of the richand the mortality rate of the poor followed aU-shaped age profile from about 85 percent forthe young (ie the rich die 15 percent less oftenthan the poor when they are 20ndash24 or 25ndash29years old) down to about 70 percent for middle-age individuals in their forties to fifties and upto 100 percent for very old individuals in theireighties to nineties30 This profile correspondsto the best available estimates in the literatureand it appears to be relatively stable over timeand across developed countries In the absenceof better data it is the best one can do31 Thebenchmark differential-mortality estimates re-ported on Figure 9 show that although addingdifferential mortality produces different levelsof inequality it does not have much impact onthe upward trend in concentration

30 See Kopczuk and Saez (2004 Table A4)31 See Kopczuk and Saez (2004 Appendix B) for refer-

ences to the US and international literature devoted to theage-wealth profile of mortality rates

FIGURE 9 WEALTH CONCENTRATION AMONG DECEDENTS AND AMONG THE LIVING IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives (see Piketty et al 2004Table A4 for detailed series)

251VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

The resulting differential-mortality inequalityestimates lie between those based on decedentsonly and those using uniform mortality for theliving (see Figure 9) Note that moving fromuniform-mortality to differential-mortality es-tate multiplier methodology can either increaseor decrease inequality Here the reason whysuch a move leads to lower wealth concentra-tion seems to be due to the fact that differentialmortality tends to put higher weights onpositive-wealth decedents (for a given age)thereby increasing the estimated fraction of liv-ing individuals with positive wealth The im-portant point however is that the resulting leveleffects are relatively small in magnitude con-stant in time and dwarfed by the upward timetrend Even if we were to assume an enormousincrease in differential mortality during thenineteenth century in the sense that differentialmortality between the rich and poor was equalto 0 percent of the benchmark differential in1807 and 100 percent of the benchmark differ-ential in 1902 the resulting wealth concentra-tion estimates would still be significantly higherin 1902 than in 1807 (see Figure 9) Yet wehave no reason to believe that differential mor-tality increased to such an extent During thenineteenth century real wages for unskilledworkers rose which would have reduced mor-tality more for the poor than for the rich After1850 public health measures (sanitation watervaccination) were in place Again these wouldhave had a significant effect on the poor and themiddle class who could not privately purchasesuch health-improving services To be sure therich could avail themselves of more medicalservices than the poor or the middle class butthe impact of these services was probably small(bear in mind that neither antibiotics nor car-diovascular interventions were available)

Finally we have applied the estate multipliermethod to available data for 1947 and 1994 Over-all the sharp decline in wealth concentration ob-served during the twentieth century (andespecially between 1914 and 1945) is very robustIf anything the decline appears to be even largerwhen one looks at wealth concentration amongthe living rather than among decedents32

V The Changing Age Profile of Wealth

In the previous sections we focused almostexclusively on aggregate top wealth shares Ourdata however also detail the characteristics ofeach decedent in particular their gender andage The evolution of wealth by gender is ofrelevance for over the past two centuries therehave been massive changes in womenrsquos laborforce participation capacity to manage theirown affairs and life expectancy relative to menThe evolution of wealth by age is of relevancebecause there was a significant increase in adultlife expectancy over the twentieth century andbecause the progressive diffusion of pensionsmay have changed savings motivations More-over age-wealth profiles also inform us aboutthe motives of wealth accumulation and theeconomic impact of high wealth concentration

A first pass at the data considers the genderbreakdown of wealth at death Remarkably inour micro data the share of women in top estatestakes its highest value on the eve of World WarI For instance the womenrsquos share in the top 05percent rose from 35 percent prior to 1850 to 45percent in 1902 only to fall to 40 percent afterWorld War II Strikingly womenrsquos share ofwealth follows almost exactly the pattern ofaggregate inequality Women were relativelyricher when inequality reached its apex inFrance than at any other time Moreover insti-tutional variables seem to have played almostno role in changing the relative wealth ofwomen Unlike in common law countriesFrench law starting with the code civil of 1804required nearly equal treatment of all children inbequests Further research will help us deter-mine to what extent women of great wealthwere heirs or part of economically very success-ful couples

The data also reveal striking changes in the

32 See Piketty et al (2004 Table A4) It is unfortunatelynot possible to construct complete series for wealth concen-

tration among the living for the twentieth century due todata limitations tables broken down by estate brackets andage of decedents are available solely for years 1943ndash1954and at the national level (no table broken down by estatebrackets and age of decedents has ever been compiled at thedepartement level except in 1931 for Seine departementsee E S Danysz 1934) and the 1994 micro sample is notlarge enough to allow for a reliable application of the estatemultiplier method at the Paris level Thus the only wealth-of-the-living concentration estimates we provide for thetwentieth century are national estimates for 1947 and 1994

252 THE AMERICAN ECONOMIC REVIEW MARCH 2006

age profile of wealth between 1807 and 1994(see Table 5) During the nineteenth century aswealth concentration was increasing the veryrich were getting older and older At the begin-ning of the nineteenth century in the aftermathof the French Revolution the richest individualswere those in their fifties they were typically100 percent richer on average than people intheir forties 25 percent richer than those in theirsixties and 40 percent richer than those in theirseventies and eighties By the 1870s howeverthe age-wealth pattern had become stronglymonotonic the richest individuals were the old-est individuals In 1902 people in their sixtiesand seventies bequeathed 150 percent morethan those in their fifties and those in theireighties 300 percent more On the eve of WorldWar I top wealth holders were old and likely tobe retired This pattern breaks some time duringthe 1914ndash1945 period33 In 1947 as well as in1994 we are back to a pattern where the richestindividuals are those in their fifties Overall theperiod of maximal wealth inequality (1860ndash1913) also appears to be a period characterizedby a very specific age profile of wealth and largeconcentration of assets among the elderly

Another way to analyze the changing age-wealth relationship is to look at average age by

top estate fractile34 In 1817 average age wasvirtually the same for the top 10 percent and thetop 1 percent of estates (or even slightly declin-ing) The average-age-per-fractile relationshipbecomes upward sloping during the nineteenthcentury and by 1902 those in the top 1 percentwere almost six years older than those in the top10 percent The relationship is flat in 1947 anddownward-sloping in 1994 Finally one canapply the estate multiplier method (see SectionIV above) and analyze how wealth concentra-tion by age group among the living changedover the course of the nineteenth century Thegeneral population in Paris did not becomeolder during the nineteenth century those aged60 or older made up about 15 percent of thepopulation in 1817 and after 1847 about 10 to11 percent35 The share of total wealth ownedby the elderly rose significantly however aswealth distribution worsened The wealth be-longing to those aged 60 or more rose fromabout 25 to 30 percent of the total at the begin-ning of the nineteenth century to about 40 to 45percent by the end of the century The wealthshare of those aged 70 or older doubled fromless than 10 percent to about 20 percent36

33 Existing evidence on the age-wealth profile for 1931(see Danysz 1934) suggests that the Great Depression andWorld War II (rather than World War I) played the leadingroles in breaking this pattern This is an issue we plan toinvestigate in future research

34 See Piketty et al (2004 Table 6)35 Although life expectancy was increasing which

should have led to large shares of population for oldergroups the city was also growing quickly The large num-ber of immigrants (who were typically in their twenties)increased the relative size of the younger cohorts (seePiketty et al 2004 Table 7 and Figure 11)

36 See Piketty et al (2004 Figure 12)

TABLE 5mdashTHE AGE PROFILE OF WEALTH AT DEATH IN PARIS 1817ndash1994(Average estate left by 50- 59-year-old 100)

20ndash29yr-old

30ndash39yr-old

40ndash49yr-old

50ndash59yr-old

60ndash69yr-old

70ndash79yr-old

80ndash89yr-old

90ndash99yr-old

1817 26 22 28 100 54 59 591827 44 50 53 100 88 87 601837 133 90 107 100 116 123 1101847 87 73 102 100 117 204 1321857 84 77 101 100 104 109 1451867 67 58 136 100 141 125 1541877 66 73 63 100 197 260 4301887 45 33 63 100 152 233 2951902 29 40 80 100 253 272 4011947 31 51 73 100 113 105 105 1091994 11 45 100 87 93 95 68

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources)

253VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

It is perhaps not surprising that inequalitybecame strongly correlated with age in the1860s Those who died at an old age in the1820s and 1830s had lived through the difficultyears of the French Revolution and the disloca-tion of the end of the Napoleonic period Theirability to accumulate wealth had been severelyimpaired Furthermore they were rather lesslikely to inherit much wealth since the Revo-lution wiped out the bond portfolios of theirparents through a prolonged period of high in-flation Those who died from the 1860s to theearly 1910s did not suffer from the adversitiesthat plagued their forebears Instead they en-joyed the fruits of the financial sector expansionthat began in the 1850s After 1947 we seem tohave returned to a situation quite like that of1817 Presumably the capital damages associ-ated with both world wars and the Great De-pression had a strong negative effect on thewealth holdings of older generations The per-sistence of a flat age-wealth profile until 1994 islikely to be associated with two factors First insocieties where income growth is rapid abso-lute wealth accumulation is faster by youngercohorts than by older ones because their in-comes are higher at every age This is an im-portant distinction between the nineteenthcentury and the twentieth century Furthermorehighly progressive rates of income and estatetaxation have probably made it more difficult toaccumulate large fortunes thereby flattening theobserved age-wealth profile

The more interesting (and more difficult)question relates to the possible efficiency im-pact of high wealth concentration and changingage-wealth profiles Although our data do notallow us to address efficiency issues in a rigor-ous way our results allow us to formulate anumber of hypotheses and to shed new light onthe ongoing debate on inequality and growth37

From a theoretical viewpoint whether highwealth concentration can have a negativegrowth impact depends critically on the exis-tence of credit constraints With first-best credit

markets money flows toward the best entrepre-neurs and investment projects irrespective ofthe initial distribution High levels of wealthconcentration can be bad from a social justiceviewpoint but they entail no efficiency lossWhen credit constraints bind however initialwealth matters and high levels of inequality canhurt growth Whether the loss is large or smalldepends on who owns the assets If the rich areefficient investors (they know which projects tofund etc) then wealth concentration may evenbe useful If the rich are retired rentiers how-ever investing their wealth in low-yield assets(or low-ability inheritors) then high wealthconcentration and credit constraints might pre-vent talented but penniless investors from un-dertaking efficient projects thereby entailingnegative growth consequences The data used inthis paper are not ideal to address whether creditconstraints were important in a country likeFrance at the end of the nineteenth century Ourresults suggest however that to the extentcredit constraints were indeed severe highwealth concentration did have a negativegrowth impact In order to investigate this hy-pothesis further one would need to gather moresystematic data on investment strategies andasset returns Preliminary evidence suggeststhat the rich elderly of the 1860ndash1913 perioddid indeed hold a disproportionate fraction oftheir wealth in low-yield assets (such as gov-ernment bonds) An alternative hypothesishowever is that steeper age-wealth profileswere the consequence of the growth of financialmarkets as their children faced fewer creditconstraints parents decided to hold on to moreof their wealth

VI Conclusion

Evidence from wealth at death in Paris and inFrance over the last two centuries reveals threekey patterns First wealth concentration haschanged dramatically over time In 1807 thetop-1-percent share of wealth (40 percent inFrance 50 percent in Paris) was twice as high asit would be in 1994 but substantially less thanin 1913 when it peaked above 55 percent inFrance and 70 percent in Paris Some of thesechanges were due to economic phenomena thathave long been emphasized as creating inequal-ity namely industrialization and financial cen-

37 Thus far this literature has concentrated upon cross-country regressions of inequality on growth a methodologythat raises serious identification problems especially giventhe low quality of available international datasets on in-equality which are neither long-run nor homogeneous (seeeg Atkinson and Andrea Brandolini 2001)

254 THE AMERICAN ECONOMIC REVIEW MARCH 2006

tralization Yet the decline comes largely fromadverse shocks rather than economic conver-gence These changes are of such magnitudethat they are not sensitive to whether one ex-amines wealth at death in Paris or in France orwhether one examines it directly rather thanconverting it to wealth of the living by an estatemultiplier method

Our second key result is that there was asignificant transition during the nineteenth cen-tury from an important role for real estate as aform of wealth to moveable assets as the keyform of wealth for the very rich Similarly theshare of wealth held by aristocrats first rose andthen was eclipsed by that of financiers and in-dustrialists in the second half of the nineteenthcentury Hence mobility within this highly un-equal society might have been quite high Yetthis conjecture is tempered by our third findingthe wealthy were getting older over time andolder relative to less wealthy decedents Suchaging among the very wealthy would have hadnegative consequences for growth if financialmarkets were imperfect This issue requires fur-ther investigation and we hope it will attractfuture research

REFERENCES

Atkinson Anthony B ldquoTop Incomes in the UKover the 20th Centuryrdquo Journal of the RoyalStatistical Society Series A (Statistics in So-ciety) 2005 168(2) pp 325ndash43

Atkinson Anthony B and Brandolini AndrealdquoPromise and Pitfalls in the Use of lsquoSecond-aryrsquo Data-Sets Income Inequality in OECDCountries as a Case Studyrdquo Journal of Eco-nomic Literature 2001 39(3) pp 771ndash99

Atkinson Anthony B and Harrison Alan J Dis-tribution of personal wealth in Britain Cam-bridge Cambridge University Press 1978

Atkinson Anthony B and Piketty Thomas Topincomes over the twentieth century A con-trast between European and English speak-ing countries Oxford Oxford UniversityPress (forthcoming)

Banerjee Abhijit V and Duflo Esther ldquoInequalityand Growth What Can the Data Sayrdquo Journalof Economic Growth 2003 8(3) pp 267ndash99

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoPourquoi la richesse ne

srsquoest-elle pas diffusee avec la croissance Ledegre zero de lrsquoinegalite et son evolution enFrance 1800ndash1940rdquo Histoire et mesure 200318(1ndash2) pp 147ndash98

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoDefense et illustration delrsquoenquete des 3000 familles Lrsquoexemple deson volet patrimonialrdquo Annales de demogra-phie historique 2004 19 pp 19ndash57

Danysz E S ldquoContribution a lrsquoetude des for-tunes privees drsquoapres les declarations de suc-cessionsrdquo Bulletin de la statistique generalede France 1934 XXIV (1) pp 111ndash171

Daumard Adeline Les fortunes francaise auXIXe siecle Enquete sur la repartition et lacomposition des capitaux prives a ParisLyon Lille Bordeaux et Toulouse drsquoapreslrsquoenregistrement des declarations de succes-sion Paris Mouton 1973

Gain Andre La restauration et les biens desemigres 2 vols Nancy Societe drsquoimpressiontypographique 1929

Jones Alice H American colonial wealth Doc-uments and methods 3 vols New York ArnoPress 1977

Kopczuk Wojciech and Saez Emmanuel ldquoTopWealth Shares in the United States 1916ndash2000 Evidence from Estate Tax ReturnsrdquoNational Tax Journal 2004 57(2) pp 445ndash87

Kuznets Simon ldquoEconomic Growth and IncomeInequalityrdquo American Economic Review1955 45(1) pp 1ndash28

Lampman Robert J The share of top wealth-holders in national wealth 1922ndash1956Princeton Princeton University Press 1962

Lescure Michel Les banques lrsquoEtat et lemarche immobilier en France a lrsquoepoquecontemporaine 1820ndash1940 Paris EditionslrsquoEcole des Hautes Etudes en Sciences Socia-les 1982

Levy-Leboyer Maurice and Bourguignon Fran-cois LrsquoEconomie francaise au XIXe siecleAnalyse macroeconomique Paris Eco-nomica 1985 English translation Cam-bridge Cambridge University Press 1990

Lindert Peter H ldquoUnequal English Wealthsince 1670rdquo Journal of Political Economy1986 94(6) pp 1127ndash62

Lindert Peter H ldquoThree Centuries of Inequalityin Britain and Americardquo in Anthony B At-kinson and Francois Bourguignon eds

255VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

bution own no wealth and the richest decileaccounts for at least 95 percent of the value ofall assets during the nineteenth century (seeTable 2) we focus on the top 1 percent Therichest 1 percent of (adult decedents) Parisiansappears to have held a stable and very highfraction of all assets during the first half of thenineteenth century (around 50 to 55 percent oftotal wealth) The 1817 spike was short-livedand was due not to a large increase in the size oftop estates but rather to a large decline inmodest estates (which apparently suffered themost from the Napoleonic Wars)18 Inequalityin Paris increased substantially after 1867 withthe top-1-percent share of wealth at death

climbing from about 52 percent to over 72 per-cent in 1913 World War I and the ensuingshocks then prompted an abrupt decline Thetop-1-percent share dropped by 34 percentagepoints between 1913 and 1947 and by about 10

18 Other spikes in the top-1-percent share are due for themost part to the volatility of the very top estates (thetop-01-percent share and mostly the top-001-percentshare) Note that with about 20000 decedents per year inParis the top-01-percent fractile includes only 20 dece-dents and the top 001 percent only 2 decedents so that theestimates for these fractiles are unstable They depend onthe identity of very wealthy individuals who happened todie in a specific year The figures reported in Table 2 are theraw figures with no adjustment whatsoever for this topwealth volatility Note however that the 1867ndash1913 up-ward trend is highly significant and does not rely on a smallnumber of very top wealth holders

FIGURE 3 WEALTH CONCENTRATION AT DEATH IN PARIS 1807ndash1994

Source Authorsrsquo computations based on estate tax returns (see Table 2)

TABLE 2mdashWEALTH CONCENTRATION AT DEATH IN PARIS1807ndash1994

Top-10-percentestate share

Top-1-percentestate share

Top-01-percentestate share

1807 960 512 1791817 976 573 2281827 973 495 1481837 977 501 1481847 983 558 2131857 969 510 1341867 971 530 1631877 969 589 2461887 971 554 2011902 991 648 2611913 996 721 3281929 949 631 2641938 904 536 2411947 767 381 1481956 750 346 1171994 669 237 65

Source Authorsrsquo computations using estate tax returns (seePiketty et al 2004 Table A2 for detailed series andsources)

243VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

percentage points between 1956 and 1994 Con-verting these wealth-at-death concentration es-timates into wealth-of-the-living concentrationestimates leaves this general picture unchanged(see Section IV below)

Who were the individuals who enjoyed sucha substantial increase in their relative wealthbetween 1867 and 1913 For the most part theirfortunes derived from large industrial and finan-cial estates As Figure 4 illustrates the share ofpersonal (non-real) estate has always been aU-shaped function of wealth This reflects thewell-known fact that real estate is a middle-class asset The poor are too poor to own land orbuildings what little they have is in furniturecash or other moveable items In contrast therich hold most of their wealth in stocks andbonds What is more interesting is that duringthe nineteenth century the relative importanceof personal wealth in Parisian estates also fol-lowed a U-shaped curve over time This wasespecially true for the very wealthy (see Figures4 and 5) where real assets became more andmore important from 1807 to 1837 Real estatethen entered a relative decline after 1837 andaccelerated after 1867

The ebb and flow of the relative importanceof real estate was linked to Parisrsquos recoveryfrom the French Revolution Prior to the Revo-lution the peripheral parts of the city had beena maze of convents monasteries and educa-tional institutions all belonging to the CatholicChurch When the wealth of the Church wasnationalized these real estate assets wereabruptly put on the private market creating aglut of buildings and low prices As the cityrsquospopulation expanded building and land valuesrecovered and the relative importance of realestate grew before being overshadowed by thefinancial boom of the last part of the century(Michel Lescure 1982)

The share of aristocratic decedents among thevery rich follows an inverted-U-shaped curveover the nineteenth century (see Figure6)19 That is nobles became more and morenumerous in top wealth fractiles from 1807 until1847 then the trend reversed and their impor-

19 We take a very broad view of aristocrats they includethe Old Regime nobility the members of the elite who weregiven titles by Napoleon (1801ndash1814) and the Bourbons(1815ndash1830)

FIGURE 4 WEALTH COMPOSITION AT DEATH IN PARIS 1807ndash1902(Share of personal (non-real) estate in total estate)

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives

244 THE AMERICAN ECONOMIC REVIEW MARCH 2006

FIGURE 5 WEALTH COMPOSITION AT DEATH IN PARIS AND FRANCE 1807ndash1902(Share of personal (non-real) estate in total estate)

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives and national aggregateestate statistics compiled by the French tax administration

FIGURE 6 ARISTOCRATIC ESTATES AT DEATH IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives

245VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

tance declined steadily To be sure aristocratsremain overrepresented throughout the periodincluding in 1902 (about 13 percent of nobles inthe top 1 percent of estates over 25 percent inthe top 01 percent versus less than 1 percent inthe population as a whole) The inverted-U pat-tern is yet another of the Revolutionrsquos legaciesIn 1807 when we first observe it aristocraticwealth was at a temporary nadir On the onehand the nobility was impoverished by theRevolutionrsquos inflation and by the sharp declineof the value of Parisian real estate On the otherhand part of the Old Regime nobility was inexile and thus if they died we do not observetheir moveable wealth Aristocrats were able torecoup part of their losses during the first half ofthe nineteenth century Napoleon providedsome assistance by conferring titles of nobilityon his chief military officers and endowingthem with wealth Later the Restoration gov-ernment (1815ndash1830) compensated individualswho fled abroad during the Revolution for thelosses they suffered when their property wasconfiscated The government distributed nearlyone billion francs in the famous milliard desemigres (Andre Gain 1929) The beneficiariesof Napoleonrsquos and the Restorationrsquos largess ap-

pear among the very rich until mid-centuryPresumably such redistribution did not contrib-ute to accelerate French industrialization

III From Paris to France

We can use the Paris data to construct wealthconcentration at death estimates for all ofFrance from 1807 to 1902 To do so we need toknow the evolution of the share of Paris estatesin top estates Between 1902 and 1994 avail-able data (broken down by departement) showsthat the evolution of top estate shares in Francewas parallel to that of top estate shares in ParisWealth inequality is always lower for the coun-try as a whole but the trends are similar (seeFigure 7) It is also striking to note that Parisrsquosshare of the top 1 percent of French estates hasremained fairly stable over the twentieth cen-tury (it fluctuates between 20 percent and 25percent with no trend) even though Parisrsquosshare of all decedents has been dwindling overtime reflecting the population decline of thecapital (see Table 3) In 1902 Paris decedentswere four times more likely to belong to thenational top 1 percent of estates than averagedecedents (26665 41) in 1994 Paris de-

FIGURE 7 WEALTH CONCENTRATION AT DEATH IN PARIS AND FRANCE 1807ndash1994

Source Authorsrsquo computations based on estate tax returns (see Tables 2 and 4)

246 THE AMERICAN ECONOMIC REVIEW MARCH 2006

cedents were seven times more likely to belongto the national top 1 percent of estates thanaverage decedents (25236 70) If anythingthe geographic concentration of fortunes waslarger at the end of the twentieth century than atthe beginning The decline of wealth concentra-tion that followed World War I was not due toredistribution between Paris and the provinces

How did the fraction of Paris estates in topestates evolve over the course of the nineteenthcentury Our estimates rely on a simple andvery conservative assumption from 1807 to1902 Parisrsquos share of estates in the top percen-tile increased at the same rate as Parisrsquos share ofFrench adult deaths More precisely let us de-note FPt(w) the cumulative distribution functionfor wealth-at-death in Paris in year t Ft(w) thecorresponding distribution for France nPt thetotal number of adult deaths in Paris in year tand nt the corresponding number for all ofFrance The ninetieth-percentile threshold P90Ptis defined by FPt(P90Pt) 09 the ninety-ninth-percentile threshold P99Pt is defined byFPt(P99Pt) 099 etc and similarly for theFrench thresholds P90t P99t etc We observeFPt(w) nPt and nt throughout the 1807ndash1994period but we do not observe Ft(w) until 1902(before this date we observe only national ag-

gregate average wealth wt Wtnt) To con-struct our benchmark estimates we assume thatthe shares s99t s995t s999t and s9999t of Parisestates in the national top 1 percent 05 percent01 percent and 001 percent of the nationalwealth-at-death distribution increased at thesame rate as nPtnt during the 1807ndash1902 period(see Table 3) Using this approximation and ourParis samples of individual tax returns we com-pute the threshold wealth levels for the toppercentiles of the national wealth distribution(eg P99 P995 P999 and P9999)20 Wealso calculate the average wealth levels for therelevant wealth classes (eg P99ndash995 P995ndash999 P999ndash9999 and P9999ndash100) usingPareto interpolation techniques These are thenweighted by the number of individuals inFrance in that wealth class in order to compute

20 For instance the number of decedents (aged 20 yearsand older) in France was 583976 in 1887 (see Piketty et al2004 Appendix Table A1) so that the top 1 percent of theestate distribution at death consists of the top 5840 estatesIf the share of Paris among French top-1-percent estates was241 percent in 1887 (see Table 3) then the national P99threshold for 1887 corresponds to the top 1410 Parisianestates (0241 5840 1410) (the national P99 thresholdfor 1887 reported in Piketty et al 2004 Appendix TableA3 was computed using this formula)

TABLE 3mdashTHE FRACTION OF PARIS ESTATES IN TOP ESTATES AT DEATH IN FRANCE 1807ndash1994

(1) Fraction of Parisdecedents in all

decedents 20-yr

(2) Fraction of Parisestates in top-10-

percent estates

(3) Fraction of Parisestates in top-1-percent estates

(4) Fraction of Parisestates in top-01-

percent estates

1807 25 101 2051817 25 103 2101827 28 116 2371837 31 126 2561847 36 146 2971857 36 146 2971867 49 199 4041877 51 211 4281887 59 241 4911902 65 75 266 5411913 65 75 255 5231929 58 83 239 5301938 53 74 216 4211947 55 110 198 3521956 55 128 223 3501994 36 89 252 352

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources) No datasource exists to compute columns (3)ndash(4) prior to 1902 and the numbers reported on this table for years 1807ndash1887 werecomputed assuming that the columns (3)ndash(4) followed the same trend as column (1) over the 1807ndash1902 period (see textSection III)

247VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

the average wealth levels for top fractiles (P99ndash100 P995ndash100 P999ndash100 and P9999ndash100)Lower thresholds of the national wealth distri-bution (P90 and P95) were computed using thenational TRA survey21 and the P90ndash95 andP95ndash99 intermediate wealth levels were alsocomputed using Pareto interpolation techniques

The national top estate shares estimates re-ported in Table 4 were computed using thismethodology They suggest that wealth concen-tration (as measured by the top-1-percent estateshare) rose throughout the nineteenth century inFrance during both the 1807ndash1867 and 1867ndash1902 periods although less sharply than in Parisduring the latter period (see Figure 7) Theseestimates are conservative in the sense that it isalmost certain that they underestimate the riseof wealth concentration that took place inFrance during the nineteenth century First weknow that the bulk of population growth in Parisduring the nineteenth century was due to theannexation of suburbs in 1860 and to populationgrowth in these peripheral arrondissements Be-

cause the outskirts of the city were poor theannexation added few top estates Thus there islittle doubt that Parisrsquos share of top estates inFrance actually increased less than its share of thetotal population This hypothesis is confirmed bynineteenth-century housing tax tabulations show-ing that the fraction of Paris taxpayers in thenational top 1 percent of taxpayers was substan-tially larger than 10 percent at the beginning of thenineteenth century22 Giving Paris a larger (andmore realistic) share of top estates in 1807 wouldboth reduce the share of wealth of the top 1percent in France at that date and lead to morerapid rise in inequality over time

Next and most importantly other estate sur-veys are consistent with the view that wealth in-equality was growing The important study byAdeline Daumard (1973) which relied on samplesof estate tax returns collected in Paris Lyon Tou-louse Lille and Bordeaux found that wealth con-centration increased in each of these five citiesduring the nineteenth century23 The TRA surveyalthough it is ill-suited for the study of top estatesis also consistent with our view Wealth dispersionwas on the rise in nineteenth-century France ac-cording to the TRA survey both in the sense thatthe fraction of decedents with positive estates de-clined over time (in spite of the sharp increase inthe value of the average estate) and that ratios suchas the P90P50 increased24 We also comparedour benchmark national P99 series extrapolatedfrom our Paris samples and the national P99series computed using the TRA survey We foundthat both series display the same overall upwardtrend in concentration (which is reassuring regard-ing the general validity of our Paris-France extrap-olation technique) except that the growth ofinequality from 1807 to 1902 in the TRA series is

21 See Bourdieu et al (2003) for full details about theTRA survey The P90 and P95 thresholds reported onPiketty et al (2004 Appendix Table A3) were computedusing ten-year moving averages around the target years inorder to make sure that each estimate was based on asufficient number of observations

22 These tabulations were published in the same FinanceMinistry official publications as the estate tabulations Wechose not to use them in our formal computations becausethe tax base of the housing tax (namely the rental value ofthe real estate property where the household lives) is onlyloosely connected to the estate tax base (in particular onecannot rule out the possibility that the housing tax baseoverrepresents Paris-based taxpayers)

23 Unfortunately Daumardrsquos samples are not available inmachine-readable format she has only two or three years ofdata for each city and she did not try to compute homog-enous inequality indicators (top fractiles shares etc) withher data Thus although her results and our work are con-sistent they cannot be compared directly

24 See Bourdieu et al (2004)

TABLE 4mdashWEALTH CONCENTRATION AT DEATH IN FRANCE1807ndash1994

Top-10-percentestate share

Top-1-percentestate share

Top-01-percentestate share

1807 791 434 1631817 810 445 1811827 824 452 1631837 796 438 1471847 816 479 1841857 829 495 1741867 810 480 1741877 838 471 2011887 839 487 1921902 839 516 2311913 863 549 2601929 820 502 2471938 776 420 1991947 699 299 1101956 694 304 1101994 610 213 63

Source Authorsrsquo computations using estate tax returns (seePiketty et al 2004 Table A3 for detailed series)

248 THE AMERICAN ECONOMIC REVIEW MARCH 2006

more severe than in our series (see Figure 8) Thisagain suggests that the latter provide a conserva-tive lower bound for the upward trend in wealthconcentration In any case the finding of a largeincrease in wealth inequality in nineteenth-centuryFrance (and up until World War I) appears to berobust25

As was mentioned earlier there exists nocomparable continuous data source covering thenineteenth and twentieth centuries in othercountries which makes it difficult to put ourFrench long-run series in international perspec-tive We note however that existing series forthe United States and the United Kingdom areconsistent with our findings on France26 Re-garding levels existing evidence suggests thatduring the nineteenth and most of the twentiethcenturies France was in an intermediate posi-tion in terms of wealth concentration in be-tween the United States (more equal) and theUnited Kingdom (the most unequal)27 These

25 Note that this continuous rise in wealth inequalitydoes not necessarily imply that a parallel rise occurredregarding income inequality Given that there exists nomicro source on incomes prior to the creation of the incometax in 1914 it is very difficult (if not impossible) to properlyaddress this issue Christian Morrisson and Wayne Snyder(2000) have attempted to link income inequality estimatesbased upon Old-Regime fiscal sources (pre-1789) withmodern income-tax-based twentieth-century estimates andthey have argued that income inequality might have startedto decline during the later part of the nineteenth century andon the eve of World War I (see also Morrisson 2000)Although our data do not allow us to rule out such apossibility we note that their nineteenth-century personaldistribution estimates are based on fragile macroeconomicdata on functional distribution and are not homogenous totheir eighteenth- and twentieth-century estimates Given theevidence that we provide on wealth inequality any signif-icant decline in aggregate income inequality would have tobe associated with severe compression of the wage distri-bution There is little research on this issue however

26 Lee Soltow and Jan L Van Zanden (1998) also find adecline in inequality in the twentieth century in the Neth-erlands Their data are consistent with a rise in inequality inthe nineteenth century but they have no direct evidenceabout its extent

27 According to our series the top-1-percent wealthshare in France rose from around 45 percent in 1800 toabout 55 percent around World War I and then fell to about20 percent by the end of the twentieth century (see Figure7) wealth concentration among the living appears to besomewhat larger (see Section V below) According to theseries pieced together by Lindert (2000 pp 181ndash82 and186) the UK top-1-percent wealth share rose from about

FIGURE 8 ESTIMATES OF THE P99 THRESHOLD FOR THE FRENCH DISTRIBUTION OF ESTATES AT DEATH EXTRAPOLATION

FROM PARIS SAMPLES VERSUS ESTIMATES FROM TRA SAMPLES (CURRENT FRENCH FRANCS)

Source Authorsrsquo computations based on estate tax returns (Paris samples and TRA samples)

249VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

differences in inequality largely hark back to dif-ferential concentration of landownership En-glandrsquos land was extremely narrowly held whilethe United States was most egalitarian The impactof access to real estate assets can also be seenwhen we contrast Paris (where it was extremelyconcentrated) with the provinces

Trends in inequality over time may be easierto compare because biases in source materialmay have a significant effect on levels ratherthan trends The historical pattern is similar ineach of the three economies In particular thereis evidence that wealth concentration increasedduring the nineteenth century in both Anglo-Saxon countries and declined during the twen-tieth century with a turning point around WorldWar I28 Neither exhibits patterns consistentwith a Kuznet process

IV From the Wealth of the Decedents to theWealth of the Living

The estimates reported thus far refer to in-equality among decedents as described in thetax returns filed by their heirs The evolution ofthe distribution of wealth among the livingmight however have followed a different pat-tern In order to convert wealth-at-death con-centration estimates into wealth-of-the-livingconcentration estimates it is standard to use theldquoestate multiplierrdquo method29 It consists ofweighting each observation of an estate at death

by the inverse of the mortality rate for this agegroup That is if the mortality rate for ages 20to 24 was 068 percent in Paris in 1902 theneach decedent aged 20 to 24 represented about147 living individuals of the same age (100068 147) Similarly if the mortality ratefor ages above 80 was 2143 percent in Paris in1902 then each decedent in that group repre-sented about 47 living individuals in the sameage group (102143 47) Applying thismethod requires mortality tables (these are eas-ily available) and estate tabulations brokendown by estate size and age at death (these arescarcer) Fortunately the cityrsquos statistical bu-reau published annual death-by-age totals theFrench censuses report the age distribution forthe capital every five years and we collectedage at death from the estate declarations Thesedata allowed us to compute estimates of wealthconcentration among the living over the 1807ndash1902 period using various assumptions aboutthe wealth profiles of mortality rates

The base population for the living is the set ofall individuals aged 20 and over living in Parisin year t which we denote pt The number ofliving individuals in age bracket a is denoted pta(a 20ndash24 25ndash29 30ndash34 75ndash79 80 andover) and the number of decedents in agebracket a is denoted nta The mortality rate forage bracket a is given by mta ntapta Webegin with a uniform-mortality benchmarkThese estimates are based on the simplifyingassumption that these mortality rates dependsolely on age and are the same for all wealthgroups (and in particular are the same forzero-wealth and positive-wealth individuals)We can then weigh each decedent with positiveestate and age a collected in the Paris archivesin year t by ptanta This allows us to computethe number of living Parisians with positivewealth in year t and also (by differentiatingwith pt) the number of living Parisians with zerowealth at year t which is used to weight zero-estate observations We then use our weighteddatasets to compute top estate fractiles amongthe living in Paris

The main conclusion is that the living expe-rienced the same upward trend in wealth con-centration as the decedents (see Figure 9) Wefind that inequality was significantly higheramong the living than among decedents be-cause survivors were on average younger than

55 percent in 1800 to 70 percent around World War I thenfell to about 20 percent in the 1990s The US top-1-percentwealth share rose from about 15 to 20 percent in 1800 toabout 40 percent around World War I then fell to about 30percent in the 1990s (and as low as 20 to 25 percentaccording to the more recent estimates from Kopczuk andSaez 2004) Wealth concentration is now larger in theUnited States than in European countries but the reversewas true during the nineteenth century up until World WarII (It is only since the 1950sndash1970s period that US wealthconcentration has been somewhat larger)

28 See Lindert (2000 pp 181ndash82 and 188)29 This method was widely used in England and France

in the late nineteenth and early twentieth centuries to com-pute the stock of national wealth on the basis of the flow ofwealth transmitted at death Standard references that use thistechnique to estimate the wealth distribution of the livingfrom estate tax data tabulated by estate size and age at deathinclude Atkinson and Harrison (1978) and Lampman(1962) For a more recent application of this technique tothe United States see Kopczuk and Saez (2004)

250 THE AMERICAN ECONOMIC REVIEW MARCH 2006

those who died and the young were on aver-age poorer In particular the estate multipliermethod leads to lower average weights forpositive-wealth decedents than for zero-wealthdecedents (the former are on average older andtherefore ldquorepresentrdquo a smaller number of livingindividuals) As a result the fraction of positive-wealth individuals is even smaller among theliving than among decedents Hence our bench-mark uniform-mortality estimates of wealthconcentration among the living are significantlylarger than corresponding estimates among de-cedents (eg top-1-percent wealth shares areabout 15 percent larger among the living)Changes over time however are similar In-creased life expectancy and declining mortalityrates over the course of the nineteenth centuryhave only a small effect on the trends

In order to make the estate multiplier methodmore reliable one would prefer to take intoaccount differential mortality by wealth Doingso would require having access to mortalityschedules based both on wealth and age at dif-ferent points in time unfortunately these are notavailable We have nonetheless reestimatedwealth of the living based upon the same as-sumption as Kopczuk and Saez (2004) That is

we assumed uniform mortality among the poor(here defined as zero-wealth individuals) andamong the rich (here defined as positive-wealthindividuals) and we assumed that the ratiomtaRmtaP between the mortality rate of the richand the mortality rate of the poor followed aU-shaped age profile from about 85 percent forthe young (ie the rich die 15 percent less oftenthan the poor when they are 20ndash24 or 25ndash29years old) down to about 70 percent for middle-age individuals in their forties to fifties and upto 100 percent for very old individuals in theireighties to nineties30 This profile correspondsto the best available estimates in the literatureand it appears to be relatively stable over timeand across developed countries In the absenceof better data it is the best one can do31 Thebenchmark differential-mortality estimates re-ported on Figure 9 show that although addingdifferential mortality produces different levelsof inequality it does not have much impact onthe upward trend in concentration

30 See Kopczuk and Saez (2004 Table A4)31 See Kopczuk and Saez (2004 Appendix B) for refer-

ences to the US and international literature devoted to theage-wealth profile of mortality rates

FIGURE 9 WEALTH CONCENTRATION AMONG DECEDENTS AND AMONG THE LIVING IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives (see Piketty et al 2004Table A4 for detailed series)

251VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

The resulting differential-mortality inequalityestimates lie between those based on decedentsonly and those using uniform mortality for theliving (see Figure 9) Note that moving fromuniform-mortality to differential-mortality es-tate multiplier methodology can either increaseor decrease inequality Here the reason whysuch a move leads to lower wealth concentra-tion seems to be due to the fact that differentialmortality tends to put higher weights onpositive-wealth decedents (for a given age)thereby increasing the estimated fraction of liv-ing individuals with positive wealth The im-portant point however is that the resulting leveleffects are relatively small in magnitude con-stant in time and dwarfed by the upward timetrend Even if we were to assume an enormousincrease in differential mortality during thenineteenth century in the sense that differentialmortality between the rich and poor was equalto 0 percent of the benchmark differential in1807 and 100 percent of the benchmark differ-ential in 1902 the resulting wealth concentra-tion estimates would still be significantly higherin 1902 than in 1807 (see Figure 9) Yet wehave no reason to believe that differential mor-tality increased to such an extent During thenineteenth century real wages for unskilledworkers rose which would have reduced mor-tality more for the poor than for the rich After1850 public health measures (sanitation watervaccination) were in place Again these wouldhave had a significant effect on the poor and themiddle class who could not privately purchasesuch health-improving services To be sure therich could avail themselves of more medicalservices than the poor or the middle class butthe impact of these services was probably small(bear in mind that neither antibiotics nor car-diovascular interventions were available)

Finally we have applied the estate multipliermethod to available data for 1947 and 1994 Over-all the sharp decline in wealth concentration ob-served during the twentieth century (andespecially between 1914 and 1945) is very robustIf anything the decline appears to be even largerwhen one looks at wealth concentration amongthe living rather than among decedents32

V The Changing Age Profile of Wealth

In the previous sections we focused almostexclusively on aggregate top wealth shares Ourdata however also detail the characteristics ofeach decedent in particular their gender andage The evolution of wealth by gender is ofrelevance for over the past two centuries therehave been massive changes in womenrsquos laborforce participation capacity to manage theirown affairs and life expectancy relative to menThe evolution of wealth by age is of relevancebecause there was a significant increase in adultlife expectancy over the twentieth century andbecause the progressive diffusion of pensionsmay have changed savings motivations More-over age-wealth profiles also inform us aboutthe motives of wealth accumulation and theeconomic impact of high wealth concentration

A first pass at the data considers the genderbreakdown of wealth at death Remarkably inour micro data the share of women in top estatestakes its highest value on the eve of World WarI For instance the womenrsquos share in the top 05percent rose from 35 percent prior to 1850 to 45percent in 1902 only to fall to 40 percent afterWorld War II Strikingly womenrsquos share ofwealth follows almost exactly the pattern ofaggregate inequality Women were relativelyricher when inequality reached its apex inFrance than at any other time Moreover insti-tutional variables seem to have played almostno role in changing the relative wealth ofwomen Unlike in common law countriesFrench law starting with the code civil of 1804required nearly equal treatment of all children inbequests Further research will help us deter-mine to what extent women of great wealthwere heirs or part of economically very success-ful couples

The data also reveal striking changes in the

32 See Piketty et al (2004 Table A4) It is unfortunatelynot possible to construct complete series for wealth concen-

tration among the living for the twentieth century due todata limitations tables broken down by estate brackets andage of decedents are available solely for years 1943ndash1954and at the national level (no table broken down by estatebrackets and age of decedents has ever been compiled at thedepartement level except in 1931 for Seine departementsee E S Danysz 1934) and the 1994 micro sample is notlarge enough to allow for a reliable application of the estatemultiplier method at the Paris level Thus the only wealth-of-the-living concentration estimates we provide for thetwentieth century are national estimates for 1947 and 1994

252 THE AMERICAN ECONOMIC REVIEW MARCH 2006

age profile of wealth between 1807 and 1994(see Table 5) During the nineteenth century aswealth concentration was increasing the veryrich were getting older and older At the begin-ning of the nineteenth century in the aftermathof the French Revolution the richest individualswere those in their fifties they were typically100 percent richer on average than people intheir forties 25 percent richer than those in theirsixties and 40 percent richer than those in theirseventies and eighties By the 1870s howeverthe age-wealth pattern had become stronglymonotonic the richest individuals were the old-est individuals In 1902 people in their sixtiesand seventies bequeathed 150 percent morethan those in their fifties and those in theireighties 300 percent more On the eve of WorldWar I top wealth holders were old and likely tobe retired This pattern breaks some time duringthe 1914ndash1945 period33 In 1947 as well as in1994 we are back to a pattern where the richestindividuals are those in their fifties Overall theperiod of maximal wealth inequality (1860ndash1913) also appears to be a period characterizedby a very specific age profile of wealth and largeconcentration of assets among the elderly

Another way to analyze the changing age-wealth relationship is to look at average age by

top estate fractile34 In 1817 average age wasvirtually the same for the top 10 percent and thetop 1 percent of estates (or even slightly declin-ing) The average-age-per-fractile relationshipbecomes upward sloping during the nineteenthcentury and by 1902 those in the top 1 percentwere almost six years older than those in the top10 percent The relationship is flat in 1947 anddownward-sloping in 1994 Finally one canapply the estate multiplier method (see SectionIV above) and analyze how wealth concentra-tion by age group among the living changedover the course of the nineteenth century Thegeneral population in Paris did not becomeolder during the nineteenth century those aged60 or older made up about 15 percent of thepopulation in 1817 and after 1847 about 10 to11 percent35 The share of total wealth ownedby the elderly rose significantly however aswealth distribution worsened The wealth be-longing to those aged 60 or more rose fromabout 25 to 30 percent of the total at the begin-ning of the nineteenth century to about 40 to 45percent by the end of the century The wealthshare of those aged 70 or older doubled fromless than 10 percent to about 20 percent36

33 Existing evidence on the age-wealth profile for 1931(see Danysz 1934) suggests that the Great Depression andWorld War II (rather than World War I) played the leadingroles in breaking this pattern This is an issue we plan toinvestigate in future research

34 See Piketty et al (2004 Table 6)35 Although life expectancy was increasing which

should have led to large shares of population for oldergroups the city was also growing quickly The large num-ber of immigrants (who were typically in their twenties)increased the relative size of the younger cohorts (seePiketty et al 2004 Table 7 and Figure 11)

36 See Piketty et al (2004 Figure 12)

TABLE 5mdashTHE AGE PROFILE OF WEALTH AT DEATH IN PARIS 1817ndash1994(Average estate left by 50- 59-year-old 100)

20ndash29yr-old

30ndash39yr-old

40ndash49yr-old

50ndash59yr-old

60ndash69yr-old

70ndash79yr-old

80ndash89yr-old

90ndash99yr-old

1817 26 22 28 100 54 59 591827 44 50 53 100 88 87 601837 133 90 107 100 116 123 1101847 87 73 102 100 117 204 1321857 84 77 101 100 104 109 1451867 67 58 136 100 141 125 1541877 66 73 63 100 197 260 4301887 45 33 63 100 152 233 2951902 29 40 80 100 253 272 4011947 31 51 73 100 113 105 105 1091994 11 45 100 87 93 95 68

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources)

253VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

It is perhaps not surprising that inequalitybecame strongly correlated with age in the1860s Those who died at an old age in the1820s and 1830s had lived through the difficultyears of the French Revolution and the disloca-tion of the end of the Napoleonic period Theirability to accumulate wealth had been severelyimpaired Furthermore they were rather lesslikely to inherit much wealth since the Revo-lution wiped out the bond portfolios of theirparents through a prolonged period of high in-flation Those who died from the 1860s to theearly 1910s did not suffer from the adversitiesthat plagued their forebears Instead they en-joyed the fruits of the financial sector expansionthat began in the 1850s After 1947 we seem tohave returned to a situation quite like that of1817 Presumably the capital damages associ-ated with both world wars and the Great De-pression had a strong negative effect on thewealth holdings of older generations The per-sistence of a flat age-wealth profile until 1994 islikely to be associated with two factors First insocieties where income growth is rapid abso-lute wealth accumulation is faster by youngercohorts than by older ones because their in-comes are higher at every age This is an im-portant distinction between the nineteenthcentury and the twentieth century Furthermorehighly progressive rates of income and estatetaxation have probably made it more difficult toaccumulate large fortunes thereby flattening theobserved age-wealth profile

The more interesting (and more difficult)question relates to the possible efficiency im-pact of high wealth concentration and changingage-wealth profiles Although our data do notallow us to address efficiency issues in a rigor-ous way our results allow us to formulate anumber of hypotheses and to shed new light onthe ongoing debate on inequality and growth37

From a theoretical viewpoint whether highwealth concentration can have a negativegrowth impact depends critically on the exis-tence of credit constraints With first-best credit

markets money flows toward the best entrepre-neurs and investment projects irrespective ofthe initial distribution High levels of wealthconcentration can be bad from a social justiceviewpoint but they entail no efficiency lossWhen credit constraints bind however initialwealth matters and high levels of inequality canhurt growth Whether the loss is large or smalldepends on who owns the assets If the rich areefficient investors (they know which projects tofund etc) then wealth concentration may evenbe useful If the rich are retired rentiers how-ever investing their wealth in low-yield assets(or low-ability inheritors) then high wealthconcentration and credit constraints might pre-vent talented but penniless investors from un-dertaking efficient projects thereby entailingnegative growth consequences The data used inthis paper are not ideal to address whether creditconstraints were important in a country likeFrance at the end of the nineteenth century Ourresults suggest however that to the extentcredit constraints were indeed severe highwealth concentration did have a negativegrowth impact In order to investigate this hy-pothesis further one would need to gather moresystematic data on investment strategies andasset returns Preliminary evidence suggeststhat the rich elderly of the 1860ndash1913 perioddid indeed hold a disproportionate fraction oftheir wealth in low-yield assets (such as gov-ernment bonds) An alternative hypothesishowever is that steeper age-wealth profileswere the consequence of the growth of financialmarkets as their children faced fewer creditconstraints parents decided to hold on to moreof their wealth

VI Conclusion

Evidence from wealth at death in Paris and inFrance over the last two centuries reveals threekey patterns First wealth concentration haschanged dramatically over time In 1807 thetop-1-percent share of wealth (40 percent inFrance 50 percent in Paris) was twice as high asit would be in 1994 but substantially less thanin 1913 when it peaked above 55 percent inFrance and 70 percent in Paris Some of thesechanges were due to economic phenomena thathave long been emphasized as creating inequal-ity namely industrialization and financial cen-

37 Thus far this literature has concentrated upon cross-country regressions of inequality on growth a methodologythat raises serious identification problems especially giventhe low quality of available international datasets on in-equality which are neither long-run nor homogeneous (seeeg Atkinson and Andrea Brandolini 2001)

254 THE AMERICAN ECONOMIC REVIEW MARCH 2006

tralization Yet the decline comes largely fromadverse shocks rather than economic conver-gence These changes are of such magnitudethat they are not sensitive to whether one ex-amines wealth at death in Paris or in France orwhether one examines it directly rather thanconverting it to wealth of the living by an estatemultiplier method

Our second key result is that there was asignificant transition during the nineteenth cen-tury from an important role for real estate as aform of wealth to moveable assets as the keyform of wealth for the very rich Similarly theshare of wealth held by aristocrats first rose andthen was eclipsed by that of financiers and in-dustrialists in the second half of the nineteenthcentury Hence mobility within this highly un-equal society might have been quite high Yetthis conjecture is tempered by our third findingthe wealthy were getting older over time andolder relative to less wealthy decedents Suchaging among the very wealthy would have hadnegative consequences for growth if financialmarkets were imperfect This issue requires fur-ther investigation and we hope it will attractfuture research

REFERENCES

Atkinson Anthony B ldquoTop Incomes in the UKover the 20th Centuryrdquo Journal of the RoyalStatistical Society Series A (Statistics in So-ciety) 2005 168(2) pp 325ndash43

Atkinson Anthony B and Brandolini AndrealdquoPromise and Pitfalls in the Use of lsquoSecond-aryrsquo Data-Sets Income Inequality in OECDCountries as a Case Studyrdquo Journal of Eco-nomic Literature 2001 39(3) pp 771ndash99

Atkinson Anthony B and Harrison Alan J Dis-tribution of personal wealth in Britain Cam-bridge Cambridge University Press 1978

Atkinson Anthony B and Piketty Thomas Topincomes over the twentieth century A con-trast between European and English speak-ing countries Oxford Oxford UniversityPress (forthcoming)

Banerjee Abhijit V and Duflo Esther ldquoInequalityand Growth What Can the Data Sayrdquo Journalof Economic Growth 2003 8(3) pp 267ndash99

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoPourquoi la richesse ne

srsquoest-elle pas diffusee avec la croissance Ledegre zero de lrsquoinegalite et son evolution enFrance 1800ndash1940rdquo Histoire et mesure 200318(1ndash2) pp 147ndash98

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoDefense et illustration delrsquoenquete des 3000 familles Lrsquoexemple deson volet patrimonialrdquo Annales de demogra-phie historique 2004 19 pp 19ndash57

Danysz E S ldquoContribution a lrsquoetude des for-tunes privees drsquoapres les declarations de suc-cessionsrdquo Bulletin de la statistique generalede France 1934 XXIV (1) pp 111ndash171

Daumard Adeline Les fortunes francaise auXIXe siecle Enquete sur la repartition et lacomposition des capitaux prives a ParisLyon Lille Bordeaux et Toulouse drsquoapreslrsquoenregistrement des declarations de succes-sion Paris Mouton 1973

Gain Andre La restauration et les biens desemigres 2 vols Nancy Societe drsquoimpressiontypographique 1929

Jones Alice H American colonial wealth Doc-uments and methods 3 vols New York ArnoPress 1977

Kopczuk Wojciech and Saez Emmanuel ldquoTopWealth Shares in the United States 1916ndash2000 Evidence from Estate Tax ReturnsrdquoNational Tax Journal 2004 57(2) pp 445ndash87

Kuznets Simon ldquoEconomic Growth and IncomeInequalityrdquo American Economic Review1955 45(1) pp 1ndash28

Lampman Robert J The share of top wealth-holders in national wealth 1922ndash1956Princeton Princeton University Press 1962

Lescure Michel Les banques lrsquoEtat et lemarche immobilier en France a lrsquoepoquecontemporaine 1820ndash1940 Paris EditionslrsquoEcole des Hautes Etudes en Sciences Socia-les 1982

Levy-Leboyer Maurice and Bourguignon Fran-cois LrsquoEconomie francaise au XIXe siecleAnalyse macroeconomique Paris Eco-nomica 1985 English translation Cam-bridge Cambridge University Press 1990

Lindert Peter H ldquoUnequal English Wealthsince 1670rdquo Journal of Political Economy1986 94(6) pp 1127ndash62

Lindert Peter H ldquoThree Centuries of Inequalityin Britain and Americardquo in Anthony B At-kinson and Francois Bourguignon eds

255VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

percentage points between 1956 and 1994 Con-verting these wealth-at-death concentration es-timates into wealth-of-the-living concentrationestimates leaves this general picture unchanged(see Section IV below)

Who were the individuals who enjoyed sucha substantial increase in their relative wealthbetween 1867 and 1913 For the most part theirfortunes derived from large industrial and finan-cial estates As Figure 4 illustrates the share ofpersonal (non-real) estate has always been aU-shaped function of wealth This reflects thewell-known fact that real estate is a middle-class asset The poor are too poor to own land orbuildings what little they have is in furniturecash or other moveable items In contrast therich hold most of their wealth in stocks andbonds What is more interesting is that duringthe nineteenth century the relative importanceof personal wealth in Parisian estates also fol-lowed a U-shaped curve over time This wasespecially true for the very wealthy (see Figures4 and 5) where real assets became more andmore important from 1807 to 1837 Real estatethen entered a relative decline after 1837 andaccelerated after 1867

The ebb and flow of the relative importanceof real estate was linked to Parisrsquos recoveryfrom the French Revolution Prior to the Revo-lution the peripheral parts of the city had beena maze of convents monasteries and educa-tional institutions all belonging to the CatholicChurch When the wealth of the Church wasnationalized these real estate assets wereabruptly put on the private market creating aglut of buildings and low prices As the cityrsquospopulation expanded building and land valuesrecovered and the relative importance of realestate grew before being overshadowed by thefinancial boom of the last part of the century(Michel Lescure 1982)

The share of aristocratic decedents among thevery rich follows an inverted-U-shaped curveover the nineteenth century (see Figure6)19 That is nobles became more and morenumerous in top wealth fractiles from 1807 until1847 then the trend reversed and their impor-

19 We take a very broad view of aristocrats they includethe Old Regime nobility the members of the elite who weregiven titles by Napoleon (1801ndash1814) and the Bourbons(1815ndash1830)

FIGURE 4 WEALTH COMPOSITION AT DEATH IN PARIS 1807ndash1902(Share of personal (non-real) estate in total estate)

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives

244 THE AMERICAN ECONOMIC REVIEW MARCH 2006

FIGURE 5 WEALTH COMPOSITION AT DEATH IN PARIS AND FRANCE 1807ndash1902(Share of personal (non-real) estate in total estate)

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives and national aggregateestate statistics compiled by the French tax administration

FIGURE 6 ARISTOCRATIC ESTATES AT DEATH IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives

245VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

tance declined steadily To be sure aristocratsremain overrepresented throughout the periodincluding in 1902 (about 13 percent of nobles inthe top 1 percent of estates over 25 percent inthe top 01 percent versus less than 1 percent inthe population as a whole) The inverted-U pat-tern is yet another of the Revolutionrsquos legaciesIn 1807 when we first observe it aristocraticwealth was at a temporary nadir On the onehand the nobility was impoverished by theRevolutionrsquos inflation and by the sharp declineof the value of Parisian real estate On the otherhand part of the Old Regime nobility was inexile and thus if they died we do not observetheir moveable wealth Aristocrats were able torecoup part of their losses during the first half ofthe nineteenth century Napoleon providedsome assistance by conferring titles of nobilityon his chief military officers and endowingthem with wealth Later the Restoration gov-ernment (1815ndash1830) compensated individualswho fled abroad during the Revolution for thelosses they suffered when their property wasconfiscated The government distributed nearlyone billion francs in the famous milliard desemigres (Andre Gain 1929) The beneficiariesof Napoleonrsquos and the Restorationrsquos largess ap-

pear among the very rich until mid-centuryPresumably such redistribution did not contrib-ute to accelerate French industrialization

III From Paris to France

We can use the Paris data to construct wealthconcentration at death estimates for all ofFrance from 1807 to 1902 To do so we need toknow the evolution of the share of Paris estatesin top estates Between 1902 and 1994 avail-able data (broken down by departement) showsthat the evolution of top estate shares in Francewas parallel to that of top estate shares in ParisWealth inequality is always lower for the coun-try as a whole but the trends are similar (seeFigure 7) It is also striking to note that Parisrsquosshare of the top 1 percent of French estates hasremained fairly stable over the twentieth cen-tury (it fluctuates between 20 percent and 25percent with no trend) even though Parisrsquosshare of all decedents has been dwindling overtime reflecting the population decline of thecapital (see Table 3) In 1902 Paris decedentswere four times more likely to belong to thenational top 1 percent of estates than averagedecedents (26665 41) in 1994 Paris de-

FIGURE 7 WEALTH CONCENTRATION AT DEATH IN PARIS AND FRANCE 1807ndash1994

Source Authorsrsquo computations based on estate tax returns (see Tables 2 and 4)

246 THE AMERICAN ECONOMIC REVIEW MARCH 2006

cedents were seven times more likely to belongto the national top 1 percent of estates thanaverage decedents (25236 70) If anythingthe geographic concentration of fortunes waslarger at the end of the twentieth century than atthe beginning The decline of wealth concentra-tion that followed World War I was not due toredistribution between Paris and the provinces

How did the fraction of Paris estates in topestates evolve over the course of the nineteenthcentury Our estimates rely on a simple andvery conservative assumption from 1807 to1902 Parisrsquos share of estates in the top percen-tile increased at the same rate as Parisrsquos share ofFrench adult deaths More precisely let us de-note FPt(w) the cumulative distribution functionfor wealth-at-death in Paris in year t Ft(w) thecorresponding distribution for France nPt thetotal number of adult deaths in Paris in year tand nt the corresponding number for all ofFrance The ninetieth-percentile threshold P90Ptis defined by FPt(P90Pt) 09 the ninety-ninth-percentile threshold P99Pt is defined byFPt(P99Pt) 099 etc and similarly for theFrench thresholds P90t P99t etc We observeFPt(w) nPt and nt throughout the 1807ndash1994period but we do not observe Ft(w) until 1902(before this date we observe only national ag-

gregate average wealth wt Wtnt) To con-struct our benchmark estimates we assume thatthe shares s99t s995t s999t and s9999t of Parisestates in the national top 1 percent 05 percent01 percent and 001 percent of the nationalwealth-at-death distribution increased at thesame rate as nPtnt during the 1807ndash1902 period(see Table 3) Using this approximation and ourParis samples of individual tax returns we com-pute the threshold wealth levels for the toppercentiles of the national wealth distribution(eg P99 P995 P999 and P9999)20 Wealso calculate the average wealth levels for therelevant wealth classes (eg P99ndash995 P995ndash999 P999ndash9999 and P9999ndash100) usingPareto interpolation techniques These are thenweighted by the number of individuals inFrance in that wealth class in order to compute

20 For instance the number of decedents (aged 20 yearsand older) in France was 583976 in 1887 (see Piketty et al2004 Appendix Table A1) so that the top 1 percent of theestate distribution at death consists of the top 5840 estatesIf the share of Paris among French top-1-percent estates was241 percent in 1887 (see Table 3) then the national P99threshold for 1887 corresponds to the top 1410 Parisianestates (0241 5840 1410) (the national P99 thresholdfor 1887 reported in Piketty et al 2004 Appendix TableA3 was computed using this formula)

TABLE 3mdashTHE FRACTION OF PARIS ESTATES IN TOP ESTATES AT DEATH IN FRANCE 1807ndash1994

(1) Fraction of Parisdecedents in all

decedents 20-yr

(2) Fraction of Parisestates in top-10-

percent estates

(3) Fraction of Parisestates in top-1-percent estates

(4) Fraction of Parisestates in top-01-

percent estates

1807 25 101 2051817 25 103 2101827 28 116 2371837 31 126 2561847 36 146 2971857 36 146 2971867 49 199 4041877 51 211 4281887 59 241 4911902 65 75 266 5411913 65 75 255 5231929 58 83 239 5301938 53 74 216 4211947 55 110 198 3521956 55 128 223 3501994 36 89 252 352

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources) No datasource exists to compute columns (3)ndash(4) prior to 1902 and the numbers reported on this table for years 1807ndash1887 werecomputed assuming that the columns (3)ndash(4) followed the same trend as column (1) over the 1807ndash1902 period (see textSection III)

247VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

the average wealth levels for top fractiles (P99ndash100 P995ndash100 P999ndash100 and P9999ndash100)Lower thresholds of the national wealth distri-bution (P90 and P95) were computed using thenational TRA survey21 and the P90ndash95 andP95ndash99 intermediate wealth levels were alsocomputed using Pareto interpolation techniques

The national top estate shares estimates re-ported in Table 4 were computed using thismethodology They suggest that wealth concen-tration (as measured by the top-1-percent estateshare) rose throughout the nineteenth century inFrance during both the 1807ndash1867 and 1867ndash1902 periods although less sharply than in Parisduring the latter period (see Figure 7) Theseestimates are conservative in the sense that it isalmost certain that they underestimate the riseof wealth concentration that took place inFrance during the nineteenth century First weknow that the bulk of population growth in Parisduring the nineteenth century was due to theannexation of suburbs in 1860 and to populationgrowth in these peripheral arrondissements Be-

cause the outskirts of the city were poor theannexation added few top estates Thus there islittle doubt that Parisrsquos share of top estates inFrance actually increased less than its share of thetotal population This hypothesis is confirmed bynineteenth-century housing tax tabulations show-ing that the fraction of Paris taxpayers in thenational top 1 percent of taxpayers was substan-tially larger than 10 percent at the beginning of thenineteenth century22 Giving Paris a larger (andmore realistic) share of top estates in 1807 wouldboth reduce the share of wealth of the top 1percent in France at that date and lead to morerapid rise in inequality over time

Next and most importantly other estate sur-veys are consistent with the view that wealth in-equality was growing The important study byAdeline Daumard (1973) which relied on samplesof estate tax returns collected in Paris Lyon Tou-louse Lille and Bordeaux found that wealth con-centration increased in each of these five citiesduring the nineteenth century23 The TRA surveyalthough it is ill-suited for the study of top estatesis also consistent with our view Wealth dispersionwas on the rise in nineteenth-century France ac-cording to the TRA survey both in the sense thatthe fraction of decedents with positive estates de-clined over time (in spite of the sharp increase inthe value of the average estate) and that ratios suchas the P90P50 increased24 We also comparedour benchmark national P99 series extrapolatedfrom our Paris samples and the national P99series computed using the TRA survey We foundthat both series display the same overall upwardtrend in concentration (which is reassuring regard-ing the general validity of our Paris-France extrap-olation technique) except that the growth ofinequality from 1807 to 1902 in the TRA series is

21 See Bourdieu et al (2003) for full details about theTRA survey The P90 and P95 thresholds reported onPiketty et al (2004 Appendix Table A3) were computedusing ten-year moving averages around the target years inorder to make sure that each estimate was based on asufficient number of observations

22 These tabulations were published in the same FinanceMinistry official publications as the estate tabulations Wechose not to use them in our formal computations becausethe tax base of the housing tax (namely the rental value ofthe real estate property where the household lives) is onlyloosely connected to the estate tax base (in particular onecannot rule out the possibility that the housing tax baseoverrepresents Paris-based taxpayers)

23 Unfortunately Daumardrsquos samples are not available inmachine-readable format she has only two or three years ofdata for each city and she did not try to compute homog-enous inequality indicators (top fractiles shares etc) withher data Thus although her results and our work are con-sistent they cannot be compared directly

24 See Bourdieu et al (2004)

TABLE 4mdashWEALTH CONCENTRATION AT DEATH IN FRANCE1807ndash1994

Top-10-percentestate share

Top-1-percentestate share

Top-01-percentestate share

1807 791 434 1631817 810 445 1811827 824 452 1631837 796 438 1471847 816 479 1841857 829 495 1741867 810 480 1741877 838 471 2011887 839 487 1921902 839 516 2311913 863 549 2601929 820 502 2471938 776 420 1991947 699 299 1101956 694 304 1101994 610 213 63

Source Authorsrsquo computations using estate tax returns (seePiketty et al 2004 Table A3 for detailed series)

248 THE AMERICAN ECONOMIC REVIEW MARCH 2006

more severe than in our series (see Figure 8) Thisagain suggests that the latter provide a conserva-tive lower bound for the upward trend in wealthconcentration In any case the finding of a largeincrease in wealth inequality in nineteenth-centuryFrance (and up until World War I) appears to berobust25

As was mentioned earlier there exists nocomparable continuous data source covering thenineteenth and twentieth centuries in othercountries which makes it difficult to put ourFrench long-run series in international perspec-tive We note however that existing series forthe United States and the United Kingdom areconsistent with our findings on France26 Re-garding levels existing evidence suggests thatduring the nineteenth and most of the twentiethcenturies France was in an intermediate posi-tion in terms of wealth concentration in be-tween the United States (more equal) and theUnited Kingdom (the most unequal)27 These

25 Note that this continuous rise in wealth inequalitydoes not necessarily imply that a parallel rise occurredregarding income inequality Given that there exists nomicro source on incomes prior to the creation of the incometax in 1914 it is very difficult (if not impossible) to properlyaddress this issue Christian Morrisson and Wayne Snyder(2000) have attempted to link income inequality estimatesbased upon Old-Regime fiscal sources (pre-1789) withmodern income-tax-based twentieth-century estimates andthey have argued that income inequality might have startedto decline during the later part of the nineteenth century andon the eve of World War I (see also Morrisson 2000)Although our data do not allow us to rule out such apossibility we note that their nineteenth-century personaldistribution estimates are based on fragile macroeconomicdata on functional distribution and are not homogenous totheir eighteenth- and twentieth-century estimates Given theevidence that we provide on wealth inequality any signif-icant decline in aggregate income inequality would have tobe associated with severe compression of the wage distri-bution There is little research on this issue however

26 Lee Soltow and Jan L Van Zanden (1998) also find adecline in inequality in the twentieth century in the Neth-erlands Their data are consistent with a rise in inequality inthe nineteenth century but they have no direct evidenceabout its extent

27 According to our series the top-1-percent wealthshare in France rose from around 45 percent in 1800 toabout 55 percent around World War I and then fell to about20 percent by the end of the twentieth century (see Figure7) wealth concentration among the living appears to besomewhat larger (see Section V below) According to theseries pieced together by Lindert (2000 pp 181ndash82 and186) the UK top-1-percent wealth share rose from about

FIGURE 8 ESTIMATES OF THE P99 THRESHOLD FOR THE FRENCH DISTRIBUTION OF ESTATES AT DEATH EXTRAPOLATION

FROM PARIS SAMPLES VERSUS ESTIMATES FROM TRA SAMPLES (CURRENT FRENCH FRANCS)

Source Authorsrsquo computations based on estate tax returns (Paris samples and TRA samples)

249VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

differences in inequality largely hark back to dif-ferential concentration of landownership En-glandrsquos land was extremely narrowly held whilethe United States was most egalitarian The impactof access to real estate assets can also be seenwhen we contrast Paris (where it was extremelyconcentrated) with the provinces

Trends in inequality over time may be easierto compare because biases in source materialmay have a significant effect on levels ratherthan trends The historical pattern is similar ineach of the three economies In particular thereis evidence that wealth concentration increasedduring the nineteenth century in both Anglo-Saxon countries and declined during the twen-tieth century with a turning point around WorldWar I28 Neither exhibits patterns consistentwith a Kuznet process

IV From the Wealth of the Decedents to theWealth of the Living

The estimates reported thus far refer to in-equality among decedents as described in thetax returns filed by their heirs The evolution ofthe distribution of wealth among the livingmight however have followed a different pat-tern In order to convert wealth-at-death con-centration estimates into wealth-of-the-livingconcentration estimates it is standard to use theldquoestate multiplierrdquo method29 It consists ofweighting each observation of an estate at death

by the inverse of the mortality rate for this agegroup That is if the mortality rate for ages 20to 24 was 068 percent in Paris in 1902 theneach decedent aged 20 to 24 represented about147 living individuals of the same age (100068 147) Similarly if the mortality ratefor ages above 80 was 2143 percent in Paris in1902 then each decedent in that group repre-sented about 47 living individuals in the sameage group (102143 47) Applying thismethod requires mortality tables (these are eas-ily available) and estate tabulations brokendown by estate size and age at death (these arescarcer) Fortunately the cityrsquos statistical bu-reau published annual death-by-age totals theFrench censuses report the age distribution forthe capital every five years and we collectedage at death from the estate declarations Thesedata allowed us to compute estimates of wealthconcentration among the living over the 1807ndash1902 period using various assumptions aboutthe wealth profiles of mortality rates

The base population for the living is the set ofall individuals aged 20 and over living in Parisin year t which we denote pt The number ofliving individuals in age bracket a is denoted pta(a 20ndash24 25ndash29 30ndash34 75ndash79 80 andover) and the number of decedents in agebracket a is denoted nta The mortality rate forage bracket a is given by mta ntapta Webegin with a uniform-mortality benchmarkThese estimates are based on the simplifyingassumption that these mortality rates dependsolely on age and are the same for all wealthgroups (and in particular are the same forzero-wealth and positive-wealth individuals)We can then weigh each decedent with positiveestate and age a collected in the Paris archivesin year t by ptanta This allows us to computethe number of living Parisians with positivewealth in year t and also (by differentiatingwith pt) the number of living Parisians with zerowealth at year t which is used to weight zero-estate observations We then use our weighteddatasets to compute top estate fractiles amongthe living in Paris

The main conclusion is that the living expe-rienced the same upward trend in wealth con-centration as the decedents (see Figure 9) Wefind that inequality was significantly higheramong the living than among decedents be-cause survivors were on average younger than

55 percent in 1800 to 70 percent around World War I thenfell to about 20 percent in the 1990s The US top-1-percentwealth share rose from about 15 to 20 percent in 1800 toabout 40 percent around World War I then fell to about 30percent in the 1990s (and as low as 20 to 25 percentaccording to the more recent estimates from Kopczuk andSaez 2004) Wealth concentration is now larger in theUnited States than in European countries but the reversewas true during the nineteenth century up until World WarII (It is only since the 1950sndash1970s period that US wealthconcentration has been somewhat larger)

28 See Lindert (2000 pp 181ndash82 and 188)29 This method was widely used in England and France

in the late nineteenth and early twentieth centuries to com-pute the stock of national wealth on the basis of the flow ofwealth transmitted at death Standard references that use thistechnique to estimate the wealth distribution of the livingfrom estate tax data tabulated by estate size and age at deathinclude Atkinson and Harrison (1978) and Lampman(1962) For a more recent application of this technique tothe United States see Kopczuk and Saez (2004)

250 THE AMERICAN ECONOMIC REVIEW MARCH 2006

those who died and the young were on aver-age poorer In particular the estate multipliermethod leads to lower average weights forpositive-wealth decedents than for zero-wealthdecedents (the former are on average older andtherefore ldquorepresentrdquo a smaller number of livingindividuals) As a result the fraction of positive-wealth individuals is even smaller among theliving than among decedents Hence our bench-mark uniform-mortality estimates of wealthconcentration among the living are significantlylarger than corresponding estimates among de-cedents (eg top-1-percent wealth shares areabout 15 percent larger among the living)Changes over time however are similar In-creased life expectancy and declining mortalityrates over the course of the nineteenth centuryhave only a small effect on the trends

In order to make the estate multiplier methodmore reliable one would prefer to take intoaccount differential mortality by wealth Doingso would require having access to mortalityschedules based both on wealth and age at dif-ferent points in time unfortunately these are notavailable We have nonetheless reestimatedwealth of the living based upon the same as-sumption as Kopczuk and Saez (2004) That is

we assumed uniform mortality among the poor(here defined as zero-wealth individuals) andamong the rich (here defined as positive-wealthindividuals) and we assumed that the ratiomtaRmtaP between the mortality rate of the richand the mortality rate of the poor followed aU-shaped age profile from about 85 percent forthe young (ie the rich die 15 percent less oftenthan the poor when they are 20ndash24 or 25ndash29years old) down to about 70 percent for middle-age individuals in their forties to fifties and upto 100 percent for very old individuals in theireighties to nineties30 This profile correspondsto the best available estimates in the literatureand it appears to be relatively stable over timeand across developed countries In the absenceof better data it is the best one can do31 Thebenchmark differential-mortality estimates re-ported on Figure 9 show that although addingdifferential mortality produces different levelsof inequality it does not have much impact onthe upward trend in concentration

30 See Kopczuk and Saez (2004 Table A4)31 See Kopczuk and Saez (2004 Appendix B) for refer-

ences to the US and international literature devoted to theage-wealth profile of mortality rates

FIGURE 9 WEALTH CONCENTRATION AMONG DECEDENTS AND AMONG THE LIVING IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives (see Piketty et al 2004Table A4 for detailed series)

251VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

The resulting differential-mortality inequalityestimates lie between those based on decedentsonly and those using uniform mortality for theliving (see Figure 9) Note that moving fromuniform-mortality to differential-mortality es-tate multiplier methodology can either increaseor decrease inequality Here the reason whysuch a move leads to lower wealth concentra-tion seems to be due to the fact that differentialmortality tends to put higher weights onpositive-wealth decedents (for a given age)thereby increasing the estimated fraction of liv-ing individuals with positive wealth The im-portant point however is that the resulting leveleffects are relatively small in magnitude con-stant in time and dwarfed by the upward timetrend Even if we were to assume an enormousincrease in differential mortality during thenineteenth century in the sense that differentialmortality between the rich and poor was equalto 0 percent of the benchmark differential in1807 and 100 percent of the benchmark differ-ential in 1902 the resulting wealth concentra-tion estimates would still be significantly higherin 1902 than in 1807 (see Figure 9) Yet wehave no reason to believe that differential mor-tality increased to such an extent During thenineteenth century real wages for unskilledworkers rose which would have reduced mor-tality more for the poor than for the rich After1850 public health measures (sanitation watervaccination) were in place Again these wouldhave had a significant effect on the poor and themiddle class who could not privately purchasesuch health-improving services To be sure therich could avail themselves of more medicalservices than the poor or the middle class butthe impact of these services was probably small(bear in mind that neither antibiotics nor car-diovascular interventions were available)

Finally we have applied the estate multipliermethod to available data for 1947 and 1994 Over-all the sharp decline in wealth concentration ob-served during the twentieth century (andespecially between 1914 and 1945) is very robustIf anything the decline appears to be even largerwhen one looks at wealth concentration amongthe living rather than among decedents32

V The Changing Age Profile of Wealth

In the previous sections we focused almostexclusively on aggregate top wealth shares Ourdata however also detail the characteristics ofeach decedent in particular their gender andage The evolution of wealth by gender is ofrelevance for over the past two centuries therehave been massive changes in womenrsquos laborforce participation capacity to manage theirown affairs and life expectancy relative to menThe evolution of wealth by age is of relevancebecause there was a significant increase in adultlife expectancy over the twentieth century andbecause the progressive diffusion of pensionsmay have changed savings motivations More-over age-wealth profiles also inform us aboutthe motives of wealth accumulation and theeconomic impact of high wealth concentration

A first pass at the data considers the genderbreakdown of wealth at death Remarkably inour micro data the share of women in top estatestakes its highest value on the eve of World WarI For instance the womenrsquos share in the top 05percent rose from 35 percent prior to 1850 to 45percent in 1902 only to fall to 40 percent afterWorld War II Strikingly womenrsquos share ofwealth follows almost exactly the pattern ofaggregate inequality Women were relativelyricher when inequality reached its apex inFrance than at any other time Moreover insti-tutional variables seem to have played almostno role in changing the relative wealth ofwomen Unlike in common law countriesFrench law starting with the code civil of 1804required nearly equal treatment of all children inbequests Further research will help us deter-mine to what extent women of great wealthwere heirs or part of economically very success-ful couples

The data also reveal striking changes in the

32 See Piketty et al (2004 Table A4) It is unfortunatelynot possible to construct complete series for wealth concen-

tration among the living for the twentieth century due todata limitations tables broken down by estate brackets andage of decedents are available solely for years 1943ndash1954and at the national level (no table broken down by estatebrackets and age of decedents has ever been compiled at thedepartement level except in 1931 for Seine departementsee E S Danysz 1934) and the 1994 micro sample is notlarge enough to allow for a reliable application of the estatemultiplier method at the Paris level Thus the only wealth-of-the-living concentration estimates we provide for thetwentieth century are national estimates for 1947 and 1994

252 THE AMERICAN ECONOMIC REVIEW MARCH 2006

age profile of wealth between 1807 and 1994(see Table 5) During the nineteenth century aswealth concentration was increasing the veryrich were getting older and older At the begin-ning of the nineteenth century in the aftermathof the French Revolution the richest individualswere those in their fifties they were typically100 percent richer on average than people intheir forties 25 percent richer than those in theirsixties and 40 percent richer than those in theirseventies and eighties By the 1870s howeverthe age-wealth pattern had become stronglymonotonic the richest individuals were the old-est individuals In 1902 people in their sixtiesand seventies bequeathed 150 percent morethan those in their fifties and those in theireighties 300 percent more On the eve of WorldWar I top wealth holders were old and likely tobe retired This pattern breaks some time duringthe 1914ndash1945 period33 In 1947 as well as in1994 we are back to a pattern where the richestindividuals are those in their fifties Overall theperiod of maximal wealth inequality (1860ndash1913) also appears to be a period characterizedby a very specific age profile of wealth and largeconcentration of assets among the elderly

Another way to analyze the changing age-wealth relationship is to look at average age by

top estate fractile34 In 1817 average age wasvirtually the same for the top 10 percent and thetop 1 percent of estates (or even slightly declin-ing) The average-age-per-fractile relationshipbecomes upward sloping during the nineteenthcentury and by 1902 those in the top 1 percentwere almost six years older than those in the top10 percent The relationship is flat in 1947 anddownward-sloping in 1994 Finally one canapply the estate multiplier method (see SectionIV above) and analyze how wealth concentra-tion by age group among the living changedover the course of the nineteenth century Thegeneral population in Paris did not becomeolder during the nineteenth century those aged60 or older made up about 15 percent of thepopulation in 1817 and after 1847 about 10 to11 percent35 The share of total wealth ownedby the elderly rose significantly however aswealth distribution worsened The wealth be-longing to those aged 60 or more rose fromabout 25 to 30 percent of the total at the begin-ning of the nineteenth century to about 40 to 45percent by the end of the century The wealthshare of those aged 70 or older doubled fromless than 10 percent to about 20 percent36

33 Existing evidence on the age-wealth profile for 1931(see Danysz 1934) suggests that the Great Depression andWorld War II (rather than World War I) played the leadingroles in breaking this pattern This is an issue we plan toinvestigate in future research

34 See Piketty et al (2004 Table 6)35 Although life expectancy was increasing which

should have led to large shares of population for oldergroups the city was also growing quickly The large num-ber of immigrants (who were typically in their twenties)increased the relative size of the younger cohorts (seePiketty et al 2004 Table 7 and Figure 11)

36 See Piketty et al (2004 Figure 12)

TABLE 5mdashTHE AGE PROFILE OF WEALTH AT DEATH IN PARIS 1817ndash1994(Average estate left by 50- 59-year-old 100)

20ndash29yr-old

30ndash39yr-old

40ndash49yr-old

50ndash59yr-old

60ndash69yr-old

70ndash79yr-old

80ndash89yr-old

90ndash99yr-old

1817 26 22 28 100 54 59 591827 44 50 53 100 88 87 601837 133 90 107 100 116 123 1101847 87 73 102 100 117 204 1321857 84 77 101 100 104 109 1451867 67 58 136 100 141 125 1541877 66 73 63 100 197 260 4301887 45 33 63 100 152 233 2951902 29 40 80 100 253 272 4011947 31 51 73 100 113 105 105 1091994 11 45 100 87 93 95 68

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources)

253VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

It is perhaps not surprising that inequalitybecame strongly correlated with age in the1860s Those who died at an old age in the1820s and 1830s had lived through the difficultyears of the French Revolution and the disloca-tion of the end of the Napoleonic period Theirability to accumulate wealth had been severelyimpaired Furthermore they were rather lesslikely to inherit much wealth since the Revo-lution wiped out the bond portfolios of theirparents through a prolonged period of high in-flation Those who died from the 1860s to theearly 1910s did not suffer from the adversitiesthat plagued their forebears Instead they en-joyed the fruits of the financial sector expansionthat began in the 1850s After 1947 we seem tohave returned to a situation quite like that of1817 Presumably the capital damages associ-ated with both world wars and the Great De-pression had a strong negative effect on thewealth holdings of older generations The per-sistence of a flat age-wealth profile until 1994 islikely to be associated with two factors First insocieties where income growth is rapid abso-lute wealth accumulation is faster by youngercohorts than by older ones because their in-comes are higher at every age This is an im-portant distinction between the nineteenthcentury and the twentieth century Furthermorehighly progressive rates of income and estatetaxation have probably made it more difficult toaccumulate large fortunes thereby flattening theobserved age-wealth profile

The more interesting (and more difficult)question relates to the possible efficiency im-pact of high wealth concentration and changingage-wealth profiles Although our data do notallow us to address efficiency issues in a rigor-ous way our results allow us to formulate anumber of hypotheses and to shed new light onthe ongoing debate on inequality and growth37

From a theoretical viewpoint whether highwealth concentration can have a negativegrowth impact depends critically on the exis-tence of credit constraints With first-best credit

markets money flows toward the best entrepre-neurs and investment projects irrespective ofthe initial distribution High levels of wealthconcentration can be bad from a social justiceviewpoint but they entail no efficiency lossWhen credit constraints bind however initialwealth matters and high levels of inequality canhurt growth Whether the loss is large or smalldepends on who owns the assets If the rich areefficient investors (they know which projects tofund etc) then wealth concentration may evenbe useful If the rich are retired rentiers how-ever investing their wealth in low-yield assets(or low-ability inheritors) then high wealthconcentration and credit constraints might pre-vent talented but penniless investors from un-dertaking efficient projects thereby entailingnegative growth consequences The data used inthis paper are not ideal to address whether creditconstraints were important in a country likeFrance at the end of the nineteenth century Ourresults suggest however that to the extentcredit constraints were indeed severe highwealth concentration did have a negativegrowth impact In order to investigate this hy-pothesis further one would need to gather moresystematic data on investment strategies andasset returns Preliminary evidence suggeststhat the rich elderly of the 1860ndash1913 perioddid indeed hold a disproportionate fraction oftheir wealth in low-yield assets (such as gov-ernment bonds) An alternative hypothesishowever is that steeper age-wealth profileswere the consequence of the growth of financialmarkets as their children faced fewer creditconstraints parents decided to hold on to moreof their wealth

VI Conclusion

Evidence from wealth at death in Paris and inFrance over the last two centuries reveals threekey patterns First wealth concentration haschanged dramatically over time In 1807 thetop-1-percent share of wealth (40 percent inFrance 50 percent in Paris) was twice as high asit would be in 1994 but substantially less thanin 1913 when it peaked above 55 percent inFrance and 70 percent in Paris Some of thesechanges were due to economic phenomena thathave long been emphasized as creating inequal-ity namely industrialization and financial cen-

37 Thus far this literature has concentrated upon cross-country regressions of inequality on growth a methodologythat raises serious identification problems especially giventhe low quality of available international datasets on in-equality which are neither long-run nor homogeneous (seeeg Atkinson and Andrea Brandolini 2001)

254 THE AMERICAN ECONOMIC REVIEW MARCH 2006

tralization Yet the decline comes largely fromadverse shocks rather than economic conver-gence These changes are of such magnitudethat they are not sensitive to whether one ex-amines wealth at death in Paris or in France orwhether one examines it directly rather thanconverting it to wealth of the living by an estatemultiplier method

Our second key result is that there was asignificant transition during the nineteenth cen-tury from an important role for real estate as aform of wealth to moveable assets as the keyform of wealth for the very rich Similarly theshare of wealth held by aristocrats first rose andthen was eclipsed by that of financiers and in-dustrialists in the second half of the nineteenthcentury Hence mobility within this highly un-equal society might have been quite high Yetthis conjecture is tempered by our third findingthe wealthy were getting older over time andolder relative to less wealthy decedents Suchaging among the very wealthy would have hadnegative consequences for growth if financialmarkets were imperfect This issue requires fur-ther investigation and we hope it will attractfuture research

REFERENCES

Atkinson Anthony B ldquoTop Incomes in the UKover the 20th Centuryrdquo Journal of the RoyalStatistical Society Series A (Statistics in So-ciety) 2005 168(2) pp 325ndash43

Atkinson Anthony B and Brandolini AndrealdquoPromise and Pitfalls in the Use of lsquoSecond-aryrsquo Data-Sets Income Inequality in OECDCountries as a Case Studyrdquo Journal of Eco-nomic Literature 2001 39(3) pp 771ndash99

Atkinson Anthony B and Harrison Alan J Dis-tribution of personal wealth in Britain Cam-bridge Cambridge University Press 1978

Atkinson Anthony B and Piketty Thomas Topincomes over the twentieth century A con-trast between European and English speak-ing countries Oxford Oxford UniversityPress (forthcoming)

Banerjee Abhijit V and Duflo Esther ldquoInequalityand Growth What Can the Data Sayrdquo Journalof Economic Growth 2003 8(3) pp 267ndash99

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoPourquoi la richesse ne

srsquoest-elle pas diffusee avec la croissance Ledegre zero de lrsquoinegalite et son evolution enFrance 1800ndash1940rdquo Histoire et mesure 200318(1ndash2) pp 147ndash98

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoDefense et illustration delrsquoenquete des 3000 familles Lrsquoexemple deson volet patrimonialrdquo Annales de demogra-phie historique 2004 19 pp 19ndash57

Danysz E S ldquoContribution a lrsquoetude des for-tunes privees drsquoapres les declarations de suc-cessionsrdquo Bulletin de la statistique generalede France 1934 XXIV (1) pp 111ndash171

Daumard Adeline Les fortunes francaise auXIXe siecle Enquete sur la repartition et lacomposition des capitaux prives a ParisLyon Lille Bordeaux et Toulouse drsquoapreslrsquoenregistrement des declarations de succes-sion Paris Mouton 1973

Gain Andre La restauration et les biens desemigres 2 vols Nancy Societe drsquoimpressiontypographique 1929

Jones Alice H American colonial wealth Doc-uments and methods 3 vols New York ArnoPress 1977

Kopczuk Wojciech and Saez Emmanuel ldquoTopWealth Shares in the United States 1916ndash2000 Evidence from Estate Tax ReturnsrdquoNational Tax Journal 2004 57(2) pp 445ndash87

Kuznets Simon ldquoEconomic Growth and IncomeInequalityrdquo American Economic Review1955 45(1) pp 1ndash28

Lampman Robert J The share of top wealth-holders in national wealth 1922ndash1956Princeton Princeton University Press 1962

Lescure Michel Les banques lrsquoEtat et lemarche immobilier en France a lrsquoepoquecontemporaine 1820ndash1940 Paris EditionslrsquoEcole des Hautes Etudes en Sciences Socia-les 1982

Levy-Leboyer Maurice and Bourguignon Fran-cois LrsquoEconomie francaise au XIXe siecleAnalyse macroeconomique Paris Eco-nomica 1985 English translation Cam-bridge Cambridge University Press 1990

Lindert Peter H ldquoUnequal English Wealthsince 1670rdquo Journal of Political Economy1986 94(6) pp 1127ndash62

Lindert Peter H ldquoThree Centuries of Inequalityin Britain and Americardquo in Anthony B At-kinson and Francois Bourguignon eds

255VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

FIGURE 5 WEALTH COMPOSITION AT DEATH IN PARIS AND FRANCE 1807ndash1902(Share of personal (non-real) estate in total estate)

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives and national aggregateestate statistics compiled by the French tax administration

FIGURE 6 ARISTOCRATIC ESTATES AT DEATH IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives

245VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

tance declined steadily To be sure aristocratsremain overrepresented throughout the periodincluding in 1902 (about 13 percent of nobles inthe top 1 percent of estates over 25 percent inthe top 01 percent versus less than 1 percent inthe population as a whole) The inverted-U pat-tern is yet another of the Revolutionrsquos legaciesIn 1807 when we first observe it aristocraticwealth was at a temporary nadir On the onehand the nobility was impoverished by theRevolutionrsquos inflation and by the sharp declineof the value of Parisian real estate On the otherhand part of the Old Regime nobility was inexile and thus if they died we do not observetheir moveable wealth Aristocrats were able torecoup part of their losses during the first half ofthe nineteenth century Napoleon providedsome assistance by conferring titles of nobilityon his chief military officers and endowingthem with wealth Later the Restoration gov-ernment (1815ndash1830) compensated individualswho fled abroad during the Revolution for thelosses they suffered when their property wasconfiscated The government distributed nearlyone billion francs in the famous milliard desemigres (Andre Gain 1929) The beneficiariesof Napoleonrsquos and the Restorationrsquos largess ap-

pear among the very rich until mid-centuryPresumably such redistribution did not contrib-ute to accelerate French industrialization

III From Paris to France

We can use the Paris data to construct wealthconcentration at death estimates for all ofFrance from 1807 to 1902 To do so we need toknow the evolution of the share of Paris estatesin top estates Between 1902 and 1994 avail-able data (broken down by departement) showsthat the evolution of top estate shares in Francewas parallel to that of top estate shares in ParisWealth inequality is always lower for the coun-try as a whole but the trends are similar (seeFigure 7) It is also striking to note that Parisrsquosshare of the top 1 percent of French estates hasremained fairly stable over the twentieth cen-tury (it fluctuates between 20 percent and 25percent with no trend) even though Parisrsquosshare of all decedents has been dwindling overtime reflecting the population decline of thecapital (see Table 3) In 1902 Paris decedentswere four times more likely to belong to thenational top 1 percent of estates than averagedecedents (26665 41) in 1994 Paris de-

FIGURE 7 WEALTH CONCENTRATION AT DEATH IN PARIS AND FRANCE 1807ndash1994

Source Authorsrsquo computations based on estate tax returns (see Tables 2 and 4)

246 THE AMERICAN ECONOMIC REVIEW MARCH 2006

cedents were seven times more likely to belongto the national top 1 percent of estates thanaverage decedents (25236 70) If anythingthe geographic concentration of fortunes waslarger at the end of the twentieth century than atthe beginning The decline of wealth concentra-tion that followed World War I was not due toredistribution between Paris and the provinces

How did the fraction of Paris estates in topestates evolve over the course of the nineteenthcentury Our estimates rely on a simple andvery conservative assumption from 1807 to1902 Parisrsquos share of estates in the top percen-tile increased at the same rate as Parisrsquos share ofFrench adult deaths More precisely let us de-note FPt(w) the cumulative distribution functionfor wealth-at-death in Paris in year t Ft(w) thecorresponding distribution for France nPt thetotal number of adult deaths in Paris in year tand nt the corresponding number for all ofFrance The ninetieth-percentile threshold P90Ptis defined by FPt(P90Pt) 09 the ninety-ninth-percentile threshold P99Pt is defined byFPt(P99Pt) 099 etc and similarly for theFrench thresholds P90t P99t etc We observeFPt(w) nPt and nt throughout the 1807ndash1994period but we do not observe Ft(w) until 1902(before this date we observe only national ag-

gregate average wealth wt Wtnt) To con-struct our benchmark estimates we assume thatthe shares s99t s995t s999t and s9999t of Parisestates in the national top 1 percent 05 percent01 percent and 001 percent of the nationalwealth-at-death distribution increased at thesame rate as nPtnt during the 1807ndash1902 period(see Table 3) Using this approximation and ourParis samples of individual tax returns we com-pute the threshold wealth levels for the toppercentiles of the national wealth distribution(eg P99 P995 P999 and P9999)20 Wealso calculate the average wealth levels for therelevant wealth classes (eg P99ndash995 P995ndash999 P999ndash9999 and P9999ndash100) usingPareto interpolation techniques These are thenweighted by the number of individuals inFrance in that wealth class in order to compute

20 For instance the number of decedents (aged 20 yearsand older) in France was 583976 in 1887 (see Piketty et al2004 Appendix Table A1) so that the top 1 percent of theestate distribution at death consists of the top 5840 estatesIf the share of Paris among French top-1-percent estates was241 percent in 1887 (see Table 3) then the national P99threshold for 1887 corresponds to the top 1410 Parisianestates (0241 5840 1410) (the national P99 thresholdfor 1887 reported in Piketty et al 2004 Appendix TableA3 was computed using this formula)

TABLE 3mdashTHE FRACTION OF PARIS ESTATES IN TOP ESTATES AT DEATH IN FRANCE 1807ndash1994

(1) Fraction of Parisdecedents in all

decedents 20-yr

(2) Fraction of Parisestates in top-10-

percent estates

(3) Fraction of Parisestates in top-1-percent estates

(4) Fraction of Parisestates in top-01-

percent estates

1807 25 101 2051817 25 103 2101827 28 116 2371837 31 126 2561847 36 146 2971857 36 146 2971867 49 199 4041877 51 211 4281887 59 241 4911902 65 75 266 5411913 65 75 255 5231929 58 83 239 5301938 53 74 216 4211947 55 110 198 3521956 55 128 223 3501994 36 89 252 352

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources) No datasource exists to compute columns (3)ndash(4) prior to 1902 and the numbers reported on this table for years 1807ndash1887 werecomputed assuming that the columns (3)ndash(4) followed the same trend as column (1) over the 1807ndash1902 period (see textSection III)

247VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

the average wealth levels for top fractiles (P99ndash100 P995ndash100 P999ndash100 and P9999ndash100)Lower thresholds of the national wealth distri-bution (P90 and P95) were computed using thenational TRA survey21 and the P90ndash95 andP95ndash99 intermediate wealth levels were alsocomputed using Pareto interpolation techniques

The national top estate shares estimates re-ported in Table 4 were computed using thismethodology They suggest that wealth concen-tration (as measured by the top-1-percent estateshare) rose throughout the nineteenth century inFrance during both the 1807ndash1867 and 1867ndash1902 periods although less sharply than in Parisduring the latter period (see Figure 7) Theseestimates are conservative in the sense that it isalmost certain that they underestimate the riseof wealth concentration that took place inFrance during the nineteenth century First weknow that the bulk of population growth in Parisduring the nineteenth century was due to theannexation of suburbs in 1860 and to populationgrowth in these peripheral arrondissements Be-

cause the outskirts of the city were poor theannexation added few top estates Thus there islittle doubt that Parisrsquos share of top estates inFrance actually increased less than its share of thetotal population This hypothesis is confirmed bynineteenth-century housing tax tabulations show-ing that the fraction of Paris taxpayers in thenational top 1 percent of taxpayers was substan-tially larger than 10 percent at the beginning of thenineteenth century22 Giving Paris a larger (andmore realistic) share of top estates in 1807 wouldboth reduce the share of wealth of the top 1percent in France at that date and lead to morerapid rise in inequality over time

Next and most importantly other estate sur-veys are consistent with the view that wealth in-equality was growing The important study byAdeline Daumard (1973) which relied on samplesof estate tax returns collected in Paris Lyon Tou-louse Lille and Bordeaux found that wealth con-centration increased in each of these five citiesduring the nineteenth century23 The TRA surveyalthough it is ill-suited for the study of top estatesis also consistent with our view Wealth dispersionwas on the rise in nineteenth-century France ac-cording to the TRA survey both in the sense thatthe fraction of decedents with positive estates de-clined over time (in spite of the sharp increase inthe value of the average estate) and that ratios suchas the P90P50 increased24 We also comparedour benchmark national P99 series extrapolatedfrom our Paris samples and the national P99series computed using the TRA survey We foundthat both series display the same overall upwardtrend in concentration (which is reassuring regard-ing the general validity of our Paris-France extrap-olation technique) except that the growth ofinequality from 1807 to 1902 in the TRA series is

21 See Bourdieu et al (2003) for full details about theTRA survey The P90 and P95 thresholds reported onPiketty et al (2004 Appendix Table A3) were computedusing ten-year moving averages around the target years inorder to make sure that each estimate was based on asufficient number of observations

22 These tabulations were published in the same FinanceMinistry official publications as the estate tabulations Wechose not to use them in our formal computations becausethe tax base of the housing tax (namely the rental value ofthe real estate property where the household lives) is onlyloosely connected to the estate tax base (in particular onecannot rule out the possibility that the housing tax baseoverrepresents Paris-based taxpayers)

23 Unfortunately Daumardrsquos samples are not available inmachine-readable format she has only two or three years ofdata for each city and she did not try to compute homog-enous inequality indicators (top fractiles shares etc) withher data Thus although her results and our work are con-sistent they cannot be compared directly

24 See Bourdieu et al (2004)

TABLE 4mdashWEALTH CONCENTRATION AT DEATH IN FRANCE1807ndash1994

Top-10-percentestate share

Top-1-percentestate share

Top-01-percentestate share

1807 791 434 1631817 810 445 1811827 824 452 1631837 796 438 1471847 816 479 1841857 829 495 1741867 810 480 1741877 838 471 2011887 839 487 1921902 839 516 2311913 863 549 2601929 820 502 2471938 776 420 1991947 699 299 1101956 694 304 1101994 610 213 63

Source Authorsrsquo computations using estate tax returns (seePiketty et al 2004 Table A3 for detailed series)

248 THE AMERICAN ECONOMIC REVIEW MARCH 2006

more severe than in our series (see Figure 8) Thisagain suggests that the latter provide a conserva-tive lower bound for the upward trend in wealthconcentration In any case the finding of a largeincrease in wealth inequality in nineteenth-centuryFrance (and up until World War I) appears to berobust25

As was mentioned earlier there exists nocomparable continuous data source covering thenineteenth and twentieth centuries in othercountries which makes it difficult to put ourFrench long-run series in international perspec-tive We note however that existing series forthe United States and the United Kingdom areconsistent with our findings on France26 Re-garding levels existing evidence suggests thatduring the nineteenth and most of the twentiethcenturies France was in an intermediate posi-tion in terms of wealth concentration in be-tween the United States (more equal) and theUnited Kingdom (the most unequal)27 These

25 Note that this continuous rise in wealth inequalitydoes not necessarily imply that a parallel rise occurredregarding income inequality Given that there exists nomicro source on incomes prior to the creation of the incometax in 1914 it is very difficult (if not impossible) to properlyaddress this issue Christian Morrisson and Wayne Snyder(2000) have attempted to link income inequality estimatesbased upon Old-Regime fiscal sources (pre-1789) withmodern income-tax-based twentieth-century estimates andthey have argued that income inequality might have startedto decline during the later part of the nineteenth century andon the eve of World War I (see also Morrisson 2000)Although our data do not allow us to rule out such apossibility we note that their nineteenth-century personaldistribution estimates are based on fragile macroeconomicdata on functional distribution and are not homogenous totheir eighteenth- and twentieth-century estimates Given theevidence that we provide on wealth inequality any signif-icant decline in aggregate income inequality would have tobe associated with severe compression of the wage distri-bution There is little research on this issue however

26 Lee Soltow and Jan L Van Zanden (1998) also find adecline in inequality in the twentieth century in the Neth-erlands Their data are consistent with a rise in inequality inthe nineteenth century but they have no direct evidenceabout its extent

27 According to our series the top-1-percent wealthshare in France rose from around 45 percent in 1800 toabout 55 percent around World War I and then fell to about20 percent by the end of the twentieth century (see Figure7) wealth concentration among the living appears to besomewhat larger (see Section V below) According to theseries pieced together by Lindert (2000 pp 181ndash82 and186) the UK top-1-percent wealth share rose from about

FIGURE 8 ESTIMATES OF THE P99 THRESHOLD FOR THE FRENCH DISTRIBUTION OF ESTATES AT DEATH EXTRAPOLATION

FROM PARIS SAMPLES VERSUS ESTIMATES FROM TRA SAMPLES (CURRENT FRENCH FRANCS)

Source Authorsrsquo computations based on estate tax returns (Paris samples and TRA samples)

249VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

differences in inequality largely hark back to dif-ferential concentration of landownership En-glandrsquos land was extremely narrowly held whilethe United States was most egalitarian The impactof access to real estate assets can also be seenwhen we contrast Paris (where it was extremelyconcentrated) with the provinces

Trends in inequality over time may be easierto compare because biases in source materialmay have a significant effect on levels ratherthan trends The historical pattern is similar ineach of the three economies In particular thereis evidence that wealth concentration increasedduring the nineteenth century in both Anglo-Saxon countries and declined during the twen-tieth century with a turning point around WorldWar I28 Neither exhibits patterns consistentwith a Kuznet process

IV From the Wealth of the Decedents to theWealth of the Living

The estimates reported thus far refer to in-equality among decedents as described in thetax returns filed by their heirs The evolution ofthe distribution of wealth among the livingmight however have followed a different pat-tern In order to convert wealth-at-death con-centration estimates into wealth-of-the-livingconcentration estimates it is standard to use theldquoestate multiplierrdquo method29 It consists ofweighting each observation of an estate at death

by the inverse of the mortality rate for this agegroup That is if the mortality rate for ages 20to 24 was 068 percent in Paris in 1902 theneach decedent aged 20 to 24 represented about147 living individuals of the same age (100068 147) Similarly if the mortality ratefor ages above 80 was 2143 percent in Paris in1902 then each decedent in that group repre-sented about 47 living individuals in the sameage group (102143 47) Applying thismethod requires mortality tables (these are eas-ily available) and estate tabulations brokendown by estate size and age at death (these arescarcer) Fortunately the cityrsquos statistical bu-reau published annual death-by-age totals theFrench censuses report the age distribution forthe capital every five years and we collectedage at death from the estate declarations Thesedata allowed us to compute estimates of wealthconcentration among the living over the 1807ndash1902 period using various assumptions aboutthe wealth profiles of mortality rates

The base population for the living is the set ofall individuals aged 20 and over living in Parisin year t which we denote pt The number ofliving individuals in age bracket a is denoted pta(a 20ndash24 25ndash29 30ndash34 75ndash79 80 andover) and the number of decedents in agebracket a is denoted nta The mortality rate forage bracket a is given by mta ntapta Webegin with a uniform-mortality benchmarkThese estimates are based on the simplifyingassumption that these mortality rates dependsolely on age and are the same for all wealthgroups (and in particular are the same forzero-wealth and positive-wealth individuals)We can then weigh each decedent with positiveestate and age a collected in the Paris archivesin year t by ptanta This allows us to computethe number of living Parisians with positivewealth in year t and also (by differentiatingwith pt) the number of living Parisians with zerowealth at year t which is used to weight zero-estate observations We then use our weighteddatasets to compute top estate fractiles amongthe living in Paris

The main conclusion is that the living expe-rienced the same upward trend in wealth con-centration as the decedents (see Figure 9) Wefind that inequality was significantly higheramong the living than among decedents be-cause survivors were on average younger than

55 percent in 1800 to 70 percent around World War I thenfell to about 20 percent in the 1990s The US top-1-percentwealth share rose from about 15 to 20 percent in 1800 toabout 40 percent around World War I then fell to about 30percent in the 1990s (and as low as 20 to 25 percentaccording to the more recent estimates from Kopczuk andSaez 2004) Wealth concentration is now larger in theUnited States than in European countries but the reversewas true during the nineteenth century up until World WarII (It is only since the 1950sndash1970s period that US wealthconcentration has been somewhat larger)

28 See Lindert (2000 pp 181ndash82 and 188)29 This method was widely used in England and France

in the late nineteenth and early twentieth centuries to com-pute the stock of national wealth on the basis of the flow ofwealth transmitted at death Standard references that use thistechnique to estimate the wealth distribution of the livingfrom estate tax data tabulated by estate size and age at deathinclude Atkinson and Harrison (1978) and Lampman(1962) For a more recent application of this technique tothe United States see Kopczuk and Saez (2004)

250 THE AMERICAN ECONOMIC REVIEW MARCH 2006

those who died and the young were on aver-age poorer In particular the estate multipliermethod leads to lower average weights forpositive-wealth decedents than for zero-wealthdecedents (the former are on average older andtherefore ldquorepresentrdquo a smaller number of livingindividuals) As a result the fraction of positive-wealth individuals is even smaller among theliving than among decedents Hence our bench-mark uniform-mortality estimates of wealthconcentration among the living are significantlylarger than corresponding estimates among de-cedents (eg top-1-percent wealth shares areabout 15 percent larger among the living)Changes over time however are similar In-creased life expectancy and declining mortalityrates over the course of the nineteenth centuryhave only a small effect on the trends

In order to make the estate multiplier methodmore reliable one would prefer to take intoaccount differential mortality by wealth Doingso would require having access to mortalityschedules based both on wealth and age at dif-ferent points in time unfortunately these are notavailable We have nonetheless reestimatedwealth of the living based upon the same as-sumption as Kopczuk and Saez (2004) That is

we assumed uniform mortality among the poor(here defined as zero-wealth individuals) andamong the rich (here defined as positive-wealthindividuals) and we assumed that the ratiomtaRmtaP between the mortality rate of the richand the mortality rate of the poor followed aU-shaped age profile from about 85 percent forthe young (ie the rich die 15 percent less oftenthan the poor when they are 20ndash24 or 25ndash29years old) down to about 70 percent for middle-age individuals in their forties to fifties and upto 100 percent for very old individuals in theireighties to nineties30 This profile correspondsto the best available estimates in the literatureand it appears to be relatively stable over timeand across developed countries In the absenceof better data it is the best one can do31 Thebenchmark differential-mortality estimates re-ported on Figure 9 show that although addingdifferential mortality produces different levelsof inequality it does not have much impact onthe upward trend in concentration

30 See Kopczuk and Saez (2004 Table A4)31 See Kopczuk and Saez (2004 Appendix B) for refer-

ences to the US and international literature devoted to theage-wealth profile of mortality rates

FIGURE 9 WEALTH CONCENTRATION AMONG DECEDENTS AND AMONG THE LIVING IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives (see Piketty et al 2004Table A4 for detailed series)

251VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

The resulting differential-mortality inequalityestimates lie between those based on decedentsonly and those using uniform mortality for theliving (see Figure 9) Note that moving fromuniform-mortality to differential-mortality es-tate multiplier methodology can either increaseor decrease inequality Here the reason whysuch a move leads to lower wealth concentra-tion seems to be due to the fact that differentialmortality tends to put higher weights onpositive-wealth decedents (for a given age)thereby increasing the estimated fraction of liv-ing individuals with positive wealth The im-portant point however is that the resulting leveleffects are relatively small in magnitude con-stant in time and dwarfed by the upward timetrend Even if we were to assume an enormousincrease in differential mortality during thenineteenth century in the sense that differentialmortality between the rich and poor was equalto 0 percent of the benchmark differential in1807 and 100 percent of the benchmark differ-ential in 1902 the resulting wealth concentra-tion estimates would still be significantly higherin 1902 than in 1807 (see Figure 9) Yet wehave no reason to believe that differential mor-tality increased to such an extent During thenineteenth century real wages for unskilledworkers rose which would have reduced mor-tality more for the poor than for the rich After1850 public health measures (sanitation watervaccination) were in place Again these wouldhave had a significant effect on the poor and themiddle class who could not privately purchasesuch health-improving services To be sure therich could avail themselves of more medicalservices than the poor or the middle class butthe impact of these services was probably small(bear in mind that neither antibiotics nor car-diovascular interventions were available)

Finally we have applied the estate multipliermethod to available data for 1947 and 1994 Over-all the sharp decline in wealth concentration ob-served during the twentieth century (andespecially between 1914 and 1945) is very robustIf anything the decline appears to be even largerwhen one looks at wealth concentration amongthe living rather than among decedents32

V The Changing Age Profile of Wealth

In the previous sections we focused almostexclusively on aggregate top wealth shares Ourdata however also detail the characteristics ofeach decedent in particular their gender andage The evolution of wealth by gender is ofrelevance for over the past two centuries therehave been massive changes in womenrsquos laborforce participation capacity to manage theirown affairs and life expectancy relative to menThe evolution of wealth by age is of relevancebecause there was a significant increase in adultlife expectancy over the twentieth century andbecause the progressive diffusion of pensionsmay have changed savings motivations More-over age-wealth profiles also inform us aboutthe motives of wealth accumulation and theeconomic impact of high wealth concentration

A first pass at the data considers the genderbreakdown of wealth at death Remarkably inour micro data the share of women in top estatestakes its highest value on the eve of World WarI For instance the womenrsquos share in the top 05percent rose from 35 percent prior to 1850 to 45percent in 1902 only to fall to 40 percent afterWorld War II Strikingly womenrsquos share ofwealth follows almost exactly the pattern ofaggregate inequality Women were relativelyricher when inequality reached its apex inFrance than at any other time Moreover insti-tutional variables seem to have played almostno role in changing the relative wealth ofwomen Unlike in common law countriesFrench law starting with the code civil of 1804required nearly equal treatment of all children inbequests Further research will help us deter-mine to what extent women of great wealthwere heirs or part of economically very success-ful couples

The data also reveal striking changes in the

32 See Piketty et al (2004 Table A4) It is unfortunatelynot possible to construct complete series for wealth concen-

tration among the living for the twentieth century due todata limitations tables broken down by estate brackets andage of decedents are available solely for years 1943ndash1954and at the national level (no table broken down by estatebrackets and age of decedents has ever been compiled at thedepartement level except in 1931 for Seine departementsee E S Danysz 1934) and the 1994 micro sample is notlarge enough to allow for a reliable application of the estatemultiplier method at the Paris level Thus the only wealth-of-the-living concentration estimates we provide for thetwentieth century are national estimates for 1947 and 1994

252 THE AMERICAN ECONOMIC REVIEW MARCH 2006

age profile of wealth between 1807 and 1994(see Table 5) During the nineteenth century aswealth concentration was increasing the veryrich were getting older and older At the begin-ning of the nineteenth century in the aftermathof the French Revolution the richest individualswere those in their fifties they were typically100 percent richer on average than people intheir forties 25 percent richer than those in theirsixties and 40 percent richer than those in theirseventies and eighties By the 1870s howeverthe age-wealth pattern had become stronglymonotonic the richest individuals were the old-est individuals In 1902 people in their sixtiesand seventies bequeathed 150 percent morethan those in their fifties and those in theireighties 300 percent more On the eve of WorldWar I top wealth holders were old and likely tobe retired This pattern breaks some time duringthe 1914ndash1945 period33 In 1947 as well as in1994 we are back to a pattern where the richestindividuals are those in their fifties Overall theperiod of maximal wealth inequality (1860ndash1913) also appears to be a period characterizedby a very specific age profile of wealth and largeconcentration of assets among the elderly

Another way to analyze the changing age-wealth relationship is to look at average age by

top estate fractile34 In 1817 average age wasvirtually the same for the top 10 percent and thetop 1 percent of estates (or even slightly declin-ing) The average-age-per-fractile relationshipbecomes upward sloping during the nineteenthcentury and by 1902 those in the top 1 percentwere almost six years older than those in the top10 percent The relationship is flat in 1947 anddownward-sloping in 1994 Finally one canapply the estate multiplier method (see SectionIV above) and analyze how wealth concentra-tion by age group among the living changedover the course of the nineteenth century Thegeneral population in Paris did not becomeolder during the nineteenth century those aged60 or older made up about 15 percent of thepopulation in 1817 and after 1847 about 10 to11 percent35 The share of total wealth ownedby the elderly rose significantly however aswealth distribution worsened The wealth be-longing to those aged 60 or more rose fromabout 25 to 30 percent of the total at the begin-ning of the nineteenth century to about 40 to 45percent by the end of the century The wealthshare of those aged 70 or older doubled fromless than 10 percent to about 20 percent36

33 Existing evidence on the age-wealth profile for 1931(see Danysz 1934) suggests that the Great Depression andWorld War II (rather than World War I) played the leadingroles in breaking this pattern This is an issue we plan toinvestigate in future research

34 See Piketty et al (2004 Table 6)35 Although life expectancy was increasing which

should have led to large shares of population for oldergroups the city was also growing quickly The large num-ber of immigrants (who were typically in their twenties)increased the relative size of the younger cohorts (seePiketty et al 2004 Table 7 and Figure 11)

36 See Piketty et al (2004 Figure 12)

TABLE 5mdashTHE AGE PROFILE OF WEALTH AT DEATH IN PARIS 1817ndash1994(Average estate left by 50- 59-year-old 100)

20ndash29yr-old

30ndash39yr-old

40ndash49yr-old

50ndash59yr-old

60ndash69yr-old

70ndash79yr-old

80ndash89yr-old

90ndash99yr-old

1817 26 22 28 100 54 59 591827 44 50 53 100 88 87 601837 133 90 107 100 116 123 1101847 87 73 102 100 117 204 1321857 84 77 101 100 104 109 1451867 67 58 136 100 141 125 1541877 66 73 63 100 197 260 4301887 45 33 63 100 152 233 2951902 29 40 80 100 253 272 4011947 31 51 73 100 113 105 105 1091994 11 45 100 87 93 95 68

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources)

253VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

It is perhaps not surprising that inequalitybecame strongly correlated with age in the1860s Those who died at an old age in the1820s and 1830s had lived through the difficultyears of the French Revolution and the disloca-tion of the end of the Napoleonic period Theirability to accumulate wealth had been severelyimpaired Furthermore they were rather lesslikely to inherit much wealth since the Revo-lution wiped out the bond portfolios of theirparents through a prolonged period of high in-flation Those who died from the 1860s to theearly 1910s did not suffer from the adversitiesthat plagued their forebears Instead they en-joyed the fruits of the financial sector expansionthat began in the 1850s After 1947 we seem tohave returned to a situation quite like that of1817 Presumably the capital damages associ-ated with both world wars and the Great De-pression had a strong negative effect on thewealth holdings of older generations The per-sistence of a flat age-wealth profile until 1994 islikely to be associated with two factors First insocieties where income growth is rapid abso-lute wealth accumulation is faster by youngercohorts than by older ones because their in-comes are higher at every age This is an im-portant distinction between the nineteenthcentury and the twentieth century Furthermorehighly progressive rates of income and estatetaxation have probably made it more difficult toaccumulate large fortunes thereby flattening theobserved age-wealth profile

The more interesting (and more difficult)question relates to the possible efficiency im-pact of high wealth concentration and changingage-wealth profiles Although our data do notallow us to address efficiency issues in a rigor-ous way our results allow us to formulate anumber of hypotheses and to shed new light onthe ongoing debate on inequality and growth37

From a theoretical viewpoint whether highwealth concentration can have a negativegrowth impact depends critically on the exis-tence of credit constraints With first-best credit

markets money flows toward the best entrepre-neurs and investment projects irrespective ofthe initial distribution High levels of wealthconcentration can be bad from a social justiceviewpoint but they entail no efficiency lossWhen credit constraints bind however initialwealth matters and high levels of inequality canhurt growth Whether the loss is large or smalldepends on who owns the assets If the rich areefficient investors (they know which projects tofund etc) then wealth concentration may evenbe useful If the rich are retired rentiers how-ever investing their wealth in low-yield assets(or low-ability inheritors) then high wealthconcentration and credit constraints might pre-vent talented but penniless investors from un-dertaking efficient projects thereby entailingnegative growth consequences The data used inthis paper are not ideal to address whether creditconstraints were important in a country likeFrance at the end of the nineteenth century Ourresults suggest however that to the extentcredit constraints were indeed severe highwealth concentration did have a negativegrowth impact In order to investigate this hy-pothesis further one would need to gather moresystematic data on investment strategies andasset returns Preliminary evidence suggeststhat the rich elderly of the 1860ndash1913 perioddid indeed hold a disproportionate fraction oftheir wealth in low-yield assets (such as gov-ernment bonds) An alternative hypothesishowever is that steeper age-wealth profileswere the consequence of the growth of financialmarkets as their children faced fewer creditconstraints parents decided to hold on to moreof their wealth

VI Conclusion

Evidence from wealth at death in Paris and inFrance over the last two centuries reveals threekey patterns First wealth concentration haschanged dramatically over time In 1807 thetop-1-percent share of wealth (40 percent inFrance 50 percent in Paris) was twice as high asit would be in 1994 but substantially less thanin 1913 when it peaked above 55 percent inFrance and 70 percent in Paris Some of thesechanges were due to economic phenomena thathave long been emphasized as creating inequal-ity namely industrialization and financial cen-

37 Thus far this literature has concentrated upon cross-country regressions of inequality on growth a methodologythat raises serious identification problems especially giventhe low quality of available international datasets on in-equality which are neither long-run nor homogeneous (seeeg Atkinson and Andrea Brandolini 2001)

254 THE AMERICAN ECONOMIC REVIEW MARCH 2006

tralization Yet the decline comes largely fromadverse shocks rather than economic conver-gence These changes are of such magnitudethat they are not sensitive to whether one ex-amines wealth at death in Paris or in France orwhether one examines it directly rather thanconverting it to wealth of the living by an estatemultiplier method

Our second key result is that there was asignificant transition during the nineteenth cen-tury from an important role for real estate as aform of wealth to moveable assets as the keyform of wealth for the very rich Similarly theshare of wealth held by aristocrats first rose andthen was eclipsed by that of financiers and in-dustrialists in the second half of the nineteenthcentury Hence mobility within this highly un-equal society might have been quite high Yetthis conjecture is tempered by our third findingthe wealthy were getting older over time andolder relative to less wealthy decedents Suchaging among the very wealthy would have hadnegative consequences for growth if financialmarkets were imperfect This issue requires fur-ther investigation and we hope it will attractfuture research

REFERENCES

Atkinson Anthony B ldquoTop Incomes in the UKover the 20th Centuryrdquo Journal of the RoyalStatistical Society Series A (Statistics in So-ciety) 2005 168(2) pp 325ndash43

Atkinson Anthony B and Brandolini AndrealdquoPromise and Pitfalls in the Use of lsquoSecond-aryrsquo Data-Sets Income Inequality in OECDCountries as a Case Studyrdquo Journal of Eco-nomic Literature 2001 39(3) pp 771ndash99

Atkinson Anthony B and Harrison Alan J Dis-tribution of personal wealth in Britain Cam-bridge Cambridge University Press 1978

Atkinson Anthony B and Piketty Thomas Topincomes over the twentieth century A con-trast between European and English speak-ing countries Oxford Oxford UniversityPress (forthcoming)

Banerjee Abhijit V and Duflo Esther ldquoInequalityand Growth What Can the Data Sayrdquo Journalof Economic Growth 2003 8(3) pp 267ndash99

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoPourquoi la richesse ne

srsquoest-elle pas diffusee avec la croissance Ledegre zero de lrsquoinegalite et son evolution enFrance 1800ndash1940rdquo Histoire et mesure 200318(1ndash2) pp 147ndash98

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoDefense et illustration delrsquoenquete des 3000 familles Lrsquoexemple deson volet patrimonialrdquo Annales de demogra-phie historique 2004 19 pp 19ndash57

Danysz E S ldquoContribution a lrsquoetude des for-tunes privees drsquoapres les declarations de suc-cessionsrdquo Bulletin de la statistique generalede France 1934 XXIV (1) pp 111ndash171

Daumard Adeline Les fortunes francaise auXIXe siecle Enquete sur la repartition et lacomposition des capitaux prives a ParisLyon Lille Bordeaux et Toulouse drsquoapreslrsquoenregistrement des declarations de succes-sion Paris Mouton 1973

Gain Andre La restauration et les biens desemigres 2 vols Nancy Societe drsquoimpressiontypographique 1929

Jones Alice H American colonial wealth Doc-uments and methods 3 vols New York ArnoPress 1977

Kopczuk Wojciech and Saez Emmanuel ldquoTopWealth Shares in the United States 1916ndash2000 Evidence from Estate Tax ReturnsrdquoNational Tax Journal 2004 57(2) pp 445ndash87

Kuznets Simon ldquoEconomic Growth and IncomeInequalityrdquo American Economic Review1955 45(1) pp 1ndash28

Lampman Robert J The share of top wealth-holders in national wealth 1922ndash1956Princeton Princeton University Press 1962

Lescure Michel Les banques lrsquoEtat et lemarche immobilier en France a lrsquoepoquecontemporaine 1820ndash1940 Paris EditionslrsquoEcole des Hautes Etudes en Sciences Socia-les 1982

Levy-Leboyer Maurice and Bourguignon Fran-cois LrsquoEconomie francaise au XIXe siecleAnalyse macroeconomique Paris Eco-nomica 1985 English translation Cam-bridge Cambridge University Press 1990

Lindert Peter H ldquoUnequal English Wealthsince 1670rdquo Journal of Political Economy1986 94(6) pp 1127ndash62

Lindert Peter H ldquoThree Centuries of Inequalityin Britain and Americardquo in Anthony B At-kinson and Francois Bourguignon eds

255VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

tance declined steadily To be sure aristocratsremain overrepresented throughout the periodincluding in 1902 (about 13 percent of nobles inthe top 1 percent of estates over 25 percent inthe top 01 percent versus less than 1 percent inthe population as a whole) The inverted-U pat-tern is yet another of the Revolutionrsquos legaciesIn 1807 when we first observe it aristocraticwealth was at a temporary nadir On the onehand the nobility was impoverished by theRevolutionrsquos inflation and by the sharp declineof the value of Parisian real estate On the otherhand part of the Old Regime nobility was inexile and thus if they died we do not observetheir moveable wealth Aristocrats were able torecoup part of their losses during the first half ofthe nineteenth century Napoleon providedsome assistance by conferring titles of nobilityon his chief military officers and endowingthem with wealth Later the Restoration gov-ernment (1815ndash1830) compensated individualswho fled abroad during the Revolution for thelosses they suffered when their property wasconfiscated The government distributed nearlyone billion francs in the famous milliard desemigres (Andre Gain 1929) The beneficiariesof Napoleonrsquos and the Restorationrsquos largess ap-

pear among the very rich until mid-centuryPresumably such redistribution did not contrib-ute to accelerate French industrialization

III From Paris to France

We can use the Paris data to construct wealthconcentration at death estimates for all ofFrance from 1807 to 1902 To do so we need toknow the evolution of the share of Paris estatesin top estates Between 1902 and 1994 avail-able data (broken down by departement) showsthat the evolution of top estate shares in Francewas parallel to that of top estate shares in ParisWealth inequality is always lower for the coun-try as a whole but the trends are similar (seeFigure 7) It is also striking to note that Parisrsquosshare of the top 1 percent of French estates hasremained fairly stable over the twentieth cen-tury (it fluctuates between 20 percent and 25percent with no trend) even though Parisrsquosshare of all decedents has been dwindling overtime reflecting the population decline of thecapital (see Table 3) In 1902 Paris decedentswere four times more likely to belong to thenational top 1 percent of estates than averagedecedents (26665 41) in 1994 Paris de-

FIGURE 7 WEALTH CONCENTRATION AT DEATH IN PARIS AND FRANCE 1807ndash1994

Source Authorsrsquo computations based on estate tax returns (see Tables 2 and 4)

246 THE AMERICAN ECONOMIC REVIEW MARCH 2006

cedents were seven times more likely to belongto the national top 1 percent of estates thanaverage decedents (25236 70) If anythingthe geographic concentration of fortunes waslarger at the end of the twentieth century than atthe beginning The decline of wealth concentra-tion that followed World War I was not due toredistribution between Paris and the provinces

How did the fraction of Paris estates in topestates evolve over the course of the nineteenthcentury Our estimates rely on a simple andvery conservative assumption from 1807 to1902 Parisrsquos share of estates in the top percen-tile increased at the same rate as Parisrsquos share ofFrench adult deaths More precisely let us de-note FPt(w) the cumulative distribution functionfor wealth-at-death in Paris in year t Ft(w) thecorresponding distribution for France nPt thetotal number of adult deaths in Paris in year tand nt the corresponding number for all ofFrance The ninetieth-percentile threshold P90Ptis defined by FPt(P90Pt) 09 the ninety-ninth-percentile threshold P99Pt is defined byFPt(P99Pt) 099 etc and similarly for theFrench thresholds P90t P99t etc We observeFPt(w) nPt and nt throughout the 1807ndash1994period but we do not observe Ft(w) until 1902(before this date we observe only national ag-

gregate average wealth wt Wtnt) To con-struct our benchmark estimates we assume thatthe shares s99t s995t s999t and s9999t of Parisestates in the national top 1 percent 05 percent01 percent and 001 percent of the nationalwealth-at-death distribution increased at thesame rate as nPtnt during the 1807ndash1902 period(see Table 3) Using this approximation and ourParis samples of individual tax returns we com-pute the threshold wealth levels for the toppercentiles of the national wealth distribution(eg P99 P995 P999 and P9999)20 Wealso calculate the average wealth levels for therelevant wealth classes (eg P99ndash995 P995ndash999 P999ndash9999 and P9999ndash100) usingPareto interpolation techniques These are thenweighted by the number of individuals inFrance in that wealth class in order to compute

20 For instance the number of decedents (aged 20 yearsand older) in France was 583976 in 1887 (see Piketty et al2004 Appendix Table A1) so that the top 1 percent of theestate distribution at death consists of the top 5840 estatesIf the share of Paris among French top-1-percent estates was241 percent in 1887 (see Table 3) then the national P99threshold for 1887 corresponds to the top 1410 Parisianestates (0241 5840 1410) (the national P99 thresholdfor 1887 reported in Piketty et al 2004 Appendix TableA3 was computed using this formula)

TABLE 3mdashTHE FRACTION OF PARIS ESTATES IN TOP ESTATES AT DEATH IN FRANCE 1807ndash1994

(1) Fraction of Parisdecedents in all

decedents 20-yr

(2) Fraction of Parisestates in top-10-

percent estates

(3) Fraction of Parisestates in top-1-percent estates

(4) Fraction of Parisestates in top-01-

percent estates

1807 25 101 2051817 25 103 2101827 28 116 2371837 31 126 2561847 36 146 2971857 36 146 2971867 49 199 4041877 51 211 4281887 59 241 4911902 65 75 266 5411913 65 75 255 5231929 58 83 239 5301938 53 74 216 4211947 55 110 198 3521956 55 128 223 3501994 36 89 252 352

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources) No datasource exists to compute columns (3)ndash(4) prior to 1902 and the numbers reported on this table for years 1807ndash1887 werecomputed assuming that the columns (3)ndash(4) followed the same trend as column (1) over the 1807ndash1902 period (see textSection III)

247VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

the average wealth levels for top fractiles (P99ndash100 P995ndash100 P999ndash100 and P9999ndash100)Lower thresholds of the national wealth distri-bution (P90 and P95) were computed using thenational TRA survey21 and the P90ndash95 andP95ndash99 intermediate wealth levels were alsocomputed using Pareto interpolation techniques

The national top estate shares estimates re-ported in Table 4 were computed using thismethodology They suggest that wealth concen-tration (as measured by the top-1-percent estateshare) rose throughout the nineteenth century inFrance during both the 1807ndash1867 and 1867ndash1902 periods although less sharply than in Parisduring the latter period (see Figure 7) Theseestimates are conservative in the sense that it isalmost certain that they underestimate the riseof wealth concentration that took place inFrance during the nineteenth century First weknow that the bulk of population growth in Parisduring the nineteenth century was due to theannexation of suburbs in 1860 and to populationgrowth in these peripheral arrondissements Be-

cause the outskirts of the city were poor theannexation added few top estates Thus there islittle doubt that Parisrsquos share of top estates inFrance actually increased less than its share of thetotal population This hypothesis is confirmed bynineteenth-century housing tax tabulations show-ing that the fraction of Paris taxpayers in thenational top 1 percent of taxpayers was substan-tially larger than 10 percent at the beginning of thenineteenth century22 Giving Paris a larger (andmore realistic) share of top estates in 1807 wouldboth reduce the share of wealth of the top 1percent in France at that date and lead to morerapid rise in inequality over time

Next and most importantly other estate sur-veys are consistent with the view that wealth in-equality was growing The important study byAdeline Daumard (1973) which relied on samplesof estate tax returns collected in Paris Lyon Tou-louse Lille and Bordeaux found that wealth con-centration increased in each of these five citiesduring the nineteenth century23 The TRA surveyalthough it is ill-suited for the study of top estatesis also consistent with our view Wealth dispersionwas on the rise in nineteenth-century France ac-cording to the TRA survey both in the sense thatthe fraction of decedents with positive estates de-clined over time (in spite of the sharp increase inthe value of the average estate) and that ratios suchas the P90P50 increased24 We also comparedour benchmark national P99 series extrapolatedfrom our Paris samples and the national P99series computed using the TRA survey We foundthat both series display the same overall upwardtrend in concentration (which is reassuring regard-ing the general validity of our Paris-France extrap-olation technique) except that the growth ofinequality from 1807 to 1902 in the TRA series is

21 See Bourdieu et al (2003) for full details about theTRA survey The P90 and P95 thresholds reported onPiketty et al (2004 Appendix Table A3) were computedusing ten-year moving averages around the target years inorder to make sure that each estimate was based on asufficient number of observations

22 These tabulations were published in the same FinanceMinistry official publications as the estate tabulations Wechose not to use them in our formal computations becausethe tax base of the housing tax (namely the rental value ofthe real estate property where the household lives) is onlyloosely connected to the estate tax base (in particular onecannot rule out the possibility that the housing tax baseoverrepresents Paris-based taxpayers)

23 Unfortunately Daumardrsquos samples are not available inmachine-readable format she has only two or three years ofdata for each city and she did not try to compute homog-enous inequality indicators (top fractiles shares etc) withher data Thus although her results and our work are con-sistent they cannot be compared directly

24 See Bourdieu et al (2004)

TABLE 4mdashWEALTH CONCENTRATION AT DEATH IN FRANCE1807ndash1994

Top-10-percentestate share

Top-1-percentestate share

Top-01-percentestate share

1807 791 434 1631817 810 445 1811827 824 452 1631837 796 438 1471847 816 479 1841857 829 495 1741867 810 480 1741877 838 471 2011887 839 487 1921902 839 516 2311913 863 549 2601929 820 502 2471938 776 420 1991947 699 299 1101956 694 304 1101994 610 213 63

Source Authorsrsquo computations using estate tax returns (seePiketty et al 2004 Table A3 for detailed series)

248 THE AMERICAN ECONOMIC REVIEW MARCH 2006

more severe than in our series (see Figure 8) Thisagain suggests that the latter provide a conserva-tive lower bound for the upward trend in wealthconcentration In any case the finding of a largeincrease in wealth inequality in nineteenth-centuryFrance (and up until World War I) appears to berobust25

As was mentioned earlier there exists nocomparable continuous data source covering thenineteenth and twentieth centuries in othercountries which makes it difficult to put ourFrench long-run series in international perspec-tive We note however that existing series forthe United States and the United Kingdom areconsistent with our findings on France26 Re-garding levels existing evidence suggests thatduring the nineteenth and most of the twentiethcenturies France was in an intermediate posi-tion in terms of wealth concentration in be-tween the United States (more equal) and theUnited Kingdom (the most unequal)27 These

25 Note that this continuous rise in wealth inequalitydoes not necessarily imply that a parallel rise occurredregarding income inequality Given that there exists nomicro source on incomes prior to the creation of the incometax in 1914 it is very difficult (if not impossible) to properlyaddress this issue Christian Morrisson and Wayne Snyder(2000) have attempted to link income inequality estimatesbased upon Old-Regime fiscal sources (pre-1789) withmodern income-tax-based twentieth-century estimates andthey have argued that income inequality might have startedto decline during the later part of the nineteenth century andon the eve of World War I (see also Morrisson 2000)Although our data do not allow us to rule out such apossibility we note that their nineteenth-century personaldistribution estimates are based on fragile macroeconomicdata on functional distribution and are not homogenous totheir eighteenth- and twentieth-century estimates Given theevidence that we provide on wealth inequality any signif-icant decline in aggregate income inequality would have tobe associated with severe compression of the wage distri-bution There is little research on this issue however

26 Lee Soltow and Jan L Van Zanden (1998) also find adecline in inequality in the twentieth century in the Neth-erlands Their data are consistent with a rise in inequality inthe nineteenth century but they have no direct evidenceabout its extent

27 According to our series the top-1-percent wealthshare in France rose from around 45 percent in 1800 toabout 55 percent around World War I and then fell to about20 percent by the end of the twentieth century (see Figure7) wealth concentration among the living appears to besomewhat larger (see Section V below) According to theseries pieced together by Lindert (2000 pp 181ndash82 and186) the UK top-1-percent wealth share rose from about

FIGURE 8 ESTIMATES OF THE P99 THRESHOLD FOR THE FRENCH DISTRIBUTION OF ESTATES AT DEATH EXTRAPOLATION

FROM PARIS SAMPLES VERSUS ESTIMATES FROM TRA SAMPLES (CURRENT FRENCH FRANCS)

Source Authorsrsquo computations based on estate tax returns (Paris samples and TRA samples)

249VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

differences in inequality largely hark back to dif-ferential concentration of landownership En-glandrsquos land was extremely narrowly held whilethe United States was most egalitarian The impactof access to real estate assets can also be seenwhen we contrast Paris (where it was extremelyconcentrated) with the provinces

Trends in inequality over time may be easierto compare because biases in source materialmay have a significant effect on levels ratherthan trends The historical pattern is similar ineach of the three economies In particular thereis evidence that wealth concentration increasedduring the nineteenth century in both Anglo-Saxon countries and declined during the twen-tieth century with a turning point around WorldWar I28 Neither exhibits patterns consistentwith a Kuznet process

IV From the Wealth of the Decedents to theWealth of the Living

The estimates reported thus far refer to in-equality among decedents as described in thetax returns filed by their heirs The evolution ofthe distribution of wealth among the livingmight however have followed a different pat-tern In order to convert wealth-at-death con-centration estimates into wealth-of-the-livingconcentration estimates it is standard to use theldquoestate multiplierrdquo method29 It consists ofweighting each observation of an estate at death

by the inverse of the mortality rate for this agegroup That is if the mortality rate for ages 20to 24 was 068 percent in Paris in 1902 theneach decedent aged 20 to 24 represented about147 living individuals of the same age (100068 147) Similarly if the mortality ratefor ages above 80 was 2143 percent in Paris in1902 then each decedent in that group repre-sented about 47 living individuals in the sameage group (102143 47) Applying thismethod requires mortality tables (these are eas-ily available) and estate tabulations brokendown by estate size and age at death (these arescarcer) Fortunately the cityrsquos statistical bu-reau published annual death-by-age totals theFrench censuses report the age distribution forthe capital every five years and we collectedage at death from the estate declarations Thesedata allowed us to compute estimates of wealthconcentration among the living over the 1807ndash1902 period using various assumptions aboutthe wealth profiles of mortality rates

The base population for the living is the set ofall individuals aged 20 and over living in Parisin year t which we denote pt The number ofliving individuals in age bracket a is denoted pta(a 20ndash24 25ndash29 30ndash34 75ndash79 80 andover) and the number of decedents in agebracket a is denoted nta The mortality rate forage bracket a is given by mta ntapta Webegin with a uniform-mortality benchmarkThese estimates are based on the simplifyingassumption that these mortality rates dependsolely on age and are the same for all wealthgroups (and in particular are the same forzero-wealth and positive-wealth individuals)We can then weigh each decedent with positiveestate and age a collected in the Paris archivesin year t by ptanta This allows us to computethe number of living Parisians with positivewealth in year t and also (by differentiatingwith pt) the number of living Parisians with zerowealth at year t which is used to weight zero-estate observations We then use our weighteddatasets to compute top estate fractiles amongthe living in Paris

The main conclusion is that the living expe-rienced the same upward trend in wealth con-centration as the decedents (see Figure 9) Wefind that inequality was significantly higheramong the living than among decedents be-cause survivors were on average younger than

55 percent in 1800 to 70 percent around World War I thenfell to about 20 percent in the 1990s The US top-1-percentwealth share rose from about 15 to 20 percent in 1800 toabout 40 percent around World War I then fell to about 30percent in the 1990s (and as low as 20 to 25 percentaccording to the more recent estimates from Kopczuk andSaez 2004) Wealth concentration is now larger in theUnited States than in European countries but the reversewas true during the nineteenth century up until World WarII (It is only since the 1950sndash1970s period that US wealthconcentration has been somewhat larger)

28 See Lindert (2000 pp 181ndash82 and 188)29 This method was widely used in England and France

in the late nineteenth and early twentieth centuries to com-pute the stock of national wealth on the basis of the flow ofwealth transmitted at death Standard references that use thistechnique to estimate the wealth distribution of the livingfrom estate tax data tabulated by estate size and age at deathinclude Atkinson and Harrison (1978) and Lampman(1962) For a more recent application of this technique tothe United States see Kopczuk and Saez (2004)

250 THE AMERICAN ECONOMIC REVIEW MARCH 2006

those who died and the young were on aver-age poorer In particular the estate multipliermethod leads to lower average weights forpositive-wealth decedents than for zero-wealthdecedents (the former are on average older andtherefore ldquorepresentrdquo a smaller number of livingindividuals) As a result the fraction of positive-wealth individuals is even smaller among theliving than among decedents Hence our bench-mark uniform-mortality estimates of wealthconcentration among the living are significantlylarger than corresponding estimates among de-cedents (eg top-1-percent wealth shares areabout 15 percent larger among the living)Changes over time however are similar In-creased life expectancy and declining mortalityrates over the course of the nineteenth centuryhave only a small effect on the trends

In order to make the estate multiplier methodmore reliable one would prefer to take intoaccount differential mortality by wealth Doingso would require having access to mortalityschedules based both on wealth and age at dif-ferent points in time unfortunately these are notavailable We have nonetheless reestimatedwealth of the living based upon the same as-sumption as Kopczuk and Saez (2004) That is

we assumed uniform mortality among the poor(here defined as zero-wealth individuals) andamong the rich (here defined as positive-wealthindividuals) and we assumed that the ratiomtaRmtaP between the mortality rate of the richand the mortality rate of the poor followed aU-shaped age profile from about 85 percent forthe young (ie the rich die 15 percent less oftenthan the poor when they are 20ndash24 or 25ndash29years old) down to about 70 percent for middle-age individuals in their forties to fifties and upto 100 percent for very old individuals in theireighties to nineties30 This profile correspondsto the best available estimates in the literatureand it appears to be relatively stable over timeand across developed countries In the absenceof better data it is the best one can do31 Thebenchmark differential-mortality estimates re-ported on Figure 9 show that although addingdifferential mortality produces different levelsof inequality it does not have much impact onthe upward trend in concentration

30 See Kopczuk and Saez (2004 Table A4)31 See Kopczuk and Saez (2004 Appendix B) for refer-

ences to the US and international literature devoted to theage-wealth profile of mortality rates

FIGURE 9 WEALTH CONCENTRATION AMONG DECEDENTS AND AMONG THE LIVING IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives (see Piketty et al 2004Table A4 for detailed series)

251VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

The resulting differential-mortality inequalityestimates lie between those based on decedentsonly and those using uniform mortality for theliving (see Figure 9) Note that moving fromuniform-mortality to differential-mortality es-tate multiplier methodology can either increaseor decrease inequality Here the reason whysuch a move leads to lower wealth concentra-tion seems to be due to the fact that differentialmortality tends to put higher weights onpositive-wealth decedents (for a given age)thereby increasing the estimated fraction of liv-ing individuals with positive wealth The im-portant point however is that the resulting leveleffects are relatively small in magnitude con-stant in time and dwarfed by the upward timetrend Even if we were to assume an enormousincrease in differential mortality during thenineteenth century in the sense that differentialmortality between the rich and poor was equalto 0 percent of the benchmark differential in1807 and 100 percent of the benchmark differ-ential in 1902 the resulting wealth concentra-tion estimates would still be significantly higherin 1902 than in 1807 (see Figure 9) Yet wehave no reason to believe that differential mor-tality increased to such an extent During thenineteenth century real wages for unskilledworkers rose which would have reduced mor-tality more for the poor than for the rich After1850 public health measures (sanitation watervaccination) were in place Again these wouldhave had a significant effect on the poor and themiddle class who could not privately purchasesuch health-improving services To be sure therich could avail themselves of more medicalservices than the poor or the middle class butthe impact of these services was probably small(bear in mind that neither antibiotics nor car-diovascular interventions were available)

Finally we have applied the estate multipliermethod to available data for 1947 and 1994 Over-all the sharp decline in wealth concentration ob-served during the twentieth century (andespecially between 1914 and 1945) is very robustIf anything the decline appears to be even largerwhen one looks at wealth concentration amongthe living rather than among decedents32

V The Changing Age Profile of Wealth

In the previous sections we focused almostexclusively on aggregate top wealth shares Ourdata however also detail the characteristics ofeach decedent in particular their gender andage The evolution of wealth by gender is ofrelevance for over the past two centuries therehave been massive changes in womenrsquos laborforce participation capacity to manage theirown affairs and life expectancy relative to menThe evolution of wealth by age is of relevancebecause there was a significant increase in adultlife expectancy over the twentieth century andbecause the progressive diffusion of pensionsmay have changed savings motivations More-over age-wealth profiles also inform us aboutthe motives of wealth accumulation and theeconomic impact of high wealth concentration

A first pass at the data considers the genderbreakdown of wealth at death Remarkably inour micro data the share of women in top estatestakes its highest value on the eve of World WarI For instance the womenrsquos share in the top 05percent rose from 35 percent prior to 1850 to 45percent in 1902 only to fall to 40 percent afterWorld War II Strikingly womenrsquos share ofwealth follows almost exactly the pattern ofaggregate inequality Women were relativelyricher when inequality reached its apex inFrance than at any other time Moreover insti-tutional variables seem to have played almostno role in changing the relative wealth ofwomen Unlike in common law countriesFrench law starting with the code civil of 1804required nearly equal treatment of all children inbequests Further research will help us deter-mine to what extent women of great wealthwere heirs or part of economically very success-ful couples

The data also reveal striking changes in the

32 See Piketty et al (2004 Table A4) It is unfortunatelynot possible to construct complete series for wealth concen-

tration among the living for the twentieth century due todata limitations tables broken down by estate brackets andage of decedents are available solely for years 1943ndash1954and at the national level (no table broken down by estatebrackets and age of decedents has ever been compiled at thedepartement level except in 1931 for Seine departementsee E S Danysz 1934) and the 1994 micro sample is notlarge enough to allow for a reliable application of the estatemultiplier method at the Paris level Thus the only wealth-of-the-living concentration estimates we provide for thetwentieth century are national estimates for 1947 and 1994

252 THE AMERICAN ECONOMIC REVIEW MARCH 2006

age profile of wealth between 1807 and 1994(see Table 5) During the nineteenth century aswealth concentration was increasing the veryrich were getting older and older At the begin-ning of the nineteenth century in the aftermathof the French Revolution the richest individualswere those in their fifties they were typically100 percent richer on average than people intheir forties 25 percent richer than those in theirsixties and 40 percent richer than those in theirseventies and eighties By the 1870s howeverthe age-wealth pattern had become stronglymonotonic the richest individuals were the old-est individuals In 1902 people in their sixtiesand seventies bequeathed 150 percent morethan those in their fifties and those in theireighties 300 percent more On the eve of WorldWar I top wealth holders were old and likely tobe retired This pattern breaks some time duringthe 1914ndash1945 period33 In 1947 as well as in1994 we are back to a pattern where the richestindividuals are those in their fifties Overall theperiod of maximal wealth inequality (1860ndash1913) also appears to be a period characterizedby a very specific age profile of wealth and largeconcentration of assets among the elderly

Another way to analyze the changing age-wealth relationship is to look at average age by

top estate fractile34 In 1817 average age wasvirtually the same for the top 10 percent and thetop 1 percent of estates (or even slightly declin-ing) The average-age-per-fractile relationshipbecomes upward sloping during the nineteenthcentury and by 1902 those in the top 1 percentwere almost six years older than those in the top10 percent The relationship is flat in 1947 anddownward-sloping in 1994 Finally one canapply the estate multiplier method (see SectionIV above) and analyze how wealth concentra-tion by age group among the living changedover the course of the nineteenth century Thegeneral population in Paris did not becomeolder during the nineteenth century those aged60 or older made up about 15 percent of thepopulation in 1817 and after 1847 about 10 to11 percent35 The share of total wealth ownedby the elderly rose significantly however aswealth distribution worsened The wealth be-longing to those aged 60 or more rose fromabout 25 to 30 percent of the total at the begin-ning of the nineteenth century to about 40 to 45percent by the end of the century The wealthshare of those aged 70 or older doubled fromless than 10 percent to about 20 percent36

33 Existing evidence on the age-wealth profile for 1931(see Danysz 1934) suggests that the Great Depression andWorld War II (rather than World War I) played the leadingroles in breaking this pattern This is an issue we plan toinvestigate in future research

34 See Piketty et al (2004 Table 6)35 Although life expectancy was increasing which

should have led to large shares of population for oldergroups the city was also growing quickly The large num-ber of immigrants (who were typically in their twenties)increased the relative size of the younger cohorts (seePiketty et al 2004 Table 7 and Figure 11)

36 See Piketty et al (2004 Figure 12)

TABLE 5mdashTHE AGE PROFILE OF WEALTH AT DEATH IN PARIS 1817ndash1994(Average estate left by 50- 59-year-old 100)

20ndash29yr-old

30ndash39yr-old

40ndash49yr-old

50ndash59yr-old

60ndash69yr-old

70ndash79yr-old

80ndash89yr-old

90ndash99yr-old

1817 26 22 28 100 54 59 591827 44 50 53 100 88 87 601837 133 90 107 100 116 123 1101847 87 73 102 100 117 204 1321857 84 77 101 100 104 109 1451867 67 58 136 100 141 125 1541877 66 73 63 100 197 260 4301887 45 33 63 100 152 233 2951902 29 40 80 100 253 272 4011947 31 51 73 100 113 105 105 1091994 11 45 100 87 93 95 68

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources)

253VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

It is perhaps not surprising that inequalitybecame strongly correlated with age in the1860s Those who died at an old age in the1820s and 1830s had lived through the difficultyears of the French Revolution and the disloca-tion of the end of the Napoleonic period Theirability to accumulate wealth had been severelyimpaired Furthermore they were rather lesslikely to inherit much wealth since the Revo-lution wiped out the bond portfolios of theirparents through a prolonged period of high in-flation Those who died from the 1860s to theearly 1910s did not suffer from the adversitiesthat plagued their forebears Instead they en-joyed the fruits of the financial sector expansionthat began in the 1850s After 1947 we seem tohave returned to a situation quite like that of1817 Presumably the capital damages associ-ated with both world wars and the Great De-pression had a strong negative effect on thewealth holdings of older generations The per-sistence of a flat age-wealth profile until 1994 islikely to be associated with two factors First insocieties where income growth is rapid abso-lute wealth accumulation is faster by youngercohorts than by older ones because their in-comes are higher at every age This is an im-portant distinction between the nineteenthcentury and the twentieth century Furthermorehighly progressive rates of income and estatetaxation have probably made it more difficult toaccumulate large fortunes thereby flattening theobserved age-wealth profile

The more interesting (and more difficult)question relates to the possible efficiency im-pact of high wealth concentration and changingage-wealth profiles Although our data do notallow us to address efficiency issues in a rigor-ous way our results allow us to formulate anumber of hypotheses and to shed new light onthe ongoing debate on inequality and growth37

From a theoretical viewpoint whether highwealth concentration can have a negativegrowth impact depends critically on the exis-tence of credit constraints With first-best credit

markets money flows toward the best entrepre-neurs and investment projects irrespective ofthe initial distribution High levels of wealthconcentration can be bad from a social justiceviewpoint but they entail no efficiency lossWhen credit constraints bind however initialwealth matters and high levels of inequality canhurt growth Whether the loss is large or smalldepends on who owns the assets If the rich areefficient investors (they know which projects tofund etc) then wealth concentration may evenbe useful If the rich are retired rentiers how-ever investing their wealth in low-yield assets(or low-ability inheritors) then high wealthconcentration and credit constraints might pre-vent talented but penniless investors from un-dertaking efficient projects thereby entailingnegative growth consequences The data used inthis paper are not ideal to address whether creditconstraints were important in a country likeFrance at the end of the nineteenth century Ourresults suggest however that to the extentcredit constraints were indeed severe highwealth concentration did have a negativegrowth impact In order to investigate this hy-pothesis further one would need to gather moresystematic data on investment strategies andasset returns Preliminary evidence suggeststhat the rich elderly of the 1860ndash1913 perioddid indeed hold a disproportionate fraction oftheir wealth in low-yield assets (such as gov-ernment bonds) An alternative hypothesishowever is that steeper age-wealth profileswere the consequence of the growth of financialmarkets as their children faced fewer creditconstraints parents decided to hold on to moreof their wealth

VI Conclusion

Evidence from wealth at death in Paris and inFrance over the last two centuries reveals threekey patterns First wealth concentration haschanged dramatically over time In 1807 thetop-1-percent share of wealth (40 percent inFrance 50 percent in Paris) was twice as high asit would be in 1994 but substantially less thanin 1913 when it peaked above 55 percent inFrance and 70 percent in Paris Some of thesechanges were due to economic phenomena thathave long been emphasized as creating inequal-ity namely industrialization and financial cen-

37 Thus far this literature has concentrated upon cross-country regressions of inequality on growth a methodologythat raises serious identification problems especially giventhe low quality of available international datasets on in-equality which are neither long-run nor homogeneous (seeeg Atkinson and Andrea Brandolini 2001)

254 THE AMERICAN ECONOMIC REVIEW MARCH 2006

tralization Yet the decline comes largely fromadverse shocks rather than economic conver-gence These changes are of such magnitudethat they are not sensitive to whether one ex-amines wealth at death in Paris or in France orwhether one examines it directly rather thanconverting it to wealth of the living by an estatemultiplier method

Our second key result is that there was asignificant transition during the nineteenth cen-tury from an important role for real estate as aform of wealth to moveable assets as the keyform of wealth for the very rich Similarly theshare of wealth held by aristocrats first rose andthen was eclipsed by that of financiers and in-dustrialists in the second half of the nineteenthcentury Hence mobility within this highly un-equal society might have been quite high Yetthis conjecture is tempered by our third findingthe wealthy were getting older over time andolder relative to less wealthy decedents Suchaging among the very wealthy would have hadnegative consequences for growth if financialmarkets were imperfect This issue requires fur-ther investigation and we hope it will attractfuture research

REFERENCES

Atkinson Anthony B ldquoTop Incomes in the UKover the 20th Centuryrdquo Journal of the RoyalStatistical Society Series A (Statistics in So-ciety) 2005 168(2) pp 325ndash43

Atkinson Anthony B and Brandolini AndrealdquoPromise and Pitfalls in the Use of lsquoSecond-aryrsquo Data-Sets Income Inequality in OECDCountries as a Case Studyrdquo Journal of Eco-nomic Literature 2001 39(3) pp 771ndash99

Atkinson Anthony B and Harrison Alan J Dis-tribution of personal wealth in Britain Cam-bridge Cambridge University Press 1978

Atkinson Anthony B and Piketty Thomas Topincomes over the twentieth century A con-trast between European and English speak-ing countries Oxford Oxford UniversityPress (forthcoming)

Banerjee Abhijit V and Duflo Esther ldquoInequalityand Growth What Can the Data Sayrdquo Journalof Economic Growth 2003 8(3) pp 267ndash99

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoPourquoi la richesse ne

srsquoest-elle pas diffusee avec la croissance Ledegre zero de lrsquoinegalite et son evolution enFrance 1800ndash1940rdquo Histoire et mesure 200318(1ndash2) pp 147ndash98

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoDefense et illustration delrsquoenquete des 3000 familles Lrsquoexemple deson volet patrimonialrdquo Annales de demogra-phie historique 2004 19 pp 19ndash57

Danysz E S ldquoContribution a lrsquoetude des for-tunes privees drsquoapres les declarations de suc-cessionsrdquo Bulletin de la statistique generalede France 1934 XXIV (1) pp 111ndash171

Daumard Adeline Les fortunes francaise auXIXe siecle Enquete sur la repartition et lacomposition des capitaux prives a ParisLyon Lille Bordeaux et Toulouse drsquoapreslrsquoenregistrement des declarations de succes-sion Paris Mouton 1973

Gain Andre La restauration et les biens desemigres 2 vols Nancy Societe drsquoimpressiontypographique 1929

Jones Alice H American colonial wealth Doc-uments and methods 3 vols New York ArnoPress 1977

Kopczuk Wojciech and Saez Emmanuel ldquoTopWealth Shares in the United States 1916ndash2000 Evidence from Estate Tax ReturnsrdquoNational Tax Journal 2004 57(2) pp 445ndash87

Kuznets Simon ldquoEconomic Growth and IncomeInequalityrdquo American Economic Review1955 45(1) pp 1ndash28

Lampman Robert J The share of top wealth-holders in national wealth 1922ndash1956Princeton Princeton University Press 1962

Lescure Michel Les banques lrsquoEtat et lemarche immobilier en France a lrsquoepoquecontemporaine 1820ndash1940 Paris EditionslrsquoEcole des Hautes Etudes en Sciences Socia-les 1982

Levy-Leboyer Maurice and Bourguignon Fran-cois LrsquoEconomie francaise au XIXe siecleAnalyse macroeconomique Paris Eco-nomica 1985 English translation Cam-bridge Cambridge University Press 1990

Lindert Peter H ldquoUnequal English Wealthsince 1670rdquo Journal of Political Economy1986 94(6) pp 1127ndash62

Lindert Peter H ldquoThree Centuries of Inequalityin Britain and Americardquo in Anthony B At-kinson and Francois Bourguignon eds

255VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

cedents were seven times more likely to belongto the national top 1 percent of estates thanaverage decedents (25236 70) If anythingthe geographic concentration of fortunes waslarger at the end of the twentieth century than atthe beginning The decline of wealth concentra-tion that followed World War I was not due toredistribution between Paris and the provinces

How did the fraction of Paris estates in topestates evolve over the course of the nineteenthcentury Our estimates rely on a simple andvery conservative assumption from 1807 to1902 Parisrsquos share of estates in the top percen-tile increased at the same rate as Parisrsquos share ofFrench adult deaths More precisely let us de-note FPt(w) the cumulative distribution functionfor wealth-at-death in Paris in year t Ft(w) thecorresponding distribution for France nPt thetotal number of adult deaths in Paris in year tand nt the corresponding number for all ofFrance The ninetieth-percentile threshold P90Ptis defined by FPt(P90Pt) 09 the ninety-ninth-percentile threshold P99Pt is defined byFPt(P99Pt) 099 etc and similarly for theFrench thresholds P90t P99t etc We observeFPt(w) nPt and nt throughout the 1807ndash1994period but we do not observe Ft(w) until 1902(before this date we observe only national ag-

gregate average wealth wt Wtnt) To con-struct our benchmark estimates we assume thatthe shares s99t s995t s999t and s9999t of Parisestates in the national top 1 percent 05 percent01 percent and 001 percent of the nationalwealth-at-death distribution increased at thesame rate as nPtnt during the 1807ndash1902 period(see Table 3) Using this approximation and ourParis samples of individual tax returns we com-pute the threshold wealth levels for the toppercentiles of the national wealth distribution(eg P99 P995 P999 and P9999)20 Wealso calculate the average wealth levels for therelevant wealth classes (eg P99ndash995 P995ndash999 P999ndash9999 and P9999ndash100) usingPareto interpolation techniques These are thenweighted by the number of individuals inFrance in that wealth class in order to compute

20 For instance the number of decedents (aged 20 yearsand older) in France was 583976 in 1887 (see Piketty et al2004 Appendix Table A1) so that the top 1 percent of theestate distribution at death consists of the top 5840 estatesIf the share of Paris among French top-1-percent estates was241 percent in 1887 (see Table 3) then the national P99threshold for 1887 corresponds to the top 1410 Parisianestates (0241 5840 1410) (the national P99 thresholdfor 1887 reported in Piketty et al 2004 Appendix TableA3 was computed using this formula)

TABLE 3mdashTHE FRACTION OF PARIS ESTATES IN TOP ESTATES AT DEATH IN FRANCE 1807ndash1994

(1) Fraction of Parisdecedents in all

decedents 20-yr

(2) Fraction of Parisestates in top-10-

percent estates

(3) Fraction of Parisestates in top-1-percent estates

(4) Fraction of Parisestates in top-01-

percent estates

1807 25 101 2051817 25 103 2101827 28 116 2371837 31 126 2561847 36 146 2971857 36 146 2971867 49 199 4041877 51 211 4281887 59 241 4911902 65 75 266 5411913 65 75 255 5231929 58 83 239 5301938 53 74 216 4211947 55 110 198 3521956 55 128 223 3501994 36 89 252 352

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources) No datasource exists to compute columns (3)ndash(4) prior to 1902 and the numbers reported on this table for years 1807ndash1887 werecomputed assuming that the columns (3)ndash(4) followed the same trend as column (1) over the 1807ndash1902 period (see textSection III)

247VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

the average wealth levels for top fractiles (P99ndash100 P995ndash100 P999ndash100 and P9999ndash100)Lower thresholds of the national wealth distri-bution (P90 and P95) were computed using thenational TRA survey21 and the P90ndash95 andP95ndash99 intermediate wealth levels were alsocomputed using Pareto interpolation techniques

The national top estate shares estimates re-ported in Table 4 were computed using thismethodology They suggest that wealth concen-tration (as measured by the top-1-percent estateshare) rose throughout the nineteenth century inFrance during both the 1807ndash1867 and 1867ndash1902 periods although less sharply than in Parisduring the latter period (see Figure 7) Theseestimates are conservative in the sense that it isalmost certain that they underestimate the riseof wealth concentration that took place inFrance during the nineteenth century First weknow that the bulk of population growth in Parisduring the nineteenth century was due to theannexation of suburbs in 1860 and to populationgrowth in these peripheral arrondissements Be-

cause the outskirts of the city were poor theannexation added few top estates Thus there islittle doubt that Parisrsquos share of top estates inFrance actually increased less than its share of thetotal population This hypothesis is confirmed bynineteenth-century housing tax tabulations show-ing that the fraction of Paris taxpayers in thenational top 1 percent of taxpayers was substan-tially larger than 10 percent at the beginning of thenineteenth century22 Giving Paris a larger (andmore realistic) share of top estates in 1807 wouldboth reduce the share of wealth of the top 1percent in France at that date and lead to morerapid rise in inequality over time

Next and most importantly other estate sur-veys are consistent with the view that wealth in-equality was growing The important study byAdeline Daumard (1973) which relied on samplesof estate tax returns collected in Paris Lyon Tou-louse Lille and Bordeaux found that wealth con-centration increased in each of these five citiesduring the nineteenth century23 The TRA surveyalthough it is ill-suited for the study of top estatesis also consistent with our view Wealth dispersionwas on the rise in nineteenth-century France ac-cording to the TRA survey both in the sense thatthe fraction of decedents with positive estates de-clined over time (in spite of the sharp increase inthe value of the average estate) and that ratios suchas the P90P50 increased24 We also comparedour benchmark national P99 series extrapolatedfrom our Paris samples and the national P99series computed using the TRA survey We foundthat both series display the same overall upwardtrend in concentration (which is reassuring regard-ing the general validity of our Paris-France extrap-olation technique) except that the growth ofinequality from 1807 to 1902 in the TRA series is

21 See Bourdieu et al (2003) for full details about theTRA survey The P90 and P95 thresholds reported onPiketty et al (2004 Appendix Table A3) were computedusing ten-year moving averages around the target years inorder to make sure that each estimate was based on asufficient number of observations

22 These tabulations were published in the same FinanceMinistry official publications as the estate tabulations Wechose not to use them in our formal computations becausethe tax base of the housing tax (namely the rental value ofthe real estate property where the household lives) is onlyloosely connected to the estate tax base (in particular onecannot rule out the possibility that the housing tax baseoverrepresents Paris-based taxpayers)

23 Unfortunately Daumardrsquos samples are not available inmachine-readable format she has only two or three years ofdata for each city and she did not try to compute homog-enous inequality indicators (top fractiles shares etc) withher data Thus although her results and our work are con-sistent they cannot be compared directly

24 See Bourdieu et al (2004)

TABLE 4mdashWEALTH CONCENTRATION AT DEATH IN FRANCE1807ndash1994

Top-10-percentestate share

Top-1-percentestate share

Top-01-percentestate share

1807 791 434 1631817 810 445 1811827 824 452 1631837 796 438 1471847 816 479 1841857 829 495 1741867 810 480 1741877 838 471 2011887 839 487 1921902 839 516 2311913 863 549 2601929 820 502 2471938 776 420 1991947 699 299 1101956 694 304 1101994 610 213 63

Source Authorsrsquo computations using estate tax returns (seePiketty et al 2004 Table A3 for detailed series)

248 THE AMERICAN ECONOMIC REVIEW MARCH 2006

more severe than in our series (see Figure 8) Thisagain suggests that the latter provide a conserva-tive lower bound for the upward trend in wealthconcentration In any case the finding of a largeincrease in wealth inequality in nineteenth-centuryFrance (and up until World War I) appears to berobust25

As was mentioned earlier there exists nocomparable continuous data source covering thenineteenth and twentieth centuries in othercountries which makes it difficult to put ourFrench long-run series in international perspec-tive We note however that existing series forthe United States and the United Kingdom areconsistent with our findings on France26 Re-garding levels existing evidence suggests thatduring the nineteenth and most of the twentiethcenturies France was in an intermediate posi-tion in terms of wealth concentration in be-tween the United States (more equal) and theUnited Kingdom (the most unequal)27 These

25 Note that this continuous rise in wealth inequalitydoes not necessarily imply that a parallel rise occurredregarding income inequality Given that there exists nomicro source on incomes prior to the creation of the incometax in 1914 it is very difficult (if not impossible) to properlyaddress this issue Christian Morrisson and Wayne Snyder(2000) have attempted to link income inequality estimatesbased upon Old-Regime fiscal sources (pre-1789) withmodern income-tax-based twentieth-century estimates andthey have argued that income inequality might have startedto decline during the later part of the nineteenth century andon the eve of World War I (see also Morrisson 2000)Although our data do not allow us to rule out such apossibility we note that their nineteenth-century personaldistribution estimates are based on fragile macroeconomicdata on functional distribution and are not homogenous totheir eighteenth- and twentieth-century estimates Given theevidence that we provide on wealth inequality any signif-icant decline in aggregate income inequality would have tobe associated with severe compression of the wage distri-bution There is little research on this issue however

26 Lee Soltow and Jan L Van Zanden (1998) also find adecline in inequality in the twentieth century in the Neth-erlands Their data are consistent with a rise in inequality inthe nineteenth century but they have no direct evidenceabout its extent

27 According to our series the top-1-percent wealthshare in France rose from around 45 percent in 1800 toabout 55 percent around World War I and then fell to about20 percent by the end of the twentieth century (see Figure7) wealth concentration among the living appears to besomewhat larger (see Section V below) According to theseries pieced together by Lindert (2000 pp 181ndash82 and186) the UK top-1-percent wealth share rose from about

FIGURE 8 ESTIMATES OF THE P99 THRESHOLD FOR THE FRENCH DISTRIBUTION OF ESTATES AT DEATH EXTRAPOLATION

FROM PARIS SAMPLES VERSUS ESTIMATES FROM TRA SAMPLES (CURRENT FRENCH FRANCS)

Source Authorsrsquo computations based on estate tax returns (Paris samples and TRA samples)

249VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

differences in inequality largely hark back to dif-ferential concentration of landownership En-glandrsquos land was extremely narrowly held whilethe United States was most egalitarian The impactof access to real estate assets can also be seenwhen we contrast Paris (where it was extremelyconcentrated) with the provinces

Trends in inequality over time may be easierto compare because biases in source materialmay have a significant effect on levels ratherthan trends The historical pattern is similar ineach of the three economies In particular thereis evidence that wealth concentration increasedduring the nineteenth century in both Anglo-Saxon countries and declined during the twen-tieth century with a turning point around WorldWar I28 Neither exhibits patterns consistentwith a Kuznet process

IV From the Wealth of the Decedents to theWealth of the Living

The estimates reported thus far refer to in-equality among decedents as described in thetax returns filed by their heirs The evolution ofthe distribution of wealth among the livingmight however have followed a different pat-tern In order to convert wealth-at-death con-centration estimates into wealth-of-the-livingconcentration estimates it is standard to use theldquoestate multiplierrdquo method29 It consists ofweighting each observation of an estate at death

by the inverse of the mortality rate for this agegroup That is if the mortality rate for ages 20to 24 was 068 percent in Paris in 1902 theneach decedent aged 20 to 24 represented about147 living individuals of the same age (100068 147) Similarly if the mortality ratefor ages above 80 was 2143 percent in Paris in1902 then each decedent in that group repre-sented about 47 living individuals in the sameage group (102143 47) Applying thismethod requires mortality tables (these are eas-ily available) and estate tabulations brokendown by estate size and age at death (these arescarcer) Fortunately the cityrsquos statistical bu-reau published annual death-by-age totals theFrench censuses report the age distribution forthe capital every five years and we collectedage at death from the estate declarations Thesedata allowed us to compute estimates of wealthconcentration among the living over the 1807ndash1902 period using various assumptions aboutthe wealth profiles of mortality rates

The base population for the living is the set ofall individuals aged 20 and over living in Parisin year t which we denote pt The number ofliving individuals in age bracket a is denoted pta(a 20ndash24 25ndash29 30ndash34 75ndash79 80 andover) and the number of decedents in agebracket a is denoted nta The mortality rate forage bracket a is given by mta ntapta Webegin with a uniform-mortality benchmarkThese estimates are based on the simplifyingassumption that these mortality rates dependsolely on age and are the same for all wealthgroups (and in particular are the same forzero-wealth and positive-wealth individuals)We can then weigh each decedent with positiveestate and age a collected in the Paris archivesin year t by ptanta This allows us to computethe number of living Parisians with positivewealth in year t and also (by differentiatingwith pt) the number of living Parisians with zerowealth at year t which is used to weight zero-estate observations We then use our weighteddatasets to compute top estate fractiles amongthe living in Paris

The main conclusion is that the living expe-rienced the same upward trend in wealth con-centration as the decedents (see Figure 9) Wefind that inequality was significantly higheramong the living than among decedents be-cause survivors were on average younger than

55 percent in 1800 to 70 percent around World War I thenfell to about 20 percent in the 1990s The US top-1-percentwealth share rose from about 15 to 20 percent in 1800 toabout 40 percent around World War I then fell to about 30percent in the 1990s (and as low as 20 to 25 percentaccording to the more recent estimates from Kopczuk andSaez 2004) Wealth concentration is now larger in theUnited States than in European countries but the reversewas true during the nineteenth century up until World WarII (It is only since the 1950sndash1970s period that US wealthconcentration has been somewhat larger)

28 See Lindert (2000 pp 181ndash82 and 188)29 This method was widely used in England and France

in the late nineteenth and early twentieth centuries to com-pute the stock of national wealth on the basis of the flow ofwealth transmitted at death Standard references that use thistechnique to estimate the wealth distribution of the livingfrom estate tax data tabulated by estate size and age at deathinclude Atkinson and Harrison (1978) and Lampman(1962) For a more recent application of this technique tothe United States see Kopczuk and Saez (2004)

250 THE AMERICAN ECONOMIC REVIEW MARCH 2006

those who died and the young were on aver-age poorer In particular the estate multipliermethod leads to lower average weights forpositive-wealth decedents than for zero-wealthdecedents (the former are on average older andtherefore ldquorepresentrdquo a smaller number of livingindividuals) As a result the fraction of positive-wealth individuals is even smaller among theliving than among decedents Hence our bench-mark uniform-mortality estimates of wealthconcentration among the living are significantlylarger than corresponding estimates among de-cedents (eg top-1-percent wealth shares areabout 15 percent larger among the living)Changes over time however are similar In-creased life expectancy and declining mortalityrates over the course of the nineteenth centuryhave only a small effect on the trends

In order to make the estate multiplier methodmore reliable one would prefer to take intoaccount differential mortality by wealth Doingso would require having access to mortalityschedules based both on wealth and age at dif-ferent points in time unfortunately these are notavailable We have nonetheless reestimatedwealth of the living based upon the same as-sumption as Kopczuk and Saez (2004) That is

we assumed uniform mortality among the poor(here defined as zero-wealth individuals) andamong the rich (here defined as positive-wealthindividuals) and we assumed that the ratiomtaRmtaP between the mortality rate of the richand the mortality rate of the poor followed aU-shaped age profile from about 85 percent forthe young (ie the rich die 15 percent less oftenthan the poor when they are 20ndash24 or 25ndash29years old) down to about 70 percent for middle-age individuals in their forties to fifties and upto 100 percent for very old individuals in theireighties to nineties30 This profile correspondsto the best available estimates in the literatureand it appears to be relatively stable over timeand across developed countries In the absenceof better data it is the best one can do31 Thebenchmark differential-mortality estimates re-ported on Figure 9 show that although addingdifferential mortality produces different levelsof inequality it does not have much impact onthe upward trend in concentration

30 See Kopczuk and Saez (2004 Table A4)31 See Kopczuk and Saez (2004 Appendix B) for refer-

ences to the US and international literature devoted to theage-wealth profile of mortality rates

FIGURE 9 WEALTH CONCENTRATION AMONG DECEDENTS AND AMONG THE LIVING IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives (see Piketty et al 2004Table A4 for detailed series)

251VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

The resulting differential-mortality inequalityestimates lie between those based on decedentsonly and those using uniform mortality for theliving (see Figure 9) Note that moving fromuniform-mortality to differential-mortality es-tate multiplier methodology can either increaseor decrease inequality Here the reason whysuch a move leads to lower wealth concentra-tion seems to be due to the fact that differentialmortality tends to put higher weights onpositive-wealth decedents (for a given age)thereby increasing the estimated fraction of liv-ing individuals with positive wealth The im-portant point however is that the resulting leveleffects are relatively small in magnitude con-stant in time and dwarfed by the upward timetrend Even if we were to assume an enormousincrease in differential mortality during thenineteenth century in the sense that differentialmortality between the rich and poor was equalto 0 percent of the benchmark differential in1807 and 100 percent of the benchmark differ-ential in 1902 the resulting wealth concentra-tion estimates would still be significantly higherin 1902 than in 1807 (see Figure 9) Yet wehave no reason to believe that differential mor-tality increased to such an extent During thenineteenth century real wages for unskilledworkers rose which would have reduced mor-tality more for the poor than for the rich After1850 public health measures (sanitation watervaccination) were in place Again these wouldhave had a significant effect on the poor and themiddle class who could not privately purchasesuch health-improving services To be sure therich could avail themselves of more medicalservices than the poor or the middle class butthe impact of these services was probably small(bear in mind that neither antibiotics nor car-diovascular interventions were available)

Finally we have applied the estate multipliermethod to available data for 1947 and 1994 Over-all the sharp decline in wealth concentration ob-served during the twentieth century (andespecially between 1914 and 1945) is very robustIf anything the decline appears to be even largerwhen one looks at wealth concentration amongthe living rather than among decedents32

V The Changing Age Profile of Wealth

In the previous sections we focused almostexclusively on aggregate top wealth shares Ourdata however also detail the characteristics ofeach decedent in particular their gender andage The evolution of wealth by gender is ofrelevance for over the past two centuries therehave been massive changes in womenrsquos laborforce participation capacity to manage theirown affairs and life expectancy relative to menThe evolution of wealth by age is of relevancebecause there was a significant increase in adultlife expectancy over the twentieth century andbecause the progressive diffusion of pensionsmay have changed savings motivations More-over age-wealth profiles also inform us aboutthe motives of wealth accumulation and theeconomic impact of high wealth concentration

A first pass at the data considers the genderbreakdown of wealth at death Remarkably inour micro data the share of women in top estatestakes its highest value on the eve of World WarI For instance the womenrsquos share in the top 05percent rose from 35 percent prior to 1850 to 45percent in 1902 only to fall to 40 percent afterWorld War II Strikingly womenrsquos share ofwealth follows almost exactly the pattern ofaggregate inequality Women were relativelyricher when inequality reached its apex inFrance than at any other time Moreover insti-tutional variables seem to have played almostno role in changing the relative wealth ofwomen Unlike in common law countriesFrench law starting with the code civil of 1804required nearly equal treatment of all children inbequests Further research will help us deter-mine to what extent women of great wealthwere heirs or part of economically very success-ful couples

The data also reveal striking changes in the

32 See Piketty et al (2004 Table A4) It is unfortunatelynot possible to construct complete series for wealth concen-

tration among the living for the twentieth century due todata limitations tables broken down by estate brackets andage of decedents are available solely for years 1943ndash1954and at the national level (no table broken down by estatebrackets and age of decedents has ever been compiled at thedepartement level except in 1931 for Seine departementsee E S Danysz 1934) and the 1994 micro sample is notlarge enough to allow for a reliable application of the estatemultiplier method at the Paris level Thus the only wealth-of-the-living concentration estimates we provide for thetwentieth century are national estimates for 1947 and 1994

252 THE AMERICAN ECONOMIC REVIEW MARCH 2006

age profile of wealth between 1807 and 1994(see Table 5) During the nineteenth century aswealth concentration was increasing the veryrich were getting older and older At the begin-ning of the nineteenth century in the aftermathof the French Revolution the richest individualswere those in their fifties they were typically100 percent richer on average than people intheir forties 25 percent richer than those in theirsixties and 40 percent richer than those in theirseventies and eighties By the 1870s howeverthe age-wealth pattern had become stronglymonotonic the richest individuals were the old-est individuals In 1902 people in their sixtiesand seventies bequeathed 150 percent morethan those in their fifties and those in theireighties 300 percent more On the eve of WorldWar I top wealth holders were old and likely tobe retired This pattern breaks some time duringthe 1914ndash1945 period33 In 1947 as well as in1994 we are back to a pattern where the richestindividuals are those in their fifties Overall theperiod of maximal wealth inequality (1860ndash1913) also appears to be a period characterizedby a very specific age profile of wealth and largeconcentration of assets among the elderly

Another way to analyze the changing age-wealth relationship is to look at average age by

top estate fractile34 In 1817 average age wasvirtually the same for the top 10 percent and thetop 1 percent of estates (or even slightly declin-ing) The average-age-per-fractile relationshipbecomes upward sloping during the nineteenthcentury and by 1902 those in the top 1 percentwere almost six years older than those in the top10 percent The relationship is flat in 1947 anddownward-sloping in 1994 Finally one canapply the estate multiplier method (see SectionIV above) and analyze how wealth concentra-tion by age group among the living changedover the course of the nineteenth century Thegeneral population in Paris did not becomeolder during the nineteenth century those aged60 or older made up about 15 percent of thepopulation in 1817 and after 1847 about 10 to11 percent35 The share of total wealth ownedby the elderly rose significantly however aswealth distribution worsened The wealth be-longing to those aged 60 or more rose fromabout 25 to 30 percent of the total at the begin-ning of the nineteenth century to about 40 to 45percent by the end of the century The wealthshare of those aged 70 or older doubled fromless than 10 percent to about 20 percent36

33 Existing evidence on the age-wealth profile for 1931(see Danysz 1934) suggests that the Great Depression andWorld War II (rather than World War I) played the leadingroles in breaking this pattern This is an issue we plan toinvestigate in future research

34 See Piketty et al (2004 Table 6)35 Although life expectancy was increasing which

should have led to large shares of population for oldergroups the city was also growing quickly The large num-ber of immigrants (who were typically in their twenties)increased the relative size of the younger cohorts (seePiketty et al 2004 Table 7 and Figure 11)

36 See Piketty et al (2004 Figure 12)

TABLE 5mdashTHE AGE PROFILE OF WEALTH AT DEATH IN PARIS 1817ndash1994(Average estate left by 50- 59-year-old 100)

20ndash29yr-old

30ndash39yr-old

40ndash49yr-old

50ndash59yr-old

60ndash69yr-old

70ndash79yr-old

80ndash89yr-old

90ndash99yr-old

1817 26 22 28 100 54 59 591827 44 50 53 100 88 87 601837 133 90 107 100 116 123 1101847 87 73 102 100 117 204 1321857 84 77 101 100 104 109 1451867 67 58 136 100 141 125 1541877 66 73 63 100 197 260 4301887 45 33 63 100 152 233 2951902 29 40 80 100 253 272 4011947 31 51 73 100 113 105 105 1091994 11 45 100 87 93 95 68

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources)

253VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

It is perhaps not surprising that inequalitybecame strongly correlated with age in the1860s Those who died at an old age in the1820s and 1830s had lived through the difficultyears of the French Revolution and the disloca-tion of the end of the Napoleonic period Theirability to accumulate wealth had been severelyimpaired Furthermore they were rather lesslikely to inherit much wealth since the Revo-lution wiped out the bond portfolios of theirparents through a prolonged period of high in-flation Those who died from the 1860s to theearly 1910s did not suffer from the adversitiesthat plagued their forebears Instead they en-joyed the fruits of the financial sector expansionthat began in the 1850s After 1947 we seem tohave returned to a situation quite like that of1817 Presumably the capital damages associ-ated with both world wars and the Great De-pression had a strong negative effect on thewealth holdings of older generations The per-sistence of a flat age-wealth profile until 1994 islikely to be associated with two factors First insocieties where income growth is rapid abso-lute wealth accumulation is faster by youngercohorts than by older ones because their in-comes are higher at every age This is an im-portant distinction between the nineteenthcentury and the twentieth century Furthermorehighly progressive rates of income and estatetaxation have probably made it more difficult toaccumulate large fortunes thereby flattening theobserved age-wealth profile

The more interesting (and more difficult)question relates to the possible efficiency im-pact of high wealth concentration and changingage-wealth profiles Although our data do notallow us to address efficiency issues in a rigor-ous way our results allow us to formulate anumber of hypotheses and to shed new light onthe ongoing debate on inequality and growth37

From a theoretical viewpoint whether highwealth concentration can have a negativegrowth impact depends critically on the exis-tence of credit constraints With first-best credit

markets money flows toward the best entrepre-neurs and investment projects irrespective ofthe initial distribution High levels of wealthconcentration can be bad from a social justiceviewpoint but they entail no efficiency lossWhen credit constraints bind however initialwealth matters and high levels of inequality canhurt growth Whether the loss is large or smalldepends on who owns the assets If the rich areefficient investors (they know which projects tofund etc) then wealth concentration may evenbe useful If the rich are retired rentiers how-ever investing their wealth in low-yield assets(or low-ability inheritors) then high wealthconcentration and credit constraints might pre-vent talented but penniless investors from un-dertaking efficient projects thereby entailingnegative growth consequences The data used inthis paper are not ideal to address whether creditconstraints were important in a country likeFrance at the end of the nineteenth century Ourresults suggest however that to the extentcredit constraints were indeed severe highwealth concentration did have a negativegrowth impact In order to investigate this hy-pothesis further one would need to gather moresystematic data on investment strategies andasset returns Preliminary evidence suggeststhat the rich elderly of the 1860ndash1913 perioddid indeed hold a disproportionate fraction oftheir wealth in low-yield assets (such as gov-ernment bonds) An alternative hypothesishowever is that steeper age-wealth profileswere the consequence of the growth of financialmarkets as their children faced fewer creditconstraints parents decided to hold on to moreof their wealth

VI Conclusion

Evidence from wealth at death in Paris and inFrance over the last two centuries reveals threekey patterns First wealth concentration haschanged dramatically over time In 1807 thetop-1-percent share of wealth (40 percent inFrance 50 percent in Paris) was twice as high asit would be in 1994 but substantially less thanin 1913 when it peaked above 55 percent inFrance and 70 percent in Paris Some of thesechanges were due to economic phenomena thathave long been emphasized as creating inequal-ity namely industrialization and financial cen-

37 Thus far this literature has concentrated upon cross-country regressions of inequality on growth a methodologythat raises serious identification problems especially giventhe low quality of available international datasets on in-equality which are neither long-run nor homogeneous (seeeg Atkinson and Andrea Brandolini 2001)

254 THE AMERICAN ECONOMIC REVIEW MARCH 2006

tralization Yet the decline comes largely fromadverse shocks rather than economic conver-gence These changes are of such magnitudethat they are not sensitive to whether one ex-amines wealth at death in Paris or in France orwhether one examines it directly rather thanconverting it to wealth of the living by an estatemultiplier method

Our second key result is that there was asignificant transition during the nineteenth cen-tury from an important role for real estate as aform of wealth to moveable assets as the keyform of wealth for the very rich Similarly theshare of wealth held by aristocrats first rose andthen was eclipsed by that of financiers and in-dustrialists in the second half of the nineteenthcentury Hence mobility within this highly un-equal society might have been quite high Yetthis conjecture is tempered by our third findingthe wealthy were getting older over time andolder relative to less wealthy decedents Suchaging among the very wealthy would have hadnegative consequences for growth if financialmarkets were imperfect This issue requires fur-ther investigation and we hope it will attractfuture research

REFERENCES

Atkinson Anthony B ldquoTop Incomes in the UKover the 20th Centuryrdquo Journal of the RoyalStatistical Society Series A (Statistics in So-ciety) 2005 168(2) pp 325ndash43

Atkinson Anthony B and Brandolini AndrealdquoPromise and Pitfalls in the Use of lsquoSecond-aryrsquo Data-Sets Income Inequality in OECDCountries as a Case Studyrdquo Journal of Eco-nomic Literature 2001 39(3) pp 771ndash99

Atkinson Anthony B and Harrison Alan J Dis-tribution of personal wealth in Britain Cam-bridge Cambridge University Press 1978

Atkinson Anthony B and Piketty Thomas Topincomes over the twentieth century A con-trast between European and English speak-ing countries Oxford Oxford UniversityPress (forthcoming)

Banerjee Abhijit V and Duflo Esther ldquoInequalityand Growth What Can the Data Sayrdquo Journalof Economic Growth 2003 8(3) pp 267ndash99

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoPourquoi la richesse ne

srsquoest-elle pas diffusee avec la croissance Ledegre zero de lrsquoinegalite et son evolution enFrance 1800ndash1940rdquo Histoire et mesure 200318(1ndash2) pp 147ndash98

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoDefense et illustration delrsquoenquete des 3000 familles Lrsquoexemple deson volet patrimonialrdquo Annales de demogra-phie historique 2004 19 pp 19ndash57

Danysz E S ldquoContribution a lrsquoetude des for-tunes privees drsquoapres les declarations de suc-cessionsrdquo Bulletin de la statistique generalede France 1934 XXIV (1) pp 111ndash171

Daumard Adeline Les fortunes francaise auXIXe siecle Enquete sur la repartition et lacomposition des capitaux prives a ParisLyon Lille Bordeaux et Toulouse drsquoapreslrsquoenregistrement des declarations de succes-sion Paris Mouton 1973

Gain Andre La restauration et les biens desemigres 2 vols Nancy Societe drsquoimpressiontypographique 1929

Jones Alice H American colonial wealth Doc-uments and methods 3 vols New York ArnoPress 1977

Kopczuk Wojciech and Saez Emmanuel ldquoTopWealth Shares in the United States 1916ndash2000 Evidence from Estate Tax ReturnsrdquoNational Tax Journal 2004 57(2) pp 445ndash87

Kuznets Simon ldquoEconomic Growth and IncomeInequalityrdquo American Economic Review1955 45(1) pp 1ndash28

Lampman Robert J The share of top wealth-holders in national wealth 1922ndash1956Princeton Princeton University Press 1962

Lescure Michel Les banques lrsquoEtat et lemarche immobilier en France a lrsquoepoquecontemporaine 1820ndash1940 Paris EditionslrsquoEcole des Hautes Etudes en Sciences Socia-les 1982

Levy-Leboyer Maurice and Bourguignon Fran-cois LrsquoEconomie francaise au XIXe siecleAnalyse macroeconomique Paris Eco-nomica 1985 English translation Cam-bridge Cambridge University Press 1990

Lindert Peter H ldquoUnequal English Wealthsince 1670rdquo Journal of Political Economy1986 94(6) pp 1127ndash62

Lindert Peter H ldquoThree Centuries of Inequalityin Britain and Americardquo in Anthony B At-kinson and Francois Bourguignon eds

255VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

the average wealth levels for top fractiles (P99ndash100 P995ndash100 P999ndash100 and P9999ndash100)Lower thresholds of the national wealth distri-bution (P90 and P95) were computed using thenational TRA survey21 and the P90ndash95 andP95ndash99 intermediate wealth levels were alsocomputed using Pareto interpolation techniques

The national top estate shares estimates re-ported in Table 4 were computed using thismethodology They suggest that wealth concen-tration (as measured by the top-1-percent estateshare) rose throughout the nineteenth century inFrance during both the 1807ndash1867 and 1867ndash1902 periods although less sharply than in Parisduring the latter period (see Figure 7) Theseestimates are conservative in the sense that it isalmost certain that they underestimate the riseof wealth concentration that took place inFrance during the nineteenth century First weknow that the bulk of population growth in Parisduring the nineteenth century was due to theannexation of suburbs in 1860 and to populationgrowth in these peripheral arrondissements Be-

cause the outskirts of the city were poor theannexation added few top estates Thus there islittle doubt that Parisrsquos share of top estates inFrance actually increased less than its share of thetotal population This hypothesis is confirmed bynineteenth-century housing tax tabulations show-ing that the fraction of Paris taxpayers in thenational top 1 percent of taxpayers was substan-tially larger than 10 percent at the beginning of thenineteenth century22 Giving Paris a larger (andmore realistic) share of top estates in 1807 wouldboth reduce the share of wealth of the top 1percent in France at that date and lead to morerapid rise in inequality over time

Next and most importantly other estate sur-veys are consistent with the view that wealth in-equality was growing The important study byAdeline Daumard (1973) which relied on samplesof estate tax returns collected in Paris Lyon Tou-louse Lille and Bordeaux found that wealth con-centration increased in each of these five citiesduring the nineteenth century23 The TRA surveyalthough it is ill-suited for the study of top estatesis also consistent with our view Wealth dispersionwas on the rise in nineteenth-century France ac-cording to the TRA survey both in the sense thatthe fraction of decedents with positive estates de-clined over time (in spite of the sharp increase inthe value of the average estate) and that ratios suchas the P90P50 increased24 We also comparedour benchmark national P99 series extrapolatedfrom our Paris samples and the national P99series computed using the TRA survey We foundthat both series display the same overall upwardtrend in concentration (which is reassuring regard-ing the general validity of our Paris-France extrap-olation technique) except that the growth ofinequality from 1807 to 1902 in the TRA series is

21 See Bourdieu et al (2003) for full details about theTRA survey The P90 and P95 thresholds reported onPiketty et al (2004 Appendix Table A3) were computedusing ten-year moving averages around the target years inorder to make sure that each estimate was based on asufficient number of observations

22 These tabulations were published in the same FinanceMinistry official publications as the estate tabulations Wechose not to use them in our formal computations becausethe tax base of the housing tax (namely the rental value ofthe real estate property where the household lives) is onlyloosely connected to the estate tax base (in particular onecannot rule out the possibility that the housing tax baseoverrepresents Paris-based taxpayers)

23 Unfortunately Daumardrsquos samples are not available inmachine-readable format she has only two or three years ofdata for each city and she did not try to compute homog-enous inequality indicators (top fractiles shares etc) withher data Thus although her results and our work are con-sistent they cannot be compared directly

24 See Bourdieu et al (2004)

TABLE 4mdashWEALTH CONCENTRATION AT DEATH IN FRANCE1807ndash1994

Top-10-percentestate share

Top-1-percentestate share

Top-01-percentestate share

1807 791 434 1631817 810 445 1811827 824 452 1631837 796 438 1471847 816 479 1841857 829 495 1741867 810 480 1741877 838 471 2011887 839 487 1921902 839 516 2311913 863 549 2601929 820 502 2471938 776 420 1991947 699 299 1101956 694 304 1101994 610 213 63

Source Authorsrsquo computations using estate tax returns (seePiketty et al 2004 Table A3 for detailed series)

248 THE AMERICAN ECONOMIC REVIEW MARCH 2006

more severe than in our series (see Figure 8) Thisagain suggests that the latter provide a conserva-tive lower bound for the upward trend in wealthconcentration In any case the finding of a largeincrease in wealth inequality in nineteenth-centuryFrance (and up until World War I) appears to berobust25

As was mentioned earlier there exists nocomparable continuous data source covering thenineteenth and twentieth centuries in othercountries which makes it difficult to put ourFrench long-run series in international perspec-tive We note however that existing series forthe United States and the United Kingdom areconsistent with our findings on France26 Re-garding levels existing evidence suggests thatduring the nineteenth and most of the twentiethcenturies France was in an intermediate posi-tion in terms of wealth concentration in be-tween the United States (more equal) and theUnited Kingdom (the most unequal)27 These

25 Note that this continuous rise in wealth inequalitydoes not necessarily imply that a parallel rise occurredregarding income inequality Given that there exists nomicro source on incomes prior to the creation of the incometax in 1914 it is very difficult (if not impossible) to properlyaddress this issue Christian Morrisson and Wayne Snyder(2000) have attempted to link income inequality estimatesbased upon Old-Regime fiscal sources (pre-1789) withmodern income-tax-based twentieth-century estimates andthey have argued that income inequality might have startedto decline during the later part of the nineteenth century andon the eve of World War I (see also Morrisson 2000)Although our data do not allow us to rule out such apossibility we note that their nineteenth-century personaldistribution estimates are based on fragile macroeconomicdata on functional distribution and are not homogenous totheir eighteenth- and twentieth-century estimates Given theevidence that we provide on wealth inequality any signif-icant decline in aggregate income inequality would have tobe associated with severe compression of the wage distri-bution There is little research on this issue however

26 Lee Soltow and Jan L Van Zanden (1998) also find adecline in inequality in the twentieth century in the Neth-erlands Their data are consistent with a rise in inequality inthe nineteenth century but they have no direct evidenceabout its extent

27 According to our series the top-1-percent wealthshare in France rose from around 45 percent in 1800 toabout 55 percent around World War I and then fell to about20 percent by the end of the twentieth century (see Figure7) wealth concentration among the living appears to besomewhat larger (see Section V below) According to theseries pieced together by Lindert (2000 pp 181ndash82 and186) the UK top-1-percent wealth share rose from about

FIGURE 8 ESTIMATES OF THE P99 THRESHOLD FOR THE FRENCH DISTRIBUTION OF ESTATES AT DEATH EXTRAPOLATION

FROM PARIS SAMPLES VERSUS ESTIMATES FROM TRA SAMPLES (CURRENT FRENCH FRANCS)

Source Authorsrsquo computations based on estate tax returns (Paris samples and TRA samples)

249VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

differences in inequality largely hark back to dif-ferential concentration of landownership En-glandrsquos land was extremely narrowly held whilethe United States was most egalitarian The impactof access to real estate assets can also be seenwhen we contrast Paris (where it was extremelyconcentrated) with the provinces

Trends in inequality over time may be easierto compare because biases in source materialmay have a significant effect on levels ratherthan trends The historical pattern is similar ineach of the three economies In particular thereis evidence that wealth concentration increasedduring the nineteenth century in both Anglo-Saxon countries and declined during the twen-tieth century with a turning point around WorldWar I28 Neither exhibits patterns consistentwith a Kuznet process

IV From the Wealth of the Decedents to theWealth of the Living

The estimates reported thus far refer to in-equality among decedents as described in thetax returns filed by their heirs The evolution ofthe distribution of wealth among the livingmight however have followed a different pat-tern In order to convert wealth-at-death con-centration estimates into wealth-of-the-livingconcentration estimates it is standard to use theldquoestate multiplierrdquo method29 It consists ofweighting each observation of an estate at death

by the inverse of the mortality rate for this agegroup That is if the mortality rate for ages 20to 24 was 068 percent in Paris in 1902 theneach decedent aged 20 to 24 represented about147 living individuals of the same age (100068 147) Similarly if the mortality ratefor ages above 80 was 2143 percent in Paris in1902 then each decedent in that group repre-sented about 47 living individuals in the sameage group (102143 47) Applying thismethod requires mortality tables (these are eas-ily available) and estate tabulations brokendown by estate size and age at death (these arescarcer) Fortunately the cityrsquos statistical bu-reau published annual death-by-age totals theFrench censuses report the age distribution forthe capital every five years and we collectedage at death from the estate declarations Thesedata allowed us to compute estimates of wealthconcentration among the living over the 1807ndash1902 period using various assumptions aboutthe wealth profiles of mortality rates

The base population for the living is the set ofall individuals aged 20 and over living in Parisin year t which we denote pt The number ofliving individuals in age bracket a is denoted pta(a 20ndash24 25ndash29 30ndash34 75ndash79 80 andover) and the number of decedents in agebracket a is denoted nta The mortality rate forage bracket a is given by mta ntapta Webegin with a uniform-mortality benchmarkThese estimates are based on the simplifyingassumption that these mortality rates dependsolely on age and are the same for all wealthgroups (and in particular are the same forzero-wealth and positive-wealth individuals)We can then weigh each decedent with positiveestate and age a collected in the Paris archivesin year t by ptanta This allows us to computethe number of living Parisians with positivewealth in year t and also (by differentiatingwith pt) the number of living Parisians with zerowealth at year t which is used to weight zero-estate observations We then use our weighteddatasets to compute top estate fractiles amongthe living in Paris

The main conclusion is that the living expe-rienced the same upward trend in wealth con-centration as the decedents (see Figure 9) Wefind that inequality was significantly higheramong the living than among decedents be-cause survivors were on average younger than

55 percent in 1800 to 70 percent around World War I thenfell to about 20 percent in the 1990s The US top-1-percentwealth share rose from about 15 to 20 percent in 1800 toabout 40 percent around World War I then fell to about 30percent in the 1990s (and as low as 20 to 25 percentaccording to the more recent estimates from Kopczuk andSaez 2004) Wealth concentration is now larger in theUnited States than in European countries but the reversewas true during the nineteenth century up until World WarII (It is only since the 1950sndash1970s period that US wealthconcentration has been somewhat larger)

28 See Lindert (2000 pp 181ndash82 and 188)29 This method was widely used in England and France

in the late nineteenth and early twentieth centuries to com-pute the stock of national wealth on the basis of the flow ofwealth transmitted at death Standard references that use thistechnique to estimate the wealth distribution of the livingfrom estate tax data tabulated by estate size and age at deathinclude Atkinson and Harrison (1978) and Lampman(1962) For a more recent application of this technique tothe United States see Kopczuk and Saez (2004)

250 THE AMERICAN ECONOMIC REVIEW MARCH 2006

those who died and the young were on aver-age poorer In particular the estate multipliermethod leads to lower average weights forpositive-wealth decedents than for zero-wealthdecedents (the former are on average older andtherefore ldquorepresentrdquo a smaller number of livingindividuals) As a result the fraction of positive-wealth individuals is even smaller among theliving than among decedents Hence our bench-mark uniform-mortality estimates of wealthconcentration among the living are significantlylarger than corresponding estimates among de-cedents (eg top-1-percent wealth shares areabout 15 percent larger among the living)Changes over time however are similar In-creased life expectancy and declining mortalityrates over the course of the nineteenth centuryhave only a small effect on the trends

In order to make the estate multiplier methodmore reliable one would prefer to take intoaccount differential mortality by wealth Doingso would require having access to mortalityschedules based both on wealth and age at dif-ferent points in time unfortunately these are notavailable We have nonetheless reestimatedwealth of the living based upon the same as-sumption as Kopczuk and Saez (2004) That is

we assumed uniform mortality among the poor(here defined as zero-wealth individuals) andamong the rich (here defined as positive-wealthindividuals) and we assumed that the ratiomtaRmtaP between the mortality rate of the richand the mortality rate of the poor followed aU-shaped age profile from about 85 percent forthe young (ie the rich die 15 percent less oftenthan the poor when they are 20ndash24 or 25ndash29years old) down to about 70 percent for middle-age individuals in their forties to fifties and upto 100 percent for very old individuals in theireighties to nineties30 This profile correspondsto the best available estimates in the literatureand it appears to be relatively stable over timeand across developed countries In the absenceof better data it is the best one can do31 Thebenchmark differential-mortality estimates re-ported on Figure 9 show that although addingdifferential mortality produces different levelsof inequality it does not have much impact onthe upward trend in concentration

30 See Kopczuk and Saez (2004 Table A4)31 See Kopczuk and Saez (2004 Appendix B) for refer-

ences to the US and international literature devoted to theage-wealth profile of mortality rates

FIGURE 9 WEALTH CONCENTRATION AMONG DECEDENTS AND AMONG THE LIVING IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives (see Piketty et al 2004Table A4 for detailed series)

251VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

The resulting differential-mortality inequalityestimates lie between those based on decedentsonly and those using uniform mortality for theliving (see Figure 9) Note that moving fromuniform-mortality to differential-mortality es-tate multiplier methodology can either increaseor decrease inequality Here the reason whysuch a move leads to lower wealth concentra-tion seems to be due to the fact that differentialmortality tends to put higher weights onpositive-wealth decedents (for a given age)thereby increasing the estimated fraction of liv-ing individuals with positive wealth The im-portant point however is that the resulting leveleffects are relatively small in magnitude con-stant in time and dwarfed by the upward timetrend Even if we were to assume an enormousincrease in differential mortality during thenineteenth century in the sense that differentialmortality between the rich and poor was equalto 0 percent of the benchmark differential in1807 and 100 percent of the benchmark differ-ential in 1902 the resulting wealth concentra-tion estimates would still be significantly higherin 1902 than in 1807 (see Figure 9) Yet wehave no reason to believe that differential mor-tality increased to such an extent During thenineteenth century real wages for unskilledworkers rose which would have reduced mor-tality more for the poor than for the rich After1850 public health measures (sanitation watervaccination) were in place Again these wouldhave had a significant effect on the poor and themiddle class who could not privately purchasesuch health-improving services To be sure therich could avail themselves of more medicalservices than the poor or the middle class butthe impact of these services was probably small(bear in mind that neither antibiotics nor car-diovascular interventions were available)

Finally we have applied the estate multipliermethod to available data for 1947 and 1994 Over-all the sharp decline in wealth concentration ob-served during the twentieth century (andespecially between 1914 and 1945) is very robustIf anything the decline appears to be even largerwhen one looks at wealth concentration amongthe living rather than among decedents32

V The Changing Age Profile of Wealth

In the previous sections we focused almostexclusively on aggregate top wealth shares Ourdata however also detail the characteristics ofeach decedent in particular their gender andage The evolution of wealth by gender is ofrelevance for over the past two centuries therehave been massive changes in womenrsquos laborforce participation capacity to manage theirown affairs and life expectancy relative to menThe evolution of wealth by age is of relevancebecause there was a significant increase in adultlife expectancy over the twentieth century andbecause the progressive diffusion of pensionsmay have changed savings motivations More-over age-wealth profiles also inform us aboutthe motives of wealth accumulation and theeconomic impact of high wealth concentration

A first pass at the data considers the genderbreakdown of wealth at death Remarkably inour micro data the share of women in top estatestakes its highest value on the eve of World WarI For instance the womenrsquos share in the top 05percent rose from 35 percent prior to 1850 to 45percent in 1902 only to fall to 40 percent afterWorld War II Strikingly womenrsquos share ofwealth follows almost exactly the pattern ofaggregate inequality Women were relativelyricher when inequality reached its apex inFrance than at any other time Moreover insti-tutional variables seem to have played almostno role in changing the relative wealth ofwomen Unlike in common law countriesFrench law starting with the code civil of 1804required nearly equal treatment of all children inbequests Further research will help us deter-mine to what extent women of great wealthwere heirs or part of economically very success-ful couples

The data also reveal striking changes in the

32 See Piketty et al (2004 Table A4) It is unfortunatelynot possible to construct complete series for wealth concen-

tration among the living for the twentieth century due todata limitations tables broken down by estate brackets andage of decedents are available solely for years 1943ndash1954and at the national level (no table broken down by estatebrackets and age of decedents has ever been compiled at thedepartement level except in 1931 for Seine departementsee E S Danysz 1934) and the 1994 micro sample is notlarge enough to allow for a reliable application of the estatemultiplier method at the Paris level Thus the only wealth-of-the-living concentration estimates we provide for thetwentieth century are national estimates for 1947 and 1994

252 THE AMERICAN ECONOMIC REVIEW MARCH 2006

age profile of wealth between 1807 and 1994(see Table 5) During the nineteenth century aswealth concentration was increasing the veryrich were getting older and older At the begin-ning of the nineteenth century in the aftermathof the French Revolution the richest individualswere those in their fifties they were typically100 percent richer on average than people intheir forties 25 percent richer than those in theirsixties and 40 percent richer than those in theirseventies and eighties By the 1870s howeverthe age-wealth pattern had become stronglymonotonic the richest individuals were the old-est individuals In 1902 people in their sixtiesand seventies bequeathed 150 percent morethan those in their fifties and those in theireighties 300 percent more On the eve of WorldWar I top wealth holders were old and likely tobe retired This pattern breaks some time duringthe 1914ndash1945 period33 In 1947 as well as in1994 we are back to a pattern where the richestindividuals are those in their fifties Overall theperiod of maximal wealth inequality (1860ndash1913) also appears to be a period characterizedby a very specific age profile of wealth and largeconcentration of assets among the elderly

Another way to analyze the changing age-wealth relationship is to look at average age by

top estate fractile34 In 1817 average age wasvirtually the same for the top 10 percent and thetop 1 percent of estates (or even slightly declin-ing) The average-age-per-fractile relationshipbecomes upward sloping during the nineteenthcentury and by 1902 those in the top 1 percentwere almost six years older than those in the top10 percent The relationship is flat in 1947 anddownward-sloping in 1994 Finally one canapply the estate multiplier method (see SectionIV above) and analyze how wealth concentra-tion by age group among the living changedover the course of the nineteenth century Thegeneral population in Paris did not becomeolder during the nineteenth century those aged60 or older made up about 15 percent of thepopulation in 1817 and after 1847 about 10 to11 percent35 The share of total wealth ownedby the elderly rose significantly however aswealth distribution worsened The wealth be-longing to those aged 60 or more rose fromabout 25 to 30 percent of the total at the begin-ning of the nineteenth century to about 40 to 45percent by the end of the century The wealthshare of those aged 70 or older doubled fromless than 10 percent to about 20 percent36

33 Existing evidence on the age-wealth profile for 1931(see Danysz 1934) suggests that the Great Depression andWorld War II (rather than World War I) played the leadingroles in breaking this pattern This is an issue we plan toinvestigate in future research

34 See Piketty et al (2004 Table 6)35 Although life expectancy was increasing which

should have led to large shares of population for oldergroups the city was also growing quickly The large num-ber of immigrants (who were typically in their twenties)increased the relative size of the younger cohorts (seePiketty et al 2004 Table 7 and Figure 11)

36 See Piketty et al (2004 Figure 12)

TABLE 5mdashTHE AGE PROFILE OF WEALTH AT DEATH IN PARIS 1817ndash1994(Average estate left by 50- 59-year-old 100)

20ndash29yr-old

30ndash39yr-old

40ndash49yr-old

50ndash59yr-old

60ndash69yr-old

70ndash79yr-old

80ndash89yr-old

90ndash99yr-old

1817 26 22 28 100 54 59 591827 44 50 53 100 88 87 601837 133 90 107 100 116 123 1101847 87 73 102 100 117 204 1321857 84 77 101 100 104 109 1451867 67 58 136 100 141 125 1541877 66 73 63 100 197 260 4301887 45 33 63 100 152 233 2951902 29 40 80 100 253 272 4011947 31 51 73 100 113 105 105 1091994 11 45 100 87 93 95 68

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources)

253VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

It is perhaps not surprising that inequalitybecame strongly correlated with age in the1860s Those who died at an old age in the1820s and 1830s had lived through the difficultyears of the French Revolution and the disloca-tion of the end of the Napoleonic period Theirability to accumulate wealth had been severelyimpaired Furthermore they were rather lesslikely to inherit much wealth since the Revo-lution wiped out the bond portfolios of theirparents through a prolonged period of high in-flation Those who died from the 1860s to theearly 1910s did not suffer from the adversitiesthat plagued their forebears Instead they en-joyed the fruits of the financial sector expansionthat began in the 1850s After 1947 we seem tohave returned to a situation quite like that of1817 Presumably the capital damages associ-ated with both world wars and the Great De-pression had a strong negative effect on thewealth holdings of older generations The per-sistence of a flat age-wealth profile until 1994 islikely to be associated with two factors First insocieties where income growth is rapid abso-lute wealth accumulation is faster by youngercohorts than by older ones because their in-comes are higher at every age This is an im-portant distinction between the nineteenthcentury and the twentieth century Furthermorehighly progressive rates of income and estatetaxation have probably made it more difficult toaccumulate large fortunes thereby flattening theobserved age-wealth profile

The more interesting (and more difficult)question relates to the possible efficiency im-pact of high wealth concentration and changingage-wealth profiles Although our data do notallow us to address efficiency issues in a rigor-ous way our results allow us to formulate anumber of hypotheses and to shed new light onthe ongoing debate on inequality and growth37

From a theoretical viewpoint whether highwealth concentration can have a negativegrowth impact depends critically on the exis-tence of credit constraints With first-best credit

markets money flows toward the best entrepre-neurs and investment projects irrespective ofthe initial distribution High levels of wealthconcentration can be bad from a social justiceviewpoint but they entail no efficiency lossWhen credit constraints bind however initialwealth matters and high levels of inequality canhurt growth Whether the loss is large or smalldepends on who owns the assets If the rich areefficient investors (they know which projects tofund etc) then wealth concentration may evenbe useful If the rich are retired rentiers how-ever investing their wealth in low-yield assets(or low-ability inheritors) then high wealthconcentration and credit constraints might pre-vent talented but penniless investors from un-dertaking efficient projects thereby entailingnegative growth consequences The data used inthis paper are not ideal to address whether creditconstraints were important in a country likeFrance at the end of the nineteenth century Ourresults suggest however that to the extentcredit constraints were indeed severe highwealth concentration did have a negativegrowth impact In order to investigate this hy-pothesis further one would need to gather moresystematic data on investment strategies andasset returns Preliminary evidence suggeststhat the rich elderly of the 1860ndash1913 perioddid indeed hold a disproportionate fraction oftheir wealth in low-yield assets (such as gov-ernment bonds) An alternative hypothesishowever is that steeper age-wealth profileswere the consequence of the growth of financialmarkets as their children faced fewer creditconstraints parents decided to hold on to moreof their wealth

VI Conclusion

Evidence from wealth at death in Paris and inFrance over the last two centuries reveals threekey patterns First wealth concentration haschanged dramatically over time In 1807 thetop-1-percent share of wealth (40 percent inFrance 50 percent in Paris) was twice as high asit would be in 1994 but substantially less thanin 1913 when it peaked above 55 percent inFrance and 70 percent in Paris Some of thesechanges were due to economic phenomena thathave long been emphasized as creating inequal-ity namely industrialization and financial cen-

37 Thus far this literature has concentrated upon cross-country regressions of inequality on growth a methodologythat raises serious identification problems especially giventhe low quality of available international datasets on in-equality which are neither long-run nor homogeneous (seeeg Atkinson and Andrea Brandolini 2001)

254 THE AMERICAN ECONOMIC REVIEW MARCH 2006

tralization Yet the decline comes largely fromadverse shocks rather than economic conver-gence These changes are of such magnitudethat they are not sensitive to whether one ex-amines wealth at death in Paris or in France orwhether one examines it directly rather thanconverting it to wealth of the living by an estatemultiplier method

Our second key result is that there was asignificant transition during the nineteenth cen-tury from an important role for real estate as aform of wealth to moveable assets as the keyform of wealth for the very rich Similarly theshare of wealth held by aristocrats first rose andthen was eclipsed by that of financiers and in-dustrialists in the second half of the nineteenthcentury Hence mobility within this highly un-equal society might have been quite high Yetthis conjecture is tempered by our third findingthe wealthy were getting older over time andolder relative to less wealthy decedents Suchaging among the very wealthy would have hadnegative consequences for growth if financialmarkets were imperfect This issue requires fur-ther investigation and we hope it will attractfuture research

REFERENCES

Atkinson Anthony B ldquoTop Incomes in the UKover the 20th Centuryrdquo Journal of the RoyalStatistical Society Series A (Statistics in So-ciety) 2005 168(2) pp 325ndash43

Atkinson Anthony B and Brandolini AndrealdquoPromise and Pitfalls in the Use of lsquoSecond-aryrsquo Data-Sets Income Inequality in OECDCountries as a Case Studyrdquo Journal of Eco-nomic Literature 2001 39(3) pp 771ndash99

Atkinson Anthony B and Harrison Alan J Dis-tribution of personal wealth in Britain Cam-bridge Cambridge University Press 1978

Atkinson Anthony B and Piketty Thomas Topincomes over the twentieth century A con-trast between European and English speak-ing countries Oxford Oxford UniversityPress (forthcoming)

Banerjee Abhijit V and Duflo Esther ldquoInequalityand Growth What Can the Data Sayrdquo Journalof Economic Growth 2003 8(3) pp 267ndash99

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoPourquoi la richesse ne

srsquoest-elle pas diffusee avec la croissance Ledegre zero de lrsquoinegalite et son evolution enFrance 1800ndash1940rdquo Histoire et mesure 200318(1ndash2) pp 147ndash98

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoDefense et illustration delrsquoenquete des 3000 familles Lrsquoexemple deson volet patrimonialrdquo Annales de demogra-phie historique 2004 19 pp 19ndash57

Danysz E S ldquoContribution a lrsquoetude des for-tunes privees drsquoapres les declarations de suc-cessionsrdquo Bulletin de la statistique generalede France 1934 XXIV (1) pp 111ndash171

Daumard Adeline Les fortunes francaise auXIXe siecle Enquete sur la repartition et lacomposition des capitaux prives a ParisLyon Lille Bordeaux et Toulouse drsquoapreslrsquoenregistrement des declarations de succes-sion Paris Mouton 1973

Gain Andre La restauration et les biens desemigres 2 vols Nancy Societe drsquoimpressiontypographique 1929

Jones Alice H American colonial wealth Doc-uments and methods 3 vols New York ArnoPress 1977

Kopczuk Wojciech and Saez Emmanuel ldquoTopWealth Shares in the United States 1916ndash2000 Evidence from Estate Tax ReturnsrdquoNational Tax Journal 2004 57(2) pp 445ndash87

Kuznets Simon ldquoEconomic Growth and IncomeInequalityrdquo American Economic Review1955 45(1) pp 1ndash28

Lampman Robert J The share of top wealth-holders in national wealth 1922ndash1956Princeton Princeton University Press 1962

Lescure Michel Les banques lrsquoEtat et lemarche immobilier en France a lrsquoepoquecontemporaine 1820ndash1940 Paris EditionslrsquoEcole des Hautes Etudes en Sciences Socia-les 1982

Levy-Leboyer Maurice and Bourguignon Fran-cois LrsquoEconomie francaise au XIXe siecleAnalyse macroeconomique Paris Eco-nomica 1985 English translation Cam-bridge Cambridge University Press 1990

Lindert Peter H ldquoUnequal English Wealthsince 1670rdquo Journal of Political Economy1986 94(6) pp 1127ndash62

Lindert Peter H ldquoThree Centuries of Inequalityin Britain and Americardquo in Anthony B At-kinson and Francois Bourguignon eds

255VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

more severe than in our series (see Figure 8) Thisagain suggests that the latter provide a conserva-tive lower bound for the upward trend in wealthconcentration In any case the finding of a largeincrease in wealth inequality in nineteenth-centuryFrance (and up until World War I) appears to berobust25

As was mentioned earlier there exists nocomparable continuous data source covering thenineteenth and twentieth centuries in othercountries which makes it difficult to put ourFrench long-run series in international perspec-tive We note however that existing series forthe United States and the United Kingdom areconsistent with our findings on France26 Re-garding levels existing evidence suggests thatduring the nineteenth and most of the twentiethcenturies France was in an intermediate posi-tion in terms of wealth concentration in be-tween the United States (more equal) and theUnited Kingdom (the most unequal)27 These

25 Note that this continuous rise in wealth inequalitydoes not necessarily imply that a parallel rise occurredregarding income inequality Given that there exists nomicro source on incomes prior to the creation of the incometax in 1914 it is very difficult (if not impossible) to properlyaddress this issue Christian Morrisson and Wayne Snyder(2000) have attempted to link income inequality estimatesbased upon Old-Regime fiscal sources (pre-1789) withmodern income-tax-based twentieth-century estimates andthey have argued that income inequality might have startedto decline during the later part of the nineteenth century andon the eve of World War I (see also Morrisson 2000)Although our data do not allow us to rule out such apossibility we note that their nineteenth-century personaldistribution estimates are based on fragile macroeconomicdata on functional distribution and are not homogenous totheir eighteenth- and twentieth-century estimates Given theevidence that we provide on wealth inequality any signif-icant decline in aggregate income inequality would have tobe associated with severe compression of the wage distri-bution There is little research on this issue however

26 Lee Soltow and Jan L Van Zanden (1998) also find adecline in inequality in the twentieth century in the Neth-erlands Their data are consistent with a rise in inequality inthe nineteenth century but they have no direct evidenceabout its extent

27 According to our series the top-1-percent wealthshare in France rose from around 45 percent in 1800 toabout 55 percent around World War I and then fell to about20 percent by the end of the twentieth century (see Figure7) wealth concentration among the living appears to besomewhat larger (see Section V below) According to theseries pieced together by Lindert (2000 pp 181ndash82 and186) the UK top-1-percent wealth share rose from about

FIGURE 8 ESTIMATES OF THE P99 THRESHOLD FOR THE FRENCH DISTRIBUTION OF ESTATES AT DEATH EXTRAPOLATION

FROM PARIS SAMPLES VERSUS ESTIMATES FROM TRA SAMPLES (CURRENT FRENCH FRANCS)

Source Authorsrsquo computations based on estate tax returns (Paris samples and TRA samples)

249VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

differences in inequality largely hark back to dif-ferential concentration of landownership En-glandrsquos land was extremely narrowly held whilethe United States was most egalitarian The impactof access to real estate assets can also be seenwhen we contrast Paris (where it was extremelyconcentrated) with the provinces

Trends in inequality over time may be easierto compare because biases in source materialmay have a significant effect on levels ratherthan trends The historical pattern is similar ineach of the three economies In particular thereis evidence that wealth concentration increasedduring the nineteenth century in both Anglo-Saxon countries and declined during the twen-tieth century with a turning point around WorldWar I28 Neither exhibits patterns consistentwith a Kuznet process

IV From the Wealth of the Decedents to theWealth of the Living

The estimates reported thus far refer to in-equality among decedents as described in thetax returns filed by their heirs The evolution ofthe distribution of wealth among the livingmight however have followed a different pat-tern In order to convert wealth-at-death con-centration estimates into wealth-of-the-livingconcentration estimates it is standard to use theldquoestate multiplierrdquo method29 It consists ofweighting each observation of an estate at death

by the inverse of the mortality rate for this agegroup That is if the mortality rate for ages 20to 24 was 068 percent in Paris in 1902 theneach decedent aged 20 to 24 represented about147 living individuals of the same age (100068 147) Similarly if the mortality ratefor ages above 80 was 2143 percent in Paris in1902 then each decedent in that group repre-sented about 47 living individuals in the sameage group (102143 47) Applying thismethod requires mortality tables (these are eas-ily available) and estate tabulations brokendown by estate size and age at death (these arescarcer) Fortunately the cityrsquos statistical bu-reau published annual death-by-age totals theFrench censuses report the age distribution forthe capital every five years and we collectedage at death from the estate declarations Thesedata allowed us to compute estimates of wealthconcentration among the living over the 1807ndash1902 period using various assumptions aboutthe wealth profiles of mortality rates

The base population for the living is the set ofall individuals aged 20 and over living in Parisin year t which we denote pt The number ofliving individuals in age bracket a is denoted pta(a 20ndash24 25ndash29 30ndash34 75ndash79 80 andover) and the number of decedents in agebracket a is denoted nta The mortality rate forage bracket a is given by mta ntapta Webegin with a uniform-mortality benchmarkThese estimates are based on the simplifyingassumption that these mortality rates dependsolely on age and are the same for all wealthgroups (and in particular are the same forzero-wealth and positive-wealth individuals)We can then weigh each decedent with positiveestate and age a collected in the Paris archivesin year t by ptanta This allows us to computethe number of living Parisians with positivewealth in year t and also (by differentiatingwith pt) the number of living Parisians with zerowealth at year t which is used to weight zero-estate observations We then use our weighteddatasets to compute top estate fractiles amongthe living in Paris

The main conclusion is that the living expe-rienced the same upward trend in wealth con-centration as the decedents (see Figure 9) Wefind that inequality was significantly higheramong the living than among decedents be-cause survivors were on average younger than

55 percent in 1800 to 70 percent around World War I thenfell to about 20 percent in the 1990s The US top-1-percentwealth share rose from about 15 to 20 percent in 1800 toabout 40 percent around World War I then fell to about 30percent in the 1990s (and as low as 20 to 25 percentaccording to the more recent estimates from Kopczuk andSaez 2004) Wealth concentration is now larger in theUnited States than in European countries but the reversewas true during the nineteenth century up until World WarII (It is only since the 1950sndash1970s period that US wealthconcentration has been somewhat larger)

28 See Lindert (2000 pp 181ndash82 and 188)29 This method was widely used in England and France

in the late nineteenth and early twentieth centuries to com-pute the stock of national wealth on the basis of the flow ofwealth transmitted at death Standard references that use thistechnique to estimate the wealth distribution of the livingfrom estate tax data tabulated by estate size and age at deathinclude Atkinson and Harrison (1978) and Lampman(1962) For a more recent application of this technique tothe United States see Kopczuk and Saez (2004)

250 THE AMERICAN ECONOMIC REVIEW MARCH 2006

those who died and the young were on aver-age poorer In particular the estate multipliermethod leads to lower average weights forpositive-wealth decedents than for zero-wealthdecedents (the former are on average older andtherefore ldquorepresentrdquo a smaller number of livingindividuals) As a result the fraction of positive-wealth individuals is even smaller among theliving than among decedents Hence our bench-mark uniform-mortality estimates of wealthconcentration among the living are significantlylarger than corresponding estimates among de-cedents (eg top-1-percent wealth shares areabout 15 percent larger among the living)Changes over time however are similar In-creased life expectancy and declining mortalityrates over the course of the nineteenth centuryhave only a small effect on the trends

In order to make the estate multiplier methodmore reliable one would prefer to take intoaccount differential mortality by wealth Doingso would require having access to mortalityschedules based both on wealth and age at dif-ferent points in time unfortunately these are notavailable We have nonetheless reestimatedwealth of the living based upon the same as-sumption as Kopczuk and Saez (2004) That is

we assumed uniform mortality among the poor(here defined as zero-wealth individuals) andamong the rich (here defined as positive-wealthindividuals) and we assumed that the ratiomtaRmtaP between the mortality rate of the richand the mortality rate of the poor followed aU-shaped age profile from about 85 percent forthe young (ie the rich die 15 percent less oftenthan the poor when they are 20ndash24 or 25ndash29years old) down to about 70 percent for middle-age individuals in their forties to fifties and upto 100 percent for very old individuals in theireighties to nineties30 This profile correspondsto the best available estimates in the literatureand it appears to be relatively stable over timeand across developed countries In the absenceof better data it is the best one can do31 Thebenchmark differential-mortality estimates re-ported on Figure 9 show that although addingdifferential mortality produces different levelsof inequality it does not have much impact onthe upward trend in concentration

30 See Kopczuk and Saez (2004 Table A4)31 See Kopczuk and Saez (2004 Appendix B) for refer-

ences to the US and international literature devoted to theage-wealth profile of mortality rates

FIGURE 9 WEALTH CONCENTRATION AMONG DECEDENTS AND AMONG THE LIVING IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives (see Piketty et al 2004Table A4 for detailed series)

251VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

The resulting differential-mortality inequalityestimates lie between those based on decedentsonly and those using uniform mortality for theliving (see Figure 9) Note that moving fromuniform-mortality to differential-mortality es-tate multiplier methodology can either increaseor decrease inequality Here the reason whysuch a move leads to lower wealth concentra-tion seems to be due to the fact that differentialmortality tends to put higher weights onpositive-wealth decedents (for a given age)thereby increasing the estimated fraction of liv-ing individuals with positive wealth The im-portant point however is that the resulting leveleffects are relatively small in magnitude con-stant in time and dwarfed by the upward timetrend Even if we were to assume an enormousincrease in differential mortality during thenineteenth century in the sense that differentialmortality between the rich and poor was equalto 0 percent of the benchmark differential in1807 and 100 percent of the benchmark differ-ential in 1902 the resulting wealth concentra-tion estimates would still be significantly higherin 1902 than in 1807 (see Figure 9) Yet wehave no reason to believe that differential mor-tality increased to such an extent During thenineteenth century real wages for unskilledworkers rose which would have reduced mor-tality more for the poor than for the rich After1850 public health measures (sanitation watervaccination) were in place Again these wouldhave had a significant effect on the poor and themiddle class who could not privately purchasesuch health-improving services To be sure therich could avail themselves of more medicalservices than the poor or the middle class butthe impact of these services was probably small(bear in mind that neither antibiotics nor car-diovascular interventions were available)

Finally we have applied the estate multipliermethod to available data for 1947 and 1994 Over-all the sharp decline in wealth concentration ob-served during the twentieth century (andespecially between 1914 and 1945) is very robustIf anything the decline appears to be even largerwhen one looks at wealth concentration amongthe living rather than among decedents32

V The Changing Age Profile of Wealth

In the previous sections we focused almostexclusively on aggregate top wealth shares Ourdata however also detail the characteristics ofeach decedent in particular their gender andage The evolution of wealth by gender is ofrelevance for over the past two centuries therehave been massive changes in womenrsquos laborforce participation capacity to manage theirown affairs and life expectancy relative to menThe evolution of wealth by age is of relevancebecause there was a significant increase in adultlife expectancy over the twentieth century andbecause the progressive diffusion of pensionsmay have changed savings motivations More-over age-wealth profiles also inform us aboutthe motives of wealth accumulation and theeconomic impact of high wealth concentration

A first pass at the data considers the genderbreakdown of wealth at death Remarkably inour micro data the share of women in top estatestakes its highest value on the eve of World WarI For instance the womenrsquos share in the top 05percent rose from 35 percent prior to 1850 to 45percent in 1902 only to fall to 40 percent afterWorld War II Strikingly womenrsquos share ofwealth follows almost exactly the pattern ofaggregate inequality Women were relativelyricher when inequality reached its apex inFrance than at any other time Moreover insti-tutional variables seem to have played almostno role in changing the relative wealth ofwomen Unlike in common law countriesFrench law starting with the code civil of 1804required nearly equal treatment of all children inbequests Further research will help us deter-mine to what extent women of great wealthwere heirs or part of economically very success-ful couples

The data also reveal striking changes in the

32 See Piketty et al (2004 Table A4) It is unfortunatelynot possible to construct complete series for wealth concen-

tration among the living for the twentieth century due todata limitations tables broken down by estate brackets andage of decedents are available solely for years 1943ndash1954and at the national level (no table broken down by estatebrackets and age of decedents has ever been compiled at thedepartement level except in 1931 for Seine departementsee E S Danysz 1934) and the 1994 micro sample is notlarge enough to allow for a reliable application of the estatemultiplier method at the Paris level Thus the only wealth-of-the-living concentration estimates we provide for thetwentieth century are national estimates for 1947 and 1994

252 THE AMERICAN ECONOMIC REVIEW MARCH 2006

age profile of wealth between 1807 and 1994(see Table 5) During the nineteenth century aswealth concentration was increasing the veryrich were getting older and older At the begin-ning of the nineteenth century in the aftermathof the French Revolution the richest individualswere those in their fifties they were typically100 percent richer on average than people intheir forties 25 percent richer than those in theirsixties and 40 percent richer than those in theirseventies and eighties By the 1870s howeverthe age-wealth pattern had become stronglymonotonic the richest individuals were the old-est individuals In 1902 people in their sixtiesand seventies bequeathed 150 percent morethan those in their fifties and those in theireighties 300 percent more On the eve of WorldWar I top wealth holders were old and likely tobe retired This pattern breaks some time duringthe 1914ndash1945 period33 In 1947 as well as in1994 we are back to a pattern where the richestindividuals are those in their fifties Overall theperiod of maximal wealth inequality (1860ndash1913) also appears to be a period characterizedby a very specific age profile of wealth and largeconcentration of assets among the elderly

Another way to analyze the changing age-wealth relationship is to look at average age by

top estate fractile34 In 1817 average age wasvirtually the same for the top 10 percent and thetop 1 percent of estates (or even slightly declin-ing) The average-age-per-fractile relationshipbecomes upward sloping during the nineteenthcentury and by 1902 those in the top 1 percentwere almost six years older than those in the top10 percent The relationship is flat in 1947 anddownward-sloping in 1994 Finally one canapply the estate multiplier method (see SectionIV above) and analyze how wealth concentra-tion by age group among the living changedover the course of the nineteenth century Thegeneral population in Paris did not becomeolder during the nineteenth century those aged60 or older made up about 15 percent of thepopulation in 1817 and after 1847 about 10 to11 percent35 The share of total wealth ownedby the elderly rose significantly however aswealth distribution worsened The wealth be-longing to those aged 60 or more rose fromabout 25 to 30 percent of the total at the begin-ning of the nineteenth century to about 40 to 45percent by the end of the century The wealthshare of those aged 70 or older doubled fromless than 10 percent to about 20 percent36

33 Existing evidence on the age-wealth profile for 1931(see Danysz 1934) suggests that the Great Depression andWorld War II (rather than World War I) played the leadingroles in breaking this pattern This is an issue we plan toinvestigate in future research

34 See Piketty et al (2004 Table 6)35 Although life expectancy was increasing which

should have led to large shares of population for oldergroups the city was also growing quickly The large num-ber of immigrants (who were typically in their twenties)increased the relative size of the younger cohorts (seePiketty et al 2004 Table 7 and Figure 11)

36 See Piketty et al (2004 Figure 12)

TABLE 5mdashTHE AGE PROFILE OF WEALTH AT DEATH IN PARIS 1817ndash1994(Average estate left by 50- 59-year-old 100)

20ndash29yr-old

30ndash39yr-old

40ndash49yr-old

50ndash59yr-old

60ndash69yr-old

70ndash79yr-old

80ndash89yr-old

90ndash99yr-old

1817 26 22 28 100 54 59 591827 44 50 53 100 88 87 601837 133 90 107 100 116 123 1101847 87 73 102 100 117 204 1321857 84 77 101 100 104 109 1451867 67 58 136 100 141 125 1541877 66 73 63 100 197 260 4301887 45 33 63 100 152 233 2951902 29 40 80 100 253 272 4011947 31 51 73 100 113 105 105 1091994 11 45 100 87 93 95 68

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources)

253VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

It is perhaps not surprising that inequalitybecame strongly correlated with age in the1860s Those who died at an old age in the1820s and 1830s had lived through the difficultyears of the French Revolution and the disloca-tion of the end of the Napoleonic period Theirability to accumulate wealth had been severelyimpaired Furthermore they were rather lesslikely to inherit much wealth since the Revo-lution wiped out the bond portfolios of theirparents through a prolonged period of high in-flation Those who died from the 1860s to theearly 1910s did not suffer from the adversitiesthat plagued their forebears Instead they en-joyed the fruits of the financial sector expansionthat began in the 1850s After 1947 we seem tohave returned to a situation quite like that of1817 Presumably the capital damages associ-ated with both world wars and the Great De-pression had a strong negative effect on thewealth holdings of older generations The per-sistence of a flat age-wealth profile until 1994 islikely to be associated with two factors First insocieties where income growth is rapid abso-lute wealth accumulation is faster by youngercohorts than by older ones because their in-comes are higher at every age This is an im-portant distinction between the nineteenthcentury and the twentieth century Furthermorehighly progressive rates of income and estatetaxation have probably made it more difficult toaccumulate large fortunes thereby flattening theobserved age-wealth profile

The more interesting (and more difficult)question relates to the possible efficiency im-pact of high wealth concentration and changingage-wealth profiles Although our data do notallow us to address efficiency issues in a rigor-ous way our results allow us to formulate anumber of hypotheses and to shed new light onthe ongoing debate on inequality and growth37

From a theoretical viewpoint whether highwealth concentration can have a negativegrowth impact depends critically on the exis-tence of credit constraints With first-best credit

markets money flows toward the best entrepre-neurs and investment projects irrespective ofthe initial distribution High levels of wealthconcentration can be bad from a social justiceviewpoint but they entail no efficiency lossWhen credit constraints bind however initialwealth matters and high levels of inequality canhurt growth Whether the loss is large or smalldepends on who owns the assets If the rich areefficient investors (they know which projects tofund etc) then wealth concentration may evenbe useful If the rich are retired rentiers how-ever investing their wealth in low-yield assets(or low-ability inheritors) then high wealthconcentration and credit constraints might pre-vent talented but penniless investors from un-dertaking efficient projects thereby entailingnegative growth consequences The data used inthis paper are not ideal to address whether creditconstraints were important in a country likeFrance at the end of the nineteenth century Ourresults suggest however that to the extentcredit constraints were indeed severe highwealth concentration did have a negativegrowth impact In order to investigate this hy-pothesis further one would need to gather moresystematic data on investment strategies andasset returns Preliminary evidence suggeststhat the rich elderly of the 1860ndash1913 perioddid indeed hold a disproportionate fraction oftheir wealth in low-yield assets (such as gov-ernment bonds) An alternative hypothesishowever is that steeper age-wealth profileswere the consequence of the growth of financialmarkets as their children faced fewer creditconstraints parents decided to hold on to moreof their wealth

VI Conclusion

Evidence from wealth at death in Paris and inFrance over the last two centuries reveals threekey patterns First wealth concentration haschanged dramatically over time In 1807 thetop-1-percent share of wealth (40 percent inFrance 50 percent in Paris) was twice as high asit would be in 1994 but substantially less thanin 1913 when it peaked above 55 percent inFrance and 70 percent in Paris Some of thesechanges were due to economic phenomena thathave long been emphasized as creating inequal-ity namely industrialization and financial cen-

37 Thus far this literature has concentrated upon cross-country regressions of inequality on growth a methodologythat raises serious identification problems especially giventhe low quality of available international datasets on in-equality which are neither long-run nor homogeneous (seeeg Atkinson and Andrea Brandolini 2001)

254 THE AMERICAN ECONOMIC REVIEW MARCH 2006

tralization Yet the decline comes largely fromadverse shocks rather than economic conver-gence These changes are of such magnitudethat they are not sensitive to whether one ex-amines wealth at death in Paris or in France orwhether one examines it directly rather thanconverting it to wealth of the living by an estatemultiplier method

Our second key result is that there was asignificant transition during the nineteenth cen-tury from an important role for real estate as aform of wealth to moveable assets as the keyform of wealth for the very rich Similarly theshare of wealth held by aristocrats first rose andthen was eclipsed by that of financiers and in-dustrialists in the second half of the nineteenthcentury Hence mobility within this highly un-equal society might have been quite high Yetthis conjecture is tempered by our third findingthe wealthy were getting older over time andolder relative to less wealthy decedents Suchaging among the very wealthy would have hadnegative consequences for growth if financialmarkets were imperfect This issue requires fur-ther investigation and we hope it will attractfuture research

REFERENCES

Atkinson Anthony B ldquoTop Incomes in the UKover the 20th Centuryrdquo Journal of the RoyalStatistical Society Series A (Statistics in So-ciety) 2005 168(2) pp 325ndash43

Atkinson Anthony B and Brandolini AndrealdquoPromise and Pitfalls in the Use of lsquoSecond-aryrsquo Data-Sets Income Inequality in OECDCountries as a Case Studyrdquo Journal of Eco-nomic Literature 2001 39(3) pp 771ndash99

Atkinson Anthony B and Harrison Alan J Dis-tribution of personal wealth in Britain Cam-bridge Cambridge University Press 1978

Atkinson Anthony B and Piketty Thomas Topincomes over the twentieth century A con-trast between European and English speak-ing countries Oxford Oxford UniversityPress (forthcoming)

Banerjee Abhijit V and Duflo Esther ldquoInequalityand Growth What Can the Data Sayrdquo Journalof Economic Growth 2003 8(3) pp 267ndash99

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoPourquoi la richesse ne

srsquoest-elle pas diffusee avec la croissance Ledegre zero de lrsquoinegalite et son evolution enFrance 1800ndash1940rdquo Histoire et mesure 200318(1ndash2) pp 147ndash98

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoDefense et illustration delrsquoenquete des 3000 familles Lrsquoexemple deson volet patrimonialrdquo Annales de demogra-phie historique 2004 19 pp 19ndash57

Danysz E S ldquoContribution a lrsquoetude des for-tunes privees drsquoapres les declarations de suc-cessionsrdquo Bulletin de la statistique generalede France 1934 XXIV (1) pp 111ndash171

Daumard Adeline Les fortunes francaise auXIXe siecle Enquete sur la repartition et lacomposition des capitaux prives a ParisLyon Lille Bordeaux et Toulouse drsquoapreslrsquoenregistrement des declarations de succes-sion Paris Mouton 1973

Gain Andre La restauration et les biens desemigres 2 vols Nancy Societe drsquoimpressiontypographique 1929

Jones Alice H American colonial wealth Doc-uments and methods 3 vols New York ArnoPress 1977

Kopczuk Wojciech and Saez Emmanuel ldquoTopWealth Shares in the United States 1916ndash2000 Evidence from Estate Tax ReturnsrdquoNational Tax Journal 2004 57(2) pp 445ndash87

Kuznets Simon ldquoEconomic Growth and IncomeInequalityrdquo American Economic Review1955 45(1) pp 1ndash28

Lampman Robert J The share of top wealth-holders in national wealth 1922ndash1956Princeton Princeton University Press 1962

Lescure Michel Les banques lrsquoEtat et lemarche immobilier en France a lrsquoepoquecontemporaine 1820ndash1940 Paris EditionslrsquoEcole des Hautes Etudes en Sciences Socia-les 1982

Levy-Leboyer Maurice and Bourguignon Fran-cois LrsquoEconomie francaise au XIXe siecleAnalyse macroeconomique Paris Eco-nomica 1985 English translation Cam-bridge Cambridge University Press 1990

Lindert Peter H ldquoUnequal English Wealthsince 1670rdquo Journal of Political Economy1986 94(6) pp 1127ndash62

Lindert Peter H ldquoThree Centuries of Inequalityin Britain and Americardquo in Anthony B At-kinson and Francois Bourguignon eds

255VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

differences in inequality largely hark back to dif-ferential concentration of landownership En-glandrsquos land was extremely narrowly held whilethe United States was most egalitarian The impactof access to real estate assets can also be seenwhen we contrast Paris (where it was extremelyconcentrated) with the provinces

Trends in inequality over time may be easierto compare because biases in source materialmay have a significant effect on levels ratherthan trends The historical pattern is similar ineach of the three economies In particular thereis evidence that wealth concentration increasedduring the nineteenth century in both Anglo-Saxon countries and declined during the twen-tieth century with a turning point around WorldWar I28 Neither exhibits patterns consistentwith a Kuznet process

IV From the Wealth of the Decedents to theWealth of the Living

The estimates reported thus far refer to in-equality among decedents as described in thetax returns filed by their heirs The evolution ofthe distribution of wealth among the livingmight however have followed a different pat-tern In order to convert wealth-at-death con-centration estimates into wealth-of-the-livingconcentration estimates it is standard to use theldquoestate multiplierrdquo method29 It consists ofweighting each observation of an estate at death

by the inverse of the mortality rate for this agegroup That is if the mortality rate for ages 20to 24 was 068 percent in Paris in 1902 theneach decedent aged 20 to 24 represented about147 living individuals of the same age (100068 147) Similarly if the mortality ratefor ages above 80 was 2143 percent in Paris in1902 then each decedent in that group repre-sented about 47 living individuals in the sameage group (102143 47) Applying thismethod requires mortality tables (these are eas-ily available) and estate tabulations brokendown by estate size and age at death (these arescarcer) Fortunately the cityrsquos statistical bu-reau published annual death-by-age totals theFrench censuses report the age distribution forthe capital every five years and we collectedage at death from the estate declarations Thesedata allowed us to compute estimates of wealthconcentration among the living over the 1807ndash1902 period using various assumptions aboutthe wealth profiles of mortality rates

The base population for the living is the set ofall individuals aged 20 and over living in Parisin year t which we denote pt The number ofliving individuals in age bracket a is denoted pta(a 20ndash24 25ndash29 30ndash34 75ndash79 80 andover) and the number of decedents in agebracket a is denoted nta The mortality rate forage bracket a is given by mta ntapta Webegin with a uniform-mortality benchmarkThese estimates are based on the simplifyingassumption that these mortality rates dependsolely on age and are the same for all wealthgroups (and in particular are the same forzero-wealth and positive-wealth individuals)We can then weigh each decedent with positiveestate and age a collected in the Paris archivesin year t by ptanta This allows us to computethe number of living Parisians with positivewealth in year t and also (by differentiatingwith pt) the number of living Parisians with zerowealth at year t which is used to weight zero-estate observations We then use our weighteddatasets to compute top estate fractiles amongthe living in Paris

The main conclusion is that the living expe-rienced the same upward trend in wealth con-centration as the decedents (see Figure 9) Wefind that inequality was significantly higheramong the living than among decedents be-cause survivors were on average younger than

55 percent in 1800 to 70 percent around World War I thenfell to about 20 percent in the 1990s The US top-1-percentwealth share rose from about 15 to 20 percent in 1800 toabout 40 percent around World War I then fell to about 30percent in the 1990s (and as low as 20 to 25 percentaccording to the more recent estimates from Kopczuk andSaez 2004) Wealth concentration is now larger in theUnited States than in European countries but the reversewas true during the nineteenth century up until World WarII (It is only since the 1950sndash1970s period that US wealthconcentration has been somewhat larger)

28 See Lindert (2000 pp 181ndash82 and 188)29 This method was widely used in England and France

in the late nineteenth and early twentieth centuries to com-pute the stock of national wealth on the basis of the flow ofwealth transmitted at death Standard references that use thistechnique to estimate the wealth distribution of the livingfrom estate tax data tabulated by estate size and age at deathinclude Atkinson and Harrison (1978) and Lampman(1962) For a more recent application of this technique tothe United States see Kopczuk and Saez (2004)

250 THE AMERICAN ECONOMIC REVIEW MARCH 2006

those who died and the young were on aver-age poorer In particular the estate multipliermethod leads to lower average weights forpositive-wealth decedents than for zero-wealthdecedents (the former are on average older andtherefore ldquorepresentrdquo a smaller number of livingindividuals) As a result the fraction of positive-wealth individuals is even smaller among theliving than among decedents Hence our bench-mark uniform-mortality estimates of wealthconcentration among the living are significantlylarger than corresponding estimates among de-cedents (eg top-1-percent wealth shares areabout 15 percent larger among the living)Changes over time however are similar In-creased life expectancy and declining mortalityrates over the course of the nineteenth centuryhave only a small effect on the trends

In order to make the estate multiplier methodmore reliable one would prefer to take intoaccount differential mortality by wealth Doingso would require having access to mortalityschedules based both on wealth and age at dif-ferent points in time unfortunately these are notavailable We have nonetheless reestimatedwealth of the living based upon the same as-sumption as Kopczuk and Saez (2004) That is

we assumed uniform mortality among the poor(here defined as zero-wealth individuals) andamong the rich (here defined as positive-wealthindividuals) and we assumed that the ratiomtaRmtaP between the mortality rate of the richand the mortality rate of the poor followed aU-shaped age profile from about 85 percent forthe young (ie the rich die 15 percent less oftenthan the poor when they are 20ndash24 or 25ndash29years old) down to about 70 percent for middle-age individuals in their forties to fifties and upto 100 percent for very old individuals in theireighties to nineties30 This profile correspondsto the best available estimates in the literatureand it appears to be relatively stable over timeand across developed countries In the absenceof better data it is the best one can do31 Thebenchmark differential-mortality estimates re-ported on Figure 9 show that although addingdifferential mortality produces different levelsof inequality it does not have much impact onthe upward trend in concentration

30 See Kopczuk and Saez (2004 Table A4)31 See Kopczuk and Saez (2004 Appendix B) for refer-

ences to the US and international literature devoted to theage-wealth profile of mortality rates

FIGURE 9 WEALTH CONCENTRATION AMONG DECEDENTS AND AMONG THE LIVING IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives (see Piketty et al 2004Table A4 for detailed series)

251VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

The resulting differential-mortality inequalityestimates lie between those based on decedentsonly and those using uniform mortality for theliving (see Figure 9) Note that moving fromuniform-mortality to differential-mortality es-tate multiplier methodology can either increaseor decrease inequality Here the reason whysuch a move leads to lower wealth concentra-tion seems to be due to the fact that differentialmortality tends to put higher weights onpositive-wealth decedents (for a given age)thereby increasing the estimated fraction of liv-ing individuals with positive wealth The im-portant point however is that the resulting leveleffects are relatively small in magnitude con-stant in time and dwarfed by the upward timetrend Even if we were to assume an enormousincrease in differential mortality during thenineteenth century in the sense that differentialmortality between the rich and poor was equalto 0 percent of the benchmark differential in1807 and 100 percent of the benchmark differ-ential in 1902 the resulting wealth concentra-tion estimates would still be significantly higherin 1902 than in 1807 (see Figure 9) Yet wehave no reason to believe that differential mor-tality increased to such an extent During thenineteenth century real wages for unskilledworkers rose which would have reduced mor-tality more for the poor than for the rich After1850 public health measures (sanitation watervaccination) were in place Again these wouldhave had a significant effect on the poor and themiddle class who could not privately purchasesuch health-improving services To be sure therich could avail themselves of more medicalservices than the poor or the middle class butthe impact of these services was probably small(bear in mind that neither antibiotics nor car-diovascular interventions were available)

Finally we have applied the estate multipliermethod to available data for 1947 and 1994 Over-all the sharp decline in wealth concentration ob-served during the twentieth century (andespecially between 1914 and 1945) is very robustIf anything the decline appears to be even largerwhen one looks at wealth concentration amongthe living rather than among decedents32

V The Changing Age Profile of Wealth

In the previous sections we focused almostexclusively on aggregate top wealth shares Ourdata however also detail the characteristics ofeach decedent in particular their gender andage The evolution of wealth by gender is ofrelevance for over the past two centuries therehave been massive changes in womenrsquos laborforce participation capacity to manage theirown affairs and life expectancy relative to menThe evolution of wealth by age is of relevancebecause there was a significant increase in adultlife expectancy over the twentieth century andbecause the progressive diffusion of pensionsmay have changed savings motivations More-over age-wealth profiles also inform us aboutthe motives of wealth accumulation and theeconomic impact of high wealth concentration

A first pass at the data considers the genderbreakdown of wealth at death Remarkably inour micro data the share of women in top estatestakes its highest value on the eve of World WarI For instance the womenrsquos share in the top 05percent rose from 35 percent prior to 1850 to 45percent in 1902 only to fall to 40 percent afterWorld War II Strikingly womenrsquos share ofwealth follows almost exactly the pattern ofaggregate inequality Women were relativelyricher when inequality reached its apex inFrance than at any other time Moreover insti-tutional variables seem to have played almostno role in changing the relative wealth ofwomen Unlike in common law countriesFrench law starting with the code civil of 1804required nearly equal treatment of all children inbequests Further research will help us deter-mine to what extent women of great wealthwere heirs or part of economically very success-ful couples

The data also reveal striking changes in the

32 See Piketty et al (2004 Table A4) It is unfortunatelynot possible to construct complete series for wealth concen-

tration among the living for the twentieth century due todata limitations tables broken down by estate brackets andage of decedents are available solely for years 1943ndash1954and at the national level (no table broken down by estatebrackets and age of decedents has ever been compiled at thedepartement level except in 1931 for Seine departementsee E S Danysz 1934) and the 1994 micro sample is notlarge enough to allow for a reliable application of the estatemultiplier method at the Paris level Thus the only wealth-of-the-living concentration estimates we provide for thetwentieth century are national estimates for 1947 and 1994

252 THE AMERICAN ECONOMIC REVIEW MARCH 2006

age profile of wealth between 1807 and 1994(see Table 5) During the nineteenth century aswealth concentration was increasing the veryrich were getting older and older At the begin-ning of the nineteenth century in the aftermathof the French Revolution the richest individualswere those in their fifties they were typically100 percent richer on average than people intheir forties 25 percent richer than those in theirsixties and 40 percent richer than those in theirseventies and eighties By the 1870s howeverthe age-wealth pattern had become stronglymonotonic the richest individuals were the old-est individuals In 1902 people in their sixtiesand seventies bequeathed 150 percent morethan those in their fifties and those in theireighties 300 percent more On the eve of WorldWar I top wealth holders were old and likely tobe retired This pattern breaks some time duringthe 1914ndash1945 period33 In 1947 as well as in1994 we are back to a pattern where the richestindividuals are those in their fifties Overall theperiod of maximal wealth inequality (1860ndash1913) also appears to be a period characterizedby a very specific age profile of wealth and largeconcentration of assets among the elderly

Another way to analyze the changing age-wealth relationship is to look at average age by

top estate fractile34 In 1817 average age wasvirtually the same for the top 10 percent and thetop 1 percent of estates (or even slightly declin-ing) The average-age-per-fractile relationshipbecomes upward sloping during the nineteenthcentury and by 1902 those in the top 1 percentwere almost six years older than those in the top10 percent The relationship is flat in 1947 anddownward-sloping in 1994 Finally one canapply the estate multiplier method (see SectionIV above) and analyze how wealth concentra-tion by age group among the living changedover the course of the nineteenth century Thegeneral population in Paris did not becomeolder during the nineteenth century those aged60 or older made up about 15 percent of thepopulation in 1817 and after 1847 about 10 to11 percent35 The share of total wealth ownedby the elderly rose significantly however aswealth distribution worsened The wealth be-longing to those aged 60 or more rose fromabout 25 to 30 percent of the total at the begin-ning of the nineteenth century to about 40 to 45percent by the end of the century The wealthshare of those aged 70 or older doubled fromless than 10 percent to about 20 percent36

33 Existing evidence on the age-wealth profile for 1931(see Danysz 1934) suggests that the Great Depression andWorld War II (rather than World War I) played the leadingroles in breaking this pattern This is an issue we plan toinvestigate in future research

34 See Piketty et al (2004 Table 6)35 Although life expectancy was increasing which

should have led to large shares of population for oldergroups the city was also growing quickly The large num-ber of immigrants (who were typically in their twenties)increased the relative size of the younger cohorts (seePiketty et al 2004 Table 7 and Figure 11)

36 See Piketty et al (2004 Figure 12)

TABLE 5mdashTHE AGE PROFILE OF WEALTH AT DEATH IN PARIS 1817ndash1994(Average estate left by 50- 59-year-old 100)

20ndash29yr-old

30ndash39yr-old

40ndash49yr-old

50ndash59yr-old

60ndash69yr-old

70ndash79yr-old

80ndash89yr-old

90ndash99yr-old

1817 26 22 28 100 54 59 591827 44 50 53 100 88 87 601837 133 90 107 100 116 123 1101847 87 73 102 100 117 204 1321857 84 77 101 100 104 109 1451867 67 58 136 100 141 125 1541877 66 73 63 100 197 260 4301887 45 33 63 100 152 233 2951902 29 40 80 100 253 272 4011947 31 51 73 100 113 105 105 1091994 11 45 100 87 93 95 68

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources)

253VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

It is perhaps not surprising that inequalitybecame strongly correlated with age in the1860s Those who died at an old age in the1820s and 1830s had lived through the difficultyears of the French Revolution and the disloca-tion of the end of the Napoleonic period Theirability to accumulate wealth had been severelyimpaired Furthermore they were rather lesslikely to inherit much wealth since the Revo-lution wiped out the bond portfolios of theirparents through a prolonged period of high in-flation Those who died from the 1860s to theearly 1910s did not suffer from the adversitiesthat plagued their forebears Instead they en-joyed the fruits of the financial sector expansionthat began in the 1850s After 1947 we seem tohave returned to a situation quite like that of1817 Presumably the capital damages associ-ated with both world wars and the Great De-pression had a strong negative effect on thewealth holdings of older generations The per-sistence of a flat age-wealth profile until 1994 islikely to be associated with two factors First insocieties where income growth is rapid abso-lute wealth accumulation is faster by youngercohorts than by older ones because their in-comes are higher at every age This is an im-portant distinction between the nineteenthcentury and the twentieth century Furthermorehighly progressive rates of income and estatetaxation have probably made it more difficult toaccumulate large fortunes thereby flattening theobserved age-wealth profile

The more interesting (and more difficult)question relates to the possible efficiency im-pact of high wealth concentration and changingage-wealth profiles Although our data do notallow us to address efficiency issues in a rigor-ous way our results allow us to formulate anumber of hypotheses and to shed new light onthe ongoing debate on inequality and growth37

From a theoretical viewpoint whether highwealth concentration can have a negativegrowth impact depends critically on the exis-tence of credit constraints With first-best credit

markets money flows toward the best entrepre-neurs and investment projects irrespective ofthe initial distribution High levels of wealthconcentration can be bad from a social justiceviewpoint but they entail no efficiency lossWhen credit constraints bind however initialwealth matters and high levels of inequality canhurt growth Whether the loss is large or smalldepends on who owns the assets If the rich areefficient investors (they know which projects tofund etc) then wealth concentration may evenbe useful If the rich are retired rentiers how-ever investing their wealth in low-yield assets(or low-ability inheritors) then high wealthconcentration and credit constraints might pre-vent talented but penniless investors from un-dertaking efficient projects thereby entailingnegative growth consequences The data used inthis paper are not ideal to address whether creditconstraints were important in a country likeFrance at the end of the nineteenth century Ourresults suggest however that to the extentcredit constraints were indeed severe highwealth concentration did have a negativegrowth impact In order to investigate this hy-pothesis further one would need to gather moresystematic data on investment strategies andasset returns Preliminary evidence suggeststhat the rich elderly of the 1860ndash1913 perioddid indeed hold a disproportionate fraction oftheir wealth in low-yield assets (such as gov-ernment bonds) An alternative hypothesishowever is that steeper age-wealth profileswere the consequence of the growth of financialmarkets as their children faced fewer creditconstraints parents decided to hold on to moreof their wealth

VI Conclusion

Evidence from wealth at death in Paris and inFrance over the last two centuries reveals threekey patterns First wealth concentration haschanged dramatically over time In 1807 thetop-1-percent share of wealth (40 percent inFrance 50 percent in Paris) was twice as high asit would be in 1994 but substantially less thanin 1913 when it peaked above 55 percent inFrance and 70 percent in Paris Some of thesechanges were due to economic phenomena thathave long been emphasized as creating inequal-ity namely industrialization and financial cen-

37 Thus far this literature has concentrated upon cross-country regressions of inequality on growth a methodologythat raises serious identification problems especially giventhe low quality of available international datasets on in-equality which are neither long-run nor homogeneous (seeeg Atkinson and Andrea Brandolini 2001)

254 THE AMERICAN ECONOMIC REVIEW MARCH 2006

tralization Yet the decline comes largely fromadverse shocks rather than economic conver-gence These changes are of such magnitudethat they are not sensitive to whether one ex-amines wealth at death in Paris or in France orwhether one examines it directly rather thanconverting it to wealth of the living by an estatemultiplier method

Our second key result is that there was asignificant transition during the nineteenth cen-tury from an important role for real estate as aform of wealth to moveable assets as the keyform of wealth for the very rich Similarly theshare of wealth held by aristocrats first rose andthen was eclipsed by that of financiers and in-dustrialists in the second half of the nineteenthcentury Hence mobility within this highly un-equal society might have been quite high Yetthis conjecture is tempered by our third findingthe wealthy were getting older over time andolder relative to less wealthy decedents Suchaging among the very wealthy would have hadnegative consequences for growth if financialmarkets were imperfect This issue requires fur-ther investigation and we hope it will attractfuture research

REFERENCES

Atkinson Anthony B ldquoTop Incomes in the UKover the 20th Centuryrdquo Journal of the RoyalStatistical Society Series A (Statistics in So-ciety) 2005 168(2) pp 325ndash43

Atkinson Anthony B and Brandolini AndrealdquoPromise and Pitfalls in the Use of lsquoSecond-aryrsquo Data-Sets Income Inequality in OECDCountries as a Case Studyrdquo Journal of Eco-nomic Literature 2001 39(3) pp 771ndash99

Atkinson Anthony B and Harrison Alan J Dis-tribution of personal wealth in Britain Cam-bridge Cambridge University Press 1978

Atkinson Anthony B and Piketty Thomas Topincomes over the twentieth century A con-trast between European and English speak-ing countries Oxford Oxford UniversityPress (forthcoming)

Banerjee Abhijit V and Duflo Esther ldquoInequalityand Growth What Can the Data Sayrdquo Journalof Economic Growth 2003 8(3) pp 267ndash99

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoPourquoi la richesse ne

srsquoest-elle pas diffusee avec la croissance Ledegre zero de lrsquoinegalite et son evolution enFrance 1800ndash1940rdquo Histoire et mesure 200318(1ndash2) pp 147ndash98

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoDefense et illustration delrsquoenquete des 3000 familles Lrsquoexemple deson volet patrimonialrdquo Annales de demogra-phie historique 2004 19 pp 19ndash57

Danysz E S ldquoContribution a lrsquoetude des for-tunes privees drsquoapres les declarations de suc-cessionsrdquo Bulletin de la statistique generalede France 1934 XXIV (1) pp 111ndash171

Daumard Adeline Les fortunes francaise auXIXe siecle Enquete sur la repartition et lacomposition des capitaux prives a ParisLyon Lille Bordeaux et Toulouse drsquoapreslrsquoenregistrement des declarations de succes-sion Paris Mouton 1973

Gain Andre La restauration et les biens desemigres 2 vols Nancy Societe drsquoimpressiontypographique 1929

Jones Alice H American colonial wealth Doc-uments and methods 3 vols New York ArnoPress 1977

Kopczuk Wojciech and Saez Emmanuel ldquoTopWealth Shares in the United States 1916ndash2000 Evidence from Estate Tax ReturnsrdquoNational Tax Journal 2004 57(2) pp 445ndash87

Kuznets Simon ldquoEconomic Growth and IncomeInequalityrdquo American Economic Review1955 45(1) pp 1ndash28

Lampman Robert J The share of top wealth-holders in national wealth 1922ndash1956Princeton Princeton University Press 1962

Lescure Michel Les banques lrsquoEtat et lemarche immobilier en France a lrsquoepoquecontemporaine 1820ndash1940 Paris EditionslrsquoEcole des Hautes Etudes en Sciences Socia-les 1982

Levy-Leboyer Maurice and Bourguignon Fran-cois LrsquoEconomie francaise au XIXe siecleAnalyse macroeconomique Paris Eco-nomica 1985 English translation Cam-bridge Cambridge University Press 1990

Lindert Peter H ldquoUnequal English Wealthsince 1670rdquo Journal of Political Economy1986 94(6) pp 1127ndash62

Lindert Peter H ldquoThree Centuries of Inequalityin Britain and Americardquo in Anthony B At-kinson and Francois Bourguignon eds

255VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

those who died and the young were on aver-age poorer In particular the estate multipliermethod leads to lower average weights forpositive-wealth decedents than for zero-wealthdecedents (the former are on average older andtherefore ldquorepresentrdquo a smaller number of livingindividuals) As a result the fraction of positive-wealth individuals is even smaller among theliving than among decedents Hence our bench-mark uniform-mortality estimates of wealthconcentration among the living are significantlylarger than corresponding estimates among de-cedents (eg top-1-percent wealth shares areabout 15 percent larger among the living)Changes over time however are similar In-creased life expectancy and declining mortalityrates over the course of the nineteenth centuryhave only a small effect on the trends

In order to make the estate multiplier methodmore reliable one would prefer to take intoaccount differential mortality by wealth Doingso would require having access to mortalityschedules based both on wealth and age at dif-ferent points in time unfortunately these are notavailable We have nonetheless reestimatedwealth of the living based upon the same as-sumption as Kopczuk and Saez (2004) That is

we assumed uniform mortality among the poor(here defined as zero-wealth individuals) andamong the rich (here defined as positive-wealthindividuals) and we assumed that the ratiomtaRmtaP between the mortality rate of the richand the mortality rate of the poor followed aU-shaped age profile from about 85 percent forthe young (ie the rich die 15 percent less oftenthan the poor when they are 20ndash24 or 25ndash29years old) down to about 70 percent for middle-age individuals in their forties to fifties and upto 100 percent for very old individuals in theireighties to nineties30 This profile correspondsto the best available estimates in the literatureand it appears to be relatively stable over timeand across developed countries In the absenceof better data it is the best one can do31 Thebenchmark differential-mortality estimates re-ported on Figure 9 show that although addingdifferential mortality produces different levelsof inequality it does not have much impact onthe upward trend in concentration

30 See Kopczuk and Saez (2004 Table A4)31 See Kopczuk and Saez (2004 Appendix B) for refer-

ences to the US and international literature devoted to theage-wealth profile of mortality rates

FIGURE 9 WEALTH CONCENTRATION AMONG DECEDENTS AND AMONG THE LIVING IN PARIS 1807ndash1902

Source Authorsrsquo computations using samples of estate tax returns collected in the Paris archives (see Piketty et al 2004Table A4 for detailed series)

251VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

The resulting differential-mortality inequalityestimates lie between those based on decedentsonly and those using uniform mortality for theliving (see Figure 9) Note that moving fromuniform-mortality to differential-mortality es-tate multiplier methodology can either increaseor decrease inequality Here the reason whysuch a move leads to lower wealth concentra-tion seems to be due to the fact that differentialmortality tends to put higher weights onpositive-wealth decedents (for a given age)thereby increasing the estimated fraction of liv-ing individuals with positive wealth The im-portant point however is that the resulting leveleffects are relatively small in magnitude con-stant in time and dwarfed by the upward timetrend Even if we were to assume an enormousincrease in differential mortality during thenineteenth century in the sense that differentialmortality between the rich and poor was equalto 0 percent of the benchmark differential in1807 and 100 percent of the benchmark differ-ential in 1902 the resulting wealth concentra-tion estimates would still be significantly higherin 1902 than in 1807 (see Figure 9) Yet wehave no reason to believe that differential mor-tality increased to such an extent During thenineteenth century real wages for unskilledworkers rose which would have reduced mor-tality more for the poor than for the rich After1850 public health measures (sanitation watervaccination) were in place Again these wouldhave had a significant effect on the poor and themiddle class who could not privately purchasesuch health-improving services To be sure therich could avail themselves of more medicalservices than the poor or the middle class butthe impact of these services was probably small(bear in mind that neither antibiotics nor car-diovascular interventions were available)

Finally we have applied the estate multipliermethod to available data for 1947 and 1994 Over-all the sharp decline in wealth concentration ob-served during the twentieth century (andespecially between 1914 and 1945) is very robustIf anything the decline appears to be even largerwhen one looks at wealth concentration amongthe living rather than among decedents32

V The Changing Age Profile of Wealth

In the previous sections we focused almostexclusively on aggregate top wealth shares Ourdata however also detail the characteristics ofeach decedent in particular their gender andage The evolution of wealth by gender is ofrelevance for over the past two centuries therehave been massive changes in womenrsquos laborforce participation capacity to manage theirown affairs and life expectancy relative to menThe evolution of wealth by age is of relevancebecause there was a significant increase in adultlife expectancy over the twentieth century andbecause the progressive diffusion of pensionsmay have changed savings motivations More-over age-wealth profiles also inform us aboutthe motives of wealth accumulation and theeconomic impact of high wealth concentration

A first pass at the data considers the genderbreakdown of wealth at death Remarkably inour micro data the share of women in top estatestakes its highest value on the eve of World WarI For instance the womenrsquos share in the top 05percent rose from 35 percent prior to 1850 to 45percent in 1902 only to fall to 40 percent afterWorld War II Strikingly womenrsquos share ofwealth follows almost exactly the pattern ofaggregate inequality Women were relativelyricher when inequality reached its apex inFrance than at any other time Moreover insti-tutional variables seem to have played almostno role in changing the relative wealth ofwomen Unlike in common law countriesFrench law starting with the code civil of 1804required nearly equal treatment of all children inbequests Further research will help us deter-mine to what extent women of great wealthwere heirs or part of economically very success-ful couples

The data also reveal striking changes in the

32 See Piketty et al (2004 Table A4) It is unfortunatelynot possible to construct complete series for wealth concen-

tration among the living for the twentieth century due todata limitations tables broken down by estate brackets andage of decedents are available solely for years 1943ndash1954and at the national level (no table broken down by estatebrackets and age of decedents has ever been compiled at thedepartement level except in 1931 for Seine departementsee E S Danysz 1934) and the 1994 micro sample is notlarge enough to allow for a reliable application of the estatemultiplier method at the Paris level Thus the only wealth-of-the-living concentration estimates we provide for thetwentieth century are national estimates for 1947 and 1994

252 THE AMERICAN ECONOMIC REVIEW MARCH 2006

age profile of wealth between 1807 and 1994(see Table 5) During the nineteenth century aswealth concentration was increasing the veryrich were getting older and older At the begin-ning of the nineteenth century in the aftermathof the French Revolution the richest individualswere those in their fifties they were typically100 percent richer on average than people intheir forties 25 percent richer than those in theirsixties and 40 percent richer than those in theirseventies and eighties By the 1870s howeverthe age-wealth pattern had become stronglymonotonic the richest individuals were the old-est individuals In 1902 people in their sixtiesand seventies bequeathed 150 percent morethan those in their fifties and those in theireighties 300 percent more On the eve of WorldWar I top wealth holders were old and likely tobe retired This pattern breaks some time duringthe 1914ndash1945 period33 In 1947 as well as in1994 we are back to a pattern where the richestindividuals are those in their fifties Overall theperiod of maximal wealth inequality (1860ndash1913) also appears to be a period characterizedby a very specific age profile of wealth and largeconcentration of assets among the elderly

Another way to analyze the changing age-wealth relationship is to look at average age by

top estate fractile34 In 1817 average age wasvirtually the same for the top 10 percent and thetop 1 percent of estates (or even slightly declin-ing) The average-age-per-fractile relationshipbecomes upward sloping during the nineteenthcentury and by 1902 those in the top 1 percentwere almost six years older than those in the top10 percent The relationship is flat in 1947 anddownward-sloping in 1994 Finally one canapply the estate multiplier method (see SectionIV above) and analyze how wealth concentra-tion by age group among the living changedover the course of the nineteenth century Thegeneral population in Paris did not becomeolder during the nineteenth century those aged60 or older made up about 15 percent of thepopulation in 1817 and after 1847 about 10 to11 percent35 The share of total wealth ownedby the elderly rose significantly however aswealth distribution worsened The wealth be-longing to those aged 60 or more rose fromabout 25 to 30 percent of the total at the begin-ning of the nineteenth century to about 40 to 45percent by the end of the century The wealthshare of those aged 70 or older doubled fromless than 10 percent to about 20 percent36

33 Existing evidence on the age-wealth profile for 1931(see Danysz 1934) suggests that the Great Depression andWorld War II (rather than World War I) played the leadingroles in breaking this pattern This is an issue we plan toinvestigate in future research

34 See Piketty et al (2004 Table 6)35 Although life expectancy was increasing which

should have led to large shares of population for oldergroups the city was also growing quickly The large num-ber of immigrants (who were typically in their twenties)increased the relative size of the younger cohorts (seePiketty et al 2004 Table 7 and Figure 11)

36 See Piketty et al (2004 Figure 12)

TABLE 5mdashTHE AGE PROFILE OF WEALTH AT DEATH IN PARIS 1817ndash1994(Average estate left by 50- 59-year-old 100)

20ndash29yr-old

30ndash39yr-old

40ndash49yr-old

50ndash59yr-old

60ndash69yr-old

70ndash79yr-old

80ndash89yr-old

90ndash99yr-old

1817 26 22 28 100 54 59 591827 44 50 53 100 88 87 601837 133 90 107 100 116 123 1101847 87 73 102 100 117 204 1321857 84 77 101 100 104 109 1451867 67 58 136 100 141 125 1541877 66 73 63 100 197 260 4301887 45 33 63 100 152 233 2951902 29 40 80 100 253 272 4011947 31 51 73 100 113 105 105 1091994 11 45 100 87 93 95 68

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources)

253VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

It is perhaps not surprising that inequalitybecame strongly correlated with age in the1860s Those who died at an old age in the1820s and 1830s had lived through the difficultyears of the French Revolution and the disloca-tion of the end of the Napoleonic period Theirability to accumulate wealth had been severelyimpaired Furthermore they were rather lesslikely to inherit much wealth since the Revo-lution wiped out the bond portfolios of theirparents through a prolonged period of high in-flation Those who died from the 1860s to theearly 1910s did not suffer from the adversitiesthat plagued their forebears Instead they en-joyed the fruits of the financial sector expansionthat began in the 1850s After 1947 we seem tohave returned to a situation quite like that of1817 Presumably the capital damages associ-ated with both world wars and the Great De-pression had a strong negative effect on thewealth holdings of older generations The per-sistence of a flat age-wealth profile until 1994 islikely to be associated with two factors First insocieties where income growth is rapid abso-lute wealth accumulation is faster by youngercohorts than by older ones because their in-comes are higher at every age This is an im-portant distinction between the nineteenthcentury and the twentieth century Furthermorehighly progressive rates of income and estatetaxation have probably made it more difficult toaccumulate large fortunes thereby flattening theobserved age-wealth profile

The more interesting (and more difficult)question relates to the possible efficiency im-pact of high wealth concentration and changingage-wealth profiles Although our data do notallow us to address efficiency issues in a rigor-ous way our results allow us to formulate anumber of hypotheses and to shed new light onthe ongoing debate on inequality and growth37

From a theoretical viewpoint whether highwealth concentration can have a negativegrowth impact depends critically on the exis-tence of credit constraints With first-best credit

markets money flows toward the best entrepre-neurs and investment projects irrespective ofthe initial distribution High levels of wealthconcentration can be bad from a social justiceviewpoint but they entail no efficiency lossWhen credit constraints bind however initialwealth matters and high levels of inequality canhurt growth Whether the loss is large or smalldepends on who owns the assets If the rich areefficient investors (they know which projects tofund etc) then wealth concentration may evenbe useful If the rich are retired rentiers how-ever investing their wealth in low-yield assets(or low-ability inheritors) then high wealthconcentration and credit constraints might pre-vent talented but penniless investors from un-dertaking efficient projects thereby entailingnegative growth consequences The data used inthis paper are not ideal to address whether creditconstraints were important in a country likeFrance at the end of the nineteenth century Ourresults suggest however that to the extentcredit constraints were indeed severe highwealth concentration did have a negativegrowth impact In order to investigate this hy-pothesis further one would need to gather moresystematic data on investment strategies andasset returns Preliminary evidence suggeststhat the rich elderly of the 1860ndash1913 perioddid indeed hold a disproportionate fraction oftheir wealth in low-yield assets (such as gov-ernment bonds) An alternative hypothesishowever is that steeper age-wealth profileswere the consequence of the growth of financialmarkets as their children faced fewer creditconstraints parents decided to hold on to moreof their wealth

VI Conclusion

Evidence from wealth at death in Paris and inFrance over the last two centuries reveals threekey patterns First wealth concentration haschanged dramatically over time In 1807 thetop-1-percent share of wealth (40 percent inFrance 50 percent in Paris) was twice as high asit would be in 1994 but substantially less thanin 1913 when it peaked above 55 percent inFrance and 70 percent in Paris Some of thesechanges were due to economic phenomena thathave long been emphasized as creating inequal-ity namely industrialization and financial cen-

37 Thus far this literature has concentrated upon cross-country regressions of inequality on growth a methodologythat raises serious identification problems especially giventhe low quality of available international datasets on in-equality which are neither long-run nor homogeneous (seeeg Atkinson and Andrea Brandolini 2001)

254 THE AMERICAN ECONOMIC REVIEW MARCH 2006

tralization Yet the decline comes largely fromadverse shocks rather than economic conver-gence These changes are of such magnitudethat they are not sensitive to whether one ex-amines wealth at death in Paris or in France orwhether one examines it directly rather thanconverting it to wealth of the living by an estatemultiplier method

Our second key result is that there was asignificant transition during the nineteenth cen-tury from an important role for real estate as aform of wealth to moveable assets as the keyform of wealth for the very rich Similarly theshare of wealth held by aristocrats first rose andthen was eclipsed by that of financiers and in-dustrialists in the second half of the nineteenthcentury Hence mobility within this highly un-equal society might have been quite high Yetthis conjecture is tempered by our third findingthe wealthy were getting older over time andolder relative to less wealthy decedents Suchaging among the very wealthy would have hadnegative consequences for growth if financialmarkets were imperfect This issue requires fur-ther investigation and we hope it will attractfuture research

REFERENCES

Atkinson Anthony B ldquoTop Incomes in the UKover the 20th Centuryrdquo Journal of the RoyalStatistical Society Series A (Statistics in So-ciety) 2005 168(2) pp 325ndash43

Atkinson Anthony B and Brandolini AndrealdquoPromise and Pitfalls in the Use of lsquoSecond-aryrsquo Data-Sets Income Inequality in OECDCountries as a Case Studyrdquo Journal of Eco-nomic Literature 2001 39(3) pp 771ndash99

Atkinson Anthony B and Harrison Alan J Dis-tribution of personal wealth in Britain Cam-bridge Cambridge University Press 1978

Atkinson Anthony B and Piketty Thomas Topincomes over the twentieth century A con-trast between European and English speak-ing countries Oxford Oxford UniversityPress (forthcoming)

Banerjee Abhijit V and Duflo Esther ldquoInequalityand Growth What Can the Data Sayrdquo Journalof Economic Growth 2003 8(3) pp 267ndash99

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoPourquoi la richesse ne

srsquoest-elle pas diffusee avec la croissance Ledegre zero de lrsquoinegalite et son evolution enFrance 1800ndash1940rdquo Histoire et mesure 200318(1ndash2) pp 147ndash98

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoDefense et illustration delrsquoenquete des 3000 familles Lrsquoexemple deson volet patrimonialrdquo Annales de demogra-phie historique 2004 19 pp 19ndash57

Danysz E S ldquoContribution a lrsquoetude des for-tunes privees drsquoapres les declarations de suc-cessionsrdquo Bulletin de la statistique generalede France 1934 XXIV (1) pp 111ndash171

Daumard Adeline Les fortunes francaise auXIXe siecle Enquete sur la repartition et lacomposition des capitaux prives a ParisLyon Lille Bordeaux et Toulouse drsquoapreslrsquoenregistrement des declarations de succes-sion Paris Mouton 1973

Gain Andre La restauration et les biens desemigres 2 vols Nancy Societe drsquoimpressiontypographique 1929

Jones Alice H American colonial wealth Doc-uments and methods 3 vols New York ArnoPress 1977

Kopczuk Wojciech and Saez Emmanuel ldquoTopWealth Shares in the United States 1916ndash2000 Evidence from Estate Tax ReturnsrdquoNational Tax Journal 2004 57(2) pp 445ndash87

Kuznets Simon ldquoEconomic Growth and IncomeInequalityrdquo American Economic Review1955 45(1) pp 1ndash28

Lampman Robert J The share of top wealth-holders in national wealth 1922ndash1956Princeton Princeton University Press 1962

Lescure Michel Les banques lrsquoEtat et lemarche immobilier en France a lrsquoepoquecontemporaine 1820ndash1940 Paris EditionslrsquoEcole des Hautes Etudes en Sciences Socia-les 1982

Levy-Leboyer Maurice and Bourguignon Fran-cois LrsquoEconomie francaise au XIXe siecleAnalyse macroeconomique Paris Eco-nomica 1985 English translation Cam-bridge Cambridge University Press 1990

Lindert Peter H ldquoUnequal English Wealthsince 1670rdquo Journal of Political Economy1986 94(6) pp 1127ndash62

Lindert Peter H ldquoThree Centuries of Inequalityin Britain and Americardquo in Anthony B At-kinson and Francois Bourguignon eds

255VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

The resulting differential-mortality inequalityestimates lie between those based on decedentsonly and those using uniform mortality for theliving (see Figure 9) Note that moving fromuniform-mortality to differential-mortality es-tate multiplier methodology can either increaseor decrease inequality Here the reason whysuch a move leads to lower wealth concentra-tion seems to be due to the fact that differentialmortality tends to put higher weights onpositive-wealth decedents (for a given age)thereby increasing the estimated fraction of liv-ing individuals with positive wealth The im-portant point however is that the resulting leveleffects are relatively small in magnitude con-stant in time and dwarfed by the upward timetrend Even if we were to assume an enormousincrease in differential mortality during thenineteenth century in the sense that differentialmortality between the rich and poor was equalto 0 percent of the benchmark differential in1807 and 100 percent of the benchmark differ-ential in 1902 the resulting wealth concentra-tion estimates would still be significantly higherin 1902 than in 1807 (see Figure 9) Yet wehave no reason to believe that differential mor-tality increased to such an extent During thenineteenth century real wages for unskilledworkers rose which would have reduced mor-tality more for the poor than for the rich After1850 public health measures (sanitation watervaccination) were in place Again these wouldhave had a significant effect on the poor and themiddle class who could not privately purchasesuch health-improving services To be sure therich could avail themselves of more medicalservices than the poor or the middle class butthe impact of these services was probably small(bear in mind that neither antibiotics nor car-diovascular interventions were available)

Finally we have applied the estate multipliermethod to available data for 1947 and 1994 Over-all the sharp decline in wealth concentration ob-served during the twentieth century (andespecially between 1914 and 1945) is very robustIf anything the decline appears to be even largerwhen one looks at wealth concentration amongthe living rather than among decedents32

V The Changing Age Profile of Wealth

In the previous sections we focused almostexclusively on aggregate top wealth shares Ourdata however also detail the characteristics ofeach decedent in particular their gender andage The evolution of wealth by gender is ofrelevance for over the past two centuries therehave been massive changes in womenrsquos laborforce participation capacity to manage theirown affairs and life expectancy relative to menThe evolution of wealth by age is of relevancebecause there was a significant increase in adultlife expectancy over the twentieth century andbecause the progressive diffusion of pensionsmay have changed savings motivations More-over age-wealth profiles also inform us aboutthe motives of wealth accumulation and theeconomic impact of high wealth concentration

A first pass at the data considers the genderbreakdown of wealth at death Remarkably inour micro data the share of women in top estatestakes its highest value on the eve of World WarI For instance the womenrsquos share in the top 05percent rose from 35 percent prior to 1850 to 45percent in 1902 only to fall to 40 percent afterWorld War II Strikingly womenrsquos share ofwealth follows almost exactly the pattern ofaggregate inequality Women were relativelyricher when inequality reached its apex inFrance than at any other time Moreover insti-tutional variables seem to have played almostno role in changing the relative wealth ofwomen Unlike in common law countriesFrench law starting with the code civil of 1804required nearly equal treatment of all children inbequests Further research will help us deter-mine to what extent women of great wealthwere heirs or part of economically very success-ful couples

The data also reveal striking changes in the

32 See Piketty et al (2004 Table A4) It is unfortunatelynot possible to construct complete series for wealth concen-

tration among the living for the twentieth century due todata limitations tables broken down by estate brackets andage of decedents are available solely for years 1943ndash1954and at the national level (no table broken down by estatebrackets and age of decedents has ever been compiled at thedepartement level except in 1931 for Seine departementsee E S Danysz 1934) and the 1994 micro sample is notlarge enough to allow for a reliable application of the estatemultiplier method at the Paris level Thus the only wealth-of-the-living concentration estimates we provide for thetwentieth century are national estimates for 1947 and 1994

252 THE AMERICAN ECONOMIC REVIEW MARCH 2006

age profile of wealth between 1807 and 1994(see Table 5) During the nineteenth century aswealth concentration was increasing the veryrich were getting older and older At the begin-ning of the nineteenth century in the aftermathof the French Revolution the richest individualswere those in their fifties they were typically100 percent richer on average than people intheir forties 25 percent richer than those in theirsixties and 40 percent richer than those in theirseventies and eighties By the 1870s howeverthe age-wealth pattern had become stronglymonotonic the richest individuals were the old-est individuals In 1902 people in their sixtiesand seventies bequeathed 150 percent morethan those in their fifties and those in theireighties 300 percent more On the eve of WorldWar I top wealth holders were old and likely tobe retired This pattern breaks some time duringthe 1914ndash1945 period33 In 1947 as well as in1994 we are back to a pattern where the richestindividuals are those in their fifties Overall theperiod of maximal wealth inequality (1860ndash1913) also appears to be a period characterizedby a very specific age profile of wealth and largeconcentration of assets among the elderly

Another way to analyze the changing age-wealth relationship is to look at average age by

top estate fractile34 In 1817 average age wasvirtually the same for the top 10 percent and thetop 1 percent of estates (or even slightly declin-ing) The average-age-per-fractile relationshipbecomes upward sloping during the nineteenthcentury and by 1902 those in the top 1 percentwere almost six years older than those in the top10 percent The relationship is flat in 1947 anddownward-sloping in 1994 Finally one canapply the estate multiplier method (see SectionIV above) and analyze how wealth concentra-tion by age group among the living changedover the course of the nineteenth century Thegeneral population in Paris did not becomeolder during the nineteenth century those aged60 or older made up about 15 percent of thepopulation in 1817 and after 1847 about 10 to11 percent35 The share of total wealth ownedby the elderly rose significantly however aswealth distribution worsened The wealth be-longing to those aged 60 or more rose fromabout 25 to 30 percent of the total at the begin-ning of the nineteenth century to about 40 to 45percent by the end of the century The wealthshare of those aged 70 or older doubled fromless than 10 percent to about 20 percent36

33 Existing evidence on the age-wealth profile for 1931(see Danysz 1934) suggests that the Great Depression andWorld War II (rather than World War I) played the leadingroles in breaking this pattern This is an issue we plan toinvestigate in future research

34 See Piketty et al (2004 Table 6)35 Although life expectancy was increasing which

should have led to large shares of population for oldergroups the city was also growing quickly The large num-ber of immigrants (who were typically in their twenties)increased the relative size of the younger cohorts (seePiketty et al 2004 Table 7 and Figure 11)

36 See Piketty et al (2004 Figure 12)

TABLE 5mdashTHE AGE PROFILE OF WEALTH AT DEATH IN PARIS 1817ndash1994(Average estate left by 50- 59-year-old 100)

20ndash29yr-old

30ndash39yr-old

40ndash49yr-old

50ndash59yr-old

60ndash69yr-old

70ndash79yr-old

80ndash89yr-old

90ndash99yr-old

1817 26 22 28 100 54 59 591827 44 50 53 100 88 87 601837 133 90 107 100 116 123 1101847 87 73 102 100 117 204 1321857 84 77 101 100 104 109 1451867 67 58 136 100 141 125 1541877 66 73 63 100 197 260 4301887 45 33 63 100 152 233 2951902 29 40 80 100 253 272 4011947 31 51 73 100 113 105 105 1091994 11 45 100 87 93 95 68

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources)

253VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

It is perhaps not surprising that inequalitybecame strongly correlated with age in the1860s Those who died at an old age in the1820s and 1830s had lived through the difficultyears of the French Revolution and the disloca-tion of the end of the Napoleonic period Theirability to accumulate wealth had been severelyimpaired Furthermore they were rather lesslikely to inherit much wealth since the Revo-lution wiped out the bond portfolios of theirparents through a prolonged period of high in-flation Those who died from the 1860s to theearly 1910s did not suffer from the adversitiesthat plagued their forebears Instead they en-joyed the fruits of the financial sector expansionthat began in the 1850s After 1947 we seem tohave returned to a situation quite like that of1817 Presumably the capital damages associ-ated with both world wars and the Great De-pression had a strong negative effect on thewealth holdings of older generations The per-sistence of a flat age-wealth profile until 1994 islikely to be associated with two factors First insocieties where income growth is rapid abso-lute wealth accumulation is faster by youngercohorts than by older ones because their in-comes are higher at every age This is an im-portant distinction between the nineteenthcentury and the twentieth century Furthermorehighly progressive rates of income and estatetaxation have probably made it more difficult toaccumulate large fortunes thereby flattening theobserved age-wealth profile

The more interesting (and more difficult)question relates to the possible efficiency im-pact of high wealth concentration and changingage-wealth profiles Although our data do notallow us to address efficiency issues in a rigor-ous way our results allow us to formulate anumber of hypotheses and to shed new light onthe ongoing debate on inequality and growth37

From a theoretical viewpoint whether highwealth concentration can have a negativegrowth impact depends critically on the exis-tence of credit constraints With first-best credit

markets money flows toward the best entrepre-neurs and investment projects irrespective ofthe initial distribution High levels of wealthconcentration can be bad from a social justiceviewpoint but they entail no efficiency lossWhen credit constraints bind however initialwealth matters and high levels of inequality canhurt growth Whether the loss is large or smalldepends on who owns the assets If the rich areefficient investors (they know which projects tofund etc) then wealth concentration may evenbe useful If the rich are retired rentiers how-ever investing their wealth in low-yield assets(or low-ability inheritors) then high wealthconcentration and credit constraints might pre-vent talented but penniless investors from un-dertaking efficient projects thereby entailingnegative growth consequences The data used inthis paper are not ideal to address whether creditconstraints were important in a country likeFrance at the end of the nineteenth century Ourresults suggest however that to the extentcredit constraints were indeed severe highwealth concentration did have a negativegrowth impact In order to investigate this hy-pothesis further one would need to gather moresystematic data on investment strategies andasset returns Preliminary evidence suggeststhat the rich elderly of the 1860ndash1913 perioddid indeed hold a disproportionate fraction oftheir wealth in low-yield assets (such as gov-ernment bonds) An alternative hypothesishowever is that steeper age-wealth profileswere the consequence of the growth of financialmarkets as their children faced fewer creditconstraints parents decided to hold on to moreof their wealth

VI Conclusion

Evidence from wealth at death in Paris and inFrance over the last two centuries reveals threekey patterns First wealth concentration haschanged dramatically over time In 1807 thetop-1-percent share of wealth (40 percent inFrance 50 percent in Paris) was twice as high asit would be in 1994 but substantially less thanin 1913 when it peaked above 55 percent inFrance and 70 percent in Paris Some of thesechanges were due to economic phenomena thathave long been emphasized as creating inequal-ity namely industrialization and financial cen-

37 Thus far this literature has concentrated upon cross-country regressions of inequality on growth a methodologythat raises serious identification problems especially giventhe low quality of available international datasets on in-equality which are neither long-run nor homogeneous (seeeg Atkinson and Andrea Brandolini 2001)

254 THE AMERICAN ECONOMIC REVIEW MARCH 2006

tralization Yet the decline comes largely fromadverse shocks rather than economic conver-gence These changes are of such magnitudethat they are not sensitive to whether one ex-amines wealth at death in Paris or in France orwhether one examines it directly rather thanconverting it to wealth of the living by an estatemultiplier method

Our second key result is that there was asignificant transition during the nineteenth cen-tury from an important role for real estate as aform of wealth to moveable assets as the keyform of wealth for the very rich Similarly theshare of wealth held by aristocrats first rose andthen was eclipsed by that of financiers and in-dustrialists in the second half of the nineteenthcentury Hence mobility within this highly un-equal society might have been quite high Yetthis conjecture is tempered by our third findingthe wealthy were getting older over time andolder relative to less wealthy decedents Suchaging among the very wealthy would have hadnegative consequences for growth if financialmarkets were imperfect This issue requires fur-ther investigation and we hope it will attractfuture research

REFERENCES

Atkinson Anthony B ldquoTop Incomes in the UKover the 20th Centuryrdquo Journal of the RoyalStatistical Society Series A (Statistics in So-ciety) 2005 168(2) pp 325ndash43

Atkinson Anthony B and Brandolini AndrealdquoPromise and Pitfalls in the Use of lsquoSecond-aryrsquo Data-Sets Income Inequality in OECDCountries as a Case Studyrdquo Journal of Eco-nomic Literature 2001 39(3) pp 771ndash99

Atkinson Anthony B and Harrison Alan J Dis-tribution of personal wealth in Britain Cam-bridge Cambridge University Press 1978

Atkinson Anthony B and Piketty Thomas Topincomes over the twentieth century A con-trast between European and English speak-ing countries Oxford Oxford UniversityPress (forthcoming)

Banerjee Abhijit V and Duflo Esther ldquoInequalityand Growth What Can the Data Sayrdquo Journalof Economic Growth 2003 8(3) pp 267ndash99

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoPourquoi la richesse ne

srsquoest-elle pas diffusee avec la croissance Ledegre zero de lrsquoinegalite et son evolution enFrance 1800ndash1940rdquo Histoire et mesure 200318(1ndash2) pp 147ndash98

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoDefense et illustration delrsquoenquete des 3000 familles Lrsquoexemple deson volet patrimonialrdquo Annales de demogra-phie historique 2004 19 pp 19ndash57

Danysz E S ldquoContribution a lrsquoetude des for-tunes privees drsquoapres les declarations de suc-cessionsrdquo Bulletin de la statistique generalede France 1934 XXIV (1) pp 111ndash171

Daumard Adeline Les fortunes francaise auXIXe siecle Enquete sur la repartition et lacomposition des capitaux prives a ParisLyon Lille Bordeaux et Toulouse drsquoapreslrsquoenregistrement des declarations de succes-sion Paris Mouton 1973

Gain Andre La restauration et les biens desemigres 2 vols Nancy Societe drsquoimpressiontypographique 1929

Jones Alice H American colonial wealth Doc-uments and methods 3 vols New York ArnoPress 1977

Kopczuk Wojciech and Saez Emmanuel ldquoTopWealth Shares in the United States 1916ndash2000 Evidence from Estate Tax ReturnsrdquoNational Tax Journal 2004 57(2) pp 445ndash87

Kuznets Simon ldquoEconomic Growth and IncomeInequalityrdquo American Economic Review1955 45(1) pp 1ndash28

Lampman Robert J The share of top wealth-holders in national wealth 1922ndash1956Princeton Princeton University Press 1962

Lescure Michel Les banques lrsquoEtat et lemarche immobilier en France a lrsquoepoquecontemporaine 1820ndash1940 Paris EditionslrsquoEcole des Hautes Etudes en Sciences Socia-les 1982

Levy-Leboyer Maurice and Bourguignon Fran-cois LrsquoEconomie francaise au XIXe siecleAnalyse macroeconomique Paris Eco-nomica 1985 English translation Cam-bridge Cambridge University Press 1990

Lindert Peter H ldquoUnequal English Wealthsince 1670rdquo Journal of Political Economy1986 94(6) pp 1127ndash62

Lindert Peter H ldquoThree Centuries of Inequalityin Britain and Americardquo in Anthony B At-kinson and Francois Bourguignon eds

255VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

age profile of wealth between 1807 and 1994(see Table 5) During the nineteenth century aswealth concentration was increasing the veryrich were getting older and older At the begin-ning of the nineteenth century in the aftermathof the French Revolution the richest individualswere those in their fifties they were typically100 percent richer on average than people intheir forties 25 percent richer than those in theirsixties and 40 percent richer than those in theirseventies and eighties By the 1870s howeverthe age-wealth pattern had become stronglymonotonic the richest individuals were the old-est individuals In 1902 people in their sixtiesand seventies bequeathed 150 percent morethan those in their fifties and those in theireighties 300 percent more On the eve of WorldWar I top wealth holders were old and likely tobe retired This pattern breaks some time duringthe 1914ndash1945 period33 In 1947 as well as in1994 we are back to a pattern where the richestindividuals are those in their fifties Overall theperiod of maximal wealth inequality (1860ndash1913) also appears to be a period characterizedby a very specific age profile of wealth and largeconcentration of assets among the elderly

Another way to analyze the changing age-wealth relationship is to look at average age by

top estate fractile34 In 1817 average age wasvirtually the same for the top 10 percent and thetop 1 percent of estates (or even slightly declin-ing) The average-age-per-fractile relationshipbecomes upward sloping during the nineteenthcentury and by 1902 those in the top 1 percentwere almost six years older than those in the top10 percent The relationship is flat in 1947 anddownward-sloping in 1994 Finally one canapply the estate multiplier method (see SectionIV above) and analyze how wealth concentra-tion by age group among the living changedover the course of the nineteenth century Thegeneral population in Paris did not becomeolder during the nineteenth century those aged60 or older made up about 15 percent of thepopulation in 1817 and after 1847 about 10 to11 percent35 The share of total wealth ownedby the elderly rose significantly however aswealth distribution worsened The wealth be-longing to those aged 60 or more rose fromabout 25 to 30 percent of the total at the begin-ning of the nineteenth century to about 40 to 45percent by the end of the century The wealthshare of those aged 70 or older doubled fromless than 10 percent to about 20 percent36

33 Existing evidence on the age-wealth profile for 1931(see Danysz 1934) suggests that the Great Depression andWorld War II (rather than World War I) played the leadingroles in breaking this pattern This is an issue we plan toinvestigate in future research

34 See Piketty et al (2004 Table 6)35 Although life expectancy was increasing which

should have led to large shares of population for oldergroups the city was also growing quickly The large num-ber of immigrants (who were typically in their twenties)increased the relative size of the younger cohorts (seePiketty et al 2004 Table 7 and Figure 11)

36 See Piketty et al (2004 Figure 12)

TABLE 5mdashTHE AGE PROFILE OF WEALTH AT DEATH IN PARIS 1817ndash1994(Average estate left by 50- 59-year-old 100)

20ndash29yr-old

30ndash39yr-old

40ndash49yr-old

50ndash59yr-old

60ndash69yr-old

70ndash79yr-old

80ndash89yr-old

90ndash99yr-old

1817 26 22 28 100 54 59 591827 44 50 53 100 88 87 601837 133 90 107 100 116 123 1101847 87 73 102 100 117 204 1321857 84 77 101 100 104 109 1451867 67 58 136 100 141 125 1541877 66 73 63 100 197 260 4301887 45 33 63 100 152 233 2951902 29 40 80 100 253 272 4011947 31 51 73 100 113 105 105 1091994 11 45 100 87 93 95 68

Source Authorsrsquo computations using estate tax returns (see Piketty et al 2004 Table A1 for detailed sources)

253VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

It is perhaps not surprising that inequalitybecame strongly correlated with age in the1860s Those who died at an old age in the1820s and 1830s had lived through the difficultyears of the French Revolution and the disloca-tion of the end of the Napoleonic period Theirability to accumulate wealth had been severelyimpaired Furthermore they were rather lesslikely to inherit much wealth since the Revo-lution wiped out the bond portfolios of theirparents through a prolonged period of high in-flation Those who died from the 1860s to theearly 1910s did not suffer from the adversitiesthat plagued their forebears Instead they en-joyed the fruits of the financial sector expansionthat began in the 1850s After 1947 we seem tohave returned to a situation quite like that of1817 Presumably the capital damages associ-ated with both world wars and the Great De-pression had a strong negative effect on thewealth holdings of older generations The per-sistence of a flat age-wealth profile until 1994 islikely to be associated with two factors First insocieties where income growth is rapid abso-lute wealth accumulation is faster by youngercohorts than by older ones because their in-comes are higher at every age This is an im-portant distinction between the nineteenthcentury and the twentieth century Furthermorehighly progressive rates of income and estatetaxation have probably made it more difficult toaccumulate large fortunes thereby flattening theobserved age-wealth profile

The more interesting (and more difficult)question relates to the possible efficiency im-pact of high wealth concentration and changingage-wealth profiles Although our data do notallow us to address efficiency issues in a rigor-ous way our results allow us to formulate anumber of hypotheses and to shed new light onthe ongoing debate on inequality and growth37

From a theoretical viewpoint whether highwealth concentration can have a negativegrowth impact depends critically on the exis-tence of credit constraints With first-best credit

markets money flows toward the best entrepre-neurs and investment projects irrespective ofthe initial distribution High levels of wealthconcentration can be bad from a social justiceviewpoint but they entail no efficiency lossWhen credit constraints bind however initialwealth matters and high levels of inequality canhurt growth Whether the loss is large or smalldepends on who owns the assets If the rich areefficient investors (they know which projects tofund etc) then wealth concentration may evenbe useful If the rich are retired rentiers how-ever investing their wealth in low-yield assets(or low-ability inheritors) then high wealthconcentration and credit constraints might pre-vent talented but penniless investors from un-dertaking efficient projects thereby entailingnegative growth consequences The data used inthis paper are not ideal to address whether creditconstraints were important in a country likeFrance at the end of the nineteenth century Ourresults suggest however that to the extentcredit constraints were indeed severe highwealth concentration did have a negativegrowth impact In order to investigate this hy-pothesis further one would need to gather moresystematic data on investment strategies andasset returns Preliminary evidence suggeststhat the rich elderly of the 1860ndash1913 perioddid indeed hold a disproportionate fraction oftheir wealth in low-yield assets (such as gov-ernment bonds) An alternative hypothesishowever is that steeper age-wealth profileswere the consequence of the growth of financialmarkets as their children faced fewer creditconstraints parents decided to hold on to moreof their wealth

VI Conclusion

Evidence from wealth at death in Paris and inFrance over the last two centuries reveals threekey patterns First wealth concentration haschanged dramatically over time In 1807 thetop-1-percent share of wealth (40 percent inFrance 50 percent in Paris) was twice as high asit would be in 1994 but substantially less thanin 1913 when it peaked above 55 percent inFrance and 70 percent in Paris Some of thesechanges were due to economic phenomena thathave long been emphasized as creating inequal-ity namely industrialization and financial cen-

37 Thus far this literature has concentrated upon cross-country regressions of inequality on growth a methodologythat raises serious identification problems especially giventhe low quality of available international datasets on in-equality which are neither long-run nor homogeneous (seeeg Atkinson and Andrea Brandolini 2001)

254 THE AMERICAN ECONOMIC REVIEW MARCH 2006

tralization Yet the decline comes largely fromadverse shocks rather than economic conver-gence These changes are of such magnitudethat they are not sensitive to whether one ex-amines wealth at death in Paris or in France orwhether one examines it directly rather thanconverting it to wealth of the living by an estatemultiplier method

Our second key result is that there was asignificant transition during the nineteenth cen-tury from an important role for real estate as aform of wealth to moveable assets as the keyform of wealth for the very rich Similarly theshare of wealth held by aristocrats first rose andthen was eclipsed by that of financiers and in-dustrialists in the second half of the nineteenthcentury Hence mobility within this highly un-equal society might have been quite high Yetthis conjecture is tempered by our third findingthe wealthy were getting older over time andolder relative to less wealthy decedents Suchaging among the very wealthy would have hadnegative consequences for growth if financialmarkets were imperfect This issue requires fur-ther investigation and we hope it will attractfuture research

REFERENCES

Atkinson Anthony B ldquoTop Incomes in the UKover the 20th Centuryrdquo Journal of the RoyalStatistical Society Series A (Statistics in So-ciety) 2005 168(2) pp 325ndash43

Atkinson Anthony B and Brandolini AndrealdquoPromise and Pitfalls in the Use of lsquoSecond-aryrsquo Data-Sets Income Inequality in OECDCountries as a Case Studyrdquo Journal of Eco-nomic Literature 2001 39(3) pp 771ndash99

Atkinson Anthony B and Harrison Alan J Dis-tribution of personal wealth in Britain Cam-bridge Cambridge University Press 1978

Atkinson Anthony B and Piketty Thomas Topincomes over the twentieth century A con-trast between European and English speak-ing countries Oxford Oxford UniversityPress (forthcoming)

Banerjee Abhijit V and Duflo Esther ldquoInequalityand Growth What Can the Data Sayrdquo Journalof Economic Growth 2003 8(3) pp 267ndash99

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoPourquoi la richesse ne

srsquoest-elle pas diffusee avec la croissance Ledegre zero de lrsquoinegalite et son evolution enFrance 1800ndash1940rdquo Histoire et mesure 200318(1ndash2) pp 147ndash98

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoDefense et illustration delrsquoenquete des 3000 familles Lrsquoexemple deson volet patrimonialrdquo Annales de demogra-phie historique 2004 19 pp 19ndash57

Danysz E S ldquoContribution a lrsquoetude des for-tunes privees drsquoapres les declarations de suc-cessionsrdquo Bulletin de la statistique generalede France 1934 XXIV (1) pp 111ndash171

Daumard Adeline Les fortunes francaise auXIXe siecle Enquete sur la repartition et lacomposition des capitaux prives a ParisLyon Lille Bordeaux et Toulouse drsquoapreslrsquoenregistrement des declarations de succes-sion Paris Mouton 1973

Gain Andre La restauration et les biens desemigres 2 vols Nancy Societe drsquoimpressiontypographique 1929

Jones Alice H American colonial wealth Doc-uments and methods 3 vols New York ArnoPress 1977

Kopczuk Wojciech and Saez Emmanuel ldquoTopWealth Shares in the United States 1916ndash2000 Evidence from Estate Tax ReturnsrdquoNational Tax Journal 2004 57(2) pp 445ndash87

Kuznets Simon ldquoEconomic Growth and IncomeInequalityrdquo American Economic Review1955 45(1) pp 1ndash28

Lampman Robert J The share of top wealth-holders in national wealth 1922ndash1956Princeton Princeton University Press 1962

Lescure Michel Les banques lrsquoEtat et lemarche immobilier en France a lrsquoepoquecontemporaine 1820ndash1940 Paris EditionslrsquoEcole des Hautes Etudes en Sciences Socia-les 1982

Levy-Leboyer Maurice and Bourguignon Fran-cois LrsquoEconomie francaise au XIXe siecleAnalyse macroeconomique Paris Eco-nomica 1985 English translation Cam-bridge Cambridge University Press 1990

Lindert Peter H ldquoUnequal English Wealthsince 1670rdquo Journal of Political Economy1986 94(6) pp 1127ndash62

Lindert Peter H ldquoThree Centuries of Inequalityin Britain and Americardquo in Anthony B At-kinson and Francois Bourguignon eds

255VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

It is perhaps not surprising that inequalitybecame strongly correlated with age in the1860s Those who died at an old age in the1820s and 1830s had lived through the difficultyears of the French Revolution and the disloca-tion of the end of the Napoleonic period Theirability to accumulate wealth had been severelyimpaired Furthermore they were rather lesslikely to inherit much wealth since the Revo-lution wiped out the bond portfolios of theirparents through a prolonged period of high in-flation Those who died from the 1860s to theearly 1910s did not suffer from the adversitiesthat plagued their forebears Instead they en-joyed the fruits of the financial sector expansionthat began in the 1850s After 1947 we seem tohave returned to a situation quite like that of1817 Presumably the capital damages associ-ated with both world wars and the Great De-pression had a strong negative effect on thewealth holdings of older generations The per-sistence of a flat age-wealth profile until 1994 islikely to be associated with two factors First insocieties where income growth is rapid abso-lute wealth accumulation is faster by youngercohorts than by older ones because their in-comes are higher at every age This is an im-portant distinction between the nineteenthcentury and the twentieth century Furthermorehighly progressive rates of income and estatetaxation have probably made it more difficult toaccumulate large fortunes thereby flattening theobserved age-wealth profile

The more interesting (and more difficult)question relates to the possible efficiency im-pact of high wealth concentration and changingage-wealth profiles Although our data do notallow us to address efficiency issues in a rigor-ous way our results allow us to formulate anumber of hypotheses and to shed new light onthe ongoing debate on inequality and growth37

From a theoretical viewpoint whether highwealth concentration can have a negativegrowth impact depends critically on the exis-tence of credit constraints With first-best credit

markets money flows toward the best entrepre-neurs and investment projects irrespective ofthe initial distribution High levels of wealthconcentration can be bad from a social justiceviewpoint but they entail no efficiency lossWhen credit constraints bind however initialwealth matters and high levels of inequality canhurt growth Whether the loss is large or smalldepends on who owns the assets If the rich areefficient investors (they know which projects tofund etc) then wealth concentration may evenbe useful If the rich are retired rentiers how-ever investing their wealth in low-yield assets(or low-ability inheritors) then high wealthconcentration and credit constraints might pre-vent talented but penniless investors from un-dertaking efficient projects thereby entailingnegative growth consequences The data used inthis paper are not ideal to address whether creditconstraints were important in a country likeFrance at the end of the nineteenth century Ourresults suggest however that to the extentcredit constraints were indeed severe highwealth concentration did have a negativegrowth impact In order to investigate this hy-pothesis further one would need to gather moresystematic data on investment strategies andasset returns Preliminary evidence suggeststhat the rich elderly of the 1860ndash1913 perioddid indeed hold a disproportionate fraction oftheir wealth in low-yield assets (such as gov-ernment bonds) An alternative hypothesishowever is that steeper age-wealth profileswere the consequence of the growth of financialmarkets as their children faced fewer creditconstraints parents decided to hold on to moreof their wealth

VI Conclusion

Evidence from wealth at death in Paris and inFrance over the last two centuries reveals threekey patterns First wealth concentration haschanged dramatically over time In 1807 thetop-1-percent share of wealth (40 percent inFrance 50 percent in Paris) was twice as high asit would be in 1994 but substantially less thanin 1913 when it peaked above 55 percent inFrance and 70 percent in Paris Some of thesechanges were due to economic phenomena thathave long been emphasized as creating inequal-ity namely industrialization and financial cen-

37 Thus far this literature has concentrated upon cross-country regressions of inequality on growth a methodologythat raises serious identification problems especially giventhe low quality of available international datasets on in-equality which are neither long-run nor homogeneous (seeeg Atkinson and Andrea Brandolini 2001)

254 THE AMERICAN ECONOMIC REVIEW MARCH 2006

tralization Yet the decline comes largely fromadverse shocks rather than economic conver-gence These changes are of such magnitudethat they are not sensitive to whether one ex-amines wealth at death in Paris or in France orwhether one examines it directly rather thanconverting it to wealth of the living by an estatemultiplier method

Our second key result is that there was asignificant transition during the nineteenth cen-tury from an important role for real estate as aform of wealth to moveable assets as the keyform of wealth for the very rich Similarly theshare of wealth held by aristocrats first rose andthen was eclipsed by that of financiers and in-dustrialists in the second half of the nineteenthcentury Hence mobility within this highly un-equal society might have been quite high Yetthis conjecture is tempered by our third findingthe wealthy were getting older over time andolder relative to less wealthy decedents Suchaging among the very wealthy would have hadnegative consequences for growth if financialmarkets were imperfect This issue requires fur-ther investigation and we hope it will attractfuture research

REFERENCES

Atkinson Anthony B ldquoTop Incomes in the UKover the 20th Centuryrdquo Journal of the RoyalStatistical Society Series A (Statistics in So-ciety) 2005 168(2) pp 325ndash43

Atkinson Anthony B and Brandolini AndrealdquoPromise and Pitfalls in the Use of lsquoSecond-aryrsquo Data-Sets Income Inequality in OECDCountries as a Case Studyrdquo Journal of Eco-nomic Literature 2001 39(3) pp 771ndash99

Atkinson Anthony B and Harrison Alan J Dis-tribution of personal wealth in Britain Cam-bridge Cambridge University Press 1978

Atkinson Anthony B and Piketty Thomas Topincomes over the twentieth century A con-trast between European and English speak-ing countries Oxford Oxford UniversityPress (forthcoming)

Banerjee Abhijit V and Duflo Esther ldquoInequalityand Growth What Can the Data Sayrdquo Journalof Economic Growth 2003 8(3) pp 267ndash99

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoPourquoi la richesse ne

srsquoest-elle pas diffusee avec la croissance Ledegre zero de lrsquoinegalite et son evolution enFrance 1800ndash1940rdquo Histoire et mesure 200318(1ndash2) pp 147ndash98

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoDefense et illustration delrsquoenquete des 3000 familles Lrsquoexemple deson volet patrimonialrdquo Annales de demogra-phie historique 2004 19 pp 19ndash57

Danysz E S ldquoContribution a lrsquoetude des for-tunes privees drsquoapres les declarations de suc-cessionsrdquo Bulletin de la statistique generalede France 1934 XXIV (1) pp 111ndash171

Daumard Adeline Les fortunes francaise auXIXe siecle Enquete sur la repartition et lacomposition des capitaux prives a ParisLyon Lille Bordeaux et Toulouse drsquoapreslrsquoenregistrement des declarations de succes-sion Paris Mouton 1973

Gain Andre La restauration et les biens desemigres 2 vols Nancy Societe drsquoimpressiontypographique 1929

Jones Alice H American colonial wealth Doc-uments and methods 3 vols New York ArnoPress 1977

Kopczuk Wojciech and Saez Emmanuel ldquoTopWealth Shares in the United States 1916ndash2000 Evidence from Estate Tax ReturnsrdquoNational Tax Journal 2004 57(2) pp 445ndash87

Kuznets Simon ldquoEconomic Growth and IncomeInequalityrdquo American Economic Review1955 45(1) pp 1ndash28

Lampman Robert J The share of top wealth-holders in national wealth 1922ndash1956Princeton Princeton University Press 1962

Lescure Michel Les banques lrsquoEtat et lemarche immobilier en France a lrsquoepoquecontemporaine 1820ndash1940 Paris EditionslrsquoEcole des Hautes Etudes en Sciences Socia-les 1982

Levy-Leboyer Maurice and Bourguignon Fran-cois LrsquoEconomie francaise au XIXe siecleAnalyse macroeconomique Paris Eco-nomica 1985 English translation Cam-bridge Cambridge University Press 1990

Lindert Peter H ldquoUnequal English Wealthsince 1670rdquo Journal of Political Economy1986 94(6) pp 1127ndash62

Lindert Peter H ldquoThree Centuries of Inequalityin Britain and Americardquo in Anthony B At-kinson and Francois Bourguignon eds

255VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

tralization Yet the decline comes largely fromadverse shocks rather than economic conver-gence These changes are of such magnitudethat they are not sensitive to whether one ex-amines wealth at death in Paris or in France orwhether one examines it directly rather thanconverting it to wealth of the living by an estatemultiplier method

Our second key result is that there was asignificant transition during the nineteenth cen-tury from an important role for real estate as aform of wealth to moveable assets as the keyform of wealth for the very rich Similarly theshare of wealth held by aristocrats first rose andthen was eclipsed by that of financiers and in-dustrialists in the second half of the nineteenthcentury Hence mobility within this highly un-equal society might have been quite high Yetthis conjecture is tempered by our third findingthe wealthy were getting older over time andolder relative to less wealthy decedents Suchaging among the very wealthy would have hadnegative consequences for growth if financialmarkets were imperfect This issue requires fur-ther investigation and we hope it will attractfuture research

REFERENCES

Atkinson Anthony B ldquoTop Incomes in the UKover the 20th Centuryrdquo Journal of the RoyalStatistical Society Series A (Statistics in So-ciety) 2005 168(2) pp 325ndash43

Atkinson Anthony B and Brandolini AndrealdquoPromise and Pitfalls in the Use of lsquoSecond-aryrsquo Data-Sets Income Inequality in OECDCountries as a Case Studyrdquo Journal of Eco-nomic Literature 2001 39(3) pp 771ndash99

Atkinson Anthony B and Harrison Alan J Dis-tribution of personal wealth in Britain Cam-bridge Cambridge University Press 1978

Atkinson Anthony B and Piketty Thomas Topincomes over the twentieth century A con-trast between European and English speak-ing countries Oxford Oxford UniversityPress (forthcoming)

Banerjee Abhijit V and Duflo Esther ldquoInequalityand Growth What Can the Data Sayrdquo Journalof Economic Growth 2003 8(3) pp 267ndash99

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoPourquoi la richesse ne

srsquoest-elle pas diffusee avec la croissance Ledegre zero de lrsquoinegalite et son evolution enFrance 1800ndash1940rdquo Histoire et mesure 200318(1ndash2) pp 147ndash98

Bourdieu Jerome Postel-Vinay Gilles and Suwa-Eisenmann Akiko ldquoDefense et illustration delrsquoenquete des 3000 familles Lrsquoexemple deson volet patrimonialrdquo Annales de demogra-phie historique 2004 19 pp 19ndash57

Danysz E S ldquoContribution a lrsquoetude des for-tunes privees drsquoapres les declarations de suc-cessionsrdquo Bulletin de la statistique generalede France 1934 XXIV (1) pp 111ndash171

Daumard Adeline Les fortunes francaise auXIXe siecle Enquete sur la repartition et lacomposition des capitaux prives a ParisLyon Lille Bordeaux et Toulouse drsquoapreslrsquoenregistrement des declarations de succes-sion Paris Mouton 1973

Gain Andre La restauration et les biens desemigres 2 vols Nancy Societe drsquoimpressiontypographique 1929

Jones Alice H American colonial wealth Doc-uments and methods 3 vols New York ArnoPress 1977

Kopczuk Wojciech and Saez Emmanuel ldquoTopWealth Shares in the United States 1916ndash2000 Evidence from Estate Tax ReturnsrdquoNational Tax Journal 2004 57(2) pp 445ndash87

Kuznets Simon ldquoEconomic Growth and IncomeInequalityrdquo American Economic Review1955 45(1) pp 1ndash28

Lampman Robert J The share of top wealth-holders in national wealth 1922ndash1956Princeton Princeton University Press 1962

Lescure Michel Les banques lrsquoEtat et lemarche immobilier en France a lrsquoepoquecontemporaine 1820ndash1940 Paris EditionslrsquoEcole des Hautes Etudes en Sciences Socia-les 1982

Levy-Leboyer Maurice and Bourguignon Fran-cois LrsquoEconomie francaise au XIXe siecleAnalyse macroeconomique Paris Eco-nomica 1985 English translation Cam-bridge Cambridge University Press 1990

Lindert Peter H ldquoUnequal English Wealthsince 1670rdquo Journal of Political Economy1986 94(6) pp 1127ndash62

Lindert Peter H ldquoThree Centuries of Inequalityin Britain and Americardquo in Anthony B At-kinson and Francois Bourguignon eds

255VOL 96 NO 1 PIKETTY ET AL WEALTH CONCENTRATION IN PARIS AND FRANCE 1807ndash1994

Handbook of income distribution Vol 1Amsterdam Elsevier Science North-Hol-land 2000 pp 167ndash216

Morrisson Christian ldquoHistorical Perspectiveson Income Distribution The Case of Eu-roperdquo in Anthony B Atkinson and FrancoisBourguignon eds Handbook of income dis-tribution Vol 1 Amsterdam Elsevier Sci-ence North-Holland 2000 pp 217ndash60

Morrisson Christian and Snyder Wayne ldquoLesinegalites de revenus en France du debut duXVIIIeme siecle a 1985rdquo Revue economique2000 51(1) pp 119ndash54

Piketty Thomas Les hauts revenus en Franceau XXe sieclemdashInegalites et redistributions1901ndash1998 Paris Grasset 2001

Piketty Thomas ldquoIncome Inequality in France1901ndash1998rdquo Journal of Political Economy2003 111(5) pp 1004ndash42

Piketty Thomas Postel-Vinay Gilles andRosenthal Jean-Laurent ldquoWealth Concentra-tion in a Developing Economy Paris andFrance 1807ndash1994rdquo Center for EconomicPolicy Research CEPR Discussion PapersNo 4631 2004

Piketty Thomas and Saez Emmanuel ldquoIncomeInequality in the United States 1913ndash1998rdquoQuarterly Journal of Economics 2003118(1) pp 1ndash39

Soltow Lee and van Zanden Jan L Income andwealth inequality in the Netherlands 16thndash20thcentury Amsterdam Het Spinhuis 1998

256 THE AMERICAN ECONOMIC REVIEW MARCH 2006