The Time-Varying NAIRU and its Implications for Economic Policy.pdf

8/12/2019 The Time-Varying NAIRU and its Implications for Economic Policy.pdf

1/23

Journal of Econom ic Perspectives Volume 11, Number 1 Winter 1997 Pages 1132

The Time-Varying NAIRU and itsImplications for Economic Policy

Robert J. Gordon

T he relationship between inflation and unemployment is central to the con-duct of monetary policy. More than 35 years ago, Paul Samuelson andRobert Solow (1960) coined the term Phillips curve at the 1959 AEAmeetings to describe that relationship, reacting to the publication of Phillips's(1958) seminal article a year earlier. A few years later, Milton Friedman (1968)coined the term natu ral rate of unem ploym ent, which more recently has cometo be known by the acronym NA IRU, standing for the N on-Accelerating I nflationRate ofUnemployment. If a unique NAIRU exists, then the Phillips curve tradeoffis vertical at that unemployment rate. The Federal Reserve cannot make the actualunemployment rate differ from the NAIRU in the long run, but it can maintain astable rate of inflation if it succeeds in setting the actual unem ploym ent ra te equ alto the NAIRU. If instead of maintaining a stable rate of inflation, the Fed desiresto reduce the inflation rate toward zero or some other target, then it needs to keepthe actual unemployment rate above the NAIRU. Whether the goal is steady infla-tion or lower inflation, the Fed needs to know the value of the NAIRU.

For many years it was reasonable to assume that the NAIRU was 6.0 percent. Itested that assumption repeatedly by runn ing dynamic simulations of regression equa-tions that predicted the rate of inflation, using an assumed NAIRU of 6.0 percent andother explanatory variables with their lags. These simulations were dynamic, since theestimates for one time period then created the lagged values of the inflation rateneeded for estimates of inflation in future time periods (that is, no information onthe actual inflation rate was used for the period of the simulation). Such simulations

Robert J. Gordo n is Stanley G. Harris Professor in the Social Sciences, Northwestern Uni-versity, Evanston, Illinois, and Research Associate, National Bureau of Econom ic Research ,Cambridge, Massachusetts.


2/23

12 Journal of Econom ic Perspectives

were capable of tracking the inflation rate accurately for many years after the end ofthe regression sample periodfor example, regressions using pre-1987 data couldaccurately predict inflation for the period 19871994 without any appreciable drift.Such postsample simulations provided evidence that the NAIRU had remained at6.0 percent.1The substantial acceleration of inflation that occurred in 198889, whenthe unemployment rate fell below 6.0 percent for a period of three years, is consistentwith the view that the NAIRU was at 6.0 percent or above as recently as 198889.However, the NAIRU is not carved in stone. In Friedman's (1968) interpreta-tion, the NAIRU is gro und ou t by the set of Walrasian microeconomic rela-tions in the economy, including the structure and institutions of product and labormarkets, and these relations can change. Numerous factors have changed since198889 in a way that may have reduced the NAIRU. Labor unions are weak, andtheir penetration in the labor force continues to decline. Manufacturers have beenunder intense pressure from consumers and foreign competitors to restrain priceincreases. The rest of the industrial world has experienced a sluggish recovery, andample foreign capacity exists to provide supp lies to U.S. manufacturers . Steady pricedeclines in the computer and high-tech sectors are beginning to put downwardpressure on the economywide inflation rate. Some business executives argue thatthe economy has changed drastically in th e last 10 years; as General Electric's Jo hnF.Welch, Jr., recently stated (Stevenson, 1996): Th ere is no inflation . . . thereis no pricing power at all.

The Fed also acts as if it accepts that the NAIRU can move. In early 1994, forexample, the Fed implicitly believed that the NAIRU was around 6.0 percent andsharply raised short-term interest rates when it correctly predicted that the actualunemployment rate was about to fall below 6.0 percent. But for most of 1995 andearly 1996, the Fed then allowed short-term interest rates to drift down slightly wheninflation did not accelerate in response to an average unemployment rate wellbelow 6.0 perc ent. T he absence of an acceleration of inflation in 199596 suggeststhat the NAIRU may have fallen below 6.0 percent.Has the NAIRU in fact declined? If so, from what level a decade ago to whatlevel today? Surprisingly, macroeconomists have thus far provided no answer to thisquestion that can be taken off the shelf by policymakers. In contrast to feverishresearch activity in the 1960s and 1970s, remarkably little research has been con-ducted on the U.S. inflation process in the past decade, so little that King andWatson (1994) comment on the quie scen ce of the field. The interpretation ofthe lack of attention to quantification of the inflation-unemployment tradeoff, an dof the current value of the NAIRU, differs considerably among macroeconomists.

1Th e inflation equ ation was develo ped in a series of pap ers, includ ing Go rdo n (1970, 1975, 1977a and1982b). King and Watson (1994) have called this app roa ch the Gord on-Solo w mo de l, citing the firstof my papers and Solow's 1969 book. To determine whether the inf lation relationship has changed, Ihave maintained unchanged the set of variables, lag lengths and other features of the equation intro-duced in Gordon (1982b) and Gord on a nd King (1982). Th e most recent assessments of the perfor-mance of this equation are contained in Gordon (1990a, 1994).


3/23

RobertJ.Gordon 13

Some have focused on other topics because they believe the U.S. inflation processis so stable, and the models developed in the early 1980s work so well, that therehave been no behavioral mysteries to solve. Other macroeconomists have turnedaway because they believe that searching for a link between NAIRU and inflationhas constituted a failed and unproductive line of research.In the interp retation of King and Watson (1994, p. 160), many Keynesianeconom ists have con tinued to view the Phillips curve as an essentially intact struc-tural relation, once the original econometric models of the 1960s were amended to re present supply shocks and build in a zero long-runtradeoff. Once this taskhad been accomplished (as in Gordon, 1977a), there was no agenda warrantingcontinued research, except periodically checking that the relation remained stable.Indeed, King and Watson (p. 160) note that indeed the remarkable feature of thePhillips curve in the post war period was itsstability

Across the street from the Keynesians are the neoclassical and monetarist econ-omists. Some of them have dismissed the Phillips curve as eco nometric failure ona grand scale (Lucas and Sargent, 1978), since the long-run negative correlationbetween inflation and unemployment predicted by the models of the late 1960scontrasted with the distinctly positive correlation in the data of the 1970s. At thatpoint, m any neoclassical economists stopped paying attention to empirical work onthe Phillips correlation and either did not notice or did not take seriously the newbreed of post-1975 Phillips curve estimates that incorporated a vertical long-runtradeoff and included supply shock variables. Instead of taking inflation as thevariable to be determined by their models, neoclassical economists turned to realtheories of aggregate output fluctuations in which the behavior of inflation wasneither explored nor explained. Implicitly, the price level was left as a residualthat is, as the level of nominal GDP (in turn often equal to th e money supply plusa stochastic error term) divided by whatever level of real GDP was determined bythe m odel. This treatment was the diam etric opposite to that implied in the Phillipscurve approach, in which an equation is specified to determine the inflation rate,while the growth ra te of real GDP is implicitly a residual equal to the rate of nominalGDP growth minus the rate of inflation.The NAIRU is meaningful only within a well-specified model of the inflation

process. In the nex t section, I will describe the mainstream triangle m od el of theinflation process that incorporated and resurrected the Phillips curve from whatLucas and Sargent (1978) had called the wreckage of the 1960s and early 1970s.Then, instead of assuming a value for the NAIRU and testing the validity of thatassumed value in dynamic simulations, this paper adapts an explicit econometrictechnique that allows a time-varying NAIRU to be estimated. We emerge with a setof alternative NAIRU estimates for the 19551996 period that differ only moder-ately from each other depending on which inflation index and sample period isused. I then use preferred versions of the inflation equation, together with alter-native hypothetical paths for the actual unemployment rate, to simulate the infla-tion rate in future years. The paper concludes by examining implications for thepast and future conduct of monetary policy.


4/23


Several important topics lieoutside of th eagenda of this paper. First,it isconcerned with th eU.S. inflation process a n ddoes n o ttreat th equi te differentbehavior ofinflation a n dunemployment inEurope o r Japan . Second, itestimatest h e time varying NAIRU within th econtexto fthe trian gle modelofth e inflationprocess developed in myprevious work; itdoes n o tdevelop such amo del fromscratch. Third ,itaskswhich un em ploym ent rat e shouldbethe F ed's target but doesn o t inquire intoth emethodsbywhich th eFed should attem pttoaccomplish thatgoalthat is, itdoes n o tstudy th echannelsofmo net ary policy tha t link chan gesin th eshort term interest r ate tosubsequen t lagged responses ofoutput, income,employmenta n dunem ployment.

Topreviewth e main conclusions,th e alternative estimatesallsupport the conclusion that th e NAIRUhasdeclined substantially since 1988 89, open ingup t heopportunityfor th eFedtomaintainalower un employment rat e thanwasfeasible then.Aswe shall see, th eNAIRUh asexhibited pron ounced cycles over th epostwar period,albeit within asurprisingly narrow range. F urther,t h econsequences of amistakebyt h eF ed that reduces th eactual unemploymen t rateafullpercentage point belowtheNAIRU a resurprisingly modest;th einflation rate accelerates byonly0.3percentperyear foreverypoint thatt h eactual unem ploymen t rate remains below th eNAIRU.

The Triangle ModelofInflationTh e Ph illips cur ve hasbecome ageneric termfor an yrelation betweenth e

rateofchangeof a nominal price o rwage and t helevelof a real indicator of th eintensityofdemandin th eeconomy, suchas th eunem ployment rate.I n t h e1970s,t h esimple Ph illips relationwasamendedbyincorp ora tin g supply shocks andazerolong run tradeoff.2 What emerged was aninterpretation of th ePh illips curve th atI call th e trian gle model ofinflat ion a label summarizing th edependenceo ft h e inflation rateo nthree basic det erminan ts: inertia, de ma ndan dsupply.

F o rexample, a general specification ofthis framework wouldbet =a(L)t 1 +b(L)Dt +c(L)zt +et.

The dependent variable t is th einflation rate. Inertiaisconveyed by th elagged rateof inflation t1.Dtis anindexofexcessdeman d (normalizedsothatDt= 0indicatesth eabsence ofexcessdemand),ztis avectorofsupply shock variables (normalizedsot h a tzt= 0indicatesa nabsenceofsupply shocks),a n detis aseriallyuncorrelated errort e r m .Lowercase letters designatefirstdifferences oflogarithms, uppercase letters designate logarithms oflevels,an dLis apolynomial in th e lagoperator.

Usually, this equationwillinclude severallagsofpast inflation rates.Ifth esum

Schultze (1975) a n dG o r d o n (1975) also introduced explicit variables toisolate t h eeffect offoodan denergy pricesan dprice controlson t heU.S. inflation rate.


5/23

The Time Varying NAIRU and its Implications for Economic Policy 15

of th e coefficients on these lagged inflation values equals unity, then there is a nat ura l ra te of the dem an d variable (DNt) consisten t with a const an t rat e of inflat i o n . 3While th e sum of th e coefficients onlagged inflation is usually roughly equalt ounity, that sum must be constrained to be exactlyunity for a meaningful na tu ralrat e of th e de man d variable to be calculated.

Amon g th e de man d variables th at have been en te red as proxies for Dt are the o u t p u tgap , defined as th e log ratio of actual to na tur al (or potent ial) real G DP,t h e un em plo ymen t gap, defined as the difference between the actual an d nat uralrate of unem ployment (orN A I R U ) ,and the rate of capacity utilization.4 The equations estimated in this paper use current and lagged values of the unemploymentgap as a proxy for theexcessdemand parameterDt, where the unemployment gapis defined as the difference between t he actual rate of un em ploym en t and thenatural rate, and the natural rate is allowed to vary over time. Using the unemployment rate as a predictor of inflation can bejustified by findings like those ofKing an d Watson (1994), who find th at un em ploym en t causes inflation in the

G ranger causation sense, by precedin g it in time.Although the focus here is on using the unemployment gap to predict inflation,

t h eultimate exogenous d em an d factor in this model is excessnom inal GD P growth,which is th e exten t to which growth of nomin al GD P exceeds the growth of potent ialo u t p u t In turn,excess nominal GDP growth in any time period, together with theinflation rate calculated from the triangle model's inflation equation,willdeterminet h echange in the output gap, which in turnwilldetermine the change in the unemployment gap. By treating excess nominal GDP growth as exogenous, the trianglemodel focuses on the inflation process without the distraction of building a model oft h e determinants of aggregate demand. Admittedly, this simplification sweeps twothirds of macroeconomics under the rug. Moreover, it ignores channels by whichinflation feeds back into th e determination of nom inal GDP .

We are int erested in estimating the NAIRU , which is th e un em ploym en t ratet h a t is consistent with steady inflation. The structure of the triangle model, with itsdistinction between demand and supply shocks, suggestsa particular conception oft h e NAIRU. Th e stan dard con cept is th e no supply shock NAIRU, tha t is, theunemployment rate that is consistent with steady inflation in theabsence ofsupplyshocks. To pu t it an ot he r way, if th e inflation ra te suddenly exhibits a spi ke th atis ent irely explain ed by th e supply shock variables, th en the no supply shockN A I R U measures the unemployment rate that would becom patible with steady

3 T h eintu ition beh in d this po int may be clarified by con sidering t he simple case where t he re is only on elag of inflation; t h e n ,obviously, a stable inflation rate m ean s that th e coefficient on th e inflation rate int h e previous period must be equal to 1. The same intuition holds more generally with several lags ofinflation. Ok un 's Law holds that the unem ploymen t gap and the outp ut gap are closely related. Con sider anOkun 's Law equation relating the unem ployment gap to curren t and lagged values of the log ratio ofactual to natural real GDP:Ut UNt= (L)log(Y t/ YNt) + et. Empirically the sum of the coefficients hasbeen ar ou n d 0.5 for most of th e postwar period , alth ou gh in the 1990 95 subinterval tha t sum hasbeen close to 1.0.


6/23


inflation in the abs enc e of thos e supply shocks. W ithou t this qualification, th en theNAIRU would ju m p a rou nd as supply shocks arr ived and dep arte d, which is no twhat most economists are trying to convey when they speak of the natural rate ofunemploymen t .The traditional Phillips curve specification of the 1960s and early 1970s in-cluded only lagged inflation and the unemployment rate, and omitted supplyshocks. This created an obvious problem of omitted variables, since supply shockscan create an extraneous positive correlation between inflation and unemployment.Thus, the failure to include supply shocks means that unemployment explains asmaller share of the variation of the inflation rate; in fact, the coefficient on un-em ploy me nt in such a regression will be biased toward zero and is l ikely to pr odu ceunreliable predictions in periods when supply shocks are absent. To the extent thatsupply shocks are includ ed in the equation b ut are imperfectly meas ured, an d ther eis con tem por an eou s feedback from inflation to nom inal GDP, using the un em -ployment gap (or the output gap) as a proxy for the demand variable will yield acoefficient that is biased toward zero. The m ore accurately the influence of supplyshocks is measured, the smaller the bias (Gordon, 1990b, Table 1, p. 1121). 5

Inflation depends on both the level and change in the demand variable,whe ther the une mp loym ent gap or the ou tput gap is used as the dem and var iab le.6T he rate-of-change effect is autom atically allowed to ent er as lon g as the ga p variableis en tere d with mo re than one lag; in oth er words, if the gap variable is ent ere d as,say, the current value and one lagged value, this contains precisely the same infor-mation as entering the current level and change from the previous period. Timeseries equations that do not allow for the change effect, whether by entering itdirectly or by allowing the level of the gap to enter with one or more lags, aremisspecif ied. T he c han ge effect is particularly im port ant in explaining m acro-econ om ic price beha vior in the 1930s, a result that I have foun d previously and thatRomer (1996) has recently validated.

Two final issues con cer ning the tr iangle mo del involve the role of wage ch angesand the role of expectations. After all, the original Phillips article was about therelat ion between wage changes and unemployment , and la ter formulat ions addeda term for expected inflation. But the tr iangle model as summarized here has noexpectations and no wages. These are issues of substantive significance.

T he om ission of expecta tions is delibe rate. Much at tentio n was diverted in thelate 1960s and early 1970s to the interpretation of the lagged effect of prices onwages as reflecting adaptive lags in th e formation of expec tations. Since the n, i t hasbecome clear that price and wage inertia is compatible with rational expectations.The speed of price adjustment and the speed of expectation formation are twodifferent issues. Price adjust me nt can be delayed by wage an d price con tracts , andby the time needed for cost increases to percolate through the input-output table,5Alternatively, estimates of the triangle-type inflation th at use nom ina l GDP as a proxy for the de m an d(D t) variable will yield a coefficient on nominal GDP that is biased away from zero.6 I first noted the importance of the rate-of-change effect in Gordon (1977a, pp. 270271).


7/23

Robert J. Gordon 17

and yet everyone can form expectations promptly and rationally based on full in-forma tion ab ou t th e aggregate price level. T he role of the lagged inflation ter ms isto capture the dynamics of inertia, whether related to expectation formation, con-tracts,delivery lags or anything else.The omission of wages in the triangle model is deliberate as well. The earlierfixation on wages was a mistake. The relation of prices to wages has changed overt ime;for exam ple, labor 's share in national inco me exhibits a strong upward move-ment between the mid-1960s and early 1970s that has not been adequately ex-plained. The Fed's goal is to control inflation, not wage growth, and models withseparate wage growth an d price ma rkup equations d o no t perform as well as theequation above, in which wages are only implicit. 7 By treating the relationship ofinflation to unem ploy me nt, rathe r than of wage chan ge to unem ploym ent, thetriangle approach returns to the framework of the original Samuelson and Solowarticle (1960) that coine d the term Phillips curv e and plotted U.S. data on theinflation-unemploym ent quad rant. T he earliest credit for ignorin g wages is claimedby Irving Fisher (1926 [197 3]), whose ne glec ted article discovered the Phillips curvein the form of a relationship between the unemployment rate and price changes,not wage changes.

Implications of the Triangle ModelTh e tr iangle m ode l generates several clear implications for how to think a boutinflation, unemployment and the relation between them.First, in the long run, inflation is always and everywhere an excess nominal

GDP phe no m en on . Supply shocks will come and g o. W hat rema ins to sustain long-run inflation is steady growth of nominal GDP in excess of the growth of naturalor potential real output.

Second, supply shocks can cause a positive correlation between inflation andthe unem ploy me nt gap. The observation that the Phillips curve correlation betweeninflation and unemployment was positive rather than negative in the 1970s is con-sistent with the triangle model, due to its explicit treatment of supply shocks suchas the rise and eventual fall of oil prices.

Third, since supply shocks do influence the inflation rate, targeting the no-supply-shock NAIRU, or the equivalent level of natural or potential real GDP, willlead to difficulties. For example, in a decade like the 1970s with significant, seriallycorrelated and adverse supply shocks, attempting to push unemployment down toa previously determined natural rate will lead to an acceleration in nominal GDP

7For a com plete set of wage equation s for both the Uni ted States and for German y, using the sam egen eral specification as in this pap er, see Franz and Gordo n (1993). Th at pap er det erm ine d tha t theU.S. wage NAIRU for 1990 was 6.2 percent, almost exactly the same as estimated in this paper for theGDP deflator by the time-varying approach described below.


8/23

18 Journal of Economic Perspectives

growth as the centra l bank is forced to acc om mo da te th e inflation cau sed by thesupply shocks, and hence it will lead to a permanent acceleration of inflation.

Fourth, growth in the money supply is not a unique cause of inflation. Whatmatters is excess nom inal GDP growth, which de pe nd s not jus t on th e rate of m on-etary growth but also in the growth in the velocity of money. In a literal sense, thetriangle model predicts inflation without using information on the money stock. Inan economic sense, this implies that any long-term effect of money growth on in-f lation operates through channels that are captured by the real excess demandvariables.

Fifth, in the short run , fluctuations in excess nom ina l GDP grow th lead toclockwise loops on a diagram plotting the unemployment gap on the horizontalaxis versus inflation on the vertical axis. For ex am ple , an increase in excess no mi nalGDP growth will first show up as a reduction in the unemployment gap, movingleft on the diagram, an d th en in a rise in inflation, an upw ard move men t. The loopscome from inertia, the fact that because current inflation depends partly on pastvalues of inflation, it will respond slowly to a change in the unemployment gap.

Sixth, the triangle m ode l is resolutely Keynesian. Prices are p reve nted by inertiaand by the finite Phillips curve adjustment coefficient from mimicking changes innominal GDP growth. However, the tr iangle model does not incorporate the im-plication that King and Watson (1994) attr ibute to their Key nesian straw ma nthat unem ploym ent is dom inated by aggregate dem and d is turbances . Ins tead ,both demand and supply shocks influence both the inflation rate and the unem-ployment rate.

Seventh, since excess nominal demand is the ultimate cause of inflation, asensible anti-inflation policy shou ld target this variable in a direct way. O ne straight-forward approach would be for monetary policy to target excess nominal GDPgrowth itself. Such a policy, advocated by a number of prominent economists, in-sures that the econom y has a nom ina l an ch or that prevents an acceleration ofinflation and represents a compromise response to adverse supply shocks, whichwould then cause an increase in both unemployment and inflation rather than justin one or the other.

Validation of the Triangle ModelThe textbook version of the tr iangle model came first , and the econometrics

and theory followed. A diagrammatic version of the model originated in a classroomha nd ou t that Rudiger Dornb usch d eveloped at the Chicago Business School in early1975. I laid out the basic equations in 1976, in a paper presented at the AEA meet-ings (Go rdon, 1977b). Th e e conom etric version, developed in the late 1970s, wasvalidated in 198187 when the sacrif ice rat io expe rienc ed by the econ om y thatis , the percentage loss in output associated with the deceleration of inflation thatocc urred corre spon ded almost exactly to what had been pred icted in advance on


9/23

The Time-Varying NAIRU and its Implications for Econom ic Policy 19

the basis of parameters estimated through the end of 1980. 8Versions of the equa -tion estimated through 1987 in postsample dynamic simulations tracked quite pre-cisely the acceleration of inflation observed in 19871990 and the deceleration ofinflation that occurred in 199093.T he positive perform ance of the tr iangle mode l stands in sharp contrast to theshambles in which the Phillips curve literature found itself in the mid-1970s. Acentral point of departure for Lucas's new classical revolution was the failure of the1960s Phillips curve. In the language of Lucas and Sargent (1978, pp. 4950), [T]hat these predictions were wildly incorrect, and that the doctr ine on whichthey we re based is fundam entally flawed, are now simple m atters of fact . . . thetask whic h faces cont em por ary stud ents of the bu siness cycle [is] that of sortingthrou gh the wreckage . . . of that rema rkable intellectual event called the Keynes-ian Revolution. Th e tr iangle model was in prin t in its prese nt form before Lucasand Sargent wrote those lines; it has survived and thrived, while empirical attemptsby Ro bert B arro (1978) an d ot her s to validate the new classical prop ositi on of policyineffectiveness have failed, running aground on the bedrock that inflation inertiaexists, and marke ts do n ot clear quickly eno ugh to avoid a substantial me dium -runimpact of nominal demand shocks on output and unemployment . 9

Estimating a Time-Varying NAIRU

For almost two decades, a time series for the NAIRU has been published in mymacroeconomics textbook. The series starts with a NAIRU of a bit more than5 percent in the late 1950s, and then it climbs very gradually through the 1960san d 1970s. My pro ce du re , following Perry's (1970) inno vatio n, was to use a de-mo graphic adjustm ent to the unem ploy me nt rate to reflect the r ising share of teen-agers and females in the labor force during that era. However, when I tested in thelate 1980s to see whe ther th e dem ogra phic c hange s of the 1980s (notably a redu cedshare of teenagers in the labor force) h ad re duc ed t he NAIRU accordingly, I foundthat it ha d n ot (Gord on, 1990a). W ithout any justif ication oth er th an its empiricalperformance, I arbitrarily set the textbook NAIRU equal to 6.0 percent for theentire period after 1978. Th e NAIRU series that com bines the dem ogra phic ad-just m ent thro ugh 1978 with an assumption tha t the NAIRU is consta nt at 6.0 per-cent thereafter is henc eforth called the text boo k NAIRU series.

8 Gordon and K ing (1982, Tab le 5) com put ed a sacrifice ratio of 6.2 from th eir econ om etric version ofthe triangle model. Using the data available at the time, the cumulative deviation of actual from potentialoutput d ur ing the per iod 198087 was 26.2 percen t , and inf lat ion was reduc ed by 4.1 percentag e pointsfrom 19791980 to 198586, for an actual sacrifice ratio of 6.4.9 Gord on (1982a) shows that th e new classical policy ineffectiveness pro positio n ca n be n este d in ageneral mod el of price adjustment and can b e rejected in the p resence of inflat ion iner t ia.


10/23


The BasicFrameworkThe estimation of the time varying NAIRU combines the above inflation equa

tion, with the unemployment gap servingas the proxy for excessdemand, with asecond equation that explicitly allowsthe NAIRU tovarywith time:

t = a(L ) t 1 + b(L)(Ut UNt) + c(L)zt +et,Ut =Ut 1 et.

In this formulation, the error termetin the second equation iswellbehaved, witha mean of zero and a standard deviation of #t. When this standard deviation#e =0, then the natural rate is constant. When the standard deviation#e is positive,thenthe modelallowsthe NAIRU tovaryby a limited amount each quarter. If nolimitwereplaced on the ability of the NAIRU tovaryeach time period, then thetime varying NAIRUwould jump up and down and soak up all the residual variationinthe inflation equation. This model is a standard ''stochastic time varying parameter regression model that can be estimated using maximum likelihood m ethodsdescribed by Hamilton (1994). Th e methodology was previously applied to theissueof the NAIRU, using a different specification of the inflation equation , by King,Stock and Watson (1995) and Staiger, Stock and Watson (1996).

Thefollowingare the key elements of the inflation equation. The sample period is 1955:2 1996:2,or 165 quarters. All right hand side variablesare allowed toenter with lags.10 Supply shock variables include changes in the relative price ofimports and the change in the relative price of food and energy.

11 Dummyvariablesare included for the o n and off effects of the Nixon price controls during

1971 75. These dummy variables, and indeed all the other variables, are definedexactly the same as in all my papers starting with G ordon (1982b).Alsoincludedas an explanatory variable is the difference between productivity growth and itstrend, reflecting the fact that,while most of any cyclical increase or decrease in

10 Lag lengths are chosen to be identical to those in G o r d o n (1990a). The only smoot hing cond itionimposed on th e lag distribution s involves th e lagged de pe n de n t variable, wher e 24 lagged term s en ter .R a t h e r t h a n estimating 24 uncon straine d coefficients, the lagged de pe nd en t variable is ent ere d as aseries of four quarter moving averages of rates of change; for example, the first variable is a four quarteraverage oflags t 1 to t 4, the next t 5 through t 9, an d so on . Th e coefficients on th e individualmoving averages are unc on stra ine d. Exclusion tests indic atet h a t the moving averages representinglags13 through 24 enter with a significance level of better t h a n 1 perce nt in th e three equation s displayedin Table 1 and are thus highly significant. The coefficients on lags13 through 24 represent 21 percentof the t otal lagged effect in t he equ atio n for th e GD P deflator, 34 perc en t of the tot al effect for th e PCEdeflator, an d 25 perc en t of the total effect for CPI U X1.11 T h e food en ergy effect is defined as the difference of the rat e of ch an ge of the chain weighted consumption deflator m inus the rate of change of the chain weighted consum ption deflator n et of food a ndenergy. Chain weighted deflators are available back to 1959 and ar e linked to t he implicit deflator prio rto1959. An ad dit ion al supply shock variable, t he ch an ge in sensitive raw mat erials prices, BCD series 99,was tested and foun d t o be insignificant, with a t ratio below 1. SeeG o r do n (1994, foot no te 7). Also, thech an ge in the real effective exch an ge ra te ,included in previous papers, was found to be insignificant inall versions estimated for this paper and therefore is excluded in the results presented here.


11/23

RobertJ.Gordon 21

Figure1TV NAIRUfor GDPDeflatorwith AlternativeStandardDeviations

productivity is reflected in a movement in profits in the same direction, a smallfraction remains to influence the inflation rate in the opposite direction. 12

Icalculate th ree alternative NAIRU series using three alternative price indexes.O n e is the chain weighted GDP deflator, the basic deflator con cept in the N ationalI n c o m eand Product Accoun ts since early 1996. A secon d is th e chain weighted deflatorfor personal consumption expenditures ( PCE) . The third is the inflation rate of theC o n s u m e rP rice Index con cept called C P I U X1 . 13All three series exhibit the samebasic cycles of acceleration an d decelerat ion in the inflation rat e. However, some differences exist.For example, in th e supply shock episodes of 1974 75an d 1979 1981,th eCPI inflation rate accelerates earliest an driseshighest. On average, con sumer priceinflation was more rapid t han GDP inflation from 1987 to 1994.

The SmoothnessProblemAs indicated above, an assumption must be made about thesizeof the standard

deviation of the error term in the equation for the NAIRU (#e), and this choicewilld e t e r m i n e how much t he NAIRU is allowed to move from quar ter to quar ter. Anassumption of #

e = 0 implies a completely con stant NAIRU series of 6.0 percen t, as

shown by the dotted horizontal line in Figure 1. At the other extreme, an assumptionof#e= 0.4allowsthe NAIRU to be highly variable, as shown by the line with the longdashes in Figure 1. In between is a series drawn as a solid line based on an assumed

12The productivity deviation variable was first intr odu ced in exactly the same form in Go rdo n (1970).T h eproductivity deviation is defined as the growth rate of the log ratio of actual nonfarm private outputper hour to a loglinear piecewise trend running through 1950:Q2, 1954:Q4, 1963:Q3, 1972:Q2, 1978:Q3,1987:Q3 and 1994:Q3. The 1987 1994growth rat e of this tren d is 1.07 perc ent p er an nu m.13 T h eCPI U X1 is th e same as the CPI for ur ban con sumers, called C P I U , except that it extends thepost 1983 treatm ent of the CPI shelter com po ne nt (based o n ren tal equivalence) back to 1967 and thuseliminates the pre 1983 erro r in th e treatm ent of housing in the con ventional CPI that leads to a substantial exaggeration of the inflation rate , particularly du rin g 1977 1981.


12/23

Journal ofEconomicPerspectives

Figure2Textbook NAIRU and TV NAIRU for GDP Deflator standard deviation = 0.20)

standard deviation of 0.2. Which of these (or other) possible assumptions about thestandard deviation should we make?14The most sensible standard deviation may notbe the same foreveryvariable or topic. If the NAIRU isviewed,to paraphrase Friedman,as groun d out by the microeconomic structure and behavior of the economy, thenit should shift slowly.This is especially true since the concept of the NAIRU beingestimated here is the unemployment rate consistent with steady inflationin theabsenceofsupplyshocks.F rom thisview,thezig zagsin the series assuming a standard deviationof 0.4 appear implausible; why should the no supply shocksNAIRUjump up and downfrom quarter to quarter? In essence, I propose using a smoothness prior: the NAIRUcanmove around as much as it likes, subject to the qualification that sharp quarter toquarterzig zagsare ruled out.

As shown in Figure 1, a standard deviation (#e) of 0.2 accomplishes this result,allowing a NAIRU series that exhibits substantial movements butjustavoids sharpquarter to quarterzig zags.It declines from 6.0 percent in the mid 1950s to a minimumof 5.3 percent around 1962,risesto a plateau of about 6.2 percent between1967 and 1972, declines briefly between 1972 and 1975, then exhibits a hum pof about 6.5 percent between 1978 and 1982, and then drifts down gradually to5.6 percent by mid 1996.

Figure 2 compares our preferred time varying NAIRU series based on a standard deviation of 0.2 with the textbook NAIRU series described above. The newseries is substantially higher than the textbook series until the last three years, indicating that prior to 1993 the textbook series provided too optimistic aviewof theeconomy's ability to maintain agiven unemployment rate without suffering th econsequence of accelerating inflation.

14This problem is ana logous to the choice of a smoot hn ess para me ter for t he H odrick Prescott filter sooften used to detrend time series variables.


13/23

Th eTime VaryingNAIRU and its Implications forEconomic Policy 23

Figure 3Unem ploym ent Gaps Tex tboo k NAIRU and Alternative TV-NAIRUs

Figure 3 compares the unemployment gaps implied by the textbook NAIRUseries in comparison with the various time-varying NAIRU series. We see that allthe time-varying NAIRU series indicate substantially m ore excess dem and than thetextbook series in 195557, 19651970 and 19791980. For 199596, the time-varying NAIRU series corresponding to a stand ard deviation of 0.2 indicates some-what less excess demand than the textbook series. I have previously argued that thebehavior of inflation in the 198889 expansion and 199091 recession period wasconsistent with the textbook NAIRU assumption of 6.0 percent (Gordon, 1994).But Figures 2 and 3 show that the textbook NAIRU performs well after 1987simplybecauseduring that interval it ha ppens to be quitecloseto thetime varyingNAIRU.

Staiger, Stock and Watson (1996, p. 2; this issue) have cast doubt on the en-terprise of estimating the NAIRU, concluding that a typical 95% confidence in-terval for the NAIRU in 1990 is 5.1 percent to 7.7 percent. . . . This imprecisionsuggests caution in using the NAIRU to guide monetary policy. It is true that th edifferent unem ploym ent gap series displayed in Figure 3 look almost the sam e, arevery highly correlated and result in inflation equations that fit about as well as eachother. By standard statistical criteria, they cannot be distinguished from the other.However, the smoothness criterion proposed above is a way to cut through someof this ambiguity by using an econom ic ra the r than a statistical criterion to choosebetween alternative NAIRU series.

Staiger, Stock and Watson (this issue) argue that there are two sources of un-certainty in estimates of the NAIRU: uncertainty over the proper model (for ex-ample, the specification and the smoothness parameter) and then, given the propermodel to estimate, uncertainty about the estimated parameters in the inflationequation. I consider only one pro pe r mod el, one that has performed with re-markable reliability over the past 15 years, and ignore the issue of parameter un-certainty on th e grounds th at the parame ters in this model have remained relativelystable over many years during which new data have accumulated.


14/23


Figure 4TV-NAIRUs for Alternative Price Indexes(standard deviation= 0.20)

Estimated Equations and ImplicationsTo build some intuition about the results of regressions like these, considersome explicit findings presented in Figure 4 and in Table 1.Figure 4 shows three estimates of the time-varying NAIRU, all assuming a stan-dard deviation of 0.2 but based on three different price indexes: the GDP deflator,the PCE deflator and the CPI-U-X1. Th e time-varying NAIRU series for the PCEdeflator and CPI-U-X1 are quite close to each other prior to 1980; the CPI-U-X1series for the NAIRU is lower from 1980 to 1990 and higher after 1990. By mid-1996 a substantial gap had opened up between the NAIRU for CPI-U-X1 (5.8 per-cent) and for th e PCE deflator (5.4 perce nt), with the NAIRU for the GDP deflatorin between (5.6 perc en t). Prior to 1980, the NAIRU for the GDP deflator was gen-erally lower than that for the two consumption price indexes , by as much as half apercentage point in the mid-1970s.Table 1 presents the results of regressions that use the same three price indexes:the GDP and PCE deflators, and the CPI-U-X1. Estimated sums of coefficients on theinflation inertia variable are very close to unity, while those on the unemployment gapare always highly significant and of the correct sign.15The significance of the varioussupply variables differs, but with two exceptions of insignificant coefficients, they allhave the correct sign. A one percentage point excess of productivity growth abovetrend reduces inflation by somewhat less than 0.1 percent A one percentage pointincrease in the relative price of imports raises domestic GDP inflation by 0.09 percent,not far from the average share of imports in GDP during the sample period. About

15No con stant is includ ed. This is an essential elemen t of the appro ach if the d em and variable is definedas a deviation from the NAIRU, that is , the une mp loym ent g ap U t U Nt.


15/23

RobertJ.Gordon 25

Table 1Estimated Equations for Quarterly Change in Alternative Deflators1955:21996:2

one-quarter of the food-energy relative price effect feeds through to inflation in theGDP deflator, two-thirds for the PCE deflator and more than 90 percent for CPI-U-X1.The Nixon on and off dummy variables continue to be essential elements inexplaining the dynamics of price behavior during the 197175 period, although the off variable is small and insignificant in the PCE deflator equation. Taken as a group,the inclusion of the supply side variables makes a substantial difference, especiallyduring the 19731981 period, which is influenced by adverse supply shocks. The es-timates of the time-varying NAIRU would be much higher during those years if thecontribution of the supply shock variables to inflation were to be ignored. This is shownin Figure 5, where the dashed line indicates the alternative NAIRU series that wouldbe estimated if all supply shock variables (including the Nixon control variables) wereexcluded from the estimation. As shown there, it would have taken a much higherunemployment rate of about 7 percent during the last half of the 1970s to avoid anacceleration of inflation, whereas in the 1980s the alternative NAIRU would have beenlower, reflecting the benign influence of falling real oil and im port prices in pushingdown the rate of inflation.

One way of exploring the stability and accuracy of results like these is to usethe equations as the basis for dynamic simulations of particular historical periods.This process begins by estimating an alternative set of coefficients for a sampleperiod beginning as before in 1955:2 but truncated in the third quarter of 1987,and then using these coefficients to form an estimate of inflation for the fourthquarter of 1987. Then, that estimate for 1987:4 is used in turn as a basis for fore-casting the following q uarte r, and so on for the following decade, feeding back theestimatedvaluesof lagged inflation after 1987:3 rath er than the actual values. Overthe time period from 19871996, the root-mean-squared erro r of the simulation in


16/23


Figure 5GDP Deflator-Based TV-NAIRUs With and Without Supply Shocks

forecasting the rate of inflation is 0.7 at an annual rate, actually smaller than thestandard erro r of estimate of 0.9 pe rcent (this presumably reflects the smaller vari-ance of inflation during the simulation period than du ring the sample period ). Themean error is about 0.25 percen t at an ann ual rate, m eaning tha t the actual inflationrate is on average one-quarter of a percentage point above the simulated inflationrate during 19871996. But in the last two years of the simulation (199496), themean error is only 0.07 percent, indicating no drift in the simulated inflationrate away from the actual inflation rate over the near-decade duration of the sim-ulation. The fact that it is possible to estimate inflation in 199496 based on pre-1987 data illustrates how stable the structure of the inflation process seems to havebeen during 19871996 from the perspective of the pre-1987 period.

Recently, both Eisner (1996) andAkerlof,Dickens and Perry (1996) have sug-gested that the linear specification of the inflation equation is incorrect. Eisnerargues that the Phillips curve is concave, that is, flatter when the unemploymentrate is below the conventional NAIRU and steeper when the unemployment rate isabove the conventional NAIRU.Akerlof, Dickens and Perry (1996) argue for theopposite nonlinearity, a convex Phillips curve that becomes much flatter when in-flation is low and unemployment is above the conventional NAIRU. I have testedeach possibility by allowing the coefficients on the unemployment gap to be differ-ent at low vs. high unem ploym ent rates, or at low vs. high inflation rates. None ofthese differences is statistically significant, indicating that the short-run Phillipscurve is resolutely linear, at least within the range of inflation and unemploymentvalues observed over the 19551996 period.Another set of concerns relates to the particular sample period chosen. For ex-ample, why should the Fed base its estimate of the current NAIRU on more than40 years of previous data? Why is not the more recen t past, say the last 20 years, a morerelevant interval for which to estimate the inflation equation? Splitting the sampleperiod at the first quarter of 1975 results in a sharp jump in the estimated NAIRU at


17/23


Figure 6Future Simulations Four-Quarter Moving Averages of Inflation Rate

the break point; the NAIRU for 19551974 is between 0.1 and 0.5 percentage pointslower than the full-sample estimate, and the NAIRU for 19751996 ranges between0.0 and 0.3 percentage points higher than the full-sample estimate. However, for thepurposes of conducting current monetary policy, it is reassuring tha t the NAIRU esti-mate for the second quarter of 1996 is identical in the full-sample and split-samplealternatives. It is also true that a dynamic simulation for 19871996 based only on thedata since 1975 predicts more accurately than the full-sample results in Table 1.How rapidly would inflation accelerate if the Fed, eithe r by accident or design,allowed unemploym ent to fall on e pe rcentage poin t below the NAIRU? To simplify,assume that n o supply shocks occur. F igure 6 displays the results of two simulations.The first, which yields steady inflation at a rate of 2.2 percent, is based on theassumption that the unemployment gap remains at zero forever, beginning in1996:3.Th e alternative simulation allows the gap to decline to 1.0 perce nt in thefive quarters beginn ing with 1996:3, and to remain a t 1.0 perc en t forever; this

generates a slow acceleration of inflation that starts immediately and reaches5.3 percent by the year 2005. The most no table aspect of this result is the slownessof the acceleration; after the year 2000 inflation accelerates by only 0.32 percentpe r year. The slow pace of this acceleration reflects the role of the lags in the effectsof both inflation and the unemployment gap on the current rate of inflation. 16

16Reade rs of earlier drafts questio ned th is result and a ssumed th at in the long run the rate of inflationshould accelerate by the sum of coefficients on the unemployment rate, for example, 0.61 for the GDPdeflator. H owever, a bit of simple algebra shows that this assum ption is correct o nly if the lagged inflationrate enters with a single annual lag. If there are two annual lags with equal weights, the long-run accel-eration is the 0.61 coefficient divided by 1.5. With three annual lags, the divisor is 2.0. The simulatedacceleratio n is an accu rate reflection of the estimate d coefficients on lagged inflation, which stretch o utover six years.


18/23

28 Journal o f Econom ic Perspectives

ConclusionThe inflat ion process in the United States is one of the most important mac-

roeconomic phenomena in the world, but i t is also one of the best understood. Incontrast to the gyrations of inflation in many other countries, the U.S. inflationprocess is dominated by inertia. Inflation changes little from year to year, and anydeviat ion of the actual unemployment rate from the NAIRU has only small conse-quences in the short run. The best recent example is the 19881990 period, whenunem ploym ent was on average abou t one percen tage po int below the 6 .2 percen testimated NAIRU, and the GDP deflator accelerated over the three years 19871990 from 3.1 to 4.4 percent. This implies a response of inflation of a bit less thanhal f a point per one percentage point tha t unemployment remains below theNAIRU for a single year. This is very similar to the pace of inflation's response tothe 1 perc ent une mp loym ent gap displayed in Figure 6.

Because the U.S. inflation process ha s been so stable, and is so well chara cterizedby the triangle model of inflation developed in the late 1970s and early 1980s, thatmo del has performed extremely well in dynamic postsample simulations extend ing outfor up to a decade after the en d of the sample period. In such simulations, the m odelhas proven capable of tracking the disinflation of the early and mid-1980s, the accel-eration of inflation of the late 1980s, and the subsequent deceleration of inflation inthe 1990s. Those empirical successes were achieved despite the fact that in previousresearch, the NAIRU inserted into the model was assumed arbitrarily to be constantat 6.0 percent for the entire period after 1978 rather than estimated econometrically.The reason that my previous assumption that the NAIRU was fixed at 6.0 percentperform ed so well in tracking the acceleration and deceleration of that period seemsto be that th e estimated time-varying NAIRU was fairly close to 6.0 per cen t du rin g theexpansion of the late 1980s and recession of 199091.

What would it take to reject the hypothesis that there is a NAIRU and thereforea vertical long-run Phillips curve? Formally, an inflation equation with a sum ofcoef-ficients on lagged inflation of unity and a significant negative sum of coefficients onthe unemployment gap validate the NAIRU concept. Less formally, wild gyrations ofthe estimated NAIRU over a range too wide to be explained by microeconom ic change sin market structure and institutions would lead to skepticism about the NAIRU con-cept. Within the postwar experience of the United States, the modest fluctuations inthe NAIRU seem plausible in magnitude and timing. When applied to Europe or tothe United States in the Great Depression, however, fluctuations in the NAIRU seemtoo large to be plausible and seem mainly to mimic movements in the actual unem-ployment rate. This paper is about the postwar United States, for which the NAIRUhypothesis works very well, and simply leaves open for further research the deeperreasons why the hypothesis does not seem to characterize the U.S. Great Depressionor the recent years of high unemployment in many European countries.

This paper rejects the recent argument that the band of stat ist ical uncertaintysurr ou nd ing the NAIRU is so bro ad as to ren de r the con cep t useless for the c on duc tof policy. We propose an economic cri terion based on smoothness, rather than a


19/23

RobertJ. Gordon 29

statistical criterion, to choose among alternative NAIRU estimates for any givenmeasure of inflation. The recent suggestion of Staiger, Stock and Watson (1996)that the NAIRU for the year 1990 could range from 5.1 to 7.7 percent makes noeconomic sense. If the NAIRU had been 5.1 percent since 1987, inflation wouldnot have accelerated during 19871990, since the actual unemployment rate neverfell below 5.1 percent in any calendar quarter. If the NAIRU had been 7.7 percentin the period since 1987, inflation would not have decelerated during 199093,since the actual unemployment rate never rose above 7.7 percent in any calendarquarter. The fact that the inflation rate for the GDP deflator was roughly constantduring the six quarters from the end of 1994 to the beginning of 1996, when theactual unemployment rate was approximately constant at 5.6 percent, suggests thatthe time-varying NAIRU for the GDP deflator during those six quarters was veryclose to 5.6 percent (our point estimate for 1996:2).

What evaluation of past and current monetary policy is implied by this newresearch on the NAIRU? According to our new time-varying NAIRU m easures, therewas considerably m ore excess dem and in 195557 and 19651970 than implied bythe previous textbook NAIRU series (see Figure 3 ), suggesting that monetary policywas even more overly expansionary in those periods than was previously thought.Th e new time-va rying NAIRU series also boosts modestly the extent of estimatedexcess dem and in 19791980 and 19881990. However, the new series implies tha tmonetary policy in 199596 has been almost precisely on target, with an averageunemployment rate during the seven quarters 1994:41996:2 of 5.6 percent, onlyslightly below the average estimated time-varying NAIRU of 5.7 perc ent for the GDPdeflator in that interval. If the Fed considers its goal as the stabilization of the rateof change of the PCE deflator rather than the GDP deflator, then the estimatedaverage time-varying NAIRU during the same interval was 5.5 percent, implyingthat m onetary policy was just slightly too restrictive in 199596.

The time-va rying NAIRU by any measure has declined in the 1990s. This raisesan issue of whether the Fed can run an easier monetary policy to encourage a fasterexpansion, with little fear of triggering inflation. The newime-varyingNAIRU seriesallows a new series for potential output to be created, measuring the real GDP thatcan be produced each quarter when the economy is operating at the time-varyingNAIRU. Since the time-varying NAIRU declined from about 6.2 percent in 1990 to5.6 percent in mid-1996, this potential output series grows at about 0.1 percentagepoint per year faster (that is, 2.1 percent per annum rather than 2.0 percent) thanwould be implied by a NAIRU fixed at 6.0 percent.17Thus, caution is advised regarding

17As sup por t for this statem ent, I have estimated an O ku n's Law equatio n similar to that in no te 4. Indoing so, I invert the equation to regress the unknown output gap on current and leading values of theunemployment gap implied by the new TV-NAIRU series based on the GDP deflator and a standarddeviation of 0.2.1 allow the sum of coefficients on the unemployment gap to differ during 1990-96 fromtheir values in 19721990. I calculate the fitted output gap, and then compute a trial value of potentialoutput as actual real GDP minus the fitted output gap. The final potential output series is a 12-quartercentered moving average of the trial series.


20/23


the advice of growth hawks that the U.S. econom y could grow at 3 perc ent or m oreper annum if only the Fed's monetary policy were less restrictive. The decline in theNAIRU is not nearly enough to allow potential real output growth to take off from itspresent level of about 2 percent to 3 percent or higher. Moreover, the decline inNAIRU is a one-time event; it can make room for the economy to have faster nonin-flationary grow th for a few years, bu t no t in perpetuity.

If the Fed's current deliberat ions about interest rate changes are intended toinfluence the actual unemployment rate roughly one year from now, should theFed extrapolate the rec ent decline in the NAIRU into the future, or should the Fedset its estimate of the NAIRU one year hence equal to the current value? The time-varying NAIRU derived h ere is a ran do m walk an d thu s is jus t as likely to in creaseover the sub seque nt year as to contin ue to decrease. Th ere is no information abo utfuture inflation available beyond that contained in the lagged values of the explan-atory variables in the inflation equation that are already used to derive the NAIRU.

While the time-varying NAIRU technique does not provide a magic crystal ballthat allows the Fed to see into the future, it makes two valuable contributions tothe conduct of monetary policy. First, it quantifies in a systematic way the Fed'sbelief that the NAIRU must have fallen in the 1990s, because as of mid-1996, infla-tion ha d no t accelerated as it did in 1988 1990. Second, it highlights the differencesin the time-varying NAIRU series imp lied by alternative inflation inde xes an d forcesthe Fed to take a stand on what inflation concept it is trying to stabilize.

Estimated movements in the NAIRU over time naturally raise the question as towhich factors caused these m ovem ents. The two especially large changes in the N AIRU,as shown in Figure 4 for all three alternative price indexes, are the increase betweenthe early and late 1960s and the decrease in the 1990s. The late 1960s were a time oflabor m ilitancy, relatively strong un ions , a relatively high m inim um wage and a m arke dincrease in labor's share in national income. The 1990s have been a time of laborpeace, relatively weak unions, a relatively low minimum wage and a slight decline inlabor's income share. Oth er factors that may have contrib uted to weaker de m an d andincreased supply in product and labor markets include global competition and im-migration of unskilled labor. A final factor that is beginning to play a significant rolein lowering the NAIRU is the growing share of computer output in both GDP andpersonal consumption;18 by mid-1996 the rapidly declining prices of computers weresubtracting about half a percentage point from the growth rate of the deflators androughly 0.4 percent from the time-varying NAIRUs estimated in this paper.

18In th e year ending in 1996:2, the implicit deflator for business purchases of com puters declined23 perce nt and for consum er purchases of computers declined 35 percen t. If one subtracts nom inal andreal com pute r outp ut from bo th nom inal and real GDP, one can calculate an alternative implicit deflatorfor the non com put er p art of the economy. In the four quarters end ing in 1996:2, this rose 0.7 p ercentag epoints faster than the implicit deflator for all of GDP including computers. Using this alternative ex-computer implicit GDP deflator results in a NAIRU for 1996:2 of 6.0 percent rather than the 5.6 percentestimated in this paper . The same technique adds about 0.4 percent to the NAIRU estimated for thePCE deflator. Com pute rs make n o difference to the CPI-U-X1, which still uses obsolete 198284 expe n-diture w eights for cons ume r purch ases of compu ters and thus gives them close to a zero weight.


21/23


Thisresearch is supported by the National ScienceFoundation. I am grateful to TomonoriIshikawa for research assistance, to Mark Watson for helpful discussions, and to BradDe Long, Alan Krueger and TimothyTaylorfor their many invaluable com ments and sug-gestionsfor focussing the content of the first draft and making itmorereadable.

References kerlof George A. William T. Dickens andGeorge L. Perry The Macroeconomics of Low

Inflation, Brookings Papers on Economic Activity,1996, 27 :1, 126.

Barro Robert J. Unanticipated Money, Out-put, and the Price Level in the United States,Journal ofPoliticalEconomy , August 1978,67 , 10115 .

Barro Robert J. and Hers chel I. Grossma nMoney,Employment and Inflation . Cambridg e, UK:Cambridge University Press, 1976.

Eisner Robert A New View of the NAIRU,manuscript, Northwestern University, Ju ne 1996;in Davidson, P., andJ.Kregel, eds., Improving theGlobalEconomy:Keynesianismand the Growthin Out-put and Employment. Ch eltin ham , U.K., andBrookfield: Edward Algar, forthcoming, October1997; Cahiers de l Espace Euro pe , for thcoming,March 1997.

Fisher Irving A Statistical Relation BetweenUnemployment and Pr ice Changes , Interna-tional Labor Review , Ju ne 1926, 13 , 78592; re-printed in Journal of Political Economy , M a r c h /April 1973,81 , 496502.

Franz Wolfgang and Robert J. GordonWage and Price Dynamics in Germany andAmerica: Differences and Common Themes,

European Economic Review , May 1993, 37 , 71954.Friedman Milton The Role of Monetary Pol-

icy, American E conomic Review , March 1968, 58 ,117.

Gordon Robert J. The Recent Accelerationof Inflation and its Lessons for the Future,Brookings Paperson EconomicActivity,1970,1 :1, 841.

Gordon Robert J. The Impact of AggregateDemand on Prices, Brookings Paperson EconomicActivity, 1975, 6 :3, 61362.

Gordon Robert J. Can the Inflation of the1970s Be Explained?, Brookings Papers on Eco-nomic Activity, 1977a, 8: 1, 25377.

Gordon Robert J. The Theory of DomesticInflation, American Economic Review , February1977b, Papers and Proceedings, 67 , 12834.

Gordon Robert J. Price Inertia and PolicyIneffectiveness in the United States, 18901980, Journal of Political Economy , December1982a, 90 , 1087117.

Gordon Robert J. Inflation, FlexibleExchange Rates, and the Natural Rate of Un-em plo ym ent . In Baily, M. N., ed., Workers, Jobs,and Inflation . Washington, D.C.: Brookings Insti-tution, 1982b, pp. 88152.

Gordon Robert J. U.S. Inflation, Labor'sShare, and the Natural Rate of Un emp loym ent.In Knig, Heinz, ed., Economics of W age Determi-nation. Berlin and New York: Springer Verlag,1990a, pp. 134.

Gordon Robert J. What is New-KeynesianEconomics?, Journal ofEconomicLiterature , Sep-tember, 1990b, 28 , 111571.

Gordo n Robert J. Inflation and Unemploy-ment: Where is the NAIRU? Paper presented toBoard of Governors, Fed eral Reserve System, Meet-ing of Academic Consultants, December 1, 1994.

Gordon Robert J. and Stephen R. King T h eOutput Cost of Disinflation in Traditional andVector Autoregressive Models, Brookings Paperson EconomicActivity , 1982, 13 :1, 20542.

Hamilton James D. Time Series Analysis.Princeton: Princeton University Press, 1994.

King Robert G. and Mark W. Watson T h ePost-War U.S. Phillips Curve: A Revisionist Econ-ometr ic History, Carnegie Rochester ConferenceS e-ries on Public Policy, 1994, 41 , 157219.

King Robert G. J ames H. Stock and Mark W.Watson Temporal Instability of the Unemploy-ment-Inflation Relationship, Economic PerspectivesoftheFederal ReserveBank of Chicago , May/June 1995,19 :3, 212.Lucas Robert E. Jr . and Tho ma s J. Sar-gent After Keynesian Eco no met rics . In After


22/23


the Phillips Curve:Persistence of High Inflation andHigh Unemployment,Conference Series 19 . Boston:Federal Reserve Bank of Boston, 1978, pp. 4972 .

Perry George L. Changing Labor Marke tsand Inflation, Brookings Paperson EconomicActiv-ity, 1970, 1 :3, 41141.

Phillips A. W. The Relation Between Un-employment and the Rate of Change of MoneyWage Rates in the United Kingdom, 18611957, Economica , Nov embe r 1958, 25 , 28399.

Rom er Christina D. Inflation and theGrowth Rate of Ou tpu t. NBER Working Pap erN o.5 575, May 1996.

Sam uelson Paul A. and Robert M. SolowAnalytical Aspects of Anti-Inflation Policy,

American Econom ic Review , May 1960, Papers andProceedings, 50 , 17794.

Schultze Charles L. Falling Profits, RisingProfit Margins, and the Full Employment ProfitRa te , Brookings Paperso nEconomicActivity , 1975,6:2, 44969.

Solow Robert M. Price Expectationsan dthePhil-lips Curve. Manchester , UK: Manchester Univer-sity Press, 1969.

Staiger D ouglas Jam es H. Stock and Mark W.Watson How Precise are Estimates of the Nat-ura l Ra te of Unem ploymen t? NBER WorkingPaper No. 5477, March 1996.

Stevenson Richard W. It 's Heresy at Fed ButCritics Say: Step on the Gas, New York Times,June 7, 1996, Chicago ed., 15.


23/23
http://pubs.aeaweb.org/doi/pdfplus/10.1257/jep.11.2.3http://pubs.aeaweb.org/doi/pdf/10.1257/jep.11.2.3http://dx.doi.org/10.1257/jep.11.2.3http://pubs.aeaweb.org/doi/pdfplus/10.1257/jel.37.4.1661http://pubs.aeaweb.org/doi/pdfplus/10.1257/jel.37.4.1661http://pubs.aeaweb.org/doi/pdf/10.1257/jel.37.4.1661http://dx.doi.org/10.1257/jel.37.4.1661http://pubs.aeaweb.org/doi/pdfplus/10.1257/jep.14.4.49http://pubs.aeaweb.org/doi/pdf/10.1257/jep.14.4.49http://dx.doi.org/10.1257/jep.14.4.49http://pubs.aeaweb.org/doi/pdfplus/10.1257/000282806777212251http://pubs.aeaweb.org/doi/pdf/10.1257/000282806777212251http://dx.doi.org/10.1257/000282806777212251http://pubs.aeaweb.org/doi/pdfplus/10.1257/aer.100.2.11http://pubs.aeaweb.org/doi/pdf/10.1257/aer.100.2.11http://dx.doi.org/10.1257/aer.100.2.11

The Time-Varying NAIRU and its Implications for Economic Policy.pdf

Documents

Transcript of The Time-Varying NAIRU and its Implications for Economic Policy.pdf