Structural Change and Economic Dynamics 23 (2012) 170– 179

Contents lists available at SciVerse ScienceDirect

Structural Change and Economic Dynamics

j ourna l ho me pag e: www.elsev ier .com/ locate /sced

The opening of the North–South divide: Cumulative causation,household income disparity and the regional bonus in Taiwan1976–2005

Martin Anderssona,∗, Martin Klinthäll a,b,1

a Department of Economic History, Lund University, Box 7083, S-22007 Lund, Swedenb Department of Management and Engineering, Linköping University, SE-58183 Linköping, Sweden

a r t i c l e i n f o

Article history:Received February 2010Received in revised form September 2011Accepted February 2012Available online 15 February 2012

JEL classification:N00O14O53R11R12

a b s t r a c t

During the last decades regional income divergence seems to have reappeared in bothdeveloped and developing countries. In Taiwan – a renowned case of growth with equity– regional per capita income was converging until the early 1990s after which it beganto diverge. With the help of modeled annual household survey data from 1976 to 2005we indicate the magnitude of a regional bonus and discuss reasons behind the re-openingof the North–South income divide in Taiwan. Our analysis suggests that this process is aconsequence of cumulative causation connected to the advent of the rise of ICT industryin conjunction with changes in Taiwan’s political economy which provided relatively moreadvantageous economic opportunities for the industrial structure of the leading region.

© 2012 Elsevier B.V. All rights reserved.

Keywords:Regional income dynamicsCumulative causation

Spatial inequalityIndustrialisationTaiwan

1. Introduction

Over the last decades, regional income divergence isobserved to be on the rise in both developed and develop-ing countries (e.g. Fagerberg and Verspagen, 1996; Puga,2002; Sánchez-Reaza and Rodríguez-Pose, 2002; Lall andChakravorty, 2005). On the basis of industry-location dataor national accounts, many of these studies suggest that

liberalisation and increasing openness to trade and foreigninvestments has been spurring a tendency of higher spa-tial inequality in developing countries. As for the developed

∗ Corresponding author. Tel.: +46 46 222 44 85; fax: +46 46 222 73 39.E-mail addresses: martin.andersson@ekh.lu.se (M. Andersson),

martin.klinthall@ekh.lu.se, martin.klinthall@liu.se (M. Klinthäll).1 Tel.: +46 73 256 33 88; fax: +46 11 36 30 29.

0954-349X/$ – see front matter © 2012 Elsevier B.V. All rights reserved.doi:10.1016/j.strueco.2012.02.001

countries, however, the reasons behind diverging regionalper capita income remain inconclusive (see Puga, 2002;Kim, 2008).

The purpose of this study is to analyse the underlyingforces of regional income dynamics in Taiwan, a renownedcase of regional balance and “growth with equity”. Overthe last couple of decades a trend of increased incomedisparity has been observed (see for instance Corniaet al., 2005). Inspired by theories of cumulative causation,we investigate the regional dimension of this incomedivergence. We analyse effects of agglomeration onlong-term patterns of household income, dissecting theextent to which income disparity is due to inter-regional

differences in household characteristics, such as regionalconcentrations of human capital, and the extent to whichthey adhere to regional effects not explained by suchcharacteristics, but to a “regional bonus”. Using data from

nge and

aowdcwitdp

trgTisarwii

2cc

recwfiadnfiissdrActSpvsgttTs(tlwi

both Myrdal and Hirschman explicitly acknowledged Mar-shall’s ideas). This basic suggestion later reappeared amongearly proponents of cumulative causation.2 One expositor

M. Andersson, M. Klinthäll / Structural Cha

national survey including thirty annual cross-sectionsf household-level information for the period 1976–2005,e study effects of household characteristics and pathependencies in regional development. The survey dataovers the period from the heydays of Taiwanese growthith equity, based on small and medium sized labour

ntensive enterprises catering the export markets underhe guidance of development policies of KMT, to the ICT-ominated industrial structure of the post-authoritarianost-developmental state.

The paper is organised as follows: after a review ofhe major debates and findings of the last two decadesegarding the ideas behind forces of convergence and diver-ence, we present the approach of cumulative causation.his is followed by a discussion of major structural andnstitutional changes in the Taiwanese economy during thetudied period, after which we present data and model fornalysis of the components of convergence and divergenceegarding inter-regional household income. The paper endsith a summary of our findings and a short discussion of the

mplications of our findings for regional income dynamicsn the industrialisation process.

. Conceptual framework: neo-classicalonvergence, endogenous growth divergence andumulative causation in regional growth dynamics

According to neo-classical economic theory, both inter-egional and international convergence of income isxpected. In the standard neo-classical growth model forlosed economies, international convergence of incomeill take place through diminishing returns to individual

actors of production. In an oft-cited study of US regionalncome, Williamson (1965) argued within a neo-classicalpproach that after an initial tendency towards regionalivergence, convergence naturally follows. Under inter-ational market integration, due to Heckscher–Ohlin-type

actor price equalisation, convergence is also the theoret-cally likely long-term outcome. The lack of convergencen the real world, however, made some of the pioneeringcholars of the long-term growth process to propose rea-ons for the absence of convergence; such as the nationalifferences in the “peculiarities of backwardness” or theelative lack of “social capabilities” (Gerschenkron, 1962;bramovitz, 1986). The search for empirical support foronvergence was, however, only in its infancy and withhe advent of more available and comparable data (e.g.ummers and Heston, 1991; Maddison, 1982, 1995), theossibilities to scrutinise the extent of neo-classical con-ergence in the international economy increased. Someupport for convergence was then found albeit only amongroups of nations (Baumol, 1986; Chatterji, 1992). Takinghe world as a whole, however, the global economy overhe long run was clearly characterised by divergence—Bigime (Pritchett, 1997, see also DeLong, 1988). Influentialtudies by Mankiw et al. (1992) and Barro and Sala-i-Martin1992), however, found support for neo-classical condi-

ional convergence using augmented Solow-models. In thisiterature and the empirical studies that followed, efforts

ere made to make distinctions between different mean-ngs of convergence—so-called �- and ˇ-convergence. The

Economic Dynamics 23 (2012) 170– 179 171

former meant that the variation of GDP per capita across agroup of countries/regions diminishes over time, while thelatter denoted that the growth rate for an initially poorercountry/region is higher compared to an initially wealthierone (see Barro et al., 1991). Thus, �-convergence neces-sarily implies ˇ-convergence whereas beta ˇ-convergencedoes not necessarily imply �-convergence. It was theˇ-convergence that received most attention. The aug-mented Solow-models then showed that structural condi-tions and differences in, e.g. human capital characteristicsproduced different steady-state growth rates. By holdingconstant the steady-state of individual countries, Barro andSala-i-Martin (1992) found support for both cross-countryand regional conditional ˇ-convergence. Everybody was,however, not convinced by the convergence evidence, sug-gesting biases in the statistical models resulting in mislead-ing conclusions (Quah, 1993, 1996; Temple, 1999). Somealso pointed out that overlooking the role of technology– typical of “capital fixated” Solow-models – underminesa thorough understanding of the process of convergence(Bernard and Jones, 1996). A clearer break with the assump-tions of diminishing marginal returns to capital in theneo-classical models was characterised by the adventof endogenous growth theory, by introducing increasingreturns via, e.g. knowledge spillover effects (Romer, 1986).The insights from this theoretical development helpedboost the new economic geography approach and showedtheoretical reasons of regional divergence due to agglom-eration, increasing returns and spillovers in general equi-librium models (see e.g. Krugman, 1991; Fujita et al., 2001).

The new economic geography is, however, reminiscentof the ideas of cumulative causation and the discussionof dynamics of regional income growth of Gunnar Myrdaland Albert O Hirschman. This line of thought, allowingfor both converging and diverging forces in the develop-ment process was prominent in development economicsof the 1950s and 1960s. Spatial inequality was seen asa natural consequence of a country’s dynamic develop-ment, for instance captured in the notion of “nothingsucceeds like success” (see Perroux, 1955; Myrdal, 1957;Hirschman, 1958). What advocates of cumulative causa-tion tried to disentangle was the reasons behind differencesin inter-regional dynamics and income. Since theories ofcumulative causation embrace processes of both regionalconvergence and divergence and, furthermore, includeinstitutional and “non-economic” aspects, the assumptionof an economy in (or towards) equilibrium became sig-nificantly relaxed in analyses of social development (seeMyrdal, 1957).

However, the theory of cumulative causation did notoriginate with Myrdal or Hirschman. Marshall’s externaleconomies coming from the “atmosphere” of industrial dis-tricts is one of its original underpinnings (Marshall, 1890;

2 The idea of cumulative causation has also been central in influen-tial contributions in institutional theory (Veblen, 1898), monetary theory(Wicksell, 1898) and macro economic theory (Kaldor, 1966).

ange and

172 M. Andersson, M. Klinthäll / Structural Ch

of the idea of cumulative causation in terms of increasingreturns was Allyn Young (1928): “Every important advancein the organisation of production, regardless of whetherit is based upon anything which, in a narrow or techni-cal sense, would be called a new “invention,” or involvesa fresh application of the fruits of scientific progress toindustry, alters the conditions of industrial activity andinitiates responses elsewhere in the industrial structurewhich in turn have a further unsettling effect. Thus changebecomes progressive and propagates itself in a cumulativeway.”

Geographic agglomeration, caused by Marshallianexternal economies, resulting in interregional growthimplied that spatial inequality was inevitable but also acondition of growth itself (Hirschman, 1958). For Myrdal,the normal development tendencies of a society was,through the workings of cumulative causation, to stray fur-ther and further away from equilibrium unless measuresare undertaken that counteract continuous polarisation.A stable equilibrium is according to Myrdal thereforeuntypical and unrealistic: cumulative processes are alwayspresent.

As forerunners to the current discussion on agglom-eration and regional income dynamics, Myrdal (1957)and Hirschman (1958) theorised over regional disparitiesthrough backwash effects (Myrdal) of cumulative causa-tion or polarisation effects (Hirschman). While Hirschmanpainted a fairly optimistic picture in that the polarisa-tion effects eventually may be counteracted by tendenciestowards inter-regional convergence, so-called tricklingdown effects, Myrdal’s conceptual equivalent – the spreadeffect – would have a much harder time to materialise,especially in the less advantaged countries.

For Hirschman, structural change in an advancingeconomy in conjunction with adequate economic poli-cies induced by these changes would eventually close thenational North–South per capita income gap. For Myrdal,however, at least in developing countries, the North–Southdivide is likely to become persistent because “nothing failslike failure” (Myrdal, 1956). Regional inequality increasesbecause the returns of interdependent economic activi-ties such as demand, income, investment and productionin an expansive region are above the national average.The external economies tend to be concentrated in cer-tain regions. Such regions thus attract the most able labourmigrants from the backward areas, which not only loosesproductive manpower but also might become continuouslydisadvantaged by a skewed age structure. In the backwardregions, less demand cause less investment and aggregateincome drops causing demand to slow down even more.Trade opportunities and market expansion thus increasesin one area but slows down in other. For Myrdal, externaleconomies in a progressive region were not the only forcethat increased regional inequality but also that the successof one region could come at the expense of the backwardones. Both these forces, regional success and failure, weredue to cumulative causation in which tendencies of social

development of the past naturally increase in strength overtime, especially in a situation where the market mecha-nisms are given free reign. But with proper governmentinitiatives, such as extension of infrastructure, health care,

Economic Dynamics 23 (2012) 170– 179

sanitation, schools, etc., regional imbalance can be coun-teracted.

Less advantaged regions might also begin to enjoy exter-nal economies themselves if such regions are endowedwith natural resources that come into productive useand attracts a large enough labour force which in turnstimulates growth of consumption industry and eventu-ally income. The trickling down/spread effects includesincreasing willingness for the North to invest in the Southbut also external diseconomies of agglomeration in theNorth may slow down the process of the increasing returnsin a leading region. Here the cumulative causation ofincreasing returns is weakened in the leading area due toincreasing wages, more of public spending due to conges-tion or that a leading industrial area too long continueto use evermore obsolete technology without sufficientupgrading.

Cumulative causation thus suggests that tendencies ofspatial convergence and divergence are always present.Processes of convergence or divergence are shifting overtime. Taiwan is a renowned case of growth with equity,where tendencies of increasing income divergence havebeen observed lately. In the following we will analyse theseprocesses of cumulative causation during 30 years of eco-nomic development in Taiwan.

3. Regional development and inequality inTaiwan—from growth with equity to theNorth–South divide

Many have observed the increasing inequality in Taiwanof the recent years and some have expressed concern overwhat appears to be the emergence of the so-called M-society, where the middle class is in relative decline whilethe upper and most prominently the lower middle classesare growing in relative numbers (Ohmae, 2006; China Post,2008). Others have noted the increasing regional divide(Hsu, 2009; Taipei Times, 2011). Such concerns constitute anew phenomenon in modern Taiwan which for decades hasbeen an oft-cited case of growth with equity. Despite per-vasive structural change and long periods of rapid growth,Taiwan progressed from the 1950s to the early 1980s with alargely constant, or a slightly equalising, pattern in the dis-tribution of personal income. One key reason why Taiwanmaintained a fairly equal distribution of income is heldto be the rent reduction schemes and land-to-the-tillerreform of the 1949–1953-period that paved the way forsubsequent egalitarian growth path (see e.g. Cheng, 1961;Fei et al., 1979). As a result, significant rural–urban andregional income gaps could be avoided (Fei et al., 1979).Hence the industrialisation process was characterised bystrong trickling-down/spread effects. It is a well observedfact that much of the Taiwanese industrialisation effortswere centred on the voluminous presence of rurally situ-ated small and medium scale enterprises (Ho, 1979; Parkand Johnston, 1995). Thus, the existence of rural industryin Taiwan facilitated an agriculture-to-industry transfer of

labour without inducing corresponding locational shifts.In the words of Fei et al. (1979: 63–64) “Throughout the1951–71 period the distribution of establishments did notbecome concentrated in any one area, especially in such

M. Andersson, M. Klinthäll / Structural Change and Economic Dynamics 23 (2012) 170– 179 173

Table 1Agriculture as percentage share of total regional GDP.

1976 1981 1986 1991 1994

Northern 4.3 1.9 1.5 0.8 0.8Central 23.8 18.1 9.6 7.1 6.9Southern 17.3 12.6 10.4 7.7 7.4

S

lwTfrgiAetc

orcgowpietr

cAratrfkeSaSboiCsmnaut

ccN(

Table 2Industry as percentage share of regional GDP.

1976 1981 1986 1991 1994

Northern 38.7 37.3 39.8 36.2 32.6Central 48.2 54.1 58.4 48.6 44.7Southern 50.0 61.5 55.1 47.5 43.8

Eastern 26.8 29.7 17.7 10.0 9.3

ource: Wang (1996).

arge urban centers as Taipei City and Kaohsiung City,here heavy concentrations might have been expected.

his spatially dispersed growth pattern enabled farmamilies almost anywhere in Taiwan to move easily intoural industries that were intensive in labor.” The annualrowth rate of employment in manufacturing was, accord-ng to Ho (1979) higher in rural compared to urban areas.

second equalising factor was that non-agricultural wagemployment in rural industry was more important for theotal income of relatively poorer agricultural householdsompared to richer ones (Fei et al., 1979).

Many small and medium sized enterprises (SMEs) wereffspring of agricultural activities and initially in closeelation to demand on domestic markets for not only agri-ultural inputs but also local services and simple consumeroods (Park and Johnston, 1995). The equal distributionf land contributed to a geographical spread of SMEs andith purchasing power increasing, thanks to agriculturalroductivity growth and increasing income opportunities

n the rural industries, the SME sector could continue toxpand in a dynamic and virtuous circle. Table 1 showshe decrease of the relative importance of agriculture toegional GDP in the four regions of Taiwan.

A large portion of the SMEs were so-called ‘guerrillaapitalists’ (see Lam and Lee, 1992; Lam and Clark, 1994).s such, local capitalist SME-entrepreneurs evaded KMTestrictions and regulations by keeping distance to thereas where the party-state bureaucracy had control, i.e.he big cities and in particular Taipei in the North. It seemseasonable to argue that the guerrilla capitalists where bothorced into extralegal activities by the KMT and deliberatelyeeping Taipei and other cities at a distance. The costs ofntry of such firms were low and flexibility high (Mao andchive, 1995). Many firms could enter manufacturing, suchs textiles, with relatively little capital (Lam and Lee, 1992).mall firms, perhaps due to the disadvantages experiencedy belonging to the semi-informal economy, used networksf subcontracting which increased flexibility while allow-ng for numerous under-capitalised firms to exist (Lam andlark, 1994). While being relatively independent from thetate, small firms thus circumvented obligations and per-its and other red tape of the strictly legal businesses. Rural

on-agricultural employment opportunities thus hindered devastating per capita income gap between rural andrban localities and as an extension also provided forceshat tended to increase regional convergence.

By the mid-1980s things started to change. Structuralhange in Taiwan took a marked spurt in the wake of the

omparative cost shift following the appreciation of theew Taiwan Dollar (NTD) after the Plaza Accord 1985

Andersson and Gunnarsson, 2003). The Plaza Accord,

Eastern 38.0 30.9 30.9 28.7 26.9

Source: Wang (1996).

where the G5-economies successfully managed to sign anagreement for currency reform, was aiming at depreciatingthe US dollar in order to tackle the American trade deficit,in particular with East Asia. As a consequence, the CentralBank of Taiwan abandoned its quasi-peg vis-à-vis the USdollar after which the NTD followed the Japanese Yenand appreciated from 40 NTD to the US dollar in 1985to 26 in 1989. Faced with the competitive pressure frommanufacturers elsewhere in Asia, the appreciation drove upprices of Taiwanese exports, striking the industry that fordecades had formed the centre of gravity in the Taiwaneseeconomy—the rural small and medium scaled export-oriented manufacturing enterprises. In the late 1980sto early 1990s, most of these firms closed down, movedabroad or restructured. New and old entrepreneurs nowhad to enter new industrial and service activities that weremore intensive in the use of technology and knowledge.

Fig. 1 shows how earnings of employees in traditionalindustry (textiles) start to fall behind earnings in modernmanufacturing industry (electronics) in the beginning ofthe 1990s.

Since the bulk of the traditional enterprises were sit-uated in the Central and Southern regions, structuralchanges after the mid 1980s meant a process of de-industrialisation that affected the Central and Southernregions in particular (Table 2). The importance of indus-try in the South, traditionally the most industrial regionand since the 1930s the home of heavy industry, deterio-rated. The North, on the other hand, became increasinglyimportant as the geographical site for technology and skillintensive industry and services.

Simultaneously, the political economy landscapeshifted towards a more liberal and open regime whenKMT-authoritarianism and centralism began to weaken.In addition, opportunities for investing and setting upbusinesses in mainland China increased sharply in theearly 1990s (Zhang, 2005) and to an increasing extentthese investments were ICT-related (Chen, 2004). Thecombined effect was, at least initially, a closer collabo-ration between firms and government in efforts to enterthe new opportunities given not least in the ICT-sector(Hsu, 2004). Due to this, firms tended to locate in the mostadvanced region – the North – following the logic of spatialconcentration in the early phases of industrial develop-ment to enjoy first mover advantage and inter-industryspillover effects. This might be even more relevant in asector which is skill-biased when a concentration of skillsmight be desirable. The industrial structure experienced a

generation shift where the rurally oriented, labour inten-sive, local (curb-market) capital dependent SMEs graduallydeclined while high-tech FDI oriented SMEs were gaining

174 M. Andersson, M. Klinthäll / Structural Change and Economic Dynamics 23 (2012) 170– 179

Source: DGBAS 2007

0

5000

10000

15000

20000

25000

30000

35000

40000

45000

50000

1973

1975

1977

1979

1981

1983

1985

1987

1989

1991

1993

1995

1997

1999

2001

2003

2005

Textiles mills Electronic parts & components manufacturing Services

ees on p

Fig. 1. Average earnings of employSource: DGBAS (2007).

strength. Importantly, the 1980s and early 1990s was alsoa period when the capital markets were deregulated andaccessibility to formal capital increased also for SMEs thatpreviously had relied on the curb markets. The industrialstructure and the small and medium enterprises becameincreasingly technology-intensive and, contrary to theSME sector of the previous decades, more integrated anddetached from the rural sector.

The successful rise of the IT and electronics sector, in the1990s and 2000s world leading in, e.g. semi-conductors,is markedly more skill intensive and dependent on for-eign technology, in sharp contrast to the traditional labourintensive sector, e.g. textile production (see Chen andLin, 2007). Not only do we find a higher proportion ofskilled labour in the North, but also a higher accessi-bility to both domestic and international capital and amore advanced infrastructure (of the 228 members onthe Taiwan Venture Capital Association member list, 224were situated in Taipei).3 Concerning geographical distri-bution of operation headquarters, there is a strong biastowards the North (Lee, 2006). These enterprises alsobecame more dependent on the returns of clustering andconnections to state-sponsored R&D. For instance, the coreof Taiwan’s technology sector is situated in the northernregion between Hsinchu and Taipei where both competi-tion and vertical cooperation among local firms is strong(see Saxenian and Hsu, 2001). More than 85 percent ofthe semiconductor firms are concentrated in the HsinchuScience based Industrial park, an area which during the1960s and 1970s hosted the export-oriented light bulbindustry (Hsu and Cheng, 2002; see also Lin, 1997). Net-works previously engaged in traditional industries became

important in the rise of high-tech industries (see Guerrieriand Pietrobelli, 2006). According to Lam and Clark (1994:418), “many of the electronics and computer firms today

3 http://www.tvca.org.tw/english/tvca Member List 20070410.pdf.Downloaded 2009-10-9.

ayrolls-annual by industry (NT$).

are former textile makers that gradually learned how toassemble computers”.

The change in the political economy, including increas-ing cross-strait economic integration, and the dynamics ofnew technology has reshaped the industrialisation process.The government has been active in subsidising high-techstate-of-the-art industries when the heavy industrialisa-tion had ran out of steam and Taiwan had lost its formeradvantages in producing labour-intensive goods (see e.g.Hsu and Cheng, 2002; Hsu, 2004). SME operators, eventu-ally convinced by the potential gains from agglomerationin this type of production, given the increasing possi-bilities of getting access to the capital markets, becamewilling to operate more intensively in the North. Para-doxically, state initiative therefore became more directlyimportant for many SMEs than during the state guidanceof (authoritarian) KMT rule, even if the role of state policiesgradually became less and less important for the deepen-ing of high-technology firm activities (see Hsu, 2004). Withthe relative demise of state dirigisme, state capacities toinfluence regional development generally weakened andpolicies to counter regional imbalances have been largelyneglected both under Democratic Progressive Party (DPP)rule (2000–2008) and post-2008 KMT rule (see Hsu, 2009;Taipei Times, 2011).

To sum up, cumulative causation implies that increas-ing returns/diseconomies of agglomeration are consistentfeatures of the industrialisation process. These differencesneed not, however, be constant nor linear over the longerterm but rather marked by discontinuity. Growth withequity in Taiwan up until the 1980s was characterisedby strong trickling down/spread effects as the South-ern regions were catching up with North through rurallybased and labour intensive industrialisation. Productivitygains through the transfer of surplus labour from agricul-

ture into industry made incomes in the Southern regionsincrease relative to the North. Convergence was, however,halted when labour intensive industry lost its competitiveedge through the appreciation of the NTD after the Plaza

M. Andersson, M. Klinthäll / Structural Change and

Source: Survey of Family I ncome and E xpenditure 1976-2005

South

Cen tral

0,65

0,7

0,75

0,8

0,85

0,9

1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004

FtS

Asefibt

edcic

4

ItGw(oaiphcif

at8mhthaap

ig. 2. Mean household income per capita in Central and South relativeo North. Annual figures and 5-year moving averages (North = 1).ource: Survey of Family Income and Expenditure (1976–2005).

ccord and there was a switch towards knowledge inten-ive production. Cumulative processes regarding politicalconomy, cross-strait opening, demography, technology,nance, and other factors, meant that high tech industryecame increasingly concentrated in the North and newrends of divergence emerged.

In the next section, we analyse how regional differ-nces in household income over time reflect relative skillsistribution and regional variation in returns to humanapital and to what extent there exists a regional bonus,.e. externalities to agglomeration due to cumulativeausation.

. Patterns of regional household income disparity

There is no single measurement of regional inequality.n the empirical literature, commonly used indicators arehe location of industries often measured by the locationini coefficient (Krugman, 1991) or Ellison–Glaeser indexhere the size distribution of plants are controlled for

Ellison and Glaeser, 1997). Based on national accounts datan regional income per capita, regional real per capita GDPnd the coefficient of variation are widely used methodsn measuring regional inequality. In this study we com-are household incomes across regions, using a traditionaluman capital approach that takes account of householdharacteristics. We use a Mincerian wage equation, wherendicators for region of residence capture geographical dif-erences in household per capita income.

The analysis is based on the Survey of Family Incomend Expenditure in Taiwan, (SFIET), a national surveyhat has been launched annually since 1976, covering933–14,974 households per year. The data includes infor-ation on household characteristics, various types of

ousehold incomes and expenditures, housing classifica-ion, fixed assets, characteristics of household head and of

ousehold members. We use 30 cross-sections 1976–2005nd our sample covers 380,340 households, with a total ofpproximately 1.6 million household members. The studyopulation is limited to households where the age of the

Economic Dynamics 23 (2012) 170– 179 175

household head is 18–65 years and where the householdhas a maximum of 10 members. The data set, after theselimitations, still includes over 95 percent of the total num-ber of households included in the survey.

First, we calculate mean per capita household incomesannually by region. Fig. 2 shows the ratio of per capitahousehold incomes in the Central and South regions1976–2005 relative to North.

The thin black lines show annual figures and the boldgrey lines show 5-year moving averages, in order to visu-alise the long-term trends. The time-series of income ratiosin Fig. 2 show how income levels converge with Northuntil the late 1980s in Central and until the early 1990sin the South, after which there is a trend of increasing dif-ferences in mean per capita household incomes betweenNorth and the Central/South regions. The catch-up pro-cess of Central/South comes to and end and a new patternof divergence appears from the early 1990s onwards. TheCentral region is consistently on a lower income level com-pared to South, with an average gap of 5.4 percentagepoints to South, although annual variation is significant.The income levels in the South are in their turn consis-tently on a lower level compared to the North, moving fromaround 76 percent of the northern income levels in the late1970s to a peak of close to 86 percent in 1991 and then backto a level around 79 percent at the end of the period. Thisis indicative of a process of cumulative causation hostingfirst decreasing then increasing returns to agglomerationin the North.

Next, in order to investigate to what extent theseincome differences of agglomeration are due to aregional bonus effect or a composition effect regardingdemographic and human capital characteristics of thehouseholds, we estimate a model where those householdcharacteristics are accounted for. As dependent variable weuse the logarithm of per capita household income, i.e. totalincome of the household divided by the number of mem-bers in the household. Total household income includestransfers from employers, government, individuals, com-panies, private and social insurances, entrepreneurialincome, property income, as well as transfers from abroad.Independent variables include age, sex, marital status andlevel of education of the household head and of the spouse,as well as the size and location of the household.

In order to capture the regional bonus effect over time,we estimate a linear regression model

Y = ̨ + ˇ′X + ı′R + ε

where Y is the logged annual per capita household income, ̨ is intercept, X is the matrix of household-specific vari-

ables, R is an indicator of region of residence of thehousehold, and ε is the random error term. This is the“full” model, where “full model (a) only includes educa-tional level of the household, and “full model (b)” alsoincludes educational level of the spouse. For comparison,we also estimate an “empty” model, where only household

size is included, but no variables for the demographic andhuman capital characteristics of the household. The differ-ence between the coefficient of R of the full model and thatof the empty model is regarded as the “human capital”

176 M. Andersson, M. Klinthäll / Structural Change and Economic Dynamics 23 (2012) 170– 179

Source: Survey of Family I ncome and E xpenditure 1976-2005

0.1

0.14

0.18

0.22

0.26

0.3

0.34

1976 197 8 198 0 198 2 198 4 198 6 198 8 199 0 199 2 199 4 199 6 199 8 200 0 200 2 200 4

Empty model

Full model (a)

Full model (b)

Fig. 3. Regional income disparity in Taiwan 1976–2005, absolute dif-Source: S urvey of Family I ncome and E xpenditure 1976-2005

0,08

0,12

0,16

0,2

0,24

0,28

1976 197 9 198 2 198 5 198 8 199 1 199 4 199 7 200 0 200 3

Empty model

Ful l model (a)

Fig. 4. Regional income disparity within the manufacturing sector,Taiwan 1976–2005, absolute difference between region indicators.Annual figures and 5-year moving averages.

ference between region indicators. Annual figures and 5-year movingaverages.Source: Survey of Family Income and Expenditure (1976–2005).

effect. Remaining differences suggest a “regional bonus”which is not accounted for by household characteristics.

The models are estimated for each cross-section1976–2005 and the coefficients of the region indicators areused as a measure of the effect of location on householdincome (in regard to, e.g. the regional industrial structureand labour market). The larger the absolute differencebetween the coefficients of the richest and poorest region,the greater the disparity. The absolute differences betweenthe region indicators for the richest and the poorest regionare plotted by calendar year in Fig. 3.4

Fig. 3 shows how inter-regional per capita householdincomes converged in the course of the 1980s and thenchanged into divergence in the 1990s. The two trends arenoticeable contemplating the annual figures, and the 5-year moving averages show the long-term pattern quiteclearly. The coefficients of the region indicators in theempty model show that living in the North was associatedwith a 30 percent higher per capita household income inthe beginning of the period, decreasing to 24 percent in thelate 1980s to early 1990s and then increasing again, backto 30 percent in the end of the period. The coefficients ofthe region indicators of the full model (b) move from 19percent in the 1970s down to 13 percent in the early 1990sand then up to 21 percent at the end of the period.

The figure also reveals significant level differences;the coefficients of the full model, where control variablesfor demographic and human capital characteristics of the

household are included, are significantly lower than thoseof the empty model. Over time, the average differencebetween the empty and the full model (b) is 10 units,

4 Households in the eastern region are excluded. The regional dispar-ities in Taiwan are greater if East is included, but since it is a rural andscarcely populated region, it is not the most relevant reference in thediscussion of industrial development. In addition, the eastern region rep-resents few observations in the dataset (only about 3 percent), makingthe estimation results shaky when East is used as reference category.

Source: Survey of Family Income and Expenditure (1976–2005).

implying that age, sex, marital status and education lev-els of the household head and spouse explains 36 percentof the income advantage of the North. Hence, part of theregional income disparity is due to population composition,since the North has an advantage concerning demographicand human capital characteristics. Still, even after control-ling for these characteristics, the income advantage of theNorth is significant throughout the period, implying thatthere is a regional bonus effect in combination with thehuman capital effect.

Furthermore, the trend pattern stays basically the sameafter controlling for the characteristics of the household, i.e.the gap between the empty and the full models is relativelyconstant over time. Hence, the convergence/divergencepattern is not due to changing regional human capital com-position, for instance due to selective migration. Rather,the pattern of income disparity seems to be linked tostructural change of the regional economies, since theregional bonus-component is the one showing conver-gence followed by divergence. The first trend of regionalconvergence indicates inter-regional catching-up as partof long term structural changes and growth with equityin Taiwan. Regional spread/trickling down effects werestrong. The following trend of divergence shows how theNorth moves ahead as the ICT industries gain groundand become concentrated in the North. Regional back-wash/polarisation effects are cumulated, resulting in anincreased northern regional bonus.

Since the structural change of the economy that beginsin the North is due to the spread of ICT-based industryfrom the late 1980s, we expect manufacturing to be thefirst sector to display a growing inter-regional income gap.Fig. 4 shows the plotted difference between the coefficients

of the region indicators for the richest and the poorestregion, when only households where the household headis employed within the manufacturing sector is included in

M. Andersson, M. Klinthäll / Structural Change and

Source: Survey of Family I ncome and E xpenditure 1976-2005

0,1

0,15

0,2

0,25

0,3

0,35

0,4

0,45

1976 197 8 198 0 198 2 198 4 198 6 198 8 199 0 199 2 199 4 199 6 199 8 200 0 200 2 200 4

Ful l model (a)

Empty model

Fig. 5. Regional income disparity outside the manufacturing sector,Taiwan 1976–2005, absolute difference between region indicators.AS

tt

aotrtloaavsSoiiitowfc

ioiigtmtotos

trialisation produced a cumulated tendency of regional

nnual figures and 5-year moving averages.ource: Survey of Family Income and Expenditure (1976–2005).

he estimation. Since households in Fig. 4 are sampled byhe sector of the household head, we use full model (a).

The values display a high degree of annual variation, but U-shaped pattern of income disparity is distinguishablever the period of investigation, in particular when con-emplating the 5-year moving averages. The empty modeleveals an increased inter-regional income disparity fromhe late 1980s, whereas the trend break of the full model isocated in the early 1990s, patterns that correspond to thever-all picture of Fig. 3. However, over time there is also

pattern of convergence–divergence between the emptynd the full model estimations. Hence, inter-regional con-ergence of household income within the manufacturingector is partly explained by economic catch-up of theouth and Central regions and partly by a convergencef inter-regional returns to human capital. The follow-ng divergence is partly due to regional bonus effects –ndustrial change in the North – and partly explained byncreasing returns to human capital in the North. The indus-rial development in the North has a general positive effectn incomes, relative to the other regions, of householdsithin manufacturing. This positive effect is even stronger

or households where the head has a high level of humanapital.

Fig. 5 displays the results of the two models when thencome disparity was estimated including only householdsutside the manufacturing sector. The time-series show anncome convergence from the beginning of the period ofnvestigation, reflecting a general trend of income conver-ence in Taiwan until the late 1980s/early 1990s. However,he following pattern of increased divergence found in the

anufacturing sector is not as dramatic outside manufac-uring. There is a slight upward trend in the second halff the period, but the development within the manufac-

uring sector does not seem to affect incomes as much inther sectors in the North. In addition, we do not find aimilar convergence-divergence pattern regarding returns

Economic Dynamics 23 (2012) 170– 179 177

to human capital, rather the opposite; the regional bonusaccounts for the divergence after 1992. On the other hand,the regional income disparity over the studied period as awhole is larger outside manufacturing.

Hence, the inter-regional income divergence since theearly 1990s is primarily located within manufacturingindustry, but there seems to be spill-over effects also toother sectors since the regional bonus to households in theNorth is increasing also outside manufacturing.

5. Conclusions and interpretation

The aim of this paper was to analyse the regionalincome disparities over time in Taiwan, inspired by thetheory of cumulative causation. Myrdal, himself an explicitadvocate for equity – not only for reasons of moralitybut also because he was convinced that equity facilitatesgrowth – suggested that regional divergence is a nat-ural result of open and liberalised markets. Cumulativecausation would go in the direction of divergence untilexternal diseconomies in the North set in or spread effectsof Northern expansion will stimulate also developmentin the lagging regions. Such stimuli would more rapidlycome into existence if policy makers are attentive to theregional inequality problem. Another proponent of cumu-lative causation, Hirschman, agreeing with Myrdal on thenatural tendency of divergence and also on the role of poli-cies to speed up trickling down, would perhaps be lesspessimistic on the possibilities for such measures to beadvanced. They both argued strongly for analysing suchprocesses in the context of also institutional and politi-cal factors. Sympathetic as we are to such an approachwe set out to investigate in more detail the empiricalnature underlying central aspects of the forces of cumula-tive causation. One caveat might first be in order related tothe theoretical concepts proposed by Hirschman (tricklingdown/polarisation) and Myrdal (spread/backwash effects)that are supposed to capture the forces of income con-vergence/divergence. Although both the converging anddiverging regional income dynamics are observed in thecase of Taiwan, trickling down or spread are misleadingnotions when it comes to the process of convergenceuntil the early 1990s in Taiwan. The catching-up of thelagging regions to the North would more correctly becharacterised as some sort of decentralised or multi-coreagglomeration where the impetus of industrial develop-ment of the relatively more backward regions originatedin the rural sector rather than merely inter-regional spill-over effects or investments from the North. Inter-regionalincome dynamics during the convergence phase until themid-1980s seem in the case of Taiwan to be less of adirect function of the leading region. Such theoreticaland conceptual misgivings aside, however, the cumulativecausation-approach still offers a theoretical understandingof observed phenomena.

The Taiwanese growth with equity model based on anegalitarian land reform and labour intensive rural indus-

income convergence. Our analysis of household datasuggests that this converging trend converted into diver-gence in the early 1990s. Both trends suggest cumulative

ange and

178 M. Andersson, M. Klinthäll / Structural Ch

processes based on the different dynamics of consecu-tive phases of the industrialisation process and that thebreak in the trend of regional income convergence coincidewith major events in East Asian and Taiwanese develop-ment. Liberalisation, forced appreciation of the currencyafter the Plaza Accord, political democratisation, politicalopening and changing economic opportunities vis-à-vismainland China and industrial restructuring are all signif-icant aspects of the period around late 1980s and early1990s and our analysis supports the notion that Taiwan’srenowned growth with equity model seems to have beendebilitated.

Our analysis shows the income effects of regionalagglomeration and how much of inter-regional incomedifferences are explained by human capital effects anda regional bonus, respectively. In accordance with thetheory of cumulative causation, we also see a contin-uous change over time: the relative regional positionsare never constant. We also noted that regional catch-ing up turned into regional divergence in the early 1990s.This tendency of regional divergence seems driven by theincomes in the northern manufacturing sector, i.e. the ICTindustries in the North. This lends support to the notionthat industrial development is uneven and that increasingreturns to agglomeration is due both to concentration ofskill and knowledge and to a regional bonus. When welook at the manufacturing and non-manufacturing sec-tor, respectively, the human capital effect accounts for alarger proportion of the regional income differences dur-ing the period of inter-regional divergence compared to theperiod of inter-regional convergence. Increasing returnsto human capital thus seem to be stronger during phasesof industrial upgrading. This should come as no surpriseto scholars in the tradition of endogenous growth andthe new geography. More interesting might be that theregional bonus also increases in absolute terms duringthe period of divergence and that the major explanationof the regional effect is made up of the regional bonus.The North–South divide has become a Taiwanese realityand an argument for the DPP to regain its power position(see Taipei Times, 2011).

The fact that new economic activities are increasinglyconcentrated to the North does not necessarily make trick-ling down/spread effects impossible if the other regions canattract capital inflows from the North as production costsincrease. Additionally, the North might experience conges-tion and downward pressure on disposable income as costsof living increase, changing opportunity costs in favour ofmoving out of the North (or not moving into the North).However, at least in the early ICT-oriented industrialisa-tion phase, the trickling down/spread effects seem to beweaker since the domestic linkages in terms of final desti-nation markets play a more marginal role. A higher wagepremium and skill-intensive industrial change also makelabour less substitutable. These forces tend to be polaris-ing. The agglomeration effect is likely to persist and thehuman capital effect will probably continue to be strong.

If these cumulative processes are to be found in other ICT-oriented economies remains to be seen and to advance ourunderstanding of the regional bonus is a task ahead for bothscholars and policy makers.

Economic Dynamics 23 (2012) 170– 179

Acknowledgement

The authors thank Christer Gunnarsson, Lennart Schönand three anonymous referees for insightful comments.Support from the Swedish Research Council is gratefullyacknowledged.

References

Abramovitz, M., 1986. Catching up, forging ahead and falling behind. Jour-nal of Economic History 46, 385–406.

Andersson, M., Gunnarsson, C., 2003. Development and Structural Changein Asia-Pacific: Globalising Miracles or End of a Model? Routledge,London.

Barro, R., Sala-i-Martin, X., Blanchard, O.J., Hall, R.E., 1991. Convergenceacross states and regions. Brookings Paper on Economic Activity 1,107–182.

Barro, R., Sala-i-Martin, X., 1992. Convergence. Journal of Political Econ-omy 100 (2).

Baumol, W.J., 1986. Productivity growth, convergence and welfare: whatthe long-run data show. American Economic Review 76, 1072–1085.

Bernard, A.B., Jones, C.I., 1996. Technology and convergence. The EconomicJournal 106 (July), 1037–1044.

Chatterji, M., 1992. Convergence clubs and endogenous growth. OxfordReview of Economic Policy 8, 57–69.

Chen, S-H., 2004. Taiwanese IT firms’ offshore R&D in China and theconnection with the global innovation network. Research Policy 33,337–349.

Chen, Y.-M., Lin, F.-J., 2007. Regional development and sources of superiorperformance across textile and IT sectors in Taiwan. Entrepreneurshipand Regional Development 18 (3), 227–248.

Cheng, C., 1961. Land reform in Taiwan. China Publishing Company,Taiwan.

China Post, 2008. M-shape society trend getting worse. DGBAS (September4, 2008).

Cornia, G.A., Addison, T., Kiiski, S., 2005. Income distribution changes andtheir impact in the post-Second World War period. In: Cornia, G.A.(Ed.), Inequality, Growth and Poverty in an Era of Liberalization andGlobalization. Oxford University Press, United States.

DeLong, B., 1988. Productivity growth, convergence and welfare: a com-ment. American Economic Review 78, 1138–1154.

DGBAS, 2007. Directorate-General of Budget, Accounting and Statistics.Ellison, G., Glaeser, E., 1997. Geographic concentration in U.S. manufac-

turing industries: a dartboard approach. Journal of Political Economy105, 889–927.

Fagerberg, J., Verspagen, B., 1996. Heading for divergence? Regionalgrowth in Europe reconsidered. Journal of Common Market Studies34, 431–448.

Fei, J.C.H., Ranis, G., Kuo, S.W.Y., 1979. Growth with Equity—The TaiwanCase. Oxford University Press, Washington, DC.

Fujita, M., Krugman, P., Venables, A.J., 2001. The Spatial Economy: Cities,Regions, and International Trade. The MIT Press, Cambridge, MA.

Gerschenkron, A., 1962. Economic Backwardness in Historical Perspective.Harvard University Press, Cambridge.

Guerrieri, P., Pietrobelli, C., 2006. Old and new forms of clustering andproduction networks in changing technological regimes: contrastingevidence from Taiwan and Italy. Science Technology and Society 11(1), 9–38.

Hirschman, A.O., 1958. The strategy of economic development. Yale Uni-versity Press, New Haven.

Ho, S.P.S., 1979. Decentralized industrialization and rural development:evidence from Taiwan. Economic Development and Cultural Change28 (October).

Hsu, J-Y., 2004. The evolving institutional embeddedness of a late-industrial district in Taiwan. Tijdschrift voor Economische en SocialeGeografie 95 (2), 218–232.

Hsu, J-Y., 2009. The spatial encounter between neoliberalism and pop-ulism in Taiwan: regional restructuring under the DPP regime in thenew millennium. Political Geography 28, 296–308.

Hsu, J-Y., Cheng, L-L., 2002. Revisiting economic development in post-war Taiwan: the dynamic process of geographical industrialization.

Regional Studies 36 (8), 897–908.

Kaldor, N., 1966. Cases of the Slow Rate of Economic Growth of the UnitedKingdom. Cambridge University Press.

Kim, S., 2008. Spatial inequality and economic development: theo-ries, facts, and policies. Commission on Growth and Development,

nge and

K

L

L

L

L

L

M

MM

M

MM

M

O

P

M. Andersson, M. Klinthäll / Structural Cha

Working Paper No 16. The International Bank for Reconstruction andDevelopment/The World Bank.

rugman, P., 1991. Increasing returns and economic geography. Journalof Political Economy 99, 483–499.

all, S.V., Chakravorty, S., 2005. Industrial location and spatial inequality:theory and evidence from India. Review of Development Economics9, 47–68.

am, D., Lee, I., 1992. Guerrilla capitalism and the limits of Statist theory:comparing the Chinese NICs. In: Clark, C., Chan, S. (Eds.), The Evolv-ing Pacific Basin in the Global Political Economy—Domesticand International Linkages. Lynne Rienner Publishers,Boulder.

am, D., Clark, C., 1994. Beyond the developmental state: the culturalroots of guerrilla capitalism in Taiwan. Governance: An InternationalJournal of Policy and Administration 7 (4), 412–430.

ee, T-L., 2006. Action strategies for strengthening industrial clusters insouthern Taiwan. Technology in Society 28, 533–552.

in, C.-Y., 1997. Technopolis development: an assessment of the Hsinchuexperience. International Planning Studies (June 1).

addison, A., 1982. Phases of Capitalist Development. Oxford UniversityPress, Oxford.

addison, A., 1995. Monitoring the World Economy. 1820–1992. OECD.ankiw, N.G., Romer, D., Weil, D.N., 1992. A contribution to the empir-

ics of economic growth. The Quarterly Journal of Economics 107 (2),407–438.

ao, Y.-K., Schive, C., 1995. Agricultural growth and industrial devel-opment in Taiwan. In: Mellor, J. (Ed.), Agriculture on the Road toIndustrialization. The Johns Hopkins University Press, Baltimore.

arshall, A., 1890. Principles of Economics. Macmillan, London.yrdal, G., 1956. Development and underdevelopment: a note on the

mechanism of national and international economic inequality. In:National Bank of Egypt 50th Anniversary Commemoration Lectures,Cairo.

yrdal, G., 1957. Economic Theory and Underdeveloped Regions. Duck-worth, London.

hmae, K., 2006. M-shape Society. Business Weekly Publications, CitéPublishing Group.

ark, A., Johnston, B., 1995. Rural development and dynamic external-ities in Taiwan’s structural transformation. Economic developmentand Cultural Change 44 (1 (October)), 181–208.

Economic Dynamics 23 (2012) 170– 179 179

Perroux, F., 1955. Note sur la notion de pole de croissance. EconomieAppliqué 8, 307–320.

Pritchett, L., 1997. Divergence, big time. Journal of Economic Perspectives11 (3), 3–17.

Puga, D., 2002. European regional policies in light of recent location the-ories. Journal of Economic Geography 2 (4), 372–406.

Quah, D., 1993. Galton’s fallacy and test of the convergence hypothesis.Scandinavian Journal of Economics 95 (4), 427–443.

Quah, D., 1996. Empirics for economic growth and convergence. EuropeanEconomic Review 40, 1353–1375.

Romer, P.M., 1986. Increasing returns and long-run growth. Journal ofPolitical Economy 94 (5), 1002–1037.

Sánchez-Reaza, J., Rodríguez-Pose, A., 2002. The impact of trade liberal-ization on regional disparities in Mexico. Growth and Change 33 (1),72–90.

Saxenian, A.-L., Hsu, J.-Y., 2001. The Silicon valley–Hsinchu connection:technical communities and industrial upgrading. Industrial and Cor-porate Change 10 (4), 893–920.

Summers, R., Heston, A., 1991. The Penn World Table (Mark 5): anexpanded set of international comparisons, 1950–1988. QuarterlyJournal of Economics 106, 327–368.

Taipei Times, 2011. Tsai makes Policy Proposals on Balanced Development(August 18, 2011).

Temple, J., 1999. The new growth evidence. Journal of Economic Literature37 (1), 112–156.

The Survey of Family Income and Expenditure, 1976–2006. DGBAS.Veblen, T., 1898. Why is economics not an evolutionary science? Quarterly

Journal of Economics 12, 373–397.Wang, T.-F., February 1996. Report of A Study on the Estimation of Regional

Economic Indicators. Urban and Housing Development Department,Council for Economic Planning and Development.

Wicksell, K., 1898. Interest and Prices. Macmillan, London (Reprinted NewYork: Augustus M. Kelley).

Williamson, J.G., 1965. Regional inequalities and the process of nationaldevelopment: a description of the patterns. Economic Development

and Cultural Change 13 (4), 1–84.

Young, A., 1928. Increasing returns and economic progress. The EconomicJournal 38, 527–542.

Zhang, K.H., 2005. Why does so much FDI from Hong Kong and Taiwan goto Mainland China? China Economic Review 16, 293–307.