is the export-led growth hypothesis valid …the export-led growth hypothesis (elgh) postulates that...

UNITED NATIONS CONFERENCE ON TRADE AND DEVELOPMENT

POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES

STUDY SERIES No. 7

IS THE EXPORT-LED GROWTH HYPOTHESISVALID FOR DEVELOPING COUNTRIES?

A CASE STUDY OF COSTA RICA

by

Emilio J. Medina-Smith

University of Sussex, United Kingdomand

Universidad de Carabobo, Venezuela

iii

NOTE

The views expressed in this study are those of the authors and do not necessarilyreflect the views of the UNCTAD secretariat.

The designations employed and the presentation of the material do not imply theexpression of any opinion whatsoever on the part of the Secretariat of the United Nationsconcerning the legal status of any country, territory, city or area, or of its authorities, orconcerning the delimitation of its frontiers or boundaries.

Material in this publication may be freely quoted or reprinted, but acknowledgementis requested, together with a reference to the document number. A copy of the publicationcontaining the quotation or reprint should be sent to the UNCTAD secretariat:

ChiefTrade Analysis Branch

Division on International Trade in Goods and Services, and CommoditiesUnited Nations Conference on Trade and Development

Palais des NationsCH – 1211 Geneva

UNCTAD/ITCD/TAB/8

UNITED NATIONS PUBLICATION

Sales No. E.01.II.D.8

ISBN 92-1-112518-9

ISSN 1607-8291

Copyright 8 United Nations 2000All rights reserved

iv

ABSTRACT

The export-led growth hypothesis (ELGH) postulates that export growth is one of thekey determinants of economic growth. This study goes beyond the traditional neoclassicaltheory of production by estimating an augmented Cobb−Douglas production function. Theinclusion of exports as a third input provides an alternative procedure to capture total factorproductivity (TFP) growth. The study tests the hypothesis by analysing the case of CostaRica, using annual data for the period 1950−1997. In using several procedures to test forcointegration, it goes beyond the traditional time series studies by examining empirically theshort-term as well as the long-run relationship. The study finds that the ELGH is valid in thisparticular case; however, the empirical results show that physical investment and populationmainly drove Costa Rica's overall economic performance from 1950 onwards. From areview of the literature we find that the empirical evidence regarding the relationshipbetween exports and growth is not robust, and although the results of the study suggest thatexports have a positive effect on the overall rate of economic growth and could beconsidered an “engine of growth” as the ELGH advocates, their impact was quantitativelyrelatively small, in both the short and the long-run. The evidence presented clearly supportsthe neoclassical theory of production and, to a lesser extent, the so-called new-fashionedeconomic wisdom. Moreover, it challenges the empirical literature regarding the ELGH andexpresses serious doubts with regard to promoting exports as a comprehensive developmentstrategy. The ELGH is probably beneficial only for a limited number of developingcountries, and only to a certain extent.

Keywords: export-led growth hypothesis (ELGH); economic growth; neoclassical theory of production; cointegration; Costa Rica

JEL classification codes: D62; O30; 041

v

ACKNOWLEDGEMENTS

I am grateful to Susan Teltscher, Bijit Bora and seminar participants at the UnitedNations Conference on Trade and Development (UNCTAD). I am also very much in debt toMichael T. Sumner for very useful discussions, suggestions, insightful comments andconstructive criticism. I would also like to thank two anonymous referees for their helpfulcomments on an earlier version of this paper. Finally, editorial support from Cynthia Littleis greatly acknowledged. This research was supported by the Division on Trade in Goodsand Services, and Commodities (DITC) and carried out during August and September 1999while I was a Visiting Fellow at UNCTAD. Any errors are mine and the views expressedhere are my own and do not reflect the opinion of the United Nations or its affiliatedinstitutions. Comments and critiques are welcome and can be sent to me at the followingaddress: University of Sussex, The Graduate School in Social Sciences (GSiSS), Arts E,Office E313, Falmer, Brighton, East Sussex BN1 9SN, United Kingdom. Telephone: +44(1273) 606755 or 678891, ext. 2428; fax: +44 (1273) 673563; e-mail: [email protected]

vi

CONTENTS

I. INTRODUCTION ................................................................................................... 1

II. THEORETIAL REVIEW....................................................................................... 2A. Trade and growth........................................................................................... 2B. Exports and growth........................................................................................ 4

III. EMPIRICAL STUDIES.......................................................................................... 5A. Cross-section analysis....................................................................................10B. Country specific studies.................................................................................11C. Summary of the empirical literature ..............................................................13

IV. CASE STUDY..........................................................................................................15A. Variables and data sources.............................................................................15B. Methodology and results................................................................................17

1. Testing for unit roots .........................................................................182. Cointegration.....................................................................................203. Johansen maximum likelihood (ML) approach.................................234. Error correction model and cointegration..........................................265. Cointegration tests: An assessment ...................................................34

V. CONCLUDING REMARKS ..................................................................................35

REFERENCES ....................................................................................................................37

APPENDIX...........................................................................................................................43

Tables

1. A brief framework of the related economic literatureon the export-led hypothesis...................................................................................... 6

2. Unit root tests.............................................................................................................193. Static long-run relationship........................................................................................224. Johansen cointegration tests ......................................................................................255. Engle and Granger two-step procedure .....................................................................296. Unrestricted error correction model...........................................................................317. EG two-step procedure and unrestricted error correction model

with DUM81..............................................................................................................33

1

I. INTRODUCTION

It is widely accepted amongeconomists that economic growth is anextremely complex process, which depends onmany variables such as capital accumulation(both physical and human), trade, pricefluctuations, political conditions and incomedistribution, and even more on geographicalcharacteristics.

The export-led growth hypothesis(ELGH) postulates that export expansion is oneof the main determinants of growth. It holdsthat the overall growth of countries can begenerated not only by increasing the amounts oflabour and capital within the economy, but alsoby expanding exports. According to itsadvocates, exports can perform as an “engineof growth”.

The association between exports andgrowth is often attributed to the possiblepositive externalities for the domestic economyarising from participation in world markets, forinstance from the reallocation of existingresources, economies of scale and variouslabour training effects. However, thesemechanisms are frequently invoked without anytheoretical support or any empirical proof.

A substantial amount of researchconcerning the ELGH in developing countries(DCs) has been carried out during the past 30years. In fact, during the 1990s a new series ofempirical studies has been conducted on anumber of divergent lines of research,methodologies, time periods and countries.

A key aspect concerning early studiesis related to both the methodology and the

econometric technique used. The theoreticalbenchmark can be considered in general weakand based on bivariate and ad hoc productionfunctions, while the empirical results derivedfrom traditional econometrics have been highlycriticized for being spurious. Therefore, earlystudies could have been misleading in that theyadvocated export expansion in anindiscriminate way. In fact, the evidenceavailable is far from conclusive and this situationexplains to some extent why this debate stillexists in the economic literature.

Consequently, the purpose of this studyis to examine and test the ELGH, using the caseof Costa Rica. The study has three distinctivefeatures, in contrast to the hundreds ofempirical studies on growth that have beenpublished. First, we have gone beyond thetraditional neoclassical theory of production byestimating an augmented Cobb−Douglasfunctional form, which includes exports, usingannual data for the period 1950−1997. Theinclusion of exports as a third input ofproduction provides an alternative procedure tocapture total factor productivity (TFP) growth.Secondly, the study focuses on a singledeveloping country, examining empirically therelationship between export expansion andeconomic growth. Thirdly, it has gone beyondthe traditional short-term effects, and usesextensively modern time series to examineempirically the long-run relationship, employingseveral procedures to test for cointegration.Thus, the final aim of this study is to quantify theimportance of exports in the economicperformance of Costa Rica in the second partof the twentieth century.

2

II. THEORETICAL REVIEW

A. Trade and growth

Although the theoretical links betweentrade and economic growth have beendiscussed for over two centuries, controversystill persists regarding their real effects. Theinitial wave of favourable arguments withrespect to trade can be traced to the classicalschool of economic thought that started withAdam Smith and which was subsequentlyenriched by the work of Ricardo, Torrens,James Mill and John Stuart Mill in the first partof the nineteenth century. Since then, thejustification for free trade and the various andindisputable benefits that internationalspecialization brings to the productivity ofnations have been widely discussed and arewell documented in the economic literature (seee.g. Bhagwati, 1978; Krueger, 1978).

However, in the last decade there hasbeen a surprising and impressive resumption ofactivity in the economic growth literaturetriggered by the endogenous growth theory,which has led to an extensive inventory ofmodels that stress the importance of trade inachieving a sustainable rate of economicgrowth. These models have focused ondifferent variables, such as degree of openness,real exchange rate, tariffs, terms of trade andexport performance, to verify the hypothesisthat open economies grow more rapidly thanthose that are closed (see e.g. Edwards, 1998).

Although most models emphasized thenexus between trade and growth, they stressedthat trade is only one of the variables that enterthe growth equation. However, the advocatesof the ELGH have stated that trade was in factthe main engine of growth in South-East Asia.They argue that, for instance, Hong Kong(China), Taiwan Province of China, Singapore

and the Republic of Korea, the so-called FourTigers, have been successful in achieving highand sustained rates of economic growth sincethe early 1960s because of their free-market,outward-oriented economies (see e.g. WorldBank, 1993).

The extensive literature concerning therelationship between trade and growth is alsothe consequence of the many changes that havetaken place in the fields of developmenteconomics and international trade policy in thelast two decades. An example of these changesis the tremendous modification from inward-oriented policies to export promotion (EP)strategy.1

By the early 1980s export-ledorientation and export promotion had alreadysecured a wide consensus among researchersand policy makers, to such an extent that theyhad become “conventional wisdom” amongmost economists in the developing world (seee.g. Tyler, 1981; Balassa, 1985). This is stillthe case in some international organizations, theinternational bank community and multilaterallenders such as the World Bank and theInternational Monetary Fund (IMF), andamong the mainstream policy makers.

The advocates of the export-ledstrategy and free trade point out that mostdeveloping countries that followed inward-oriented policies under the import substitutionstrategy (ISS), mostly in Latin America, hadpoor economic achievements (Balassa, 1980).Some of them showed on average a completelack of growth, while real income declined

1 According to Heitger (1987, p. 249), the ELGH wassuggested initially by Kindelberger in 1962.

3

between 1960 and 1990 (see e.g. Barro andSala-i-Martin, 1995).

These facts were partly responsible forthe substantial change that occurred in the tradeliterature in the 1980s. For example, Bruton(1989) states that as the first stage of importsubstitution came to an end, those countriesthat continued with this strategy, particularly inLatin America, or that were unable to shift to amore outward approach became increasinglyvulnerable to external events. Most of thembecame increasingly dependent on short-runcapital inflows, in particular from private banks,in order to maintain their levels of imports andthus of consumption. This was particularly thecase of most Latin American countries thatwere greatly affected by the debt crisis of theearly 1980s.

Thereafter, many DCs were forced tostimulate their export-led orientation even morebecause most of them had to rely on multilateralorganizations to implement adjustment andstabilization programmes to correct imbalancesin their basic macroeconomic indicators. Thestrategy was to encourage a free marketthrough policies that relied heavily on the exportpromotion approach as one of the most suitableand trustworthy mechanisms. Promotingexports would enable DCs to correctimbalances in the external sector and at thesame time assist them in ensuring that theirdomestic economies made a full recovery.

As part of an outward strategy, a newset of policies rapidly became a key componentfor policy makers in DCs involved inadjustment and stabilization programmes. Inthis atmosphere, numerous Governmentsstarted at this time to stimulate exports usingdiverse mechanisms and instruments, such assubsidies and tax exemptions.2

2 It is clear, however, that the ability of Governments

Consequently, by the mid-1980s, theeconomic literature concerning developmenteconomics, economic growth, adjustment andstabilization programmes had quickly rejectedthe inward-oriented approach and wassuddenly placing great emphasis on export-ledstrategy. Most macroeconomic theorists andpolicy makers in DCs rapidly embraced thenew wisdom, in the belief that by following thisscheme their countries would achieve or regainthe high rates of growth of the past.

Each strategy has been subject of anextensive theoretical survey and that theliterature examining the relationship betweentrade and growth has increased substantially inthe last decade with the impetus provided bythe endogenous growth theory. However, it isnot the intention of the present study toparticipate in or contribute to the discussionconcerning the advantages and disadvantagesof both economic strategies, which recentlygained a new impetus (see e.g. Bruton, 1998;Edwards, 1998; Frankel and Romer, 1999;Rodrik, 1999).3

In addition, although the theoreticalliterature has frequently focused on therelationship between trade and economicgrowth (see e.g. Adams, 1973; Crafts, 1973;Edwards, 1992; Scott, 1992), the interestingphenomenon is that “empirical examinationshave typically examined the relationship

to promote exports through these mechanisms hasdiminished substantially since the late 1980s, wheneconomic integration agreements started to becomeincreasingly popular in both the developed world anddeveloping countries (i.e. EU, NAFTA, MERCOSUR),a situation which continued during the 1990scharacterized by the creation of economic blocs.3 Although the advantages and disadvantages ofeach strategy began to be an unavoidable topic to beexamined by policy makers in DCs, especially afterthe Asian economic crisis started at the end of 1997,becoming fully developed since then.

4

between exports and growth” (Levine andRenelt, 1992, p. 953).

Therefore, the next section brieflyreviews the empirical literature related to theexport-led strategy, considering in particular therole that exports played in output growth andpaying close attention to the issue of causallinks between exports and economic growth.

B. Exports and growth

Since the late 1960s studies have beenconducted to examine the role of exportperformance in the economic growth process.Although the empirical literature can beconsidered to be vast, its results are clearlycontradictory for both DCs and industrializedeconomies, a feature that could explain why this

topic is still at the top of the agenda for manyeconomists.

According to the so-called neworthodoxy, most authors as well as multilateralinstitutions would agree that promoting exportsand achieving export expansion are beneficialfor both developed and DCs for many reasons,including the following (i) they generate agreater capacity utilization; (ii) they takeadvantage of economies of scale; (iii) they bringabout technological progress; (iv) they createemployment and increase labour productivity;(v) they improve allocation of scarce resourcesthroughout the economy; (vi) they relaxe thecurrent account pressures for foreign capitalgoods by increasing the country’s externalearnings and attracting foreign investment; and(vii) they increase the TFP and consequentlythe well-being of the country (World Bank,1993).

5

III. EMPIRICAL STUDIES

Table 1 presents a summary of a set of42 empirical studies conducted between 1967and 1998, which includes time period,methodology, variables, econometric techniqueand conclusions reached by the researchers.Although a substantial part of the earlier studiesfound evidence of a correlation betweenexports and growth which was used to supportthe ELGH, this tends to hold only for cross-section studies. In fact, the recent evidence ontime series, which makes extensive use ofcointegration techniques, casts doubts on thepositive effects of exports on growth in the longrun, and is thus not as conclusive as it waspreviously thought to be. Therefore,explanations regarding this extensive empiricalliterature are in order.

Among earlier empirical studies Emery(1967, 1968), Syron and Walsh (1968),Serven (1968), Kravis (1970), Michaely(1977), Heller and Porter (1978), Bhagwati(1978) and Krueger (1978) should bementioned. This first group of studies explainedeconomic growth in terms of export expansionalone, in a two-variable framework. That is,they used bivariate correlation theSpearman rank correlation test in cross-country format to illustrate the alleged superioreffects of the ELGH (Lussier, 1993, p. 107).

A second group of researchers, whichincludes Balassa (1978, 1985), Tyler (1981),Feder (1983), Kavoussi (1984), Ram (1985,1987) and Moschos (1989), studied therelationship between export and outputperformance within a neoclassical framework.In most of these studies exports were includedin an ad hoc manner in the production function,together with labour and capital. They claimedthat by including exports they were taking intoconsideration a broad measure of externalities

and productivity gains generated by this sectorwhich stimulated the domestic economy. Themajority of these investigations aimed atanalysing DCs by using ordinary least squares(OLS) on cross-section data and used theirresults to demonstrate the advantages of theexport promotion strategy in comparison withthe import substitution policy.

It was not until recently that this line ofresearch began to focus on country-specificstudies, for both industrialized countries andDCs. Surprisingly, more than half the empiricalinvestigations published in the 1990s found nolong-run relationship between exports andeconomic growth; rather, the studies suggestthat it arises only from a positive short-termrelationship between export expansion andgrowth of gross domestic product (GDP).

The studies of industrialized nationshave analysed the cases of Canada, France,Germany, the United Kingdom, the UnitedStates and Switzerland, among others. In onlya few cases have the empirical resultsconfirmed that export expansion was a keyelement in the economic success of thosecountries (see e.g. Kugler, 1991; Afxentiou andSerletis, 1991; Henriques and Sadorsky,1996). Even more astonishing is the finding inrelation to Japan, which is that internal forceswere the handmaidens of the great Japaneseeconomic success in the twentieth century,including the post-war period, and not trade asmany have claimed in the recent past (seeBoltho, 1996).

Table 1. A brief framework of the related economic literature on the export-led hypothesis

MethodologyStudy Samplea Period of

studyData set Economic growth Exports Econometric

techniqueOther variables Conclusions

Emery (1967) 50 1953−1963Averages

Cross-section GNP growth Export growth OLS Current account Support for the export-led hypothesis.

Syron & Walsh (1968) 50 1953−1963 Cross-section GNP growth Exports OLS Support for the hypothesis but the resultsare sensitive depending on the type ofcountry under scrutiny � LDCs ordeveloped countries.

Serven (1968) 50 1953−1963 Cross-section GNP growth Export growth and/orexportchange/output

OLS Support for the hypothesis andrecommends the use of export growth andexport change/output.

Kravis (1970) 37 1835−1966 Cross-section GNP Export growth Spearman rankcorrelation

None Supports the export-led hypothesis; however,indicates that LDCs that have been capableof diversifying their exports have been moresuccessful in terms of growth.

Michaely (1977) 41 1950−1973 Cross-section Per capita GNPgrowth

Growth of exportshare

Spearman rankcorrelation

None Support for the export growth hypothesis andsuggests the existence of a threshold effect.

Balassa (1978) 11 1960−1973 Cross-section Real GNP growth Real export growth Spearman rankcorrelation, OLS,production function

Labour force, domesticinvestment and foreigninvestment/output

Support for the export growth hypothesis.

Heller & Porter (1978) 41 1950−1973 Cross-section Output growth rateGNP

Per capita exports Spearman rankcorrelation

None Little support for export growth causinggrowth.

Fajana (1979) 1 1954−1974Nigeria

Time series GDP growth Export share of GDPand exportchange/GDP

OLS,two-gap model

Foreign capital Supports the export-led hypothesis andsuggests that it is due to changes in domesticinvestment resources.

Tyler (1981) 55 & 49 1960−1977Middle-income

LDCs

Cross-section Real GNP growthand GNP per capita

Real export growth Pearson andSpearman rankcorrelation,OLS, productionfunction

Labour force growth,investment growth

Supports the export growth hypothesis andsuggest the existence of an threshold effect.

Feder (1983) 32 1964−1973 Cross-section GDP growth Export growth andexportchange/output

OLS,production function

Labour force growth,investment/output

Supports the export growth hypothesis.

Kavoussi (1984) 73 1960−1978Low- and

Cross-section Real GDP growth Real export growth Spearman Rankcorrelation, OLS,

Labour growth, capital growth Support for the hypothesis, however, theeffects tend to diminish according to the level

6




middle-incomeLDCs

production function of development.

Balassa (1985) 43 1973−1979Semi

industrializedcountries

Cross-section Real GNP growth Real export growth OLS, productionfunction

Savings, labour GDP percapita, share of exports(manufactured products)

Supports the hypothesis and suggests thatoutward trade orientation is beneficial.

Jung & Marshall (1985) 37 1950−1981LDCs

Time series Real GNP or GDP Lagged real exportgrowth

OLS,Granger causalitytest

Lagged GNP and GDP growth Only in 4 cases out of 37 was there evidencethat supported the export-led hypothesis(Indonesia, Egypt, Costa Rica and Ecuador).

Ram (1985) 73 1960−19701970−1977

Low- andmiddle-income

LDCs

Time seriestwo-sub periods

Real GDP growth Real export growth OLS, White test forspecificationbias andheteroskedasticity

Labour force growth andInvestment growth

Supports the export growth hypothesis andsuggests the existence of an threshold effect.

Chow (1987) 8 1960−1980NICs

Time series Manufacturing outputgrowth

Export growth ofmanufactured goods

Sim’s causality Test(1972), bivariatemodel

None Support for reciprocal causality hypothesisregarding export growth and industrialdevelopment.

Darrat (1987) 4 1955−1982Four-littledragons

Time series Real GDP growth Lagged real exportgrowth

OLS,White test,bivariate model

None Rejects the export growth hypothesis in 3 outof 4 cases. Is able to support it in only onecase (Republic of Korea) on the basis of thecausality test.

Heitger (1987) 36 1950−1970Averages

Cross-section Real GDP per capita Export share of GDP OLS, ad hocproduction function

GDI/GDPeffective rate of protection,labour force, technologicaladaptation and adult literacy

Supports the case for trade liberalization.

Ram (1987) 88 1960−19721973−1982

Low- andmiddle-income

LDCs

Cross-sectiontwo

sub-periods

Real GDP growth Real export growth OLS, productionfunction

Government size, GDI/GDP,labour growth

Supports the export ledhypothesis butasserts that the huge intercountry differencesand diversity suggest caution wheninterpreting the results.

Moschos (1989) 71 1970−1980Averages

Cross-section Real GDP growth Real export growth OLS, productionfunction

Labour growth,real domestic investmentgrowth

Supports the export-led growth hypothesisand suggests the existence of an thresholdeffect. The rate of growth seems unaffected bylabour because of its magnitude, while capitalhas limited effects owing to its low productivitylevels.

Colombatto (1990) 47 1971, 1978and 1985

Cross-section3 separate years

OLS, correlationcoefficients

Government consumption,agricultural exports and

Rejects the export -led growth hypothesis.

7




degree of opennessFosu (1990) 28 1960−1970

1970−1980African

countries

Pooledcross-sectional

two periods

GDP growth Rate of growth ofmerchandise exports

OLS, production function

Rate of growth of GDI, labourgrowth

Supports the export -led hypothesis.

Kugler (1991) 6 1970(1)-1987(4)Industrialcountries

Time series GDP exception in thecase of the US (GNP)

Real export growth ADF unit roots,Johansen’sprocedure, VARs

Consumption (durable, non-durable and services),investment (business fixed)

There is only weak empirical evidencesupporting the export-led growth hypothesis.In only 2 cases out of 6 is a long-run relationverified (France, West Germany).

Afxentiou & Serletis(1991)

16 1950−1985Industrialcountries

Time series Real GNP growth Real export growth Phillips-Perron unitroots,EG procedure,Granger causalitytests

None No systematic relationship between exportsand GDP is verified. Only in 2 cases out of 16was a bidirectional causality manifested (USand Norway).

Sengupta (1991) 5 1967−1986South-East

Asia(Republic of

Korea)

Time series Real GDP growth Real export growth OLS, productionfunction

Labour growth andcapital growth.

Supports the export-led hypothesis andsuggests the positive externality effects ofexports on growth.

Serletis (1992) 1 1870-1985Canada

Time series Real GNP growth Real export growth ADF unit roots, EGprocedure,Granger causalitytests

Imports Supports the export-led growth hypothesis inthe short run; however, no cointegrationbetween the variables was found.

Khan & Saqib (1993) 1 1972−1988Pakistan

Time series GDP growth Real export growth:primary products and manufactured goods

3SLS, productionfunction

Labour growth,capital growth,World GDP Index, relativeprices

Supports the hypothesis of a strongassociation between exports and growthperformance.

Lussier (1993) 24 & 19 1960−1990African

economies

Cross-section andpanel data

GDP growth Real export growth OLS, 4 versions ofproduction function

Labour growth,GDI/GDP,export share of GDP

Supports the hypothesis in panel data butfails to find any positive association whenusing export growth as a share of GDP.

Sheehey (1993) 31 & 65 1960−1970Semi-

industrializedcountries

Cross-section GDP growth Real export growth OLS, productionfunction

Labour growth, GDI/GDP,export share of GDP

Inconsistent evidence of higher productivity inthe export sector compared with the non-export sector; thus, suggests caution whenanalysing empirical results.

Greenaway & Sapsford(1994)

19 1957−19851970−19851971−1985

Time series Real GDP growth Real export growthand exportchange/output

OLS, 3 versions ofproduction function

Labour growth, rate of growthof investment, dummy forliberalization episodes

Little support for the export-led growthhypothesis and for the positive liberalizationeffects on growth.

8




Lee & Cole (1994) 73 1960−19701970−1977

Cross sectiontwo sub-periods

Real GNP growth Real export growth 2SLS, productionfunction,Hausman’s test

Labour growth,GDI/GDP

Supports the existence of a bidirectionalcausality between exports and growth.

Van den Berg & Schmidt(1994)

17 1960-1987Latin America

Time series Real GDP growth Real export growth Phillips-Perron unitroot, EG two-stepprocedure, OLS,VARs, productionfunction

GDI/GDP,population growth

Points to a positive long-run relationshipbetween exports and growth in 11 of the 16cases analysed. Costa Rica is among thosecountries where the hypothesis was verified.

Jin (1995) 4 1976(2)-1993(2)

Four little tigersof Asia

Time series Real GDP Real exports F-tests, ADF, impulseresponse function,VARs, EG two-stepprocedure

Real exchange rate, foreignprice shock, output shock

Bidirectional causality was found in the shortrun but no cointegration was detected;therefore, no long-run relationship is proved.

Figueroa de la Barra &Letelier-Saavedra (1994)

1 1979(1)−1993(4)

Chile

Time seriesquarterly

Real GDP Real exports andexportchange/output

ADF unit root, VARs,Johansen’sprocedure

Labour force,capital,exports + imports/GDP

Supports the hypothesis of export-led growth.The results do not change independently ofthe indexes of outward orientation used.

Henriques & Sadorsky(1996)

1 1870-1991Canada

Time series Real GDP growth Real export growth ADF unit roots,VARs, Johansen’sprocedure, Grangercausality test

Terms of trade No support for the export-growth hypothesisbut failed to reject it.

Al-Yousif (1997) 4 1973−1993Arab Gulfcountries

Time series Real GDP growth Real growth ofexports and exportchange/ output

ADF unit roots tests,White test,production function

Labour force and GDI/GDP Evidence that supports the hypothesis in theshort run; however, it fails to find any long-runrelationship, i.e. does not find cointegration.

Islam (1998) 15 1967−1991NICs of Asia

Time series Real GDP growth Export growth andexportchange/output

ADF unit roottests, Grangercausality test, errorcorrectionmodel

Imports, government non-defence expenditures,trade orientation, investment,instability in exports earnings.

Evidence that supports the hypothesis in theshort-run but only in 5 cases was a long-runrelation (no cointegration) found.

Shan & Sun (1998) 1 1978(5)−1996(5)China

Time seriesmonthly

Real industrialoutput

Export growth Ad hoc productionfunction, VAR

Labour force, investment andenergy consumption

Indicates a bidirectional causality betweenexport and real output. Therefore, theexport-led hypothesis defined as aunidirectional causal ordering from exports togrowth is rejected.

Begum & Shamsuddin(1998)

1 1961−1992Bangladesh

Time series Real GDP Export growth andexportchange/output

OLS, VAR productionfunction, MLEestimation and archmodel

Labour force, GDI/GDP,dummy and trend

Supports the hypothesis.

9

Sources: Based partially on the studies of Balassa (1985), Greenaway & Sapsford (1994) and Shan and Sun (1998). a Number of countries included in the study.

10

Similarly, the empirical results from theanalysis of DCs do not confirm exportexpansion as being significant. For example,Catão (1998) has analysed the case of Mexicoduring the period 1870−1911. Using a new setof macroeconomic data, the author shows thatthe country’s rate of growth in that period wastwice as fast as its historical trend andcoincided with a substantial expansion ofexports; but he indicates that the size of theexport sector was very small and had weaklinkages with the rest of the economy. Thus, itis unlikely that exports could have propelled thedomestic sector of the Mexican economy, asmany researchers have claimed in the past.

In general, these empirical studiesregarding the relationship between exports andgrowth can be separated into two categories.The first type of empirical investigation focuseson cross-section analysis, and the secondpoints to country-specific studies.

A. Cross-section analysis

The first group has employed thegrowth of exports as a proxy for policyorientation in order to judge the advantages anddisadvantages of different types of tradestrategies, mostly the inward strategy asopposed to one with an outward orientation.Some studies have combined cross-sectionanalysis with time series (see e.g. Ram, 1987).Most of these studies published in the late1970s found a significant positive relationshipbetween export performance and the growth ofnational income. Balassa (1980, p. 18)summarized them, stating that “The evidence isquite conclusive: countries applying outward-oriented development strategies performedbetter in terms of exports, economic growthand employment than countries with continuedinward orientation”.

Many of the earlier studies, whichinclude Syron and Walsh (1968), Kravis(1970, 1973a, 1973b) Michaely (1977), Hellerand Porter (1978) and Balassa (1978), claimedthat these positive effects flourish only aftercountries have achieved a certain level ofeconomic development. Consequently, theirresults indicate that nations heavily dependenton agricultural commodities are less likely tobenefit from exports, in comparison withcountries that have a higher level ofdevelopment and whose exports contain ahigher domestic value added (see e.g. Kohliand Singh, 1989).

Although such empirical investigationscan explain to some extent why growth differsacross a wide spectrum of countries, this typeof cross-section investigation has severaldeficiencies, which raise doubts about theirusefulness.

The first deficiency is that these studiesdo not provide any useful country-specificinformation to policy makers in DCs. Byassuming the same production function acrossdifferent types of economies they do not takeinto account the level of technology, which islikely to differ across countries. Therefore, theempirical results obtained are averages that donot capture the particularities of manydeveloping countries. Second, those results areoften disputed because of the limited size oftheir samples. Most of these investigationsincluded fewer than 12 countries (see e.g.Balassa, 1978; Bhagwati, 1978; Krueger,1978; Chow, 1987, 1989). Third, even thosestudies in which the sample was larger werelimited to specific types of DCs, i.e. mostresearchers chose a priori middle-incomecountries and excluded low-income countriesand major oil exporters (see e.g. Feder, 1983;Kavoussi, 1984).

11

Because of the use of this strategy theempirical results reported in the economicliterature based on cross-section data areclearly susceptible to criticism from analysts oflow-income nations such as China and India,and especially those that study major petroleumexporters. It is obvious that such results cannotexplain the effects of different trade strategies,and in particular the importance of the exportsector and its performance, on the rate ofgrowth of many DCs.

The exclusion of oil exporters, inparticular those that are members of theOrganization of the Petroleum ExportingCountries (OPEC), from cross-section studieshas been highly arbitrary, since most of thesestudies included in their samples middle-incomecountries which are also highly dependent onexports of primary products and particularlyminerals. Examples are abundant: most of theseinvestigations include countries such asBotswana (diamonds), Chile (copper) andSouth Africa (gold), which still depend to alarge extent on exports of minerals to financeimports, these exports representing a largeshare of total public revenues.

All of these economic activities havefour distinctive common characteristics. First,the export sector is highly capital-intensive;second, the ownership, management andtechnology were frequently under foreigncontrol; third, the export sector is consideredan “enclave” which has limited linkages with thedomestic economy (although notdemonstrated); and fourth, they basicallyexport non-renewable natural resources withlow value added.

One of the most popular empiricalreasons for excluding major oil exporters isappraised by Tyler (1981, p. 124), who arguesthat the “statistical relationship is stronger”when major oil exporters, such as the OPEC

members, are omitted from the sample incross-section investigations. This positionindicates clearly a prejudice against petroleumexporters that does not have solid support ineconomic theory. However, it could beinterpreted also as a tacit recognition that thisgroup of countries are special cases amongDCs, which have to be studied separately andif possible in a country case framework.

B. Country specific studies

Although, like several other authors,Caves (1971, p. 424) stated many decadesago that “Tests of the export-led model, then,must intrinsically involve country case studies present industrial countries, or now-wealthynations in their years of rapid growth, or ofpresently underdeveloped countries ”, thissecond type of examination is still less frequentin the literature. In fact, it is only during the1990s that a modification has started to occur,for both developing and industrialized countries.These investigations have examined theconnection between export performance andthe rate of economic growth in particularnations, in some cases using modern time seriesanalysis (see e.g. Khan and Saqib, 1993;Serletis, 1992; Henriques and Sadorsky, 1996;Al-Yousif 1997; Begum and Shamsuddin,1998).

While the results that emanate fromcross-section studies, based on bivariatemodels or ad hoc aggregate productionfunction, generally agree on the positiverelationship between export performance andeconomic growth, it is odd that the empiricalresults obtained by researchers involved incountry case studies strongly differ betweennations and periods of time studied (see e.g.Shan and Sun, 1998). This disparity mightimply that although cross-section studies areempirically attractive for researchers, they

12

could obscure intercountry differences andsacrifice revealing information about thebehaviour of many countries.

It is clear, therefore, that cross-sectionstudies might be an unreliable source ofknowledge for scholars and policy makers,especially in DCs.

Finally, we come to the issue ofcausality and in particular whether there isempirical evidence that exports and economicgrowth have a common trend in the long run(see e.g. Chow, 1987,1989; Sephton, 1989).

The most recent time seriesinvestigations concerning DCs that have usedthe econometric methodology of cointegrationhave not been able to establish unequivocallythat a robust relationship between thesevariables indeed exists in the long term, namelythat the variables are cointegrated (see e.g.Islam, 1998). While some have been able tofind a long-run relationship, many others haverejected the export-led hypothesis, i.e. thatexport expansion causes growth in the longterm. In fact, in most studies the results suggestthat this arises owing to a simple short-termrelationship, a feature that is not surprising if wetake into account the fact that the studies thathave concentrated their attention onindustrialized nations have also been unable tofind a robust relationship between thesevariables (see e.g. Kugler, 1991).

Al-Yousif (1997) attempted to remedythe lack of empirical evidence related to majoroil exporters by analysing four Arab Gulfcountries: Saudi Arabia, Kuwait, the UnitedArab Emirates and Oman, which are allmembers of OPEC. As in other empiricalstudies in this field, he was unable to verify theexistence of a long-term relationship betweenexports and economic growth in the four majorpetroleum exporters of the Persian Gulf. Thus,

one tentative explanation could be that theirexports are highly concentrated on oil andpetroleum derivatives; thus, exports, terms oftrade and government expenditure tend to havevery similar patterns in countries that are greatlydependent on the export of a single mineral orraw material which, moreover, is mostly ownedand managed by the State. By the ad hocinclusion of five variables in the augmentedproduction function, three of which are highlycorrelated, the model might have been mis-specified, and this could have distorted theresults. However, as mentioned before, theseresults are not significantly different from othersthat have been published recently, as shown intable 1.

There are very few time series studiesconcerning Latin American countries whichhave used modern econometric techniques, andan augmented neoclassical production functionas a theoretical framework.

In the case of Chile, Figueroa de laBarra and Letelier-Saavedra (1994), usingquarterly data, were able to corroborate theexistence of a long-run relationship betweenexports and growth independent of the indexemployed. Equally, Van den Berg and Schmidt(1994) found cointegration in 11 of the 16LACs analysed. In fact, in the case of CostaRica they were able to verify the existence of along-term relationship. Although the resultsseem to endorse in general the export-ledhypothesis, they seem to deviate from thoserecently reported by the empirical literature(Rodrik, 1999).

However, a possible justification of thepositive results obtained in the investigationconducted by Van den Berg and Schmidt(1994) is that these researchers employedpopulation and investment as proxies for theappropriate aggregate inputs, i.e. labour forceand capital stock. Although they have been

13

widely used in many cross-section growthstudies as well as time series analysis (see e.g.Al-Yousif, 1997), many researchers have hadserious doubts about them and have thusexpressed their suspicion regarding studies thathave tested the export promotion hypothesis byusing, for instance, the investment−output ratio,i.e. gross domestic investment (GDI)/grossdomestic product (GDP), as opposed tocapital stock or population instead of labourforce.

For instance, Alexander (1994) amongothers, rejected the use of these proxies ingrowth studies not only on theoretical groundsbut also from an empirical point of view, andsuggested that if capital stock data areavailable, they should be used instead ofinvestment because of the “significantmeasurement errors” present in these types ofempirical growth studies. However, if dataregarding the stock of capital are not available,a common recommendation nowadays is toconstruct a series of capital stock (Khan andSaqib, 1993).

Even though this is a sensible andlogical strategy, the basic constraint thatresearchers have encountered when trying toconstruct a series of capital stock for DCs isthe non-existence of two vital sets ofinformation: the initial base year for the capitalstock and the rate of depreciation.

The use of population as a proxy forlabour force is based on very strongassumptions concerning the rate ofunemployment, the participation rates and thesignificance of the underground economy.Although in the case of many DCs all thesesuppositions could be considered unrealistic,they can be defensible, particularly when theseries of population employed as well as labourforce is not available for the entire period underinvestigation or exists only for a limited period.

C. Summary of the empiricalliterature

From the review of the empiricalliterature on exports and growth since the late1960s, which is summarized in table 1, it isclear that the recent evidence available suggeststhat exports do not necessarily cause growth,as many economists believed and maintaineduntil recently and as early studies suggested.

The results reported are clearlysensitive to the variables employed, e.g.investment instead of capital, population insteadof labour force, and also to the theoreticalframework assumed, i.e. bivariate models andad hoc production functions instead of anaugmented neoclassical production function.

Although an augmentedCobb−Douglas production function could beconsidered ad hoc, we can tackle this issue byconstructing a simple two-sector growth model,which is based on the following assumptions.First, the economy is composed of two sectors,each of which produces a single good. One isa tradable good and the other is non-tradablemerchandise; that is, the first one is producedfor the foreign market, while the second isentirely for the domestic market. Second, bothsectors demand inputs from the economy,essentially labour and capital. Third, there aresignificant productivity differences between thetwo sectors. Fourth, the production of thedomestic sector (non-export sector) dependson the volume of exports. This type of modelhas been widely used since Feder (1983) firstpresented it. It focuses on the likelihood ofnon-optimum allocation of resources due to adifferential of productivity between the twosectors and where exports can capture a rangeof positive spillovers and externalities which arenot measured by the conventional national

14

accounts.

From the voluminous literature on therelationship between export expansion andeconomic growth that is summarized in table 1,it is clear that the results obtained depend notonly on the theoretical approach used but also even on the econometric

methodology employed. For example, cross-section studies are more likely to corroboratea positive relationship between exports andgrowth, while the results of time series studiesdepend substantially on the countries analysed,

the period chosen and the econometric methodused. In addition, since cross-section studiescan obscure particularities of DCs, especiallythose that are low-income countries as well asmajor oil-exporting countries, the correctstrategy to follow from an empirical point ofview is to address the issue in a country caseframework, using as much as possible therecent developments in time series analysis.

15

IV. CASE STUDY

There are four main reasons forchoosing Costa Rica as a case study. First, asufficiently long series of macroeconomic datais available. Second, during the period underinvestigation, the country has had an enviablerecord of political stability among DCs;therefore, the political factor can be excluded apriori from the analysis. Third, exports are tosome extent diversified and the country doesnot depend on exports of minerals. Lastly, thecountry is considered to some extent a successstory among LACs because of the systematicincrease in GDP and GDP per capita, whichhas led to substantial improvements in mosteconomic and social indicators. Severalquestions therefore arise. What were the mainengines of growth? What was the role playedby exports during the second part of thetwentieth century? Furthermore, how did apoor and backward country that experienceda violent civil war in 1948 become the mostsuccessful country in Latin America during thesecond part of the twentieth century?

A. Variables and data sources

The data are derived from bothnational and international statistical yearbooks.The principal national source was the dataavailable from the Banco Central de CostaRica (BCC) through publications such asActualidad Económica and Evolución de lasPrincipales Variables Macroeconómicas.The principal international source of data wasthe International Financial Statistics (IFS)published by the International Monetary Fund(IMF). In addition, there were the WorldTables, Global Development Finance(formerly known as World Debt Tables),World Development Indicators and World

Bank Atlas, published by the InternationalBank for Reconstruction and Development(IBRD). Other international sources used in thisstudy include the International TradeStatistics Yearbook issued by the UnitedNations (UN) and the Statistical Yearbook ofLatin America and the Caribbean publishedby the UN’s Economic Commission for LatinAmerica and the Caribbean (ECLAC).4

The data used in this analysis have anumber of limitations, and they should behighlighted. First, the sample period is limited to1950−1997 because of the non-availability ofofficial national account data prior to thisperiod. Consequently, the estimates obtainedusing some of the current econometrictechniques have some limitations that must betaken into account.

Second, owing to the shortage ofreliable quarterly data for most of the variablesunder consideration for the entire period, theperiodicity of all the data used in thisinvestigation is annual.

Third, because of the inherentdifficulties in measuring the stock of physicalcapital (KT), the lack of official and credibleseries of aggregated and disaggregated termsfor the period studied restricted the inclusion ofcertain variables and limited the testing ofcertain models and hypotheses. Thus, onestrategy would have been to construct a capitalstock series; however, for that task we needtwo basic sets of information that to ourknowledge do not exist: the initial base year forthe capital stock and the rate of depreciation.

4 All the data used in this study are available from theauthor upon request.

16

Therefore, the only plausible strategy at thisstage to overcome these obstacles was to usedata related to investment, specifically GDI andgross fixed capital formation (GFCF), atcurrent prices in millions of Colons, takenmainly from data published by the BCC. It isimportant to note that this strategy has beenwidely used by researchers engaged in testingthe ELGH for both cross-section and countrycase studies of DCs and even for industrializednations.5

Fourth, the level of prices was obtainedfrom the deflator of the GDP index or theimplicit deflator of GDP. This uses 1990 as thebase year and it was taken from the IFS (line99bi.p). It is constructed by the BCC by takingthe ratio of GDP at current prices (line 99b)and at constant prices in millions of Colons,which are also published by the IFS (line99b.p).

Fifth, the information related to thelabour force comes in the first instance fromseveral national censuses and surveys. In thisstudy the figures regarding employmentgathered by national sources were notconsistent between different publications, andtherefore in order to employ a consistent series,the statistics were taken primarily from thelatest Yearbook of Labour Statisticspublished by the International LabourOrganization (ILO), and were evaluated bycomparing them with the data constructed bythe World Bank. Some estimations were madefor the period 1960–1997, using the labourforce series published by the World Bank.6

Unfortunately, neither the breakdown of thelabour force nor the statistics regardingemployment within the economy was

5 Table 1 shows in the column "Other variables" thedifferent variables used by researchers as proxies forthe rate of physical capital accumulation.6 These results are available from the author uponrequest.

obtainable for the period under investigation;therefore, we decided not to use the labourforce series and relied on population for thisinvestigation and used as a proxy.7

Although this procedure could beconsidered by some unrealistic from aneconomic point of view, it can be defendedfrom an econometric point of view. If we takeinto consideration the limited size of the bothseries: labour force (1960−1997) andemployment (1976−1997), i.e. 38 and 22observations respectively, this would raiseserious questions concerning the robustness ofthe empirical results obtained throughcointegration tests, which are extremelysensitive to finite sample sizes.8

7 The first population census carried out in thesecond part of the twentieth century was in 1950.Since then several censuses have been carried out bythe Dirección General de Estadísticas y Censos in1963, 1973, 1984 and, most recently, in 1994.Unfortunately, the surveys regarding employmentonly started to be organized systematically in 1976. Itis important to note that Costa Rica's rate of naturalpopulation, which was already high up to 1950,increased even more during the period 1950−1970 byapproximately 4 per cent, a figure which issubstantially higher than the average populationgrowth in Latin America (2.7 per cent) with theexception of Venezuela. This was because these twocountries experienced an earlier start in the reductionof mortality than other DCs, while the birth rateoverall (43 per 1,000) remained the same as in Asia. Inaddition, in both countries an open-door immigrationpolicy was introduced after the Second World War.However, after 1975 the population growth rate inCosta Rica dropped to 2.5 per cent and during the1990s it decreased even more, reaching 2.1 per centper year (Collier et al., 1992).8 Furthermore, in many of the studies mentioned inthe review of the literature, the average rate of growthof the population has been included as a proxy forlabour growth. This is especially important when theresearcher has considered that the data regarding thelabour force is unreliable or is simply not available.The disadvantages of using population growth (∆p)in this particular case are similar in relation to otherstudies concerning DCs. As a result, it is important tobear in mind that the use of population in an empiricalstudy such as this could result in overestimating the

17

We now turn our attention to theproblem of how the period for the estimationswas chosen and the ultimate sample size usedto estimate the model. A priori, there were twooptions for selecting the period: one wasstraightforward and consisted in using thewhole sample period available (1950−1997),and the other was to focus on a specific periodwhich had a substantial and distinctiveeconomic and, possibly, political regime.

Although Costa Rica had timidlypromoted industrialization since the 1940s, itwas not until the early 1950s that its effects onthe entire economy started to operate. Incommon with the rest of Central America aswell as the rest of Latin America, the countrypursued import-substitution industrializationafter the Second World War as a developmentstrategy. Successive Governments quicklystarted to offer incentives for the establishmentof industries inside the country through variousmechanisms, such as tariffs, subsidies, and localand national tax concessions, all of which werean integral part of a broad and aggressive ISSto protect so-called infant industry frominternational competition.

The ISS was kept in place bysuccessive Governments until the early 1980s,when a newly elected Government was forcedto implement a severe adjustment programmein order to correct major macroeconomicimbalances which were clearly evident by1981. However, it is important to state thatmost measures were incremental rather than thetypical shock therapy that most of LatinAmerica had to pursue. Since then successivepolicy-makers have embraced the exportpromotion (EP) strategy slowly but steadily. Anexample of their shift of development effortstowards export expansion was the

contribution of labour as a factor of production to therate of economic growth.

implementation from the mid-1980s of the so-called export contract system, which has beenused by the State since then to promoteexports.9

However, it is clear that the residue ofthe ISS endured through the early 1990s anddiminished at a very slow rate. An interestingfact to take into account is that Costa Rica isnow the most industrialized country in CentralAmerica and during the entire period studiedwas characterized by an enviably stabledemocratic system. Consequently, the period1950−1997 was used to estimate severalmodels, which coincided to a great extent withthe epoch of the ISS and was characterized bya stable democratic system.

Finally, it appropriately to mention thatall the empirical estimations in this study werecarried out using the time series econometricpackage Microfit 4.0, developed by Pesaranand Pesaran (1997).

B. Methodology and results

Prior to testing for a causal relationshipbetween the time series, the first step is tocheck the stationarity of the variables used asregressors in the models to be estimated. Theaim is to verify whether the series had astationary trend, and, if non-stationary, toestablish orders of integration.

For this purpose, all the variables areexamined through graphical inspection of theirtime series plots. The variables are real grossdomestic product (y), real export of goods andservices (x), real gross domestic investment or

9 The system was introduced in 1984 as theGovernment's principal instrument to promoteexports, particularly to extraregional markets (Wu andChuang, 1998).

18

real gross fixed capital formation (i), and theseries of population (p).

All the series are expressed inlogarithms and annual rates of growth of all thevariables are approximated by first differencesof the logarithms of the corresponding variablevalue of successive years.

All the variables were transformed toconstant prices, with the obvious exception ofpopulation, by using the GDP price indexreferred to in the previous section.

The plots of the variables underscrutiny are presented in Figures 1.1 to 1.10 inthe appendix. The inspections of all thevariables in levels clearly suggest that the seriesare linearly trended and, given that eachvariable seems to have a non-constant mean, itappears from the graphs that they are notstationary in levels, i.e. their distributiondepends on time.

Subsequently, the plots of the variablesin first differences, in contrast, show noevidence of trending time series, different meanvalues at different points in time or considerablechanging variances. The visual evidenceprovided by the diagrams is consistent with thevariables being integrated at an order of 1denoted as I(1).

Although graphical evidence is useful asa first approximation to decide whether thevariables are non-stationary, mosteconometricians agree that this is clearly anunreliable method to use to make inferences

about unit roots and, therefore, at this stage weturn to the formal testing procedures currentlyavailable in order to examine each of thevariables under scrutiny (see e.g. Harris, 1995).

1. Testing for unit roots

To test the level of integration of thevariables that will be employed in the growthequations, the well-known Dickey−Fuller (DF)and the augmented Dickey−Fuller (ADF) testsare used. The aim is to determine whether thevariables follow a non-stationary trend and arein fact of the order of 1 denoted as I(1) orwhether the series are stationary, i.e. of theorder of 0 denoted as I(0).

First, if the series are non-stationarythe use of classical methods of estimation suchas OLS could lead us to mistakenly acceptspurious relationships, and thus their resultswould be meaningless.

Second, in cases where the series arenon-stationary around their mean, thetraditional suggestion was to differentiate theseries. This usually leads to stationarity,allowing the researcher to apply conventionaleconometrics (Granger and Newbold, 1974).

However, first differencing is certainlynot an appropriate solution to the aboveproblem and has a major disadvantage: itprevents detection of the long-run relationshipthat may be present in the data, i.e. the long-runinformation is lost, which is precisely the mainquestion being addressed.

19

Table 2a. Unit root tests Time period 1950−− 1997Variables in levels(in natural logarithms)

DFa ADF(3)a

GDP (LY) -0.77305 -0.93825Population (LP) -2.1639 -2.1218Gross domestic investment (LGDI) -2.1721 -0.98142Gross fixed capital formation (LGFCF) -1.3986 -1.2207Exports of goods and services (LXGS) -2.7721 -1.7953

Table 2b. Unit root tests Time period 1950−− 1997Variables in first differences (Rates of growth)

DFb ADF(3)b

GDP (DLY) -4.9505(***) -3.3436(**)Population (DLP) -6.4818(***) -3.3894(***)Gross domestic investment (DLGDI) -8.3884(***) -3.3722(**)Gross fixed capital formation (DLGFCF) -6.9170(***) -3.2473(**)Exports of goods and services (DLX) -7.7139(***) -4.1579(***)

Note: The number of lags included in both tests was 3.a The tests include a constant (intercept) and a linear trend.b The tests include a constant (intercept) but not a trend.

* Significant at a 10% level.** Significant at a 5% level.*** Significant at a 1% level.

Tables 2a and 2b present the results ofboth tests, namely the DF and the ADF. Theresults obtained provide strong evidence that allthe time series in levels are non-stationary,which means they are integrated at an order of1, i.e. I(1) at the 95 per cent confidence level.Thus, they have a stochastic trend and theyindicate that the null hypothesis cannot berejected for any of the variables under scrutiny.In addition, when taking first differences, thetests strongly reject the unit root, which meansthat they are integrated at an order of 0, i.e.I(0) at the 95 per cent confidence level, whichmeans that they are stationary.

However, it is important to note that inall cases, irrespective of the order of theaugmentation chosen for both tests, the DF andthe ADF statistics are all well below the 95 percent critical value in table 2a or above in table2b respectively.

The results of the unit root testsperformed corroborate previous findings in the

empirical literature, i.e. as with mostmacroeconomic series, the variables underconsideration in this study appear to be non-stationary and trended in levels. Only their firstdifferences are stationary. Considering that thedata appear to be stationary in first differences,no further tests are performed.

Since the series are I(1), the use oftraditional econometric techniques such as OLSand the use of tests such as t-tests and F-testscan lead to mistaken (false) acceptance ofspurious relationships between the variables.Actually, these regressions produce empiricalresults that are characterized by high levels ofR2, which suggests the existence of astatistically significant relationship between thevariables in the estimated model.

The spurious problem has otherentanglements; for instance, Phillips (1986)demonstrated that the DW statistics convergetowards zero, and thus equations that report

20

high R2 and low value of DW are typicalcharacteristics of spurious regressions.

Nevertheless, the only fact that inreality could emanate from this type ofestimations at this stage is the existence of acontemporaneous correlation between thevariables, rather than meaningful economiccausal relationships between them.

If, by contrast, the variables are foundto have been stationary, it is not necessary toproceed to testing for cointegration sinceclassical regression methods of estimation suchas OLS are appropriate and can be applied tostationary variables in levels. Ultimately, if thevariables are found to be integrated at differentorders, it is possible to conclude that varioussubsets of variables under consideration maybe cointegrated (only where there are morethan two variables under consideration).However, further analysis would obviously berequired to test this conjecture.

The contribution of Engle and Granger(1987) was to demonstrate that although theindividual series could be non-stationary, i.e.they are I(1), like those previously examined, alinear combination of them might be stationary,i.e. I(0).

Consequently, the next section of theempirical study investigates whether the seriesunder scrutiny are cointegrated, so that a well-defined linear relationship exists among them inthe long run. Thus, we proceed to test forcointegration between the variables on levelsusing several tests, all of which are based onthe null hypothesis of no cointegration.10

2. Cointegration

10 Although tests with cointegration as nullhypothesis do exist, they have not been widely usedin the empirical literature (see Maddala and Kim, 1998p. 205−210).

Although finding cointegration inempirical studies is not a frequent result, it isone that has attracted the greatest attentionamong applied econometricians andmacroeconomists. This implies that if we wishto estimate the long-run relationship betweenthe two variables, let us suppose, yt and x t, it isnecessary only to estimate a static model suchas the equation (1.1) or (1.2) and checkwhether the residuals ε t from the regression arestationary, i.e. I(0).

Taking into account that both DF andADF tests suggest that all the variables appearto be integrated at an order of 1, i.e. I(1), andthus have a stochastic trend, and in additiontheir changes or first differences appear to bestationary, they are all candidates for inclusionin a long-run relationship concerning theinterdependence between exports and output,using as a theoretical benchmark an augmentedneoclassical production function. Thus, the aimat this stage is to test whether these variablesare indeed cointegrated.

Not only has the economic literatureadopted a supply-side approach as the basicframework to test empirically the relationshipbetween export and growth, but also nearly allthe studies mentioned in the review of theliterature have specified a linear relation.Consequently, we will follow this strategy andin the first instance estimate a simple Cobb–Douglas production function using a linearequation of the following form:

yt = φ0 + φ1 pt + φ2 it + µt (1.1)

where yt, pt, it are real GDP, population, realGDI (subcase a) or GFCF (subcase b) as aproxy of the stock of physical capitalrespectively. Subsequently, we estimate anaugmented Cobb−Douglas production function,such as the following equation, which includes

21

real exports of goods and services denoted byx t. As usual, all the variables are expressed innatural logarithms:

yt = φ0 + φ1 pt + φ2 it + φ3 xt + µt (1.2)

The results are obtained after estimatingequations in levels using two alternativespecifications of the so-called static orcointegrating regression that employed GDIand GFCF, such as equations (1.1) and (1.2)through OLS which are shown in tables 3a and3b.

It is extremely important to note that,with the exception of the adjusted R2 and theDW statistics, the customary diagnostic testshave not been reported. In addition, eventhough the φi coefficients reported in thefollowing table could be interpreted asapproximations of partial elasticities, they donot provide any kind of basis for sensible andvalid inferences at this stage. Furthermore, theycannot be used to draw any kind of inferenceswithout confirming a priori that the variables arein fact cointegrated. Even if the variables are infact cointegrated, although the estimatesobtained through the cointegration regressionusing OLS are “super-consistent”, i.e. theestimates of φ obtained converged faster thanin the case of OLS models using stationaryvariables, the estimated standard errors are not(Stock, 1987). By contrast, if the variables arenot cointegrated, the results are meaninglessand show only a “spurious correlation” that hasno economic significance.

Although from a theoretical point ofview the appropriate investment variable isGFCF, in this case we decided to estimateboth specifications by using both variables, i.e.using GDI and GFCF for the entire period

(1950−1997).11 The results are shown in tables3a and 3b, which set out the basic results andin addition include two cointegration tests,namely the CRDW and the EG.

The two cointegration tests are singleequation methods amongst various residuals-based tests which have been proposed in theeconometric literature since the mid-1980s.They are obtained after estimating the equationsin levels using the two alternative specificationsof the so-called static or cointegrationregression that employed GDI and GFCF.

In all four cases, independently of thespecification taken into account and theinvestment variable employed, the CRDWclearly exceeds the value of 0.99, which is theapproximate critical value for n = 50 at the0.05 level of significance. Therefore, theCRDW test is able to reject the null hypothesisthat the variables are not cointegrated and, infact, the residuals estimated suggest that thevariables have a long-term relationship in allcases for the 1950−1997 period.

Using the EG cointegration test forequations concerning the neoclassical theory ofproduction, i.e. (1.1a) and (1.1b), the nullhypothesis of no cointegration can be rejectedat the 5 per cent significance level in one of twocases and can be easily rejected at the 10 percent significance level in the other one.

11 Conceptually, GDI includes inventories and thiscategory of investment clearly does not add tooutput; thus, its inclusion in a production function isquestionable and suggests that it is thereforepreferable to rely on models that use GFCF from atheoretical point of view.

22

Table 3a. Static long-run relationship (using GDI) Time period: 1950−− 1997Regressions (1.1a) (1.2a)

Dependent variable LY LY

Number of observations 48 48

Variables

C 6.3529 6.0285

LP 0.57646 0.50803

LGDI 0.52568 0.48373

LXGS 0.073454

Adjusted R2 0.99584 0.99594

DW-statistics (CRDW) 1.2587 1.0895Engle and Granger cointegration test

DFADF (2)

-5.1636 (***)-3.9368(**)

-4.8702 (***)-3.9924(*)

Table 3b. Static long-run relationship (using GFCF) Time period: 1950−− 1997Regressions (1.1b) (1.2b)

Dependent variable LY LY

Number of observations 48 48

Variables

C 7.1029 6.5397LP 0.82170 0.69559LGFCF 0.44872 0.41197LXGS 0.092371Adjusted R2 0.99780 0.99808DW-statistics (CRDW) 1.0382 1.0050Engle and Granger cointegration test

DFADF (2)

-4.2200 (**)-3.6131(*)

-4.5562 (**)-4.2268 (*)

Notes: Regressions (1.1a) and (1.1b) are based on the standard neoclassical framework (Cobb−Douglasproduction function).Regressions (1.2a) and (1.2b) represent the main case under scrutiny, which estimates anaugmented production function that includes exports as a third input of production.

* Significant at a 10% level.** Significant at a 5% level.*** Significant at a 1% level.

It is also important to mention that thesystem variables appear to be cointegratedindependently of the investment variable takeninto account, i.e. GDI or GFCF. Thus, bothtests suggest that a linear combination of theseries of output, population and investmentexists in the long term.

Furthermore, in the main case underscrutiny, the so-called ELGH, represented byequations (1.2a) and (1.2b), both cointegrationsub-tests are able to find evidence of a long-run

relationship between exports and output,independently of the investment variableemployed GDI or GFCF.

In general, evidence of cointegrationincludes high R2, "apparently" significantcoefficients,12 significantly non-zero CRDW

12 It is extremely important to note that because thevariables are non-stationary, the standard propertiesof OLS do not hold. Furthermore, because of theautocorrelation of the residuals, the t-statistics fromthe static long-run relationship are biased upwards

23

and significant DF and ADF tests on theresiduals from the static long-run regressions.Since all of them are present in all thespecifications shown in tables 3a and 3b, theevidence at this stage strongly suggests that acointegration relationship or relationshipsexist.

However, it is important to mentionthat although both CRDW and the EGprocedure have distinct advantages and in spiteof the positive results mentioned earlier, bothtests have several important defects. This issueemerged after several Monte Carlo studies thatconsidered the robustness of these tests (andothers not employed in this analysis) showedthat in general the most standard tests are notpowerful. Moreover, most of the studies cometo the conclusion that no one test predominatesover the others. Thus, the literature holds that itis very important for empirical studies to carryout several tests for cointegration instead ofusing one single procedure (Maddala and Kim,1998). In fact, in cases where the sample sizeis finite, the estimations conducted through theEG procedure are sensitive to the imposition ofnormalization.

Thus, before making any kind ofjudgement, some further cointegration tests areemployed to verify the existence ofcointegration, which will be shown in the nextsections.

In the following section, the Johansenprocedure will be briefly explained. Themethod is completely different from the

and it is thus not possible to determine at this stagethe true significance of the coefficient estimates.Nevertheless, if the variable is insignificant when theoriginal t-value is used, it is obvious that when the"true" or corrected values are employed the variableswill still be insignificant; thus, it is feasible toacknowledge the insignificance of the coefficients atthe levels stage.

previous ones because it is a multiple equationmethod where the objective is to identify thecointegration space which is based oncanonical correlation methods, a procedurewhich enables us to test how manycointegration relationships there are.

3. Johansen maximum likelihoodapproach

The Johansen procedure is a multipleequation method that permits the identificationof the cointegration space using a canonicalcorrelation method, which enables the testing ofhow many cointegration relationships exist.

To briefly illustrate it, let us define St =(yt, pt, it, x t), a vector of four elements (P = 4)and consider the following autoregressiverepresentation of St:

kYt = π0 + Σ πi Υt-i + ut (1.3) i =1where Γi, = − ( I − π0,,……...., π0) , andπ = ( I + π1,, ……….. , πk).

The Johansen procedure involvesestimating equation (1.3) by employing themaximum likelihood (ML) technique and testingthe null hypothesis of no cointegration, that isthat H0: (π = ψξ ) of r cointegratingrelationships, ξY t-i = ηit, and where r is therank of the matrix π (0 < r < P), ψ is the matrixof weights with which the variables entercointegrating relationships, and ξ is the matrixof the cointegrating vector. As stated in theprevious section, this procedure could lead usto find up to three independent cointegratingvectors.

The null hypothesis of no cointegrationbetween the system of variables is rejected

24

when the estimated likelihood ratio testsstatistic, φ i, exceeds the critical value, where

pφ i = – n Σ ln (1- λi) (1.4) i = r + 1

The Johansen ML technique hasseveral distinct advantages in comparison withthe EG method illustrated in the previoussection to test for cointegration. First, it is aninvariant test, which permits the existence ofcointegration between the system variableswithout imposing bias on the estimates. Thus, itdoes not assume somewhat arbitrarily thedirection of the regression, which may lead todifferent and misleading results. Second, it canidentify whether more than one cointegratingvector really exists. Third, it can also estimatethe long run or cointegrating relationshipsbetween the non-stationary variables using aML procedure. This last feature could be usefulfor comparing the estimates obtained with theones obtained using, for instance, the EG two-step procedure and the unrestricted errorcorrection model.

Summing up, the Johansen test forcointegration is a multivariate unit root testwhich estimates the cointegrating rank r in themultivariate case, and which is also able toestimate the parameters β of these cointegratingrelationships.

To test for cointegration this procedureuses two test statistics. The first is called themaximum eigenvalue test (λmax), which teststhe null hypothesis that there are r + 1cointegrating vectors versus the alternativehypothesis that there are r cointegratingvectors. The second, labelled the trace test, isemployed to test the hypothesis that there areat most r cointegrating vectors.

Even though Johansen and Juselius(1990) initially indicated that the first test mightperform better, the Monte Carlo experimentsreported by Cheung and Lai (1993, p. 326)suggest that regarding non-normality, skewnessin innovations has a statistically significant effecton the test sizes of both the trace and themaximal eigenvalue test. However, they statethat between the two Johansen procedures totest “for cointegration, the trace test showsmore robustness to both the skewness andexcess kurtosis in innovations than themaximum eigenvalue test”. Since there is notcomplete agreement among econometricians, inthis case we have preferred to be cautious andprudent, and report and rely on both sub-tests.

Before turning to the empiricalestimations, we had to determine the lag K ofthe vector autoregressive (VAR) model inlevels, which is a critical stage of the JohansenML procedure. The literature recommends theuse of the Akaike Information Criterion (AIC)and the Schwarz Bayesian Criterion (SBC) toselect the lag length of the VAR system, whichis achieved by minimizing the AIC and SBC. Inmost cases, incidentally, both criteria concur insuggesting the use of a VAR with a lag of 2,while in those few cases where the choicecriteria are different, we have decided to usethe one that suggests the smaller order. This isbecause if, for instance, we use a VAR of agreater order, i.e. 4, 5, or 6, we could betaking an unnecessary risk of over-parameterization, a situation which is moreacute in cases where the sample size is finitesuch as this one. Moreover, since the data areof annual periodicity, an inspection of theresults suggests that serial correlation is not aproblem when we set the order of the VAR at2.13

13 The results of the AIC and the SBC are notreported in this study.

25

Tables 4a and 4b contain the resultsobtained by the application of the Johansenprocedure to test for cointegration using a VARat an order of 2. The results correspond to theentire time period (1950−1997). The tests areperformed by the use of the so-calledaugmented production function, which includesexports in real terms and is represented by thespecification given by regressions (1.2a) and(1.2b).

Both tests the maximum eigenvalueand trace statistics are used to determine thenumber of cointegrating vectors (r), from which

it is concluded that the results support theexistence of two cointegrating relationships,which clearly could lead to interpretationproblems in this case.

However, Cheung and Lai (1993, p.326), among other researchers, have suggestedthat the critical values used to test the numberof cointegration relationships through theJohansen procedure can be misleading (see e.g.Enders, 1995). Therefore, corrections to thecritical values are strictly necessary whenapplied to sample sizes of 100 or smaller,typical of finite sample sizes.

Table 4a. Johansen cointegration tests Equation (1.2a)List of variables included: LY, LP, LGDI, LX, intercept Time period 1950−− 1997

Null hypothesis Alternative

hypothesis

Maximum

eigenvalue

test

Adjusted

95% critical

values

Trace

test

Adjusted

95% critical

values

r = 0 r = 1 36.3943 35.7095 77.9446 67.5537

r <= 1 r = 2 20.3606 27.8400 41.5503 44.0463

r <= 2 r = 3 15.7565 20.0463 21.1897 25.4905

r <= 3 r = 4 5.4332 11.5705 5.4332 11.5705

Table 4b. Johansen cointegration tests Equation: (1.2b)List of variables included: LY, LP, LGFCF, LX, intercept Time period 1950−− 1997Null hypothesis Alternative

hypothesis

Maximum

eigenvalue

test

Adjusted

95% critical

values

Trace

test

Adjusted

95% critical

values

r = 0 r = 1 35.2820 35.7095 80.2205 67.5537

r <= 1 r = 2 27.5330 27.8400 44.0385 44.0463

r <= 2 r = 3 12.7096 20.0463 17.4055 25.4905

r <= 3 r = 4 4.6959 11.5705 4.6959 11.5705

Note: Table 4a exhibits the estimations using GDI as a proxy of real capital stock, while table 4b showsthe estimations using GFCF.

The justification for this correction isthe fact that the Johansen cointegration tests are“sensitive to under-parametrization in the laglength”. This is due to the fact that the Johansenlikelihood ratio (LR) test for cointegration isderived from asymptotic results and statisticalinferences which might be inappropriate. Thus,

the ML procedure tends to over-reject the nullhypothesis of non-cointegration when this istrue. This is critical when the system beingestimated contains more than two variables (n)or many lags (k), and even more when thesample size is finite. Consequently, the normalcritical values based on the asymptotic

26

distributions are dubious, and the results areoften biased and are thus misleading in thatcointegration is too often found when they areused. Thus, in this case we have followed theadvice given by Cheung and Lai (1993) andhave adjusted the critical values, which areshown in table 4a and 4b.14

Although this situation is commonlyencountered by most applied economists,especially when using time seriesmacroeconomic data, it must be taken intoconsideration when final judgements are to bemade from the empirical results of this study.

Using the normal critical values, theexistence of two or even three cointegratingvectors is found. However, when using theadjusted critical values, instead of the normalcritical values, we are able only to find onecointegrating vector. That is, the null hypothesisof no cointegration can be rejected at the 5 percent significance level in all cases. It is alsoimportant to mention that the system variablesappear to be cointegrated, independently of thespecification taken into account and, moreover,in the period of estimation, a situation similar tothat encountered in other cointegration tests.That is, the results of the Johansen procedure

14 According to Cheung and Lai (1993, p. 322), theresponse surface estimation shows that the finitesample bias of Johansen’s tests is a positive functionof T/(T- np), where T is the sample size, n is thenumber of variables and p is the number of lags in theVAR. Since T/(T- np) is greater than unity for anyfinite sample size T, both tests the maximumeigenvalue tests and the trace tests “are seriouslybiased toward spuriously finding cointegration” toooften when using the critical values based onasymptotic results and statistical inferences. Thus, intables 4a and 4b, T = 48, n = 5 and p = 2 the criticalvalues for the entire period of estimations have beencorrected by 1.2631. However, there is no completeagreement in this respect; for instance, Doornik andHendry, amongst other econometricians, have raiseddoubts about whether this is the preferred correction(see Harris, 1995, p. 88).

seem consistent with the previous cointegrationtests.

The Johansen procedure can also beemployed to obtain long-run parameterestimates that could be used to compare theestimates obtained with the ones obtainedthrough the error correction models. However,in this case we prefer to rely on the errorcorrection approach because of the intrinsiclimitations of the Johansen procedure in smallsample sizes.

4. Error correction model andcointegration

The initial concept of this type of modelcan be traced back to the work done bySargan in the mid 1960s, who considered aclass of models subsequently to be labellederror correction mechanisms (ECMs).However, it was the work of David Hendryand his many collaborators during the late1970s and 1980s that popularized their useamong econometric practitioners and especiallyamong applied macroeconomists (see e.g.Enders, 1995; Maddala and Kim, 1998).

Almost at the same time themethodology pioneered by Granger (1981,1986), Hendry (1986) and Engle and Granger(1987), among others, opened a new channelfor testing for cointegration. The Grangerrepresentation theorem, broadly speaking,states that if a linear combination of variables isstationary or I(0), then the variables are said tobe cointegrated and can therefore beconsidered to be generated by an ECM.Consequently, they proved that ECM generatecointegrated series and that, to be expressedconversely, cointegrated series have an ECMrepresentation, which allows the short-termdisequilibrium relationship within an ECMframework (see e.g. Engle and Granger, 1991).

27

According to Hendry (1986, p. 204)one of the most important consequences of theseminal work of Engle and Granger (1987) hasbeen “thus reconciling the two approaches aswell as clarifying when level information couldlegitimately be retained in econometricequations”. Furthermore, this led in the 1990sto the development of cointegration tests basedon the Granger representation theorem, whichare based directly on ECMs.

The first method, which involvesestimating an ECM, is the Engle and Granger(EG) two-step procedure, which providesinformation about the short-term dynamicsresponses of the variables. The method isstraightforward and involves runningregressions using stationary time series I(0),which in this case are achieved by using firstdifferences of the variables (rates of growth)and including in the regressions as anexplanatory variable the lagged residuals fromthe levels regressions. This lagged term, RES(-1), is intended to capture the error correctionprocess.

Kremers, Ericsson and Dolado, amongother researchers, initiated and developed thesecond method during the early 1990s (seeKremers et al. 1992). This new approach totest for cointegration suggests the use of theECM test instead of the residual-based tests.The main reason given by these researcherswas that the residual-based tests, which areoften used to test for cointegration and whichhave been discussed, employed and judged inthis study in previous sections, had been foundto have low power. They argued that theseresidual-based tests (as well as the DF andADF test for testing unit roots) imposed acommon factor restriction by ignoring equationdynamics in the so-called cointegrationregression or static regression given by

equation (1.1) or (1.2), which carries over tothe second stage of the cointegration systems.

If the common factor restriction doesnot hold, i.e. the observed error dynamics arespurious because equation dynamics areignored, then the ECM-based tests can bemore powerful than the residual-based test,since the Granger representation theorem statesthat there is an error correction representationfor every cointegration relationship. Althoughthe procedure was initially suggested byKremers et al. (1992), Zivot and Banerjeeamong others, have recently highlighted newways of applying it (see Maddala and Kim,1998, p. 203-205).

Consequently, we first make use of theEG two-step procedure and subsequently weestimate all the long- and short-run parametersin the ECM in one single step. It is important tonote that in both circumstances OLS can beapplied. This is because in the first case all thevariables included are I(0) and therefore thestandard properties of OLS hold, and in thesecond procedure, although LYt-1, LPt-1,LGDIt-1, LGFCFt-1 and LXGSt-1 are I(1)variables, OLS can still be employed,particularly since we have verified that thevariables are cointegrated using other methods(CRDW, DF, ADF and Johansen), and thusthere is a linear combination of them that is I(0)(see e.g. Thomas, 1997).

In this case, we have preferred, forseveral reasons, to follow both methods insteadof using only one of them. First, even thoughthe EG two-step procedure is still by far themost popular method for testing forcointegration, it has been widely shown to bebiased in small sample sizes. This is due to thefact that the estimations of the initialcointegration regression using OLS are biased,which carries over into the second stage

28

through the disequilibrium errors andconsequently to the short-run parameters.

Second, estimating the ECM in onestep will produce an alternative set of estimatesto those previously obtained from thecointegration regression and, furthermore, willenable testing for the "true" significance of thevariables both in levels and in difference forms.In addition, some studies have suggested thatthe estimates obtained through this method aresuperior to those obtained through the EG two-step procedure, in particular when using smallsample sizes. However, a word of warning isnecessary at this stage: since we are dealingwith a small sample size, there is no assurancethat the two approaches the EG two-stepprocedure and the unrestricted error correctionmodel, also described as the general-to-specific approach will lead to the samemodel. Moreover, it is important to note that inboth ECMs an implicit assumption is made thatthe right-hand side variables of the model areexogenous.15

Table 5 shows the results of theestimations using the EG two-step method,utilizing first GDI and afterwards directly theseries of GFCF for the entire period(1950−1997). From the four different dynamicequations for growth reported in the table, themost important points that emerge from thisestimate are the following.

15 It is very important to note that if this assumptionholds, the parameters' estimates will be efficient andtheir distribution asymptotically normal. If so thiswould allow use of estimated standard errors todetermine the true significance of the coefficientestimates. However, if this is not the case, the resultswill be biased. Note that even though severalexogeneity tests have been developed and are widelyemployed, they were considered beyond the primeobjective of this study and therefore, not made use of(see e.g. Engle and Granger, 1991; Maddala and Kim,1998).

The first two columns show the resultsof a simple Cobb−Douglas production function,while the last two columns exhibit theregression results that are based on anaugmented production function that includesexports as a third input of production.

All the residuals from the four levelregressions estimated by OLS are included inlagged form and labelled as RES(-1), with theobjective of capturing the process by which theeconomic agents adjust their prediction errorsfrom the last period. This represents the short-term adjustment mechanism from theequilibrium point, which is always significant,regardless of the specification employed. Thesignificance of the lagged residuals providesstrong evidence of the adequacy of an errorcorrection framework.

This implies that an ECM existswhereby the economic agents adjust theirbehaviour to unanticipated changes in output. Inthis case, around 50 per cent on average of theadjustment is achieved during the first period.Additionally, when we check the DF and theADF tests from the residuals of all theregressions estimated based on the EG two-step procedure also provides further proofregarding cointegration, given that the DF isalways significant even at the 1 per cent level,which indicates once again that the variablesare co-integrated.

If an error correction mechanism existswhereby the economic agents adjust theirexpectations to unanticipated changes in output,then on the basis of the Granger representationtheorem, this also implies that the variables arecointegrated, and vice versa.

29

Table 5. Engle and Granger two-step procedureRegressions (1.1a) (1.1b) (1.2a) (1.2b)

Years (sample) 1952-97 1952-97 1952-97 1952-97No. of observations (n) 46 46 46 46Dependent variable DLY DLY DLY DLYVariables

C 0.011464(1.0420)

0.0041865(0.43941)

0.016741(2.3399)

0.5793E-3(0.064478)

RES(-1) -0.44974(-4.1114)

-0.51421(-4.2585)

-0.40272(-3.5696)

-0.54901(-4.3941)

DLY(-1) 0.30432(2.9381)

0.20460(2.2672)

0.28926(2.8427)

0.19127(2.2786)

DLP 0.26352(0.98850)

0.63423(2.6710)

0.62874(2.8690)

DLGDI 0.27070(6.3895)

0.24892(6.1021)

DLGFCF 0.31776(8.3014)

0.31525(8.7649)

DLX 0.073156(1.9099)

0.076611(2.4823)

Basic statisticsR2 0.56363 0.68254 0.58374 0.73438Adjusted R2 0.52106 0.65157 0.54313 0.70118RSS 0.034133 0.024831 0.032560 0.020777F-statistics 13.2391 22.0376 14.3739 22.1181DW-statistics 1.9954 1.9968 1.8200 2.0131Durbin’s h-statistics 0.021752 0.013908 0.84341 -0.054222

Diagnostic testsSerial correlation

LM versionF version

0.144480.12603

0.156820.13683

0.979130.86994

0.0315960.026807

Functional formLM versionF version

0.965880.85791

0.450580.39568

0.863050.76483

0.0548320.046543

NormalityLM versionF version

2.0391NA

0.36476NA

5.8982(***)NA

2.8031NA

HeteroskedasticityLM versionF version

0.0636760.060992

0.455390.43995

0.3247E-30.3106E-3

0.703310.68317

Serial correlation LMStatistics CHSQ(3)F-statistics (3, n)

1.76550.50555

2.97290.87520

1.90710.54785

0.291790.078733

Unit roots test for residuals DF

ADF (2)-6.7375(***)

-3.3008-7.1810(***)

-3.3438-6.0679(***)

-3.6511-6.6917(***)

-3.4846Notes: Regressions (1.1a and 1.1b) are based on the standard neoclassical equation based on the

Cobb−Douglas production function, while regressions (1.2a and 1.2b) are based on the augmentedneoclassical production function that includes exports as a third input of production. Also note thatwhile regressions (1.1a and 1.2a) use GDI as a proxy of real capital stock, regressions (1.1b and1.2b) are estimated using GFCF.

The normal t-values of the coefficients are in parenthesis.NA = not applicable.* Significant at a 10% level.** Significant at a 5% level.*** Significant at a 1% level.

30

In this case, the coefficients of the errorcorrection term always have the correct sign(negative), and it is statistically different fromzero, in terms of both magnitude andsignificance. The significance of the errorcorrection term might be suggesting at this stagethat the variables are cointegrated; however, itis important to note that the ECM test forcointegration requires a new set of criticalvalues to judge accurately for significance andthus to test for cointegration. Even so, since thecoefficients are significant at 1 per cent usingnormal t-value tables, this is considered apreliminary confirmation of the existence ofcointegration. In addition, when we check theDF and the ADF tests from the residuals of allthe regressions, the DF is always significant atthe 1 per cent level, which clearly indicatesonce again that the variables are cointegrated.

The coefficients of the variables are ingeneral significant and have the correct sign.16

They confirm that the short-term effects ofcapital and labour (proxied by population) andinvestment are extremely importantindependently of the specification taken intoaccount. The coefficient of population growth(DLP) has the greatest magnitude, followed bythe rate of growth of investment (both DLGDIand DLGFCF) and finally the rate of growth ofexports (DLX). However, when GDI is usedas the investment variable independently of theproduction function employed, the coefficient ofDLP is not significant, and it only becomesstatistically significant when the variableemployed is GFCF.

16 In this case, since the variables are stationary, thestandard properties of OLS do in fact hold; therefore,the t-statistics are unbiased, and thus it is possible todetermine at this stage the true significance of thecoefficient estimates and rely on the diagnostic testswhich are disclosed.

Although all the models suggest thatexports have a positive effect in the short termon growth, the magnitude is very small, afeature which is also present in the staticcointegration equation.

In general, evidence includes mediumhigh R2 that are close to 0.7, which suggeststhat the overall fit of all the regressions is quitegood and that they are able to explain up to 70per cent of the variability of the growth rates.Despite the fact that neoclassical specificationis good overall with some problems ofnormality in regression (1.2a), it is evident thatthe overall fit of the regressions is higher whenexports are included. In addition, the results ofthe diagnostic tests employed improve whenexports are included and are in general alsovery good, which suggests the importance ofthis variable. The tests apparently suggest noproblem of autocorrelation of first or higherorder, heteroskedasticity, normality orfunctional form. The only exception isregression (1.2a), which violates theassumption of normality as already mentioned;nevertheless, this problem is completelycorrected when a dummy for the year 1981 isincluded (see table 7).

Now we start from an unrestrictederror correction model and test down themodel using the general-to-specific approach.Among the four different specifications forgrowth reported in table 6, the most importantaspects of these estimates are the following.

The findings are favourable when usingboth categories of investment (GDI and GFCF)in the so-called neoclassical productionfunction, but they improve when exports areincluded and, in particular, when using GFCF.In fact, all the coefficients of the variables inlevels LYt-1, LPt-1, LGDIt-1,

31

Table 6. Unrestricted error correction modelRegressions (1.1a) (1.1b) (1.2a) (1.2b)

Years (sample) 1952-97 1952-97 1952-97 1952-97

No. of observations (n) 46 46 46 46Dependent variable DLY DLY DLY DLYVariables

C 2.8689(4.0793)

3.7275(4.2617)

2.4046(3.5761)

3.7797(4.9059)

LY(-1) -0.42613(-3.9772)

-0.50106(-4.1515)

-0.39224(-3.7336)

-0.59369(-5.1975)

LP(-1) 0.24682(2.9524)

0.42497(3.9618)

0.17752(2.2063)

0.36723(3.9768)

LGDI(-1) 0.21146(3.1009)

0.13583(1.9440)

LGFCF(-1) 0.20960(3.5771)

0.20254(3.9092)

LXGS(-1) 0.081096(2.0555)

0.10775(3.5231)

DLP 0.48593(1.8572)

0.49363(2.2430)

DLGDI 0.23941(5.2774)

0.19834(4.5477)

DLGFCF 0.29052(6.8064)

0.26620(7.2638)

DLX 0.088455(2.1247)

0.088400(2.8567)

DLX(-1) -0.10622(-2.5978)

-0.10183(-3.0642)

DLY(-1) 0.22259(2.0043)

0.16970(1.7118)

0.32352(2.9348)

0.27610(3.0340)

Basic statisticsR2 0.59855 0.69995 0.68942 0.80670

Adjusted R2 0.54836 0.65379 0.62226 0.75838

RSS 0.031401 0.023469 0.024294 0.015120F-statistics 11.9276 15.1633 10.2663 16.6935DW-statistics 1.9057 1.9400 1.6906 1.9395Durbin’s h-statistics 0.48597 0.27504 1.5797 0.26077 Diagnostic testsSerial correlation

LM version F version

0.239450.20408

0.0246180.020347

2.94582.4631

0.1232780.094056

Functional form LM version

F version1.01770.88231

0.358030.29808

0.962270.76917

0.636620.49448

Normality LM version

F version3.1076

NA0.97266

NA11.1519(*)

NA1.6130

NAHeteroskedasticity


0.110420.10587

0.119050.11417

0.210070.20185

0.612390.59367

Notes: Regressions (1.1a and 1.1b) are based on the standard neoclassical equation production function,while regressions (1.2a and 1.2b) are based on the augmented production function that includesexports as a third input of production.

The normal t-values of the coefficients are in parentheses.NA = not applicable.* Significant at a 1% level.

32

LGFCFt-1 and LXGS t-1 are significant and havethe correct sign, independently of thespecification and the framework used, a featurethat is clearly desirable.

This evidence corroborates once againthe fact that, in the long run, labour (proxied bypopulation) and investment were the mainforces that drove the Costa Rican economyduring the second part of the twentieth century.Moreover, the coefficients are relatively stableindependently of the specification taken intoaccount. The coefficient of population (LP) hasthe greatest magnitude, followed by investment(LGDI or LGFCF) and finally exports(LXGS).

With regard to the coefficientscapturing the short-term effects, they showonce again that population growth (DLP) hasthe greatest magnitude, followed by investmentindependently of the variable taken into account(DLGDI or DLGFCF), and ultimately the rateof growth of exports (DLX).

The first two regressions, based strictlyon the neoclassical framework, include mediumhigh R2 that are between 0.6 and 0.7, whichsuggests that the overall fit of the equations isfairly good and that they are able to explain upto 70 per cent of the growth of output. Despitethe fact that neoclassical specification is goodoverall, it is obvious that the overall fit of theregressions improves when exports areincluded with R2 that are between 0.7 and 0.8,which suggests once again the importance ofthis variable, in terms of both significance andmagnitude. Additionally, all the diagnostic testsused are in general also very good, whichindicates that there is no problem ofautocorrelation of first or higher order,heteroskedasticity, normality or functional form.

The only exception is regression (1.2a), whichviolates the assumption of normality.Nevertheless, it is extremely important to statethat this problem is also corrected when adummy for the year 1981 is included (see table7).

The economic justification for theinclusion of a dummy variable rests entirely onthe economic crisis that started to severelyafflict the Costa Rican economy in 1981. Thissituation forced a newly elected administrationto implement a harsh adjustment programme in1981 and, therefore, the dummy (DUM81) thatis included in the estimations affects the shortterm not the long term. However, it is importantto note that the coefficients of the variables donot change significantly in terms of magnitudeand their significance, a feature that suggeststhat the dummy variable is capturing in realitythe negative short-run effects that the economystumbled upon during that year.

In addition, the estimation results showthat all the coefficients of y1,t-1 are statisticallydifferent from zero in terms of both magnitudeand significance. The significance of thecoefficients suggests once again that thevariables are cointegrated across differentequations, independently of the specification.

As mentioned earlier, even thoughthere is no guarantee that both procedures willlead us to the same model, in this case there aredefinite similarities that should be highlighted.From both procedures it is obvious that thecoefficient of DLP becomes insignificant whenthe investment variable employed is GDIindependently of the specification, and it onlybecomes statistically significant when GFCF isused as the investment variable.

33

Table 7. EG two-step procedure and unrestricted error correction model with DUM81Regressions (1.2a-D) (1.2a-D)

Years (sample) 1952-97 1952-97

No. of observations (n) 46 46Dependent variable DLY DLYVariables

C 0.019333(3.2911)

1.4847(2.6876)

RES(-1) -0.28684(-3.0019)

LY(-1) -0.29359(-3.5038)

LP(-1) 0.059281(0.88939)

LGDI(-1) 0.11876(2.1872)

LXGS(-1) 0.076829(2.5098)

DLP

DLGDI 0.22789(6.7728)

0.20005(5.9136)

DLX 0.18157(4.6061)

0.19286(5.0300)

DLX(-1) -0.097382(-3.0659)

DLY(-1) 0.18091(2.0939)

0.19343(2.1662)

DUM81 -0.13677(-4.6239)

-0.13476(5.0502)

Basic statisticsR2 0.72873 0.81821Adjusted R2 0.69482 0.77276RSS 0.021218 0.014220F-statistics 21.4912 18.0032DW-statistics 1.9111 2.0046Durbin’s h-statistics 0.37191 -0.019489

Diagnostic testsSerial correlation


0.0414990.035215

0.0122300.0093080

Functional form LM version

F version2.36542.1141

0.141600.10807

Normality LM version

F version0.63813

NA0.11321

NAHeteroskedasticity


0.472530.45667

0.612230.59351

Notes: Both regressions are based on the augmented production function that includes exports as a thirdinput of production, in addition both include a dummy for 198 (DUM81).

The normal t-values of the coefficients are in parentheses.NA = not applicable.

34

5. Cointegration tests: An assessment

The consistency of the results obtainedusing five different tests employed in thisinvestigation suggests that the results concerningthe existence of a long term relationshipbetween investment, labour (proxied bypopulation), exports and growth appears to beclear-cut across different regressions. In fact,they are definitely robust to differentspecifications taken into account and themethod employed to test for cointegration.Furthermore, it is evident that even though thelong-term effects of exports on growth aredifferent from zero, they are smaller if wecompare them with the effects of traditionalfactors of production.

Given that we have confirmed that thevariables are cointegrated, we can use thecoefficients estimated from the static long-runequations to verify whether the overallproduction function followed constant returnsto scale. This can be done because thecoefficients from the static cointegratingregression are "super-consistent" (Stock,1987).

The Cobb−Douglas productionfunction assumes that the sum of elasticities withrespect to all inputs is equal to unity. In this

case, the contribution of population andinvestment to total output was a little less thantwo thirds and over one third respectively,which is what most studies would expect if wehad assumed a priori constant returns to scale.It is important to emphasize that the sum ofelasticities did not change substantially whenexports were included in the regressions as anadditional input of production and, furthermore,in all the cases, the value shares of inputs in thevalue share of outputs sums to unity or veryclose thereto. All of this clearly suggests thatoverall growth in this case exhibited constantreturns to scale in the 1950−1997 period.

To recapitulate, the empirical evidenceobtained in this case clearly suggests thatexports operated as an additional engine ofgrowth. However, it is very difficult to acceptthe idea that growth in the long run was export-led as some studies have attempted to indicatein the case of Costa Rica (see e.g. Jung andMarshall, 1985; Van den Berg and Schmidt,1994). The findings imply that this was clearlynot the case, and in addition it is certain that theoverall economic performance of this CentralAmerican country from 1950 onwards wasmainly driven by traditional inputs ofproduction.

35

V. CONCLUDING REMARKS

The main objective of this study hasbeen to investigate the relationship betweenexports and output using time series data onCosta Rica. The empirical results have shownthat the ELGH hypothesis can be verified in thisparticular case.

On the basis of annual data extendingfrom 1950 to 1997, various tests detect theexistence of a long-term relationship betweenGDP, investment, population and exports. Thatis, the results suggest that the variables underconsideration are cointegrated and thereforeshare a linear common trend, i.e. they movetogether in the long term.

Furthermore, the available evidenceindicates that the strong correlation between theseries is not spurious as many empirical studieshave found it to be and that the co-movementbetween these variables reflects much morethan an accounting identity. Additionally, theexistence of cointegration between exports andoutput through different test justifies theapplication of the error correction approach. Infact, both methods the EG two-stepprocedure and the unrestricted error correctionmodel make it possible to distinguishbetween short-run and long-run effects ofexports on growth but also allow furtherchecking for cointegration.

What are the economic interpretationsof these empirical findings? Can these results beexplained in terms of the economicdevelopment of Costa Rica in the second partof the twentieth century?

The first and most obvious answer isthat exports can explain not only cyclicalchanges in output (short-term) but also thelong-term trend of output. Moreover, the fact

that the results obtained through theunrestricted error correction model indicate thatall the variables had the correct sign, and weresignificant, corroborates the view thatinvestment and population were significant indetermining the overall rate of growth of outputin the long run, but also indicates that exportswere a significant variable in the growthprocess.

It is reasonable to recognize that alarge number of factors, such as capitalaccumulation, entrepreneurship, innovation,learning by doing and human capitalaccumulation, determine economic growth.However, in this particular case it should beemphasized to the reader that the evidenceobtained from the supply side implies thatgrowth was driven primarily by traditionalfactors of production and, although exportsacted as an additional engine of growth, theimpact was relatively small and limited.

The evidence obtained is in fact inaccordance with recent studies of South-EastAsia which highlight the importance ofinvestment and in particular physical capitalaccumulation in the extraordinary growthexperienced by the so-called Four Tigers (seee.g. Krugman, 1994; Young, 1995). It alsoopenly raise questions regarding the traditionalstory and express serious reservations aboutthe ELGH in general and the so-called neworthodoxy. In addition, this exercise clearlyendorses the neoclassical theory of productionand supports to a lesser extent the so-callednew-fashioned economic wisdom representedby the advocates of free trade and the ELGH(see e.g. Edwards, 1998; Frankel and Romer,1999).

36

However, a crucial question arises. Ifinvestment was the main force driving thiseconomy in the last 50 years, and we acceptand embrace the neoclassical economic theory,then marginal productivity of capital must bediminishing as time goes by. What does thismean? It means that in the long run the CostaRican economy is not going to be able to growas it did in the recent past unless there is asystematic increase in its overall productivity.As a matter of fact, the meagre and sluggishperformance of the economy during most of the1990s could be an indication of this and opensthe door to other studies concerning TFPgrowth and determinants of investment.

Even though, since the 1980s manyresearchers, including Balassa (1983) andEdwards (1992) among others, have employedthe growth of exports as a proxy for policyorientation (see e.g. Levine and Renelt, 1992).This led to classifying studies that investigatedempirically the relationship between the pace ofexports expansion and the overall economicperformance as a distinct category of workrelated to trade policy orientation. In this studyI do not take this into consideration becausethe use of export growth as a proxy for tradeorientation is first of all highly suspect.Furthermore, the issue of categorising the typeof trade orientation followed by Costa Ricawas not an objective of this investigation.

It should be emphasized that theintention of this study was not to derive agrowth model for the Costa Rican economy,nor identify growth determinants, participatedirectly in the export promotion-importsubstitution controversy or determine theeffects of trade policy orientation. Its scopewas more limited and thus the results should notbe used for any of the purposes mentionedabove and furthermore the empirical resultsshould be taken with the appropriate caution.Rather the driving force was to study the long-term properties of the main generating forces ofgrowth and in particular to focus on the role ofexports. The paper’s main objective was toinquire about the validity of the ELGH at theaggregate level in the particular case of adeveloping country, which is considered bymany to be a success story.

As for future research, however, ananalysis of Costa Rica’s growth requires anunderstanding of the reasons that motivatedforeign and domestic entrepreneurs to invest inthis economy. Was the overall physical capitalaccumulation driven by direct investment orwas domestic capital formation mainlyresponsible for the high rates of growth? Whichtype of capital accumulation was crucial inreaching a sustained rate of growth from 1950onwards?

37

REFERENCES

Adams, N. A. (1973). “A note on trade as a handmaiden of growth”, Economic Journal, 83, 329(March): 210−12.

Afxentiou, P. C. and Serletis, A. (1991). “Exports and GNP Causality in the industrial countries: 1950-1985”, Kyklos, 44, 2 (May): 167-79.

Alexander, W. R. J. (1994). “The Investment−Output Ratio in Growth Regressions”, AppliedEconomics Letters, 1, 5 (May): 74-76.

Al-Yousif, Y. K. (1997). “Exports and economic growth: Some empirical evidence from the Arab Gulfcountries”, Applied Economics, 29, 6 (June): 693-97.

Balassa, B. (1978). “Exports and economic growth: Further evidence”, Journal of DevelopmentEconomics, 5, 2 (June): 181-89.

Balassa, B. (1980). “The process of industrial development and alternative development stategies”,Princeton Essays in International Finance, Nº 141, December (Princeton, NJ: PrincetonUniversity, Department of Economics).

Balassa, B. (1985). “Exports, policy choices, and economic growth in developing countries after the1973 oil shock”, Journal of Development Economics, 4, 1 (June): 23-35.

Barro, R. J. and Sala-i-Martin, X. (1995). Economic Growth. New York: McGraw-Hill.

Begum, S. and Shamsuddin, A. F. M. (1998). “Exports and economic growth in Bangladesh”,Journal of Development Studies, 35, 1 (October): 89-114.

Bhagwati, J. (1978). Anatomy and Consequences of Exchange Control Regimes: LiberalizationAttempts and Consequences. Cambridge, MA: Ballinger.

Boltho, A. (1996). “Was Japanese growth export-led?”, Oxford Economic Papers, 48, 3 (July):415-32.

Bruton, H. J. (1989). “Import substitution as a development strategy”, in Handbook of DevelopmentEconomics, Vol. II, edited by H. B. Chenery and T. N. Srinivasan. Amsterdam: North Holland:1601-44.

Bruton, H. J. (1998). “A reconsideration of import substitution”, Journal of Economic Literature,36, 2 (June), 903-36.

Catão, L. A. V. (1998). “Mexico and export-led growth: The Porfirian Period revisited”, CambridgeJournal of Economics, 22, 1 (January): 59-78.

38

Caves, R. E. (1971). “Export-led growth and the new economic history”, in Trade, Balance ofPayments and Growth, eds. J. N. Bhagwati, R. W Jones, R. A. Mundell and J. Vanek.Amsterdam: North Holland, Chapter 19, pp. 403-42.

Cheung, Y. W and Lai, K. S. (1993). “Finite-Sample Sizes of Johansen’s Likelihood Ratio Tests forCointegration”, Oxford Bulletin of Economic and Statistics, 55, 3 (August): 313-28.

Chow, P. C. Y. (1987). “Causality between export growth and industrial development: Empiricalevidence from the NICs”, Journal of Development Economics, 26, 1 (June): 55-63.

Chow, P. C. Y. (1989). “Causality between export growth and industrial development: Empiricalevidence from the NICs − Reply”, Journal of Development Economics, 31, 2 (October):413-15.

Collier, S. et al. (1992). The Cambridge Encyclopaedia of Latin America and the Caribbean.New York: Cambridge University Press, Second edition.

Colombatto, E. (1990). “An analysis of exports and growth”, Kyklos, 43, 4 (November): 579-97.

Crafts, N. F. R. (1973). “Trade as a handmaiden of growth: An alternative view”, Economic Journal,83, 331 (September): 875-84.

Darrat, A. F. (1987). “Are exports an engine of growth? Another look at the evidence”, AppliedEconomics, 19, 2 (February): 277-83.

Edwards, S. (1992). “Trade orientation, distortions and growth in developing countries”, Journal ofDevelopment Economics, 39, 1 (July): 31-57.

Edwards, S. (1998). “Openness, productivity and growth: What do we really know?”, EconomicJournal, 108, 2, (March): 383-98.

Emery, R. F. (1967). “The relation of exports and economic growth”, Kyklos, 20, 2 (May): 470-86.

Emery, R. F. (1968). “The relation of exports and economic growth: A Reply”, Kyklos, 21, 4(November): 757-60.

Enders, W. (1995). Applied Econometrics Time Series. New York: John Wiley & Sons.

Engle, R. F. and Granger, C. W. J. (1987). “Co-integration and Error Correction: Representation,Estimation and Testing”, Econometrica, 55, 2 (March): 251-76.

Engle, R. F. and Granger, C. W. J. (1991). Long-Run Economic Relationships: Readings inCointegration. Oxford: Oxford University Press.

Fajana, O. (1979). “Trade and growth: The Nigerian experience”, World Development, 7, 1,(January): 73-78.

39

Feder, G. (1983). “On exports and economic growth”, Journal of Development Economics, 12,2, (February/April): 59-73.

Figueroa de la Barra, L. and Letelier-Saavedra, L. (1994). “Exportaciones, Orientación al Comercioy Crecimiento: Un Enfoque de Cointegración”, Cuadernos de Economía (PontificiaUniversidad Católica de Chile), 31, 94 (December): 401-21.

Fosu, A. K. (1990). “Exports and economic growth: The African case”, World Development, 18,6 (June): 831-35.

Frankel, J. A. and Romer, D. (1999). “Does trade cause growth”, American Economic Review, 89,3 (June): 379-99.

Granger, C. W. J. (1981). “Some properties of Time Series Data and their use in Econometric ModelSpecification”, Journal of Econometrics, Annals of Applied Econometrics, 16: 121-30.

Granger, C. W. J. (1986). “Developments in the Study of Cointegrated Economic Variables”, OxfordBulletin of Economics and Statistics, 48, 3 (August): 213-28.

Granger, C. W. J. and Newbold, P. (1974). “Spurious regressions in econometrics”, Journal ofEconometrics, 2, 1 (January): 111-20.

Greenaway, D. and Sapsford, D. (1994). “What does liberalisation do for exports and growth”,Weltwirtschaffliches Archiv, 130, 1: 152-74.

Harris, R. (1995). Using Cointegration Analysis in Econometric Modelling. London: Prentice Hall.

Heitger, B. (1987). “Import protection and export performance: Their Impact on Economic Growth”,Weltwirtschaffliches Archiv, 123, 2: 249-61.

Heller, P. S. and Porter, R. C. (1978). “Exports and growth: An empirical re-investigation”, Journalof Development Economics, 5, 2 (June): 191-93.

Hendry, D. F. (1986). "Econometric Modelling with Cointegrated Variables: An overview", OxfordBulletin of Economics and Statistics, 48, 3 (August): 201-12.

Henriques, I. and Sadorsky, P. (1996). “Export-led growth or growth-driven exports? The CanadianCase”, Canadian Journal of Economics, 29, 3 (August): 541-55.

Islam, M. N. (1998). “Exports expansion and economic growth: Testing for Cointegration andCausality”, Applied Economics, 30, 3 (March): 415-25.

Jin, J. C. (1995). “Export-led growth and the Four Little Dragons”, Journal of International Tradeand Economic Development, 4, 2 (July): 203-15.

Johansen, S. and Juselius, K. (1990). “Maximum Likelihood Estimation and Inference on Cointegrationwith applications to the Demand for Money”, Oxford Bulletin of Economics and Statistics,52, 2 (May): 169-210.

40

Jung, W. S. and Marshall, P. J. (1985). “Exports, growth and causality in developing countries”,Journal of Development Economics, 18, 1 (May/June): 1-12.

Kavoussi, R. M. (1984). “Export expansion and economic growth: Further empirical evidence”,Journal of Development Economics, 14, 1/2 (January/February): 241-50.

Khan, A. H. and Saqib, N. (1993). “Exports and economic growth: The Pakistan experience”,International Economic Journal, 7, 3 (Autumn): 53-64.

Kohli, I. and Singh, N. (1989). “Exports and growth: Critical minimum effort and diminishing returns”,Journal of Development Economics, 30, 2 (April): 391-400.

Kravis, I. B. (1970). “Trade as a handmaiden of growth: Similarities between the nineteenth andtwentieth centuries”, Economic Journal, 80, 320 (December): 850-72.

Kravis, I. B. (1973a). “A reply to Mr. Adams”, Economic Journal, 83, 329 (March): 212-17.

Kravis, I. B. (1973b). “A reply to Mr. Crafts’ Note”, Economic Journal, 83, 331 (September): 885-89.

Kremers, J. J. M., Ericsson, N. R. and Dolado, J. J. (1992). “The Power Cointegration Tests”,Oxford Bulletin of Economics and Statistics, 54, 3 (August): 325-48.

Krueger, A. O. (1978). Foreign Trade Regimes and Economic Development: LiberalizationAttempts and Consequences. Cambridge, MA: Ballinger.

Krugman, P. (1994). “The myth of Asia’s miracle”, Foreign Affairs, 73, 6 (November/December):62-78.

Kugler, P. (1991). “Growth, exports and Cointegration: An empirical investigation”,Weltwirtschaffliches Archiv, 127, 1: 73-82.

Kwan, A. C. C., Cotsomitis, J. A. and Kwok, B. K. C. (1999). “Exports, economic growth andstructural invariances: Evidence from some Asian NICs”, Applied Economics, 31, 4 (April):493-98.

Lee, F. Y. and Cole, W. E. (1994). “Simultaneity in the study of exports and economic growth”,International Economic Journal, 8, 1 (Summer): 33-41.

Levine, R. and Renelt, D. (1992). “A sensitivity analysis of Cross-Country Growth Regressions”,American Economic Review, 82, 4 (September): 942-63.

Lussier, M. (1993). “Impacts of exports on economic performance: A comparative study”, Journalof African Economies, 2, 1 (May): 106-27.

Maddala, G. S. and Kim, I. M. (1998). Unit Roots, Cointegration and Structural Change.Cambridge: Cambridge University Press.

41

Michaely, M. (1977). “Exports and growth: An empirical investigation”, Journal of DevelopmentEconomics, 4, 1 (March): 49-53.

Moschos, D. (1989). “Export expansion, growth and the level of economic development: An empiricalanalysis”, Journal of Development Economics, 30, 1 (January): 93-102.

Pesaran, M. H. and Pesaran, B. (1997). Working with Microfit 4.0: Interactive EconometricAnalysis (User’s Manual). Oxford: Oxford University Press.

Phillips, P. C. B. (1986). “Understanding Spurious Regressions in Econometrics”, Journal ofEconometrics, 33 : 311-40.

Ram, R. (1985). “Exports and economic growth: Some additional evidence”, EconomicDevelopment and Cultural Change, 33, 2 (January): 415-25.

Ram, R. (1987). “Exports and economic growth in developing countries: Evidence from Time-Seriesand Cross-Section Data”, Economic Development and Cultural Change, 36, 1 (October):51-63.

Rana, P. B. (1988). “Exports, policy changes and economic growth in developing countries after the1973 oil shock”, Journal of Development Economics, 28, 2 (March): 261-64.

Rodrik, D. (1999). The New Global Economy and Developing Countries: Making OpennessWork. Washington D.C.: Overseas Development Council, Essay No. 24.

Rotemberg, J. J. (1987). “Export promotion as a development strategy”, Journal of DevelopmentEconomics, 26, 2 (August): 343-55.

Sephton, P. S. (1989). “Causality between export growth and industrial development: empiricalevidence from the NICs − A Comment”, Journal of Development Economics, 31, 2(October): 413-15.

Sengupta J. K. (1991). “Rapid growth in NICs in Asia: Tests of New Growth Theory for Korea,”Kyklos, 44, 4 (November): 561-79.

Serletis, A. (1992). “Export growth and Canadian economic development”, Journal of DevelopmentEconomics, 38, 1 (January): 133-45.

Serven, A. K. (1968). “The relation of exports and economic growth: Comment”, Kyklos, 21, 3(August): 546-48.

Shan, J. and Sun, F. (1998). “On the Export-Led Growth Hypothesis: The Econometric Evidencefrom China”, Applied Economics, 30, 8 (August): 1055-65.

Sheehey, E. J. (1990). “Exports and growth: A glawed framework", Journal of DevelopmentStudies, 27, 1 (October): 111-16.

42

Sheehey, E. J. (1993) “Exports as a factor of production: A consistency test”, World Development,21, 1 (January): 155-60.

Stock, J. H. (1987). “Asymptotic Properties of Least Squares Estimators of Cointegrating Vectors”,Econometrica, 55 : 1035-56.

Syron, R. F. and Walsh, B. M. (1968). “The relation of exports and economic growth: A note”,Kyklos, 21, 3 (August): 541-45.

Thomas, R. L. (1997). Modern Econometrics: An Introduction. Harlow, England: Addison-Wesley.

Tyler, W. G. (1981). “Growth and export expansion in developing countries: Some empiricalevidence”, Journal of Development Economics, 9, 1 (August): 121-30.

Van den Berg, H. and Schmidt, J. R. (1994). “Foreign trade and economic growth: Time seriesevidence from Latin America”, Journal of International Trade and Economic Development,3, 3 (November): 121-30.

World Bank (1993). The East Asian Miracle: Economic Growth and Public Policy, Oxford:Oxford University Press.

Wu, C.S. and Chuang, S. C. (1998). “Duty drawback mechanisms: The system in Taiwan (China) andrecommendations for Costa Rica", in Trade Policy Reform, edited by J. Nash and W. Takacs.Washington D.C. World Bank: 189-210.

Young, A. (1995). “The tyranny of numbers: Confronting the statistical realities of the East Asiangrowth experience”, Quarterly Journal of Economics, 110, 3 (August): 641-80.

43

APPENDIX

44

Figures

1.1. Gross domestic product in 1990 prices

1.2. Rate of growth of GDP

1.3. Population

1.4. Rate of growth of population

1.5. Gross domestic investment in 1990 prices

1.6. Rate of growth of GDI

1.7. Gross fixed capital formation in 1990 prices

1.8. Rate of growth of GFCF

1.9. Exports of goods and services in 1990 prices

1.10. Rate of growth of exports of goods and services

45

Figure 1.1. Gross domestic product in 1990 prices

Figure 1.2. Rate of growth of GDP

46

Figure 1.3. Population

Figure 1.4. Rate of growth of population

47

Figure 1.5. Gross domestic investment in 1990 prices

Figure 1.6. Rate of growth of GDI

48

Figure 1.7. Gross fixed capital formation in 1990 prices

Figure 1.8. Rate of growth of GFCF

49

Figure 1.9. Exports of goods and services in 1990 prices

Figure 1.10. Rate of growth of exports of goods and services

51

UNCTAD Study Series on

POLICY ISSUES IN INTERNATIONAL TRADE AND COMMODITIES

No. 1 Erich Supper, Is there effectively a level playing field for developing country

exports?, 2001.

No. 2 Arvind Panagariya, E-commerce, WTO and developing countries, 2000.

No. 3 Joseph Francois, Assessing the results of general equilibrium studies of

multilateral trade negotiations , 2000.

No. 4 John Whalley, What can the developing countries infer from the Uruguay

Round models for future negotiations?, 2000.

No. 5 Susan Teltscher, Tariffs, taxes and electronic commerce: Revenue

implications for developing countries, 2000.

No. 6 Bijit Bora, Peter J. Lloyd, Mari Pangestu, Industrial policy and the WTO, 2000.

No. 7 Emilio J. Medina-Smith, Is the export-led growth hypothesis valid fordeveloping countries? A case study of Costa Rica, 2001.

No. 8 Christopher Findlay, Service sector reform and development strategies:Issues and research priorities, forthcoming.

is the export-led growth hypothesis valid …the export-led growth hypothesis (elgh) postulates that...

Documents