Public Disclosure Authorized Public Disclosure Authorized

RESEARCH

POLICY

WORKING

PAPER

Foreign TechnologyImports and Economic Growth DeVelOping Countries

1412

developingcountry's economicgrowth rate

- increasesas foreign

technology importsincrease. In developing countries,

XiaomingZhang

dependnoton innovation

Heng-fu Zou

but on importingforeign plantsandequipmentandon borrowing foreign technology.

Public Disclosure Authorized

Public Disclosure Authorized

in productmty increases

The World Bank PolicyReseach Department Public Econom.icsDivion.

January 1995

POL.ICY RIESE-ARCH WORKING PAPEI

1412

Summary findings Zhang and Zou investigatethe rclationship between foreigntechnology imports and economicgrowth in developingcountries. They develop an intertemporal endogenousgrowth model that explicitly accepts foreign tcchnologyimports as a factor of production. The model establishesa link between the growth rate of productivityin a developing country and the country's intensity of learning to use foreign technologies. T hey hypothesize that a developingcountry's economicgrowth rate increasesas foreign technology imports increasie.

They run eegressionswith data for about S0 devclopingcountries, using different econometric methods and time spans. These cmpirical tests confirm the hypothesis that foreign technology transfers boost incomc growth rates. Moreover, economicdcveloping in developing countries differs from that in industrial countries. In developingcountries, increases in productivitydepend not on innovation but on importing foreignplants and equipment and on borrowing foreign technology.

Thispaper-a productofthe PublicEconomicsDivision,PolicyResearchDcpartmcnt-is partofa largereftortinthe department to understandeconomicgrowthand foreigntrade.Copiesof the papcr are availablcfree from the World Bank,1818 H Street NW,Washington,DC 20433. Pleasecontac CarlinaJones, room NlO-063,extension37754 (38 pag,.s).January 199S.

ThePOIKy ResacnhWorkingPaperSeoies dsmminatesthe findingsof work in press to encourage the exchange of ideasabout deuelopment issu An obecative of theseris isto getthefindingsout quickly,een if thepresentations ae essthanfully polishecL The papers cany thenamesof theauthorsand shouldbeusedandcitedaccordingly.Thefindings,interpretations, and condusions arethe autbors'ownandshouldnot beattributedto theWorldBank its Fxecutive Boardof Directors,or any of its membercountries.

Producedby the PolicyResearchDisseminationCenter

FOREIGN TECHNOLOGYIMPORTS AND ECONOMIC GROWTHIN DEVELOPING COUNTRIES

Xiaoming Zhang and

Heng-fu Zou Policy ResearchDepariment The World Bank Washington,D.C. 20433

Mailing Address: Mr. Heng-fu Zou Economist The World Bank Room N1O-075 Washington,D.C. 20433

We thank Lance Taylor, Richard Eckaus and ShantayananDevarajan for very helpful comments.

FOW-NTCEO.DN

BO

ADW

sNI

GILOW1Z IN DSVEOI!ING CtNRE By XiaomingZhang and Heng-fu Zou 1. NTRODUCTION The relationship between dhe transfer of foreign technology and economic growth in developingcountrieshas long been studiedby economists. In the gap-modelapproach,Cheneryand Bruno (1962), McKinmon(1964),Bacha (1984),and Taylor(1990, 1993)focus on foreignexchange resources as of the most important constrainton economi growth in developingcountries. Their argument is based on the idea that most developingcounties, becase they camot produce the eeded technology-embodied capital goodsdomestically,rely on importedcapitl goods n acquirng advancedtechnology;thus, importedcapitalgoodsand intermediategoodsare indispensableinputs; and if there is not sufficientforeign exchangeto fmancethe desired technology-embodiedforeign capitalgoods and intmediate goods, the economycannotoperate properly, not to mentionachieve high growth.1 Some economistseven claim that foreign technologyimports are the most importantfactor in explaiinig the rapid economic growth of Japan, Taiwan, South Korea, and other newly industrializedcountries. For example, Amsden(1989, p. V) argues that the commoncharacter of the economicdevelopmentprocess of all the 'late industialiUs their industrializationis based on learng.

(i.e., developingcoantries) is that

Suchcountriesas Japan, SouthKorea, Brazl, Turkey,

India, and Mexico"all industrializedbyborrowingforeigntechnologyrather tan by generatingnew lBochove(1982)also arguesthat manyimportsare indspensableinputsin developingeconomies, thereforeimportsshouldbe treated explicitlyas a factorof productionin long-rungrowthmodels.

-2productsor processes[.)" She suggeststhat a growthmodelappropriatefor late industrializersshould incorporatenot technologicalinnovation,but foreign technologyimports. While the idea of imports as a factor of productionhas been put forward in some simple models, to our knowledge, there does not exist an intertemporalendogenous growth model incorporatingthis idea, nor are there any systematicstudiesto test ihis hypothesis.Althoughmany new growth models try to tacklethe importantissue of endogenousproductivitygrowth, they fail to explain the important linkage between foreign technologytransfer and the phenomenaleconomic growth in countries like Japan, South Korea, Taiwan, and many others. Thus the conventional growthmodel is inappropriatefor developingcountriesbecauseit throws away valuable information on the source of productivity increase in these countries: borrowed foreign technologythrough import and transfer. On the empiricalside, there are few studies based on the two-gap model approachin testing the linkagebetween imports and growth. For example, Esfahani(1991)conducteda simultaneoussystemanalysis testingthe relationshipamong exports, intermediate Worts, and economicgrowth using a sampleof 31 semi-industrialized countries, andfoundthat 'export promotionpoliciesin these countriescan be quite valuablein supplyingforeign exchange,which relieves importshortagesand permits output expansion."(p. 114) However, there existsno empirical studiesdirectly testingthe hypothesisthat foreign technologyimports are the most importantfactor in explainingeconomic growth process in developingcountries. In this paper, we first developan intertemporalendogenousgrowthmodel that explicitlylinks foreign capital importsto economicgrowthin developingcountries. Then we conductan empirical test on the model using a sampleof about 50 developingcountries.

-3-

In section 2, we present a two-goodsmodel of optimal growth along the lines of the technologyargumentby dividingcapitalaccumulationin a typicaldevelopingcountry into two parts: the accumulationof traditionally,home-producedcapitaland the accumulationof importedforeign technology. Revenues from exports are used to purchase foreign consumer goods and foreig technologyimports. We formallyshow that a developingcountry's economicgrowth rate increases as foreign technologyimports increase. In section 3, we conduct empiricaltests of the hypotheses generatedby the model, using panel data from developiL,countries. Section4 concludes.

2. AN ENDOGENDUS GROWT0H MOE0M FOREIGNTECHIOLOGY IMPORTS The model developed in this sectionhas its orgins m the neoclassicalgrowth model. The standardversion of the neoclassicalgrowthimodel developedfirst by Solow(1956)has the property that the only potential sourcesof growth are susained exogenousincreasesin factor supplies(e.g., populationgrowth)and exogenouslygiventechnologicalchange(see, for example,Jones & Mamuelli 1989). Thus, except for the possibilityof exogenoustecbnicalchange, ihesemodels of growthlead to the starting conclusionthat there is no per capita growth in the long rn.

Rather, dependingon

inital conditions, there is growth until the capital stock reaches a steady state where things settle down permanenty. The fundamentalproblemwith he neoclassicalgrowthmodel, as Solow (1970) acknowledged,is that it is not able to explain the wide differencesin rates of productivity growth across countries. Faced with the phenomenalsustined growth in per capita output that many developingcouries have experienced,the only explanationthe model has to offer is exogenous technologicalchange, which sheds no new light on cross-countrydifferences.

*4 Since the mid-1980s,many economistshave tried to endogenizethe processof technological change. Three differentgroups of modelshave been proposedto deal with this problem. The first group relies on externalityand increasingreturns to scale (Romer 1986). In the second group are the models of human capital formation pioneered by Lucas (1988). The third focuses on the introduction of new goods with learning by doing advancedby Grossman and Helpman (1990). However, aL of these models fail to explain the important linkage between foreign technology transfer and thephenomenaleconomicgrowthin countriessuch as Japan, South Korea. Taiwan, and many other developingcountries. As Amsden(1989)points out, the conventionalgrowthmodel is inappropriatefor developingcountriesbecause it trows away valuable informationon the source of productivityincreasein thesecountries:borrowedforeigntechnologythrough importand transfer. We constructa model that addressesthis shortcoming.Two featuresdistinished our model from all the other growth models. First, we explicitly assume that foreign capital goods are indispensableinputs in developingcountries' production. Foreign capitalgoods are not perfectly substitutableby home capital goods. Second, we build into the model a direct linkage between foreign technologicalimportsand productivityincreasesin developingcountriesby assumingthat the rate of technologicalgrowthis a positive functionof foreign capital imports. 2.1 The Model There are two economiesin this model:the homecountryand the foreign country. The home country is a developingeconomy, and foreign countryis a developedone. There are two goods the home good and the foreign good; and the homegood price in the foreign market is Px, which, as will be discussedlater, is a negativefunctionof the quantityexported.

- 5At time t there are N(t) identicalpersons in the home countryproducingthe home good with a technologygiven by the productionfunction,

Y(t)

=

Kh(t)aKf(t)O[A(t)N(t)]-cr+, (a +0 < 1)

(1) Where Kh(t) is home capitalstock at time t and, K(t) is foreigncapital. While allowingsubstititiori between home capital and foreign capital in production, in general, foreign capital through its embodimentof modem technologyis more efficientthan home capital. The idea of putting foreign capital into the productionfunctionas an input is taken from the paper by Devarajanand Zou (1993). N(t) is the total populationin the home country. The populationis growing at a constantrate n, i.e., N/N = n. A(t) is an index of labor-augmentingtechnologyat time t. A t) is growing at rate 4: A(t) ewt. We can defne N(t)e'Oas the effective labor force at time t, and denote it as .

k=N(O)e(f+O)t, because N=N(O)et

The effective labor force grows at the rate of n

+

=

Thus,2

(2)

k. For a given size of physical

population, there will be more effectiveunits of labor as time passes. But the number of physical bodies increases at the constantrate n. Now we can rewrite the productionfunction as follows: Y=pKKRFA-(cx+0) h f

2 For

(3)

notationalsimplicity,we will drop the time indexfor all the currentvariablesfrom here on. So unless specifiedotherwise,N is equivalentto N(t).

-6-

Dividingboth sides of (3) by

h,

and defining y = Y/I, kh - KhIAT,and kf= Kf 1,

the constant

return assumptionimplies: h -f

~~~~~~~~~~~~~~(4

Note that here y, kh, kf are all variablesmeasuredin effective units of labor. Now we need to fiuther examinethe technologyindexA = eWt. In the standardneoclassical growth model, 4 is assumedto be exogenouslygiven. In our model, in order to capture a stylized fact of developingcountries, we assume that the technologicalgrowth rate is a function of the importedforeign capital stock, i.e.,

|t b(k(t-l))

'Of

#SYh< Yf

(5)

if Yh 2Yf

whereky(t-1) is the foreign capitalstock measuredin efficientunits of labor at time t - 1, and $'(.) > 0, r(.)

< 0.

f is the developedcountry's technologicalgrowthrate, which is assumedto be

exogenouslygiven and constant for simplicity. Equation(5) says that the growth rate of the laboraugmentingtechnologyin the developmgcountry at tuMet is a positive fimction of the stock of importedforeign capital goods at time t - 1. This is an inportant assumptionin our model. It establishes a direct lik between foreign technology transfer and home country's techological growth. We can identify hee chamnelsthrough which foreign capitalimportaffects the growth rate of technologyin developingcountries: first, foreign plant and equipment investmentsgenerally embodyadvancedforeign echnology, advanceddesigns and advancedmanagementmethods. More

-7 investmentin importingforeign plant and equipmentwill raise the home country's technologyby havingmore embodiedforeign technology. Furthermore,investmentin foreignplant and equipment involves training techniciansin the foreign country, so a higher stock of foreign capital means a proportionallylarger numberof peoplebeing trained in a foreign country. The second channel by which the stock of foreign capital affects the growth rate of technologyis by scale economies. A higher level of foreign capital importsmakes it more likelyfor the homecountry to operate foreign technologyon a scale sufficient to minimizeunit costs. The third channel is through experience accumulation. How efficientlyforeign technologyis used will depend on the experienceof the user. The higher the level of foreign capitalimports,the more intensivelypeoplehave to learn to operate foreign equipment,and hence the faster experienceaccumulates. That is to say, learning-by-doing, which is one critical aspect of learning in general, dependson the accuulation of foreign capital. The assumptionthat the growthrate of technologyin the developedcountry is exogenously fixed is for analytical simplicity. Alternatively, we could have used an endogenous growth formulationalong the lines of Lucas (1988)or Grossman& Helpman(1991). However, the focus of this paper is on the developingcountry's "catching-up"process; whathappensafter the developing country becomesa developedone is not importanthere. We assumethat there is no foreign direct investmentin the homecountry. To obtain foreign technology,the home countryrelies on its export earnings. This assumptionis for simplicity;it can be relaxed withoutchangingthe results in our model. The home country's foreign earnings are: E = PxX

(6)

where Px is the price of the home good in the foreign market. Let Ch and Cf be the aggregatehome good consumptionand foreign good consumptionat time t, respectively. The dynamicequationsfor the a

lation

of home capital and foreign capital are:

-8-

kh = KaKOf(eOfN)l (1*I) hh :f(XNl(t

~ Kh - X,(8 Ch-kX

lff =PxX-Cf-Kf.

(8) (9)

Expressed in effective units of labor, the dynamicequationsfor the accumulationof home capital and foreign capitalbecome: kh =kak

h f

-Ch

kf = Px,x-cf-

-

(n+O)kh (n+O)kf.

(10) (11)

Note that we have assumedaway capital depreciationfor simplicity. Consumers maximizean instantaneousudlity functionspecifiedas:

log(1gCh /NeO)+0log(Cf Ne Otfle-Ptdt.

(12)

The separabilityof utilityfunctionis also purely for analyticalsimplicity. The constnt 0 is positive and measures the preferencefor foreign good consumption. Note that the utility functionis defined in consumptionper capita (per physicalbody) terms while the dynamicequationsare defined in terms of consumptionper effective labor unit. We can transform the utility functionusing the equality CNe't0= c.

J°[lg0ch+Ologcf] dt t

(13)

The representativeagent in the home country chooses ch and cf so as to maximize (13) subject to the dynamicconstraints(10) and (11), and the initial values of home capital and foreign capital (kh(O).kf (O)).

-9-

The currentvalueHamiltonian functionis: [ k-Ptkogc^.iogc c - x - (fn+j)kh]

H * +Xf[Pxx

Cf -

-

(n+ )kfJ

(14)

Notethat although4 changesin eachperiod,the representative agenttakes+ as givenbecauseit is an externalityas in Romer(1986)and Lucas(1988). Thenecessaryconditionsfor maximization are: Cf

PxcCh,

_

akfkk

__-

.k O

kh

kf

(15)

If

-

=AO

(n+*+p),

(16)

h- (n+4)+P),

(17)

-

ch -

P-x - Cf-

In the steadystate, th = ef

(n+4O)kh -

(n+ )kj

=

kd = kf

x,

(18)

(19)

0O. So the necessaryconditionsfor optimization

in equilibriumare: Cf -

-,,j3-a-1

akf kh -cf

k

-

Pxch =°

(20)

(n+4+p)-O

(21)

-(n+

+P) = 0

(22)

- 10-

khkf -ch-(n+O)khPxx-

Cf-

x =0

(n+O)kf=0

(23) (24)

where a bar over a variable denotes its steady-statevalue, and all derivativesare evaluatedat the steady state. Condition (20) gives the optimalrelationshipbetweenhome good consumptionand foreign good consumption. Conditions(21) and (22) are the modifiedgolden rules. Condition (23) gives the steady-statelevel of per effectiveunit of labor consumptionof home good. Condition(24) says that exports is the only sourcesof incomefor purchasingthe foreign consumptiongood and foreign capitalgood. 2.2 Growth rate at the steady state We define the steady state(or balancedgrowthpath) as the statewhereall the variablesgrow at a constant rate. Thus we rule out paths with ever increasinggrowth rates. Equations(21), (22), (23), and (24) tell us that in the steadystate, the consumptionof home and foreign goods, and the home capitalstock and foreign capitalstock measured at per effective labor unit are constant, i.e.,

Ch=C, cf=

Cf.

kh = kh,kf= kf

Hence the growth rates of all per effectiveunit of labor variablesare zero. Knowingthis, we can find the growth rate of all the variablesmeasured in per capita (i.e., per physical body) from the relationbetweenper capitavariablesandper effectiveunit variables. Takingtime derivativeof both sides of equation(25) and then dividingthe resultby (25), we get the growth rate of per capita home good consumptionat steady state:

d(ChIN)Idt

Ch =

)

(kfp

(26)

4]=

Similarly,we can show that all the per capita variablesgrow at the same rate when the economyis at the steady state: d(CfIN)Idt

d(Kh IN)/dt

Cf/N

Kh IN

(27

d(KfIN)ldt = d(YlN)ldt=

Kf IN

YIN

Equation(26) and (27) say that in the steadystate, per capita consumptionof homegood and foreign good, and per capitahome and capitalstocks,and thusper capita income,are growingat the samerate 4(

z

+(kf)), whichis determinedby the steady-stateforeign capitalstock per effective

labor unit. If a countryhas a higher steady-stateper effective labor unit foreign capital stock, its per capita income growth rate in the steady state is higher. It is conceivabledtat given the right parametersthe home country's growthrate can be higher than that in the developedcountry, i.e., *(kf) >.O.

Then the income gap between the two countries will decrease, unti the home

country's per capita incomelevel is equal to that of the foreign countryat whichpoint there wiUlbe no particularadvantageof importingforeign technologyand the growth rate of the home economy will be the same as the foreign country O = gy This scenariocaptures the essence of the catchingup experienceby many late industrializers(Japan, South Korea, Taiwan, Singapore), which was based on learning and borrowing foreign technologyfrom developedcountries. The aggregatevariablesof this economy-aggregate income Y, total consumptionsof home good Ch and foreign good Cf, and total home capital stock Kh and foreign capital stock K,

-

are

all growing at the rate of 4,+n, until the home country's aggregate incomelevel reaches that of the foreign country, at which point the growthrates of the two economieswill converge.

- 12 -

The above result is a very powerfulone. It links a developingcountry's long-runeconomic growthrate with its effortsin learningadvancedtechnologyfrom foreign technologicalimports. The model can explain why some developingcounties succeededin catching up with the developed countrieswhile others laggedbehind. We pause for a momentand compare the model constructedhere with all existing growth models. There are three distinctivefeatures which set this model apart from new growth models. As mentionedin the beginning,in the existinggrowththeory, the growdhrate of technologyis eitier assumed to be exogenouslydetermined (Solow, 1956), or to be determined endogenouslyby postulating some externality effects (Lucas 1988, Romer 1986). All of them have one dting in common: they assumeaway the importantfact that developingcountnes can usually take advantage of the existing advancedtechnologyin the developedcountriesby intensivelearning, insteadof by investing in R&D and innovation. Althoughin the modelsdeveloped by Grossmanand Helpman (1991), the technologicaldifferencebetweenthe North and South is a central focus, they model the learning process as a rather mechanicalone: the North always creates new products and the South always imitates. The developing countries can never catch up and surpass the income and technologicallevel of the developedones. Our model is a drasticdeparture from growthmodels on technologicalprogress. In our model, it is the quality gap betweenthe developingcountry's home technologyand importedtechnologyfrom developedcountrythat propels the former to catch up with the latter. Through active learning, the developingcountrycan reduce the technologicalgap and eventuallybecome a "NIC". By explicitlylinkingproductivitygrowth with increasesin output, our model is a long distancedescendantof models developedby Kaldor (1967, 1978). Another importantfeatureof this model is that the steadystate is givena new meaninghere. In most growth models, the steady state means an ideal state existing only in the futre.

All

- 13 developingcountries are usually assumednot to be in such a state, as if the long histories of these countries do not count. In our model, we do not assume that economic growth starts from the beginningof the 20th century or the end of World War II or some arbitrary date. After all, most developingcountrieshave severalhundred years history; many even have several thousandyears of civilization. If after such a long history a country is still in some mid-way to the steady state, then what is the use of studyingthe steadystate? In our model, we postulatethat all developingcountries are in their steady state development. Each country's steadystate per capita income is growingat a rate determinedwithinthe economicsystem. The differentgrowthrates we observe are the results of each nation's different preferences and tastes (which are related to culture and history) and different foreign exchangeresources. 3. EMPIRICAL TESTS OF THE RELATIONSHIP BETWEEN FOREIGN TECHNOLOGY IMPORTS AND ECONOMIC GROWTH In this part, we conduct enpirical tests on the predicdons generated by the model in the previoussection. We test the relationshipbetweenthe economicgrowthrate and foreign technology imports. We first develop statisticalmodel specifications,then discuss the data and the empirical results, and then discuss policy implicationsand suggestionsfor future research. 3.1 The Model and StatisficalSpecification Our empiricalmodel specificationfollowsthe generalapproachused in the studyby Mankiw, Romer, and Weil (M-R-Wthereafter)(1992), althoughwe do not adoptmany of their assumptions. Let the productionfunctionbe: Y(t)= K(t)K((28)(t)N(t)) Let Sk be the fraction of income invested in Foreigncapital imports. The dynamicsof the economy is given by:

- 14 -

K(t)h =S(OhY(t)

(29)

-K(Q)h

ke(tyf= S(t)hY(t) -K(t)f

(30)

Equations(29) and (30) imply that the economyconvergesto a steadystate given by:

-a-

K(th =A(t)N(t) |

(31)

K(tof=A(t)N(t)|Q)ftS(f

I-a(32)

Substituting(31) and (32) into the productionfumntion,and taking logs, we get an equation for per capita income: InY(t) =lnA(t)+InN(t)-

a-'43 1-a1-U

a

1

a

-li

h+

I

-11

nf

(33)

Equation(33) relates a country's level of income with the rate of home capital investment and that of foreign capital investment,and its population. This equationwill be the basis of our empiricalstudy. Our modelpredicts that the coefficientson homecapital investmentand on foreign capital importare positive, and the latter shouldbe bigger than the former in magnitude. We have to first make assumptionson the parametersbefore we can test the model. We assume that a is country specificbut constantover time withinthe same country. Thus by takingfirst differencesof individualcountry observationsover time we can eliminatethe 6:

- 15 -

bnYr-InYt-I =

[IUnA(t)-nA(t - 1)] + +

1

6

(nSht-

l

InSh,t-1)

(InSf,b-SfnSyt- ) +[InN()- InN(r- 1)]

i.e., y-[bMA(t)-

A(t-1)]+ 1-.

s|

+ l-a-ti[fI

+

(35)

The term A(.) in equation(35) is in fact an all encompassingvariable. It reflects not only technology, but also many unobservablefactors. These include resource endowments,climates, social institutions,and oiter random effects. In M-R-W's study, they assumethat IA (t) = a +

(36)

f

where a is assumedto be a constant both cross-countryand over time, and e is a random shock includingall country-specificfactors that are independentof the rate of investmentand population growth. In growth form, their assumptionmeans: LnA

(t) - nA(t

-

)= -t -f

t_

(37)

That is, all the unobservedinstitutional variables are assumed away in this formulation. This assumptionallowsthem to proceed with the simpleOLS estimation. M-R-Wprovidethree reasons for this assumption. First, this assumptionis made not only by Solow, but also in many other growthmodels. They also argue that in models where investment is endogenousbut preferencesare isoleastic,Sh and Sf are independentof e. Second, this

-

16-

assumptionis necessary for testingdifferent hypothesein their paper. Third, because the model predictionsare very precise, they can use the result to test whetherthe OLS is a mis-specification. Many economistshave questionedthis assumptionand the three supportingarguments. For example, Islam (1992) argues that investmentand fertility behavior is apparently affected by the variables includedin the A(t). Indeeda theoreticalcase can be made against M-R-W's assumption. By standard formulation, M(t)

-

bzA(t -

1) is the technical growth rateD , which is country

specific. In fact, 4 can be decomposedinto:

'O=

(38)

xci + git

Where i denotes countriesin the sample, t is index for time.

C*is a country-specificconstant, and

Pft is all the unobservedvariablesthat are not correlatedwith the explanatoryvariables, and is i.i.d. 2 with variance equal to cr;. Substituting(37) and (38) into equation (35), we have:

y c + y1 XC 1 +i-a-li

shl+____ |hJ

1-a-#

[4 + I+Ait

(39)

| S[] N

Equation(39) specifiesa model with heterogeneousinmtercepts, homogeneousslopes. If this specificationis true, then M-R-W's specificationof an independente is equivalentto a restrictionthat all interceptsare the same. And their estimateswould be biased. In what follows, we will use the specificationin equation (39) to study the relationship between a country's level of income and its foreign capitalimport share in GDP, albeit expressed in growth rate form. The dependentvanable is the income growthrate, the independentvariables are the growth rates of dte share of foreign capital importsin GDP, of home investmentis share in GDP, and of population. The term Ct is an unobservableconstantfor each couty.

We will use

- 17 -

variable-interceptmodels with panel data to deal widhthis issue. By assumingthat the effectsof the numerous omitted country-specificvariables are each individuallyunimportant but collectively significantand possess the propertyof a randomvariablethat is uncorrelatedwith all other included and excludedvariables, we can specifyour model as having commonslopesfor all countriesbut the interceptvaries over individualcountries. This methodis called the variable-interceptmethod.3 We will also run simple OLS regressionsbased on M-R-W's assumptionand compare the results from different methods, whichwould providea test on their assumption. 3.2 The Data and Samples The dataare assembledfrom the UnitedNationsStatisticalOffice, the World Bank, Summers and Heston (1991), and some other sources. Definitionsfor all variablesand data sources appear in Appendix1. The data do not includeOECD countries,since manydevelopmenteconomistsargue that the developmentprocess in developingcountriesis different from that of developedcountries. We also exclude major oil producers fron our sample (as defined by World Bank in World DevelopmentReport). Countries with a populationless than 1 million in early 1980sare excluded the samplebecausethe determinationof their real incomemay be dominatedby idiosyncraticfactors. The data include annal variablesand cover the period of 1965-1988. The panel data set albowsus to conduct tests based on variable-interceptmodels, which can control for unobserved country-specificeffects. We measure Sf as the share of current foreign capital goods imports in current GDP. The data on foreign capital imports are obtained from the United Nation's

3 SeeHsiao(1986) for the detailsof the variable-intercept methodin panelregression.

- 18 -

InternationalTrade Statistics.4 Sh is calculatedas the differencebetween the share of current total investmentin current GDP minus Sf. The data on current total investmentshare in GDP are from the Summers and Heston (1991) data set. We measure SfISf as changes in the share of foreign capital import in GDP, Sh ISh the change in the share of home investment in GDP. Y/Y is real annualgrowth rate of GDP, which are from the WorldBank's World Tables (1990). The population growth rate N/Nis from the populationdata in the 1990 World Tables. Table 1 lists all the countriesin our sampleand the mean valuesof Sf Sf, ShISh, YlIY,andNIN. We also list the quality rating of the databy Summersand Heston, since manyof our variablesare taken from their data set. This information should be useful in helping readers make judgement on the reliability of the statistical inferencesfrom the data. The number of developingcountries includedin our empirical study varies amongdifferent model specifications,dependingon the variablesincludedin a specification. Some countriesmay not have informationon certain important variables so we have to exclude them from a particular equation. 3.3 The Result 3.3.1 Initial regressions Table 2 presents three different regressionsof the growth rate of incomeon the growth rate of foreign capital import, growth rate of home investment,and growth rate of population. Before

4We

divide the SITC two digit level import commoditydata into three main categories:capital equipmentiWports(includingSITCcommodities71,72, 73, part of 86, 87, and part of 9), intermediate good inport (includingSITC commodities2, 3, 4, 5, 6, and part of 9), and final consumtion good import(includingcommoditiesin the SITC groups0, 1, 81-85, part of 86, 89, and part of 9).

-

19 -

Table 1: The list of Countries In the Sample (All the variables are averages over the period 1965-1988)

Country Greece Portugal Israel Hong Kong SouthKorea Kenya Costa Rica DominicanRep. El Salvador Guatemala Honduras Jamaica Mexico Panama Argentina Bolivia Chile Colombia Ecuador Paraguay Peru India Indonesia Malaysia Philippines Singapore Turkey

Sf/Sf

Sh/Sh

Y/Y

N/ N

DataQuality

0.071 0.065 0.098 0.144 0.051 0.043 0.050 0.036 0.031 0.027 0.058 0.051 0.023 0.060 0.021 0.049 0.038 0.023 0.045 0.043 0.034 0.007 0.027 0.075 0.025 0.179 0.019

0.186 0.176 0.152 0.060 0.223 0.101 0.097 0.138 0.045 0.059 0.077 0.106 0.178 0.176 0.097 0.119 0.089 0.143 0.205 0.081 0.124 0.163 0.195 0.227 0.171 0.105 0.201

0.041 0.043 0.052 0.079 0.086 0.052 0.045 0.014 0.021 0.035 0.016 -0.002 0.046 0.050 0.021 0.023 0.024 0.045 0.025 0.027 0.028 0.037 0.060 0.064 0.042 0.109 0.051

0.007 0.004 0.024 0.020 0.018 0.038 0.026 0.025 0.023 0.028 0.033 0.014 0.027 0.024 0.015 0.025 0.017 0.022 0.028 0.029 0.026 0.022 0.022 0.025 0.027 0.019 0.024

aab bbC c C C c c C C c c c C C C C c C C c c c c

- 20 (Continued) Table 1: The list of Countries In the Sample (All the variables are averages over the period 1965-1988)

Country

S/ Sf

Sh / Sh

Y/Y

N/N

Data Quality

Cameroon Ivory Coast Morocco Senegal South Africa Tanzania Brazil Uruguay Pakistan Sri Lanka Thailand Egypt Ethiopia Madacascar Malawi Mali

0.047 0.063 0.034 0.045 0.084 0.067 0.011 0.024 0.016 0.015 0.028 0.041 0.023 0.030 0.038 0.035

0.059 0.044 0.060 0.028 0.192 0.156 0.179 0.146 0.119 0.202 0.129 0.022 0.023 0.057 0.092 0.035

0.052 0.049 0.043 0.021 0.008 0.033 0.059 0.011 0.056 0.044 0.065 0.055 0.024 0.012 0.045 0.038

0.027 0.040 0.025 0.026 0.022 0.033 0.024 0.004 0.030 0.018 0.025 0.024 0.026 0.027 0.032 0.022

cccccccccccd+ d+ d+ d+ d+

Mauritius

0.037

0.094

0.052

0.014

d+

Sierra Leon Zambia Ghana Sudan Uganda Zaire Haiti Nacaragua

0.017 0.106 0.042 0.026 0.112 0.063 0.014 0.031

0.001 0.240 0.029 -0.008 -0.072 0.030 0.063 0.146

-0.031 0.013 0.007 0.028 0.003 0.001 0.035 0.009

0.021 0.030 0.024 0.027 0.026 0.029 0.018 0.030

d+ d+ d d d d d d

- 21 -

Table 2: Panel Data Regresions (AnnualData)

Dependentvariable: annual growth rate of income Method of Estimte

PooledOLS

Fixed-effects

Random-effects

Countries:

53

53

53

Observations:

989

989

989

0.059

0.051

0.053

(0.007)

(0.006)

(0.006)

SI/Sf

0.012 (0.004)

0.013 (0.004)

0.013 (0.004)

NIN

0.421 (0.169)

0.529 (0.311)

0.430 (0.240)

C

0.033 (0.004)

RF

0.090

0.088

0.085

F2

0.087

0.034

0.031

F(52,933)=4.33

X2(3)= 17.00

Sf/Sf

Test of Restrictions:

0.032 (0.006)

Note: Standarderrors are in parenthesis.

we discuss empirical findings, we explain the different econometric methods used in the three regressions. The first column is the result from a simpleOLS regressionusing pooled data. The second and third columnsare results from panel data regressionsusing variable-interceptsmethod. The differencebetweenthe second columnand the third cohlmn is that we use a fixed-effectsmodel for the regression in the second column, and a random-effectsmodel in the third. That is, in the

- 22 secondcolumn, we assumethat the omittedcountry-specificvariable (C,) are fixed over time, while in the third column regressionwe treat the country-specificeffects, like the error term, as random variables. Generally the two types of specificationsproducequite different results.5 At the bottom of the third column, we providethe chi-squarestatistic which can be used to test whetherthe data favor a fixed-effectsmodelor a random-effectsone. The null hypothesisis that the true model is a random-effectsmodel. If the computedchi-square statistic is larger than the critical value at a predeterminedsignificancelevel, the null hypothesisshould be rejected. From Table 2 we see the computedChi-squarestatisticis 17.0, which well exceeds the critical value for the 1 percent significancelevel at 3 degreesof freedom, which is 11.34. Thus we should reject the random-effectsspecificationin the third column and accept the fixed-effectsmodel in the second rolumn. At the bottom of the second column, we also provide the F-statistic for testing the hypothesisthat the interceptsfor differentcountriesare different(i.e., the pooled OLS model is misspecified). The computedF value is 4.33, which is much larger than the I percent critical value, This indicatesthat the pooled OLS regression,which is based on the M-R-W's assumption,is indeed mis-specified. We should reject the result in column one and accept the result from the second column. However, if we look at the estimatedcoefficientsacross Table 2, we find that, econometric theory notwithstanding,the results from all the differentregressionare very similar. That, is to say, the pooled regression producesresults similar to the panel data regression. Now consider at the estimatedcoefficients. Both estimatedcoefficientson foreign capital imports and home investment are positive and statisticallyvery significant. Furthermore, the estimated coefficient on foreign capital imports is indeed much higher than the one on domestic 5 For a detaileddiscussionaboutthe differencebetweenfixedeffectsand randomeffectsmodels,see

Hsiao 1986.

- 23 capital investment,as is predictedby our model. Thus the empiricaldata from 53 countriesshows that the level of foreign capital imports has a positive impacton the growth rate of income. The estimatedcoefficienton the populationgrowth rate is positivebut not statisticallysignificantin the fixed-effectsmodel (th,esecondcolumn of Table 2), which is the favored model. Although the results from Table 2 produce the right signs for the coefficients on the investmentof foreign capitalequipmentand that of home capital,there are severalproblems. First, as mentionedabove, the estimatedcoefficienton populationgrowth turns out to be insignificant. Second, the magnitudesof the estimatedcoefficientson the three variables (Sf 1Sf,Sh hSh, and kl/N) are too small. The implieda and (3, which are the reladve share of home capital and imported capital in production, are smallerthan 0.02 and 0.06 respectively.And the estimatedcoefficienton populationgrowth is also much smaller than 1, as the model predicted. The third problem is that the independentvariablesin all three regressionsexplainvery littleof the variationof the dependent variable, as indicatedby the extremelylow i2s. 6

3.3.2 Omitted variable problem We suspect that the above problems may arise becauseof the many omittedvariables. As mentionedin the last section,our model specificationare based on strong neoclassicalassumptions that are not true in the real world. In reality, the economicdevelopmentprocess in developing countries is affected not only by factors of production, but also by many social and institutional factors. These omittedvariablesmay cause biased estimatesin our m6del.

6Please note that the smaller R2 s

in the variable-intercept modelsare due to the fact that a large numberof constantsare used in thesemodels.

-

24 -

Thus, in Table 3, we present the regressionresults with moreexogenousvariables included in the model. The new variablesintroducedinto the regressionsare: annualinflationrate (INFLAT), black market foreign exchangerate premium(EXCHPREM),changesin the terms of trade (TOT), primary school enrollmentrate in the population(SCHOOL),growth rate of export (EXPORT). All these variablesare widelyusedby othereconomistsin empiricalstudiesgrowth. Fischer (1993)has argued that the inflationrate is a good measure of the long-runeconomicgrowth rate, because it is the best indicatorof the overallability of the governmentto manageand stabilizethe economy. If macroeconomicstabilityis good for growth, then a high inflationrate tends to lower growth rate. Levine and Renelt (1992) show that high growth countriesare also lower inflation countries,and have lower blackmarketexchangerate premia. The negativeimpactof adverseterms of trade shocks on developingcountries'seconomicgrowth has been a widely accepted fact. Th-% inclusion of the SCHOOL variable was introduced first by M-R-W (1992), and has become a standard variable in growth studies ever since. Many studies have found a positive relationship between the growth rate of export and economicgrowth. Zhang (1994)found that different sectors of export (i.e., primary exports and macturi

exports) have different impacts on the

long-run growth rate. However, because we do not have annual sectoral cross-countrydata on developingcountries' exports, we will only use a single export variablein this study. Now we look at the results in Table 3. It once again containsthree regressions. The first one is the simpleOLS regression,and the last two are panel data regressions. Note that the sample size of regressionsin Table 3 are smallerthan these in Table 2. Nine countrieswhich were in the Table 2 sampledo not have informationon some of the new variables,so they are excludedin Table 3 regressions.

- 25 Table 3: Panel Data Regressions(Annual Data) With Added Variables Dependentvariable:annualgrowthrate of income Methodof Estimate

PooledOLS

Fixed-effects

Random-effects

Sf/Sf

0.058 (0.007)

0.058 (0.007)

0.058 (0.007)

Sf/S (t -1)

0.024 (0.007)

0.023 (0.007)

0.024 (0.007)

S/

0.017 (0.007)

0.016 (0.007)

0.016 (0.007)

Si/Sf ft-3J

0.011 (0.007)

0.007 (0.007)

0.010 (0.007)

S,/Sh

0.017 (0.004)

0.016 (0.004)

0.016 (0.004)

N/N

0.901 (0.2176) -0.023 (0.005) -0.022 (0.005) 0.030 (0.011) 0.00008 (0.00008) 0.056 (0.009) 0.005 (0.006) 0.0006 (0.007) -0.0097 (0.0052) -0.017 (0.006) 0.025 (0.010) 0.325 0.310

1.170 (0.510) -0.023 (0.006) -0.010 (0.004) 0.028 (0.011) -. 00027 (0.0002) 0.045 (0.009)

0.262 0.192

0.908 (0.271) -0.024 (0.005) -0.014 (0.004) 0.030 (0.011) -0.00004 (0.0001) 0.051 (0.009) 0.006 (0.008) 0.0005 (0.010) -0.009 (0.007) -0.018 (0.008) 0.028 (0.013) 0.286 0.218

44 772

44 772 F(43,614)= 1.96

44 772 X2 (16)=28.37

f

(t - 2) J

INFLAT EXCHPREM TOT SCHOOL EXPORT DEASIA DSAS1A DLATIN DSAFRIC C R2 1?2 Countries: Observations: Test of Restrictions:

Note: Standarderrors are in parenthesis.

- 26 The F and chi-squarestatisticsare shown at the bottomof the table. The F-test once again rejects the pooled OLS regression in favor of variable-interceptmodels. The chi-squarestatistic, however, indicatesthat the fixed-effectsmodel shouldbe rejected in favor of the random-effects model. But once again we find similaritiesamongthe results in the three regressions. Results in Table 3 show several improvementsover the regressionsin Table 2. First, after introducing new explanatory variables, both R2 and R2 are indeed much higher than the correspondingregressionsexcludingnew variables. Second,the estimatedcoefficientson population growth (NIN) are very significantand close to I in magnitudein all three regressions. The estimatedcoefficientson the other key variables- foreigncapital importsand home investmentare again positiveand very signficat, and are larger than those in Table 2 in magnittde. All the newlyaddedvariablesexceptthe SCHOOLvariablehave the expectedsignsand are statisticallysignificant. The SCHOOL variable is a proxy for human capital, which should be positivelycontributingto growth. But in Table 3, the SCHOOLvariableis either insignificant(first column),or has a wrong sign (in the secondand third columns). One possiblereasonfor this result is that primary-schoolenrollmentrate in a country is not a good proxy for the measurementof human capital.7 We also include severallaggedforeign capitalimportsas exogenousvariablesin the Table 3 regressions. The estimatedcoefficientsfor these laggedforeign capital imports provide a very interestng result. They show that the current changein foreign capital imports has the strongest positiveimpact on incomegrowth, and the impactsbecomeweaker as one goesback further. This can also serveas a test of the causalrelationshipbetweenincomegrowthand foreigncapitalimports. 7Tbisnegativesignhas appearedin manyotherrecentstudies;see Jorgensonand Fraumeni(1992) andBenhabiband Spiegel(1994)for morediscussions.

- 27 -

Sincebothone-year and two-yearlaggedforeign capitalinvestmenthave positiveimpactson income growth,the causalrelationshipis likelyto be from the formerto the latter, rather than the other way around. Finally, notice that we put regionaldummy variablesfor differentregions in the equation (East Asia, SouthAsia, Latin America,and Sub-SaharanAfrica).8 The countriesin the base group are non-OECDEuropeancountries(Greece, Portugal, Turkey), North African countries(Egypt, Morocco), and South Africa and Israel. Table 3 shows that only the coefficientfor Sub-Saharan Region is significantlynegative. 3.3.3 Annualdata vs. longer time span AlthoughTable 3 resultsshowa significantimprovementthanthose in Table 2, thereremains the problemthat the estimatedcoefficientson the growthrate of foreigncapital importsand on that of home capitalinvestmentare still too small in magnitude. Furthermore,the reportedR2 's are still not very high relativeto the onesin othersimilarstudies(for example,see Levineand Renelt 1992). We suspect that the problem may arise from the use of the annualdata, which containtoo much noise and short term disturbancesthat do not reflectlong-runtrends, and are not captmredin the exogenousvariables in the model. One way to smooththeseshort term disturbancesis to use a longer time span. We thus divide the total period of 1965-88into several5-year time intervals. More specifically,we will have four observationsfor each country, i.e., 1970, 1975, 1980, and 1985. Whent = 1985,t - 1 is 1980. All the growthrate variablesare averagesover the five year time span. This set-upwouldalso reducethe serial correlationbetweenthe JiA's.

8 Thefixed-effects modeldoesnotprovideestimatesfor regionaldummiesbecausethe fixedindividual

country-specific interceptsalreadyaccountfor theseindividualcountryeffects.

- 28 Table 4: Panel Data Regressions (5-Year Time Interval)

Dependentvariable: 5-year average growth rate of income Method of E timate

PooledOLS

Fixed-effects

Random-effects

St/Ssf

0.165 (0.020)

0.147 (0.016)

0.155 (0.015)

ShiSh

0.024 (0.019)

0.051 (0.020)

0.030 (0.016)

N/N

0.715 (0.271) 0.026 *(0.007) 0.400

1.207 (0.533) 0.609

0.833 (0.312) 0.023 (0.080) 0.498

0.385

0.335

0.147

49 125 F(48,73)=3.29

49 125 X2(3)=5.52

C 2 R2 Countries: Observations: Test of Restrictions:

49 125

Note: Standarderrors are in parenthesis.

Table 4 presents the regressionresults using 5-year time intervaldata. The regressions in Table 4 use the basic model withoutadded variables,correspondingto these in Table 2. The first apparent result in Table 4 is that the R2 's are improved greatly comparedto the corresponding results in Table 2 or even the larger-variableregressionsin Table 3. The secondthing to notice is that how once again similar the estimatesfrom the three regressionsare. The most important result in Table 4 regressions is that the estimated coefficients on SfISf and Sh ISh are not only positive and very significant, they are also much larger in magnitudethan those estimatedwith annualdata.

- 29 Table 5 shows the fixed-effectspanel data regressionsusing 5-year time intervalsdata with different groups of added explanatoryvariables. Since we have seen in all the previoustables that the results from fixed-effectsmodel and random-effectsmodel are very similar, we do not present the results from the random-effectsregressions. Again, all the estimatedcoefficientson foreign capital importsare positiveand significant. This result stronglysupports our modelpredictionthat foreign technologytransferis one of the most importantfactors in explainingthe differenteconomic growth rates among developingcountries. For comparison,in Table 6, we present thepooledOLS regressionswiththe sameexogenous variables as in Table 5. One can see that the resultsin Table 6 are very similar in those in Table 5. Thuswe have demonstratedthat for practicalpurposes,pooled regressionsproduceresultssimilar to results from panel regressions.

4. CONCLUSIONS In summary,we first developeda model specificationthat links the growthrate of income with that of foreign capital imports' share in GDP andhome investment'sshare in GDP. Then we ran regressions with a sample of around 50 developingcounties, using different econometric methods and differenttime spans. Severalconclusionscan be drawn from this study. First, our empirical tests confirm our theoretical model predictionthat foreign technologytransfer has a positive impact on the income growth rate in developingcounties. All the results confirm the hypothesisthat foreign technologyimportsare a key element in explainingthe differencesin the growth rates of income among developingcountries. The economic developmentprocess in developingcountriesis differentfrom that in developedcountries. More specifically,the increases

- 30 -

Table S: Fixed-EffectsPanel Data ReMressions(5-Year Time Intervafl

Dependentvariable:5-year averagegrowthrate of income

0.132 (0.016)

0.111 (0.018)

0.115 (0.017)

0.102 (0.018)

0.197 (0.036)

Sh / Sh

0.037 (0.019)

0.027 (0.018)

0.036 (0.019)

0.028 (0.018)

0.030 (0.017)

N/N

1.107 (0.559) -0.036 (0.011) -0.021

0.819 (0.492) -0.020 (0.011) -0.028

1.157 (0.596)

1.394 (0.643) -0.021 (0.012) -0.024

2.260 (1.140) 0.004 (0.016) -0.012

(0.010)

(0.009)

INFLAT EXCHPREM TOT

0.031 (0.046)

(0.010)

-0.124 (0.063) -0.018 (0.018) -0.000003 0.00001 -0.0002 (0.0004) -0.012 (0.019)

0.721 0.495

0.735 0.518

0.681 0.439

0.794 0.562

0.120 (0.032) 0.926 0.634

42 106 F(41,58) =2.81

44 112 F(43,61) =3.26

46 119 F(45,67) =2.61

40 101 F(39,47) -1.83

37 65 F(36,12) =1.91

F.CONSUM GDP(t- 1) SCHOOL EXPORT Sf/ Sf (t -1) R2 R2 Countries: Observations: Test of Restrictions:

(0.019)

-0.051 (0.045) -0.011 -0.004 (0.015) (0.016) -0.000009 -0.00001 -0.00002 (0.000004) (0.0000D4) (0.000006) -0.0003 -0.0002 (0.0003) (0.00032) 0.060 0.040 (0.030) (0.030)

Note: Standarderrors are in parenthesis.

- 31 Table 6: Pooled OLS Regressions(5-Year Time Interval) Dependentvariable:5-year average growthrate of income

Sf/JSf

0.137 (0.018)

0.135 (0.020)

0.126 (0.019)

0.113 (0.017)

0.145 (0.021)

Sh / Sh

0.047 (0.029)

0.032 (0.020)

0.050 (0.019)

0.039 (0.017)

0.028 (0.016)

N/N

0.717 (0.250) -0.027

0.805 (0.265) -0.031

1.076 (0.282)

0.948 (0.287) -0.025

.998 (0.338) -0.022

(0.009)

(0.010)

(0.009)

(0.009)

-0.028 (0-.008) -0.013

-0.028 (0.009)

-0.014 (0.009) -0.0003

-0.014 (0.010) -0.018

(0.039)

(0.050)

-0.0003 (0.016) -0.000006 (0.000002) -0.0002

-0.036 (0.020) -0.000004 (0.000003) -0.00009

(0.00009)

(0.00010)

(0.0001)

0.150 (0.028)

0.124 (0.027)

0.146 (0.038)

0.0016 (0.0067) -0.0027 (0.0076) -0.0022 (0.0057) -(0.0149)

0.054 (0.024) -0.0003 (0.007) -0.010 (0.008) -0.012 (0.006) -0.022

INFLAT EXCHPREM TOT F.CONSUM

(0.044)

GDP(t -1) SCHOOL

0.019 (0.018) 0.000002 -0.000002 (0.000002) (0.000002) 0.0002

EXPORT Sf/ Sf (t-1) DEASIA DSASIA DLATIN DSAFRIC C R2 R2 Countries: Observations:

0.036 (0.007) 0.582 0.557 42 106

0.033 (0.009) 0.579 0.550

-0.0056 (0.011) 0.563 0.540

44 112

46 119

Note: Stndard errors are in parenthesis.

(0.0067)

(0.007)

0.022 (0.013) 0.744 0.703

0.026 (0.015) 0.842 0.793

49 101

37 65

- 32 -

of productivityin developingcountriesdo not rely on innovationbut on inporting foreign plant and equipmentand on borrowingforeign technology. Second, althougheconometrictheory shows that M-R-W's OLS assumptionwouldproducea biasedresult, for all practicalpurpose, OLS regression results are as good as panel regression results. However, one thing to note is that in all our regressions, the F-tests demonstratethat although the heterogeneousintercept and homogeneous slopes specificationis a better model than the simpleOLS, it shouldbe rejectedin favor of models allowed for hetergeneityof intercept and slopes. That is, the data call for individualcountry regressions. However, since we do not have enough observations for each country to allow individualcountryregressions,ibis is a candidatefor futurestudy of the relationhip betweenimcome growth and foreign technologyimports.

- 33 -

Oxford Amsden,Alice. 1989. Asia's Next Giant: South Koreaand Late Industrializadon. UniversityPress. Arrow, K.J. 1962. "Tbe EconomicImplicationof Learning by Doing. Review of Economic

Studies,29 (June).pp. 155-173. Bacha,Edar L. "Growthwith LimitedSuppliesof Foreign Exchange:A Reappraisalof the TwoGap Model" in Moshe Syrquin, Lance Taylor, and Lany Wespal, eds., Economic

Structureand Performance: Essaysin Honorof HollisB. Chenery.New York:Acadmc Press, 1984.

Economics Balassa,Bela 1978."ExportsandGrowth:FurtherEvdence",JournalofDevelopment 5, no. 2, 181-189. Barro, R. 1991. 'Economic Growth in a Cross Secdon of Countries." Quartery Jouma of Economis. May.

Benhabib,J. andM. Spiegel. 1994.'The roleof hma capitalin econmic development: Evidence from aggregatecross-countrydata.' Joura of MonetaryEconoics 34, pp. 143-173.

Bochove,ConelisA. van. 1982, Importsand EconomicGrowth. The Haue: Marims Nijhoff Publishers. Chenery, Hollis, and MichealBnmo. 1962. 'DevelopmentAlternativein an Open Economy:tbe Case of Israel,' EconomicJournal. Vol. 57, March 1962, 79-103. de Melo, Jaimeand Sherma Robinson. 1990."ProductivityandExternis: Growth," Workingpaper no. 387, The World Bank.

Modelsof Exrt-led

- 34 Devarajan,Shantaand Heng-fu Zou. 1993. "Exports,Foreign TechnologyImports, and Long-run Growth", mimeo, Policy ResearchDepartment,The World Bank. 1994. "ExportExternalities,ExportPricing and EndogenousGrowth", mimeo,Policy

______.

ResearchDepartment,The WorldBank. Esfahani, Hadi S.

1991. "Exports, Imports, and Economics Growth in Semi-industrialized

Countries," Journal of DevelopmentEconomics,35(1), pp. 93-116. Fischer, Stanley. 1993. "The role of macroeconomicfactors in growth." Unpublishedpaper, Departmentof Economics,MIT. Grossman, G.M., and Helpman, E.

1991. Innovation and Growth in the Gla

Economry.

Cambridge,MA: MIT Press. Hsiao, Cheng. 1986. Analysisof PanelData. CambridgeUniversityPress. Islam, Naznul.1992. "GrowthEmpirics:A PanelData Approach."Unpublishedpaper, Deparment of Economics,Harvard University. Jones, L. and R. Manuelli. 1990. " A Convex Model of Equilibrium Growth.' Journal of Poliical Economy, October. Jorgenson,Dale W. andBarbaraM. Fraumeni. 1992. "Investmentin Educationand U.S. Economiic Growth." The ScandinavianJournal of Economics,Vol. 94, Supplement,pp. 51-70. Kaldor, N. 1967. StrategicFactorsin EconomicDevelopment. Ithaca, NY: Cornell University Press. _

. 1978. "Causes of the Slow Rate of Growthin U.K." Essays in Economic Theory.

London:Duchworth.

- 35 King, R. and Rebelo, S. 1990. 'Public Policy and EconomicGrowth.' Journal of Political Economy,October. Levine, Ross, and Renelt David. 1992. 'A SensitivityAnalysis of Cross-Country Growth Regressions," 7he AmericanEconomicReview, 82:942-63. Lucas, R.E., Jr. 1988. "On the Mechanicsof EconomicDevelopment." Journal of Monetary Economics.No. 22, pp. 3-42. Mankiw, G., D. Rommer, and D. Weil. 1991. "A Contributionto the Empirics of Economic Growth." QuarterlyJournal of Economics. McKinnon, Ronald. 1964. "ForeignExchangeConstraintsin EconomicDevelopmentand Efficient Aid Allocation," EconomicJournal, Vol. 74, p. 388-409. Romer, P. 1986. "IncreasingReturnsand Long-RunGrowth." Joumal of Politcal Economy,94, pp. 1002-1037. Solow, R. 1956. "A contributionto the theory of economicgrowth." The QuarterlyJournal of Economics,70 (February):65-94. *957.

"TechnicalChange and The AggregateProductionFunction." The Review of

Economicsand Statistics39, pp. 312-320. ______*

1970. GrowthT7heory: An Exposition.New York and Oxford: Oxford UniversityPress.

Summers, Robert, and Alan Heston. 1991. "The Penn World Table (Mark 5): an ExpandedSet of Iernational Comparisons,1950-1988," The QuarterlyJournal of Economics(May), pp. 327-68. Taylor, Lance. 1991. IncomeDistribution,Inflaion, and Growth. Cambridge,MA: MIT Press.

- 36 Tyler, WilLiam. 1981. "Growth and Export Expansionin DevelopingCountries," Journal of DevelopmentEconomics9, 1981, pp. 121-130. United Nations Conferenceon Trade and Development. Handbook of InternationalTrade and DevelopmentStatisftcs. New York: United NatinnsPublicaton. Variousyears. United Nations.Internaonal Trade StadsticalYearbook. 1992. United Nations StatisticalOffice. Uzawa, H.

1965. "OptimumTechnologicalChange in an Aggregative Model of Economic

Growth.' InternatonalEconomicReview 6 (January). pp. 18-31. The World Bank. WorldDevelopmentReport. Variousyears. The World Bank. 1990. WorldTablesof Economicand Social Indicators. InternationalEconomics Department,Computerfile. Am Arbor, MI: Inter-universityConsortiumfor Political and Social Research(distributor).

Zhang, Xiaoming. 1994. "Different Sectors of Exports and Economic Growth.'

Mimeo,

Departmentof Economics,MIT. Zou, Heng-fu. 1993. "Product Inovation, Capital Accumulationand Endogenous Growth." Mimeo, Policy ResearchDqetment, The World Bank.

Deflnltlonsand Sourcesof Variables Y/Y:

Averageannualgrowthrateof GDP. Source: WorldTables,WorldBank,1992.

Sf / Sf:

Averageannualchangeof theshareof foreigncapitalimportin GDP. Thisvariable is calculatedbasedon data from two sources:data on the dolUarvalueof foreign capitalare from UnitedNations'Intrtional TradeStatisticalYearbook;and the share of totalimportin GDPis from SummersandHeston(1991).

Sh / Sh:

Averageannualchangeof theshareof homeinvestment in GDP. Thisis calculated as the log differenceof shareof homeinvestmentin GDP, and the shareof home investmentin GDP in urVin calculatedby subtractingSf from the share of total investmentin GDP. Thelatteris frm SummersandHestondataset.

i

N I N:

Averageannualgrowthrateof population.Soure: Summersand Hestondataset.

INFLAT:

Averageamual inflationrate, computedas the log-difference of CPI. Source: Intermational FmawcialStatisdics, CD-ROM,June, 1993. GDPdeflatordata from the WorldBankwereusedto extendinflationsenes for Malawi.

EXCPREM: Averageblackmarketexchangerate premium.Source: WorldBank,World DevelopmentReport,1991.[Computerfile] TOT:

Changein termsof trade,calculatedas the logdifferenceof the net termsof trade in a timeperiod. Sources: WorldBank,WorldDevelopment Report,1991 dataset.

F.CONSUM:Annualaveragechangein the shareof foreignconsumption iMportin GDP. Sources:UN'sInternational TradeStatisticalYearbook,and Summersand Heston.

- 38 GDP(t-l):

Real GDP 5 years before current 5-year period. This variable is used here as an substitutionfor the initial GDP level in Mankiw's single period regressionmodel. Source: Summersand Heston(1991).

SCHOOL:

Primary schoolenrollmentas percentageof age group. Source:same as above.

EXPORIT:

Average annualgrowth rate of export, weightedby the share of export in GDP. Source: Summersand Heston.

Policy Research Working Paper Series Tbtle

Author

Date

Contact for paper

UnDadush WPS1397 ArePrvateCapitalFlowsto CountriesSustainable? AshokDhareshwar Developing RonJohannes

December1994

J. Queen 33740

WPS1398TheCostof Air PollutionAbatement RaymondS. Hartman DavidWheeler ManjulaSingh

December 1994

E. Schaper 33457

WPS1399HowImportantto India'sPooris the MartinRavallion Urban-Rural Compositonof Growth? GauravDatt

December 1994

P. Cook 33902

WPS1400TechnicalandMarketingSupport Systemsfor SuccessfulSmalland Enterprises in Four Medium-Size Countries

BrianLevywith December 1994 AlbertBerry,MotoshigeItoh, UnsuKim,JeffreyNugent, andShujiroUrata

D. Evans 38526

WPS1401Colombia's SmallandMediurn-Size AlbertBerry ExportersandTheirSupportSystemsJose Escandon

December 1994

D. Evans 38526

SmallandMedium-Size AlbertBerry WPS1402Indonesia's ExportersandTheir SupportSystemsBrianLevy

December 1994

D. Evans 38526

WPS1403SmallandMedium-Size Enterprise SupportPoliciesin Japan

MotoshigeItoh ShujiroUrata

December 1994

D. Evans 38526

WPS1404The Republicof Korea'sSmalland Medium-Size Enterprises andTheir SupportSystems

LinsuKim JeffreyB. Nugent

December 1994

D. Evans 38526

WPS1405GrowthandPovertyin RuralIndia

MarlinRavallion GauravDatt

January1995

WDR 31393

WPS1406StructuralBreaksandLong-Run Trendsin Commodity Prices

JavierLe6n Raimundo Soto

January1995

R.Luz 31320

WPS1407Pakistan'sAgricultureSector Is 3 to 4 PercentAnnualGrowth Sustainable'

RashidFaruqee

January1995

F.Willie 82262

WPS1408Macroecon' c Management and lntergoverr ntalRelationsin China

JunMa

January1995

C.Jones 37754

WPS1409Restructurir,lUganda'sDebt DebtBuy-Back TheCommercial Operation

KapilKapoor

January1995

E. Spano 35538

WPS1410Macroeconomic Effectsof Termsof-TradeShocks:TheCaseof OilExportinig Countries

NikolaSpatafora AndrewWamer

January1995

J. Queen 33740

Policy

Research Working

Title

Paper Series Contact for paper

Author

Date

WPS1411 Income Inequality,Wellarm,and Poverty: An IllustrationUsing Ukrainian Data

NanakKakwanl

January 1995

G. Evans 85783

WPS1412 ForeignTechnologyImportsand EconomicGrowth in Developing Countries

XiaomingZhang Heng-iuZou

January 1995

C.Jones 37754