Journal of Agricultural and Applied Economics, 44,4(November 2012):517-531

© 2012 Southern Agricultural Economics Association

From Coffee Beans to Microchips: Export

Diversification and Economic Growth in
Costa Rica
Gustavo F.C. Ferreira and R. Wes Harrison

In the wake of a severe economic crisis in the 1980s Costa Rica abandoned an import sub-
stitution model of development adopted in the 1960s and implemented policies supporting
foreign investment and the diversification of its exports. This study presents an application
of the model proposed by Herzer and Nowak-Lehnmann to test the hypothesis that export
diversification has contributed to economic growth in Costa Rica via externalities of learning-
by-doing and learning-by-exporting over the period of 1965-2006. After using the autore-
gressive distributed lags and dynamic ordinary least squares models no long-run relationship
was found between export diversification and growth.

Key Words: Central America, Costa Rica, export diversification, economic growth, time series

JEL Classifications: F14, 054

Costa Rica is an interesting case study in eco- a well-educated labor force. According to the
nomic development not only because of its long Heckscher-Ohlin trade model, given Costa
democratic tradition and relative economic sta- Rica's endowment of a well-educated work-
bility, but also because the economy of this force, this country has a comparative advantage
small nation has evolved from being heavily in the production of knowledge-intensive goods
reliant on exports of coffee and bananas to be- (Learner, 1995). Another important character-
coming the largest software exporter per capita istic of Costa Rica is the small size of its domestic
in Latin America. As the World Bank states market. This limits the capability of sustained
".. .it has evolved from the production of its growth in its Gross Domestic Product (GDP) and
"golden bean" (high quality coffee beans) to reduces the chances of producing certain goods
the "Golden chip" (World Bank, 2006). In ad- that are subject to economies of scale. Export
dition, and as a result of decades of policies growth and export diversification could be the
emphasizing universal education and health solution to these constraints, and a possible ex-
care to its population, Costa Rica, today, has planation of why international trade has played
such an important role in this country's economy.
Well aware of these realities, Costa Rican
Gustavo F.C. Ferreira is an instructor and extension authorities have been playing a very active role
economist, Department of Agricultural and Applied in the development of a domestic industrial sec-
Economics, Virginia Tech, Blacksburg, Virginia. R.
Wes Harrison is a Warner L. Bruner Professor, De- tor during the 1960s and 1970s, in the recent
partment of Agricultural Economics and Agribusiness, diversification of the nation's economic activ-
Louisiana State University Agricultural Center, Baton ities, and in the attraction of investments from
Rouge, Louisiana. high-tech multinational firms. The outcome of
We thank Christopher F. Baum, P. Lynn Kennedy, these policies has been decades of sustained
and the anonymous reviewers for their comments.
518 Journal of Agricultural and Applied Economics, November 2012

economic growth. This paper seeks to test the is often cited as a mechanism through which in-
hypothesis that both vertical and horizontal stability of export earnings can be reduced, and
export diversification has positively influenced thus mitigate any unfavorable effects on gov-
economic growth in Costa Rica via externali- ernment revenues, investment, import capacity,
ties of learning-by-exporting and learning- and producers' incomes. In a seminal paper, Love
by-doing. This hypothesis is tested using two (1986) proposed that countries should avoid
econometric procedures, an autoregressive dis- having a heavy concentration of their exports in
tributed lags (ARDL) model and a dynamic a few products because this reduces a nation's
ordinary least squares (DOLS) model. capability of partially offsetting fluctuations in
This paper is organized as follows: Section some export sectors with counter fluctuations,
two discusses the literature on the linkages be- or stability in other sectors. His findings con-
tween export diversification and economic cluded that export concentration had a positive
growth. Section three gives an overview of and significant influence on instability of ex-
Costa Rica's export diversification experience. port earnings. Jansen (2004) demonstrates that
Section four presents the theoretical model and income volatility in small economies is ex-
data sources, while section five describes the plained, to a great extent, by their high level of
econometric methodology employed in this pa- economic openness and by their lack of export
per. Section six offers the empirical results, and diversification. In another study, Al-Marhubi
section seven concludes and discusses some of (2000) hypothesizes that instability in export
the policy implications. earnings is a major source of economic uncer-
tainty in many commodity-exporting nations
because under unstable domestic markets, in-
Review of Empirical Literature
vestment becomes riskier in those nations. In
other words, increasing instability in a nation's
There is a large body of literature that has in-
export earnings may discourage investment,
vestigated the linkages between export diver-
which in turn negatively impacts economic
sification and long-term economic growth, and
growth. Using a cross-country sample of 91
attempted to answer two important questions.
countries, a positive and robust relationship
Does export diversification have any effect on
between export diversification and economic
long-run economic growth? Is it possible for
growth was found. Hesse (2008) presents an
a country to improve its economic performance
extensive literature review on export diversi-
by exporting different types of goods? (Gutierrez-
fication and economic growth, and estimates
de-Pineres and Ferrantino, 2000). a simple augmented Solow growth model to
investigate the relationship between export di-
The Effects of Export Diversification versification and income per capita growth.
on Long-Run Economic Growth Once again, evidence that export concentration
is detrimental to economic growth in developing
A number of studies have presented evidence countries was found. Feenstra and Kee (2004)
that export diversification is conducive to higher analyze the effects of sectoral export variety on a
per capita income growth. The generally pro- country's productivity. After estimating a trans-
posed hypothesis is that nations with more di- log GDP function system for a sample of 34
verse economic structures are more likely to countries, their results show that a 10% increase
consistently sustain periods of high economic in export variety of all industries leads to a 1.3%
growth than nations with more concentrated increase of a country's productivity.
export structures. Empirical growth literature
has shown that income volatility has a negative Other empirical studies have tested the pos-
impact on a nation's economic growth. Along itive links between export diversification and
this line of thought, the so-called "portfolio ef- economic growth for specific regions or coun-
fect" is a widely accepted argument in favor of tries. The study of Gutierrez-de-Pineres and
export diversification that has been borrowed Ferrantino (2000) for Latin American countries
from the finance literature. The portfolio effect finds associations between episodes of export
Ferreira and Harrison: Export Diversification and Economic Growth in Costa Rica 519

diversification and rapid economic growth. produces larger externalities than other eco-
Moreover, export specialization was signifi- nomic sectors. These externalities are impor-
cantly and negatively correlated with economic tant in the sense that they may result in further
growth after controlling for other common de- horizontal diversification and improvements in
terminants of growth. Also in Latin America, the ability of all industries to compete inter-
Gutierrez-de-Pifieres and Ferrantino (1999) iden- nationally (Matthee and Naude, 2007).
tified examples of countries where knowledge The ratio of manufactured exports to total
gained from exporting activities was later uti- exports is a good indicator of the degree to which
lized by other exporters. This knowledge can an economy managed to develop forward link-
take several forms such as the diffusion and ages and reduced its dependence on the primary
awareness of export opportunities, diffusion of sector. Levin and Raut (1997) conclude that
transportation and production technologies, and increases in the ratio of manufactured exports
development of domestic services (i.e., insur- to total exports have positive and significant
ance, banking, etc.). In the case of Colombia, impact on economic growth, whereas growth of
exports of fresh cut flowers were followed by the primary goods exports share has a negligi-
other highly perishable goods. However, after ble effect. Fosu (1990) also tests the effect of
applying cointegration and error-correction meth- manufactured exports on growth comparatively
odologies, the authors found no long-run effect of to ones from the primary sector and concludes
export diversification on economic growth. In that, in developing countries, the former has a
Chile, the export success of table grapes was positive impact in the economy. In another
later followed by the export of an array of fresh study, Moreno-Brid and Perez (2003) studied
fruits. Herzer and Nowak-Lehnmann (2006) the effect of the external sector on the long-run
studied the Chilean experience and tested the rate of economic growth for three Central
hypothesis that export diversification has had American countries: Costa Rica, El Salvador,
an impact on economic growth via externalities and Guatemala. In the case of Costa Rica, the
of learning-by-doing and learning-by-exporting. shift from exports of primary commodities to
Using time series econometrics their results more manufacturing/high-technology goods
show that both horizontal and vertical export was found to increase the income-elasticity
diversification have a positive effect on growth. of its exports. Finally, Balaguer and Cantavella-
At the regional level, Matthee and Naude (2007) Jorda (2004) demonstrated that the structural
show that South African regions with more di- transformation in export composition was a key
versified export supplies experienced higher eco- factor for Spain's economic development. In
nomic growth rates and contributed more to the addition, their findings lend support to the idea
nation's overall exports. Furthermore, only hori- that allocation of resources toward more in-
zontal diversification, and not vertical diversifi- dustrialized export sectors has a positive impact
cation, is associated with higher economic growth. on the economy.
Despite the fact that the literature has iden-
Linkages between Economic Performance tified strong links between export diversification
and the Exports of Different Products and economic growth, the discussion is by no
means closed. Empirical research is still limited
Several studies have tested the hypothesis that to a few cross-country and country level studies,
the exports of certain products have different and Costa Rica represents a very specific case
effects on a nation's economic growth. Greena- where this relationship is not clear and merits
way, Morgan, and Wright (1999) disaggregated investigation.
exports into key components based on the ar-
gument that different components have different Overview of Costa Rica's Export
effects on GDP growth. Their findings suggest Diversification Experience
not only that export growth is an important driver
of economic growth, but also corroborate the Until the second half of the twentieth century,
widely held view that the manufacturing sector Costa Rica was an agro-exporting economy
520 Journal of Agricultural and Applied Economics, November 2012

highly dependent on the exports of a few national economy. 1 The FTZs coupled with
agricultural products. Coffee and bananas alone Costa Rica's relatively educated population,
accounted for almost 90% of the value of its political stability, and a series of pro-investment
total exports, and drove economic growth through public policies allowed the country to become
the 1960s (Mesa-Lago, Arenas-de-Mesa, and an important offshore manufacturing and cus-
Brenes, 2003). Aware of the vulnerability of this tomer service location for a number of multi-
commodity-export model to external shocks, national corporations.
Costa Rican authorities implemented a new de- The implementation of these export promo-
velopment strategy that would lead the country tion and export diversification policies during
through an economic transition during the 1960s the second half of the 1980s, and throughout
and 1970s. The country veered toward a model of 1990s transformed Costa Rica's export supply.
development based on industrialization through The share of manufactured exports to total ex-
import substitution, in particular of consumer ports increased substantially, and for the period
goods. For that, Costa Rica imposed high tariff of 1992-2000 these exports became the main
rates for consumer goods, and maintained low contributor to economic growth. At the same
import taxes for intermediate and capital goods. time Costa Rica managed to reduce its depen-
In addition, export taxes were applied on those dency on the exports of few primary goods, and
goods in which Costa Rica had a strong com- now has flourishing high-tech and medical
parative advantage (Cattaneo, Hinojosa-Ojeda, equipment manufacturing export sectors as well
and Robinson, 1999). as diversified agricultural and service sectors.
The import substitution industrialization (ISI) However, and despite these achievements, overall
strategy was relatively successful in creating a economic growth never reached the levels of the
domestic industrial sector and resulted in high ISI period. The average rate of economic growth
rates of economic growth for more than two between 1965 and 1979 was 6.15% in compari-
decades. However, in the beginning of the son with a 5% average observed between 1984
1980s, Costa Rica went through its worst eco- and 2007. According to Vos et al. (2006), once
nomic crisis since World War II, which evi- the production from the export-processing re-
denced some of the shortcomings of the ISI gimes is excluded, a more modest economic
model. With the support of international finan- performance is revealed. This may be explained
cial and developmental organizations, Costa by the fact that export diversification in Costa
Rica adopted new policies of development Rica has been directly dependent on the estab-
that would include export promotion and ex- lishment of foreign firms in FTZs and linkages
port diversification. This new economic out- between the export enclaves and the rest of the
ward orientation secured a wide consensus economy may be very limited in nature. Figure 1
among Costa Rican policy makers, and im- shows the ratio of foreign direct investment
portant structural reforms were implemented (FDI) to Costa Rica's GDP. A steady increase of
throughout the 1980s. As part of this new this measure can be observed since the appli-
export-led model, Costa Rican authorities cation of the economic structural reforms in the
established free trade zone (FTZ) regimes 1980s. This increase in FDI was only interrupted
where fiscal and economic incentives were around the year 2000 due to an economic re-
granted to those firms locating there. This cession that affected major industrialized na-
policy was an important step toward the pro-
tions. Nevertheless, a few years later FDI in
motion of new exports and attraction of foreign
Costa Rica resumed its growth, and actually
firms. Mitchell and Pentzer (2008) observe that
accelerated in more recent years. Today, multi-
most exporting firms located there are large
national firms operating in the FTZs are the
foreign companies attracted by the incentives
offered by the Costa Rican authorities. The
most illustrative example was the investment
1
of Intel in a microprocessor plant in Costa Rica For good discussions on the impact that Intel has
in 1997, and its indisputable impact on the had on Costa Rica's economy see Larrain, Lopez-Calva,
and Rodriguez-Clare (2000) and World Bank (2006).
Ferreira and Harrison: Export Diversification and Economic Growth in Costa Rica 521

7.00

^ x ^& A*
,°> & ^d> ^J ? ^& ^ ^ ^ #'# .<& <# .«#.«# -c^ * & j & ^
N^ ^ ^ N«° &> & N«° ^ rp r^ $
Year

Figure 1. Foreign Direct Investment as a Percentage of Costa Rica GDP (1970-2007) (Source:
World Development Indicators, World Bank [2008])

nation's main exporters surpassing Costa Rican as a positive externality in Equation (1). This
firms, which remain more oriented toward the knowledge externality has two main properties.
domestic and Central American markets. This One is that knowledge spillovers are primarily
anecdotal evidence raises questions as to what generated by export sectors as a result of both
degree export diversification in Costa Rica has learning-by-exporting and learning-by-doing.
generated the externalities and spillovers iden- Learning-by-exporting arises when an export
tified by the economic literature. sector acquires knowledge from its foreign
purchasers who share part of their know-how
Theoretical Model and Data and offer advice on productivity enhancement.
On the other hand, the central premise of learning-
The Model by-doing is that knowledge creation occurs as a
byproduct of production and it depends on the
This section presents a generalization of the firm's cumulative output. Hence, firms will in-
model proposed by Herzer and Nowak- crease their stock of knowledge as they expand
Lehnmann (2006) to test the hypothesis that their exports, and this accumulation process
export diversification has influenced economic will accelerate as a firm exposes itself to com-
growth in Costa Rica via externalities of learning- petitive international markets.
by-exporting and learning-by-doing. The model It is assumed that each export sector St pro-
assumes that the economy is composed of a duces an equal amount of public knowledge p.
total of n sectors from which S are export Hence, a nation's level of aggregated knowl-
sectors, thus S € n. It is also assumed that there edge is given by the following equation:
is only one firm in each sector, and that at a
(2) P, = StPt.
given point in time, t, the production function
of each s e c t o r / € [1, n] is characterized by Because p, is not directly observable and is
a neoclassical production function: assumed as a constant parameter, the level of
knowledge in the economy can be expressed
(1) Yfi = Fft(Kfi,Llt,Pl), instead as a function of the number export
sectors without including pt:
where Yfi is the output of a sector, and Kft and
Lft are standard capital and labor inputs, re- (3) P, = Z(S),
spectively. The input P, corresponds to an index In their study, Herzer and Nowak-Lehnmann
of public knowledge in period t, and is regarded (2006) assumed that primary goods tend to have
522 Journal of Agricultural and Applied Economics, November 2012

a lower potential for learning-by-doing and and vertical export diversification influence
learning-by-exporting comparatively to man- economic growth via externalities of learning-
ufactured goods. Consequently, they hypothe- by-doing and learning-by-exporting. That is, v|/
size that the pace of knowledge creation in the and y are greater than zero.
economy will increase with increases in the share To empirically test the long-run relationship
of manufactured products in total exports. Based between growth and export diversification Equa-
upon this premise, a new knowledge equation can tion (7) is transformed into a log-linear regression
take the following form form
(4) P, = Z{S„ MX,), \nY, = a + pin/f ( -l-einL, + \|/lnS,
where the share of manufactured products in (8) + ylnMX, + nt,
total exports (MXt), and the number of export where In is the natural logarithm of the vari-
sectors (St) are used as proxies for the stock of ables, and the estimates of P, 0, \\f, and y rep-
knowledge in the economy. resent elasticities. The error term (Xt is assumed
The second main property of this model is to be white noise normally and identically dis-
that the level of aggregated knowledge Pt is tributed. Equation (8) will be used to test the
considered a public good and constant within
diversification-led growth hypothesis for the
all sectors. It is assumed that Pt affects all
manufacturing sector:
sectors equally, and by treating Pt as a given, Fft
H0: \|/, y = 0
behaves as a constant-returns-to-scale production
function. It is also assumed that all firms operate Hi: \|/, Y > 0.
in perfect competition and are price takers. Next, Accordingly, it is hypothesized that the esti-
the components of the production function are mates of V|/ and y are both positive and statistically
set: significant, thus confirming the diversification-led
growth hypothesis.
K
(5) Y, = £ Yf„ K, = fl f" L> = E **>
/=i /=i /=i
The Data
where Y, is Costa Rican real GDP measured in
constant prices (2000 U.S. dollars). Now, Yt can
To estimate Equation (8) annual data for the
be rewritten as function:
period of 1965-2006 are used for all variables.
n
S, represents the number of export sectors
(6) Y, = J2Yft = W, Lt, Pt) classified by the Standard International Trade
/=i
Classification at the three-digit level, and has
Equation (7) is obtained by inserting the
been gathered from the United Nations dataset.
public knowledge parameter of Equation (4) into
The data for the remaining variables are from
the production function expressed by Equation
the 2008 World Development Indicators online
(6). Equation (7) is then expressed as a Cobb-
version. First, Yt represents Costa Rica's real
Douglas production function.
GDP, while K, represents gross fixed capital
(7) Y, = F{(K„ L,)(S„ MX,)) = aKfl*$MXj formation and is used as a proxy for capital
accumulation. These two variables are mea-
K, and L, represent the stock of accumulated sured in inflation-adjusted U.S. dollars and the
capital and labor force of the economy re- year 2000 is used as the base year. The series L,
spectively, and the parameters p, 8, \|/, and y are corresponds to Costa Rica's total labor force
constants. The shift parameter a is included in given by the economically active population
Equation (7) to account for all the influences on (EAP). The EAP comprises persons of either sex
growth other than capital and labor. By adding above a specified age who furnish the supply of
the number of export sectors and the share of labor for the production of economic goods and
manufactured exports as explanatory variables services. Finally, MX, corresponds to the share
to Equation (7), it is implied that both horizontal of manufactured exports to total exports and it
Ferreira and Harrison: Export Diversification and Economic Growth in Costa Rica 523

is expressed in percentages. Complete variable Test for Multivariate Cointegration -Auto

definitions and data sources are provided in Regressive Distributed Lag
Appendix 1.
Before testing the proposed empirical model, a
Econometric Methodology discussion of the ARDL approach to cointegration
is presented to justify the choice of this meth-
Tests for Univariate Integration odology over other cointegration methods. First,
Pesaran and Shin (1995) and Pesaran, Shin, and
The first step in the empirical analysis is the ex- Smith (2001) showed that the ARDL models
amination of the time series properties of all the yield consistent estimates of the long run co-
variables in logarithmic terms. A visual inspection efficients that are asymptotically normal irre-
of all variables in levels and logs in Figure 2 spective of whether the underlying regressors
suggests that they are trending, and therefore are are purely stationary, 7(0), purely integrated of
nonstationary. That is, their variances and co- order one, 7(1), or fractionally cointegrated.
variances are not finite or independent of time. This represents an advantage over the Johansen
The sample autocorrelation functions and procedure, which allows for testing for the ab-
the partial autocorrelation functions provide sence of a long-run relationship only under the
further evidence that the series are not stationary restrictive assumption that all the model's vari-
in levels or logs and may contain unit roots. ables are integrated of order one. Furthermore,
When variables are nonstationary the standard the ARDL circumvents the low power of unit
ordinary least squares (OLS) model cannot be root tests and the resulting degree of uncer-
applied and there might be a spurious regres- tainty regarding the order of integration of the
sion. Spurious regressions are normally char- underlying variables. Additionally, the ARDL
acterized by having a high R 2 and statistically methodology provides unbiased estimates of the
significant t-statistics but their results have no long-run model coefficients and valid t-statistics
economic meaning (Granger and Newbold, by the inclusion of dynamics in the model, even
1974). The stationarity of the series is first in- when some of the regressors are endogenous
vestigated by applying the augmented Dickey- (Inder, 1993). This is advisable for this model
Fuller unit root test (ADF) and the Phillips and because of potential endogeneity of the ex-
Perron (1988) test (PP). However, recent stud- port diversification variables due to potential
ies have found that these standard unit root tests linkages with the inflows of foreign direct in-
tend to perform poorly in the presence of small vestment in Costa Rica. Lastly, when compared
samples as the one used in this paper. In addi- with other alternative techniques, this method-
tion, these tests suffer from a well-known ology performs better with small samples like the
weakness when testing stationarity of a series that one in this study.
exhibits a structural break. More specifically, they
To conduct the bounds test, Equation (8)
tend to identify a structural break in the series as
is converted into an unrestricted error cor-
evidence of nonstationarity, and thus fail to reject
rection model (UECM) form represented by
the null hypothesis. To deal with this problem,
Equation (9):
a number of methods were developed to improve
the statistical tests in the presence of structural n n

breaks. The Zivot and Andrews (1992) and the A In Y, = a + Y2 Si A In y,_fc + ] T 52A In K,-k
k=l k=0
Perron and Vogelsang (1992) unit root tests are n n
undertaken in this study. Both procedures allow for + ]T8 3 AlnL,_*+ ]T8 4 AlnS,_ t
formal evaluation of the time series properties in (9) k=0 k=0
n
the presence of a structural break at an unknown + ^8 5 AlnMX,_t + pin/s:,_i + 61nL,_i
point in time. The results from the four unit root k=0
tests will be compared so that valid conclusions + V|/lnS,_i +YlnMX,_i + e,,
can be drawn on the order of integration of the
variables in the model. where a is the drift component, A represents the
first differences, and et are white noise errors
524 Journal of Agricultural and Applied Economics, November 2012

^ # ^ # ^ ** tf ^ & <? ^ # J" ** # •& S t X S w w S S w W S S K g S S S S v K O '

YMr

Log of Costa Rican Real GDP at 2000 US$ Constant Prices (1960- Log of the Number of Exports Sectors in Costa Rica (1965-2006).
2007).

Log of Percentage Share of Manufactured Exports Relative to Total Log of Costa Rican Total Labor Force (1965-2006).
Costa Rican Exports (1965-2006).

i iiiii
Vkar

Log of Gross Fixed Capital Formation in Costa Rica (1965-2006).

Figure 2. Time Series of Variables Used

uncorrelated with the variables in the right- " "

AlnY
hand side of the equation. In this setup, the ' = « + ^ + l > A l n y , _ * + ^8 2 Alntf,_*
short-run effects are inferred by the sign and
significance of the estimates of 81, 8 2 , 83, 8 4 , (10) + ]C*=o §3 lnL '-* + 5Z § 4 A l n S,~k
and 85, while the long-run effects are inferred by
+
the sign and significance of the estimates of p, 8, HLo&sA ln MX-k+P ln K,
~•
\j/, and y. Because all the variables in the model + einL,_! + \|/lnS,_, +ylnMX,_, + £t
appear to be trended, a second ARDL-UECM
including a trend term (£,,) is presented in The implementation of the ARDL approach
Equation (10). to cointegration procedure requires two steps.
Ferreira and Harrison: Export Diversification and Economic Growth in Costa Rica 525

The first step involves estimating Equations (9) its use in the present study are now presented.
and (10) using OLS, and the second step in- First, evidence from Monte Carlo simulations
cludes tracing the presence of cointegration has shown that estimators from this procedure
among the variables by restricting all estimated are superior to a number of alternative estima-
coefficients of lagged level variables so that the tors of long-run parameters, including those
inclusion of the lagged level of variables is proposed by Engle and Granger (1987), Johansen
warranted. Thus, the null hypothesis of no (1988), and Phillips and Hansen (1990). More-
cointegration (H 0 = |J = 0 = \|r = y = O)is over, DOLS allows for variables of different
tested against the alternative (H^ 3 ^ 0 ^ \j/ ^ integration order. It accounts for any possible
y ^ 0) using the familiar F-test with critical simultaneity bias within regressors, and it guar-
values tabulated by Pesaran, Shin, and Smith antees valid estimations even in the presence of
(2001). Two asymptotic critical value bounds endogenous independent variables. Finally, DOLS
provide a test for cointegration when the de- is asymptotically equivalent to Johansen's max-
pendent variables are 1(d) with zero < d < 1. imum likelihood estimator, but it tends to perform
The upper bound assumes all variables are 1(1) well with small samples like the one in this study.
while the lower bound assumes that all the The DOLS procedure involves regressing
variables are 7(0). If the computed F-statistics any 1(1) variable on other 1(1) variables, on 7(0)
exceed their respective upper critical values, the variables, and on the leads and lags of the first
null hypothesis of no cointegration is rejected. If differences of any 1(1) variables. The final equa-
the test statistics fall below the lower critical tion of DOLS model is presented in the fol-
values, the null hypothesis cannot be rejected. If lowing section of the paper, and it is constructed
the statistics fall wimin their respective bounds, based on the results from the unit root tests for
inference would be inconclusive and the order of each series.
integration of the underlying variables has to be
investigated more deeply.
Empirical Results

Estimation of Long-Run Elasticities: Tests for Unit Roots

Stock-Watson Dynamic OLS
Given that all variables exhibit upward trends
Stock and Watson (1993) developed a powerful overtime, the ADF and PP tests were under-
and practically convenient modeling procedure taken with and without the inclusion of a de-
known as DOLS. Several arguments that validate terministic trend. Table 1 reports the ADF and

Table 1. Augmented Dickey-Fuller and Phillips-Perron Tests for Units Roots

Augmented Dickey-Fuller Phillips- Perron
Variable Levels Z(t)df Z(t*)df Z(t)pp Z(t*)pp Result
LY, -0.95 -2.02 -0.84 -2.282
LL, -0.85 -2.22 -1.12 -2.058
LK, -0.425 -1.99 -0.52 -1.9
LS, -2.17 -4.20** -2.17 —4.17** 1(0) + trend
LMX, -0.61 -1.69 -0.71 -1.9
First Differences
ALY, — 3 78*** -3.69** —3 72*** -3.62** /(I)
ALL, —7 72*** 7.74*** —7.98*** -8.08*** /(I)
ALK, — 5 27*** —5.19*** -5.25*** -5.18*** /(I)
ALS, —7.47*** —7.46*** —7.81*** —7.79*** 1(0) + trend
ALMX, -5.34*** -5.32*** -5.31*** —5.27*** /(I)
Note: Z(t)df is the ADF test allowing for a drift term, whereas Z(t*)df is the ADF test allowing for a drift and a deterministic
trend. Z(t)pp is the PP test allowing for a drift term, whereas Z(t*) df is the PP test allowing for a drift and a deterministic trend.
*, **, and *** denote the rejection of the null hypothesis of a unit root at the 10%, 5%, and 1% level, respectively.
526 Journal of Agricultural and Applied Economics, November 2012

the PP test statistics for the log levels and first The results in Table 2 show that results from
differences of all variables. The results from the Zivot and Andrews (1992) test suggest that,
both tests indicate that the null hypothesis of when a structural break is considered, all vari-
a unit root cannot be rejected for all variables ables are 7(0) in levels with the exception of the
in levels, with the exception of the number of labor force variable which becomes 7(0) only
export sectors variable, which is trend station- after being differenced. The Perron and Vogelsang
ary in levels. When the tests were computed unit root test shows that both export diversi-
using first-differenced data, the null hypothesis fication variables are stationary at the levels,
was strongly rejected in all cases. In sum, the while GDP, labor, and capital variables are in-
results from these two unit root tests suggest that tegrated of order one. These results question the
all variables, with the exception of S„ are 7(1) in integration orders found by the ADF and PP unit
levels but 7(0) in first differences. Despite the root tests, and at the same time provide evidence
consistency of the results of these two tests, one that both vertical and export diversification
needs to be cautious in interpreting them. variables are likely to be stationary in levels,
Literature on Costa Rican economy iden- while GDP, labor, and capital variables are 7(1).
tifies two potential structural breaks in the last
40 years. The first break occurred when a se- Multivariate Integration: ARDL
vere economic crisis affected the country be-
tween the late 1970s and early 1980s, resulting To determine the optimal number of lags to be
in important structural reforms in the mid-1980s. included in the estimation of the ARDL-UECM
The other potential break was likely to have procedure, the Akaike's Information criterion, the
happened in the late 1990s when the American Schwarz's Bayesian information criterion, and the
multinational, Intel, began its operations in Hanna and Quinn information criterion were used.
Costa Rica. Once again, a visual inspection of Nevertheless, because there was no agreement
Figure 2 reveals that at least one of the above among the criteria on whether to include one or
mentioned structural breaks may be present in two lags, the ARDL-UECM was estimated with
the series, with the exception of labor force both orders of lags. The computed F-statistics for
variable. Based on this, two further unit root the joint significance of lagged levels in
tests are computed to check if, in the presence Equations (9) and (10) lags are presented in
of a structural break, the series are integrated of Table 3 for each order along with the 10% level
order one or otherwise. critical values. The results in Table 3 indicate that

Table 2. Zivot and Andrews, Perron and Vogelsang Unit Root Tests with Structural Break
Zivot and Andrews Perron and Vogelsang
Minimum Minimum
t-Statistic Break Year Result t-Statistic Break Year Result
Variable Levels
LY, -4.96** 1981 1(0) -2.34 1994
LL, -4.24 1991 -1.97 1989
LK, -5.59*** 1982 1(0) -2.33 1984
LS, -6.368*** 1987 1(0) -4.02** 1988 1(0)
LMX, -7.221*** 1997 1(0) —6.41*** 1995 1(0)
First Differences
ALY, -5.82*** 1980 /(I)
ALL, -7.078*** 1996 /(I) -7.06*** 1989 /(I)
ALK, —5.99*** 1981 I(D
Note: Critical values for the Zivot and Andrews test are taken from Zivot and Andrews (1992). Critical values for the Perron and
Vogelsang test are taken from Perron and Vogelsang (1992). The lag length used in the test for each series was determined by the
Akaike's information criterion, the Schwarz's Bayeasian information criterion, and the Hanna and Quinn information criterion.
*, **, and *** denote the rejection of the null hypothesis of a unit root at the 10%, 5% and 1% level, respectively.
Ferreira and Harrison: Export Diversification and Economic Growth in Costa Rica 527

Table 3. Bounds Test for the Existence of a Long- and lags. 2 A step dummy, duSO, and an impulse
Run Relationship between Economic Growth and e?80 are also included in Equation (11) to ac-
the Hypothesized Explanatory Variables count for the severe economic downturn af-
3
10% Critical fecting Costa Rica in the early 1980s and for
Bounds the resulting economic transformation. The
results in Table 4 show that capital and labor
Lag F-Statistics 7(0) 7(1) have a positive and significant effect on Costa
ARDL with 2 1.84 2.45 3.52 Rica's economic growth. On the other hand,
no trend 1 1.43 2.45 3.52 neither vertical nor horizontal export diversi-
ARDL 2 3.01 3.03 4.06 fications significantly influence economic growth.
with trend 1 1.66 3.03 4.06 Interestingly, horizontal export diversification
Note: The relevant critical value bounds are obtained from appears with a negative sign, which contradicts
Table Cl.iii (with an unrestricted intercept and no trend, with the hypothesized relationship with economic
four regressors) and from Table Cl.v (with an unrestricted growth, while vertical export diversification has
intercept and unrestricted trend, with four regressors) in
Pesaran, Shin, and Smith (2001).
the hypothesized positive impact. The Shapiro-
*, **, and *** indicate significance at the 1, 5, and 10% levels. Wilk test for normality shows a p value larger
than 0.05, which indicates that the data are
normally distributed. The Breusch-Godfrey test
the computed F-statistics are not significant at the was included and found no evidence of serial
10% level. More specifically, given that the test correlation in the disturbance. Finally, in the
statistics fell below the lower critical values, the context of time-series data, two specification tests
null hypothesis that all coefficients ((3, 8, \\r, and on residuals were included: The Durbin-Watson
y) equal zero cannot be rejected. These results (DW) test on residual autocorrelation and the
suggest that there is no cointegration between real Engle's Lagrange Multiplier test for autore-
GDP, capital, labor, and the export diversification gressive conditional heteroskedasticity. A DW
variables. The conclusions do not change for the value of 1.03 lies between the zone of inde-
ARDL model including a trend term. Both results cision and the acceptance of the null hypoth-
suggest the lack of a linear long-run impact of esis of no autocorrelation. Finally, the results
export diversification on economic growth in from the Engle's Lagrange Multiplier test do
Costa Rica, and contradict the evidence found in not indicate any problems with autocorrelation
other countries and regions. On the other hand, conditional heteroskedasticity. Equation (11)
these findings seem to confirm the evidence that was again estimated using now robust standard
Costa Rica experienced lower economic growth errors, and its results are shown in Table 5. No
rates during the years of exports expansion and noteworthy changes in the statistical signifi-
diversification in comparison with the ISI pe- cances or signs of the estimated elasticities were
riod. To confirm the robustness of this finding, found. In summary, the DOLS procedure con-
the DOLS procedure is applied to Equation (8). firms the lack of a long-run causality between
export diversification and economic growth in
Costa Rica over the period of 1965-2006.
Long-Run Elasticities: Stock-Watson DOLS
It is important to understand why the pres-
To estimate the long-run parameters using the ent results differ from those found for Chile by
DOLS procedure the growth Equation (8) is
transformed into Equation (11).
2
The DOLS model was also estimated using one
lny, = a + filriK, + XlnL, + \\rlnS, + y\nMX, and three leads and lags without altering the results to
k=n k=n any significant degree,
^2AlnKt~k+du80
3
(11) + Y2 CiAlnL,_/+ ^2 The year of 1980 was chosen based on the
k=—n k~—n literature on the economic crisis that affected Costa
+ dSO + w, Rica, and on visual observation of the plots of the
series in log levels. duSO is one from 1980 onwards and
Given that annual data are used, the model is zero otherwise, while dSO has a value of one in 1980
estimated with inclusion of n = ± two leads and zero otherwise.
528 Journal of Agricultural and Applied Economics, November 2012

Table 4. Stock-Watson DOLS Long-Run Pa- multinationals poses limitations to the amount
rameter Estimates of knowledge spillovers generated by the ex-
port sectors. Consequently, Costa Rica has not
p X •F Y
been able to use its high-tech and high value-
0.26*** 0.81*** -0.18 0.24 added exports to trigger a sustained process of
(3.14) (4.38) -(1.11) (0.44) economic growth. This corroborates the argu-
Notes: The parentheses under the coefficients denote t statis- ment of Sanchez-Ancochea (2006) that although
tics. Adj. R2 = 0.99; DW = 1.03; SW = 0.96 (0.15); Intel and other multinational corporations op-
ARCH(l) = 0.99; ARCH(2) = 0.98; ARCH(3) = 0.99;
BG(1) = 0.00; BG(2) = 0.00; BG(3) = 0.00. BG is the erating in Costa Rica contributed to an increase
Breush-Godfrey test for higher order serial correlation in the in exports and generated direct employment,
disturbance and ARCH is Engle's Lagrange Multiplier test for they failed to generate substantial linkages with
autocorrelation conditional heteroskedasticity, with k = 1,2,
the rest of the economy. In the particular case of
and 3 lags. SW is the Shapiro-Wilk test for normality.
*, **, and *** indicate significance at the 10, 5, and 1% levels. Intel, some economists maintain that this firm
has operated as an enclave, importing most of
Herzer and Nowak-Lehnmann's (2006). Even its components for its assembly, and generating
though these two countries are regarded as a low economic multiplier (World Bank, 2006).
successful in terms of their economic perfor- Furthermore, despite the surge of non-traditional
mance and diversification of their exports, a agricultural exports in the last decades, Costa
closer look at their economies will reveal im- Rica is still exporting mainly raw agricultural
portant differences. These differences can ex- products with litde value added (Barquero, 2006).
plain why export diversification has played such Finally, Mitchell and Pentzer (2008) make an
an important role in Chile and not so in Costa important observation that despite the fact that
Rica. In the case of the former, the most im- the range of export products in Costa Rica has
portant source of export diversification has been grown, a small group of products, including
the emergence of non-traditional agricultural ex- manufactured and agricultural products, con-
ports. Examples of exported resource-based tinues to account for the majority of the export
goods are those produced by forestry and value. Thus, progress made in terms of hori-
mining conglomerates, a thriving wine sector, zontal and vertical export diversification may
and an expanding salmon-farming industry. Al- fail to reveal a persistent concentration in terms
though these products have low levels of tech- of value. For instance, in 2005, 84% of the total
nological content, they often are produced by value of all goods exported was produced by
domestic firms. On the other hand, Costa Rica large corporations, which account only for 20%
went from being highly reliant on exports of a of the total number of manufacturers in Costa
few primary goods to a country with flourishing Rica (Promotora del Comercio Exterior de Costa
high-tech and medical equipment manufactur- Rica, 2005).
ing export sectors, and well diversified agricul-
tural and service sectors. However, this was Conclusion and Policy Implications
likely the result of the creation of export pro-
cessing zones by Costa Rican authorities, which By estimating an augmented Cobb-Douglas
attracted foreign capital in sectors with high production function using time series data, this
technological contents throughout the 1990s. study has presented new empirical evidence that
Such interdependence between export diver- in Costa Rica neither vertical nor horizontal di-
sification and foreign investment by large versification is associated with faster economic
growth over the period of 1965-2006. These
Table 5. Stock-Watson DOLS Long-Run Pa- findings are contrary to what was hypothesized
rameter Estimates with Robust Standard Errors and to what has been observed in other econo-
mies. On the other hand, this evidence generates
p X Y Y
a discussion on whether a government-backed
0.26*** 0.81*** -0.18 0.24 export diversification based on FDI is conducive
(4.37) (5.65) -(1.20) (0.80) to long-term economic growth.
Ferreira and Harrison: Export Diversification and Economic Growth in Costa Rica 529

In terms of policy implications, the results Fiscal Imbalances, and Macroeconomic Policy:
from this study suggest that expansion and di- A Computable General Equilibrium Model."
versification of exports per se may not be suf- The North American Journal of Economics and
ficient to promote economic growth, unless Finance 10,l(1999):39-67.
they lead to the creation of new productive Engle, R.F., and C.W.J. Granger. "Cointegration
and Error Correction: Representation, Estima-
capabilities in other sectors of the economy via
tion and Testing." Econometrica: Journal of
knowledge externalities. Although Costa Rica
the Econometric Society 55,2(1987):251-76.
has been fairly successful in attracting FDI in Feenstra, R., and H.L. Kee. "Export Variety and
manufactured goods, this study identifies the Country Productivity." Working Paper 10830,
limitations of the hitherto export-led model of National Bureau of Economic Research,
development. Hence, Costa Rica should design Cambridge M A , 2004.
and implement a new set of policies seeking Fosu, K. "Export Composition and the Impact
to improve the nation's long-term economic of Exports on Economic Growth of Developing
growth potential and increase the role of the Economies." Economics Letters 34,l(1990):71-76.
domestic industrial sector in its exports. First, Granger, C.W.J., and P. Newbold. "Spurious Re-
further linkages between the export sector and gression in Econometrics." Journal of Econo-
the rest of the economy should be created so metrics 2,3(1974): 111-20.
Greenaway, D., W. Morgan, and P. Wright. "Ex-
that new channels for knowledge spillovers
ports, Export Composition and Growth." The
may be generated. Furthermore, the presence
Journal of International Trade & Economic
of multinational companies in the country Development 8,1(1999):41-51.
should be used by Costa Rica to spur the de- Gutierrez-de-Pifieres, S.A., and M. Ferrantino.
velopment of clusters of domestic-owned sup- "Export Sector Dynamics and Domestic Growth:
pliers and other satellite businesses. Financial The Case of Colombia." Review of Development
instruments and access to credit should also be Economics 3,3(1999):268-80.
facilitated for potential domestic firms that . Export Dynamics and Economic Growth
wish to produce and export products that are in Latin America: A Comparative Perspective.
more technologically involved. Finally, regu- Ashgate Publishing, the University of California,
lation reforms should take place to simplify the 2000.
creation of small and medium domestic export- Herzer, D., and D.F. Nowak-Lehnmann. "What
Does Export Diversification do for Growth?
oriented firms. All these efforts should be done
An Econometric Analysis." Applied Economics
in close coordination with higher education in-
38,15(2006): 1825-38.
stitutions so that they can supply domestic firms
Hesse, H. "Export Diversification and Economic
with qualified workers. Growth." Working Paper No. 21. The International
Bank for Reconstruction and Development/World
[Received February 2010; Accepted February 2012.] Bank, Commission on Growth and Development,
2008.
Inder, B . "Estimating Long-Run Relationships
References
in Economics: A Comparison of Different Ap-
proaches." Journal of Econometrics 57,1-3(1993):
Al-Marhubi, F. "Export Diversification and 53-68.
Growth: An Empirical Investigation." Applied Jansen, M. "Income Volatility in Small and De-
Economics Letters 7,9(2000):559-62. veloping Economies: Export Concentration
Balaguer, J., and M. Cantavella-Jorda. "Structural Matters." Discussion Papers No 3. Geneva:
Change in Exports and Economic Growth: Coin- World Trade Organization, 2004.
tegration and Causality Analysis for Spain (1961- Johansen, S. "Statistical Analysis of Cointegra-
2000)." Applied Economics 36,5(2004):473-77. tion Vectors." Journal of Economic Dynamics
Barquero, S.M. "Producers Earn Less than 10% and Control 12,2-3(1988):231-54.
of Sales Revenues from Coffee and Canta- Larrain, F.B., L.F. Lopez-Calva, and A. Rodriguez-
loupes." La Nacion (newspaper), Economics Clare. "Intel: A Case Study of Foreign Direct
Section (2006). Investment in Central America." Working Paper
Cattaneo, A., R.A. Hinojosa-Ojeda, and S. No. 58. Center for International Development,
Robinson. "Costa Rica Trade Liberalization, Harvard University, 2000.
530 Journal of Agricultural and Applied Economics, November 2012

Learner, E.E. The Heckscher-Ohlin Model in Frisch Centennial Symposium. Cambridge, MA:
Theory and Practice. Princeton Studies in In- Cambridge University Press, 1995.
ternational Finance. 77. Princeton, NJ: Princeton Pesaran, M.H., Y. Shin, and R.J. Smith. "Bounds
University Press, 1995. Testing Approaches to the Analysis of Level
Levin, A., and L.K. Raut. "Complementarities Relationships." Journal of Applied Economet-
between Exports and Human Capital in Economic rics 16,3(2001):289-326.
Growth: Evidence from the Semi-industrialized Phillips, P.C.B., and B.E. Hansen. "Statistical
Countries." Economic Development and Cultural Inference in Instrumental Variables Regression
Change 46,1(1997): 155-74. with 7(1) Processes." The Review of Economic
Love, J. "Commodity Concentration and Export Studies 57,1(1990):99-125.
Earnings Instability: A Shift from Cross-Section Phillips, P.C., and P. Perron. "Testing for a Unit
to Time Series Analysis." Journal of Develop- Root in Time Series Regression." Biometrika
ment Economics 24(I986):239-48. 75,2(1988):335-^6.
Matthee, M., and W. Naude. "Export Diversity Promotora del Comercio Exterior de Costa
and Regional Growth: Empirical Evidence Rica—Costa Rica. Estadisticas de Exportacion.
from South Africa." Research Paper No. 2007/ Internet site: http://www.procomer.com/est/
11. United Nations University, 2007. mercados/libros.cfm/ (Accessed February 16,
Mesa-Lago, C , A. Arenas-de-Mesa, and I. 2008).
Brenes. Market, Socialist, and Mixed Econo- Sanchez-Ancochea, D. "Development Trajecto-
mies: Comparative Policy and Performance— ries and New Comparative Advantages: Costa
Chile, Cuba, and Costa Rica. Baltimore, MD: Rica and the Dominican Republic." World
Johns Hopkins University Press, 2003. Development 34,6(2006):996-1015.
Mitchell, M.T., and S. Pentzer. Costa Rica: A Stock, J.H., and M. Watson. "A Simple Estimator
Global Studies Handbook. Santa Barbara, CA: of Cointegrating Vectors in Higher Order In-
ABC-CLIO Inc., 2008. tegrated Systems." Econometrica: Journal of
Moreno-Brid, J.C., and E. Perez. Trade Liber- the Econometric Society 61,4(1993):783-820.
alization and Economic Growth in Central Vos, R., E. Ganuza, S. Morley, and S. Robinson.
America. CEPAL Review 8 1 . Santiago, Chile: Who Gains From Free Trade? Export-led Growth,
The Economic Commission for Latin America Inequality and Poverty in Latin America. London,
(ECLA), 2003. New York: Routledge, 2006.
Perron, P., and T. Vogelsang. "Nonstationarity World Bank. The Impact of Intel in Costa Rica:
and Level Shifts with an Application to Pur- Nine Years after the Investment. Washington,
chasing Power Parity." Journal of Business & DC: World Bank, 2006.
Economic Statistics 10(1992):301-20. Zivot, E., and D.W.K. Andrews. "Further Evi-
Pesaran, M.H., and Y. Shin. "Chapter 11: An dence on the Great Crash, the Oil-Price Shock,
Autogressive Distributed Lag Modeling Ap- and the Unit-Root Hypothesis." Journal of
proach to Cointegration Analysis." Econometrics Business & Economic Statistics 10,3(1992):
and Economic Theory in 20* Century: The Ragnar 251-70.

Appendix 1: Variable Definitions rates. For a few countries where the official exchange
and Data Sources rate does not reflect the rate effectively applied to
actual foreign exchange transactions, an alternative
conversion factor is used.
G D P at Constant Prices (2000 U.S. $) (J)
Source: World Development Indicators online ver-
GDP at purchaser's prices is the sum of gross value sion, World Bank 2008.
added by all resident producers in the economy plus
any product taxes and minus any subsidies not in- Gross Fixed Capital Formation
cluded in the value of the products. It is calculated (2000 U.S. $) (K)
without making deductions for depreciation of fab-
ricated assets or for depletion and degradation of Gross fixed capital formation is measured by the total
natural resources. Data are in constant 2000 U.S. value of a producer's acquisitions, less disposals, of
dollars. Dollar figures for GDP are converted from fixed assets during the accounting period plus certain
domestic currencies using the 2000 official exchange additions to the value of non-produced assets (such as
Ferreira and Harrison: Export Diversification and Economic Growth in Costa Rica 531

subsoil assets or major improvements in the quantity, forces and seasonal or part-time workers, in general
quality, or productivity of land) realized by the pro- the labor force includes the armed forces, the un-
ductive activity of institutional units. employed, and first-time job-seekers, but excludes
homemakers and other unpaid caregivers and workers
Source: World Development Indicators online ver- in the informal sector.
sion, World Bank 2008.
Source: World Development Indicators online ver-
Step Time Dummy Variable (dw80) sion, World Bank 2008.

This variable assumes the value 1 from 1980 onwards Vertical Export Diversification (MX)
and zero otherwise.
Source: Author's own calculations. This is the manufactured exports as a percentage
of total merchandise exports. The manufactured
comprise commodities in Standard International
Impulse Time Dummy Variable (rf80) Trade Classification sections five (chemicals), six
(basic manufactures), seven (machinery and trans-
This variable has a value of 1 in 1980 and zero port equipment), and eight (miscellaneous manu-
otherwise. factured goods), excluding division 68 (non-ferrous
Source: Author's own calculations. metals).

Source: World Development Indicators online ver-

Labor Force, Total (L) sion, World Bank 2008.

Total labor force comprises people who meet the Horizontal Export Diversification (S)
International Labor Organization definition of the
economically active population: all people who sup- This is the number of export sectors classified by the
ply labor for the production of goods and services
Standard International Trade Classification at the
during a specified period. It includes both the em-
three-digit level.
ployed and the unemployed. While national practices
vary in the treatment of such groups as the armed Source: United Nations dataset.