Professional Documents
Culture Documents
3, 2010, 565–583
doi: 10.1111/j.1477-9552.2010.00257.x
Abstract
This paper quantifies the contribution of agricultural exports to economic growth
in developing countries. We estimate the relationship between GDP and agricul-
tural and non-agricultural exports for 42 countries using panel cointegration
methods. Results show that a long-run relationship exists, the agricultural export
elasticity of GDP is 0.07 whereas that of non-agricultural exports is 0.13, and
total exports Granger-cause GDP, which supports the export-led growth hypo-
thesis. Structural differences exist in the relationship by broad income group.
Balanced export-promotion polices are implied for the poorest countries, but, for
those with higher incomes, higher economic growth is achieved from non-agricul-
tural exports.
1. Introduction
Economic growth is a primary aim of developing countries and a recurrent theme
in the trade and development literature is the role of exports in this process. Much
of this literature focuses on total exports as an engine of growth, but agriculture’s
contribution to total exports is often substantial in developing countries: in 2004 for
example, this proportion was 90% for Guinea-Bissau, 85% for Nicaragua and 75%
1
Ana Sanjuán-López is a Visiting Fellow in the School of Agriculture, Food and Rural
Development, Newcastle University, Newcastle NE1 7RU, UK, and a Research Fellow
in the Centre of Agrofood Research and Technology of Aragon (CITA), Zaragoza, Spain.
E-mail: aisanjuan@aragon.es for correspondence. Phil Dawson is a Reader in the School of
Agriculture, Food and Rural Development, Newcastle University, Newcastle NE1 7RU, UK.
E-mail: p.j.dawson@ncl.ac.uk. Ana Sanjuán-López gratefully acknowledges financial support
from the Spanish Ministry of Science and Education, within the programme ‘Fomento de
movilidad de profesores e investigadores’. We are grateful to two referees and an External
Editor for comments on a previous draft.
2010 The Authors. Journal compilation 2010 The Agricultural Economics Society.
566 Ana I. Sanjuán-López and P. J. Dawson
for both Benin and The Gambia (FAO, 2007). It is surprising that the empirical
relationship between agricultural exports and economic growth has been somewhat
neglected in the literature despite its role in the development process being long
recognised. Johnston and Mellor (1961, p. 575) argue that expanding agricultural
exports is one of the most promising means of increasing incomes. Moreover, where
a developing country’s exports are a small proportion of world trade, as is typical,
export demand for that country is elastic and policies that seek to stimulate agricul-
tural exports are not irrational even when world conditions are unfavourable, par-
ticularly where few alternatives exist. Agricultural exports therefore can play an
important role in economic growth, and export-led growth from agriculture may
represent optimal resource allocation for those countries that have a comparative
advantage in agricultural production.
Two empirical studies that examine the short-run relationship between agricultural
export growth and economic growth with panel data are Levin and Raut (1997) and
Dawson (2005). We contribute to this literature by examining the long-run relation-
ship between gross domestic product (GDP) and exports in developing countries.
Exports are disaggregated into agricultural and non-agricultural components, and the
main focus is to assess the impact of agricultural exports on economic growth. Our
framework is the export-led growth hypothesis where increases in exports lead to eco-
nomic growth. We use recently developed panel cointegration methods to address two
criticisms of conventional cointegration tests, namely that they tend to under-reject in
the presence of structural breaks and have poor size and power properties. Westerl-
und’s (2006a) multivariate panel cointegration procedure is used to test for the exis-
tence of a long-run relationship where structural breaks are permitted. Its parameters
are then estimated using fully modified ordinary least squares (FMOLS). In essence,
our method seeks unknown breaks in the series and then models them in a panel coin-
tegrating relationship. Westerlund’s method is particularly appropriate here because it
allows for the estimation of endogenously determined and heterogeneous structural
breaks, caused, for example, by the oil crisis of 1973 ⁄ 1974 and the debt crisis of
1981 ⁄ 1982, which may have affected individual countries differently. In addition,
there is some evidence to suggest that the level of economic development gives rise to
structural differences in the export–income relationship (Ram, 1987) and we divide
our dataset into three subsamples by the broad income groupings categorised by the
World Bank as ‘low’, ‘lower middle’ and ‘upper middle’ income countries.
There are three lines of enquiry. First, we test for cointegration between GDP and
agricultural and non-agricultural exports, thereby identifying any long-run export–
income relationship in developing countries. Second, we estimate this relationship and
examine whether it differs by broad income groups. Third, we test the export-led
growth hypothesis by testing causality from exports to GDP. The paper is organised as
follows: Section 2 outlines explanations of the export–income relationship and reviews
some empirical literature, section 3 details the empirical method, section 4 discusses
the data and presents the results and section 5 provides a summary and discussion.
2. Background
The export-led growth hypothesis has dominated the export–income literature and
there are four explanations. First following short-run Keynesian arguments, export
growth leads to income growth via the foreign trade multiplier. Second, foreign
exchange from exports can be used to finance imported manufactured and capital
2010 The Authors. Journal compilation 2010 The Agricultural Economics Society.
Agricultural Exports and Economic Growth in Developing Countries 567
goods and technology, which contribute to growth (Chenery and Strout, 1966).
Third, competition leads to scale economies, technological advance and growth
(Helpman and Krugman, 1985). Fourth following endogenous growth theory, the
export sector creates positive externalities, such as more efficient production meth-
ods, which lead to growth (Balassa, 1978; Grossman and Helpman, 1993).
Alternative hypotheses regarding causality in the export–income relationship have
also been postulated. First, technology trade theory hypothesises growth-led exports
which occur when domestic demand lags income growth and technological advance
results in output increasing faster than domestic demand so that exports increase.
Similarly, economic growth could also reduce export growth when there is an
increase in domestic demand which is concentrated in exportable and non-tradable
goods so that economic growth increases but exports fall (Jung and Marshall,
1985). Second, intra-industry trade theory predicts that when economies of scale
increase productivity, exports also increase; if the resulting market structure involves
fewer firms, economies of scale lead to cost reductions, and there is bi-directional
causality between export growth and economic growth (Helpman and Krugman,
1985; Kunst and Marin, 1989).
There are numerous empirical studies of the export–income relationship and Giles
and Williams (2000a,b) and Ahmad (2001) provide comprehensive surveys. Rather
than replicating these surveys, we highlight the diverse results. Early studies gener-
ally support the export-led growth hypothesis and include Balassa (1978) whose
estimate of the export elasticity of income is 0.05, and its validity gained widespread
acceptance, notably by the World Bank (1987). Further support for the export-led
growth hypothesis is found from Solow-type sources-of-growth equations. Here, an
aggregate production function is specified with labour and capital as conventional
inputs and exports as an additional ‘input’. Sources-of-growth equations are then
derived where income growth is determined by growth rates of both conventional
inputs and exports. Examples include Tyler (1981), Feder (1983), Ram (1987) and
Dawson and Hubbard (2004), and export elasticities vary between 0.06 and 0.57.
More formal statistical tests of causality, often with simple bivariate export–income
models, provide mixed results. For example, Jung and Marshall (1985) provide
only weak evidence in support of the export-led hypothesis, whereas Xu (1996) and
Balaguer and Cantavella-Jordá (2004) find more general support. Sharma and
Dhakal (1994) find evidence of causality in both directions, of bi-directional causal-
ity and of no causality. Ghirmay et al. (2001) estimate long-run export elasticities
between )0.21 and 0.67; there is much evidence of bi-directional causality, but there
is also evidence of unidirectional causality in both directions. A further line of
enquiry is to examine the impact of increasing income on the export elasticity of
income, and Ram (1987) finds some evidence that this impact decreases as income
increases.
De Piñeres and Cantavella-Jordá (2007) highlight these disparate conclusions
by presenting results for Latin American countries. Findings show that long-run
relationships between GDP and exports exist only for some countries. For these,
there is evidence of export-led growth where long-run export elasticities vary
between )2.07 and 4.41, but there is also evidence of growth-led exports, bi-direc-
tional causality and no causality. These inconsistent results, and by extension
those elsewhere, arise because of differences in methodologies, countries examined,
variable definitions and data sources.
2010 The Authors. Journal compilation 2010 The Agricultural Economics Society.
568 Ana I. Sanjuán-López and P. J. Dawson
Few studies examine the relationship between agricultural exports and income,
but two exceptions that use panel data are Levin and Raut (1997) and Dawson
(2005).2 Levin and Raut examine the effect on GDP of exports of both manufac-
tured and primary goods, where the latter includes agricultural commodities. The
panel dataset comprises of 30 semi-industrialised countries and two time periods for
the 10-year growth rates of 1965–1974 and 1975–1984. A Feder-type sources-of-
growth equation is estimated and findings show that GDP growth can be increased
by manufactured export growth, but not by expanding primary commodity exports.
Export promotion therefore should focus on manufactured output. Dawson devel-
ops a Feder-type sources-of-growth equation from a dual economy model with agri-
cultural and non-agricultural sectors. This equation is estimated from annual data
for 62 developing countries for 1974–1995. Results show that the short-run agricul-
tural export elasticity of GDP is similar to the non-agricultural export elasticity,
and that their impacts fall as income increases. Both studies estimate sources-of-
growth equations where GDP growth is a function of export growth. Such differ-
encing of the variables throws away information contained in the data and the
focus is on the short run only. This paper, by contrast, examines the long-run rela-
tionship between GDP and agricultural and non-agricultural exports in levels using
annual data from 42 developing countries for 1970–2004. We also test the export-
led growth hypothesis, that is, we test non-causality from total exports to GDP.
3. Empirical Method
Many economic time series are non-stationary and are integrated of order one, I(1).
OLS regressions between such data are often spurious but where two or more I(1)
series move together and their linear combination is stationary, they are cointegrated
and a meaningful long-run equilibrium exists (Granger, 1988). The augmented
Dickey–Fuller (ADF) test (Dickey and Fuller, 1981) is often used to test for unit
roots and the null hypothesis of non-cointegration can be tested using the Engle–
Granger (1987) method. These tests have two well-known shortcomings. First, when
a series is subject to a deterministic trend and exogenous shocks cause structural
breaks, the ADF test tends to under-reject (Perron, 1989), as does the Engle–
Granger (1996) test. Second, both tests have poor size and power properties when
applied to time series of moderate length. Panel data can be used to address these
criticisms as they provide more variability and less collinearity between variables,
and parameter estimates are more efficient as degrees of freedom are greater
(Baltagi, 2001, pp. 5–7). A number of panel tests for both unit roots and cointegra-
tion have been developed and those used here are by Westerlund (2006a,b).3
Conditionally, the cointegrating relationships for each country and for the panel are
2
Balaguer and Cantavella-Jordá (2004) examine the relationship between GDP and exports
using annual data for Spain for 1961–2000. Exports are disaggregated into five categories,
one of which is agricultural exports. Inter alia, they find bi-directional causality between agri-
cultural exports and GDP.
3
We choose Westerlund (2006a,b) tests because our time series consists of 35 annual observa-
tions for 1970–2004 and it is reasonable to assume that structural breaks may have occurred
during that period due, for example, to the oil crisis of 1973 ⁄ 1974 or the debt crisis of
1981 ⁄ 1982.
2010 The Authors. Journal compilation 2010 The Agricultural Economics Society.
Agricultural Exports and Economic Growth in Developing Countries 569
then estimated following Pedroni (2001), and Canning and Pedroni’s (2008) test is
used to test for non-causality from exports to GDP.
X
pi
Dtit ¼ /i tit1 þ cij Dtitj þ eit ð2Þ
j¼1
i i
where DUiqt is a step dummy equal to unity for t > Tbq where Tbq denotes the date
of the qth break in country i (q = 1,…, mi) and mi is the number of breaks (mi ‡ 1)
in the ith country of the panel; gi is a constant that represents the level of the ith
series before any break; gi + hi is a constant that represents the level of the ith ser-
ies after any break; and eit is the disturbance which has zero mean, is serially uncor-
related and is uncorrelated across i.4 The null hypothesis is that the individual series
for all countries, that is, yit for i = 1,…, N and t = 1,…, T are non-stationary, and
the alternative hypothesis is that at least one of these series is stationary. Specifi-
cally, the null hypothesis is that H0: /i = 0 for all i, and the alternative hypothesis
is that HA: /i < 0 for some i. The Lagrange multiplier (LM) unit-root statistic for
each country’s series, s/i , is the t-statistic of /i estimated from an auxiliary regres-
sion in equation (2) following Amsler and Lee (1995) who illustrate for the case of
one break. Westerlund (2006b) then uses the cross-sectional average of the individ-
ual country LM s/i statistics, s/. Under cross-independence of the disturbances eit,
as T fi ¥ followed by N fi ¥, the standardised form follows an asymptotic normal
distribution:
pffiffiffiffi
N ðs/ HÞ
Z LM ¼ pffiffiffiffiffi ) Nð0; 1Þ ð3Þ
R
where H and R are the average of the mean and variance of the limiting distribution
of s/ which are estimated by simulation. Values of the Z-LM statistic smaller than
the appropriate left tail critical value reject the null hypothesis. The selection of the
number and location of the breaks is carried out through an outlier detection proce-
dure.
4
The condition that eit is uncorrelated across i is useful to derive the asymptotical distribu-
tion of the panel unit-root test, but its validity is questionable here as it implies that econo-
mies of developing countries are independent.
2010 The Authors. Journal compilation 2010 The Agricultural Economics Society.
570 Ana I. Sanjuán-López and P. J. Dawson
where
X
t
^Sit ¼ ^eij
i
j¼Tb;j1 þ1
is the partial sum of efficient estimates of the residuals in equation (4) and x ^ 2i is a
consistent estimate of the long-run variance of eit. The corresponding standardised
statistic, denoted as Z-LMC, under cross-independency as T fi ¥ followed by
N fi ¥ is:
pffiffiffiffi
N ðLMC HÞ
Z LMC ¼ pffiffiffiffi ) Nð0; 1Þ ð8Þ
R
where H and R are the average of the mean and variance of the limiting distribu-
tion of the LMC statistic in equation (7). Response surface moments are obtained
by Westerlund (2006a) using Monte Carlo simulations of the limiting distribution
which depend on the deterministic specification of the model and the number of
regressors in xit, but not on the location of the breakpoints. The Z-LMC statistic is
compared with the right tail of the normal distribution. The number and location
of breaks is selected using Bai and Perron’s (1998, 2003) iterative procedure.7
5
In our case, yit is GDP and xit is a two-dimensional vector containing agricultural and non-
agricultural exports.
6
Note that /i can be zero for some countries and it is not required that /i = / „ 0.
7
The asymptotic normal distribution is only valid under the assumption of cross-country
independence and bootstrapping is necessary to obtain appropriate critical values.
2010 The Authors. Journal compilation 2010 The Agricultural Economics Society.
Agricultural Exports and Economic Growth in Developing Countries 571
where ^eit1 is the estimated long-run disequilibrium, that is, the residuals from equa-
tion (4); ai is the error-correction term that gives the reaction of yit to return the
system to long-run equilibrium; and Iqit and DUqit are impulse and step dummies to
provide consistency with the breaks in levels and trends in the long-run relationship
in equation (4); and li is a constant for each country which is included in the esti-
mation of the (single) heterogeneous panel ECM only.
First, consider testing for non-causality for each country. The ECM in equation
(9) is estimated separately for each i = 1,…, N country where the number of coun-
try-specific lags, k, with a maximum of k = 4, is determined by the Schwartz
Bayesian criterion. The joint null hypothesis of no country short- or long-run cau-
sality from xit to yit is:
H01 : C2ij ¼ 0 and ai ¼ 0 for each i ¼ 1; . . . ; N and j ¼ 1; . . . ; k ð10Þ
and F Fk+1. The null hypothesis of no long-run causality from xit to yit for each
country is:
H02 : ai ¼ 0 for each i ¼ 1; . . . ; N ð11Þ
and t tT)npar where npar is the number of estimated parameters in equation (9).
More general insights about causality are provided by corresponding panel tests.
We estimate the single heterogeneous ECM in equation (9) jointly for i = 1,…, N
countries where the number of lags, k, are determined from country ECMs as
before. The joint null hypothesis of no short- or long-run causality in the panel is:
H03 : C2ij ¼ 0 and ai ¼ 0 for all i ¼ 1; . . . ; N and all j ¼ 1; . . . ; k ð12Þ
and the log-likelihood ratio is LLR v2RþN where R is the total number of lagged
2010 The Authors. Journal compilation 2010 The Agricultural Economics Society.
572 Ana I. Sanjuán-López and P. J. Dawson
Dxi terms in all N equations in equation (9). The null hypothesis of no long-run
causality from xit to yit in the panel is:
H04 : ai ¼ 0 for all i ¼ 1; . . . ; N ð13Þ
and LLR v2N .
The null hypotheses in equations (10) and (12) are Granger-causality
tests as yit evolves exogenously with respect to xit at all non-contemporaneous time
horizons.8
Following Canning and Pedroni (2008), we also test the ‘pervasiveness’ of long-
run casual effects in the panel. The group mean test is based on the average of
adjustment coefficients for each country, ai, in equation (9) and the null hypothesis
is of no long-run causality in the Ppanel on average. Denote the average of1the PNadjust-
ment parameters as: a ¼ N 1 N i¼1 a i . The test statistic is:
t ðaÞ ¼ N i¼1 tðai Þ
where t(ai) is the test statistic for each country and the null hypothesis is:
H05 : a ¼ 0: ð14Þ
The t ðaÞ statistic has a standard normal distribution under the null hypothesis of
no long-run causality and rejection in favour of the alternative hypothesis is at
either tail. The null hypothesis of no causality in equation (14) concerns the average
long-run effect and non-rejection may be a consequence of heterogeneous long-run
coefficients with positive and negative values that cancel each other out. To rule out
this possibility,
P Canning and Pedroni propose a Wald-type homogeneity test:
W¼ N r2
i¼1 ai ða i aÞ2 , where
a is the estimated group mean estimates of the N
error-correction terms and r2 ai is the inverse of the sample variance of the country-
specific error-correction terms. Under the null hypothesis of parameter homogeneity
across countries, W v2N .
8
In equation (9), yit is GDP and xit is a two-dimensional vector containing agricultural and
non-agricultural exports. To test no long-run causality, we test the significance of ai which is
the coefficient associated with the error term, ^eit1 , that is, the lagged residuals in equation
(4). The error-correction term, ai, measures the adjustment of GDP in each period to the dif-
ference between GDP and a linear combination of exports. Thus, we test the null hypothesis
of no causality from xit to yit, that is, from total exports to GDP. It does not seem possible
to test causality from each export variable to GDP separately. Likewise, in conventional
cointegrated systems, Granger-causality can only be tested between two exclusive subsets
of variables (Lütkepohl, 1993; Zapata and Gil, 1999).
9
The World Bank divides economies according to per capita gross national income (GNI). In
2008, low-income countries are for GNI of $975 or less; lower middle income countries are
for GNI between $976–$3,855 and upper middle income countries are for GNI between
$3,856-$11,905 (http://web.worldbank.org). During the sample period, not all countries were
categorised in the same income group. In 1990, for example, India, Indonesia and Sri Lanka
were classified as low-income countries, whereas now they have lower middle incomes.
2010 The Authors. Journal compilation 2010 The Agricultural Economics Society.
Agricultural Exports and Economic Growth in Developing Countries 573
Table 1
Unit-root tests
GDP
Full sample )0.39 (0.34)*
Low-income countries 0.69 (0.59)*
Lower middle income countries )0.19 (0.37)*
Upper middle income countries )1.54 (0.16)*
Agricultural exports
Full sample )3.65 (0.05)*
Low-income countries )2.97 (0.05)*
Lower middle income countries )1.57 (0.23)*
Upper middle income countries )2.20 (0.07)*
Non-agricultural exports
Full sample )2.15 (0.15)*
Low-income countries 0.15 (0.57)*
Lower middle income countries )2.14 (0.09)*
Upper middle income countries )1.68 (0.14)*
GDP and the total exports of goods and services are measured in constant local
currency units (source: World Bank, 2006). Agricultural and non-agricultural
exports in constant local currency units are obtained as follows. Data are obtained
for total agricultural exports, which includes all crop and livestock products, and
for total merchandise exports in US$ (source: FAO, 2007). Their ratio is then multi-
plied by the total exports of goods and services in constant local currency units to
obtain agricultural exports in constant local currency units. Subtracting agricultural
exports from total exports gives non-agricultural exports in constant local currency
units.
On average, agricultural exports as a proportion of total exports varies between
90% for Malaysia to 1% for Gabon. This ratio falls as income increases: it is 47%
for low-income countries, 37% for those with lower middle incomes and 26% for
those with upper middle incomes. Similarly, the ratio of agricultural exports to
GDP varies on average between 77% for Guinea-Bissau and 2% for Algeria. This
ratio also falls as income increases: it is 10% for low-income countries, 8% for
those with lower middle incomes and 6% for those with upper middle incomes.
Using logarithms throughout, we first examine the order of integration of each
variable using the panel Z-LM test of Westerlund (2006b) and results are shown in
Table 1.10 A maximum of one break in each equation is permitted. A deterministic
trend is included to account for possible trend stationary behaviour under the null
hypothesis and 5,000 replications are run to obtain bootstrapped critical values
under general conditions of cross-dependency (Chang, 2004; Westerlund, 2006b).
For the full sample and for each subsample by income group, all three variables
are I(1).
10
We are grateful to Joakim Westerlund who provided GAUSS code for both unit-root and
cointegration tests.
2010 The Authors. Journal compilation 2010 The Agricultural Economics Society.
574 Ana I. Sanjuán-López and P. J. Dawson
Table 2
Cointegration tests
Z-LMC statistic
Note: Bootstrapped critical values are given in parentheses at the 5% significance level.
We now apply Westerlund’s (2006a) panel Z-LMC test to test the null hypothesis
of panel cointegration. Again in each country-specific equation, we allow for a max-
imum of one break (and two regimes) in both levels and trends. As our maintained
hypothesis is export-led growth, we normalise on GDP and the estimated equa-
tion is:
GDPit ¼ g1i þ g2i þ d1i t 1 þ d2i t 2 þ b1i AXit þ b2i
NAXit þ eit ; i ¼ 1; . . . ; N; t ¼ 1; . . . ; T ð15Þ
where g1i and g2i are the country-specific intercepts in the two regimes (q = 1, 2);
t1 and t2 are time trends in the two regimes with corresponding country-specific
parameters of d1i and d2i; and AX and NAX are agricultural and non-agricultural
exports with corresponding vectors of long-run parameters of b1i and b2i. Westerl-
und’s (2006a) bootstrap approach is applied to obtain critical values under general
conditions of cross-sectional dependence, and 5,000 replications are run. The results
are shown in Table 2. For the full sample and for each subsample, the null hypothe-
ses of panel cointegration are not rejected and there are meaningful long-run rela-
tionships between GDP and agricultural and non-agricultural exports. Of the
42 countries, 40 show evidence of a significant structural break with Ghana and
Sri Lanka being the exceptions.
Fully modified OLS estimates of equation (15) are shown in Table 3.11,12 There is
no particular pattern to the breaks, although 30 are between 1974 and 1988, with
five occurring around the oil crisis of 1973 ⁄ 1974 and the subsequent world recession
of 1974 ⁄ 1975 and six occurring shortly after the debt crisis of 1981 ⁄ 1982. Wald
tests, which are not reported, are used to test the null hypothesis that the intercept
and trend coefficients are equal across regimes. They show that 40 relationships
have a statistically significant broken level at the 5% significance level, 32 have
a broken trend and 30 have both. The modelling of the long-run export–income
relationship with structural breaks appears appropriate.
Under the export-led growth hypothesis, we expect the estimated export elastici-
ties to be positive. Those for agricultural exports are positive for 33 countries, of
11
The t-statistic for each panel parameter is the sum of the t-statistics across countries,
divided by the square of the number of countries. When negative individual t-statistics (which
correspond to negative coefficients) are larger in absolute terms than those that are positive,
negative panel t-statistics occur.
12
Results for DOLS are similar and are not reported.
2010 The Authors. Journal compilation 2010 The Agricultural Economics Society.
Table 3
Fully modified ordinary least squares results
Low-income countries
Bangladesh )0.014 (0.42) 0.156* (2.60) 24.010* (16.00) 23.917* (15.71) )0.011 (0.54) 0.025* (3.72) 1974 5.41* [0.03]
Benin )0.017 (0.65) )0.014 (0.73) 27.047* (30.56) 27.592* (30.98) 0.033* (9.64) 0.046* (12.43) 1988 0.01 [0.93]
Burkina Faso 0.005 (0.97) )0.005 (0.98) 26.729* (153.66) 27.430* (148.47) 0.038* (38.45) 0.041* (45.43) 1989 2.64 [0.11]
Cote d’Ivoire 0.193 (1.34) 0.023 (0.27) 22.778* (4.17) 23.151* (4.12) 0.065* (5.73) 0.013* (3.48) 1979 1.82 [0.19]
Gambia 0.028 (1.29) 0.021* (2.20) 19.419* (35.56) 19.797* (36.04) 0.057* (4.34) 0.032* (38.36) 1974 0.15 [0.70]
Ghana 0.178* (11.28) 0.126* (5.65) 16.226 (43.20) – 0.020 (15.13) – 2.37 [0.13]
Guinea-Bissau 0.077* (4.43) )0.004 (0.46) 18.675* (51.21) 19.295* (49.60) 0.031* (28.96) 0.006 (0.80) 1997 20.68* [0.00]
India 0.015* (0.64) 0.037 (0.90) 27.406* (35.52) 27.651* (34.98) 0.034* (7.83) 0.051* (14.37) 1978 0.13 [0.72]
Kenya 0.106 (1.84) 0.034 (0.64) 22.949* (9.75) 23.585* (10.02) 0.052* (8.14) 0.024* (5.32) 1980 1.68 [0.20]
Madagascar 0.070* (2.83) 0.075* (5.05) 23.082* (32.30) 23.234* (32.69) 0.003* (2.96) 0.019* (4.49) 1996 0.02 [0.88]
Pakistan 0.094* (5.60) 0.049* (2.45) 23.866* (42.21) 25.116* (41.75) 0.052* (27.09) 0.029* (14.05) 1993 2.53 [0.12]
Senegal 0.083* (9.94) )0.033 (0.72) 26.198* (21.45) 26.706* (21.32) 0.031* (12.90) 0.034* (10.77) 1988 6.84* [0.01]
Zambia 0.004 (0.36) 0.144* (2.38) 24.203* (14.25) 24.369* (14.65) 0.012* (6.95) 0.022* (4.59) 1993 5.21* [0.03]
Panel 0.091* (15.62) 0.062* (7.98) – – – – – 0.02 [0.88]
Lower middle income countries
Algeria )0.033* (2.46) 0.631* (7.14) 10.444* (4.67) 10.900* (4.84) 0.036* (6.17) )0.002 (0.63) 1980 58.47* [0.00]
2010 The Authors. Journal compilation 2010 The Agricultural Economics Society.
Bolivia 0.144* (4.51) 0.088 (1.12) 18.404* (11.75) 18.611* (11.95) 0.029* (6.44) 0.012* (2.70) 1985 0.34 [0.56]
Brazil )0.007 (0.12) 0.025 (0.54) 25.636* (14.64) 26.299* (14.57) 0.091* (7.04) 0.023* (4.28) 1976 0.24 [0.63]
Cameroon 0.174* (3.80) 0.092 (1.94) 20.561* (11.15) 21.210* (10.80) 0.050* (5.62) 0.015* (3.41) 1987 2.10 [0.16]
Colombia 0.117* (2.22) )0.011 (0.42) 27.819* (14.33) 28.821* (14.17) 0.039* (10.84) 0.027* (4.51) 1997 8.03* [0.01]
Congo Republic )0.003 (0.11) 0.264* (1.98) 19.269* (5.72) 19.933* (5.68) 0.032* (2.49) 0.005 (0.87) 1981 3.88 [0.06]
Costa Rica 0.103* (2.31) 0.108* (4.65) 21.535* (15.87) 21.827* (15.82) 0.042* (10.52) 0.026* (6.62) 1980 0.01 [0.91]
Dominican Republic 0.166* (3.59) 0.037 (1.17) 19.539* (15.69) 20.005* (15.50) 0.046* (7.88) 0.027* (5.33) 1983 5.57* [0.02]
Ecuador 0.033 (1.81) 0.024* (2.22) 21.249* (50.71) 21.878* (51.44) 0.067* (12.37) 0.019* (11.96) 1978 0.16 [0.69]
Agricultural Exports and Economic Growth in Developing Countries
Egypt 0.000 (0.00) 0.023 (0.35) 23.734* (13.14) 24.648* (12.95) 0.070* (5.53) 0.042* (10.00) 1981 0.10 [0.75]
El Salvador 0.133* (2.83) 0.157* (5.30) 16.317* (17.94) 16.376* (17.99) 0.033* (5.64) 0.015* (3.50) 1979 0.14 [0.71]
Guatemala 0.295* (7.64) 0.148* (4.47) 12.853* (16.82) 13.281* (17.37) 0.027* (19.20) 0.019* (6.85) 1988 5.94* [0.02]
575
Table 3
576
(Continued)
Indonesia )0.015 (1.27) 0.094* (4.31) 30.474* (40.95) 32.036* (41.36) 0.064* (53.17) 0.042* (9.80) 1997 17.02* [0.00]
Lesotho 0.046 (0.96) 0.060 (1.40) 18.525* (13.48) 19.162* (13.36) 0.070* (3.51) 0.038* (7.61) 1976 0.09 [0.77]
Morocco 0.016 (0.42) 0.173 (1.67) 20.196* (7.92) 20.569* (7.96) 0.046* (2.42) 0.020* (2.72) 1975 2.21 [0.15]
Nicaragua 0.194* (2.80) 0.129* (2.98) 17.021* (8.62) 16.849* (8.87) 0.007 (1.18) 0.007 (0.80) 1988 0.84 [0.37]
Paraguay 0.075* (4.78) 0.009 (0.42) 24.293* (53.32) 25.161* (52.91) 0.074* (16.24) 0.021* (14.02) 1979 3.96* [0.05]
Peru 0.105* (2.14) 0.086 (1.01) 20.758* (8.23) 20.976* (8.31) 0.025* (6.51) 0.021* (2.22) 1988 0.06 [0.81]
Philippines 0.050 (1.69) 0.004 (0.11) 25.194* (22.19) 25.700* (21.93) 0.052* (11.65) 0.034* (9.08) 1983 0.71 [0.40]
Sri Lanka 0.023 (0.70) 0.040 (1.71) 24.665* (20.60) – 0.040* (14.48) – – 0.34 [0.56]
Thailand 0.055 (0.46) 0.213* (3.67) 20.338* (5.69) 21.239* (5.52) 0.038* (3.08) 0.024 (1.54) 1997 1.84 [0.18]
Tunisia 0.034 (1.79) 0.139* (3.59) 18.546* (19.29) 19.163* (19.15) 0.046* (12.56) 0.036* (14.25) 1985 7.11* [0.01]
Panel 0.077* (8.63) 0.115* (10.94) – – – – – 3.20 [0.07]
Upper middle income countries
Argentina 0.135 (1.41) 0.149* (2.21) 19.765* (9.32) 19.734* (8.93) )0.010 (1.73) )0.010 (1.27) 1991 0.01 [0.92]
Chile 0.026 (0.35) 0.642* (5.11) 11.255* (3.36) 10.579* (3.09) )0.076* (2.14) 0.001 (0.06) 1974 14.94* [0.00]
Malaysia 0.100* (2.95) 0.454* (12.97) 11.532* (10.06) 11.745* (9.77) 0.021* (4.99) 0.008 (1.78) 1984 58.90* [0.00]
Mexico )0.077* (3.13) 0.060 (1.88) 27.300* (24.94) 27.815* (24.33) 0.051* (8.94) 0.030* (6.48) 1982 19.13* [0.00]
South Africa )0.075* (3.84) 0.048 (0.53) 27.526* (11.44) 27.803* (11.36) 0.017* (6.69) 0.029* (6.20) 1991 1.94 [0.17]
2010 The Authors. Journal compilation 2010 The Agricultural Economics Society.
Trinidad and Tobago )0.153* (4.43) 0.512* (3.68) 15.381* (4.21) 15.515* (4.25) 0.024* (4.65) 0.005 (0.55) 1983 29.51* [0.00]
Uruguay 0.208* (2.49) 0.256* (4.90) 11.484* (7.36) 11.189* (6.83) )0.017 (1.73) 0.002 (0.29) 1982 0.22 [0.64]
Ana I. Sanjuán-López and P. J. Dawson
Notes: t-statistics are given in parentheses; p-values are given in brackets. Significance is at the *5% level.
Agricultural Exports and Economic Growth in Developing Countries 577
which 18 are significant; nine coefficients are negative of which four – Algeria,
Mexico, South Africa and Trinidad and Tobago – are significant. The magnitudes
of the positive elasticities are generally small, as expected: 19 are less than 0.1, 12
are between 0.1 and 0.2, whereas those for Guatemala and Uruguay are 0.3 and
0.21. Similarly, elasticities for non-agricultural exports are positive for 37 countries,
of which 21 are significant; and the five negative estimates are all insignificant. The
magnitudes of the positive non-agricultural export elasticities are also generally
small: 20 are less than 0.1, 10 are between 0.1 and 0.2 and seven are greater than
0.2 with Algeria, Chile, Malaysia and Trinidad and Tobago being surprisingly large
at over 0.45.
The panel FMOLS estimates provide some general insights. As each country is
permitted its own break, there is no estimate of a break for the panel. Table 3
shows that both agricultural and non-agricultural exports significantly determine
GDP. For the full sample, a 1% increase in agricultural exports increases GDP by
0.07%. Corresponding elasticities are 0.09 for low-income countries, 0.08 for those
with lower middle incomes and 0.02 for those with upper middle incomes. Thus, the
impact of an increase in agricultural exports on GDP seems to fall as income
increases, although it is not possible to test whether these differences are statistically
significant. Similarly for the full sample, a 1% increase in non-agricultural exports
increases GDP by 0.13%. Corresponding elasticities for low-, lower middle and
upper middle income countries are 0.06, 0.12 and 0.29; so, the impact of an increase
in non-agricultural exports on GDP appears to increase as income increases.
Wald statistics are used to test the hypotheses that for each country the coeffi-
cient associated with agricultural exports is the same as that associated with non-
agricultural exports, that is, b1 = b2 for each i = 1,…, N. Table 3 shows that this
null hypothesis is rejected for 15 countries at the 5% significance level. There is no
obvious pattern of rejections: the null hypothesis is rejected for four of 13 low-
income countries, for seven of 22 lower middle income countries and for four of
seven upper middle income countries. For the full sample and for upper middle
income countries, the null hypothesis that b1 = b2 is also rejected; for lower middle
income countries, it is rejected at the 10% significance level; and for low-income
countries, it is not rejected. Thus, for low-income countries, changes in agricultural
and non-agricultural exports have a statistically identical effect on GDP, whereas
for lower and upper middle income countries, a change in non-agricultural exports
has a greater impact on GDP than a similar change in agricultural exports, as, per-
haps, would be expected.
Finally, we test for non-causality from total exports (both agricultural exports
and non-agricultural exports together) to GDP for each country from heteroge-
neous panel ECMs in equation (9). The results are shown in Table 4. The null
hypothesis of non-causality in both the short and long run for each country, that is,
H01 in equation (10), is rejected for 26 countries, whereas the null hypothesis in the
long run only, that is, H02 in equation (11), is also rejected for 26, but not the same,
countries. There is no pattern of rejections between income groups. Corresponding
tests for the panels are H03 in equation (12) and H04 in equation (13). Both for the
full sample and for all sub-samples by income group, the null hypotheses are con-
vincingly rejected. Similarly, the group mean panel tests, where the null hypothesis
is of no long-run causality for the panel on average, that is, H05 in equation (14),
imply that total exports cause GDP. Homogeneity tests show evidence of heteroge-
neous long-run coefficients within each sample. Our findings therefore lead us to
2010 The Authors. Journal compilation 2010 The Agricultural Economics Society.
578 Ana I. Sanjuán-López and P. J. Dawson
Table 4
Non-causality tests from total exports to GDP
Low-income countries
Bangladesh 3 3.77 [0.01]* )1.28 [0.22]
Benin 1 4.75 [0.01]* )2.79 [0.01]*
Burkina Faso 1 6.58 [0.00]* )4.09 [0.00]*
Cote d’Ivoire 1 0.89 [0.46] )1.06 [0.30]
Gambia 1 4.49 [0.01]* )3.43 [0.00]*
Ghana 1 2.16 [0.11] )2.45 [0.02]*
Guinea-Bissau 1 5.71 [0.00]* )2.67 [0.01]*
India 1 7.79 [0.00]* )4.40 [0.00]*
Kenya 1 4.25 [0.01]* )3.32 [0.00]*
Madagascar 1 10.43 [0.00]* )3.78 [0.00]*
Pakistan 1 1.58 [0.22] 2.07 [0.05]*
Senegal 1 1.61 [0.21] )0.77 [0.45]
Zambia 1 3.35 [0.03]* )2.38 [0.02]*
Panel – 157.37 [0.00]* 87.61 [0.00]*
Group mean test )2.65 [0.00]*
Homogeneity test 46.74 [0.00]*
Lower middle income countries
Algeria 2 1.33 [0.29] )0.96 [0.35]
Bolivia 1 3.43* [0.03]* )1.66 [0.11]
Brazil 1 6.57* [0.00]* )4.26 [0.00]*
Cameroon 1 1.89 [0.16] )1.51 [0.14]
Colombia 1 4.61 [0.01]* )3.70 [0.00]*
Congo Republic 1 0.99 [0.41] )1.72 [0.10]
Costa Rica 1 7.56 [0.00]* )4.55 [0.00]*
Dominican Republic 1 5.34 [0.01]* )3.98 [0.00]*
Ecuador 1 9.64 [0.00]* )3.77 [0.00]*
Egypt 1 7.94 [0.00]* )4.71 [0.00]*
El Salvador 1 3.73 [0.02]* )0.97 [0.34]
Guatemala 1 0.83 [0.49] )1.50 [0.15]
Indonesia 1 0.39 [0.76] )1.00 [0.33]
Lesotho 1 6.83 [0.00]* )4.44 [0.00]*
Morocco 1 3.90 [0.02]* )3.34 [0.00]*
Nicaragua 1 4.48 [0.01]* )3.48 [0.00]*
Paraguay 1 12.31 [0.00]* )5.47 [0.00]*
Peru 1 4.23 [0.01]* )3.41 [0.00]*
Philippines 1 1.73 [0.18] )1.75 [0.09]
Sri Lanka 1 3.32 [0.03]* )2.80 [0.01]*
Thailand 1 1.88 [0.16] )1.32 [0.20]
Tunisia 2 1.13 [0.37] )1.54 [0.14]
Panel – 256.59 [0.00]* 216.12 [0.00]*
Group mean test )2.81 [0.00]*
Homogeneity test 36.88 [0.00]*
Upper middle income countries
Argentina 1 11.47 [0.00]* )5.61 [0.00]*
Chile 1 0.77 [0.52] )1.05 [0.31]
Malaysia 1 2.67 [0.07] )2.67 [0.01]*
Mexico 1 1.44 [0.25] )1.58 [0.12]
2010 The Authors. Journal compilation 2010 The Agricultural Economics Society.
Agricultural Exports and Economic Growth in Developing Countries 579
Table 4
(Continued)
Notes: P-values are given in square brackets. Significance is at the *5% level.
conclude that there is strong evidence to support the export-led hypothesis that
total exports cause GDP.
13
Balassa (1981) discusses in detail the effects of the oil crisis on key macroeconomic vari-
ables, including exports and economic growth, for newly industrialised countries.
2010 The Authors. Journal compilation 2010 The Agricultural Economics Society.
Agricultural Exports and Economic Growth in Developing Countries 581
References
Ahmad, J. ‘Causality between exports and economic growth: What do the econometric stud-
ies tell us?’ Pacific Economic Studies, Vol. 6, (2001) pp. 147–167.
Amsler, C. and Lee, J. ‘An LM test for a unit root in the presence of a structural break’,
Econometric Theory, Vol. 11, (1995) pp. 359–368.
Bai, J. and Perron, P. ‘Estimating and testing linear models with multiple structural changes’,
Econometrica, Vol. 66, (1998) pp. 47–78.
Bai, J. and Perron, P. ‘Computing and analysis of multiple structural change models’, Journal
of Applied Econometrics, Vol. 18, (2003) pp. 1–22.
2010 The Authors. Journal compilation 2010 The Agricultural Economics Society.
582 Ana I. Sanjuán-López and P. J. Dawson
2010 The Authors. Journal compilation 2010 The Agricultural Economics Society.
Agricultural Exports and Economic Growth in Developing Countries 583
Jung, W. S. and Marshall, P. J. ‘Exports, growth and causality in developing countries’, Jour-
nal of Development Economics, Vol. 18, (1985) pp. 1–12.
Kaplinsky, R. ‘Revisiting the revisited terms of trade: Will China make a difference?’ World
Development, 34, (2006) pp. 981–995.
Kunst, R. M. and Marin, D. ‘On exports and productivity: A causal analysis’, Review of
Economics and Statistics, Vol. 71, (1989) 699–703.
Kwiatkowski, D., Phillips, P. C. B., Schmidt, P. and Shin, Y. ‘Testing the null hypothesis of
stationarity against the alternative of a unit root: How sure are we that economic time
series have a unit root?’ Journal of Econometrics, Vol. 54, (1992) pp. 159–178.
Levin, A. and Raut, L. K. ‘Complementarities between exports and human capital in eco-
nomic growth: Evidence from the semi-industrialized countries’, Economic Development and
Cultural Change, Vol. 46, (1997) pp. 155–174.
Love, J. ‘Engines of growth: The export and government sectors’, The World Economy,
Vol. 17, (1994) pp. 203–218.
Lütkepohl, H. Introduction to Multiple Time Series (Berlin: Springer-Verlag, 1993).
Michaely, M. ‘Exports and growth: An empirical investigation’, Journal of Development Eco-
nomics, Vol. 4, (1977) pp. 49–53.
Pedroni, P. ‘Purchasing power parity tests in cointegrated panels’, Review of Economics and
Statistics, Vol. 83, (2001) pp. 727–731.
Pedroni, P. ‘Fully modified OLS for heterogeneous cointegrated panels’, in B. H. Baltagi
(ed.), Nonstationary Panels, Panel Cointegration and Dynamic Panels (Advances in Econo-
metrics, Vol. 15) (New York: Elsevier Science Inc, 2002, pp. 93–130).
Perron, P. ‘The great crash, the oil price shock and the unit root hypothesis’, Econometrica,
Vol. 57, (1989) pp. 1361–1401.
Ram, R. ‘Exports and economic growth in developing countries: Evidence from time-series
and cross-section data’, Economic Development and Cultural Change, Vol. 36, (1987)
pp. 51–72.
Sharma, S. C. and Dhakal, D. ‘Causal analyses between exports and economic growth in
developing countries’, Applied Economics, Vol. 26, (1994) pp. 1145–1157.
Sheehey, E. J. ‘Exports and growth: A flawed framework’, Journal of Development Studies,
Vol. 27, (1990) pp. 111–116.
Tyler, W. ‘Growth and export expansion in less developed countries: Some empirical evi-
dence’, Journal of Development Economics, Vol. 9, (1981) pp. 121–130.
Westerlund, J. ‘Testing for panel cointegration with multiple structural breaks’, Oxford Bulle-
tin of Economics and Statistics, Vol. 68, (2006a) pp. 101–132.
Westerlund, J. Simple Unit Root Tests with Multiple Breaks (Mimeo, University of Lund,
Sweden, 2006b). Available at: http://www.nek.lu.se/nekjwe/papers/imetal.pdf.
World Bank. World Development Report 1987 (Washington, DC: World Bank, 1987).
World Bank. World Development Indicators (Edition: April 2006). Accessed through Eco-
nomic and Social Data Service, http://www.esds.ac.uk/.
Xu, Z. ‘On the causality between export growth and GDP growth: An empirical re-investiga-
tion’, Review of International Economics, Vol. 4, (1996) pp. 172–184.
Zapata, H. O. and Gil, J. M. ‘Cointegration and causality in international agricultural eco-
nomics research’, Agricultural Economics, Vol. 20, (1999) pp. 1–9.
2010 The Authors. Journal compilation 2010 The Agricultural Economics Society.
Copyright of Journal of Agricultural Economics is the property of Wiley-Blackwell and its content may not be
copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written
permission. However, users may print, download, or email articles for individual use.