Narayan Chandra Pradhan*
Empirical verification of export-led growth (ELG) hypothesis by applying various
time series techniques reveals both short- and long-run relationship between export growth
and output growth. The research question that has been addressed in the study is: whether
openness has impact on growth, and if so, then in what direction? Bivariate Granger causality
test suggests that the direction of causality runs from export growth to GDP growth. This
fact implies that one can use exports to better predict the GDP than simply by the past
history of GDP. The aim is to substantiate the importance of exports in the growth process
of Indian economy after its opening up to the world economy.
JEL Classification : F1, F43, O53.
Keywords : International Trade in Goods and Services, Economic Growth, India
I. Introduction
Economists across the board have agreed with the opinion that the
process of economic growth is an extremely complex phenomenon. It depends
on many variables, such as, capital accumulation (both physical and human),
international trade, price condition, political situation, income distribution,
and even more on geographical factors. Export-led growth (ELG) hypothesis
postulates that export expansion is one of the prime determinants of economic
growth. The overall growth of countries can be generated not only by
increasing the amounts of labour and capital within the economy as the
classical economists postulates, but also by expanding exports to wider
markets. According to the proponents of ELG hypothesis, exports can perform
the function of an ‘engine of growth’. The association between exports and
economic growth is often attributed to the positive externalities for the domestic economy arising from participation in world markets, for instance,
from the re-allocation of existing resources, economies of scale and various
labour-specialisation effects (Bhagwati, 1978; Krueger, 1978).
The term ELG hypothesis is seldom explicitly defined in
economic literature. However, it is clear that most authors have in
mind some notion of a multiplier effect, whereby, an initial favourable
shock in the export sector sets in motion forces leading to additional
economic growth. Kindleberger (1962) defines trade as a leading
sector when ‘exports rise would lead to an incentive for the
establishment and expansion of other peripheral activities’. Meier
(1976) explained that the export sector acts ‘as a key propulsive
sector, propelling the rest of the economy forward’. In keeping with
the spirit of these definitions, the criterion adopted here for ‘strong
export-led growth’ (SELG) is that expansion in the export sector
should stimulate aggregate capital accumulation. This is the natural
criterion in that with a larger capital stock the increase in steady-state
growth is greater than the direct gain conferred by the terms of trade
improvement or resource discovery. An increase in the steady-state
capital stock is a necessary but not sufficient condition for growth to
be higher in the long run. If it can be established that steady-state
growth increases despite a decline in the aggregate capital stock, the
outcome will be labelled as ‘weak export-led growth’ (WELG).
Finally, when capital decumulation is severe enough to lower steadystate
growth, the outcome will be characterised as ‘export-led fizzle’
(ELF).
In fact, during the 1990s a new series of empirical studies has been
conducted on a number of divergent lines of research methodologies,
time periods and countries. A key aspect of the earlier studies is related
to both the analytical and the econometrics technique used. Earlier
studies could have been misleading in the sense that they advocated
export expansion in an indiscriminate way (Feder, 1982). In fact, the
evidence available is inconclusive and this situation explains to some
extent why this debate still exists in the economic literature. Added to
this debate is the question of whether strong economic performance is ‘export-led’ or ‘growth-driven’. This question is important because the determination of the causal pattern between export and economic
growth has important implications for policy-makers’ decision about
the appropriate growth and development strategies.
Although, most studies focus on the causal link between exports
and output growth in industrialised countries (Marin, 1992; Serletis,
1992; Henriques and Sadorsky, 1996), some researchers have
examined the export-led growth hypothesis with emphasis on
developing countries (Michaely, 1977; Balassa, 1978). Using data
from selected industrialised countries, Marin (1992) examines the
causal link between exports and productivity and finds that the exportled
growth (ELG) hypothesis cannot be rejected for Germany, Japan,
the United Kingdom, and the United States. Henriques and Sadorsky
(1996) similarly focused on the export and output growth relationship
for Canada using three variables (GDP, exports, and terms of trade).
They employ a multivariate cointegration estimation methodology
that accounted for potential feedback and simultaneity effects between
these three variables. In contrast to Serletis’s (1992) earlier result,
Henriques and Sadorsky (1996) find that ‘changes in GDP precede
changes in exports’.
The lack of consistent causal pattern between exports and output
growth in earlier studies may be due to one or more of the following
issues. The causal models in those studies may be mis-specified
because of: (i) the omission of an important variable, such as, capital
and foreign output growth; (ii) the traditional Granger causality F-test
in a regression context may not be valid if the variables in the system
are cointegrated, since the test statistic does not have a standard
distribution (Toda and Philips, 1993); and (iii) temporal aggregation
issues from the use of annual time series may yield erroneous causation
results (Bahmani-Oskooee and Alse, 1993).
Consequently, the purpose of this article is to examine the nexus
between export growth and economic growth and test the ELG
hypothesis for Indian economy. The analysis has three distinctive
features differentiated from earlier empirical studies: (i) the study has
gone beyond the traditional neo-classical theory of production function by estimating an augmented Cobb-Douglas functional form,
which includes exports using annual data for the period 1970-71 to
2009-10. This study also includes services exports to that of
merchandise one, as earlier studies generally based upon; (ii) the
analysis carried out by focussing on a single country – India, instead
of cross-country comparison; (iii) the study has examined empirically
the long-run relationship, using procedures like unit root tests,
stationarity, cointegration, Granger causality and vector autoregresion
(VAR). Thus, the aim is to substantiate the importance of exports in
the ‘growth process of Indian economy’ after its opening up to the
world economy.
II. Literature Survey
II.1 Literature on ELG Hypothesis
For the last two decades, there has been massive resumption of
economic growth literature triggered by the ‘endogenous growth
theory’, which has led to the propagation of models that stress the
importance of trade in achieving a sustainable rate of economic
growth. These models have focused on different variables, such as,
degree of openness, real exchange rate, tariffs, terms of trade and
export performance, to verify the hypothesis that open economies
grow more rapidly than the closed ones (Edwards, 1998). Although,
most models emphasised the nexus between trade and growth, they
stressed that trade is only one of the variables that enter the growth
equation. However, the advocates of the ELG hypothesis have stated
that trade, in fact, was the main engine of growth in Southeast Asia.
They argue that, for instance, Hong Kong, Taiwan, Singapore and
South Korea, the so-called Four Tigers, have been successful in
achieving high and sustained rates of economic growth since the early
1960s because of their free-market and the outward-oriented
economies (World Bank, 1993). The literature concerning the
relationship between trade and growth is also the consequence of the
many changes that have taken place in the fi elds of development
economics and international trade policy.
Although, a substantial part of the earlier studies found evidence
of a correlation between exports and growth which was used to support
the ELG hypothesis, this tends to hold only for cross-sectional studies.
In fact, the recent evidence on time series, which makes extensive use
of cointegration techniques, casts doubts on the positive effects of
exports on growth in the long-run, and is thus not as conclusive as it
was previously thought.
Among earlier major empirical studies, Emery (1967), Syron
and Walsh (1968), Heller and Porter (1978), Bhagwati (1978) and
Krueger (1978) can be mentioned. These studies explained economic
growth in terms of export expansion alone in a two-variable framework.
That is, they used bi-variate correlation - the Spearman’s rank
correlation test - in cross-country format to illustrate the superior
effects of the ELG hypothesis (Lussier, 1993). A second group of
researchers, which includes Balassa (1978), Tyler (1981), Feder
(1982), Kavoussi (1984), Ram (1985, 1987) and Moschos (1989)
studied the relationship between export and output performance
within a neo-classical framework. In most of these studies exports
were included in an ad hoc manner in the production function, together
with labour and capital. They claimed that by including exports they
were taking into consideration a broad measure of externalities and
productivity gains generated by this sector which stimulated the
domestic economy. The majority of these investigations aimed at
analysing developing countries by using ordinary least squares (OLS)
on cross-section data and used their results to demonstrate the
advantages of the export promotion strategy in comparison with the
import substitution policy.
For most of the country-specific studies, both industrialised and
developing, the empirical investigations found no long-run relationship
between exports and economic growth; rather, the studies suggest that
it arises only from a positive short-term relationship between export
expansion and growth of gross domestic product (GDP). The studies
of industrialised nations have analysed the cases of Canada, France,
Germany, the United Kingdom, the United States and Switzerland,
among others. In only a few cases have the empirical results confirmed that export expansion was a key element in the economic success of
those countries (Kugler, 1991; Afxentiou and Serletis, 1991; Henriques
and Sadorsky, 1996). Even more surprising is the finding in relation to
Japan, which states that internal forces were the handmaidens of the
great Japanese economic success in the twentieth century, including
the post-war period, and not trade as many have claimed in the recent
past (Boltho, 1996).
The most recent time series investigations concerning developing
countries that have used the econometric methodology of cointegration
have not been able to establish unequivocally that a robust relationship
between these variables indeed exists in the long term, namely that
the variables are cointegrated (Islam, 1998). While some have been
able to find a long-run relationship, many others have rejected the
ELG hypothesis i.e., that export expansion causes growth in the long
run. In fact, in most studies the results suggest that this arises owing
to a simple short-term relationship, a feature that is not surprising, if
we take into account the fact that the studies that have concentrated
their attention on industrialised nations have also been unable to find
a robust relationship between these variables (Kugler, 1991).
Berg and Schmidt (1994) found cointegration in 11 of the 16 Latin
American Countries analysed. In fact, in the case of Costa Rica they
were able to verify the existence of a long-term relationship. Although
the result seems to endorse in general the export-led hypothesis, they
seem to deviate from those recently reported by the empirical literature
(Rodrik, 1999). However, a possible justifi cation of the positive results
obtained in the investigation conducted by Berg and Schmidt (1994) is
that these researchers employed population and investment as proxies
for the appropriate aggregate inputs, i.e. labour force and capital stock.
Although they have been widely used in many cross-section growth
studies as well as time series analysis (Al-Yousif, 1997), many
researchers have had serious doubts about them and have thus expressed
their suspicion regarding studies that have tested the export promotion
hypothesis by using, for instance, the investment-output ratio, i.e.
gross domestic investment (GDI)/gross domestic product (GDP), as
opposed to capital stock or population instead of labour force.
Though, there are numerous facets to the trade-growth nexus,
most of the literature has concentrated on disturbances connected with
the export sector. The ELG hypothesis has frequently been invoked to
explain differences in development patterns among developing
countries. The trade theorists have also examined the growing concern
over the potentially adverse effects of a booming natural resource
based export sector termed as Dutch Disease phenomenon. The
literature on this special aspect focus on the impact of a rise in export
revenues from an inelastically supplied, resource intensive product
that uses little capital or labour and is not consumed domestically -
and tends to make de-industrialisation, not aggregate growth, its
principal concern (Buffie, 1992).
Despite the sizeable literature generated by the ELG hypothesis,
little is known about how various export shocks might affect the
economic growth. The numerous case studies done by development
economists and economic historians are full of suggestive ideas but
do not point to any firm conclusions. Country experiences have
varied widely and in the absence of any explicit theoretical framework
linking export shocks and the main determinants of economic growth.
It is, thus, difficult to judge whether export sector expansion has
stimulated growth, retarded growth, or merely accompanied growth
or contraction in the rest of the economy (Kindleberger, 1961; Kravis,
1970; Meier, 1976).
There have been studies on the existence of a threshold effect as
well (Kavoussi, 1984; Moschos, 1989). These studies have been
supplemented by causality tests (Jung and Marshall, 1985). The
econometric methods employed in these analyses have been
significantly influenced by the work of Granger (1969), Engle and
Granger (1987), and Johansen and Juselius (1990), among others. The
studies such as Jung and Marshall (1985), Afxentiou and Serletis
(1991), and Dodaro (1993) have cast some doubt on the validity of the
ELG hypothesis. Others such as Serletis (1992), Henrique and
Sadorsky (1996), Bahmani-Oskooee and Alse (1993), and Nidugala
(2001) provide fairly robust evidence in favour of the ELG hypothesis.
Most of the time series studies employ the Granger method, while only a few studies combine Granger’s test with the Akaike’s
Information Criterion (AIC) to determine the optimal lag length in the
Granger causality test. The latter approach removes the ambiguity
involved in the arbitrary choice of lag lengths.
The idea that export growth is one of the major determinants of
output growth - ELG hypothesis - is a recurrent one. Export growth
may affect output growth through positive externalities on nonexports,
through the creation of more efficient management styles,
improved production techniques, increased scale economies, improved
allocative efficiency, and better ability to generate dynamic
comparative advantage. If there are incentives to increase investment
and improve technology, this would imply a productivity differential
in favour of the export sector. It is thus argued that an expansion of
exports, even at the cost of other sectors, will have a net positive effect
on the rest of the economy (Balassa, 1978). It may also ease the
foreign exchange constraint. There could also be positive spillover
effects on the rest of the economy. These factors, notwithstanding, the
empirical evidence for the ELG hypothesis is mixed.
II.2 ELG Hypothesis: Studies on India
There is fair amount of literature on ELG hypothesis pertaining
to Indian economy. Majority of the empirical studies found lack of
causality between export and economic growth in both directions.
Jung and Marshall (1985) and Dodaro (1993) reported an insignificant
F-statistic for real export growth to real income growth as well as in
other way round, although the sign is positive in both cases. Similarly,
Dhawan and Biswal (1999) investigate the ELG hypothesis using a
vector autoregressive (VAR) model by considering the relationship
between real GDP, real exports and terms of trade during 1961-1993.
They employ a multivariate framework using Johansen’s
cointegration procedure and find a long-run equilibrium relationship
between these three variables and the causal relationship flows from
the growth in GDP and terms of trade to the growth in exports.
However, they conclude that the causality from exports to GDP
appears to be a short-run phenomenon. In a similar framework, Asafu-Adjaye et al. (1999) consider three variables: exports, real
output and imports for the period 1960-1994. They do not find any
evidence of the existence of a causal relationship between these
variables in case of India and no support for the ELG hypothesis,
which is not too surprising given India’s economic history and trade
policies. Anwer and Sampath (2001) also find evidence against the
ELG hypothesis for India.
Mallick (1996), using annual data for the period 1950-92 and
employing Engle-Granger cointegration cum error-correction
procedure, finds a strong cointegration between income and exports,
and that the direction of causality runs from income growth to export
growth (i.e., growth-led exports). While the Granger-causality tests,
in his study, are sensitive to the lag length chosen and do not show
consistent causal flow from income growth to export growth, the
results of the error-correction model show that causation runs from
income growth to export growth (as the error-correction term is
significant) irrespective of the lag length chosen. This seems to
suggest that the causality found by Mallick (1996) is a long-term
phenomenon.
Nidugala (2001) finds evidence in support of the ELG hypothesis,
particularly in the 1980s. His study reveals that growth of manufactured
exports had a significant positive relationship with GDP growth, while
the growth of primary exports had no such infl uence. Ghatak and
Price (1997) test the ELG hypothesis for India during 1960-1992,
using ‘GDP net of exports’ as regressor, along with exports and
imports as additional variables. Their results indicate that real export
growth Granger-caused by non-export real GDP growth over the
period 1960-1992. Their cointegration tests confirmthe long-run
nature of this relationship. However, imports do not appear to be
important in those studies.
Chandra (2002), on the other hand, finds that export growth
and GDP growth are interlocked in a two-way relationship as
opposed to Mallick (1996). Chandra also finds that real exports and
real GDP are not cointegrated in lndia, implying that there is no long-run relationship between them. Sinha (1996) envisaged the
relationship between openness and economic growth in India,
wherein, two types of analysis were performed. First, long run
relationship between GDP and openness was studied. Secondly, tests
were performed to find the causal relationship between GDP and
openness. The result of the Granger causality test shows that there is
a two-way causality between the growth of GDP and openness.
Thus, a reduction in trade barriers is likely to promote economic
growth i.e., the results of the study show that openness contribute to
growth of GDP which implies that both exports and imports
contribute to economic growth in the long run.
Marjit and Raychaudhari (1997) have analysed the implications
of specific trade policies on exports and gross domestic products.
They assumed that all the domestic demand will be catered by
domestic production which leads to a decline in exports to some
extent. In case of India, GDP granger causes export growth (yearly
data for 1951 to 1994), but not vice versa according to their results.
The volume of imports was hypothesised to be an increasing function
of the real GDP and foreign exchange reserves and decreasing function
of the relative price notion. A dummy variable was also introduced to
account for the introduction of economic reforms. The study found
that a growth in exports volume was due to growth in real income.
The ordinary least square results of the study indicate that income
elasticity of exports (with respect to world real income) is higher than
the income elasticity of imports.
Sharma and Panagiotidis (2004) re-examines the sources of
growth for the period 1971-2001 based upon Feder’s (1982) model to
investigate empirical relationship between export growth and GDP
growth (the export led growth hypothesis). They investigate the
following hypotheses: (i) whether exports, imports and GDP are
cointegrated using the Johansen approach and Breitung’s nonparametric
cointegration test, (ii) whether export growth Granger
causes GDP growth, and (iii) whether export growth Granger causes
investment. They fail to find support for the hypothesis that exports
Granger cause GDP, using two measures for GDP (GDP with exports and GDP without exports). The same also holds for the relationship
between exports and investment.
From the review of empirical literature on exports and growth, it
is clear that the exports do not necessarily cause growth. The results
reported are clearly sensitive to the variables employed, theoretical
approach used and even on the econometric methodology employed.
For example, cross-section studies are more likely to corroborate a
positive relationship between exports and growth, while the results of
time series studies depend substantially on the countries analysed, the
period chosen and the econometric methods used. In addition, since
cross-section studies can obscure particularities of developing
countries, especially, those that are low-income countries, the correct
strategy to follow from an empirical point of view is to address the
issue in a single country framework, using as much as possible the
recent developments in time series analysis.
III. Empirical Analysis and Results
III.1 Data, Source and Explanations
The data set consists of observation on India’s export of goods
and services, real GDP, gross domestic capital formation, real effective
exchange rate, and the world GDP. It may be mentioned that, there are
two basic sources for data on India’s exports. One set is compiled by
the Directorate General of Commercial Intelligence and Statistics
(DGCI&S), Ministry of Commerce and Industry, Government of India
and the other set is compiled by the Reserve Bank of India (RBI). The
DGCI&S compiles information on real transactions, reporting
quantities/volumes of exports as well as export earnings in Rupee
terms. Merchandise exports are decomposed into headings compatible
with the ITC (HS)1 Standard Industrial Classification (SIC) codes.
Thus, merchandise exports are disaggregated by SIC categories and by destination (i.e. according to commodity and country Classification).
The RBI export data (both goods and services) is compiled by
aggregating the economy-wide financial transactions related to
exports, as reported by exporting firms. Exporters and financial
intermediaries have to provide this information to RBI on the basis of
statutory provision. DGCI&S data has been used frequently in the
trade analysis as the case of merchandise data is concerned. RBI’s
data based on Balance of Payments (BoP) basis has been used
relatively less frequently. As the current study is concerned with
services exports as well, it is decided to use the RBI’s data sets for our
analysis. Accordingly, the data used in this exercise has been obtained
from the ‘Handbook of Statistics on the Indian Economy 2009-10’
(HBSIE). The data for the current empirical analysis pertaining to the
period 1970-71 to 2009-10 compiled from HBSIE, partly owing to
ease of availability at this end, and partly for using exports of both
goods and services.
The time series data on real GDP and gross domestic capital
formation (GDCF) are obtained from the ‘Central Statistical
Organisation’ of the Government of India (Base Year: 1999-2000), the
same is also published in HBSIE for the period 1970-71 to 2009-10.
The time series data on real effective exchange rate (REER) are
calculated from the RBI’s HBSIE based on splicing methodology. It
may be mentioned that the data on REER up to 1992 are based on
official exchange rates and data from 1993 onwards are based on
Foreign Exchange Dealers’ Association of India (FEDAI) indicative
rates. REER indices are recalculated from April 1993 onwards using
the new wholesale price index (Base: 1993-94=100). A new 6-currency
REER series (Trade-based weights) has been introduced with effect
from December 2005.
The data set is annual and covers the period 1970-71 (financial
year data represented by 1970 in econometric analysis) to 2009-10
(similarly represented by 2009). The data description and their
specifications in empirical analysis are as follows:
(1) RGDP: Real GDP (GDP at factor cost at constant prices; Base: 1999-2000).
(2) EXGD: Exports of Goods (Merchandise exports on BoP basis)
(3) EXGS: Exports of Goods and Services (clubbing of Merchandise exports and Non-factor services receipts, both on BoP basis).
(4) GDCF: Gross Domestic Capital Formation at constant prices (Base: 1999-2000).
(5) REER: Real Effective Exchange Rate (Index).
(6) WGDP: World GDP (in value).
All the above series are subjected to logarithmic transformations.
The prefix ‘L’ stands for the natural logarithm of the respective time
series, ‘R’ stands for the residuals of the respective regression, and ‘D’ denotes differencing of the respective time series. It is appropriate
to mention that, all econometric exercises are carried out by using
EViews.
III.2 Tests of Unit Root and Stationarity
Before proceeding to the test the ELG hypothesis, it is appropriate
that all the series be tested for stationarity or the ‘same statistical
property’ - means the series have to be differenced or de-trended by
the same number of times to render them stationary. The traditional
approach of first differencing disregards potentially important
equilibrium relationships among the levels of the series to which the
hypotheses of economic theory usually apply (Engle and Granger,
1987).
The early and pioneering work on testing for a unit root in time
series was done by Dickey and Fuller (Fuller, 1976; Dickey and Fuller,
1979). The basic objective of the test is to examine the null hypothesis
![1](http://rbi.org.in/scripts/images/2EEG050511_1.gif) |
![2](http://rbi.org.in/scripts/images/2EEG050511_2.gif) |
The null hypothesis of a unit root is rejected in favour of the
stationary alternative in each case if the test statistics is more negative
than the critical value. Accordingly, Time series univariate properties
were examined using two unit root tests: augmented Dickey and Fuller
(1979) and Phillip and Perron (1988) tests. The PP tests are similar to
ADF tests, but they incorporate an automatic correction to the DF
procedure to allow for autocorrelated residuals. The tests often give the same conclusions as, and suffer from most of the same important
limitations as, the ADF tests.
Table 1 summarises the results for unit root tests on levels and in
first differences (at ‘maximum lags 2’ with ‘trend and intercept’
included in the test equation) of the data. For the ADF tests, the lag
length is based on the Akaike Information Criterion (AIC), while for
the PP test bandwidth selection is based on Newey-West. It is evidenced
from the test statistics that all the time series are I(1). Under the
classical hypothesis testing framework, the null hypothesis is never accepted, it is simply stated that it is either ‘rejected’ or ‘not rejected’.
This means that a failure to reject the null hypothesis could occur either
because the null was correct, or because there is insufficient information
in the sample to enable rejection (Brooks, 2008).
Table 1 : Unit Root Tests (1970-71 to 2009-10) |
Series |
Type |
Test-Statistics |
T-critical at 1% |
T-critical at 5% |
T-critical at 10% |
Result |
LRGDP |
ADF |
-0.8588 |
-4.2268 |
-3.5366 |
-3.2003 |
Don’t Reject Null Hypothesis |
PP |
-1.3544 |
-4.2191 |
-3.5331 |
-3.1983 |
Don’t Reject Null Hypothesis |
D(LRGDP,1) |
ADF |
-7.6665 |
-4.2268 |
-3.5366 |
-3.2003 |
Reject Null Hypothesis |
PP |
-9.0778 |
-4.2268 |
-3.5366 |
-3.2003 |
Reject Null Hypothesis |
LEXGD |
ADF |
-2.3257 |
-4.2268 |
-3.5366 |
-3.2003 |
Don’t Reject Null Hypothesis |
PP |
-1.656 |
-4.2191 |
-3.5331 |
-3.1983 |
Don’t Reject Null Hypothesis |
D(LEXGD,1) |
ADF |
-3.7838 |
-4.2268 |
-3.5366 |
-3.2003 |
Reject Null Hypothesis |
PP |
-3.7838 |
-4.2268 |
-3.5366 |
-3.2003 |
Reject Null Hypothesis |
LEXGS |
ADF |
-2.1216 |
-4.2268 |
-3.5366 |
-3.2003 |
Don’t Reject Null Hypothesis |
PP |
-1.2826 |
-4.2191 |
-3.5331 |
-3.1983 |
Don’t Reject Null Hypothesis |
D(LEXGS,1) |
ADF |
-3.3545 |
-4.2268 |
-3.5366 |
-3.2003 |
Reject Null Hypothesis |
PP |
-3.3991 |
-4.2268 |
-3.5366 |
-3.2003 |
Reject Null Hypothesis |
LGDCF |
ADF |
-1.241 |
-4.2191 |
-3.5331 |
-3.1983 |
Don’t Reject Null Hypothesis |
PP |
-1.1021 |
-4.2191 |
-3.5331 |
-3.1983 |
Don’t Reject Null Hypothesis |
D(LGDCF,1) |
ADF |
-7.1698 |
-4.2268 |
-3.5366 |
-3.2003 |
Reject Null Hypothesis |
PP |
-8.9766 |
-4.2268 |
-3.5366 |
-3.2003 |
Reject Null Hypothesis |
LREER |
ADF |
-1.0933 |
-4.2191 |
-3.5331 |
-3.1983 |
Don’t Reject Null Hypothesis |
PP |
-1.2592 |
-4.2191 |
-3.5331 |
-3.1983 |
Don’t Reject Null Hypothesis |
D(LREER,1) |
ADF |
-5.302 |
-4.2268 |
-3.5366 |
-3.2003 |
Reject Null Hypothesis |
PP |
-5.302 |
-4.2268 |
-3.5366 |
-3.2003 |
Reject Null Hypothesis |
LWGDP |
ADF |
-1.0834 |
-4.2191 |
-3.5331 |
-3.1983 |
Don’t Reject Null Hypothesis |
PP |
-1.3023 |
-4.2191 |
-3.5331 |
-3.1983 |
Don’t Reject Null Hypothesis |
D(LWGDP,1) |
ADF |
-5.2068 |
-4.2268 |
-3.5366 |
-3.2003 |
Reject Null Hypothesis |
PP |
-5.2068 |
-4.2268 |
-3.5366 |
-3.2003 |
Reject Null Hypothesis |
The most important criticism that has been levelled at unit root
tests is that their power is low if the process is stationary but with a
root close to the non-stationary boundary. Stationarity tests have
stationarity under the null hypothesis, thus reversing the null and
alternatives under the Dickey-Fuller approach. Thus under stationary
tests, the data will appear stationary by default if there is little
information in the sample. One such stationarity test proposed by
Kwaitkowski, Phillips, Schmidt, and Shin (1992), in short, the KPSS
test on the levels series presented in Table 2. We have now observed
that the test statistics exceeds the critical value even at 1% level, so
that the null hypothesis of a stationary series is strongly rejected.
The results of these tests can be compared with the ADF/PP procedure
to see if the same conclusion is obtained. The joint use of unit root
tests and stationarity is known as confirmatory data analysis. The
null and alternative hypotheses under each testing approach are as
follows:
![3](http://rbi.org.in/scripts/images/2EEG050511_3.gif) |
There are four possible outcomes: |
(1) Reject H0 |
and |
Do not reject H0 |
(2) Do not Reject H0 |
and |
Reject H0 |
(3) Reject H0 |
and |
Reject H0 |
(4) Do not Reject H0 |
and |
Do not Reject H0 |
For the conclusion to be robust, the results should fall under
outcomes (1) or (2) above.
Table 2: KPSS Stationarity Tests (1970-71 to 2009-10) |
Series |
Test-
statistics |
T-critical
at 1% |
T-critical
at 5% |
T-critical
at 10% |
Result |
LRGDP |
0.7595 |
0.7390 |
0.4630 |
0.3470 |
Reject Null Hypothesis |
LEXGD |
0.7701 |
0.7390 |
0.4630 |
0.3470 |
Reject Null Hypothesis |
LEXGS |
0.7660 |
0.7390 |
0.4630 |
0.3470 |
Reject Null Hypothesis |
LGDCF |
0.7563 |
0.7390 |
0.4630 |
0.3470 |
Reject Null Hypothesis |
LREER |
0.6908 |
0.7390 |
0.4630 |
0.3470 |
Reject Null Hypothesis |
LWGDP |
0.6752 |
0.7390 |
0.4630 |
0.3470 |
Reject Null Hypothesis |
By conducting tests under both types of the null hypotheses, the
results are much more robust than if just one of the tests is used,
provided of course that the results of the two tests are compatible. In
all cases, both the tests confirm the same conclusion – all the variables
under examination are having property I(1). The results of the unit
root tests performed corroborate previous findings in the empirical
literature, i.e. as with most macroeconomic series, the variables under
consideration in this study appear to be non-stationary and trended in
levels. Only their first differences are stationary.
Consequently, the next section of the empirical study investigates
whether the series under scrutiny are cointegrated, so that a welldefi
ned linear relationship exists among them in the long run. Thus,
we proceed to test for cointegration between the variables on levels
using several tests, all of which are based on the ‘null hypothesis of
no cointegration’.
III.3 Tests of Cointegration (Engle-Granger Approach)
In most cases, if two variables that are I(1) are linearly combined,
then the combination will also be I(1). Most generally, if variables with
differing orders of integration are combined, the combination will have
an order of integration equal to the largest. This linear combination of
I(1) variables will itself be I(1), but it would obviously be desirable to
obtain residuals that are I(0), so that the variables are cointegrated.
According to Engle and Granger (1987), a set of variables is
defined as cointegrated, if a linear combination of them is stationary. Many time series are non-stationary but ‘move together’ over time –
that is, there exist some influences on the series (for example, market
forces), which imply that the two series are bound by some relationship
in the long run. A cointegrating relationship may also be seen as a long
term or equilibrium phenomenon, since it is possible that cointegrating
variables may deviate from their relationship in the short run, but their
association would return in the long run. An interesting question is:
whether a potentially cointegrating regression should be estimated
using the levels of the variables or the logarithms of the levels of the
variables. Hendry and Juselius (2000) noted that, if a set of series is
cointegrated in levels, they will also be cointegrated in log levels. It is
common to run a regression of the log of the series rather than on the
levels; the main reason for using logarithms is that differences of the
logs are growth rates, whereas this is not true for the levels.
In the main case under scrutiny (Screenshots 1A and 1B): the
ELG hypothesis represented by cointegration sub-tests are able to
find evidence in favour of long run relationship between real GDP
and exports independently of other variables in case of the Indian economy. When variables are cointegrated, the OLS estimates from
the cointegrating regression will be super consistent, implying that
the estimates approach their true parameters at a faster rate than if
the variables were stationary and not cointegrated (Gujarati, 2003).
The presence of a cointegrating relationship forms the basis of error
correction specification. One can treat error term as equilibrium
error.
Screenshot 1A: ADF Test Result |
“Null Hypothesis: R-LRGDP-LEXGD has a unit root”
Exogenous: Constant
Lag Length: 3 (Automatic based on AIC, MAXLAG=3) |
|
t-Statistic |
Prob.* |
Augmented Dickey-Fuller test statistic |
-2.8336 |
0.0639 |
Test critical values: |
1% level |
-3.6329 |
|
5% level |
-2.9484 |
|
10% level |
-2.6129 |
|
*MacKinnon (1996) one-sided p-values. |
Screenshot 1B: ADF Test Result |
“Null Hypothesis: R-LRGDP-LEXGS has a unit root”
Exogenous: Constant
Lag Length: 3 (Automatic based on AIC, MAXLAG=3) |
|
t-Statistic |
Prob.* |
Augmented Dickey-Fuller test statistic |
-2.6676 |
0.0890 |
Test critical values: |
1% level |
-3.6156 |
|
5% level |
-2.9412 |
|
10% level |
-2.6091 |
|
*MacKinnon (1996) one-sided p-values. |
III.4 Equilibrium Correction or Error Correction Model
The error correction mechanism (ECM) was first used by Sargan
(1984) and later popularised by Engle and Granger (1987). An
important theorem known as Granger Representation Theorem states
that if two variables Y and X are cointegrated, then the relationship
between the two can be expressed as ECM. The error correction model
takes the following form of equation:
In both cases (Screenshots 2A and 2B), the coefficients of the
error correction term have the desired sign (negative). About 17 per
cent of disequilibrium is corrected every year in case of cointegration
between ‘exports of goods’ and GDP and about 14 per cent
disequilibrium corrected every year in case of ‘exports of goods and
services’ and GDP. The significance of the error correction term at 5%
level confirms that exports and GDP are cointegrated in the long run
and error correction takes place in the short run.
Screenshot 2A: Result of Error Correction Model |
Dependent Variable: DLRGDP
Sample (adjusted): 1971-72 to 2009-10
No.of observations: 39 after adjustments |
Variable |
Coefficient |
t-Statistic |
Prob. |
DLEXGD |
0.3155 |
7.1211 |
0.0000 |
R-LRGDP-LEXGD(-1) |
-0.1739 |
-2.6035 |
0.0133 |
R-squared: -0.6237
S.E. of regression: 0.0391
Durbin-Watson stat: 1.8183 |
Screenshot 2B: Result of Error Correction Model |
Dependent Variable: DLRGDP
Sample (adjusted): 1971-72 to 2009-10
No. of observations: 39 after adjustments |
Variable |
Coefficient |
t-Statistic |
Prob. |
DLEXGS |
0.3096 |
7.6026 |
0.0000 |
R-LRGDP-LEXGS(-1) |
-0.1443 |
-2.3524 |
0.0242 |
R-squared: -0.505251
S.E. of regression: 0.037637
Durbin-Watson stat: 1.797727 |
One of the major drawbacks of Engle-Granger approach is that it
can estimate only up to one cointegrating relationship between the
variables. But in other situations, if there are more variables, there
could potentially be more than one linearly independent cointegrating
relationship. Thus it is appropriate to examine the issue of cointegration
within the Johansen’s VAR framework.
III.5 Johansen Cointegrating Systems based on VAR
The Johansen procedure is a multiple equation method that
permits the identification of the cointegration space, which enables
the testing of how many cointegration relationships exist. LRGDP,
LEXGS, LGDCF, LREER and LWGDP are tested under Johansen’s
technique and results displayed in Screenshots 3A and 3B.
The trace test in Screenshot 3B indicates that the test statistics of
124.02 considerably exceeds the critical value 69.82 and so the null of no cointegrating vectors is rejected. This continues, until we do not
reject the null hypothesis of at most 2 cointegrating vectors at the 5%
level. The max test also confirms this result.
Screenshot 3A: Johansen Cointegration Test Result |
Sample (adjusted): 1974-75 to 2009-10
No. of observations: 35 after adjustments
Trend assumption: Linear deterministic trend
Series: LRGDP LEXGD LGDCF LREER LWGDP
Lags interval (in first differences): 1 to 3 |
Unrestricted Cointegration Rank Test (Trace) |
Hypothesized
No. of CE(s) |
Eigenvalue |
Trace Statistic |
0.05 Critical
Value |
Prob.** |
None * |
0.748950 |
107.3184 |
69.81889 |
0.0000 |
At most 1 * |
0.497694 |
58.94472 |
47.85613 |
0.0033 |
At most 2 * |
0.433178 |
34.84559 |
29.79707 |
0.0120 |
At most 3 |
0.331463 |
14.97573 |
15.49471 |
0.0597 |
At most 4 |
0.024900 |
0.882533 |
3.841466 |
0.3475 |
Trace test indicates 3 cointegratingeqn(s) at the 0.05 level
* denotes rejection of the hypothesis at the 0.05 level
**MacKinnon-Haug-Michelis (1999) p-values |
Unrestricted Cointegration Rank Test (Maximum Eigenvalue) |
Hypothesized
No. of CE(s) |
Eigenvalue |
Max-Eigen
Statistic |
0.05 Critical
Value |
Prob.** |
None * |
0.748950 |
48.37365 |
33.87687 |
0.0005 |
At most 1* |
0.497694 |
24.09913 |
27.58434 |
0.1313 |
At most 2* |
0.433178 |
19.86985 |
21.13162 |
0.0743 |
At most 3 |
0.331463 |
14.09320 |
14.26460 |
0.0532 |
At most 4 |
0.024900 |
0.882533 |
3.841466 |
0.3475 |
Max-eigenvalue test indicates 3 cointegratingeqn(s) at the 0.05 level
* denotes rejection of the hypothesis at the 0.05 level
**MacKinnon-Haug-Michelis (1999) p-values |
Suppose, we want to test the hypothesis that the LREER and
LWGDP do not appear in the cointegrating equation. We could test
this by specifying the restriction that their parameters are zero. In this
case there are two restrictions, so that the test statistics follows a
Chi-square distribution with 2 degrees of freedom. The p-value for
the test is 0.0004, so the restrictions are not supported by the data and we could conclude that the cointegrating relationship must also
include the LREER and LWGDP (Screenshots 4A and 4B).
Screenshot 3B: Johansen Cointegration Test Result |
Sample (adjusted): 1974-75 to 2008-09
No. of observations: 35 after adjustments
Trend assumption: Linear deterministic trend
Series: LRGDPLEXGS LGDCF LREER LWGDP
Lags interval (in first differences): 1 to 3 |
Unrestricted Cointegration Rank Test (Trace) |
Hypothesized
No. of CE(s) |
Eigenvalue |
Trace Statistic |
0.05 Critical
Value |
Prob.** |
None * |
0.8019 |
124.0243 |
69.8189 |
0.0000 |
At most 1 * |
0.5780 |
67.3487 |
47.8561 |
0.0003 |
At most 2 * |
0.4833 |
37.1484 |
29.7971 |
0.0059 |
At most 3 |
0.3028 |
14.0348 |
15.4947 |
0.0820 |
At most 4 |
0.0393 |
1.40617 |
3.8415 |
0.2357 |
Trace test indicates 3 cointegratingeqn(s) at the 0.05 level
* denotes rejection of the hypothesis at the 0.05 level
**MacKinnon-Haug-Michelis (1999) p-values |
Unrestricted Cointegration Rank Test (Maximum Eigenvalue) |
Hypothesized
No. of CE(s) |
Eigenvalue |
Max-Eigen
Statistic |
0.05 Critical
Value |
Prob.** |
None * |
0.8019 |
56.6756 |
33.8769 |
0.0000 |
At most 1 * |
0.5780 |
30.2003 |
27.5843 |
0.0225 |
At most 2 * |
0.4833 |
23.1136 |
21.1316 |
0.0260 |
At most 3 |
0.3029 |
12.6287 |
14.2646 |
0.0892 |
At most 4 |
0.0394 |
1.4062 |
3.8415 |
0.2357 |
Max-eigenvalue test indicates 3 cointegratingeqn(s) at the 0.05 level
* denotes rejection of the hypothesis at the 0.05 level
**MacKinnon-Haug-Michelis (1999) p-values |
The result thus demonstrate that the considered variables are
cointegrated in that there is a long-run equilibrium relationship among
them (these series cannot move too far away from each other or they
cannot move independently of each other). The fact that the variables
are cointegrated implies that there is some adjustment process in the
short run, preventing the errors in the long run relationship from
becoming larger and larger.
Screenshot 4A: Vector Error Correction Estimates |
Sample (adjusted): 1974-75 to 2009-10
No. of observations: 35 after adjustments
Cointegration Restrictions: B(1,4)=0, B(1,5)=0
Convergence achieved after 10 iterations.
Not all cointegrating vectors are identified
LR test for binding restrictions (rank = 1):
Chi-square(2): 16.80826
Probability: 0.000224 |
CointegratingEq: |
CointEq1 |
LRGDP(-1) |
2.293244 |
LEXGD(-1) |
6.076050 |
LGDCF(-1) |
-12.08954 |
LREER(-1) |
0.000000 |
LWGDP(-1) |
0.000000 |
C |
59.60599 |
Screenshot 4B: Vector Error Correction Estimates |
Sample (adjusted): 1974-75 to 2009-10
No. of observations: 36 after adjustments
Cointegration Restrictions: B(1,4)=0, B(1,5)=0
Convergence achieved after 10 iterations.
Not all cointegrating vectors are identified
LR test for binding restrictions (rank = 1):
Chi-square(2): 15.55530
Probability: 0.000419 |
CointegratingEq: |
CointEq1 |
LRGDP(-1) |
5.992756 |
LEXGS(-1) |
5.299681 |
LGDCF(-1) |
-14.34167 |
LREER(-1) |
0.000000 |
LWGDP(-1) |
0.000000 |
C |
42.55284 |
III.6 Granger Causality Test: Empirical Finding
The Null Hypothesis (Ho) in each case is: the variable under
consideration does not Granger cause the other variable.
The result in Tables 3A and 3B suggests that the direction of
causality is from export growth to GDP growth; since the estimatedF-statistics is significant, at the 5% level up to 4 lags, at the 10% level
at lag 5. On the other hand, there is no “reverse causation” from GDP
growth to export growth, since the F-statistics is statistically
insignificant. It can be assessed that, at lag 6, there is no statistically
discernible relationship between the two variables. This reinforces the
point made earlier that the outcome of the Granger causality test is
sensitive to the number of lags introduced in the model. In the next Table, we have presented the Granger causality between GDP and
Exports of Goods and services. This indicates that one can use exports
to better predict the GDP than simply by the past history of GDP.
Table 3A : Granger Causality between DLRGDP and DLEXGD |
Direction of Causality |
No. of Lags |
F-Statistic |
Probability |
Decision Regarding Ho |
Exports→GDP |
1 |
6.95666 |
0.01250 |
Rejected |
GDP→Exports |
1 |
0.69292 |
0.41098 |
Not Rejected |
Exports→GDP |
2 |
3.62001 |
0.03864 |
Rejected |
GDP→Exports |
2 |
1.69715 |
0.19979 |
Not Rejected |
Exports→GDP |
3 |
3.34858 |
0.03308 |
Rejected |
GDP→Exports |
3 |
1.80044 |
0.17001 |
Not Rejected |
Exports→GDP |
4 |
3.33842 |
0.02542 |
Rejected |
GDP→Exports |
4 |
0.88408 |
0.48770 |
Not Rejected |
Exports→GDP |
5 |
2.39229 |
0.07073 |
Rejected |
GDP→Exports |
5 |
0.81603 |
0.55113 |
Not Rejected |
Exports→GDP |
6 |
1.87782 |
0.13730 |
Not Rejected |
GDP→Exports |
6 |
1.07856 |
0.40961 |
Not Rejected |
Note: Variables are in Δlogs.
|
Table 3B: Causality between DLRGDP and DLEXGS |
Direction of Causality |
No. of Lags |
F-Statistic |
Probability |
Decision Regarding Ho |
Exports→GDP |
1 |
8.58354 |
0.00602 |
Rejected |
GDP→Exports |
1 |
0.10059 |
0.75306 |
Not Rejected |
Exports→GDP |
2 |
5.14572 |
0.01176 |
Rejected |
GDP→Exports |
2 |
0.63338 |
0.53753 |
Not Rejected |
Exports→GDP |
3 |
4.06956 |
0.01614 |
Rejected |
GDP→Exports |
3 |
0.70741 |
0.55568 |
Not Rejected |
Exports→GDP |
4 |
4.13654 |
0.01045 |
Rejected |
GDP→Exports |
4 |
0.32326 |
0.85968 |
Not Rejected |
Exports→GDP |
5 |
3.31053 |
0.02225 |
Rejected |
GDP→Exports |
5 |
0.40970 |
0.83686 |
Not Rejected |
Exports→GDP |
6 |
2.75318 |
0.04251 |
Rejected |
GDP→Exports |
6 |
0.78254 |
0.59400 |
Not Rejected |
Note: Variables are in Δlogs. |
III.7 Block Exogeneity/Granger Causality in VAR: Empirical
Estimates
The first step in the construction of any VAR model, once the
variables that will enter the VAR have been decided, will be to
determine the appropriate lag length. This can be achieved in a variety
of ways, but one of the easiest is to employ a multivariate information
criterion (Screenshot 5). EViews presents the values of various
information criteria and other methods for determining the lag order.
In this case, the Schwartz criteria select a zero order as optimal, while
Akaike’s and Hannan-Quinn criterion chooses VAR(5).
Following the lag order selection criteria, I have tested Granger
causality/Block Exogeneity in VAR framework. The result indicates
lead-lag relationship between exports and GDP and Granger causality
is significant at 5% level from exports of Goods and Services to GDP; ‘significant at 10% from exports to GDCF’ but no causality in the opposite direction (Screenshot 6). The result can be interpreted as
movements in the exports of goods and services appear to lead that of
GDP in case of Indian economy.
Screenshot 5: VAR Lag Order Selection Criteria |
Endogenous variables: DLRGDP, DLEXGS, DLGDCF, DLREER, DLWGDP
Exogenous variables: Constant
Sample: 1970-71 to 2009-10
Included observations: 34 |
Lag |
LogL |
LR |
FPE |
AIC |
SC |
HQ |
0 |
232.4803 |
NA |
7.08e-13 |
-13.78668 |
-13.55994* |
-13.71039 |
1 |
268.5359 |
59.00008* |
3.70e-13 |
-14.45672 |
-13.09626 |
-13.99897 |
2 |
289.1497 |
27.48505 |
5.40e-13 |
-14.19089 |
-11.69671 |
-13.35167 |
3 |
311.7413 |
23.27616 |
8.53e-13 |
-14.04492 |
-10.41703 |
-12.82425 |
4 |
361.5677 |
36.23744 |
3.86e-13 |
-15.54956 |
-10.78794 |
-13.94742 |
5 |
421.2525 |
25.32083 |
2.39e-13* |
-17.65167* |
-11.75634 |
-15.66807* |
* indicates lag order selected by the criterion
LR: sequential modified LR test statistic (each test at 5% level)
FPE: Final prediction error
AIC: Akaike information criterion
SC: Schwarz information criterion
HQ: Hannan-Quinn information criterion |
Screenshot 6: VAR Granger Causality/Block Exogeneity Wald Tests |
Sample period: 1970-71 to 2009-10
Included observations: 37 |
Dependent variable: DLRGDP |
Excluded |
Chi-sq |
df |
Prob. |
DLEXGS |
6.571840 |
2 |
0.0374 |
DLGDCF |
1.930558 |
2 |
0.3809 |
DLREER |
1.145787 |
2 |
0.5639 |
DLWGDP |
0.570733 |
2 |
0.7517 |
All |
13.05493 |
8 |
0.1100 |
Dependent variable: DLEXGS |
DLRGDP |
4.335449 |
2 |
0.1144 |
DLGDCF |
1.992873 |
2 |
0.3692 |
DLREER |
0.243723 |
2 |
0.8853 |
DLWGDP |
5.318795 |
2 |
0.0700 |
All |
10.36116 |
8 |
0.2406 |
Dependent variable: DLGDCF |
DLRGDP |
4.388943 |
2 |
0.1114 |
DLEXGS |
4.610782 |
2 |
0.0997 |
DLREER |
2.158529 |
2 |
0.3398 |
DLWGDP |
0.090009 |
2 |
0.9560 |
All |
9.672450 |
8 |
0.2888 |
Dependent variable: DLREER |
DLRGDP |
0.850660 |
2 |
0.6536 |
DLEXGS |
0.993505 |
2 |
0.6085 |
DLGDCF |
2.986425 |
2 |
0.2246 |
DLWGDP |
1.981283 |
2 |
0.3713 |
All |
6.374739 |
8 |
0.6053 |
Dependent variable: DLWGDP |
DLRGDP |
10.70434 |
2 |
0.0047 |
DLEXGS |
3.213572 |
2 |
0.2005 |
DLGDCF |
1.462160 |
2 |
0.4814 |
DLREER |
2.041921 |
2 |
0.3602 |
All |
25.79547 |
8 |
0.0011 |
IV. Concluding Observations
Application of stationarity/unit root tests, viz., ADF, PP and
KPSS, confirms that all the variables are non-stationary at log levels
and there is existence of unit root in the series used in the study. In
other words, all the macroeconomic variables used in this study are
I(1) in log levels and become stationary after first differencing their
log levels. Subsequent residual-based cointegration test on log levels
between exports and GDP confirms their long run relationship. This
result sets the stage for application of error correction model in bivariate
as well as multivariate frameworks. The bivariate error
correction model indicates that, in short run, if the GDP is above its
equilibrium value, it will start falling in the next period to correct the
equilibrium error. The coefficient of error correction term decides
how quickly the equilibrium is restored. About 17 per cent of
disequilibrium is corrected every year in case of exports of goods and
GDP; and about 14 per cent disequilibrium is corrected every year in
case of ‘goods and services’ and GDP. The significance of the error
correction term at 5% level suggesting the robust relationship between
export growth and growth of real GDP. This reinforces the nexus
between export and GDP growth in both short and long run.
The test of cointegrating relationship among a set of chosen
variables in Johansen’s procedure: the trace test indicates the null of
no cointegrating vectors is rejected. This continues, until we do not
reject the null hypothesis of at most 2 cointegrating vectors at the 5%
level. The max test also confirms this result. In the subsequent
specification of restriction under Vector Error Correction Model
(VECM) in VAR, we dropped LREER and LWGDP to test the
hypothesis that these two variables do not appear in the cointegrating
equation. The p-value for the test is 0.0004 indicates that the
restrictions are not supported by the data and we could conclude that
the cointegrating relationship must also include the LREER and
LWGDP. The result thus demonstrate that the considered variables are
cointegrated in that there is a long-run equilibrium relationship among
them (these series cannot move too far away from each other or they
cannot move independently of each other). The fact that the variables are cointegrated implies that there is some adjustment process in the
short run, preventing the errors in the long run relationship from
becoming larger and larger.
The test of Granger causality suggests that the direction of
causality from export growth to GDP growth; since the estimated
F-statistics is significant, at the 5% level up to 4 lags, at the 10% level
at lag 5. On the other hand, there is no “reverse causation” from GDP
growth to export growth, since the F-statistics is statistically
insignificant. It can be assessed that, at lag 6, there is no statistically
discernible relationship between the two variables. This indicates that
one can use exports to better predict the GDP than simply by the past
history of GDP. Granger causality/Block Exogeneity in VAR
framework indicates lead-lag relationship between exports and GDP
and the result can be interpreted as movements in the exports of goods
and services appear to lead that of GDP in case of Indian economy.
The conclusion supporting the validity of the ELG hypothesis is
similar to results of Serletis (1992) in case of Canada and for other
industrial countries as in Marin (1992). However, the caveat is that,
import side of openness has not taken into consideration. Given the
recent success of software exports from India along with the focus
area approach to both merchandise and services exports including its
diversification, the finding is plausible and consistent with prior
expectation that increasing exports stimulate economic growth.
* Research Officer, Department of Economic and Policy Research, Reserve Bank
of India, Central Office, Mumbai. The author would like to thank Prof. Pushpa
Trivedi and an anonymous referee for their insightful comments that helped further
improvement. The views expressed in the article are author’s own. However, the
usual disclaimer applies.
1 International Trade Classification (Harmonised System) is an extended version of
the International Classification system called ‘Harmonised Commodity Description
and Coding System’ evolved by World Custom Organisation previously known as
Customs Cooperation Council.
References:
Afxentiou, P. C., and A. Serletis (1991), ‘Exports and GNP Causality in the
Industrial Countries: 1950-1985’, Kyklos, 44(2): 167-79.
Al-Yousif, Y. K. (1997), ‘Exports and Economic Growth: Some Empirical
Evidence from the Arab Gulf Countries’, Applied Economics, 29(6): 693-97.
Anwer, M. S., and R. K. Sampath (2001), ‘Exports and Economic Growth’,
Indian Economic Journal, 47 (3): 79-88.
Asafu-Adjaye, J., and D. Chakraborty (1999), ‘Export-led Growth and
Import Compression: Further Time Series Evidence from LDCs’, Australian
Economic Papers, 38:164-75.
Bahmani-Oskooee, M., and J. Alse (1993), ‘Export Growth and Economic
Growth: Application of Cointegration and Error Correction Modelling’,
Journal of Developing Areas, 27: 535-42.
Balassa, B. (1978), ‘Exports and Growth: Further Evidence’, Journal of
Development Economics, 5(2): 181-89.
Berg, H. Van den, and J. R. Schmidt (1994), ‘Foreign trade and economic
growth: Time series evidence from Latin America’, Journal of International
Trade and Economic Development, 3(3): 121-30.
Bhagwati, J. (1978), Anatomy and Consequences of Exchange Control
Regimes: Liberalisation Attempts and Consequences, Ballinger Pub. Co.,
Cambridge, MA.
Boltho, A. (1996), ‘Was Japanese Growth Export-led?’, Oxford Economic
Papers, 48(3): 415-32.
Brooks, Chris (2008), Introductory Econometrics for Finance, Second
Edition, Cambridge University Press, New York.
Buffie, E. (1992), ‘On the Condition for Export-led Growth’, Canadian
Journal of Economics, 25: 211-25.
Chandra, R. (2002), ‘Export Growth and Economic Growth: An Investigation
of Causality in India’, Indian Economic Journal, 49(3), 64-73.
Dhawan, U., and B. Biswal (1999), ‘Re-examining Export Led Growth
Hypothesis: a Multivariate Cointegration analysis for India’, Applied
Economics, 31: 525-30.
Dickey, D. A. and W. A. Fuller (1979), ‘Distribution of the Estimators for
Autoregressive Time Series with a Unit Root’, Journal of the American
Statistical Association, 74: 427-31.
Dodaro, S. (1993), ‘Comparative Advantage, Trade and Growth: Export-
Led Growth Revisited’, World Development, 19(9): 1153-65.
Edwards, S. (1998), ‘Openness, Productivity and Growth: What do we
Really Know?’, The Economic Journal, 108: 383–98.
Emery, R. F. (1967), ‘The relation of exports and economic growth’, Kyklos,
20(2): 470-86.
Engle, R. F., and C. W. J. Granger (1987), ‘Cointegration and Error
Correction: Representation, Estimation and Testing’, Econometrica, 55:
251-76.
Feder, G. (1982), ‘On Exports and Economic Growth’, Journal of
Development Economics, 5: 59-73.
Fuller, W. A. (1976), Introduction to Statistical Time Series, John Wiley &
Sons, New York.
Ghatak, S., and S. W. Price (1997), ‘Export Composition and Economic
Growth: Cointegration and Causality Evidence for India’, Weltwirtschaftliches
Archiv, 133(3): 538-53.
Granger, C. W. J. (1969), ‘Investigating Causal Relations by Econometric
Models and Cross-Spectral Methods’, Econometrica, 37: 424-38.
Gujarati, D. N. (2003), Basic Econometrics, Fourth Edition, McGraw-Hill,
New York.
Heller, P. S., and R. C. Porter (1978), ‘Exports and growth: An Empirical
Re-investigation’, Journal of Development Economics, 5(2): 191-93.
Hendry, D. F. and K. Juselius (2000), ‘Explaining Cointegration Analysis:
Part I’, Energy Journal, 21: 1-42.
Henriques, I., and P. Sadorsky (1996), ‘Export-led Growth or Growth-driven
Exports: The Canadian case’, Canadian Journal of Economics, 29(3): 541-55.
Islam, M. N. (1998), ‘Exports expansion and Economic growth: Testing for
Cointegration and Causality’, Applied Economics, 30(3): 415-25.
Johansen, S., and K. Juselius (1990), ‘Maximum likelihood and inference on
Cointegration: with applications to the Demand for Money’, Oxford Bulletin
of Economics and Statistics, 52: 169-210.
Jung, W. S., and P. J. Marshall (1985), ‘Exports, Growth and Causality in
Developing Countries’, Journal of Development Economics, 18: 1-12.
Kavoussi, R. M. (1984), ‘Export expansion and Economic growth: Further
empirical evidence’, Journal of Development Economics, 14: 241-50.
Kindleberger, C. P. (1962), Foreign Trade and the National Economy, Yale
University Press, New Haven, CT.
Kravis, I. B. (1970), ‘Trade as a Handmaiden of Growth: Similarities
between the nineteenth and twentieth centuries’, Economic Journal, 80:
850-72.
Krueger, A. O. (1978), Foreign Trade Regimes and Economic Development:
Liberalisation Attempts and Consequences, Ballinger Pub. Co., Cambridge,
MA.
Kugler, P. (1991), ‘Growth, exports and cointegration: An empirical
investigation’, Weltwirtschaffliches Archiv, 127(1): 73-82.
Kwaitkowski, D., P. C. B. Phillips, P. Schmidt, and Y. Shin (1992), ‘Testing
the Null Hypothesis of Stationarity against the alternative of a Unit Root’,
Journal of Econometrics, 54: 159-78.
Lussier, M. (1993), ‘Impacts of exports on economic performance: A
comparative study’, Journal of African Economics, 2(1): 106-27.
Mallick, S. K. (1996), Causality between exports and economic growth in
India: Evidence from Cointegration based Error-Correction Model, Indian
Journal of Economics, 76 (302), 307-320.
Marin, D. (1992), ‘Is the export-led growth hypothesis valid for industrialized
countries?’, Review of Economics and Statistics 74, 678-88.
Marjit, S., and A. Raychaudhuri (1997), India’s Exports - An Analytical
Study, Oxford University Press, New Delhi.
Meier, Gerald M. (1976), ‘Conditions of export-led development – A Note’,
in Leading Issues in Economic Development, ed., G. Meier, Oxford
University Press, New York.
Michaely, M. (1977), “Exports and Growth: An empirical investigation”,
Journal of Development Economics, 4: 49-53.
Moschos, D. (1989), ‘Export expansion, Growth and the level of Economic
Development: An empirical Analysis’, Journal of Development Economics,
30(1): 93-102.
Nidugala, G K. (2001), ‘Exports and Economic Growth in India: An
Empirical Investigation’, Indian Economic Journal, 47(3): 67-78.
Phillips, P. C. B., and P. Perron (1988), ‘Testing for a Unit Root in Time
Series Regression’, Biometrika, Vol. 75, pp. 335-46.
Ram, R. (1985), ‘Exports and Economic Growth: Some Additional
Evidence’, Economic Development and Cultural Change, 33(2): 415-25.
──── (1987), ‘Exports and economic growth in developing countries:
Evidence from Time-Series and Cross-Section Data’, Economic Development
and Cultural Change, 36(1): 51-63.
Rodrik, D. (1999), The New Global Economy and Developing Countries:
Making Openness Work, Overseas Development Council, Washington DC.
Serletis, A. (1992). “Export growth and Canadian economic development”,
Journal of Development Economics, 38, 1 (January): 133-45.
Sharma, A., and T. Panagiotidis (2004), An analysis of exports and growth
in India: Cointegration and causality evidence (1971-2001), ESRC Seminar
Series, St Antony’s College, Oxford University.
Sinha, D. (1996), ‘Openness and Economic Growth in India’, Indian Journal
of Economics, 77: 115-24.
Syron, R. F., and B. M. Walsh (1968), ‘The relation of exports and economic
growth: A note’, Kyklos, 21(3): 541-45.
Toda, H. Y., and P. C. B. Phillips (1993), ‘Vector autoregressions and
causality’, Econometrica, 61: 1367-93.
Tyler, W. G. (1981), ‘Growth and export expansion in developing countries:
Some empirical evidence’, Journal of Development Economics, 9(1): 121-30.
World Bank (1993), The East Asian Miracle: Economic Growth and Public
Policy, Oxford University Press, New York.
|