Exports, Imports and Economic Growth in Malaysia: Empirical Evidence Based on Multivariate Time Series*

Transcription

1 EXPORTS, [Asian Economic IMPORTS Journal 1998, 1999, AND Vol. ECONOMIC No. 4] 3] GROWTH IN MALAYSIA 389 Exports, Imports and Economic Growth in Malaysia: Empirical Evidence Based on Multivariate Time Series* Ahmad Zubaidi Baharumshah University Putra Malaysia Salim Rashid University of Illinois This paper examines the relationship between export growth and income growth by including imports in the system of equations using the Johansen (1988) procedure and vector-error correction (VEC) model. Real exports were disaggregated into manufacturing and agricultural exports. The results of the multivariate cointegration indicate the presence of a stationary long-run relationship between exports, imports and GDP. The estimated VEC models suggest economic growth is driven by exports. Test results also confirm that economic growth causes export growth for manufacturing exports. Indeed, we found a feedback causal relationship between exports and economic growth for both the manufacturing and agricultural exports. The empirical findings indicate that an important determinant of long-run growth in the fast growing Malaysian economy is imports of foreign technology. I. Introduction Malaysia pursued an agricultural-diversification and industrial-promotion programme based on import substitution in the 1960s. The major impact of the government policy was reflected in greater diversification of agricultural exports, with timber, palm oil and petroleum playing an increasingly important role. Previously, the economy had relied heavily on the tin and rubber industries which generated about 75% of export earnings. The government shifted to an export promotion strategy in the early 1970s following the experience of newly * The authors are grateful to Abul M. M. Masih and the anonymous referees for useful and insightful comments. However, all remaining errors are the responsibility of the authors. One author (Baharumshah) would like to acknowledge the financial support of the Malaysia Government IRPA Research Grant. For research conducted when he was visiting the University of Illinois during the academic year.

2 ASIAN ECONOMIC JOURNAL 390 industrialized economies (NIE) that showed developing economies could catch up with advanced countries by applying the export-led growth model. 1 The result of the policy change is reflected in the composition of the gross domestic product (GDP). The share of agricultural exports declined from 37.9% in 1960 to 15.9% in 1994 while manufacturing exports increased from 8.7% to 34.7% over the same period. The export structure of the manufacturing sector has also changed over the past decades. The ratio of high technology manufactured goods to total manufacturing exports has increased from 29.0% in 1975 to 69.2% in On the import side, the average annual growth rate of imports for is 15.7%. Malaysia s import dependence has risen significantly reflecting the openness of the domestic markets. Importation of capital and intermediate goods has been the critical factor for Malaysia s economic development. Malaysia has been successful in courting foreign investments and technology by providing attractive fiscal incentives in the early 1980s. It has experienced high growth rates averaging more than 8% since 1985/86, until it slowed down because of the collapse of commodity prices. Clearly, the Malaysian economy has operated as an open economy and is dependent on international trade for its development. Malaysia now joined the top 20 trading nations and has emerged as a major export competitor in ASEAN region. 3 The purpose of this paper is to provide an empirical test of the causal relationship between export growth and output growth. The focus is on Malaysia since it is often cited as an example of a fast-growing outward-oriented economy. The analysis employed in the present study differs from previous work in at least two important ways. First, we examine the time-series properties of the variable involved. In particular, this study employs the vector-error correction (VEC) model for establishing the dynamic relationship between export expansion and economic growth. As pointed out by Riezman et al. (1996), the relationship between exports and growth is essentially a dynamic one and the bivariate causality test may not provide a comprehensive picture on the export-growth relation. 4 Second, imports are included in the causal relationship to analyse the importance of this variable on Malaysia s economic growth. This is important 1. The term export-led growth is often used in studies of causality testing between exports and economic growth. The export-led growth hypothesis argued that growth in the export sector causes national income to grow as opposed to the counter argument that it is growth in national income that leads to growth in exports (Edwards, 1992; Rodrik, 1992). Thus, if the export promotion strategy causes economic growth then the evidence may be interpreted as supporting the export-led growth hypothesis. 2. High technology manufactured goods include electrical and electronic appliances, other machinery and transport equipment. 3. The World Bank (1987) has also classified Malaysia as a moderately outward oriented country in both the and periods. 4. Except for Kunst and Marin (1989), Afxentiou and Serletis (1992), Riezman et al. (1996) and Ghartey (1993), previous studies typically focus on the bivariate relationship between export and income. As noted by Fisher (1981), the VAR technique is useful as a convenient way of summarizing empirical regularities and perhaps suggesting predominant channels through which relations work. We consider the VAR methodology suitable for Malaysian case since the relevant structural

3 EXPORTS, IMPORTS AND ECONOMIC GROWTH IN MALAYSIA 391 since the manufacturing base of the country is built on export-oriented industries and imports may play a central role in explaining the economic performance. It can be argued that by providing needed intermediate goods, imports are an important determinant of economic performance (see for example Esfahani, 1991; Serletis, 1992; Riezman et al., 1996; Liu et al., 1997). Export promotion allows accumulation of foreign exchange that permits imports of high quality goods and services, which in turn expand the production possibilities. The rest of the paper is organized as follows. Section II briefly reviews the theoretical arguments and empirical evidence on the role of exports in economic growth in LDCs. In Section III we set out the methodology and data used in the analysis. Results based on the Johansen estimation and the VEC model are provided in Section IV. Finally, Section V contains the summary and our conclusions. II. The role of exports in economic growth: theory and evidence Many researchers have concluded that export growth has made substantial contributions to export-oriented economies (Balassa, 1978; Feder, 1983; Michaely, 1977; Kavoussi, 1984; Krueger, 1978). There are numerous arguments in the literature for the pursuit of an export-oriented development strategy. First, trade expansion will bring about enhanced productivity through greater economies of scale in the export sector. Second, increased competition encountered in the international markets will undoubtedly provide greater incentives for technological advancement and better management, the effects of which will spill over into the non-export sectors, and thus raise the over-all productivity of the economy. Third, increased export earnings will ease constraints on growth by enhancing the capacity to import essential goods, especially imports of intermediate and capital goods. In other words, export expansion promotes capital accumulation and, consequently, over-all economic growth. Fourth, exports may have a positive impact on productivity owing to better allocation of resources through specialization based on comparative advantage. Fifth, an export-oriented approach in a labour-surplus economy permits rapid expansion of employment and real wages. Sixth, Lucas links trade to learning by doing and pointing out that the learning-spillover technology is consistent with the strong relationship between rapid productivity and trade openness. He concludes, Thus a large volume of trade is essential to a learning-based growth episode Lucas (1993, p. 269). The above arguments have been recently supplemented by the literature on endogenous growth theory which emphasizes the role of increasing returns to relationships remain unclear. McMillin (1988) defended the use of VAR models by pointing out that the VAR models are useful in case of forecasting, analysing the cyclical behaviour of the economy, the generation of stylized facts about the behaviour of the elements of the system which can be compared with existing theories or can be used in formulating new theories, and testing of theories that generate Granger-causality implications.

4 ASIAN ECONOMIC JOURNAL 392 scale and the dynamic spill-over effects of the export sector s growth. Exports may increase long-run growth by allowing the economy to specialize in sectors with scale economies that arise from research and development, human capital accumulation, or learning-by-doing. In this context, increasing returns to scale are associated with the use of new technology and with the complementarities between human and physical capital. Both Grossman and Helpman (1991) and Edwards (1992) have emphasized the role of free trade in generating technological progress. They show that a higher degree of openness allows smaller countries to absorb technology developed in advanced nations at a faster rate and thus grow, in equilibrium, more rapidly than economies with a lower degree of openness. 5 There is also another view suggesting that the causality runs from economic growth to exports. As pointed by Kaldor (1967), economic (productivity) growth leads to lower unit costs, which facilitates exports. In addition, Ghartey (1993) argued that economic growth causes export growth if innovation and technical progress result in well-developed markets, which in turn improve export performance in the trade sector. Producers are likely to sell goods in international markets if domestic production increases faster than domestic demand. Thus, economic growth causes export growth. Combination of all the above arguments supports a third hypothesis that suggests a bidirectional (or feedback) causal relationship between export and economic growth. In this case, export growth causes economic growth, as explained above, and growth affects export performance through technical progress and spin-off effects. According to Bhagwati (1988), increased trade (for whatever reason) produces more income, and more income facilitates more trade the result being a virtuous circle. Numerous studies have reported that there can be a two-way causality relationship between export and economic growth. Examples of these studies included Helpman and Krugman (1985), Kunst and Marin (1989), Ghartey (1993), Bahmani-Oskooee and Alse (1993), Doraisami (1996) and Liu et al. (1997), among others. The debate on the role of exports in stimulating economic growth has generated considerable interest in the literature. Empirical studies supporting exportpromoting policies fall into two broad categories. The first consists of in-depth country case studies that critically evaluate the implications of export-promotion as opposed to import-substitution strategies for economic growth. Some examples of such studies include Bhagwati (1978) and Krueger (1978). For instance, Krueger found that the economic performance of Korea and Brazil improved substantially after these two countries switched from import-substitution policies to export-promotion policies in the 1960s. 5. A recent paper by Liu et al. (1997) examined the relation between exports (openness) and economic growth for China. The results suggest a feedback causal relationship economic growth and exports and imports. They conclude that, regardless of size of the economy, opening to the world market allows the achievement of export-led growth and growth-export linkage.

5 EXPORTS, IMPORTS AND ECONOMIC GROWTH IN MALAYSIA 393 The second category of studies supporting export-promotion policies includes numerous cross-country studies that examine the relationship between growth of exports and aggregate output across many LDCs and over time. Examples are Michaely (1977), Balassa (1978, 1985), Feder (1983), Tyler (1981) and Kavoussi (1984), among others. Briefly, these studies measured the contribution of exports to economic growth based on the theoretical arguments on the export-led growth hypothesis. For instance, the contribution of exports is measured by its impact on GNP and the ability to manage external shocks (Balassa, 1978, 1985); scale economies and externalities (Tyler, 1981); intersectoral resource allocations and externalities (Feder, 1983); total factor productiveness (Kavoussi, 1984). In general, these cross-sectional studies provide uniform support for the export-led growth hypothesis: Those LDCs with higher average export growth over extended periods of time tend to grow faster than others. More recent studies that have used longer time periods and recent advances in time-series econometrics in investigating the causal relationship between exports and economic growth have failed to provide unambiguous support for the exportpromotion development strategy. For example, Jung and Marshall (1985) analysed the lead and lag timing patterns between growth rate of real exports and growth rate of real output for 37 developing economies separately; in only four cases (Indonesia, Egypt, Costa Rica and Ecuador) did they find evidence to support the export-promotion hypothesis. Jung and Marshall found little support for the hypothesis even in those NIEs such as Korea and Taiwan that have experienced both rapid growth and export expansion. Dorado (1993) using data for more than 80 countries for period, concluded that the Granger causality test offers weak support for the notion of trade as an engine of growth and trade as a handmaiden of growth. Others, notably Bahmani-Oskooee and Alse (1993), Doraisami (1996), Ghatak et al. (1997) and Liu et al. (1997), Afxentiou and Serletis (1992), and Sengupta and Espana (1994), provide positive support for the hypothesis. For instance, Bahmani-Oskooee and Alse using the Engle-Granger two-step procedure (Engle and Granger 1987) showed that there is a stable long-run relationship between export and economic growth and a bi-directional relationship exist between them. 6 Recently, Doraisami (1996) examined the long-run relationship between the two aggregates for Malaysia and showed empirical support for bi-directional causality link between exports and economic growth. Ghatak et al. (1997) also tested the relationship between export expansion and economic growth using annual data for Malaysia. 7 The results provide support for the hypothesis: aggregate exports 6. In this study nine developing countries were analysed and they found that, except for Malaysia, exports and GDP were integrarted. The evidence of long-run relationship between the two aggregates holds for the other ASEAN countries, including Singapore, Thailand and Philippines. 7. Both Bahmani-Oskooee and Alse (1993) and Doraisami (1996) tested the export-led growth hypothesis using the Engle-Granger (1987) two-step procedure. Ghatak et al. (1997) employed the Johansen maximum likelihood procedure to test the hypothesis.

6 ASIAN ECONOMIC JOURNAL 394 Granger-cause real GDP and non-export GDP. They also showed that manufacturing exports rather than the traditional exports drive the relationship. Unlike the papers mentioned above, Afxentiou and Serletis (1992) and Riezman et al. (1996) include the role of imports in the analysis. Afxentiou and Serletis found that export growth causes GNP growth for the Canadian case. However, he found no evidence that import growth causes either export growth or income growth. Conditional on import growth, Riezman et al. found a causal ordering from export growth to income growth in 30 of the 126 countries; 25 have the reverse ordering. They showed that for the Asian Tigers of the Pacific Rim, the relationship between export growth and output growth becomes clearer when conditioned on human capital and investment growth as well as import growth. Thus, despite the extensive literature analysing the relationship between exports and growth, no clear conclusion has emerged from all these studies. We present an alternative approach to test for the existence of a stable long-run relationship between output, exports and imports, both at the aggregate level and for the agricultural and manufacturing sectors. III. Methodology and data It has become standard practice to begin the analysis by examining the timeseries properties of the data. The analysis of cointegration starts with the determination of the univariate properties of the time series. If the series do not follow the same order of integration, then there can be no meaningful relationship among them. In the present context, if the three series are integrated of the same order, we can proceed to the cointegration test. We utilize two asymptotically equivalent procedures for detecting unit roots in the data: the Augmented Dickey-Fuller (ADF) and the Phillip and Perron (PP) tests (see Said and Dickey, 1984; Phillips and Perron, 1988). There are two main approaches to testing for cointegration. They are the Engle and Granger (1987) two-step procedure and the Johansen (1988) and Johansen and Juselius (1990) procedure. 8 Cointegration tests in this paper are conducted by mean of the method developed by Johansen (1988) and Johansen and Juselius (1990) since this particular method is claimed to be superior to the regression-based Engle and Granger procedure. The Johansen-Juselius method sets out a maximum likelihood procedure for estimating and determining the presence of cointegrating vectors in VAR system. Suppose the vector of p- variables, Z t = (Z 1t,..., Z pt ), is generated by the k-order vector autoregressive process with Gaussian errors: 8. Apart from being multivariate, the Johansen procedure has several appealing features. First, it allows more than one cointegrating relation among the variables being examined. Second, there are concerns about the small-sample bias in estimates from Engle-Granger procedure. Finally, unlike the Johansen procedure, the Engle-Granger two-step procedure does not easily accommodate dynamics in the cointegration analysis.

7 EXPORTS, IMPORTS AND ECONOMIC GROWTH IN MALAYSIA 395 Z t = A 1 Z t A k Z t k +µ+ε t t = 1,..., T (1) where Z t is a p 1 vector of stochastic variables, ε 1,..., ε T are iid Np (0, Σ) and µ is a vector of constants. Since we want to distinguish between stationarity by linear combinations and by differencing this process may be written in error correction form as: Z t =Γ 1 Z t Γ k 1 Z t k+1 +ΠZ t k +µ+ε t, t = 1,..., T (2) The matrix Π contains information about the long-run relationship between the variables in the vector. Information about the number of cointegrating vectors is found in the rank of Π. In other words, the rank of Π determines how many linear combinations of Z t are stationary. If the p p matrix Π has rank zero, r = 0, then all elements of Z t are non-stationary. Thus there are no cointegrating relationship between the variables. If Π is of full rank, r = p, then all elements of Z t are stationary, Thus, any combination of the variables results in a stationary series, that is, cointegrated. In the intermediate case r < p, there are r nonzero cointegrating vectors among the elements of Z t and p r common stochastic trends. If a non-zero relationship is indicated by the test, a stationary long-run relationship is implied. In the case where 0 < r < p, Π can be factored as αβ (or Π = αβ ) where α and β are both p r matrices. The matrix α contains the adjustment parameters while the matrix β is called the cointegrating matrix and has the property that βz t I(0), where I(0) indicates integrated of order zero. Thus we can interpret the relations of βz t as the stationary relations among potentially non-stationary variable, that is, as cointegrating relations. Johansen (1988) and Johansen and Juselius (1990) developed a maximum likelihood estimation procedure for µ, Γ, α, β and Σ and also provide tests for the number of cointegrating vectors. The Johansen method suggests two statistics to check for cointegrating vectors; trace and maximum eigenvalue tests. Since the method is now widely used in the literature, a description of the specific details of the test is omitted here. The appropriate critical values for the tests are provided in Osterwald-Lenum (1992). Test for causality between two variables has been coined by Granger (1969) and popularized by Sims (1972). 9 Specifically, the test is utilized to determine whether prediction of the present value of y is enhanced by using past values of x, other information being identical. If it can, then, x is said to Granger-cause y (denoted as x y). In the present context, if movements in exports can help to predict economic growth then exports Granger-cause economic growth. In the literature on this subject, various tests for causality have been proposed. In this study, we employed the vector-error correction model and variance decompositions 9. Causality is a subject of great controversy among economist. Causality as a philosophical concept implies existence of a priori laws that govern the relation between the cause and the effect (Feigel, 1953). Thus, empirical tests are at best necessary but never sufficient in establishing causality. For an excellent discussion of the issue see Zellner (1988).

8 ASIAN ECONOMIC JOURNAL 396 in order to capture both within-sample and out-of-sample Granger causality among the economic variables. 10 According to Granger, if the variables are cointegrated, the finding of Granger non-causality is ruled out. However, it does not indicate the direction of causality between variables. The direction of Granger causality can only be detected through the vector-error correction model (VECM) derived from the long-run cointegrating vectors, which is now widely used in the literature (see for example Masih and Masih, 1996). Through the error-correction term, the ECM opens an additional channel for Granger-causality to merge, a channel ignored by the standard Granger and Sims tests. In addition to the direction of causality, the VECM also allows us to distinguish between short-run and long-run Granger-causality. The F-test of the explanatory variables (in first differences) indicates the short-run causal effect, while the long-run causal relationship is implied through the significance of the lagged error-correction term (ECT) which contains the long-run information. The data are quarterly observations on real gross domestic product (GDP), real total exports (TX), real manufacturing exports (MX), real agricultural exports (AX) and real imports (IM). Quarterly data on all these variables, except for GDP are available from 1970:1 to 1994:4 (T = 108). Monthly GDP figures were extrapolated from annual series employing the procedure suggested by Gandolfo (1981). For primary commodity exports, the commodities included are palm oil, rubber, saw log and sawn-timber. These series were obtained from various issues of the Quarterly Economic Bulletin published by Bank Negara. All series are in logarithm form and were deflated by consumer price index (CPI) which is also available on quarterly basis. IV. Empirical results The variables were tested for their order of integration using Augmented Dickey- Fuller (ADF) and non-parametric Phillips-Perron (PP) tests. The unit-root properties of the series are crucial for the cointegration and causality analysis. The variables utilized in the analysis were tested for their order of integration using Augmented Dickey-Fuller (ADF) and non-parametric Phillips-Perron (PP) tests. It is generally known that the results of these tests often depend on the number of lags included, therefore special attention must paid to the lag-length selection. In this study, the optimal lag for conducting ADF test was based on the Akaike s final prediction criteria using a range of lags. In addition, the Ljung-Box Q-statistics for serial correlation is utilized to determine whether sufficient lag is included so that the residuals are not autocorrelated. The truncation lag parameter 7 used for the PP tests was selected using a window choice of: w (s, 7) = 1 [s/(7 + 1)], where the order is the highest significant lag from either the autocorrelation or partial autocorrelation function of the first-difference series. 10. See Liu et al. (1997) Kunst and Marin (1989) and Jung and Marshall (1985), among others for the application of these tests in examining the exports-growth relationship.

9 EXPORTS, IMPORTS AND ECONOMIC GROWTH IN MALAYSIA 397 Table 1 Tests for Unit Root Variables ADF Statistic PP Statistic Constant Constant with Constant Constant with without trend trend without trend trend LGDP (0) (0) (3) (3) LTX (6) (6) (6) (6) LAX (8) (8) (1) (1) LMX (4) (4) (1) (1) LIM (9) (9) (1) (1) Notes: LGDP = log of gross domestic product; LTX = log of total exports; LAX = log of agricultural exports; LMX = log of manufacturing exports; and LIM = log of imports. The test statistics that include a constant and without linear trend should be compared to 3.34 (α =0.01) and 2.86 (α =0.05). The test statistics that include a constant and linear trend should be compared to 3.96 (α=0.01) and 3.14 (α =0.05). These values are provided by SHAZAM output based on MacKinnon (1991). The numbers in parentheses for the ADF (Augmented Dickey Fuller) tests represents the number of lags of the dependent variable included in the test regression to ensure white noise. The numbers in parentheses for the PP (Phillip and Perron) test represent the choice of truncation lag length used in the test. The results of unit-root tests are presented in Table 1. The unit-root test statistics reported in Table 1 are all insignificant for variables of log-level based on conventional significant level. The test results of the ADF and PP, including constant and trend indicate that the null hypothesis of a unit root cannot be rejected for the variables under investigation at the 5% significance level. Unit root tests were also applied on the first difference of the variables (not reported here) which indicated that it is possible to reject an I (2) specification at 5% significance level in all cases. Hence all the series under investigation could be characterized as I (1) process, that is, they are all integrated of order one. The results are consistent with the view that most macroeconomics variables are nonstationary. For the LTX, there is some evidence to suggest that the series may be stationary around a trend based on the PP test at 5% significance level. However, given the power of the unit root tests, we consider the series to be I (1) process. Next, we proceed with the Johansen-Juselius cointegration test to detect any long-run relationship among exports, imports and economic growth. The initial task in testing the cointegration relationship is to determine the optimum order of lag length. This is important since under-parametrization would tend to bias the results and over-parametrization would diminish the power of tests. In order to determine the lag length of the VAR model, information theoretic model selection criteria attributed to Schwarz (1978) (Schwarz information criteria) were considered. Based on this procedure, a VAR [12] specification is selected for total exports and a VAR [10] for the agricultural and manufacturing equations. In Table 2 the Ljung-Box Q-statistic (Q [4] ), Godfrey LM-statistic (LM [8]), Engle ARCH test (ARCH [4] ) and normality test (NORM) are reported. Both

10 ASIAN ECONOMIC JOURNAL 398 Table 2 Johansen-Juselius Multivariate Cointegration Tests H 0 Trace λ-max LM[5] Q[10] ARCH [4] Normality Total exports Manufacturing exports Agriculture exports P = ** 55.87** P * 24.09* P P = ** 42.94** P P P = ** 53.54** P P Notes: The p denotes the number of cointegrating vectors. The (*) indicates statistically significant from zero at the critical levels of 5% and (**) indicates statistically significant from zero at 1%. The critical values for the Johansen-Juselius test were obtained from Osterwald-Lenum (1992). Trace: 22.0 p = 0 λ-max: 34.9 p = p p p p 2 the Ljung Q-statistic and Godfrey LM statistics show no evidence of serial correlation, indicating the appropriateness of the lag structure employed in the VAR models. The Engle s LM rejects the existence of ARCH effect in the residuals of the model. Thus, the diagnostic testing of the residuals confirmed that the residuals of VAR models appear to be well behaved. In all the models, the cointegration tests are conducted with the inclusion of a deterministic trend. As shown in Table 2, both the trace and maximum eigenvalue test statistics indicate that at least one cointegrating vector is present in each of the three models based on the 5% significance level. In all cases, the null hypothesis r = 0 against r 1 is soundly rejected by both the tests, implying that the null hypothesis of no cointegration is rejected. The same conclusion arises when the null hypothesis r = 2 is tested for total exports. Hence, we conclude that there appear to be two stationary linear combinations among the three series. However, at r = 2 both tests are unable to reject the null at the disaggregated level. At r = 3 the trace and maximal eigenvalue tests are unable to reject the null in all three cases. Experimentation with longer lag length indicates that the number of cointegrating vectors do not change the qualitative results of the VAR model. The Johansen-Juselius procedure provides a convenient framework to test several hypotheses on the coefficients by imposing restrictions using the likelihood ratio tests. The question of whether all the variables and especially the import variable enter into the cointegration relationship is important in examining the export-economic growth relation. Likelihood ratio tests for zero restrictions on

11 EXPORTS, IMPORTS AND ECONOMIC GROWTH IN MALAYSIA 399 the elements of the cointegrating vector are performed. The null hypothesis that the import variable does not enter the cointegrating relationship is soundly rejected in all the three cases. 11 The results of the multivariate cointegrating regression conclusively indicate that GDP, exports and imports are cointegrated. The long-run relationship between exports and growth confirmed the findings of Doraisami (1996) and Ghatak et al. (1997). This conclusion does not match that of Bahmani-Oskooee and Alse (1993), using the Engle and Granger methodology. There are two important implications of the above results. First, cointegration between two or more variables is sufficient to establish the presence of causality relation in at least one direction, implying that one variable helps to forecast the other. Second, the error-correction term (ECT) has to be included in the VAR models to capture the extent to which the system is out of equilibrium in order to examine the causal relationships (see Granger, 1988). 12 These and many other related issues are discussed by King et al. (1991) and Toda and Phillips (1993, 1994) who support the use of VECMs in empirical tests of Granger causality. In order to understand the nature of causation, the next technique employed is the Granger causality test. The hypothesis that the variable, say x, is influenced by y is equivalent to the test that all of the coefficient on the lagged values of y included in the regression are jointly equal to zero. The test statistic used is an F-statistic and rejection of the null hypothesis suggests that the causation runs from y to x. The hypothesis that y is not influenced by x is tested in an analogous fashion. If the null hypothesis is rejected in both cases, it suggests that a feedback relationship exist between the two variables. To determine the short-run causality and the dynamics between these series, we relied on the vector-error correction (VEC) model. Test results derived from estimating the VEC models are reported in Table 3. The hypothesis that growth in exports do not Granger-cause growth in GDP is easily rejected at the 1% level of significance for both agricultural and manufacturing exports. The Granger causality from exports to growth of GDP is established through the significance of both the error-correction adjustment (ECT) coefficient and the significance of the F-statistic on the lagged independent variables terms for total, manufacturing and agricultural exports. Thus, the inclusion of past information on exports improves the forecast of growth of GDP in the short run as well as the long run, at least for the sample period considered in an analysis. The hypothesis that growth in output does not Granger-cause exports is also rejected in all the three models based on conventional significance level. Thus, 11. The tests are performed using MICROFIT 4.0 and the results are available from the authors upon request. 12. Sims et al. (1990) and Toda and Phillips (1993) note that if two series are cointegrated, then there would exist an error-correction representation and vice versa. This is known as the Granger Representation Theorem. Granger s results showed that if the error-correction term is not included in the VAR model when the series are cointegrated then the standard causality test is invalid. Test results based on the misspecified model tend to favour the rejection of the null of Granger causality.

12 ASIAN ECONOMIC JOURNAL 400 Table 3 Granger Causality Results based on Vector-Error Correction Model Total Exports ( p = 12) Independent variable Dependent GDP TX IM ECT(T-1) Variable F-statistics ( p-value) t-values GDP 3.051(0.003)** 0.951(0.515) 1.257(0.131) TX 1.956(0.049)* 1.431(0.179) 2.787(0.001)** IM 1.854(0.063) 1.421(0.188) 1.933(0.030)* Manufacturing Exports ( p = 10) Independent variable Dependent GDP MX IM ECT(T-1) Variable F-statistics ( p-value) t-values GDP 4.263(0.000)** 0.657(0.771) 1.976(0.053)* MX 1.054(0.415) 1.770(0.084) 3.031(0.004)** IM 4.550(0.000)** 2.380(0.016)* 3.084(0.003)** Agricultural Exports ( p = 10) Independent variable Dependent GDP AX IM ECT(T-1) Variable F-statistics ( p-value) t-values GDP 2.809(0.004)** 0.881(0.560) 2.072(0.043)* AX 3.094(0.003)** 2.214(0.028)* 3.486(0.001)** IM 0.904(0.543) 1.050(0.419) 2.286(0.026)* Notes: The F-statistic tests the joint significance of the lagged values of the independent variables, and t-statistic tests the significance of the error correction term (ECT). Figures in parentheses are the p-values. Diagnostic tests (not reported) conducted for the residuals show no evidence of serial correlation. our results tend to favour a two-way (or feedback) Granger causal relationship between growth rate of export and growth rate of output rather than unidirectional causality between the two variables. The implementations of agrarian reforms in the early 1960s and the industrial policies in 1970s have positive contributions to the Malaysian economic performance and export sector, both in the short-and long-run. Indeed, our results suggested both exports and economic growth reinforce each other, thus supporting the view expressed by Bhagwati (1988) and others, trade expansion increases income and more income facilitates more trade. Table 3 also demonstrates the importance of imports in the causal link. The Granger causality test rejects non-causality from exports to imports as well as from imports to exports. Therefore a feedback relation also exits between exports and imports for all categories of exports, at least in the long run. Unlike the

13 EXPORTS, IMPORTS AND ECONOMIC GROWTH IN MALAYSIA 401 agricultural exports, the causal link in the manufacturing equation is established through both the error-correction term as well as the lagged independent variables. The manufacturing sector in Malaysia heavily relied on imported intermediate goods and capital in the short and long run. On the other hand, imports are important in the agricultural economy in the long run. The causal relationship running from exports to imports can be justified since Malaysia relied heavily on exports to finance its import requirements, especially for manufactured goods. Thus, the inflows of FDI in the last two decades have not reduced the dependency of import growth on export earnings. In order to provide further insight into the dynamic relationships of the variables in the system, the forecast error-variance decomposition is calculated. 13 The variance decomposition shows the proportion of the forecast error of each endogenous variable that is accounted by each of the other variables. A similar procedure is adopted by Masih and Masih (1996), among others, to gauge the relative strength of the Granger-causality chain or degree of exogeneity amongst the variables. As pointed by Sims (1982) a variable that is optimally forecast from it own lagged values will have all its forecast variance account by its own disturbances. In other words, if both imports and exports are important to economic growth, the impact of the shock on this variable is significant on economic growth. A five-year horizon based on evidence for a world business cycle of roughly five years is chosen for the analysis (see for example, Riezman et al., 1996). The pattern of response in the VAR analysis has been reported as quite sensitive to alternative orderings. Changing the ordering had a negligible impact in the variance decompositions and therefore, the inference. The results of the variance decompositions of the forecast-error are shown in Table 4. Here only the chain ordering imports-exports-income using a 20-quarter forecasting horizon is reported. Several conclusions are derived from the various decomposition results. First, innovations in both categories of exports lead to income growth, which is consistent with the VECM results. Table 4 indicates that changes in exports accounts for 18% and 24% of the variation in forecast error of income in manufacturing and agricultural exports, respectively. Thus, the role of the agricultural economy on the performance of the Malaysia s economy cannot be underrated. The point estimate of the effect of innovation in imports on income is higher for the manufacturing sector, more than twice as large as the effect on agriculture, thus implying that imports are more crucial to the manufacturing sector. Both imports and exports account for over 40% of the forecast error variance of income. The significant impact for this variable on income is consistent with the view that exports expansion has contributed to Malaysia s economic growth. 13. A finding that exports Granger-cause income means that the variance in the one-step-ahead forecast error, made from predicting income linearly using its own past, is reduced when lags of exports are included. There is no priori reason to think that any causal relationship between exports and imports must necessarily become apparent in a year. The problem of limited time horizon in the Granger causality test is addressed with the variance decomposition method.

14 ASIAN ECONOMIC JOURNAL 402 Table 4 Variance Decomposition Total Exports Source of Variation (Effect upon) LIM LTX LGDP LIM LTX LGDP Manufacturing Exports Source of Variation (Effect upon) LIM LMX LGDP LIM LMX LGDP Agricultural Exports Source of Variation (Effect upon) LIM LAX LGDP LIM LAX LGDP Note: The variance decomposition shows that the percentage of the five-year forecast error in the variance in each row due to the variable at the top of the column. To conserve space, the results of causal chain ordering import-exports-income are only reported here. Second, changes in income account for about 15% of the forecast error of exports. For example in the total export equation, income accounts for 20% of the forecast error in exports. Again the empirical results are consistent with those of the VECM analysis reported earlier in that a feedback relation exists in the export-growth relation. V. Summary and policy conclusions Our analysis brings out several facts that need to be considered by theorists in developing models of economic growth. First, the empirical evidence suggests that export expansion strategy can contribute to a country s growth. Malaysia is an open economy and is dependent on international trade for a large part of its economic development. This is consistent with the widely held consensus that Malaysia should expand production beyond its borders to promote economic growth. Second, the role of the growth rate of imports cannot be ignored when examining the relationship between export growth and income growth. In the VEC model, we showed that a feedback relationship existed between exports and imports. For manufacturing, the causal link holds both in the short-run and

15 EXPORTS, IMPORTS AND ECONOMIC GROWTH IN MALAYSIA 403 long-run. But in the case of agricultural exports the link is established only through the error-correction channel. Can export-promotion strategies continue to accelerate growth in the next decades, especially in the face of worldwide regionalism and limitation of world markets? Support for export-promotion strategies is not universal. Klein (1990) argued that export-led growth could not be continued in the 1990s because of the shrinking market in North America and European countries. Krugman (1994) in his article entitled The Myth of Asia s Miracle argued that the empirical growth-accounting studies prove that both earlier Soviet and more recent East Asian growth have relied on factor accumulation to yield economic growth and that East Asia will suffer economic collapse of like the former Soviet Union. Others argued that ELG can be continued if the export structure is adjusted to produce new and high quality products. Expanding trade between LDCs and forming regional trade groups can also reduce the dependency on overly narrow traditional markets. It is important to point out that Malaysia s success in adopting outwardoriented development strategies in the last 20 years or so is based on stable macroeconomic environments. It is well known that policies do not work in isolation. Both categories of exports are expected to decline because of uncertainty over inflation and exchange rates following the recent financial crisis. However, the manufacturing sector will be more adversely affected than the agricultural sector in the short run with the appreciation of the Malaysian ringgit since the sector relied heavily on imports (capital goods) from developed economies. The future success of Malaysia s export strategies will depend on the ability to penetrate into new markets, increased labour productivity and the production of quality products through product innovation and product developments. Malaysia depended heavily on the US, Japan, Singapore and the EU for manufactured export markets. However, recent trends show that the share of manufacturing exports to emerging markets like Hong Kong, Taiwan, and Brazil have increased from 11% in 1990 to 26% in The manufacturing sector in Malaysia is labour intensive and the tight labour market has led to wage inflation. For instance, the wage rate has been increasing at the rate of 11% and 7% in 1994 and 1995 respectively. If the trend continues the export sector will lose its competitiveness to other ASEAN countries as well as to China. Of course all these issues merit further research. Finally, it is worth pointing out here that the results of this paper do no more than suggest that the adoption of outward orientation and export expansion policies can facilitate rapid economic growth. This cautionary note is important because there is the risk that too much emphasis on export expansion may cause governments to underrate the other important factors that influence growth. For instance, the dynamic impacts of exports on economic growth can be negated to a large extent if forward and backward linkages are weak because of inadequate infrastructure, or investors loss of confidence, or if macroeconomic policies are inept. In addition, as pointed out by Jacobs et al. (1979), evidence against the

16 ASIAN ECONOMIC JOURNAL 404 null hypothesis of no Granger causality between two variables does not necessarily mean that there is causal relation between the two variables, it simply means that one variable contains some information about the others. 14 The test does not say anything about how much information is contained in the variable. We believe that our research provides some information on the export-economicgrowth relationship, but the precise form may be intertwined with other economic forces. References Afxentiou, P. C. and A. Serletis, 1992, Openness in Canadian economy: Applied Economics, 24, pp Bahmani-Oskooee, M. and J. Alse, 1993, Export growth and economic growth: an application of cointegration and error-correction modelling. Journal of Developing Areas, 27, pp Balassa, B., 1978, Export and economic growth: further evidence. Journal of Development Economics, 5, pp Balassa, B., 1985, Exports, policy choices, and economic growth in developing economies after the 1973 oil shock. Journal of Development Economics, 18, pp Bhagwati, J., 1978, Foreign Trade Regimes and Economic Development: Anatomy and Consequences of Exchange Control Regimes. Ballinger Publishing Co., Cambridge, MA. Bhagwati, J., 1988, Export promotion trade strategy: issues and evidence. World Bank, Research Observer, 3, pp Dorado, S., 1993, Exports and growth: a reconsideration of causality. Journal of Developing Areas, 27, pp Doraisami, A., 1996, Export growth and economic growth, a re-examination of some time-series: evidence of the Malaysian experience. Journal of Developing Areas, 30, pp Edwards, S., 1992, Trade orientation, distortions and growth in developing countries. Journal of Development Economics, 39(1), pp Engle, R. F. and C. W. J. Granger, 1987, Cointegration and error correction representation, estimation and testing. Econometrica, 5, pp Esfahani, H. S., 1991, Exports, imports, and economic growth in semi-industrialised countries. Journal of Development Economics, 35, pp Feder, G., 1983, On exports and economic growth. Journal of Development Economics, 12, pp Feigel, H., 1953, Notes on causality. In H. Feigel and M. Brodbeck, Eds., Readings in the Philosophy of Science. Appleton Century-Crotts, New York, pp Fisher, S., 1981, Relative shocks, relative price variability and inflation. Brookings Papers on Economic Activity, 2, pp Gandolfo, G., 1981, Quantitative Analysis and Econometric Estimation of Continuous Time Dynamics. North-Holland Publishing Company, Amsterdam. Ghartey, E. E., 1993, Causal relationship between exports and economic growth: some empirical evidence in Taiwan, Japan and the US. Applied Economics, 25, pp Ghatak, S., C. Milner and U. Utkulu, 1997, Exports, export composition and growth: cointegration and causality evidence for Malaysia. Applied Economics, 29, pp Granger, C. W. J., 1969, Investigating causal relations by econometric models and cross spectral methods, Econometrica, 40, pp We thank the anonymous referee for bringing this point to our attention.

17 EXPORTS, IMPORTS AND ECONOMIC GROWTH IN MALAYSIA 405 Granger, C. W. J., 1988, Causality, cointegration, and control. Journal of Economic Dynamics and Control. 12, pp Grossman, G. and E. Helpman, 1991, Innovation and Growth in the Global Economy. MIT Press, Cambridge, MA. Helpman, E and P. Krugman, Market Structure and Foreign Trade. MIT Press, Cambridge, MA. Jacobs, R. L., Edward E. Leamer and Michael P. Ward, 1979, Difficulties with testing for causation. Economic Inquiry, 7, pp Johansen, S., 1988, Statistical analysis of cointegration vectors. Journal of Economic Dynamics and Control, 12, pp Johansen, S. and K. Juselius, 1990, Maximum likelihood estimation and inference on cointegration with applications to the demand for money. Oxford Bulletin of Economics and Statistics, 52, pp Jung, W. S. and P. J. Marshall, 1985, Export, growth and causality in developing countries. Journal of Developing Economics, 18, pp Kaldor, N., 1967, Strategic Factors in Economic Development. Cornell University, New York. Kavoussi, R. M., 1984, Export expansion and economic growth: further empirical evidence. Journal of Development Economics, 14, pp King, R. G., C. I. Plosser, J. H. Stock, amd M. W. Watson, 1991, Stochastic trends and economic fluctuations. American Economic Review, 81, pp Klein, L. R., 1990, Can export-led growth continue indefinitely? An Asia-Pacific perspective. Journal of Asian Economics, 1. pp. xx xx. Krueger, A., 1978, Foreign Trade Regimes and Economic Development: Liberalisation Attempts and Consequences. Ballinger, Cambridge, MA. Krugman, P., 1994, The myth of Asia s miracle. Foreign Affairs, 73(6), pp Kunst, R. M. and D. Marin, 1989, On exports and productivity: a causal analysis. Review of Economic and Statistics, pp Liu, X., H. Song and P. Romilly, 1997, An empirical investigation of the causal relationship between openness and economic growth in China. Applied Economics, 29, pp Lucas, R. E., 1993, Making a miracle. Econometrica, 61(2), pp MacKinnon, J., 1991, Critical values for cointegration tests. In R. F. Engle and C. W. J. Granger, Eds, Long-Run Economic Relationships: Readings in Cointegration. Oxford University Press, New York. McMillin, W. D., 1988, Money growth Volatility and the Macroeconomy, Journal of Money, Credit and Banking, 20, pp Masih, A. M. M. and R. Masih, 1996, Empirical tests to discern the dynamic causal chain in macroeconomic activity: new evidence from Thailand and Malaysia based on a multivariate cointegration/ vector error-correction modelling approach. Journal of Policy Modelling, 18, pp Michaely, M., 1977, Exports and growth: an empirical investigation. Journal of Development Economics, 4, pp Osterwald-Lenum M., 1992, A note with quantiles of asymptotic distribution of the maximum likelihood cointegration rank test statistics. Oxford Bulletin of Economics and Statistics, 53, pp Phillips, P. C. B. and P. Perron, 1988, Testing for unit root in time series regression. Biometrika, 74, pp Riezman, G. R., C. H. Whiteman and P. M. Summers, 1996, The engine of growth or its handmaiden? A time-series assessment of export-led growth. Empirical Economics, 12, pp Rodrik, D., 1992, The limits of trade policy reform in developing countries. Journal of Economic Perspectives, 6, pp Said, E. S. and D. A. Dickey, 1984, Testing for unit roots in autoregressive-moving average models of unknown order. Biometrika, 71, pp Schwarz, G., 1978, Estimating the dimension of a model. Annals of Statistics, 6, pp