Estimating Export Equations for Developing Countries Sanjesh Kumar * The paper uses annual time series data to estimate the price and income elasticities of export demand for three developing countries Fiji, Papua New Guinea (PNG) and Bangladesh. The LSE-Hendry s general-to-specific approach (known as GETS) is employed by using correct specifications of relative price to find the relevant export elasticities. Income elasticities of export demand are found to be unity for all the three countries. The price elasticity estimates, in contrast, have correct signs, and give plausible magnitudes. Based on the results, exports in developing countries can be seen as an engine for growth, and thus, export promotion policies are necessary to improve trade balance and to achieve greater export-based growth. Introduction In formulating export promotion policies, the knowledge of relevant elasticities is essential. This is because higher elasticities allow a basis for greater export-based growth policies. A sizeable number of empirical studies have been carried out in this field in order to estimate export equations for individual countries and to determine their price and income elasticities. The purpose of estimating export equations has not only exclusively been to test certain theoretical hypothesis, but also to help policymakers in improving the evaluation of potential policy options. Exports is potentially seen as a growth enhancing tool, adopted by many developing countries, because of its influential role in generating sufficient levels of foreign reserves needed to fund imports and finance investment goods, for present and future capital formation. Therefore, to allow exports to act as an engine for growth in developing countries (i.e., directly contributing to GDP), proper estimation of price and income elasticities is necessary, even though there is some sense of ambiguity on the sizeable characteristic of the price and income elasticities for exports demand that prevails in the developing countries. In addition, export and import demand elasticity parameters also play an important role in the measurement of real exchange rate variation on the trade balance, and is also crucial in deriving the conclusion that the Marshall-Learner 1 phenomena proposes. A higher income elasticity of export demand suggests that exports can act an engine of growth, while a higher price elasticity suggests a greater * Graduate Assistant, Department of Economics, University of the South Pacific, Suva, Fiji. E-mail: kumar_sj@usp.ac.fi 1 The Marshall-Learner condition implies that if the sum of the absolute values of export and import demand elasticities is greater than 1, then devaluation helps to bring about an improvement in the trade balance. Estimating 2009 The Export Icfai Equations University for Press. Developing All Rights Countries Reserved. 17
competition for a country s exports at the global scene. Thus, in this situation, a successful real devaluation 2 can improve and enhance export earnings. Some leading empirical works by researchers in estimating export demand equations, have been subjected to misspecification and omitted variable bias for instance, Senhadji and Montenegro (1999) because they ignored the powerful influence of Exchange rate (E ) in the relative price variable, which is a significant determinant of the relative price. A number of papers discussed later, estimating exports demand, having omitted E, have defined their relative P price variable as ln D, where P P D is the domestic price of exports and P F is the price level of F trading partners, respectively. Therefore, ignoring the exchange rate variable implies that the income and price elasticities obtained render a biased estimate, because it results in the underestimation of price elasticity and the overestimation of income elasticity, thus having serious implications on the proposed trade policies. Rao and Singh (2006) propose the correct specification of relative price variable by incorporating exchange rate with domestic price level (P D ) and foreign price level (P F ). The purpose of this paper is to empirically estimate the income and price elasticities for the export demand function for three developing countries Fiji, Papua New Guinea (PNG) and Bangladesh by using the correct specification, as proposed by Rao and Singh (2006), for the relative price variable. The LSE-Hendry s general to specific (known as GETS) approach is used to estimate the long-run export equations for these countries. The paper briefly surveys the empirical literature on export equations, followed by discussions of the model specification and empirical results. It finally concludes, offering policy implications of the main findings. A Brief Survey of Literature Here we provide a brief survey of the existing studies for the purposes of comparing our results to the existing findings, as well as to place the current research in the context of existing literature. Most conventional trade models typically consider foreign income and relative prices as the main determinants of exports. However, problems of misspecification of relative price variable, parameter instability (Hooper et al., 1998), and the non-stationarity of the data in numerous studies has rejuvenated the interest in the study of this relationship. Therefore export demand specification is crucial for meaningful export forecasts, international trade planning and policy formulation. In their influential works, Senhadji (1998) and Senhadji and Montenegro (1999), using the conventional model, estimated export demand elasticities for a large number of developing and industrial countries using the OLS and FMOLS techniques, and found the average long-run price and income elasticities to be 1 and 1.5, respectively, with Asian countries facing the highest income and price elasticities in the sample. The authors failed to find any evidence of cointegration between export volumes, relative price and export demand, and used statistical techniques that 2 Rose (1990 and 1991), Ostry and Rose (1992), Reinhart (1995), and Marquez and McNeilly (1998) further suggested how a real devaluation affects exports and imports. 18 The Icfai University Journal of Applied Economics, Vol. VIII, No. 2, 2009
can deal with non-stationarity. In addition, the authors specification of relative price is misspecified as they do not include the exchange rate in it, and therefore, the estimates of income elasticities are perhaps overestimated, which may indicate towards wrong policy implications. Till date, there has been very limited research carried out for export demand on Fiji. Reddy (1997) developed an export and import model to estimate their relevant relative price and income elasticities, using data from 1970 to 1994. Utilizing the OLS procedure, he found that the implied income and relative price elasticities for the exports demand were 0.76 and 0.78, respectively. The author s specification of relative price variable in both the export and the import equation is inappropriate, 3 as he failed to incorporate the exchange rate variable in it. In addition, the author also failed to carry out the test on the stationarity of the variables. Thus, both equations are said to be estimated with the presence of unit roots. Similarly, Asafu-Adjaye (1999) conducted an empirical investigation of the effects of exchange rate variability on the export growth in Fiji, for the period 1981 to 1997, using error-correction and cointegrating modeling techniques. The implied long-run income elasticity was 1.4 and the relative price elasticity was around 0.2. However, similar to other empirical works, the specification of the relative price was incorrect. Prasad (2000) estimated a single equation model for exports on Fiji for a sample period, from 1968 to 1998. Using the unrestricted Error Correction Model (ECM) procedure, the estimates for the long-run income elasticity of demand were found to be around 2.4. Stylized facts suggest that for developing countries (including Fiji), income elasticity of demand is around unity. Therefore, the obtained results suggest that Fiji s export is seen as a luxury good by its trading partners. In addition, there is some element of ambiguity in the paper as to how the cointegrating relationship is obtained. In the same line, Narayan and Narayan (2004 and 2005), using the ARDL, Dynamic Least Squares and FMOLS methods, estimated income and price elasticities for exports and imports for Fiji. Using annual data from 1970-2002 and 1972-1999 for exports and imports respectively, they showed that the income elasticity for exports is around 0.80, while for imports, it ranges from 1.05 to 1.90. However, the author s specification of relative prices does not include exchange rate. Rao and Singh (2006), in their most recent work, estimated export equation for Fiji using annual data from 1970 to 2002, through the usage of proper specification of the relative price variable in the equation in which they incorporated exchange rate. Utilizing the JML, GETS and FMOLS procedure to estimate the exports demand function, their implied relative price elasticity is 0.86 in GETS and 1.02 in FMOLS, whereas their implied income elasticity is 1.16 and 0.99 for GETS and FMOLS, respectively. The authors, using the correct specification of relative price, empirically showed that ignoring the exchange rate in the relative price causes an overestimation of the income elasticity by some 40% to 60% in the export equation. Singh (2006), using the same sample and methodology as Rao and Singh, reestimated the exports and imports equation using JML, GETS and FMOLS time series methods for the sample 3 This leads to a misspecification bias in the equations, and gives an inappropriate (under and over) estimate of the price and income elasticity, signaling wrong policy implications. Estimating Export Equations for Developing Countries 19
period 1970-2002. Using proper specification of relative price variable, by including the exchange rate, he concluded that income elasticity is unity and the relative price elasticity is as high as 1.25 for exports in Fiji, while for imports, the income and price elasticities are found to be 1.20 and 0.50, respectively. A high income elasticity implies that export is an engine of growth, and the presence of the Marshall-Learner condition suggests that real devaluation can improve trade performance in Fiji. Empirical literature on estimating export demand function for PNG is relatively non-existent. This has provided the motivation for selecting such a country, so that the findings of the current research can be placed in the context of the existing literature. Metwally (1995) developed and tested a simultaneous structural equation model to assess the effects of growth in exports to the EU, on the economic development of five South-East Asian countries, including PNG. The dependent variable used in this study is the exports of Asian countries, the explanatory variables being the real rate of growth of GDP of the EU and the export price index of the Asian countries. Using the method of the two-stage least squares to estimate the equations for the period 1974-1992, the growth rate of GDP and the export prices were 0.05 and 0.007, respectively. The limitation of this study is three fold. Firstly, instead of using relative prices, the estimation is carried out using domestic export prices of Asian countries. Secondly, the sample size used to estimate the various equations is small (i.e., 18 years). Thirdly, no unit root test is carried out to determine the stationarity and non-stationarity of the variables used. Bayes et al. (1995) estimated the demand and supply models of exports with annual data, and found that Bangladesh s export is highly sensitive to the income growth of its trading partners, estimating that a 10% rise in foreign income would raise the demand for Bangladesh s exports by 23%. In the same line, Kabir (1988) used the standard regression model to investigate the Bangladesh s export demand function and found evidence for income inelastic demand for exports. Both studies have failed to include exchange rate to determine the relative price. Model Specification and Empirical Results The basic specification of exports demand function 4 for the three developing countries under study, in their log-linear form is: P D ln X t 0 1 ln YFt 2 ln t E P...(1) F where X is the quantity of real exports, YF is the income of trading partners, P D and P F are the domestic price of exports and the price level of trading partners, respectively. E is the nominal exchange rate (price of one unit of foreign currency in terms of domestic currency) and t is the error term with the standard classical properties. Since each variable is defined in logarithmic terms, the estimated coefficients of 1 and 2 are the income and price elasticities of exports, respectively. Our prior expectations 5 are that 1 will be positive and 2 will be negative. 4 Our specification of export demand function is the same as that employed by Rao and Singh (2006). 5 A rise in a trading partner s income would lead to an increase in the demand for exports. Similarly, a rise in relative prices would lead to a decline in the demand for exports. 20 The Icfai University Journal of Applied Economics, Vol. VIII, No. 2, 2009
This paper estimates the long-run export demand function for Fiji, PNG and Bangladesh, using the LSE-Hendry s general-to-specific (GETS) approach, with annual time series data 6 (See Appendix A for data details and source). All the variables in Equation (1) when tested for unit roots are found to be I(1) at their levels and I(0) at their first difference, although pre-testing the variables for unit roots is not necessary in GETS. However, the usual Augmented Dickey-Fuller (ADF) test for unit roots is used, and the results are shown in Appendix B. Microfit 4.1 of Pesaran and Pesaran (1997) is used for the estimation. The Expor ts Elasticities LSE-Hendry s GETS approach is used with Non-Linear Least Squares (NLLS) to estimate Equation (1), in order to obtain the cointegrating long-run relationship of export demand function for Fiji, PNG and Bangladesh. The cointegrating coefficients are reported in Table 1. The estimates of income and relative price elasticities indicate that they are plausible with expected correct signs and the coefficients are found to be significant at the conventional 5% level of significance. Table 1: Cointegration Coefficient of Exports (GETS) Vari able F i j i PNG Bangl ade sh Constant 0.507 (1.78)** 18.819 (27.03)* 1.436 ( 1.10) In Y t 1.080 (5.21)* 0.850 (4.87)* 1.430 (6.56)* P ln D 0.828 ( 5.70)* 1.296 ( 5.18)* 1.009 ( 2.77)* PF E N o t e: The t-ratios for coefficients are reported in parentheses; * and ** indicate significance at 5% and 10% levels, respectively. The results for the respective developing countries indicate that the income elasticity for Fiji is around 1.08, for PNG 0.90, and for Bangladesh 1.43. The relative price elasticity is 0.83 for Fiji, 1.29 for PNG, and 1.01 for Bangladesh. Thus, there is no significant difference among the income elasticities of the three countries. However, there is a marginal difference among the estimated relative price elasticities of the three countries. The results, therefore, suggest that for developing countries, the income elasticity for export is unity and the relative price elasticity is as high as 1.29. An Autoregressive Distributed Lag Structure (ARDL) and its ECM is estimated in NLLS, with a restricted intercept term in Microfit 4.1 of Pesaran and Pesaran (1997). Using GETS, and applying the standard variable deletion test, a parsimonious exports equation for the short run and the long run is obtained in one step. The results are reported in Table 2, where A1, B1 and C1 of GETS are the initial parsimonious estimates. The implied income elasticities of exports demand in A1, B1 and C1 for Fiji, PNG and Bangladesh are unity, and relative price elasticities is in the range of 0.61 to 1.29. 6 Sample periods for Fiji, PNG and Bangladesh are 1970-2002, 1977-2001 and 1976-2000, respectively. Estimating Export Equations for Developing Countries 21
Table 2: Short-Run and Long-Run Exports Equations with GETS Variable Fiji PNG Bangladesh A 1 A 2 B 1 B 2 C 1 C 2 Constant 0.475 0.507 18.836 18.819 0.183 1.436 (1.50) (1.78)** (26.05)* (27.03)* ( 0.11) ( 1.10) ln Y t 1.133 1.080 0.843 0.850 1.437 1.430 (5.16)* (5.21)* (4.66)* (4.87)* (6.40)* (6.56)* P ln D PF E 0.608 0.828 1.286 1.296 0.613 1.009 ( 1.01) ( 5.70)* ( 4.83)* ( 5.17)* ( 1.14)* ( 2.77)* 0.218 0.207 0.597 0.593 0.439 0.458 ( 2.00)** ( 2.29)* ( 3.34)* ( 3.48)* ( 3.91)* ( 4.03)* P ln D E PF 0.715 0.828 0.878 0.847 ( 4.46)* (C ) ( 2.79)* (C ) P D, t 1 ln E t 1 PF, t 1 0.311 0.301 ( 2.27)* ( 2.42)* ln X t 1 0.480 0.484 (2.51)* (2.64)* lny t 5.053 3.827 (4.28)* (6.01)* lny t 2 2.884 3.827 ( 2.91)* (C ) Coup 0.147 ( 2.13)* R 2 0.545 0.614 0.468 0.494 0.766 0.757 SER 0.096 0.089 0.109 0.106 0.053 0.054 2 ( sc ) 0.174 0.086 0.928 0.783 3.265 1.221 (0.676) (0.770) (0.335) (0.376) (0.071) (0.269) 2 ( ff ) 0.109 0.422 1.000 0.461 4.126 2.514 (0.741) (0.516) (0.371) (0.497) (0.042) (0.113) (Contd...) 22 The Icfai University Journal of Applied Economics, Vol. VIII, No. 2, 2009
Table 2: Short-Run and Long-Run Exports Equations with GETS (...contd) Variable Fiji PNG Bangladesh A 1 A 2 B 1 B 2 C 1 C 2 2 n ( ) 6.360 0.079 0.849 0.853 0.054 0.605 (0.042) (0.961) (0.654) (0.653) (0.973) (0.739) 2 ( hs) 1.535 1.141 2.650 2.745 0.041 0.428 (0.215) (0.285) (0.104) (0.098) (0.840) (0.836) Note: The t-ratios for coefficients and p-values for 2 tests for serial correlation, functional form misspecification, non-normality and heteroscedasticity in residuals are in parentheses; * and ** indicate significance at 5% and 10% levels respectively; is the speed of adjustment and (c ) is the constraint variable. Microfit 4.1 of Pesaran and Pesaran (1997) is used for estimation. All the estimated crucial long-run parameters and short-run dynamics are meaningful and significant, thus giving satisfactory results. Our preferred final equation, after some parameter restrictions using Wald test, is reported in A2, B2 and C2. The t-ratios for the coefficients and the p -values for 2 summary statistics are in parentheses. The 2 statistics indicate that there is no serial correlation, 2 (sc ), non-normality, 2 (n), heteroscedasticity, 2 (hs) and functional form misspecification, 2 (ff ). The Standard Error of Regression (SER) is found to be high, but it is plausible for small developing countries where export growth rate is highly volatile. The coefficient of error correction ( ) lies in the range of 0.21 to 0.59 and has the correct negative sign serving as the negative feedback function. This implies that if there are departures from equilibrium in the previous period, this departure takes 2 to 5 years to slowly adjust back to the equilibrium for the three countries under study. We have also tested for the effects of structural variables for the three countries and have found that structural dummies like devaluation, Asian Crisis, drought and trade liberalization were not significant for PNG and Bangladesh in our sample. However, in the case of Fiji, it was found that the political coups had a temporary negative effect on Fiji s exports. Our preferred GETS (A2, B2 and C2) equation was tested for temporal stability using the CUSUM and CUSUM of squares test. The CUSUM test showed that they are stable. The CUSUM of squares stability test results are given in Figures 1, 2 and 3 in Appendix C. Conclusion and Policy Implications The knowledge of relevant export elasticities is essential because higher elasticities allow for greater reliance on export-based growth policies. Therefore, this paper seeks to estimate the export demand function for three developing countries Fiji, PNG and Bangladesh using the time series data, and determines the crucial long-run export elasticities. LSE-Hendry s general-to-specific (GETS) approach is employed to estimate the export elasticities. Most conventional trade models typically consider foreign income and relative prices as the main determinants of exports. However, problems of misspecification of relative price variable is present in most empirical studies. Therefore, in order to avoid misspecification bias, the correct specification of relative price, as proposed by Rao and Singh (2006), has been used, in order to obtain unbiased results. The estimates of the crucial Estimating Export Equations for Developing Countries 23
parameters of export demand function suggest that the income elasticities for the three developing countries under study is around unity, whereas the relative price elasticities are found to be as high as 1.29. The relative price and income elasticities obtained in this study, suggest the scope for export growth policies for the three developing countries, which will allow exports to act as an engine of growth, by directly contributing to the GDP. While the results sound promising for the three developing countries, certain policy recommendations can be formulated to promote exports. The three developing countries need to reduce imports and attract more FDI. Trade liberalization should be vigorously undertaken in order to ensure that exports increase. Local policymakers need to consult with private sector importers and exporters in order to identify effective measures to reduce transaction and trade costs for maintaining international competitiveness. Acknowledgment: The author is grateful to Professor B B Rao and Rup Singh of the University of the South Pacific, for their comments and suggestions. However, the errors are solely the author s responsibility. References 1. Asafu-Adjaye J (1999), Exchange Rate Variability and Export Growth in Fiji, Asia Pacific School of Economics and Management Working Papers (99-4), Asia Pacific Press, Australian National University. 2. Bayes A M, Hossain I and Rahman M (1995), Independent Review of Bangladesh s Development External Sector, Centre for Policy Dialogue, Dhaka. 3. Hooper P, Johnson K and Marquez J (1998), Trade Elasticities for G-7 Countries, International Finance Discussion Papers, 609. 4. Kabir R (1988), Estimating Import and Export Demand Function: The Case of Bangladesh, The Bangladesh Development Studies, Vol. 16, pp. 115-127. 5. Marquez and McNeilly (1998), A Framework for Economic Forecasting, Econometrics Journal, Vol. 1, pp. 228-66. 6. Metwally M M (1995), The Interaction Between Economic Growth in the EU and South-East Asia, Economic Business Review, Vol. 95, No. 2, pp. 40-47. 7. Narayan P and Narayan S (2004), Determinants of Demand for Fiji s Exports: An Empirical Investigation, The Developing Economies, Vol. 42, No. 1, pp. 95-112. 8. Narayan P and Narayan S (2005), Estimating Income and Price Elasticities of Imports for Fiji in a Cointegration Framework, Economic Modeling, Vol. 22, No. 3, pp. 423-438. 9. Ostry J D and Rose A K (1992), An Empirical Evaluation of the Macroeconomic Effects of Tarrifs, Journal of International Money and Finance, Vol. 11, No. 1, pp. 63-79. 10. Pesaran M and Pesaran B (1997), Working with Microfit 4.0, Oxford University Press, Oxford. 11. Prasad S (2000), Determinants of Exports in Fiji, Staff Working Papers (04/2000), The Reserve Bank of Fiji, Suva. 24 The Icfai University Journal of Applied Economics, Vol. VIII, No. 2, 2009
12. Rao B and Singh R (2006), Estimating Exports Equations, Applied Economics Papers (Forthcoming), UK. 13. Reddy M (1997), Devaluation and Economic Simulation: The Fiji Economy Post-Coup, Pacific Economic Bulletin, Vol. 12, No. 2, pp. 85-94. 14. Reinhart C (1995), Devaluation, Relative Prices and International Trade, Staff Working Paper, No. 42, June, pp. 290-312. 15. Rose A K (1990), Exchange Rates and the Trade Balance: Some Evidence from Developing Countries, Economics Letters, Vol. 34, No. 3, pp. 271-275. 16. Rose A K (1991), The Role of Exchange Rates in a Popular Model of International Trade: Does the Marshall-Lerner Condition hold?, Journal of International Economics, Vol. 30, Nos. 3-4, pp. 301-316. 17. Senhadji A (1998), Time Series Estimation of Structural Import Demand Equations: A Cross-Country Analysis, IMF Staff Paper, June, The International Monetary Fund, Washington DC. 18. Senhadji A and Montenegro C (1999), Time Series Analysis of Export Demand Equations: A Cross-Country Analysis, IMF Staff Paper, September, The International Monetary Fund, Washington DC. 19. Singh R (2006), Cointegration Test on Trade Equations: Is Devaluation an Option for Fiji?, Staff Working Paper (18/2006), The University of the South Pacific, Suva. Estimating Export Equations for Developing Countries 25
Appendix A Data 7 X t = Total exports of goods and services (f.o.b) for Fiji, PNG and Bangladesh, deflated by the export price index; P D = Price of domestic export goods (2000 = 100); P F = Import weighted average of major trading partners export price index, computed as the share of respective imports to total imports; E t = Nominal exchange rate, as the price of one unit of foreign currency, in terms of domestic currency; YF = Trade weighted average real income of major trading for the three countries under study. Trade weights are computed as the share of trade to each country relative to total trade; and COUP = Temporary dummy for political instability. Data constructed as one in 1987 and zero in other periods. Source: IFS CD-ROM (2005) and UN Statistics Division. Appendix B ADF Unit Root Tests Results Fiji Vari able La gs ADF Vari able La gs ADF ln X [2] 2.98 ln X [1] 6.57 (3.58) (2.98) ln YF t [1] 2.93 ln YF t [0] 4.07 (3.58) (2.98) ln P D PF E P D ln PF E [0] 4.07 (3.58) (2.98) (Contd...) 7 The data for estimating export demand function of Fiji is obtained from Rao and Singh (2006) and EC403 Lab exercises at the University of the South Pacific, Suva, Fiji. 26 The Icfai University Journal of Applied Economics, Vol. VIII, No. 2, 2009
Appendix B (...contd) ADF Unit Root Tests Results PNG ln X [1] 1.41 ln X [0] 4.01 (3.62) (2.99) ln YF t [0] 2.16 ln YF t [0] 3.84 (3.62) (3.00) ln P D PF E [1] 0.32 P D ln PF E [0] 3.55 (3.62) (2.99) Bangladesh ln X [0] 2.45 ln X [0] 3.46 (3.62) (3.01) ln YF t [1] 2.85 ln YF t [1] 3.82 (3.69) (3.05) ln P D PF E [0] 1.72 P D ln PF E [0] 3.94 (3.71) (3.07) N o t e: ADF is the usual Augumented Dickey-Fuller test; figures in brackets are the 5% level critical values; the lag lengths are selected using AIC and SBC criteria, for example, [0] indicates that lag 0 is significant in the respective test. Intercept and trend are included in tests at the levels of the variables, but are excluded in the tests at their first differences. The null hypothesis in ADF is that the variable contains a unit root. Appendix C Figure 1: Stability Test on the GETS (A2) Equation for Fiji 1.5 Plot of Cumulative Sum of Squares of Recursive Residuals 1.0 0.5 0 0.5 1972 1977 1982 1987 1992 1997 2002 (Contd...) Estimating Export Equations for Developing Countries 27
Appendix C (...contd) Figure 2: Stability Test on the GETS (B2) Equation for PNG 1.5 Plot of Cumulative Sum of Squares of Recursive Residuals 1.0 0.5 0 0.5 1977 1982 1987 1992 1997 2001 Figure 3: Stability Test on the GETS (C2) Equation for Bangladesh 1.5 Plot of Cumulative Sum of Squares of Recursive Residuals 1.0 0.5 0 0.5 1983 1985 1987 1989 1991 1993 1995 1997 1999 2000 Note: The dotted lines represent critical bounds at 5% significance level. Reference # 05J-2009-03-02-01 28 The Icfai University Journal of Applied Economics, Vol. VIII, No. 2, 2009