INTERNATIONAL ECONOMIC JOURNAL Volume 12, Number 3, Autumn 1998

89

COINTEGRATION APPROACH TO ESTIMATE THE LONG-RUN TRADE ELASTICITIES IN LDCs MOHSEN BAHMANI-OSKOOEE* University of Wisconsin-Milwaukee

The Marshall-Lerner condition postulates that if the sum of import and export demand elasticities add up to more than one, devaluation should improve the trade balance in the long-run. This paper is the first to employ a long-run method, i.e., cointegration technique to estimate trade elasticities in less develop countries. In most cases the results reveal that indeed trade elasticities are large enough to support devaluation as a successful policy to improve the trade balance. [F10, F31]

1. INTRODUCTION An important macroeconomic policy to reduce the trade deficit is said to be devaluation. As far as the experiences of Less Developed Countries (LDCs) are concerned, it is not uncommon to find arguments for and against devaluation. Researchers have used different methodologies to assess the response of the trade balance to devaluation in LDCs, with mixed conclusions. First, there are early studies which basically concentrated on estimating the import and export demand elasticities, either directly or indirectly, to determine whether devaluation can improve the trade balance of an LDC. Gylfason and Risager (1984: Table 3) used parameter estimates of a general macro model to show that devaluations can improve the current account in LDCs. Bahmani-Oskooee (1986: Tables 2-8) estimated price elasticities through import and export demand models and showed that they are high enough to warrant improvement in the trade balance. Marquez (1990) used band spectrum analysis and estimated bilateral trade elasticities from which multilateral trade elasticities were obtained. By lumping less developed countries in one group, she estimated their trade elasticities (sum of import and export price elasticities, i.e., Marshall-Lerner condition) to be -0.78, not large enough for a successful devaluation. When new econometric methods (such as cointegration technique) are introduced, some of the old theories receive renewed attention and the import and export demand equations of LDCs are no exception. Specifically, given the contrasting results among different studies mentioned above, it is a worthy exercise to investigate the time-series properties of the trade data and estimate the trade elasticities in LDCs using relatively a new technique.1
*Thanks are due to two anonymous referees for helpful comments. Any error is my own responsibility.

90

MOHSEN BAHMANI-OSKOOEE

Therefore, the purpose of this paper is to estimate multilateral trade elasticities for as many LDCs as the data permits. Since the Marshall-Lerner condition is a long-run condition, a long-run approach such as the cointegration technique of Johansen and Juselius (1990) is the appropriate method. Section 2 introduces the models and the methodology as well as the results. It is shown that for most LDCs in our sample, price elasticities are high enough to rely upon devaluation as a successful policy to improve their trade balances. Section 3 concludes. Data definition and sources are cited in an Appendix. 2. THE MODELS, METHODOLOGY AND THE RESULTS To study the long-run equilibrium relation between volume of imports and its determinants in one relation and the volume of exports and its determinants in another relation, following the literature we assume that the import and export demand equations take the following forms:

Log Mt = F[Log (PM/PD)t, Log NEXt, Log Yt] and

(1)

Log Xt = F[Log (PX/PXW)t, Log NEX, Log YWt]

(2)

where M = volume of imports; PM = import prices; PD = domestic price level; NEX = nominal effective exchange rate; Y = domestic income; X = volume of exports; PX = export prices; PXW = world export price level; and YW = world income. Note that the price term in each equation is expected to carry a negative coefficient. Defined as units of foreign currency per unit of domestic currency a decrease in nominal effective exchange rate which indicates a depreciation of domestic currency is expected to discourage imports and encourage exports. Thus, the expected sign of the coefficient attached to NEX variable in the import demand equation is positive and in the export demand equation, negative. The income terms are expected to carry positive coefficients in both models.2
1 Bahmani-Oskooee (1991), Bahmani-Oskooee and Malixi (1992) and Himarios (1985, 1989) are other studies who rather than estimating the Marshall-Lerner condition, they established a direct link between the trade balance and the exchange rate and showed that in most LDCs devaluations improve their trade balances. 2 For more on these formulations and their different variants as well as the expected signs see Houthakker and Magee (1969), Khan (1974), Warner and Kreinin (1983), and BahmaniOskooee (1986).

LONG-RUN TRADE ELASTICITIES

91

Note that in models that include real and nominal variables, it is possible to have more than one cointegrating vectors. King et al (1991) have argued and demonstrated that plausibility of the additional relationships among all or some of the variables in the cointegrating space is dictated by economic theory. For example in the import demand model if import and income variables are insignificant, then presence of cointegration will be between exchange rate and relative prices, a test of the purchasing power parity (PPP). Bahmani-Oskooee (1995a) has shown that even if all variables are significant, still it is possible to have more than one vector, again, dictated by economic theory. For example, in model (1) one relation could be import demand equation, a second relation could be an exchange rate equation where relative prices, income and volume of imports are the determinants of nominal exchange rate. Finally, the last relation could be a price equation in which volume of imports, income and exchange rate are the main determinants of relative prices. Using quarterly data over the current floating exchange rate period of 1973I1990IV we first try to establish whether a long-run equilibrium relation exists between the variables in equations (1) and (2) for Greece, Korea, Pakistan, the Philippines, Singapore and South Africa using the Maximum-Likelihood cointegration procedure of Johansen (1988) and Johansen and Juselius (1990).3 The choice of countries is dictated by the availability of quarterly data. The first step in applying the cointegration technique is to determine the degree of integration of each variable., i.e., how many times each variable needs to be differenced in order to achieve stationarity. To this end, we applied the ADF test which included a trend term. Following Dickey et al (1986: 19) choice of lags in the ADF was governed by the level of significance of estimated lag coefficients using the standard t-test. The results not reported but available from the author upon request, showed that for all countries all variables are first difference stationary except Log X and Log (PX/PXW) for the Philippines. However, for consistency we treat them as stationary variables and review the results for the Philippines with some caution. We are now in a position to apply Johansen and Juselius method.4 Their method which is based on maximum likelihood estimation procedure amounts to calculating two test statistics known as -max and trace that are used to determine number of cointegrating vectors. When quarterly data are used, a common practice is to employ four lags in the procedure. Thus, we carry out Johansens procedure by employing four lags and following Johansen and Juselius (1992: 219) we make sure that the residuals pass the normality test using the Jarque-Bera statistic.5 The results of the Note that Hakkio and Rush (1991) have shown that the ability of cointegration tests to detect cointegration is a function of total sample length and not a function of data frequency. Thus, our results must be viewed with some caution. 4 It should be indicated that Johansen-Juselius technique is carried out by MFIT3.0, an statistical package by Pesaran and Pesaran (1991). 5 In two instances, i.e., export demand function of Greece and South Africa, the Jarque-Bera test did not satisfy the normality condition. In the former case we employed five lags and in the
3

92

MOHSEN BAHMANI-OSKOOEE

max and the trace tests, not reported but available from the author upon request, showed the null hypothesis of no cointegration (r = 0) among all variables that enter into the import and export demand equations can be rejected at the 5% or 10% level of significance, at least by one of the tests for all countries. By relying on at least one of the tests, the results showed that there was only one cointegrating vector among the variables of import demand function of Greece, and South Africa; two cointegrating vectors in the case of Korea, and Singapore; and three cointegrating vectors in the case of Pakistan and the Philippines. Among the variables of export demand equation, there was one cointegrating vector in the results for Pakistan and Singapore; two cointegrating vectors in the case of Korea, the Philippines, and South Africa and three cointegrating vectors in the case of Greece. The next step is to report the cointegrating vectors. In order to interpret the estimated cointegrating vectors, it is a common practice to normalize them on one of the variables by setting its estimated coefficient equal to -1. Since our interest is to obtain long-run trade elasticities, we normalize the cointegrating vectors on Log M in the import demand equation and Log X in the export demand equation. This practice enables us to read the elasticities directly from cointegrating vectors. Table 1 reports all possible cointegrating vectors that are associated with the largest eigenvalue of the stochastic matrix for each country. Also, reported in Table 1, in the bracket next to each coefficient is the likelihood ratio test for the exclusion of each variable from the cointegrating space. Johansen and Juselius (1990) show that the test statistic is distributed as 2 with degrees of freedom equal to number of cointegrating vectors.6 From Table 1 one forms an opinion that indeed price elasticities are high enough, therefore, the Marshall-Lerner condition is satisfied for almost all countries. This is due to the fact that the sum of absolute values of import and export own price elasticities add up to more than one. Furthermore, the null hypothesis that the coefficient of each variable could be restricted to zero is rejected in majority of cases. This is because the calculated 2 statistic for exclusion of each variable is larger than its critical value. All in all, our long-run approach to estimate the Marshall-Lerner condition lends support for the notion that devaluations could improve the trade balances of LDCs in our sample. How are these results compared to those of the others in the literature? Khan (1974) estimated the import and export demand functions for 15 LDCs using annual data over 1951-69 period by the means of two-stage least square. Pakistan and the Philippines are the only countries from our sample which were also included in Khans study. While our finding for Pakistan is inconsistent with that of Khan, we are consistent on the results for the Philippines. Greece, Korea and South Africa from our sample were also included in Bahmanilater case three lags. We also tried 2 and 6 lags for all cases, there was no substantial change in the conclusion. 6 For more on the formulation of this test and its application see, Bahmani-Oskooee (1996).

LONG-RUN TRADE ELASTICITIES

93

Oskooee (1986) who employed quarterly data over 1973-1980 period. While in the results for Korea and South Africa our findings that the M-L condition is satisfied are consistent with that of Bahmani-Oskooee (1986), in the case of Greece we are inconsistent. More precisely, while our results in the case of Greece show that the sum of price elasticities add up to more than one, Bahmani-Oskooees results from his Table 3 show that they add up to less than one. Table 1. Cointegrating Vectors Normalized on the Log M and Log X
Country Greece Korea Pakistan Loq M 1.00[12.7]* 1.00[7.99]* 1.00 1.00[11.5]* 1.00 1.00 1.00[15.9]* 1.00 1.00 1.00[15.7]* 1.00 1.00[8.31]* + Log Y 1.357[10.9]* 0.311[1.95] 0.402 1.111[11.9]* 1.443 2.116 1.970[12.5]* 3.965 1.267 1.260[15.9]* 1.054 2.172[9.78]* + Log YW 0.662[1.82] 0.527 0.609 1.277[1.27] 1.180 2.256[4.13]* 7.079[5.53]* 3.040 4.046[4.49]* 3.781[8.59]* 4.786 Log (PM/PD) 1.93[21.7]* 2.01[22.6]* 1.91 1.23[28.9]* 2.28 0.90 3.80[31.6]* 9.86 0.52 0.15[2.19] 0.30 1.37[6.82]* Log (PX/PXW) 1.93[10.2]* 12.9 1.46 1.84[5.41]* 1.41 1.84[12.7]* 4.79[11.2]* 0.90 0.83[2.75]* 2.09[9.29]* 3.07 + Log NEX 0.014[0.01] 0.235[0.45] 0.005 0.002[8.16]* 0.002 0.014 0.225[11.9]* 2.657 0.203 1.657[30.4]* 0.468 0.327[9.19]* + Log NEX 0.001[2.26] 0.943 0.479 0.401[4.81]* 4.258 2.992[13.9]* 0.523[3.70] 0.355 3.417[10.8]* 0.972[8.72]* 0.227

Philippine

Singapore S. Africa

Country Greece

Loq X 1.00[12.9]* 1.00 1.00 1.00[19.1]* 1.00 1.00[10.0]* 1.00[3.39] 1.00 1.00[4.93]* 1.00[9.40]* 1.00

Korea Pakistan Philippine Singapore S. Africa

Note: a. Number inside the bracket next to each coefficient is the 2 statistic testing the null hypothesis of restricting that coefficient to 0 in all cointegrating vectors. b. At the 10% level of significance, the critical value of 2(1) = 2.70; 2(2) = 4.60; and 2(3) = 6.25. c. * indicates significant 2 statistic indicating that the null of restricting the coefficient to 0 is rejected.

94

MOHSEN BAHMANI-OSKOOEE

Our finding for Greece is also consistent with Brissimis and Leventakis (1989) who used a general equilibrium modeling approach to estimate price elasticities. Mah (1993) estimated the Korean import demand function only. Our finding of a price elasticity of -2.01 is somewhat larger than Mahs finding of -1.029 for almost similar period. Finally, our maximum-likelihood approach to estimate the Marshall-Lerner condition is consistent with the OLS estimates of Marquez (1990), but inconsistent with her band spectrum estimates. The conflicting results in some cases could be explained not only by different sample period, but also by the fact that previous studies employed non-stationary data while we have used stationary data. 3. SUMMARY AND CONCLUSION Introduction of any new estimation technique usually sets the scene to shed additional lights on some controversial economic issues and the trade elasticities are no exception. Historically the import and export demand elasticities have been of eminent practical importance to policy makers. This paper presents the first application of Johansen and Juselius cointegration technique to estimate the trade elasticities, thus, the Marshall-Lerner condition in some less developed countries. This long-run approach to estimate the long-run Marshall-Lerner condition reveals that indeed in most LDCs considered in this paper the condition is satisfied, indicating that devaluations could improve their trade balances. APPENDIX Data Definition and Sources All data are quarterly for the period 1973-1990 and are taken from the following sources. a. International Financial Statistics of IMF, various issues. b. Bahmani-Oskooee (1995b). c. OECD Statistics of Foreign Trade and OECD Main Economic Indicators, various issues. Variables: M = import volume. Nominal imports are deflated by import price index (1985=100) to obtain this measure. All data are from source a. Y = real GNP. Quarterly GNP figures were not available for the countries included in our study. Thus, we generated quarterly nominal GNP using the method in Bahmani-Oskooee (1986). We then deflated these nominal figures by the domestic price level (CPI, 1985=100) to obtain real GNP. The annual GNP and import figures used in this process are from source a.

LONG-RUN TRADE ELASTICITIES

95

PM = index of unit value of imports, 1985=100, source a. PD = index of domestic price level measured by CPI, 1985=100, source a. NEX = index of nominal effective exchange rate (1985=100) from source b. Note that Bahmani-Oskooee (1995b) who constructed the real and nominal effective exchange rate for 22 LDCs for 1971I-1990IV period defined them as number of units of foreign currency per unit of domestic currency. X = export volume. Nominal exports are deflated by export price index (1985=100) to obtain this measure. All data are from source a. PX/PXW = the ratio of each countrys dollar-denominated export unit value index (1985=100) to the dollar denominated export unit value index (1985=100) of the IMFs industrial country aggregate. This definition is used by Bailey et al. (1986). All data come from source a. YW = world income. Following others in the literature this variable is proxied by the index of industrial production (1985=100) in OECD countries. Data come from source c. REFERENCES Bahmani-Oskooee, Mohsen, Determinants of International Trade Flows: The case of Developing Countries, Journal of Development Economics, January-February 1986, 107-123. , Is There a Long-Run Relation Between the Trade Balance and the Real Effective Exchange Rate of LDCs? Economics Letters, August 1991, 403-407. , Mohsen, Source of Inflation in Post-Revolutionary Iran, International Economic Journal, Summer 1995a, 61-72. , Mohsen, Real and Nominal Effective Exchange Rates for 22 LDCs: 1971I-1990IV, Applied Economics, July 1995b, 591-604. , Mohsen, The Black Market Exchange Rate and the Demand for Money in Iran, Journal of Macroeconomics, Winter 1996, 171-176. , Mohsen and Margaret Malixi, More Evidence on the J-Curve from LDCs, Journal of Policy Modeling, October 1992, 641-653. Bailey, Martin J., George S. Tavalas, and Michael Ulan, Exchange-Rate Variability and Trade Performance: Evidence for the Big Seven Industrial Countries, Weltwirtschaftliches Archiv, 1986, 466-477. Brissmis, Sophocles N. and John A. Leventakis, The Effectiveness of Devaluation: A General Equilibrium Assessment with Reference to Greece, Journal of Policy Modeling, Summer 1989, 247-271. Dickey, David A., William R. Bell, and Robert B. Miller, Unit Roots in Time Series Models: Tests and Implications. The American Statistician, February 1986, 12-26. Gylfason, Thorvaldur and Ole Risager, Does Devaluation Improve the Current Account? European Economic Review, June 1984, 37-64. Hakkio, Craig S., and Mark Rush, Cointegration: How Short is the Long-Run?

96

MOHSEN BAHMANI-OSKOOEE

Journal of International Money and Finance, December 1991, 571-581. Himarios, Daniel, The Effects of Devaluation on the Trade Balance: A Critical View and Reexamination of Miles New Results. Journal of International Money and Finance, December 1985, 553- 563. Himarios, Daniel, Do Devaluations Improve the Trade Balance? The Evidence Revisited, Economic Inquiry, January 1989, 143-168. Houthakker, Hendrik S. and Stephen Magee, Income and Price Elasticities in World Trade, Review of Economics and Statistics, May 1969, 111-125. Johansen, Sren, Statistical Analysis of Cointegration Vectors, Journal of Economic Dynamics and Control, June/Sept. 1988, 231-254. Johansen, Sren and Katarina Juselius, Maximum Likelihood Estimation and Inference on Cointegration-With Application to the Demand for Money, Oxford Bulletin of Economics and Statistics, May 1990, 169-210. Johansen, S ren and Katarina Juselius, Testing Structural Hypothesis in a Multivariate Cointegration Analysis of the PPP and the UIP for UK, Journal of Econometrics, July-Sept. 1992, 212- 244. King, Robert G.; Charles I. Plosser; James H. Stock, and Mark W. Watson, Stochastic Trends and Economic Fluctuations, American Economic Review, September 1991, 819-840. Khan, Mohsin, Import and Export Demand in Developing Countries, IMF Staff Papers, September 1974, 678-693. MacKinnon, James J., Critical Values for Cointegration Tests, in Long-Run Economic Relationships: Readings in Cointegration, ed. R. F. Engle and C. W. Granger, Oxford, Oxford University Press, 1991, 267-276. Mah, Jai Sheen, Structural Change in Import Demand Behavior: The Korean Experience, Journal of Policy Modeling, 1993, 223-227. Marquez, Jaime, Bilateral Trade Elasticities, The Review of Economics and Statistics, February 1990, 70-77. Pesaran, Hashem M. and Bahram Pesaran, Microfit 3.0, An Interactive Econometric Package, Oxford: Oxford University Press, 1991. Warner, Dennis and Mordechai E. Kreinin, Determinants of International Trade Flows, Review of Economics and Statistics, February 1983, 96-104.

Mailing Address: Professor Mohsen Bahmani-Oskooee, Professor of Economics, Department of Economics, The University of Wisconsin-Milwaukee, Milwaukee, WI 53201, U.S.A. Tel: 414-229-4811, Fax: 414-229-3860, e-mail: bahmani@csd.uwm.edu