THE EFFECT OF NOMINAL EXCHANGE RATE DEPRECIATION ON TRADE IN ELEVEN DEVELOPING COUNTRIES

This study examined the effect of nominal exchange rate depreciation on the trade balance in 11 Asian-African countries between 1980 and 2019, and within the context of an exogenously determined single structural break. The countries had persistently experienced both nominal exchange rate depreciation and upward trends in trade in goods. Using the Chow test to frame the discussion, these countries were found to be facing structural changes associated with external factors such as the commodity price crisis in South Asia and the global financial crisis. The time-series autoregressive distributed lag (ARDL) approach with bounds test for cointegration and 26 Malaysian Management Journal, 25 (July) 2021, pp: 25–48 error-correction mechanism (ECM) was also applied for the analysis. The results of the study showed a long-term cointegration between the trade in goods and other variables included. Specifically, the nominal exchange rate depreciation positively affected the trade in goods in both the long-run and short-run in most of the Asian-African countries studied. There was a positive relationship between trade and foreign direct investment in the short-run, but this relationship mostly became insignificant in the long-run. Gross domestic product had a significant impact on trade performance in goods in both the long-run and shortrun in all countries studied.


INTRODUCTION
The way the exchange rate affects the balance of trade has been a fundamental issue in economics. In general, the economic rationale behind this study was whether the nominal exchange rate depreciation would provide 11 Asian-African countries between 1980 and 2019 with a competitive edge in obtaining favorable effects, or instead exposed them to unfavorable contract trade in goods. To put things into perspective, the study had assumed the postulates of the elasticities approach, which comprised the Bickerdike-Robinson-Meltzer (BRM) condition, the Marshall-Lerner (ML) theory, and J-curves theoy in its attempt to understand and explain the issues to be addressed, and thus attain the objective of the study.
To start the discussion, the statistics of nominal exchange rate changes and trade in goods over the past four decades in developing countries have been summarized in Figure 1 and Figure 2. As shown in Figure  1, the selected developing countries recorded a nominal exchange rate deprecation in most cases during the period 1980-2019. The nominal exchange rate depreciation in Ghana, Swaziland, and Tunisia was much lower than in other countries. In the case of Indonesia, the exchange rate depreciation during the period of 1980-1996 was narrower than those from 1997 until 2019. Nigeria experienced a narrow exchange rate depreciation from 1980 to 1993, and it was relatively stable from 1994 to 1998. The Nigerian Naira plunked in 1999 and deteriorated from 2000 to 2019.

Figure 1
The Official Exchange Rate Per USD Note: The data for Indonesia involved larger units (in thousand) as shown on the right-hand axis, and data for other countries are indicated in the left axis.
Bangladesh experienced exchange rate depreciation during the period 1980-2012, but the rate was relatively stable from 2013 to 2019. India recorded an exchange rate depreciation during the period 1980-2002, it appreciated during the period of 2003-2008, but slightly depreciated in recent years. As for Pakistan, the exchange rate depreciated from 1980 to 2001, but appreciated in 2002 and 2003. It depreciated again in the following years. Sri Lanka recorded an exchange rate depreciation during the period 1980-2019, except in 2005, 2008, and 2011. However, trade in goods showed an increasing trend in almost all countries except Indonesia, India, and Nigeria. A steady goods trade

Figure 2
Trade in Goods of Selected Developing Countries in USD.
Note: The data for Bangladesh involved a larger amout (in trillion) as shown in the right-hand axis, and data for other countries in the left-hand axis.
The upward trend in the trade in goods was also experienced by the Philippines and Pakistan. Nigeria also showed an upward trend in trade in goods during the period of 1980-2011. However, a decline was observed between the year 2011 and 2016. It had rebounded in recent years. It was quite similar in the case of Indonesia. In summary, the main objective of this study was to examine the short-run effect and long-run effect of the nominal exchange rate depreciation on the volume in the trade in goods. The paper is organised as follows. Section 2 presents a brief review of previous studies which had highlighted the linkage between exchange rate depreciation and trade. Section 3 describes the data, model, variables, and method of analysis. The empirical findings is presented in Section 4. The conclusions and implications of the study are provided in the final section.

LITERATURE REVIEW
The roles of the exchange rate are often associated with international trade. Auboin and Ruta (2011) were of the view that changes in the exchange rate had substantial effects on the economy. It could affect the structure of output and investment, influence labour market and prices, which in turn could lead to changes in domestic absorption and external trade. Two other studies also held similar views. Nicita (2013) posited that the exchange rate played a pivotal role in a country's trade performance, particularly, when exchange rate fluctuation would bring about repercussions on international trade, the balance of payments, and overall economic performance. In Helleiner (2011), it was argued that the exchange rate was more important than trade policies in influencing the amount of trade in a country, at least in the short to medium-term period.
Theoretically, an exchange rate fluctuation will bring about changes in the relative price of exportable and importable goods, thus a country's trade volume. The elasticities approach has claimed that a change in the exchange rate will affect the trade balance, which depends on the import and export supply, demand elasticities, and the initial volume of trade. The study by Ali et al. (2014) has suggested, though implicitly, the idea that the elasticities approach provided a focus on the issues of responsiveness of trade (both in terms of volume and values) to changes in the exchange rate. Moreover, the ML theory could explain the circumstances under which a change in the exchange rate of a country's currency would lead to an improvement or worsening of a country's balance of payments. Particularly, the ML condition has stated that an exchange rate depreciation would only lead to an improvement in the balance of payments if the sum of demand elasticity for imports and exports was greater than one. Meanwhile, Magee (1973) described the J-curve phenomenon as a J-pattern trend in which a country's trade balance would initially worsen following a period of exchange rate depreciation, it would then gradually recover and finally improve on its previous performance.
The concept of the J-curve posits that trade balance will worsen in the short-run. This is because by assuming that the volume of import will remain stable, the price of importable goods will become more expensive due to the exchange rate depreciation. An exchange rate depreciation may increase exports and reduce imports, and thus leading to an improved trade balance. A surplus in trade balance would take place for some time ahead, until a certain period through the J-curve pattern (Bahmani-Oskooee, 1985;Krugman & Baldwin, 1987). The exchange rate depreciation would lead to a change in spending from foreign to domestic goods, thus, improving the trade balance (Krueger, 1983;Himarios, 1985).
According to Bahmani-Oskooee (2001) and Kandil and Mirzaie (2003), the exchange rate depreciation was a way to gain global competitiveness in domestic industries and improve the trade balance. In a similar vein, Dooley et al. (2003) noted that some Asian countries kept their exchange rate deliberately undervalued as a part of export stimulation and economic growth strategy. By maintaining the exchange rate depreciation, developing countries could divert the spending in buying importable goods to buying domestic goods, and it, in turn, would stimulate trade and improve the trade balance.
In the study reported here, aggregate data, trade and other variables were in terms of the US dollar, and the nominal exchange rate of the respective country was relative solely to the US dollar. The sum of trade in goods was the dependent variable when examining the impact of the nominal exchange rate on trade. A substantial justification for applying trade volume in goods was explained by the UNCTAD (2016) as due to nominal factors, e.g. principally the fall and rise in the price of commodities and the overall appreciation of the US dollar against the currencies of the developing countries. In other words, the trade volume in goods reflected the resilience of the nominal exchange rate depreciation with respect to the global economic environment. Thus, the volume of exports and imports would adjust to changes in exchange rates, the adjustment denoted the resilience of the nominal exchange rate changes. In Pritchett (1991), it was his theoretical explanation on structural-adjusted trade intensity that the trade volume could be estimated as a function of size and other structural characteristics.
In the context of the 11 developing countries, resilience was illustrated by their pivotal roles as suppliers for commodities markets, such as raw materials, primary agricultural products, manufacturing, and natural and mineral products. Apart from this, the developing countries were also the buyers of consumer goods, intermediate and capital goods used in the industrialisation and growth processes. Having experienced persistent exchange rate depreciation the 11 developing countries provided a suitable case study to examine the link between exchange rate depreciation and trade volume.
The present study has assumed that exports and imports responded simultaneously to changes in NEXR, the inclusion of FDI effects, and the structures for growth-generating.

METHODOLOGY
The examination of the effect of exchange rate depreciation on trade is based on the model developed by Miles (1979), Bahmani-Oskooee (1985), and Rose and Yellen (1989). However, the present study has made some modifications in line with the arguments made in the last two paragraphs in the previous section.
This study has hypothesized that trade (TR) depends on the nominal exchange rate depreciation (NEXR), foreign direct investment (FDI), gross domestic product (GDP), and dummy variable (DUM). The study acquired aggregate data from the respective countries. The dependent variable, trade was the sum of the exports and imports of goods, not including services. It was measured by deducting trade (as a percentage of GDP) to trade in services (as a percentage of GDP), then multiplying it with nominal GDP. The NEXR was measured using the local currency units per the US dollar (USD). The other two variables, the FDI was measured using the net inflow of the FDI. The GDP was represented by the sum of value added by all its producers. All data was quoted from the World Bank Indicator and all variables were in the logarithmic form to obtain the elasticities of each variable. The model is shown in Equation (1): (1) An identifying structural break was conducted before estimating the models. It was because, as Narayanan and Smyth (2006) has argued, the existence of structural break in time-series analysis would distort the long-run equilibrium relationship between the dependent variable and independent variables, thus resulting in a misleading conclusion. The Chow test was applied, testing the null hypothesis of structural stability against the alternative of a onetime structural break specified in advance. The Chow F-statistic is written as Equation (2): (2) where RSS P = the sum of squared residuals from the combined data, RSS 1 = the sum of squared residuals from the first group data, and RSS 2 = the sum of squared residuals from the second group data. N 1 and N 2 indicate the number of observations in each group and k is the total number of parameters. This study rejected the null hypothesis of no break, i.e., the null hypothesis was rejected if the calculated F-value was greater than the F-critical value.
The autoregressive distributed-lag (ARDL)-bound test method was specified with the Akaike Information Criterion (AIC) lag-length. The process encompassed two steps. The first step was to consider the existence of a long-run relationship, in which the constant trend of the ARDL model was specified. The ARDL bound test was appropriate in cases where the nature of the stationarity of the data was not clear. The bound testing procedure was conducted to draw conclusive inference without knowing whether the variables were integrated as zero or one, I(0) or I(1), respectively (Pesaran et al., 2001). Having the dummy variable, the ARDL-bound test model is written as Equation (3): ( 3) where, is a constant, and is an intercept; the terms and represent the coefficient of long-run of the model respectively, the term is the optimum lag-length, DUM is dummy (DUM=0 for pre-structural change period and DUM=1 for poststructural change period), and is the error-term.
In order to capture the cointegration among TR, NEXR, FDI, GDP, and DUM, the null hypothesis, which stated that there was no long-run relationship: was tested against an alternative hypothesis implying the existence of cointegration. If the generated value of t statistic was signifcant (higher than the upper bound), the study would reject H 0 in favor of H 1 , capturing the long-run relationship between observed variables. The second step was to derive the Error-Correction Mechanism (ECM) approach for estimating both short-run and long-run effects of the dependent variable on explanatory variables. The ECM integrated the short-run dynamics with the long-run equilibrium without losing long-run information and avoided a spurious relationship resulting from non- (2) where, 0 is a constant, and 1 is an intercept; the terms 2 , 3 , 4 , and 5 represent the coefficient of long-run of the model respectively, the term ( , , , , ), is the optimum laglength, DUM is dummy (DUM=0 for pre-structural change period and DUM=1 for poststructural change period), and − is the error-term.
In order to capture the cointegration among TR, NEXR, FDI, GDP, and DUM, the null hypothesis, which stated that there was no long-run relationship: 0 : 1 = 2 = 3 = 4 = 5 = 0 was tested against an alternative hypothesis 1 : 1 ≠ 2 ≠ 3 ≠ 4 ≠ 5 ≠ 0 implying the existence of cointegration. If the generated value of t statistic was signifcant (higher than the upper bound), the study would reject 0 in favor of 1 , capturing the longrun relationship between observed variables. (2) where, 0 is a constant, and 1 is an intercept; the terms 2 , 3 , 4 , and 5 represen coefficient of long-run of the model respectively, the term ( , , , , ), is the opti length, DUM is dummy (DUM=0 for pre-structural change period and DUM=1 fo structural change period), and − is the error-term.
where, 0 is a constant, and 1 is an intercept; the terms 2 , 3 , 4 , a coefficient of long-run of the model respectively, the term ( , , , length, DUM is dummy (DUM=0 for pre-structural change period a structural change period), and − is the error-term.
where, 0 is a constant, and 1 is an intercept; the terms 2 , 3 , 4 , and 5 repr coefficient of long-run of the model respectively, the term ( , , , , ), is the length, DUM is dummy (DUM=0 for pre-structural change period and DUM= structural change period), and − is the error-term. ms 2 , 3 , 4 , and 5 represent the e term ( , , , , ), is the optimum lagchange period and DUM=1 for post-where, 0 is a constant, and 1 is an intercept; the terms 2 , 3 , 4 , and 5 repr coefficient of long-run of the model respectively, the term ( , , , , ), is the length, DUM is dummy (DUM=0 for pre-structural change period and DUM= structural change period), and − is the error-term.
intercept; the terms 2 , 3 , 4 , and 5 represent the l respectively, the term ( , , , , ), is the optimum lagor pre-structural change period and DUM=1 for postis the error-term. d 1 is an intercept; the terms 2 , 3 , 4 , and 5 represent the coefficient of ectively, the term ( , , , , ), is the optimum lag-length, DUM is dummy l change period and DUM=1 for post-structural change period), and − is ntegration among TR, NEXR, FDI, GDP, and DUM, the null hypothesis, s no long-run relationship: 0 : 1 = 2 = 3 = 4 = 5 = 0 was tested where, 0 is a constant, and 1 is an intercept; the terms 2 , 3 , 4 , and 5 represent the coeffici long-run of the model respectively, the term ( , , , , ), is the optimum lag-length, DUM is du (DUM=0 for pre-structural change period and DUM=1 for post-structural change period), and the error-term.
In order to capture the cointegration among TR, NEXR, FDI, GDP, and DUM, the null hypot which stated that there was no long-run relationship: 0 : 1 = 2 = 3 = 4 = 5 = 0 was against an alternative hypothesis 1 : 1 ≠ 2 ≠ 3 ≠ 4 ≠ 5 ≠ 0 implying the existen cointegration. If the generated value of t statistic was signifcant (higher than the upper bound), the would reject 0 in favor of 1 , capturing the long-run relationship between observed variables second step was to derive the Error-Correction Mechanism (ECM) approach for estimating both run and long-run effects of the dependent variable on explanatory variables. The ECM integrat short-run dynamics with the long-run equilibrium without losing long-run information and avo spurious relationship resulting from non-stationary time series data. Thus, the ARDL-ECM mode stationary time series data. Thus, the ARDL-ECM model is written as Equation (4): where, indicates the current value of the dependent variable (TR) as related to the past value of itself and simultaneously the past values of other regressors, is the intercept, the terms and represent the long-run dynamics of the model in respect of the ARDL (p, q, r, s, t) lag-lenght, DUM is dummy, and ECT is the error correction-term.
In the ARDL method, the term is replaced with and Next, it replaces the long-term with the residuals to revert the estimated model into error correction model. When the ARDL model included the same lagged levels as in the error correction model, it as implies that all the long-run variables (the ) are specified and not restricted. The significance of the ECT show the evidence of causality in at least one direction. The negative coefficient of indicates the speed of adjustment, i.e., the rate at which the model corrected its previous years of disequilibrium toward the long-run equilibrium. Finally, the study also employed diagnostic tests to ascertain the robustness of the estimated model.      Table 3 reports the optimal ARDL model for each country and the result of F-statisic of bound test for cointegration. The results showed that the value of F-statistic exceeded the upper bound at 5 percent and 10 percent significance level, indicating the long-run cointegration relationships between TR and NEXR for all countries. Next, Table 4 shows the coefficient of long-run relationship. The coefficient of NEXR was positive and statistically significant in seven out of 11 countries. In the long-run, the NEXR affected positively TR in Bangladesh, Ghana, India, Kenya, Pakistan, Swaziland, and Tunisia. The impact of NEXR on TR was negative in Sri Lanka and insignificant in Indonesia, Nigeria, and the Philippines.

RESULTS AND DISCUSSION
Related to the impact of FDI in the long-run, it was statistically insignificant in determining the trade in goods in most countries, excluding Nigeria, Pakistan, Philippines, and Swaziland. The GDP was positive and statistically significant at 5 percent level of significance in influencing the trade in goods in all countries, except Kenya and Pakistan. Meanwhile, the contribution of the dummy variable was negative in Ghana, India, Philippines and Sri Lanka. With regard to the issue of short-run relationships, see Appendix 1 which shows the estimation results of the coefficient of shortrun and error-correction term for each country. In sum, the study found negative signs of the coefficient of the error-correction term in all countries. The impact of the NEXR on TR was positive and statistically significant at a 5 percent level of significance, indicating that the NEXR influenced TR positively in the short-run in Bangladesh, Ghana, India, Indonesia, Nigeria, Pakistan, and Sri Lanka. The coefficient of the NEXR was negative and statistically significant in Kenya, the Philippines, Swaziland, and Tunisia.
The error correction term shows a negative sign and statistically significant at a five percent level for each country. A negative coefficient of the ECT implied that the speed of adjustment of the dependent variable was corrected from the disequilibrium. For example, in the case of Indonesia, the coefficient of -0.36 indicated a low-speed of convergence to equilibrium, e.g. the trade balance was corrected for about 36 percent every year to converge to long-run equilibrium. The coefficient of the ECT was as follows: From the estimation results of the short-run and long-run relationships, the trade volume in goods in Kenya, Swaziland, and Tunisia, was deteriorating in the short-run and improving in the long-run later. The impact of the nominal exchange rate depreciation on the trade was positive in the long-run and short-run for three South Asia countries, namely Bangladesh, India, Pakistan, and Ghana. However, in the case of Sri Lanka, the exchange rate had a positive and significant effect in the short-run, but a negative effect in the long-run. Furthermore, the study only found shortrun relationships in Indonesia, Nigeria, and the Philippines. In the short-run, the effect of the nominal exchange rate depreciation on the trade balance was positive in Indonesia and Nigeria, but there was a negative effect in the Philippines.
The above empirical findings were in line with what what was espoused by the elasticities approach, ML theory, and J-curve theory, which helped to explain the link between the trade balance and nominal exchange rate depreciation. According to the findings of the present study, the nominal exchange rate depreciation was able to explain the trade volume. The essence of the difference in these results was the way the trade variable was determined.
The findings of this study were also corroborated in earlier studies in the context of the J-curve phenomenon. The existence of a positive long-run equilibrium relationship between trade balance and exchange rate depreciation were consistent with ealier studies which mainly focused on the Asian-African countries. Among others these were studies by Bahmani-Oskooee (2001) for eleven middle eastern countries including Tunisia; Lal and Lowinger (2002) for Bangladesh, India, and Pakistan;and Caporale, et al. (2012) for Kenya. In the case of the Philippines, the present study found that real exchange rate depreciation impacted negatively the trade balance at two lags, but there was insignificant result in the longrun. By using the ARDL method and quarterly data, Harvey (2013) showed that the trade balance deteriorated before impoving in the long-run. In the case of Nigeria, this study showed that there was no long-run relationship between the trade balance and exchange rate, but the study found a positive short-run relationship among the variables. This evidence was in line with the findings in Danmola (2013) and Onakoya and Johnson (2018).
Furthermore, in the case of Ghana, the impact of the exchange rate changes on the trade balance was varied in the short-run, but positive in the long-run. The empirical evidence in this study provided robust support for the findings in previous studies by Bhattaraj and Armah (2013) and Iyke and Ho (2017). Finally, Rose and Yellen (1989) concluded that the depreciation of the nominal exchange rate did not necessarily lead to an increase in the trade balance, which was in this stufy found in the case of Sri Lanka.
In the context of the present study, the researchers were looking at the nominal exchange rate as one of the determining variables for international trade. The empirical evidence strongly pointed to the conclusion that the nominal exchange rate was a determinant of trade. Specifically, in the policy context, for the majority of countries, policymakers should maintain the nominal exchange rate depreciation as a part of trade strategy and liberalise foreign trade for the sake of the economic benefit of the country.
The study justified the use of diagnostic tests and CUSUM/ CUSUMSQ in pursuing the stability of the model. Table 6 shows diagnostic tests for the ARDL model in developing countries. The result of the Lagrange Multiplier test shows the estimated model has no serial autocorrelation. The Breusch-Pagan showed there was no heteroscedasticity occurring in the data, indicating the error term was homogenous in nature. The Jarque-Bera test found that the residuals of the estimated models were normally distributed. Subsequently, both CUSUM and CUSUMQ show the time-series was lying well within the 0.05 critical lines, indicating the parameter of the model was stable. For a graphical description, the CUSUM and CUSUM-SQ for each countries are provided in Appendix 2 and Appendix 3, respectively.

CONCLUSION
The present study has examined the impact of the exchange rate depreciation on international trade for a number of Asian-African developing countries. The ARDL model in the form of an unrestricted error-correction model was robust in explaining the nominal exchange rate depreciation as the determinant of trade. The study found that the nominal exchange rate depreciation had a significant positive impact on the trade balance in most countries in the long-run. The negative sign of the coefficient of the error-correction term in all countries supported the model convergence from the short-run toward the long-run, i.e. the existence of a long-run equilibrium relationship between trade balance and the independent variables. The CUSUM/ CUSUMSQ parameter stability tests confirmed the estimated models were stable over the period of the study for the respective countries.