THE IMPACT OF EXCHANGE RATE DYNAMICS ON THE TRADE BALANCE IN SIERRA LEONE: AN ARDL COINTEGRATION APPROACH

Transcription

1 THE IMPACT OF EXCHANGE RATE DYNAMICS ON THE TRADE BALANCE IN SIERRA LEONE: AN ARDL COINTEGRATION APPROACH Morlai Bangura, Crispin Denison-George and Robin B. Caulker 1 Abstract This paper examines the role of exchange rate in determining the short and long- run behavior of Sierra Leone's trade balance between 1980 and The econometric procedure employed is the bound testing approach to cointegration and error correction models, developed within an Autoregressive Distributed Lag (ARDL) framework.the result suggest that there is a stable long-run relationship between trade balance, real exchange rate, domestic income, foreign income and money supply variables. More importantly, the results reveal that domestic income and money supply are the main determinants of trade balance in Sierra Leone in the long run. Furthermore, the findings also suggest that Marshall-Lerner condition does not hold in the long-run for Sierra Leone. The short-run dynamic behaviour of Sierra Leone's trade balance has been investigated by estimating an error correction model in which the error correction term has been found to be correctly signed and statistically significant. The short-run result reveals that real money supply and a one period lag of real exchange rate have a negative significant effect on trade balance. It can be inferred from the short-run dynamics that the J-curve phenomenon is supported by Sierra Leone data. These findings are relevant in the sense that policy on Sierra Leone's trade balance should be viewed from the absorption and monetary approaches. JEL Classification: F12, F14, F31 Key Words: J-curve, trade balance, Marshall-Lerner Condition, cointegration, ARDL model, Sierra Leone. INTRODUCTION Despite voluminous literature, the effects of exchange rate changes on the balance of payments, particularly those related to the trade balance, are still to be understood clearly. Neither theoretical nor empirical work has established definitively whether a nominal devaluation of a country's domestic currency improves its trade balance, or even if exchange rates play a role in determining trade flows. These issues continue to be relevant to the understanding of the short and long-run relationship between those two variables, the formulation of policy, and the applied literature in international trade and finance. Earlier attempts to resolve this debate had led to the development of the elasticity approach (Robinson, 1947 and Meltzer, 1948) to trade balance. Precisely, this issue has been traditionally studied in the Marshall-Lerner (ML) conditions and the so called J-curve 1 Bangura, Denison-George and Caulker are staff of the Research Department, Bank of Sierra Leone. The views expressed in this paper are those of the authors and do not represent the official position of the Bank. 64

2 Morlai Bangura, Crispin Denison-George and Robin B. Caulker framework. According to the ML condition, currency devaluation or depreciation improves the trade balance in the long-run only if the sum of the absolute values of imports and exports demand price elasticities exceed unity. Junz and Rhomberg (1973) later on showed that following devaluation, trade balance is expected to follow a J-curve like pattern during the short-run. Reviewing the literature, some past studies found mixed evidence on the effect of exchange rate changes on trade balance. A plausible explanation for this is that different countries' characteristics offer different results. There are two strand of empirical research on the effects of exchange rate on the trade balance (balance of payments).those that found weak empirical support for significant long-run effect of devaluation/depreciation on trade balance as well as for J-curve effect. (see, Greenwood (1984), Rose and Yellen (1989), Rose (1991), Mahdavi and Sohrabin (1993), Wilson (2001), Wilson and Tat (2001), and those that reveal a favorable long-run effect of the depreciation on the trade balance with varying degree of J-curve phenomenon (Baharumshah (2001), Bahmani Oskooee (2001), Brahmasrene (2002), Kale (2001), Lal and Lowinger (2002), Onafowora (2003). Sierra Leone s external sector has been characterized by a persistent current account deficit since the mid-1970s, arising mainly from the continued deterioration in the trade balance. One approach to correcting this disequilibrium was to induce nominal depreciation, along with trade policy reforms, in order to achieve a real depreciation of the Leone, which was thought to be overvalued and hence affecting Sierra Leone s international competitiveness. In spite of these measures taken to bring about a favourable external balance position, the deteriorating trend continued. The trade balance as a % of GDP initially improved from an average deficit of % between to a surplus of 4.90% between 1985 and This positive developments in the trade balance was reversed in subsequent periods, with the trade balance worsening again to an average of -1.23% between This worsened to % between and then further deteriorated to %, the worst external situation in decades.it average % between The interesting question that therefore arises from these observations in Sierra Leone is; given the existence of real exchange rate depreciation coupled with deficit in the trade balance over the past decades; does real exchange rate play a role in determining the trade balance in Sierra Leone? From an economic policy perspective, the motivation for the analysis of the impact of the exchange rate changes on the trade balance in Sierra Leone is as follows: The fundamental question of whether the exchange rate depreciation can improve the trade balance needs to be quantified to determine whether there exists a stable long -run relationship between the exchange rate and the trade balance. As Stucka (2003) points out, if a stable long-run relationship exits, it is beneficial to quantify the degree of trade balance improvements in order to weigh the trade balance benefits against the costs of permanent depreciation, such as higher exchange rate pass-through to domestic prices, leading to a permanent rise in inflation. Similarly, answers to these questions are important 65

3 Vol. 13, No.1 Journal of Monetary and Economic Integration to policy makers to the extent that their answers can help provide viable solutions to the imbalances in the trade balance in Sierra Leone. The primary objective of this paper is to examine the role of exchange rates in determining short- and long-run trade balance behaviour for Sierra Leone. Specifically, this paper seeks to investigate whether the Marshall-Lerner (ML) condition and J-curve exist, both for the period As far as we know, there is only one paper (Adeniyi etal, 2011) which has investigated the J-curve effect in Sierra Leone. This paper supports a long-run relationship between trade balance, real exchange rate, foreign income and domestic income for the Sierra Leone economy during the period 1980 to 2007, but failed to incorporate monetary variable. In this paper, we test not only the validity of the ML condition but also the existence of the J-curve pattern for the Sierra Leone economy using most recent available data to capture the current dynamics in the model. In addition, we have attempted to test the empirical relevance of the monetarist approach using the Autoregressive Distributed Lag (ARDL) approach to cointegration and error correction models by incorporating money supply variable in the model. Following this introduction, the rest of the paper is structured as follows: Section two explore the trends in real exchange rate and trade balance in Sierra Leone over the sample period. Section three reviews the theoretical and empirical literature on the effects of exchange rate changes on the trade balance. Section four develops the empirical model and describes ARDL approach to estimating the long-run and short-run determinants of trade balance. It also discusses the data and sample characteristics. The empirical findings are discussed in section five. The final section concludes and presents policy recommendations. REAL EXCHNAGE RATE AND TRADE BALANCE DEVELOPMENTS IN SIERRA LEONE Sierra Leone has been plagued with persistent current account deficits since the mid-1970s, arising mainly from the persistent deterioration in the trade balance, which provoked discussion among policy makers and academics on the potential causes of these imbalances. These imbalances over the decade were in part due to poor export performance, huge services account deficits, external debt service payments, low foreign direct investment, excessive domestic monetary and credit expansion, large fiscal deficit, price distortion and a deterioration in the terms of trade (IMF1977). Trade balance developments in the pre-reform era suggest that the trade balance was in deficit since 1980, reached its peak in 1981(-16.72% of GDP).In subsequent years, a sustained improvement was observed reaching a surplus of 14% of GDP in 1988.However, on a global scale the period experienced an average deterioration in the trade balance of -4.09% of GDP. 66

4 Morlai Bangura, Crispin Denison-George and Robin B. Caulker In a bid to correct the disequilibrium in the external sector, the government of Sierra Leone in 1989, embarked on trade and exchange rate reforms as part of a broad Structural Adjustment Programme (SAP). The approach to correcting the over-valuation of the Leone was to induce a nominal devaluation of the Leone, along with trade reform policies, in order to achieve the needed real depreciation of the Leone. The result was sustained and massive annual depreciation of the Leone, following the Structural Adjustment Programme. The first year of reform saw the depreciation of nominal exchange rate by about 150%, while the real exchange rate depreciated by about 25%. Over the period , the nominal exchange rate depreciated against the US dollar on average by 100.3%, while the real effective exchange rate depreciated by about 11.74% on average. A significant improvement was achieved in the trade balance (surplus of 6.25% of GDP) in 1989, only to be reversed starting 1990 though marginally by % of GDP between Despite the liberal trade and exchange rate policies introduced under the SAP, the balance of payments position continued to worsen. In 1992, a flexible exchange rate regime was introduced. This was in itself a desperate attempt to resolve the problem of persistent macroeconomic imbalances(external and internal).the rate of depreciation was moderate during the exchange rate liberalization era , when on average, the nominal exchange rate depreciated by about 12.8% per annum while the real effective exchange rate appreciated by about 0.18%. Despite the adoption of the managed floating exchange rate regime and the continuous depreciation of the nominal exchange rate during this regime, external sector performance in Sierra Leone remained unsatisfactory. During the flexible exchange rate regime, trade balance deteriorated from -4.58% of GDP between to % between 1997 and It then attained its climax between 2002 and 2006 (13.46%) before moderating to % of GDP between From the afore mentioned analysis, it is observed that Sierra Leone did not fully succeed in adjusting its export structure to new demand, whilst huge imports were necessary to satisfy the domestic demand. Figure 1 shows the trends in real exchange rate changes and trade balance as a percentage of GDP in Sierra Leone. Real exchange rate appreciated continuously from the early 1980s reaching its peak in 1984, this period coincided with a sustained improvement in the trade balance. This was followed by a continuous depreciation of the real exchange rate between the periods , a period of improved trade balance as a percentage of GDP. From the real exchange rate exhibited an appreciating trend; however the trade balance continued to improve. The real exchange rate was fluctuating over the period before registering a sustained appreciation over the period from , a period that recorded a sustained trade deficit as a percentage of GDP. Recent trend indicates that the real exchange rate is on a depreciating path, though the trade balances continue to be in deficit. The discussion above shows that real exchange rate and trade balance in Sierra 67

5 Vol. 13, No.1 Journal of Monetary and Economic Integration Leone do not exhibit any clear discernable pattern. It is therefore important to investigate the role of the real exchange rate in the determination of the trade balance in Sierra Leone. Figure1. Real exchange rate changes and Trade Balance as a % of GDP trade deficit/gdp(%) REER(%) Source: World Development Indicator Database 2011 and authors calculation. REVIEW OF THEORITICAL AND EMPIRICAL LITERATURE Theoretical Literature Within the body of literature on international trade, it is not surprising to still find arguments about whether currency depreciation or devaluation will improve the balance of trade. The three main approaches identified in the literature are the elasticities, absorption and monetary approaches all bare testimony to this with each having its own arguments. The elasticity approach was first proposed by Bickerdike (1920); Robinson (1947) and Metzler (1948) and popularized by Kreuger (1983) dubbed the BRM model. This BRM model is a partial equilibrium version of a standard two-country (domestic and foreign), two-goods (exports and imports) model, where the effects of exchange rates changes are analyzed through the separation of markets for exports and imports. The elasticity approach focuses on the trade balance. It is built on the Marshall-Lerner condition, which states that the sum of elasticity of demand for a country s exports and imports must be greater than unity for a devaluation to have a positive effect on a country s balance of 68

6 Morlai Bangura, Crispin Denison-George and Robin B. Caulker payments. If we assume the elasticity of exports to be zero, in this case, exports in domestic currency are the same as before the devaluation. If the sum of these elasticities is greater than one, the elasticity of demand for imports must be greater than one, so that the value of imports falls. With no fall in the value of exports and a fall in the value of imports, the balance of trade will improve. Suppose now the demand for imports has zero elasticity. The value of imports will rise by the full percentage of devaluation. If the elasticity of demand for exports is greater than unity, the value of exports will expand by more than the percentage of devaluation. Hence, the trade balance will improve. If each element of the elasticity of demand is less than unity, but the sum is greater than unity, the trade balance will improve because expansion of exports in domestic currency will exceed the value of imports. Absorption approach due to Harberger (1950), Meade (1951), Alexander (1952) and popularized by Miles (1979) was developed to overcome some of the shortcomings of the elasticity approach. The major purpose of the absorption approach is to integrate the balance of payments with the functioning of the total economy in a general equilibrium framework, in which balance of payments disequilibrium on current account is viewed as the outcome of the difference between decisions to produce and spend, or to save and invest. The core of this approach is the proposition that any improvement in the trade balance requires an increase of income over total domestic expenditures. The monetarist approach (Polak, 1957; Mundell, 1971; Durnbusch, 1973; Frenkel and Rodriequesz, 1975) emerged at the end of the 1950s and considers balance of payments as essentially a monetary phenomenon. The fundamental basis of the monetary approach to the balance of payments is that the balance of payments is a monetary phenomenon and not a real phenomenon. It is argued that any disequilibrium in the balance of payments is a reflection of disequilibrium in money markets. In very simple terms, if people demand more money than is supplied by the Central Bank then excess demand for money would be satisfied by inflows of money from abroad and hence an improvement in the trade balance. On the otherhand, if the Central Bank is supplying more money than is demanded, the excess supply of money is eliminated by outflows to other countries and this will worsen the trade balance. Three key assumptions that underlie the monetary model are the stable money demand function, vertical aggregate supply schedule and purchasing power parity. It can be seen from the theoretical review that a country s balance of trade will be affected by changes in domestic income level, foreign income, money supply and exchange rate. Thus, the present study develops a model that incorporates all three approaches to empirically verify their relevance and validity in Sierra Leone. 69

7 Vol. 13, No.1 Journal of Monetary and Economic Integration Empirical Literature Many empirical analyses, both multi -country panel regressions and econometric models applied to individual countries, have been conducted to show how exchange rate changes affect the trade balance of developing and developed countries. Despite these large volumes of theoretical and empirical researches into how exchange rate changes affect trade balance, there is still considerable disagreement concerning the relationships between these economic variables and the effectiveness of currency devaluation as a tool for increasing a country's balance of trade (Onafowora, 2003). In the case of the developed countries, Haynes and Stone (1982) estimated the impact of terms of trade on the U.S trade balance. Their results showed no improvement in the trade balance following a deterioration of the terms of trade for the period between 1947 and In a similar vein, the empirical findings of Rose and Yellen (1989) and Bahmani-Oskooee and Brooks (1999), with disaggregated bilateral trade data and autoregressive Distributed Lag approach, do not support the existence of J-curve effect for USA, although the later study reports a long-run positive relationship between devaluation and trade balance. Gupta-Kapoor and Ramakrishnan (1999) used the error correction model and the impulse response function to determine the J-curve effect on Japan using quarterly data from 1975:1-1996:4. Their analysis showed the existence of the J-curve on the Japanese trade balance. In a recent study, Hsing and Sergi (2010) apply the vector error correction model and generalized impulse response function to establish that in the case of the US and six of its western European trading partner countries, there is support for the J-curve hypothesis with regards to Germany, but find no evidence in the case of Belgium, France, Italy, Spain and the Netherlands. Some studies have also been conducted on the impact of exchange rate on trade balance in developing countries. Onafowora (2003), also examined the short run and long run effects of real exchange rate changes on the real trade balance of three Asian countries in their bilateral trade with Japan and USA and found improvement in their trade balance but with time lag. Stucka (2003) using a reduced form model to estimate the impact of a permanent shock on the merchandise trade balance, supports evidence of J-curve on trade balance for Croatia. Rabeya Khatoon (2009), using cointegration technique to estimate the impact of the taka depreciation on the trade balance in Bangladesh found no evidence of J-curve effect. Similar results were observed by Duasa, J. (2007) using the ARDL approach to cointegration on Malaysian data. However, Waliullah etal.(2010) in a bid to determine the effect of exchange rate on trade balance in Pakistan using the ARDL approach for the period 1970 to 2005, found evidence of a long-run positive relationship between exchange rate and trade balance. Many studies have also been conducted on the effect of exchange rate on trade balance in Sub-Saharan African countries. The study by Rawlins and Praveen (1993), examined the 70

8 Morlai Bangura, Crispin Denison-George and Robin B. Caulker impact of devaluation on trade balance of a sample of 19 countries in Sub-Saharan Africa by specifying and estimating an Almon Distributed lag process of trade balance using annual data. They found in no case did real exchange rates revert to their pre -devaluation levels and in seventeen of nineteen countries real exchange rate depreciation did improve a country's trade balance in the year of the devaluation. Damoense and Agbola (2004) came with evidence that supports the view that devaluation of exchange rate worsens trade balance. In their study of the impact of devaluation on trade balance of South Africa, they found that in the long run, devaluation of exchange rate worsens trade balance. Similarly, the empirical study by Agbola (2004), by using the Johansen multivariate cointegration procedure and the Stock-Watson dynamic Ordinary Least Square model (DOLS), revealed that devaluation did not improve the trade balance of Ghana. Bhattari and Armah (2005) studied the effects of real exchange rates on the trade balance in Ghana using cointegration analyses of both single equation models and vector-error correction model. They concluded that a long-run relationship existed between both exports and imports and the real exchange rate and that the short-run elasticities of exports and imports indicated contractionary effects of devaluation in terms of Marshall-Lerner-Robinson condition. Omojimite etal (2010) studied the effect of exchange rate reforms on Nigeria s trade performance during the period using Ordinary Least Square and General Methods of Moments. They found a small positive effect of exchange rate reforms on non-oil exports through the depreciation of the value of the country s currency. Drama Bedi Guy Herve etal (2010) examined the effect of real exchange rate on the trade balance of Cote d Ivoire for the period using multivariate cointegration and vector error correction models.they result confirmed the existence of Marshall-Lerner condition through the J- curve. Umoru and Oseme (2013) analyzed the J-curve effect based on Nigeria data by adopting the vector error correction methodology. They found no support for the J-curve hypothesis in the Nigeria trade balance. Danmola etal (2013) examines the validity of J-curve Hypothesis in the Nigerian economy; the study employs cointegration, Vector Auto regression Estimate, Granger Causality and Variance Decomposition to analyze the hypothesis. the study shows that there is absence of long-run relationships among variables under consideration but the study found short run relationship between exchange rate devaluation and trade balance through Granger causality test and therefore confirming the existence of J-curve hypothesis but with little effect in the long-run. Even though this large body of literature exists on the effect of exchange rate on trade balance in both developed and developing countries, studies on the Sierra Leone economy are very sparse and virtually non-existent. A deep scan through the literature reveal only one study carried out in recent times aimed at determining the effect of exchange rate on trade balance. Adeniyi etal (2011), queried the existence or otherwise of a J-Curve in four West African countries( The Gambia, Nigeria, Ghana and Sierra Leone).Using quarterly data from 1980 to 2007 with bound testing approach to cointegration, they concluded that the J-curve hypothesis is supported only for Nigeria with Sierra Leone exhibiting no J- 71

9 Vol. 13, No.1 Journal of Monetary and Economic Integration Curve pattern. Our study differs from this study, as this study failed to incorporate monetary variable in the model. This paper therefore has extended the frontier coverage of the studies by testing for the relevance of the elasticity, absorption and monetary approaches to the Sierra Leone economy by incorporating money supply in the model. MODEL SPECFICATION, ARDL APPROACH AND DATA Model Specification Both theoretical and empirical literatures propose a number of key variables that have significant effects on export and import and hence on trade balance. Trade balance is usually measured as the difference between receipts for exports of goods and expenditure for imports of a nation during a specific period of time. As we know, Sierra Leone is small open economy in the sense that it cannot influence prices in international market, imports prices are given in the world market and the prices are independent of the imports volume. Thus the imports demand depends on mainly real domestic income. However, the demand for imports can also be determined by real exchange rates. The most common definition of the real exchange rate take into account the prices ratio of tradable to non-tradable goods. Hence the real exchange rate was constructed as follow: e p * RER (1) P Where e denotes the nominal exchange rate, p and p* express respectively domestic price denotes foreign price. We assume also that Sierra Leone is small open economy and price taker in the global market. Trade balance is usually measured as the difference between receipts for exports of goods and expenditure for imports of a nation during a specific period of time. Based on the works of Rawlins and Praveen (1993) and Agbola (2004), trade balance can be specified as: TB X M X PX *, Y M P. M e. Y (2) e Trade therefore the functional form of the import and export demand models is written as follow: p X t. Y * t (3) * e p 72

10 Morlai Bangura, Crispin Denison-George and Robin B. Caulker e p * M t. (4) Y t p Where X and M represent the volume of exports and imports, e denotes the nominal exchange rate and P, p * and Y, Y * express the domestic and foreign price levels and incomes respectively, and are the real exchange rate elasticities for exports and imports then and measure the income elasticities for imports and exports. As illustrated above, the model estimates elasticity then, if we incorporate natural logarithm on both sides, our export and import demand models can be written in this form: * * - (5) ln X t ln Pt ln Pt ln et ln Y t * - (6) ln et ln M t ln P ln Pt ln Y t Here, ln P * t ln e t ln P t measures the natural logarithm of real exchange rate. If we consider that the trade balance TB is defined as the ratio between exports and imports we have: * ln TB ln Y t ln Y t ln et - (7) Where and the coefficient indicates whether the ML conditions is fulfilled, then and are assume to be negative. Precisely, and are positive thus whenever 0 mean that a depreciation or devaluation of real exchange rate appears to improve trade balance over time. In another words, an increase of real exchange rate will improve the balance of trade. Incorporating monetary variable that affect trade balance into equation (7) in line with economic theory and recent empirical investigation, this study uses modified form of equation (7) specified as: lntbt 0 1ln RGDPt 2 ln REERt 3lnUSRGDPt 4 ln RM2t 5WAR t (8) Where ln is natural logarithm, TB is trade balance in Sierra Leone, RGDP is real gross domestic product a proxy for real domestic national income, RM2 is real domestic money 73

11 Vol. 13, No.1 Journal of Monetary and Economic Integration supply, USRGDP is real gross domestic product of US as a proxy for world real income, REER is real effective exchange rate index, WAR is war dummy to capture the effect of the civil war from and t is the random error term. Following classical theory, the sign of could be positive or negative. If the estimate 1 of is negative, it means that an increase in real income in will increase imports volume. 1 However, if the estimate is positive, it means that an increase in domestic real income is due to increase in the production of import-substituted goods. The effect of changes in real effective exchange rate on balance of trade is ambiguous. Thus, could take any 2 sign, positive or negative. In general, if real depreciation/devaluation occurs, which causes the real effective exchange rate to increase, the exports go up, the imports fall as a consequence and it improves the trade balance. The sign of could be either positive 3 or negative depending on whether supply side factors dominate demand side factors. Even though there are differences on the rationale between schools of thought, there is a consensus that the signs on should be negative. According to the Monetarist view, 4 increases in the money supply propel real balances above levels considered optional by economic agents, resulting in increased expenditure out of a given income thus stimulating imports and causing the trade balance to deteriorate. For Keynesians, increases in the money supply reduce interest rates thus stimulating increased absorption which puts negative pressure on the trade balance. Estimation Technique The study employs the Autoregressive Distributed Lag (ARDL), the bounds testing approach to cointegration proposed by (Pesaran etal 1997). This method has certain econometric advantages over the Engle- Granger (1987) framework and the multivariate cointegration framework proposed by Johansen and Juselius (1990) and Johansen (1991). First, the bounds framework does not require pre-testing the series to determine the order of integration since the test can be done regardless of whether the variables are purely I(0), I(1), or mutually integrated. Second, endogeneity problems and inability to test hypotheses on the estimated coefficients in the long run associated with the Engle-Granger method are circumvented. Third, the ARDL framework incorporates a sufficient number of lags to capture the data generating process general to specific modeling framework (see Kargbo and Adamu, 2009).Fourth, the error correction model (ECM) can be derived from ARDL through a simple linear transformation (Banerjee etal 1998). The ECM integrates the short run adjustments with long run equilibrium without losing long run information. Moreover, 74

12 Morlai Bangura, Crispin Denison-George and Robin B. Caulker small sample properties of ARDL approach is far superior than that of Johansen and Juselius s cointegration approach (Pesaran and Shin 1999). In addition the ARDL procedure can distinguish between dependent and explanatory variables. Consequently, in the ARDL framework, estimation is possible even when the explanatory variables are endogenous (Pesaran etal., 1997). This is important in the current context since Real exchange rate, real domestic income, real foreign income or Money supply may be an endogenous variable. An ARDL representation of equation (8) is specified as follows: lntbt 0 1lnTBt 1 2 ln RGDPt 1 3 ln REERt 1 4 ln RM 2t 1 p p p 5 ln USRGDPt 1 1WAR 1,i ln TBt 1 2,i ln RGDPt 1 3,i ln REERt 1 i 1 i 1 i 1 p p 4i ln RM 2 t 1 5i ln USRGDP t 1 t (9) i 1 i 1 Where is the first difference operator and all other variables as earlier defined. The long-run relationship among the variables in equation (9) is estimated by means of bounds testing procedure of Pesaran etal The first step of the ARDL bounds testing framework is to estimate equation (9) by Ordinary Least Squares (OLS) in order to test for the existence of a long run relationship among the variables by conducting an F-test for the joint significance of the coefficients of the lagged levels of the variables, that is: the null hypothesis of no cointegration relationship, defined as H0 : is tested against the alternative hypothesis of the existence of cointegrating relationship defined as H1 : The cointegration test is based on the F-statistics or Wald statistics. Given a relatively small sample size in this study of 32 observations, the critical values used are as reported by Narayan (2004) which is based on small sample size between 30 and The test involves asymptotic critical value bounds, depending whether the variables are I(0) or I(1) or a mixture of both. Two sets of critical values for the cointegration test are generated. The lower critical bound assumes that all the variables are I(0), meaning that there is no cointegration among the variables, while the upper bound assumes that all the variables are I(1). If the computed F-statistic is greater than the upper critical bound, then the null hypothesis of no cointegration will be rejected. Conversely, if the computed F-statistic falls below the lower critical bounds value, and then the null hypothesis of no cointegration is accepted. The test is however inconclusive if the computed F-statistic lies between the lower and upper bounds. In this case, unit root tests 2 Pesaran and Pesaran(1997) and Pesaran et al (2001), however, generated critical values based on 500 and 1000 observations and 20,000 and 40,000 replications, respectively, which is suitable for large sample size. 75

13 4,i ln RM 2 5,i lnusrgdpt 1 ECM t 1 WAR t Vol. 13, No.1 Journal of Monetary and Economic Integration should be conducted to determine the order of integration of the variables. If all the variables are found to be I(0), then the decision is taken on the basis of the lower critical bound value. On the other hand, if all the variables are found to be I(1), then the decision is taken on the basis of the upper critical value. Once a long run relationship is established, the long run and error correction estimates of the ARDL model can be obtained from Equation (9). The parameter stability test for the appropriately selected ARDL representation of the error correction model can also be performed at this stage.a general error correction representation of Equation (9) is specified as: p p p lntbt c0 1,i TBt i 2,i ln GDPt 1 3,i ln REERt 1 i 1 i 1 i 1 p i 1 t 1 p i 1 (10) Where ECM is the residual that is obtained from the estimated cointegration model,,s are the short run dynamic coefficients of the model s convergence to equilibrium, and is the speed of adjustment. To ascertain the goodness of the fit of the ARDL model, the diagnostic and stability tests are conducted. The diagnostic test examines the serial correlation, functional form, normality and heteroscedasticity associated with the model. The stability test of the regression parameters is undertaken using the Brown et al. (1975) stability testing technique, also known as cumulative sum of recursive residuals (CUSUM) and the cumulative sum of squares of recursive residuals (CUSUMQ). Data and Sample Characteristics For estimation purposes, we use the logarithmic transformation of annual data for the period 1980 through 2011, sourced from the World Development Indicator (WDI) data base 2011.The time series include observed values of exports, imports, a real effective exchange rate index, broad money (M2), the real gross domestic product and US real gross domestic product as a proxy for foreign income. The measure of trade balance (called TB) is represented by the ratio of exports to imports.this ratio, or its inverse, has also been used in similar settings by Haynes and Stone (1982), Bahmani-Oskooee and Alse (1994). The use of this ratio has several advantages. First, it is invariant to unit of measurement and can be interpreted as the nominal or real trade balance. Second, the regression equations can be expressed in log-linear form or constant elasticity form. Accordingly, the estimated coefficients are elasticities. 76

14 Morlai Bangura, Crispin Denison-George and Robin B. Caulker EMPIRICAL FINDINGS Unit root test Before we proceed with the ARDL bound test, we test for the stationarity status of all variables to determine their order of integration. This is to ensure that the variables are not I(2) stationary so as to avoid spurious results. Therefore, the implementation of unit root tests in the ARDL procedure might still be necessary in order to ensure that none of the variables is integrated of order 2 or beyond. Both the augmented Dickey-Fuller (ADF) (1979) and Phillips-Perron (PP) (1988) unit-roots tests are employed for this purpose. The null hypothesis under both tests is that the time series are generated by unit root processes. Table1 indicate that all of our variables are I(1) except the trade balance(tb) which is I(0) from the ADF. This suggests that an ARDL approach is the appropriate method of estimation of the relationship since the variables are a mixture of I(1) and I(0). Table 1: Unit-Root Test (Augmented Dickey-Fuller and Phillips-Perron) ADF test statistics based on SBC P-P test statistics Variable Level First Difference Level First Difference LRGDP * * LREER ** ** LM * * LTB ** * * LUSRGDP ** * * Notes: The superscripts * and ** denotes rejection of the hypothesis of a unit root at 1% and 5% significance levels respectively Bounds tests for cointegration In the first stage, the order of lags on the first-differenced variables for Equation (9) is obtained from unrestricted vector autoregression (VAR) by mean of Akaike Information Criterion (AIC). Given the limited number of observations and the use of annual data, we experimented up to 2 lags on the first difference of each variable and computed the F-statistic for the joint significance of the lagged levels of variables in equation (9). A lag length of 2 years was chosen and used in the bounds testing framework. This is consistent with Pesaran and Shin (1999), who suggest a maximum of 2 lags for annual data. Following the procedure in Pesaran, M.H and B. Pesaran (1997), we first estimated an OLS regression for the first differences part of equation (9) and then test for the joint significance of the parameters of the lagged level variables when added to the first regression. According to 77

15 Vol. 13, No.1 Journal of Monetary and Economic Integration Pesaran, M.H and B. Pesaran (1997)," this OLS regression in first differences are of no direct interest" to the bounds cointegration test. The F-statistics tests the joint null hypothesis that the coefficients of the lagged level variables are zero (i.e. no long-run relationship exists between them).table2 reports the results of the calculated F-statistics when each variable is considered as a dependent variable (normalized) in the ARDL-OLS regressions. Table 2: Results from bounds tests on equation (2) Dep. Var. AIC Lags F-statistic Probability Outcome FLTB(LTB/LRGDP,LREER, Cointegration LRM2,LUSRGDP) FLRGDP(LRGDP/LREER,LR No Cointegration M2,LTB,LUSRGDP) FLRM2(LRM2/LREER, LTB, LRGDP,LUSRGDP) Cointegration FLREER(LREER/LRM2,LTB, Cointegration LRGDP,LUSRGDP) FLUSRGDP(LREER/LRM2,LT No Cointegration B,LRGDP,LUSRGDP) Notes: Asymptotic critical value bounds are obtained from Table F in appendix C, Case II: intercept and no trend for k=4 (Narayan 2004). Lower bound 1(0) = and Upper bound 1(1) = at 1% significance level. Lower bound 1(0) = and Upper bound 1(1) = at 5% significance level. Lower bound 1(0) = and Upper bound 1(1) = at 10 % significance level. The calculated F-statistic FTB(LTB/LRGDP, LREER,LM2,LUSRGDP)= is higher than the upper bound critical value at the 10 % level. Also FLREER(LREER/LM2, LTB, LRGDP, LUSRGDP)= is also higher than the upper bound critical value at the 5% level and FLRM2(LRM2/LREER, LTB, LRGDP,LUSRGDP)= is also higher than the upper bound critical value at the 10% level. Thus the null hypothesis of no cointegration is rejected, implying long-run cointegration relationships amongst the variables when the regressions are normalized on LTB, LRM2 and LREER variables (Table2). However, based on the model specified, we used LTB as the dependent variable. This suggests the existence of a long-run relationship between trade balance and real exchange rate, broad money real domestic income and real foreign income. The existence of a cointegration relationship is not found when the other variables are taken as dependent variables since the resulting F-statistic is less than the lower critical bound as indicated in Table 2. 78

16 Morlai Bangura, Crispin Denison-George and Robin B. Caulker The existence of a cointegrating relationship between the trade balance and a vector of explanatory variables suggests the estimation of both the long run coefficients and the short run dynamic parameters. Following Liew (2004) 3 the ARDL model is estimated based on the Akaike Information Criterion (AIC). The ARDL static long-run results and the diagnostic test statistics of the estimated model are presented in Table 3. Table 3: Long-run estimates based on AIC-ARDL (2,2,1,0,0) Dependent Variable (LTB) Regressor Coefficient t-values p-values Panel A: Long run results LRM LREER LRGDP LUSRGDP C R-Square= F-Stat(9,20) =5.2370(0.001) Durbin Watson= R-Bar Square= Diagnostic Test statistics A: Normality 2 (2) = [0.884] B: Serial correlation F(1,19) = [0.136] C: Heteroscedasticity F(1,28) = [0.634] D: Functional Form F (1,19) = [0.118] E: Stability F:Recursive Estimates CUSUM [Stable] CUSUM Sum of Squares [Stable] 3 Liew(2004) argues that Akaike Information Criterion(AIC) and Final Prediction Error(FPE) criteria have superior properties in small sample estimations since they minimize the chance of underestimation while maximizing the change of recovering the true lag length. 79

17 Vol. 13, No.1 Journal of Monetary and Economic Integration The empirical results of the long-run model shows that both money supply (LRM2) and domestic income (LRGDP) are important determinants of trade balance in Sierra Leone. The sign of the coefficients of money supply and domestic income are consistent with prior expectation and are both statistically significant at 5% level of significance. The result suggests that money supply and domestic income are the key variables affecting trade balance from this long- run link. The positive sign on the domestic income variable is not in agreement with what was expected by proponents of the absorption approach, however, this result supports the monetarist view. The theory indicates that a rise in domestic income increases the demand for money and will therefore increase exports and improve the trade balance. The estimated coefficients suggest that a 1 % increase in real income yields an average 1.41 % improvement in the trade balance. This positive sign agrees with the findings by Waliullah etal (2010) who got similar result for Pakistan. Similarly, a 1% increase in real money balances leads to an average 0.69 % deterioration in trade balance. This result corroborates the monetary approach to trade balance and is consistent with result obtained by Korsu and Nduka (2010). Theory suggest that as money supply increase, without a corresponding increase in domestic output, prices for non-tradables will increase causing domestic inflation. This will lead to an expenditure switching behaviour in favour of tradables (imports).hence increasing imports demand will worsen the trade balance. The estimated long-run exchange rate elasticity has the apriori sign but is statistically insignificant. Therefore, a real depreciation would not lead to an improvement in the trade balance in the long-run, thus indicating that the Marshal -Lerner condition does not hold in the long-run in the case of Sierra Leone. These results are somehow consistent with a study by Liew etal (2003) which found that the role of exchange rate is rather insignificant in initiating changes in the trade balance in the case of Malaysia, Singapore, Thailand and the Philippines. The result of the long-run relationship establishes that domestic income and money supply are the main determinants of trade balance in Sierra Leone. The dummy variable accounting for the civil war (WAR) was dropped because it was found to be insignificant though it had the expected negative sign. In order to ensure the model is an appropriate model, diagnostic tests were carried out. These tests include the test for serial correlation, heteroscedasticity, normality, and functional form. Based on the diagnostic test results, the estimated model is well fitted, follows the correct functional form and is stable (table3). Short-run Dynamics The results of Error Correction Model (ECM) associated with the ARDL (2, 2, 1, 0, 0) are presented in Table 4.This will enable us analyze the J-curve phenomenon for Sierra Leone. The result shows that the error correction term (ect) is highly significant at 1% level of significance. The speed of adjustment implied by the ECM coefficient is 75%. The short 80

18 Morlai Bangura, Crispin Denison-George and Robin B. Caulker run model provides information on how the dependent variable, trade balance adjusts to restore long-run equilibrium in response to the disturbances. Approximately, 75% of disequilibria from the previous year s shock converge back to the long- run equilibrium in the current year. The coefficients of dlrm2 and dlreer1 are statistically significant with expected signs for the short-run dynamics. Table 4: Error Correction representation of the ARDL Model: ARDL(2,2,1,0,0) selected based on Akaike Information Criterion Dependent Variable (LTB) Regressor Coefficient t-values p-values dltb dlrm dlreer dlreer dlrgdp dlusrgdp Constant ECMt R-squared Meanofdependent variable Adjusted R-squared S.D. dependent variable S.E.. of regression Akaike info criterion Sum squared residuals Schwarz criterion Log likelihood Durbin-Watson statistic F-statistic [0.021] The result suggests that real exchange rate lagged one period is statistically significant at 5%. Indicating that 1% depreciation of the real exchange rate in the previous period will worsen the current trade balance by 0.50 %.Hence suggesting that some evidence of the J-curve short-term trade balance adjustment was found for Sierra Leone. The result is consistent with the findings by Danmola etal (2013) for Nigeria, but contradict the findings by Adeniyi etal (2011) for Sierra Leone. Real money supply remains consistently negative in the short-run. The magnitude of the estimate revealed that a 1% increase in real money supply is likely to worsen Sierra Leone s trade balance by 0.52 %. Stability Tests 81

19 Vol. 13, No.1 Journal of Monetary and Economic Integration To ascertain whether the results obtained in the model can be of any use for policy analysis, we carried out coefficient stability test of the standard and the ECM model using the cumulative sum (CUSUM) test and the cumulative sum of squares test (CUSUMSQ). Figure 2(appendix1) presents the chart result of the CUSUM test, while Figure 3(appendix1) presents the result of the CUSUM of squares test. To read the chart, one simply needs to observe the movement of the cumulative sum of squares, if it lies within the 95% confidence interval, it suggest that the parameters are stable, and hence the estimates can be used for policy analysis. Using the two competing tests (i.e. CUSUM and CUSUMSQ), the sum of squares lies within the bands of 95% interval, indicating that the coefficients are stable. CONCLUSION AND POLICY RECOMMENDATIONS The objective of this study was to examine empirically the role of exchange rate in determining the short and long-run behaviour of Sierra Leone's trade balance. In particular, it tested the validity of the Marshall-Lerner (ML) conditions using a regression model which included the trade balance, exchange rate, money supply, and income. Indirectly, it tested the empirical relevance of the absorption and monetary approaches for the data used. The econometric technique employed is the bound testing Autoregressive Distributed Lag (ARDL) approach to cointegration to examine the longrun relationships between trade balance, real effective exchange rate, broad money, real foreign income and real gross domestic income using Sierra Leone as the case study for the period We find evidence of a cointegration relationship among the variables when the trade balance, real money supply and real exchange rate are the dependent variable. However, based on the model specification, we used the trade balance as the dependent variable. This allows us to examine the long and short-run elasticities of Sierra Leone's trade balance for policy implications. The major findings of this paper is that trade balance in Sierra Leone is highly responsive to money supply and domestic income in the long-run. The result further suggests that the Marshall-Lerner condition is not fulfilled in the case of Sierra Leone. A long-run relationship was also found to exist between trade balance and its determinants, indicating that the model has self-adjusting mechanism for correcting any deviation of the variables from equilibrium. In the short-run, Money supply and the one period lag of real exchange rate negatively affects the trade balance. Inference drawn from this result indicates evidence of J-curve effect in the case of Sierra Leone. The equilibrium correction is fairly fast and is restored by the following year. The findings of the study reveal some lessons for policy consideration. External sector imbalance has to be addressed through boosting supply of goods and other services as well as the management of domestic demand. 82