Keywords Exchange rate volatility, Export Quantity Index, Time rolling moving average variance, GARCH, Multivariate GARCH, VECM, Subprime crisis.

Transcription

1 Impacts of Exchange Rate Volatility on Export Quantity of Thailand Tatre Jantarakolica 1 & Porjai Chalermsook 2 1 Faculty of Economics, Thammasat University, Prachan Road, Phranakorn, Bangkok 10200, Thailand Tel: , Fax: , tatre@econ.tu.ac.th 2 The Office of Permanent Secretary, Ministry of Commerce, Thailand, porjaia@moc.go.th Abstract This research intends to determine (i) whether exchange rate volatility has impact on export quantity of Thailand and (ii) alternative methods in determining exchange volatility. The study constructs composite Thailand export quantity index by using the disaggregated export value and export price. The four alternative methods, including, monthly variance, time rolling moving average variance, univariate GARCH model of spot exchange rate, and the bivariate GARCH models of spot and forward exchange rate, are employed in determining exchange rate volatility. Static and dynamic structure models have been applied to analyze the impact of exchange rate volatility on Thai export quantity by using monthly data. Also, the study includes subprime crisis dummy variable in determining impact of subprime crisis on Thai export. The estimated results indicate that the bivariate GARCH model is the most appropriated method in determining the exchange rate volatility on Thai export. The findings reveal that two markets, future and spot exchange rate markets of Thailand, are interconnected, thus, evaluating the exchange rate risk must be computed through the movement of both markets. The dynamic structural model using VECM provides better overall significant results than the estimated results from the static model. The results implied that the Thai export is dynamically influenced by both exchange rate and the exchange rate risk. Higher exchange rate volatility can result in a reduction of export quantity. Additionally, the significant result of the subprime crisis had confirmed the contagion impacts of subprime crisis on Thailand economy through the reduction of Thai export during that period. Keywords Exchange rate volatility, Export Quantity Index, Time rolling moving average variance, GARCH, Multivariate GARCH, VECM, Subprime crisis. Work done as a collaboration between Thammasat University and The Office of Permanent Secretary, Ministry of Commerce. -1-

2 1. Introduction The promotion of export oriented industries as a mechanism of economic stimulus has been one of the Thailand s major economic development plans since By adopting the export oriented policy for over 30 years, Thailand s international trade has contributed to a high proportion of the country s Gross Domestic Product. Thus, the international trade has played an important role on Thailand s economic growth since then. Among the most crucial factors that affect the international trade is the foreign exchange rate which influences the country s competitiveness in many ways. After Thailand s financial crisis in 1997, the country has adopted the floating exchange rate policy which greatly increased the exchange rate volatility and therefore directly affected the export industries. The consequences of exchange rate risk or exchange rate volatility on international trade, especially on export, have been extensively studied. Volatility of exchange rate is believed to have negative impact on trade volume, as discussed in (McKenzie 1999; Clark 2004); intuitively, when the risk increases, the risk-averse traders would be afraid of losing money, so they would export less. The effect of exchange rate volatility on international trade also depends on the degree of competition and the relative degree of risk aversion and risk exposure of importers and exporters. If exporters bear the risk, prices will increase. If importers do, prices may fall. Invoicing in the home currency does not eliminate the exporter s risk, as quantity demanded becomes uncertain. Although the effect of exchange rate volatility is intuitively clear from the above (highlevel) discussion, both theoretical and empirical researches have failed to either support or reject the hypothesis (Cote 1994; McKenzie 1999; Clark 2004; Ozturk 2006). Theoretically, such factors as definition of exchange rate volatility, risk-aversion of traders, and presence or absence of forward exchange market seem to be crucial in determining the relationship between trade and exchange rate volitility (Broll, 1994; De Grauwe, 1988; Dellas and Zilberfarb, 1993; Ethier, 1973; Franke, 1991; Hooper and Kohlhagen, 1978; Sercu, 1992; Viaene and de Vries, 1992). For instance, in a theoretical work of De Grauwe (De Grauwe,1988), it is argued that the effects of volatility depend on the degree of risk-aversion of traders in the market; in their model, the riskaverse exporters would export more when there is more risk in the exchange rate. This result disagrees with the standard belief about behavior of risk-averse traders. -2-

3 On empirical side, the situation is similarly indecisive. There are results that support the hypothesis (Arize, 1995; Chowdhury, 1993; De Grauwe, 1987; Perée and Steinherr, 1989). On the other hands, many results do not (Asseery and Peel, 1991; IMF, 1984). Even worse, some studies cannot find a significant relationship between these two variables; see e.g. (Aristotelous, 2001; Gagnon, 1993). These previous researches suggest that the empirical results are very sensitive to a number of factors, e.g. proxies for exchange rate volatility, model specification, sample period, and the countries considered. Some recent empirical results show that the choice of statistical techniques used also matters, e.g., in (De Vita and Abbott, 2004), the empirical result does not show any significant impact of short-term exchange rate volatility, whereas the long-term effect was found. An increase in exchange rate volatility may also have a secondary effect on trade prices, reducing the pass-through of changes in competitiveness. In this paper, we emphasize the direct effects, although admittedly, the indirect effects may have an even greater role to play. It is also worth mentioning that the literature is largely based on a partial-equilibrium approach that precludes inferences about welfare. As a result, the empirical findings cannot be used to conclude that one exchange rate system is necessarily preferable to another, as other factors would have to be taken into consideration. The main purpose of this paper is to perform an empirical study of this very important problem in the case of Thailand. The questions we try to answer include: (i) whether exchange rate volatility has impact on export quantity of Thailand and (ii) alternative methods in determining exchange volatility. In this paper, we constructs composite Thailand export quantity index as the variable that represents export by using the aggregated export value and export price. The study employs four alternative methods, including, (i) monthly variance, (ii) time rolling moving average variance, (iii) univariate GARCH model of spot exchange rate, and (iv) the bivariate GARCH models of spot and forward exchange rate, to determine exchange rate volatility. Static and dynamic structure models have been applied to analyze the impact of exchange rate volatility on Thai export quantity by using monthly data. 2. Conceptual Framework To analyze the impacts of exchange rate volatility on export quantity of Thailand, this study first employs the producer theory of the firm under uncertainty. The study first begins -3-

4 with the discussion on the basic international trade models and then more literature review on more recent models. 2.1 Basic uncertainty trade models The traditional model assumes firms generally produce in order to maximizing their profits. However, the profitability of undiversified exporting firms is directly affected by the fluctuation of bilateral exchange rate. As a result, these exporting firms face the exchange rate risk in conducting their international trading business. Clark (1973) introduced the model in explaining behavior of an exporting firm that produces under perfect competition a homogeneous export product that is sold entirely abroad. The model assumes that the firm uses no imported inputs and the price of the exported good in foreign currency is an exogenous variable. The firm is paid in foreign currency without any hedging opportunity. Output is assumed to be constant over the entire period of study. The exchange rate risk is determined by the uncertainty of the future exchange rates that the firm faced when converting the future export receipts into domestic currency. Therefore, the firm has to make decision on the level of their international trade under this exchange rate uncertainty. By assuming quadratic profit function, the first-order condition of the expected utility maximization indicates that risk aversion s firm must produce at the level at which marginal revenue exceeds marginal cost. Then, the firm must be compensated for the uncertainty of the exchange rate. As a result, the export supply curve shifts to the left, and the volume of export production as well as firm international trade decrease. In attempting to reduce their exchange rate risk, the risk aversion s firms reduce their sales, so, their expected profits and variance of profits also decrease while their expected utility increases. Furthermore, in the case that inputs of export products are imported, the reduction in the supply of exports would be smaller. Variation in profits would be less influenced by the uncertainty of the exchange rate. Later study by Baron (1976) relaxes perfect competition assumption in determining the impact of exchange rate volatility on the export prices, especially invoice currency. The study explains that the exchange rate risk occurs when exporter invoices are in terms of foreign currency. With the fixed quantity of export order and domestic prices, exporters revenue and profits become uncertain as the exchange rate fluctuate, and thus, the exporter faces exchange rate risk. On the contrary, when invoicing is in terms of domestic currency, the exporter faces -4-

5 quantity risk. The export demand becomes uncertain due to variation of export price in terms of foreign currency. With quantity demand uncertainty, costs of production of the firm become uncertain. In both cases, the risk-averse firms would try to reduce their exchange rate risk exposure, but impacts of the risk are different. In the foreign currency invoicing case, exchange rate risk caused by an exchange rate appreciation will result in rising of the export price, and then, the export demand will decline. Therefore, the higher export price, the lower the firms profits. In the domestic currency invoicing case, the impacts will depend on demand function of the destination market. Export price will decrease in linear demand function case and lead to increasing in export demand. However, the profit margin of export firm declines, which might lead to a reduction of the firms profit. Employing bilateral framework and assuming monopolistic competitive market, Hooper and Kohlhagen (1978) determined impacts of exchange rate volatility on the level of international trade by using the movement of nominal exchange rate as the measurement of exchange rate uncertainty. The models are derived from demand and supply functions of individual firms and then aggregate them to obtain reduced-form equations for the market equilibrium price and quantity. Their study emphasizes on estimated results of the parameters of invoice currency, forward hedging position, and degrees of risk aversion of the exporters and importers. Their result indicates that exchange rate volatility has impacts only on the portion of the profits that is not hedged. Demand for imports is treated as the derived demand from exports by assuming that imports are used as inputs for export production. Then, an increase in exchange rate volatility will result in higher variation of profits and lower export demands. As a result, export quantity and price decrease. The level of the impacts of exchange rate volatility on export depends on export price elasticity of the foreign demand, exporters and importers degree of risk aversion, and their degree of exposure to the risk. Similar to Clark (1973), exporters are assumed to export all their products into foreign market. Exchange rate risk then leads to the reduction of export supply causing the decreasing in export quantity and rising of export price. The estimated results of the reduced-form models indicate the inverse relationship between exchange rate volatility and the volume of international trade. -5-

6 2.2 The models We applied the two-country imperfect substitution models which are based on the assumption that export goods and goods produced in the foreign country cannot be perfectly substituted (Goldstein & Khan, 1985, and Rose, 1991). Particularly for this study, we further assume that the amount of Thai export depends on the prices of Thai s products represented by export price, converted from Thai Baht to foreign currency via the exchange rate in the corresponding time period. Meanwhile, since the export quantity also depends on foreign income or importing countries income, we can write the Thai export quantity as a function of the foreign income, export price, and exchange rate as follows. (1) where denotes Thai s export quantity, is foreign income, is the Thai s export price, and is the exchange rate. Moreover, Thailand, as a developing country, always relies on factors of production from foreign countries, so the export volume also depends on the import of factors of production. (2) where, is Thailand s import. The importers of the trading partners will confront the risk of exchange rate volatility, in case it is sufficiently high (since exchange rate directly affects the export price). Therefore, the export function should also account for the exchange rate volatility as well, as shown in the following equation. (3) where, is the exchange rate volatility. The export function above can be applied to the structural model with one-directional and static relationship as follows: (4) where, is column vector of export quantity at time t. is matrix of independent variables at time t -6-

7 is column vector of the error term at time t. Due to the fact that our study involves the subprime crisis period (which may affect the export volume), we add a dummy variable to account for the subprime crisis in this model. The model then becomes (5) where, is a dummy variable for subprime crisis, which assumes the value of 0 before 2008, and 1 after Methodology 3.1 Data In this paper, we use a monthly, aggregated data set over the period between January 2000 and August quantity. Dependent Variable: We use Thai export quantity index ( ) as a proxy for export Independent Variables: We use the foreign industrial production index 1 ( ) as a proxy of importing countries income, Thai s export price index ( ) as a proxy of export price, nominal exchange rate ( ), in terms of Baht per US dollar from Bank of Thailand to proxy exchange rate, and import value ( ) to proxy import of Thailand. We apply four methods to measure exchange rate volatility ( ), as explained later in Section 3.4. Dummy variable ( ) takes the value of 0 before 2008 and 1 after Measurement of Export Quantity Index The dependent variable used in this study is export quantity. Previous studies that deal with export of Thailand mostly used export value as a dependent variable due to data availability. However, the dependent variable in export function theoretically refers to export volume or quantity not the export value, so we construct the export quantity index to be used as a dependent variable in this study via the following formula: 1 In fact, it is more common to use gross domestic product (GDP) to proxy foreign countries income, but since we are dealing with monthly data, we are unable to use GDP here. -7-

8 (6) where is export quantity index in period t. is export value in period t. is export price index in period t. is export value in the based year (in this study based year is 2007). is export price index in the based year (in this study based year is 2007). 3.3 Measurement of Foreign Industrial Production Index We will be using the weighted average of industrial production index of the selected importing countries. More specifically, we use the following formula. (7) where is the weighted average of industrial production index, is the industrial production index of country i at time t, is the industrial production index of country i in the base year (i.e. year of 2005), and finally is the weight (export share) for country i at time t computed by the following formula. 3.4 Measurement of Exchange Rate Volatility In this study, four alternative methods are employed in determining exchange rate volatility. The four alternative techniques include monthly variance, time rolling moving average variance, univariate GARCH model of spot exchange rate, and the bivariate GARCH models of spot and forward exchange rate Monthly Variance Monthly variance is computed by using daily data of each month using the sample variance as shown by the following formula: var 1 and fx fx fx 2 j j t j n j 1 t1 n j fx j n j 1 fxt (8) n j t 1-8-

9 where var j j fx is the variance of exchange rate fx t at time j Moving Variance Measurement The moving variance method (Klaassen, 2004) is widely used in measuring the exchange rate volatility, in which one calculates the moving variance from the observed exchange rate. We will be using the window widths of 3, 6, 9 and 12 months. The formula, as taken directly from Klaassen (2004), is as follows. j 1 V fx fx fx 2 and t t ti j j 1 i1 fx j 1 j fxti j i1 (9) V fx is the moving variance of exchange rate fx t at time t where t t The value of j is set to be 3, 6, 9 and 12, respectively, so we have obtained four series for exchange rate volatility; (1) the 3-month moving variance, (2) the 6-month moving variance, (3) the 9-month moving variance, and (4) the 12-month moving variance, respectively. We further examine whether the window width of the moving variance has any impact on the accuracy of the exchange rate volatility Variance Measurement using Generalize Autoregressive Conditional Heteroscedasticity (GARCH) Most research in exchange rate volatility measurement applies the Generalize Autoregressive Conditional Heteroscedasticity model (GARCH). In this study, we adapt the Mean Equation Model, so it becomes the Autoregressive Integrated Moving Average model (ARIMA) with GARCH variance equation. Therefore, we have the following equation. where d L fxt L ut (10) L is the Moving Average polynomial in the following form, L is the Autoregressive polynomial 1 2 q L L L ql (11) 1 2 p L L L pl (12) -9-

10 fxt is exchange rate at time t, GARCH(p,q) variance 2 t L d is the d th difference operator, u t is the error term at time t with L (13) 2 2 t t L is the Autoregressive polynomial L 1 2 q L 1 1L 2L ql (14) And finally is the moving average polynomial 1 2 p L 1 1L 2L pl (15) Variance Measurement using Multivariate Generalize Autoregressive Conditional Heteroscedasticity (M-GARCH) Researches concerning on exchange rate volatility that incorporated the movement of both spot and future foreign exchange market mostly employed the Multivariate Generalize Autoregressive Conditional Heteroscedasticity model (M-GARCH). In this study, we apply the bivariate GARCH model of both spot and forward premium. The model employs the Vector Autoregressive (VAR) model as mean equations for spot and forward premium and multivariate ARCH as the variance equation. Therefore, we have the following equation. Fx k Fx (16) t 0 i ti t i0 t t 1 t t ij ijt ~ N(0, H ), H h, i, j 1,2, h 2, iit i i1 it h h h 2, i j, ijt ij iit jjt where Fx fx fx, fx st represents spot rate at time t, t st pt fx pt represents forward premium at time t, 0 is a two dimensional column vector of constants, i are 2x2 matrices of coefficients, t is a normally distributed two-dimensional vector random error with covariance matrix H t -10-

11 with the sufficient conditions that H t are positive definite for all t and each conditional variances, h iit, are positive and that the constant matrix of conditional correlations is positive definite, and ij Corr it, jt t 1 constant. denotes the conditional correlation which is assumed to be 3.5 Research Processes As a recap, we use export quantity index as dependent variable. For independent variables, we use the export price index ( ) to represent the export price, the nominal exchange rate ( ) for the exchange rate, the foreign industrial production index ( -11- ) to proxy foreign countries income, the import of Thailand ( ), and the exchange rate volatility ( ). We have four methods to measure the exchange rate volatility as discussed in the previous section, i.e. (i) monthly variance, (ii) time rolling moving average variance, (iii) univariate GARCH model of spot exchange rate, and (iv) the bivariate GARCH models of spot and forward exchange rate. For each such measurement, we perform the experiment in both static and dynamic settings Static Model following equation. Next, we estimate the export quantity from (5) in the static model by the (17) The Ordinary Least Square method is employed to estimate the static model, where the first order autocorrelation and multicollinearity problems are tested and solved in order to adjust the model Dynamic Model The economic variables, especially the time series ones, tend to possess a more dynamic relationship which, together with the non-stationary time series, might cause the spurious problem. Therefore, we will be using the Vector Error Correction Mechanism (VECM) to determine the dynamic relationship and to give long run equilibrium relationship of the variable in the system. We, additionally, set the long run cointegrated vector to be Y t 1. The VECM equation is as follows.

12 where Y t is endogenous variable p Y Y Y (18) t t 1 i t i t i1 is Speed of Adjustment Parameter t is a noise vector assumed to be normally distributed Y t 1 is long run relationship vector, which can be written as follows Or (19) The estimation of (18) and (19) is done by determining the proper lag or the value of i by means of Schwarz Bayesian Information Criteria. Then we test the stability of the system. Finally, the Maximum Likelihood Estimation Method will be used to estimate the model. 4. Empirical Results We use four methods four alternative methods to measure exchange rate volatility, i.e. variance, time rolling moving average variance, univariate GARCH model of spot exchange rate, and the bivariate GARCH models of spot and forward exchange rate. The movement of the exchange rate series during 2000 to 2010 is shown in Figure 1. During the subprime crisis in 2008, Thai Baht was appreciated against US dollar. <Insert Figure 1 here> The spot exchange rate and forward exchange rate movements during 2000 to 2010 are shown in Figure 2. Forward premium can be determined by the gaps between both rates. The premiums tend to be higher during the exchange rate depreciation compare to the appreciation. <Insert Figure 2 here> -12-

13 Figure 3 compares the level of fluctuation of the movements of exchange rate volatility measured by time-rolling moving average with different time windows three-month, sixmonth, nine-month, and twelve-month. This figure shows that the longer the length of time windows the higher the volatility of exchange rate it measures. The twelve-month time-rolling moving average variance gives the highest value of volatility of exchange rate. <Insert Figure 3 here> Figure 4 illustrates the movements of exchange rate volatility measured by four alternative methods, including, monthly variance, 3-month time-rolling moving average variance, univariate GARCH model of spot exchange rate, and the bivariate GARCH models of spot and forward exchange rate. This figure shows that monthly volatility computed from daily exchange rate data (monthly variance & bivariate GARCH) are lower than those computed from monthly data (3-month time-rolling moving average variance & GARCH). The bivariate GARCH model takes into account the exchange rate forward market which allows the hedging position of the exchange rate risk, so measurement of exchange rate risk from this model provides the lowest level of the volatility. <Insert Figure 4 here> The estimation results of bivariate GARCH models are shown in Table 1, which indicates significant relationship between spot exchange rate and forward rate. The results reveal that the bivariate ARCH effects exist between both markets (spot and forward markets), which implies that both markets are interconnected, allowing importers and exporters to hedge their international trading transactions against the exchange rate risk through the mechanism of this forward derivative market. However, the results do not show any significant impacts of the bivariate GARCH effects. The insignificant results might be due to the fact that the estimated data employed in this study are daily data which might not be highly frequent enough to reveal the bivariate GARCH effects of both markets. <Insert Table 1 here> -13-

14 The results of unit root test by using Augmented Dickey-Fuller (ADF) test are shown in Table 2. According to Table 2, only five variables, which are (i) 3-month time-rolling moving average variance, (ii) 6-month time-rolling moving average variance, (iii) exchange rate variance determined by GARCH, (iv) monthly variance exchange rate, and (v) exchange rate variance determined by Bivariate GARCH, qualified for stationary at level, I(0), while other variables are all integrated series of order 1 or I(1). Therefore, estimating the export quantity model using these variables by OLS may cause spurious problem. Thus, this study employed Vector Error Correction Mechanism (VECM) technique in estimating Cointegrated long-run relationship and dynamic model of the export quantity determination model. However, this study also estimated the static models to be compared with the dynamic and cointegrated long-run models. <Insert Table 2 here> The estimations of the eight static models using OLS are shown in Table 3. There are eight different models shown in different columns of Table 3. The models consist of (1) the model without exchange rate volatility; (2) the model with 3-month time-rolling moving average variance; (3) The model with 6-month time-rolling moving average variance; (4) the model with 9-month time-rolling moving average variance; (5) the model with 12-month time-rolling moving average variance; (6) the model with variance determined by GARCH model; (7) the model with monthly variance; and (8) the model with variance determined by Bivariate GARCH model. <Insert Table 3 here> According to the estimated results of these static models, we found that the exchange rate volatility determined by 9-month time-rolling moving average is the only exchange rate risk variable that statistically significant influenced the Thai quantity export. The model with the exchange rate risk determined by Bivariate GARCH model (Model (8)) seems to be the best model among all eight static models, in terms of statistical significance of the model, best fit the data (lowest AIC and BIC), and prediction error (lowest RMSE). -14-

15 For all eight models, the coefficients of import value and subprime crisis dummy are statistically significant. Higher import value will result in increasing in export volume while the subprime crisis in 2008 caused the significant sharply contraction of Thai export. These estimated results indicate that static estimation method using OLS might not be appropriated in this case due to the high possibility of spurious problem as most control variables are nonstationary. Therefore, we employed cointegration test by using Johansen Multivariate Cointegration test and found that all variables in the models have the long-run cointegrated relationship. Table 4 reveals that there are three cointegrating vectors for all models except model (1) (model without exchange rate volatility) which has only two cointegrating vectors. These results confirm that long-run relationship model determined by Vector Error Correction better reveals the impacts of exchange rate volatility on export volume than the static models. All seven models with exchange rate volatility as explanatory variable (model (2) to model (8)) indicated that exchange rate volatility variables determined by four alternative methods are all statistically significant in determining the Thai export quantity. <Insert Table 4 here> The long-run dynamic models illustrate significant impacts of all explanatory variables (exchange rate, exchange rate volatility, and subprime crisis) and control variables (importers income determined by world industrial production index, Thai export price, and Thai import value) on the Thai export volume. Exchange rate level and its volatility both have significant influenced on Thai export demand. Exchange rate appreciation and high exchange rate volatility will both worsen the Thai export demand. Subprime crisis had significant negative impacts on Thai export. Increasing in world production can lead to an increasing in Thai export quantity. The higher Thai export price can result in contraction of Thai export demand. Thai export quantity is also determined by the level of Thai import value. Increasing in Thai import value indicates an increasing in Thai export. -15-

16 Concerning on econometric properties, as indicated by the lowest log-likelihood value, model with Bivariate GARCH as measurement of exchange rate volatility (model (8)) seems to be the best model in revealing impacts of exchange rate risk on Thai export volume. These findings confirm the dependency between the exchange rate volatility and the Thailand s export volume. This may suggest that adding the exchange rate volatility variable obtained by Bivariate GARCH can help revealing all dimensions of impacts of exchange rate (both level and volatility) on the export volume more precisely and significantly. Furthermore, it also reflects the impact of subprime crisis which significantly caused the contraction of Thai export. 5. Discussion and Conclusion This research intends to reveal the impacts of exchange rate volatility on Thai export quantity. Four alternative methods in determining exchange rate volatility consist of monthly variance, time rolling moving average variance, univariate GARCH model of spot exchange rate, and the bivariate GARCH models of spot and forward exchange rate. The results from long-run dynamic models confirm that exchange rate risk has significant impact on export quantity. This finding is consistent with the common belief that higher exchange rate risk can cause higher fluctuation in profit of the Thai export firms, thus, leading to a decline of export. Our findings also show that among all other alternative measurements, the exchange rate risk measured Bivariate GARCH model is the best measurement of the exchange rate risk in determining its impact on export volume. This result coincides with the recent study by Klaassen (2007). Additionally, the significant results of the Bivariate GARCH models imply that both spot and forward exchange rate market are interconnected, which allow Thai importers and exporters to hedge their international trading transactions against exchange rate risk. This hedging channel is also supported by the fact that exchange rate risk determined by Bivariate GARCH reveals the lowest log-likelihood level when compared with the other single-spot market methods of measurement. The structural model of export with dynamic relationship reveals better statistical performance than the static relationship ones in term of reflecting the relationship among many factors that influence the export volume. This suggests that currently the Thai economic and -16-

17 international trade variables are related in a dynamic way, and hence the estimation of the impact of these variables should take this dynamic relationship into account as well. In conclusion, this study found that the Thai export volume is dynamically influenced by both exchange rate and the exchange rate risk in a negative way, i.e. the higher exchange rate, the lower the export quantity. Additionally, the fact that dummy variable of subprime crisis is significant revealed the contagious impacts of subprime crisis on Thailand economy through the reduction of Thai export during that period. -17-

18 References Anderson, T.G., Bollerslev, T., Diebold, F.X., & Laby, P. (1998). The Distribution of Exchange Rate Volatility. Working Paper, National Science Foundation. Anderton, R. & Skudelny, F. (2001). Exchange Rate Volatility and Euro Area Imports. Working Paper, No. 64, European Central Bank. Aristotelous, K. (2001). Exchange rate volatility, exchange rate regime, and trade volume: evidence from the UK US Export Function ( ). Economics Letters, 72, pp Arize, A.C. (1995). The Effects of Exchange Rate Volatility on U.S. Exports: An Empirical Investigation. Southern Economic Journal, Vol. 62, Arize, A.C. (1998). The Effects of Exchange Rate Volatility on U.S. Imports: An Empirical Investigation. International Economic Journal, Vol. 12(3), Asseery, A. and Peel, D. A. (1991). The effects of exchange rate volatility on exports: some new estimates. Economics Letters, 37, pp Baron, D. P. (1976). Fluctuating Exchange Rates and the Pricing of Exports. Economic Inquiry, 14, pp Broll, U. (1994) Foreign Production and Forward Markets. Australian Economic Papers, 62, 1-6. Cote, A. (1994). Exchange Rate Volatility and Trade: A Survey. Working Paper 94-5, Bank of Canada. Chowdhury, A. R. (1993). Does exchange rate volatility depress trade flows? Evidence from error-correction models. Review of Economics and Statistics, 75, pp Clark, P. (1973). Uncertainty, Exchange Risk, and the Level of International Trade. Western Economic Journal, 11, pp Clark, P., Tamirisa, N., Wei, S.J., Sadikov, A. and Zeng, L. (2004) Exchange Rate Volatility and Trade Flows Some New Evidence, IMF Manuscript. De Grauwe, P. (1987). International trade and economic growth in the EMS. European Economic Review, 31, pp

19 De Grauwe P. (1988), Exchange Rate Volatility and Slowdown in Growth of International Trade. IMF Staff Papers. De Vita G. and Abbott A (2004). The Impact of Exchange Rate Volatility on UK exports to EU countries. Scottish Journal of Political Economy, Vol 51(1), Dellas, H. and Zilberfarb, B.-Z. (1993). Real exchange rate volatility and international trade: a re-examination of the theory. Southern Economic Journal, 59, pp Diebold, F.X. & Nerlove, M. (1989). The Dynamics of Exchanger Rate Volatility: A Multivariate Latent Factor ARCH Model. Journal of Applied Econometrics, Vol. 4, Franke G (1991). Exchange Rate Volatility and International Trading Strategy. Journal of International Money and Finance, 10, Gagnon, J. E. (1993). Exchange rate variability and the level of international trade. Journal of International Economics, 34, pp Gali, J. & Monacelli, T. (2004). Monetary Policy and Exchange Rate Volatility in a Small Open Economy. Working Paper, Universita Bocconi. Goldstein, M., & Khan, M.S., (1985). Income and price effects in foreign trade. in Jones, R.W., Kenen, P.B. (Eds.). Handbook of International Economics, North-Holland, Amsterdam, Hooper, P. and Kohlhagen, S. W. (1978). The effect of exchange rate uncertainty on the prices and volume of international trade.journal of International Economics, 8, pp International Monetary Fund (IMF) (1984). Exchange rate volatility and world trade. Occasional Papers, July. Klaasen, F. (2007). Why is it so difficult to find an effect of exchange rate risk on trade? Journal of International Money and Finance, Vol. 23, Kroner, F. and Lastrapes, W. D. (1993). The impact of exchange rate volatility on international trade: reduced form estimates using the GARCH-in-mean model. Journal of International Money and Finance, 12, pp Lastrapes, W. & Koray, F. (1990). Exchange Rate Volatility and U.S. Multilateral Trade Flows. Journal of Macroeconomics, Vol. 12(3),

20 McKenzie D. M.(1998), The Impact of Exchange Rate Volatility on International Trade Flows. Journal of Economics Survey, Vol. 13(1), Ozturk, I. (2006) Exchange Rate Volatility and Trade: A Literature Survey, International Journal of Applied Econometrics and Quantitative Studies 3(1): PerÉe, E. and Steinherr, A. (1989). Exchange rate uncertainty and foreign trade. European Economic Review, 33, pp Rose, A.K. (1991). The role of exchange rates in a popular model of international trade: does the Marshall-Lerner condition hold? Journal of International Economics, Vol. 30, Sercu, P. (1992). Exchange rates, volatility and the option to trade. Journal of International Money and Finance, 11, pp Tenreyro, S. (2004). On the Trade Impact of Nominal Exchange Rate Volatility. Working Paper, Federal Reserve Bank of Boston. Viaene, J.-M. and De Vries, C. (1992). International trade and exchange rate volatility. European Economic Review, 36, pp

21 Figure 1: Exchange Rate (Bath/US Dollar) Baht/US Dollar 50 Subprime Crisis Exchange Rate Year Note: During the subprime crisis in 2008, Thai Baht was appreciated against US dollar. Figure 2: Movement of Spot and Forward Exchange Rate Baht/USD 55 Depreciation Appreciation Depreciation Baht/USD forward rate spot rate Note: Forward premium can be determined by the gaps between both rates. The premiums tend to be higher during the rapidly exchange rate depreciation compare to the rapidly appreciation. -21-

22 Figure 3: Exchange Rate Volatility Measured by Time-Rolling Moving Average Exchange Rate Volatility 6 TRMA03 TRMA06 TRMA09 TRMA TRMA12 TRMA09 TRMA06 TRMA03 Year Note: TRMA03 is 3-month time-rolling moving average variance of exchange rate. TRMA06 is 6-month time-rolling moving average variance of exchange rate. TRMA09 is 9-month time-rolling moving average variance of exchange rate. TRMA12 is 12-month time-rolling moving average variance of exchange rate. This figure shows that the longer the length of time the higher the volatility of exchange rate. 12-month time-rolling moving average variance results the highest measure of volatility of exchange rate. -22-

23 Figure 4: Exchange Rate Volatility Measured by Four Different Methods Exchange Rate Volatility 2.0 Monthly TRMA03 GARCH Bi-GARCH GARCH 0.0 TRMA03 Monthly Bi-GARCH Year Note: Monthly is monthly variance computed by daily exchange rate. TRMA03 is 3-month time-rolling moving average variance of exchange rate. GARCH is monthly variance computed from GARCH model. Bi-GARCH is monthly variance computed from daily Bivariate-GARCH model. This figure shows that monthly volatility computed from daily exchange rate data (Monthly & Bi- GARCH) are lower than those computed from monthly data (TRMA03 & GARCH). Bivariate-GARCH model takes into account the future market which allows the hedging position of the exchange rate risk, thus, measurement of exchange rate risk from this model provides the lowest level of the volatility. -23-

24 Table 1: Estimated Result of Bivariate-GARCH Models Variable fx st fx pt cons ** * fx t ** *** Sigma0 1_ *** 2_ _ *** L.ARCH 1_ *** 2_ _ *** Statistics Chi2 test *** Log likelihood Note: * means statistically and significantly different from zero at 0.1 significant levels. ** means statistically and significantly different from zero at 0.05 significant levels. *** means statistically and significantly different from zero at 0.01 significant levels. The results indicate the bivariate ARCH effects exist between both markets (spot and forward markets). This can be implied that both markets are interconnected which allow importers and exporters to hedge their trading transaction against the exchange rate risk through the mechanism of this forward derivative market. However, the results do not show the significant impacts of the bivariate GARCH effects. The reason of such results might be because of the estimated data employed in this study is daily data which might not highly frequent enough to reveal the bivariate GARCH effects of both markets. -24-

25 Table 2: Unit Root Test by using Augmented Dickey-Fuller Test Variable t-statistic at level I(0) t-statistic at first difference I(1) eqi *** wipi *** epi *** fx *** im *** fxrisk *** *** fxrisk *** *** fxrisk *** fxrisk *** fxvol *** *** fxvol *** *** fxvol *** *** Note: * means statistically and significantly different from zero at 0.1 significant levels. ** means statistically and significantly different from zero at 0.05 significant levels. *** means statistically and significantly different from zero at 0.01 significant levels. eqi is Thai export quantity index. wipi is world industrial production index as a proxy for importer s income. epi is export price index as a proxy for Thai export price. fx is Baht per US as a proxy for exchange rate. im is Thai import value. fxrisk3 is 3-month time-rolling moving average variance as a proxy by exchange rate volatility. fxrisk6 is 6-month time-rolling moving average variance as a proxy by exchange rate volatility. fxrisk9 is 9-month time-rolling moving average variance as a proxy by exchange rate volatility. fxrisk12 is 12-month time-rolling moving average variance as a proxy by exchange rate volatility. fxvol1 is exchange rate variance determined by GARCH as a proxy for exchange rate volatility. fxvol2 is monthly variance exchange rate as a proxy for exchange rate volatility. fxvol3 is exchange rate variance determined by Bivariate GARCH as a proxy for exchange rate volatility. The results of unit root test by using Augmented Dickey-Fuller (ADF) test indicate that only five variables, i.e. 3- month time-rolling moving average variance, and 6-month time-rolling moving average variance, exchange rate variance determined by GARCH, monthly variance exchange rate, and exchange rate variance determined by Bivariate GARCH qualified for stationary at level, I(0), while other variables are all I(1). Therefore, estimating the model by OLS may cause a spurious problem. -25-

26 Table 3: The Static Model Estimated by OLS Static Models (1) (2) (3) (4) (5) (6) (7) (8) Constant wipi epi im *** *** *** *** *** *** *** *** fx ** ** crisis ** ** ** ** ** ** ** ** fxrrisk fxrisk fxrisk * fxrisk fxvol fxvol2 fxvol F-test *** *** *** *** *** *** *** *** R Adj. R RMSE AIC BIC Note: (1) the model without exchange rate volatility; (2) the model with 3-month time-rolling moving average variance; (3) The model with 6-month time-rolling moving average variance; (4) the model with 9-month time-rolling moving average variance; (5) the model with 12-month time-rolling moving average variance; (6) the model with variance determined by GARCH model; (7) the model with monthly variance; and (8) the model with variance determined by Bivariate GARCH model. * means statistically and significantly different from zero at 0.1 significant levels. ** means statistically and significantly different from zero at 0.05 significant levels. *** means statistically and significantly different from zero at 0.01 significant levels. eqi is Thai export quantity index. wipi is world industrial production index as a proxy for importer s income. epi is export price index as a proxy for Thai export price. fx is Baht per US as a proxy for exchange rate. im is Thai import value. fxrisk3 is 3-month time-rolling moving average variance as a proxy by exchange rate volatility. fxrisk6 is 6-month time-rolling moving average variance as a proxy by exchange rate volatility. fxrisk9 is 9-month time-rolling moving average variance as a proxy by exchange rate volatility. fxrisk12 is 12-month time-rolling moving average variance as a proxy by exchange rate volatility. fxvol1 is exchange rate variance determined by GARCH as a proxy for exchange rate volatility. fxvol2 is monthly variance exchange rate as a proxy for exchange rate volatility. fxvol3 is exchange rate variance determined by Bivariate GARCH as a proxy for exchange rate volatility. -26-

27 Table 4: The Dynamic Models Estimated by Vector Error Correction Mechanism (VECM) Dynamic Models (1) (2) (3) (4) (5) (6) (7) (8) Constant *** *** *** *** *** *** *** *** wipi *** *** ** *** * epi *** *** *** *** *** ** *** *** im ** ** *** *** *** *** *** ** fx * *** *** *** *** *** crisis *** *** *** *** *** *** *** fxrrisk * fxrisk *** fxrisk *** fxrisk * fxvol *** fxvol *** fxvol *** Loglikelihood Overall Test *** *** *** *** *** *** *** *** CE Rank Note: (1) the model without exchange rate volatility; (2) the model with 3-month time-rolling moving average variance; (3) The model with 6-month time-rolling moving average variance; (4) the model with 9-month time-rolling moving average variance; (5) the model with 12-month time-rolling moving average variance; (6) the model with variance determined by GARCH; (7) the model with monthly variance; and (8) the model with Bivariate GARCH * means statistically and significantly different from zero at 0.1 significant levels. ** means statistically and significantly different from zero at 0.05 significant levels. *** means statistically and significantly different from zero at 0.01 significant levels. eqi is Thai export quantity index. wipi is world industrial production index as a proxy for importer s income. epi is export price index as a proxy for Thai export price. fx is Baht per US as a proxy for exchange rate. im is Thai import value. fxrisk3 is 3-month time-rolling moving average variance as a proxy by exchange rate volatility. fxrisk6 is 6-month time-rolling moving average variance as a proxy by exchange rate volatility. fxrisk9 is 9-month time-rolling moving average variance as a proxy by exchange rate volatility. fxrisk12 is 12-month time-rolling moving average variance as a proxy by exchange rate volatility. fxvol1 is exchange rate variance determined by GARCH as a proxy for exchange rate volatility. fxvol2 is monthly variance exchange rate as a proxy for exchange rate volatility. fxvol3 is exchange rate variance determined by Bivariate GARCH as a proxy for exchange rate volatility. -27-