Volatility in Exchange Rate Components and the Volume of International Trade

By decomposing the changes in the real exchange rate series into fundamental and transitory components (market microstructure and stochastic element) and modeling the volatility in each via a GARCH process, this paper examines how volatility in exchange rate affects the volume of aggregate and disaggregate US trade with Canada, Germany, and Hong Kong during the 1989–2002 period. The results indicate significantly different impacts of volatility due to the fundamental and transitory components of the exchange rate series on US bilateral trade. While the findings suggest heterogeneous responses of traders to volatilities arising from different components of the real exchange rate, the impact of the volatility due to the fundamental component is also found to vary across commodities, implying disparities in the inter- and intra-trading arrangements made by traders of different goods in counteracting foreign exchange risk arising from changes in the economic fundamentals. Notes 1 The implication is that differences in the pricing behavior could give rise to a potential variation in the sensitivity of firms to exchange rate risk. 2 For example, while importing firms in Germany might be able to absorb relatively larger shocks in the foreign exchange market with little or no deviations from the normal trend, it might not be the case for traders from Hong Kong. i.e., given that Hong Kong is a small open economy, short-term shocks in the exchange market might have a significant impact on its trading activities. 3 The choice of SITC 1-digit level bilateral trade data is dictated by the availability of the data. Use of more disaggregated trade data could be of interest, but it is not often published at the higher frequency level (monthly) level used in this analysis. 4 The bilateral trade data are seasonally adjusted using the X11 procedure in SAS and were converted to real terms by deflating US imports (defined in the exporting country's currency) from each of the countries by their respective consumer price index. 5For example, the real Canadian exchange rate per US$ is computed as (SPOT)*(CPICAN)/(CPIUSA), where, SPOT refers to the average monthly Canadian dollar spot price of the US$. 6 The Kalman Filter is a state-space model that provides a linear dynamic representation of a set of state variables, which are not directly observable, being subject to contamination by noise (CitationHarvey, 1989). Alternatively, it can also be taken as a procedure for calculating linear least squares forecasts of a state variable (vector) at time t+1 given data observed up to date t. 7The decomposition of the real exchange series in to two components in this paper is consistent with CitationMussa's (1982) stochastic generalization of CitationDornbusch's (1976) exchange rate overshooting model, in which exchange rate is decomposed into the permanent (fundamentals) and the transitory (innovations) components. 8The approach employed by CitationGotur (1985) and CitationKenen and Rodrick (1986) among others, includes import prices as an explanatory variable. Given the presence of the exchange rate variable in the model, the use of import price indices may seem redundant. However, here the import price index variable is not commodity-specific; thus, the index captures more than the effect of the specific SITC group's prices being estimated. 9 Industry-specific import price data are generally not available. In the absence of such information, the use of aggregate country-specific import price index has no substitute. 10 Full estimation results are available upon request from the author. 11Each of the control variables in the model was evaluated for serial autocorrelation and heteroscedasticity. Whenever the test statistic indicates strong evidence of first order autocorrelation, the sum of the squared error search (SSESEARCH) method suggested by CitationHildreth-Lu (1960) was employed using the Prais-Winsten transformed regression estimator. The SSESEARCH method specifies the value of rho (serial autocorrelation) that minimizes the sum of squared errors of the transformed equation. The Huber-White heteroscedasticity consistent standard errors method was used to correct for the presence of heteroscedasticity. 12 Only the coefficients estimates of the two volatility components are reported here for the sake of brevity and space. Detailed estimation results for the control variable in the SITC 1-digit level regressions and the corresponding test statistics for each of the coefficients are available upon request from the author.