"... In PAGE 8: ... The relatively short lag in the VAR reported in the previous section means that the dynamics fade away within a year at most, raising the possibility that it may be difficult to disentangle deliberately incomplete pass-through from the unintended consequences of imperfect foresight. An error-correction model of the change in (log) export prices is shown in the first column of Table4 . The explanatory variables are the lagged residuals (with mean adjusted to zero) from the preferred estimate of the cointegrating vector (EC), first- differences (prefixed D) of the I (1) variables, and the index of spare capacity.... In PAGE 8: ... To allow for the possibility of asymmetrical responses, positive and negative values of the explanatory variables were entered separately; those distinctions which reduced the standard error of the regression were retained. Table4 here The most i nteresting result is the coefficients of similar magnitude and opposite sign on the current and lagged change in the effective exchange rate. A Wald test shows that the restriction of coefficients which sum to zero cannot be rejected with anything resembling a reasonable probability.... In PAGE 8: ... Restricting the positive values of these variables, and of the appropriately insignificant intercept, to zero is acceptable jointly as well as separately, as the Wald test demonstrates. The second column of Table4 shows the results of imposing all the restrictions. The high (absolute) t-ratio on the error correction term confirms the inference of cointegration.... ..."

"... In PAGE 10: ... We reduce the dynamics of the ECM by successively omitting variables with the lowest t-statistics and re-estimating. The resulting ECM (reported in Table5 ) is simply: (2) Despite the remarkably simple specification, the R-squared statistic indicates that the price of oil can account for 6.7 per cent of the month-to-month variation and most of the longer horizon systematic movements in the real exchange rate.... In PAGE 10: ...7 per cent of the month-to-month variation and most of the longer horizon systematic movements in the real exchange rate. Specifically, the residual diagnostic tests yield little evidence of autocorrelation and autoregressive conditional heteroskedasticity (see residual diagnostic tests in Table5 ).10 The test for kurtosis, however, forces us to reject the null hypothesis of no kurtosis.... ..."