both Poland and Russia the White test for heteroscedasticity and the test for serial correlation and
ARCH type errors indicate we cannot reject heteroscedasticity, serial correlation, and ARCH. As
a result we then estimate the models with exponential generalized autoregressive conditional
heteroscedasticity (EGARCH) procedures,28 with lags of the first differences of the fundamentals
and the exogenous variables. The coefficient estimates of the mean equation are presented as
specification 2 in Table 7 for Poland and Table 8 for Russia. In these specifications, all the
coefficients are statistically significant.29
Russia underwent a regime change immediately following the currency crisis. The
structural break in the long run equilibrium real exchange rate appears obvious and the Chow
break-point test on the cointegrating equation (specification 1, Table 6)30 for Russia with the
crisis month (August 1998) as the break point strongly rejects the null hypothesis of no structural
change. Thus, in order to capture the distinctive features of the pre-crisis period, another error
correction equation is specified for Russia with the inclusion of a dummy variable equal to zero
in the pre-crisis period, and equal to one from the crisis period onward. Specification 3 in Table
8 presents the modified equation. The dummy is significant and all variables that are significant
in specification 2 of Table 8 are significant in specification 3, Table 8.31
Specification 4 in Table 8 reports the coefficient estimates of the mean equation for the
error correction model for Russia in the pre-crisis period. Like the other specifications, the final
28 In the tables EGARCH(m, n) is exponential GARCH, with m lags of the variance term and n lags of the error
term in the conditional variance equation.
29 For Russia we drop TOT due to colinearity.
30 The Chow break point test could not be conducted on specification 2 due to insufficient number of
observations. The Chow break point test also confirms a structural break in the short run error correction models
reported in Table 8.
31 We also tried another specification with cross terms involving dummy that produced mixed results.
28