In the first step of the analysis I estimated, by ordinary least squares, a number of
versions of equation (5) for the seven countries in the sample.
25
The analysis of the
residuals clearly showed the presence of conditional heteroskedasticity. Engle’s LM test
rejected the null hypothesis of absence of ARCH for every country. In the second step I
identified the order of the GARCH process for each of the countries, and I verified
stability. Finally, I estimated the system (5) and (6); in all cases the dummy variables for
inflation targeting ( DIT ) and for floating regime ( FLOAT ) were lagged one period; the
results, however, are not affected in a significant way when alternative lag structures are
used (including no lags). If the adoption of inflation targeting has indeed resulted in
increased nominal effective exchange rate instability, as some critics have argued, the
estimated coefficient of DIT would be significantly positive. If floating rates increase
exchange rate volatility, as one would expect under most circumstances, the estimated
coefficient of FLOAT would be significantly positive. In the estimation of the
conditional variance equations for Australia and Canada I did not include the FLOAT
variable. The reason for this is that both of these countries have had a floating regime
since the mid 1970s.