Garch Models and the Stochastic Process Underlying Exchange Rate Price Changes

This study investigates the extent of the contribution of the original GARCH model to our understanding of the stochastic process underlying exchange rate price changes, and examines if the movement of current research to GARCH type models exclusively is warranted. GARCH(1,1) parameters are calculated on a yearly basis and used to standardize the exchange rate price change data. Frequency distributions and statistical tests indicate that independence still exists after standardization. This indicates that GARCH type models alone are inadequate since all are similar in form, and would have difficulty in accounting for such independence. It could be argued that the poor performance of the GARCH model is due to the models incorrect assumption of a normal distribution. This argument is tested by comparing the GARCH standardized data with mean -variance standardized data, which makes no assumptions about the distributional form. Results of likelihood ratio tests, question the significance of conditional volatility, in explaining exchange rate price changes. Curiously there are cases where GARCH e2(t-1) parameters are significant when tests for first-order heteroskedasticity are not significant; this suggests that the model may be misspecified. Overall, results indicate that although previous research finds that volatility clustering plays a role in determining the stochastic process, it is not the dominate factor. This study questions the contribution of the GARCH type models. We discuss implications of our results.