Parametric Distributional Flexibility and Conditional Variance Models with an Application to Hourly Exchange Rates

Abstract

This paper builds on the ARCH approach for modeling distributions with time-varying conditional variance by using the generalized Student t distribution. The distribution offers flexibility in modeling both leptokurtosis and asymmetry (characteristics seen in high-frequency financial time series data), nests the standard normal and Student t distributions, and is related to the Gram Charlier and mixture distributions. An empirical ARCH model based on this distribution is formulated and estimated using hourly exchange rate returns for four currencies. The generalized Student t is found to better model the empirical conditional and unconditional distributions than other distributional specifications.