Discriminating between GARCH and stochastic volatility via nonnested hypotheses testing

Abstract

GARCH- and Stochastic Volatility (SV)-models are the main workhorses for describing unobserved volatility in asset returns. Because economic theory behind these models is not the same and estimating SV-models is much more difficult, discriminating between these two rival models is of interest. This paper suggests a nonnested testing procedure going back to Davidson and MacKinnon (1981) that does not implicitly assume that one of the models is the correct one. We illustrate the proposed test by applying it to ten daily stock index return series and five exchange rate return series.