Testing heteroscedasticity: are parametric ARCH models appropriate?

Abstract

Many researchers have used parametric ARCH models to specify the conditional variance of financial series. However, the usual tests do not provide any information on the form of the conditional variance. The objective of this paper is to present a test for heteroscedasticity, i.e. to decide whether the use of the parametric model can be justified. The test statistic is based on the distance between a non-parametric and a parametric estimator for the conditional variance. The critical values are calculated using a bootstrap method.