Sensitivity of the Causality in Variance Test to the Garch (1,1) Parameters

Abstract

In this paper we discuss the sensitivity of causality of variance tests to the GARCH\((1,1)\) parameters. The idea of study is to investigate the impact of a number of volatile data sets on volatility spillover tests. We investigate a type of data generating process, AR(1)-GARCH(1,1), with an extensive set of Monte Carlo simulations. It is found that causation pattern, due to causality between two series, is influenced by the intensity of volatility clustering. Two testing procedures are applied for testing causality in the variance. We notice a severe size and power distortion when the clustering parameter is high and when the process is near integration. Furthermore, whenever there is a severe size distortion, there is a serial autocorrelation in the standardized residuals. This is seen when the asymptotic distribution of the statistics is used to define a critical region. So, instead of relying on the asymptotic distribution, we calculate the percentiles of the test statistic with the null hypothesis of no spillover effect and use them as a critical region for both size and power. We observe a significant improvement in the results. (Less)