A new method of structural breaks detection in GARCH-models

Abstract

The article proposes a new method of structural breaks detection in time series in the piecewise-specified GARCH-models. The method is based on the moving likelihood ratio statistics. In case of absence of structural breaks lower and upper 95%- and 99%-bounds were found for the likelihood ratio statistics. The criterion of structural breaks based on these bounds has been worked out. Good properties of the proposed method are supported by Monte Carlo numerical experiments. In the framework of performed calculations it is obtained that the method detects the correct number of structural changes approximately in 88 % of cases. In case of correct detection of number of structural changes the moments of the structural breaks are estimated quite accurately. In the absence of structural breaks the proposed method falsely detects structural breaks quite rarely - around 2,5 % of cases. The method is tested on the real data when detecting the structural breaks in the volatility of returns for “Gazprom" ordinary shares.