?
KS-метод обнаружения структурного сдвига в GARCH(1,1) моделях
We propose a new method of a structural break detection for GARCH(1,1) model. This new method is called the KS method since it is based on Kolmogorov-Smirnov statistics. By using Monte-Carlo experiments we show that the KS method has good statistical properties. We compare our method with three well-known CUSUM methods: (Kokoszka, Leipus, 1999) referred to as KT method, (Inclán, Tiao, 1994) referred to as IT method, and (Lee et al., 2004) referred to as LTM method. To make the experiments closer to real conditions, we generate GARCH processes with coefficients estimated on 26 Russian stocks time series. The results of the experiments prove the KL method to have the highest power on average to detect structural breaks among the other methods. Our KS method has a slightly lower power while IT and LTM methods are dramatically less powerful. However, we find a significant drawback of KL method: in some cases its probability of type I error may reach 42%. As opposed to KL method, our KS method demonstrates lower probability of type I error on average. As a result, we suggest that our method is highly competitive and may be placed somewhere in between the KL method which has high power and high probability of type I error, and IT and LTM methods which have low power and also low probability of type I error.