Structural break detection in autoregressional conditional heteroskedasticity model: case of Student distribution

We consider two methods of structural break detection in a piecewise generalized model of autoregressive conditional heteroscedasticity. The first method is based on Kolmogorov–Smirnov statistics and is called KS-method. The second one is based on the cumulative sums and is called KL-method. In this paper, we compare the KS- and KL-methods under the assumption of Student conditional distribution of random errors. The results of our Monte Carlo experiments were as follows: the KL-method lost to the KS-method both in terms of the average probability of first type error and in terms of the average power structural break detection.