Methods of Structural Break Detection in GARCH Models and Their Applications to Modeling of Financial Instruments Volatility

This paper proposes a hybrid method of structural breaks detection using an expert approach and two statistical algorithms: the KL-ICSS algorithm based on the cumulative sums method and the Fukuda algorithm based on the bayesian information criterion (BIC). The proposed method was tested on real data as a way of detecting structural breaks in the volatility of returns on common stocks of Russian oil and gas companies. Based on the results of the applied method, structural breaks in volatility were detected for all analyzed companies. Except for PJSC Novatek, the identified structural breaks had a significant impact on the parameters of the GARCH model, which confirms the variability of volatility structure over long time periods. The proposed method consists of combining expert and statistical approaches, and has shown its effectiveness when tested on real data. It was also found that both significant external events in the market and significant corporate news of the company can cause structural breaks in volatility.