Модель MS-LASSO для прогнозирования волатильности: преимущества в условиях нелинейности

Экономический журнал Высшей школы экономики. 2025. Т. 29. № 4. С. 691–716.

Гуревич А. М., Пьянкова М. В., Skorobogatov A., Свиридов О. И.

Forecasting and analyzing the volatility of financial instruments is one of the fundamental tasks in stock market operations. The literature most often employs linear models for predicting market volatility. However, this tool may not be the most suitable for the stated objective, as the market is inherently non-constant, with its volatility exhibiting distinct periods of high and low values. One method that allows for accounting of this instability is the Markov regime-switching model, which permits the market to exist in at least two states: high and low volatility. When combined with regularization techniques that guard against overfitting, the Markov-switching model can demonstrate superior forecasting performance compared to traditional linear models. The present study is dedicated to demonstrating this very fact. We model and forecast stock market volatility using both simulated and real-world data. For real-world examples, data from the Moscow Exchange (MOEX) and the NASDAQ exchange were taken. Simulations demonstrate that the Markov-switching model with the application of LASSO regularization forecasts at least as accurately as the linear model on linear data and significantly outperforms it on nonlinear data. The results on real data reveal that for the Russian stock market, characterized by nonlinear dependencies in the data, a model assuming a linear relationship possesses low predictive power. The Markov-switching model enhances the accuracy of volatility forecasts in the presence of nonlinear data relationships. Conversely, for the NASDAQ exchange, where the data linkages are predominantly linear, the Markov model does not yield substantial advantages over its linear counterpart.

Keywords: регуляризация, реализованная волатильность, модель Маркова с переключением режимов, нефтяные шоки, нелинейная взаимосвязь