A Markov switching long memory model of crude oil price return volatility

I propose a time series model that simultaneously captures long memory and Markov switching dynamics to analyze and forecast oil price return volatility. I compare the fit and forecasting performance of the model to that of a range of linear and nonlinear GARCH models widely adopted in the literature. Complexity-penalized likelihood criteria show that the Markov switching long memory model improves the description of the data. The out-of-sample results at several time horizons show that the model produces superior forecasts over those obtained from the selected GARCH competitors. Results are obtained using Patton's robust loss functions and the Hansen's superior predictive ability test. I conclude that the proposed model provides a useful alternative to the usually employed GARCH models.