Long Memory and Structural Breaks in Realized Volatility: An Irreversible Markov Switching Approach

Abstract

This paper examines whether it is possible to improve volatility forecasts at monthly and quarterly frequency by (i): Incorporating structural breaks in the parameters that govern volatility dynamics, (ii): Accounting for model uncertainty through a Bayesian model averaging approach. In other words, our framework simultaneously addresses structural breaks in the model parameters and uncertainty regarding the inclusion of potential predictors. Structural breaks are modeled through mixture distributions for state innovations (thus the name mixture innovation) of linear Gaussian state-space models. Results indicate that Bayesian model averaging that also incorporates structural breaks in the model parameters provides improvements in point and density forecasts compared to other alternatives. Furthermore, the predictive power associated with structural break models appear to be concentrated around the onset of recessions and periods of market turmoil such as the Stock Market Crash of 1987.