Forecast Combinations For Realized Volatility In Presence Of Structural Breaks

In this paper the problem of instability due to changes in the parameters of some Realized Volatility (RV) models has been addressed. The analysis is based on 5-minute RV of four U.S. stock market indices. Three different representations of the log-RV have been considered and, for each of them, the parameter instability has been detected by using the recursive estimates test. In order to analyse how instabilities in the parameters affect the forecasting performance, an out-of-sample forecasting exercise has been performed. In particular, several forecast combinations, designed to accommodate potential structural breaks, have been considered. All of them are based on different estimation windows, with alternative weighting schemes, and do not take into account explicitly estimated break dates. The model con_dence set has been used to compare the forecasting performances of the proposed approaches. Our analysis gives empirical evidences of the effectiveness of the combinations which make adjustments for accounting the possible most recent break point.