Title: Volatility forecasting gains from jumps: On the effect of the nature of the jumps
Authors: Marwan Izzeldin - Lancaster University Management School (United Kingdom)
Rodrigo Hizmeri - Lancaster University (United Kingdom)
Anthony Murphy - Federal Reserve Bank of Dallas (United States)
Mike Tsionas - Lancaster University (United Kingdom)
Abstract: A growing literature documents gains in forecasting return volatility using high frequency data. More recently, attention has shifted to the identification and use of jumps in forecasting. We partition jumps into finite and infinite variance (infrequent large vs frequency small) jumps, and use an extended heterogeneous, autoregressive (HAR) model to examine the forecasting gains at various horizons from this decomposition. Inter alia, we consider intraday periodicity, robust-to-noise volatility measures, different sampling frequencies and regime changes. The forecasting gains from using finite jumps exceed those from using infinite jumps except at high frequencies. Using robust-to-noise volatility measures results in better forecasts. Filtering periodicity adversely affects both types of jump measures, but affects finite jumps to a greater extent. Based on the Giacomini and White test, we also find that more significant differences in forecasting gains at higher frequencies. Finally, during the crisis period, decomposing jumps into finite and infinite components improves the forecasts.