Title: More than just errors: Accounting for state dependent mean reversion in volatility forecasting
Authors: Giampiero Gallo - NYU in Florence (Italy) [presenting]
Edoardo Otranto - University of Messina (Italy)
Fabrizio Cipollini - University of Florence (Italy)
Abstract: Recent results are revisited, where an improvement of forecasting realized volatility is attained by inserting an estimate of quarticity in a volatility model. In our approach, quarticity is recognized to be strongly correlated with volatility itself, therefore introducing a curvature effect in a volatility quadratic equation, whereby higher volatility induces a stronger mean reversion effect. Moreover, this state dependence can be linked to a low frequency component of volatility forecasting which is suitably captured by Smooth Transition of Markov Switching behavior. Our results on 30 DJ components reveal that when other forms of state dependent behavior in volatility modeling are accounted for, the quarticity term generally loses importance, showing that measurement errors are providing only incomplete explanation for this state dependent behavior.