Do Jumps Matter in Realized Volatility Modeling and Forecasting? Empirical Evidence and a New Model

Abstract

Building on an extensive empirical analysis I investigate the relevance of jumps and signed variations in predicting Realized Volatility. I show that properly accounting for intra-day volatility patterns and staleness sensibly reduces the identified jumps. Realized Variance decompositions based on intra-day return size and sign improve the in-sample fit of the models commonly adopted in empirical studies. I also introduce a novel specification based on a more informative decomposition of Realized Volatility, which offer improvements over standard models. From a forecasting perspective, the empirical evidence I report shows that most models, irrespective of their flexibility, are statistically equivalent in many cases. This result is confirmed with different samples, liquidity levels, forecast horizons and possible transformations of the dependent and explanatory variables.