The Role of Dynamic Specification in Forecasting Volatility in the Presence of Jumps and Noisy High-Frequency Data

This paper considers the performance of different long-memory dynamic models when forecasting volatility in the stock market using implied volatility as an exogenous variable in the information set. Observed volatility is separated into its continuous and jump components in a framework that allows for consistent estimation in the presence of market microstructure noise. A comparison between a class of HAR- and ARFIMA models is facilitated on the basis of out-of-sample forecasting performance. Implied volatility conveys incremental information about future volatility in both specifications, improving performance both in- and out-of-sample for all models. Furthermore, the ARFIMA class of models dominates the HAR speciations in terms of out-of-sample performance both with and without implied volatility in the information set. A vectorized ARFIMA (vecARFIMA) model is introduced to control for possible endogeneity issues. This model is compared to a vecHAR specification, re-enforcing the results from the single equation framework.