The Information Content of Realized Volatility Forecasts

Torben Andersen*(Northwestern University), Per Frederiksen(Nordea Bank), Arne Staal from(Lehman Brothers)


We examine the relative information content of monthly volatility forecasts derived from option prices and time-series forecasts of realized volatility. If volatility risk is priced the return variance process is different under the physical and risk-neutral pricing measures, which has important implications for the interpretation of implied volatility as an expectation of future volatility. We use recent developments in volatility measurement and modeling to capture and forecast the variance process. Our results suggest that forecasts of future realized volatility based on past realized volatility are unbiased, more efficient than VIX model-free implied volatility, and roughly as efficient as Black and Scholes implied volatility. Forecasts of future volatility based on historical realized volatility contain incremental information relative to both model-free and Black-Scholes implied volatility. This finding can be explained in the context of priced variance risk. Given the finding of different information contained in individual volatility forecasts, we show that conditional volatility forecasts can be improved upon by combining individual measures. We also show that predictive regressions of realized volatility on option implied volatility are likely to be misspecified due to the presence of long memory in such measures and the variance risk premium.