Volatility Forecasting and Liquidity: Large Sample Evidence Using Individual Stocks

Abstract

We find that, on average, for the full sample of stocks comprising the S&P 100 index, the theoretically superior implied volatilities are less accurate forecasts than historically based forecasts of future volatilities. However, when we split our sample based on proxies for liquidity, such as market capitalization, option trading volume, and stock trading volume, we find that implied volatility forecasts are more accurate than historical volatility forecasts for more liquid stocks, and the reverse is true for the subsample of less liquid stocks. Additionally, we document that among historical measures, Parkinson's extreme value estimator and the adjusted mean absolute deviation are more accurate than the alternative historical estimators. Overall, our results suggest that for high liquidity stocks, the implied volatility measure is likely to provide a more accurate forecast of future volatility but for low liquidity stocks Parkinson's extreme value estimator and the adjusted mean absolute deviation using historical prices are likely to provide the best forecasts.