Model-Based versus Model-Free Implied Volatility:Evidence from North American, European, and Asian Index Option Markets

Abstract

The Black–Scholes (BS) implied volatility is well known to be a rather noisy forecast of future volatility. On the one hand, it has the advantage of potentially incorporating current information that cannot be captured by any statistical technique that relies on historical data. On the other hand, it is “risk neutral,” meaning that the market’s true volatility forecasts are distorted by risk preferences. But no other pricing model has taken the place of BS, so the forecasting performance “horse race” has been between BS IVs versus statistical models such as GARCH. In recent years, “model-free” implied volatility (MFIV) is becoming more common. MFIV is extracted from the entire set of current option prices without the need to assume any specific pricing model. That is certainly a desirable property, but there are a variety of methodological issues in calculating and using MFIV. In this article, Biktimirov and Wang conduct a serious horse race among these models, looking at how much each one contributes to forecast accuracy in predicting future volatility in 13 major international stock markets over a 10-year period. The winner, somewhat surprisingly, is BS IV, both in terms of in-sample “encompassing” models that include several forecasts in the same combined specification and also in out-of-sample forecasting.