Which implied volatilities contain more information? Evidence from China

This paper aims to examine the information capacity of implied volatility (IV) estimated with model‐based and model‐free methods. Among the model‐free approaches, we further divide them into direct estimation (DE) and indirect estimation (INDE) approaches according to the computational process. We study the predictability of different implied volatilities on realized volatility over various forecasting horizons based on the Chinese option data. The out‐of‐sample results indicate that all the IVs contain the information transmitted by option market, and have predictive ability on the realized volatility. Using three different loss functions, we find that the implied volatilities estimated with DE approach contain more information. When considering different market scenarios, we find model‐based implied volatility performs best during the stable period and the DE approach always works well no matter the investor sentiment is normal or abnormal. Moreover, we also discuss the predictability of implied volatility on crash risk. Our results remain robust by using alternative realized volatility and alternative rolling windows.