Predicting Stock Jumps and Crashes Using Options

Abstract

This paper investigates the informativeness of option-implied volatility and Greeks in forecasting extreme stock returns. Using a large data set of U.S. stocks and options from 1996 to 2022 and employing Light Gradient-Boosting Machine as a machine learning algorithm, we show that option characteristics, particularly implied volatility and delta, are strong predictors of extreme returns. The long–short portfolio utilizing option variables significantly outperforms a benchmark using only stock characteristics, suggesting that options provide information beyond what can be inferred from stock characteristics. Put options are revealed to be more informative than call options, and crashes are easier to predict than jumps.