Forecasting Realized Volatility with Earnings Announcements and Overnight Returns

Abstract

In our study, we forecast realized volatility utilizing a large panel of stocks from the S&P 500, with the inclusion of overnight returns and earnings announcements. Our comparative analysis employs both the heterogeneous autoregressive model and gradient boosting. Upon evaluation, we ascertain that the inclusion of earnings announcements moderately enhances the precision of RV forecasting. Furthermore, our findings suggest that the gradient-boosting methodology demonstrates superior performance in comparison to the HAR model.