Volatility Forecasting with Machine Learning and Intraday Commonality

We apply machine learning models to forecast intraday realized volatility (RV), by exploiting commonality in via pooling stock data together, and incorporating a proxy for the market volatility. Neural networks dominate linear regressions tree-based terms of performance, due their ability uncover model complex latent interactions among variables. Our findings remain robust when we trained new stocks that have not been included training set, thus providing empirical evidence universal mechanism stocks. Finally, propose approach forecasting 1-day-ahead RVs using past as predictors, highlight interesting time-of-day effects aid mechanism. The results demonstrate proposed methodology yields superior out-of-sample forecasts over strong set traditional baselines only rely on daily RVs.