Forecasting Realized US Stock Market Volatility: Is There a Role for Economic Policy Uncertainty?

We compare the contribution of various popular economic policy uncertainty (EPU) measures with that of widely-studied realized moments (realized leverage, realized skewness, realized kurtosis, realized good and bad volatilities, realized jumps, and realized up and down tail risks) to the performance of out-of-sample forecasts of stock market volatility of the United States (US) over the sample period from 2011 to 2023. To this end, we construct optimal forecasting models by combining the popular heterogeneous autoregressive realized volatility (HAR-RV) model with optimal stepwise predictor selection algorithms and shrinkage estimators (lasso, elastic net, and ridge regression), where we control for macroeconomic factors and sentiment as well. We find that realized moments improve out-of-sample forecasting performance relative to the baseline HAR-RV model. EPU measures do not add to forecasting performance beyond realized moments, and even deteriorate forecasting performance as the length of the forecast horizon increases. The punchline is that realized moments rather than EPU measures matter for forecasting stock market volatility.