Do High-Frequency Volatility Methods Improve the Accuracies of Risk Forecasts? Evidence from Stock Indexes and Portfolio

Though the high-frequency volatility approaches are increasingly introduced to forecast financial risk in recent years, whether they can improve the accuracies of risk forecasts remains controversial. This paper compares the risk forecasting abilities of four pairs of low- and high-frequency volatility models, by calculating and evaluating the downside and upside value-at-risk and expected shortfall of stock indexes and portfolio. The empirical results show that, first, all the volatility models can well filter the serial dependence in the extremes, and the conditional standard deviation obtained from the GARCH model performs best in filtering the dependence. Secondly, the backtesting results of stock index and portfolio risk forecasts are consistent. More specifically, the traditional low-frequency volatility models produce more accurate risk forecasts in most cases, whereas the high-frequency volatility methods also manifest some advantages in the upside extreme risk forecasting.