Combining realized volatility estimators based on economic performance

We propose a forecast combination scheme that employs time-varying weights, which depend on the financial decision for which the forecasts are used. Combination weights are computed through three alternative economic performance measures, a utility-based function, an option staddle trading strategy, and a risk management loss function. We apply our model combination approach to various realized volatility estimators of the S&P500 and compare the forecasting performance of the combinations based on economic criteria to that of a variety of existing combination methods based on statistical loss functions. The results imply that our proposed combination schemes result to superior economic performance compared with the individual measures and statistical combination methods. Our findings are supported by a wide range of robustness checks and extensions Moreover, the suggested combination framework holds significant economic value during times of crisis and high volatility.