Forecasting oil price volatility: Forecast combination versus shrinkage method

Abstract

In this paper, we compare the predictive ability between forecast combination and shrinkage method in the prediction of oil price volatility. Our investigation is based on the heterogeneous autoregressive (HAR) framework. Five combination approaches combine the individual forecasts generated by the HAR model and its various extensions, while two prevailing shrinkage methods, the elastic net and lasso, employ all the predictors in our HAR framework to generate the forecast of oil price volatility. The model confidence set (MCS) test shows that the elastic net and lasso have significantly better out-of-sample forecasting performance than not only the individual extended HAR models but also the combination approaches. This result is robust across a wide range of checks. In addition, we document that the elastic net and lasso also exhibit substantially higher directional accuracy. Furthermore, a mean-variance investor can realize sizeable economic gains by using the volatility forecasts based on the shrinkage methods to allocate her portfolio.