Combining GARCH Model Forecasts of Volatility With Alternative Weighting Schemes in Electricity Markets

In this paper, we study the methods of combining different volatility forecasts using various GARCH models. Given that the major risk exposure for many investors in energy is the volatility of the electricity price, our motivation stems from the fact that there is no single best model for forecasting such volatility. Ample evidence suggests that most of the individual models produce close volatility forecasts with marginal improvement (see e.g., Hansen and Lunde, 2005). In addition, one model may outperform its peers for one country's wholesale electricity price but not necessarily for other markets since all markets have unique features. Instead of picking one model for one market, we combine them all together for forecasting volatility with different weighting schemes in the spirit of Nowotarski et al. (2014a). We find that the Bayesian model averaging is a very stable and reliable combination method in producing statistically superior forecasts. The results are robust to a series of specifications. We believe that this is the first attempt to combine GARCH models to forecast volatility in electricity markets