Forecasting realized volatility in electricity markets using logistic smooth transition heterogeneous autoregressive models

We apply the non-parametric realized volatility technique and the associated jump detection test to measure volatility and jumps in electricity prices. Then, we propose a group of logistic smooth transition heterogeneous autoregressive (LSTHAR) models of realized volatility. The models can simultaneously approximate long memory behavior and describe sign and size asymmetries. They differ in the underlying heterogeneous autoregressive structure and the transition variable specification. The out-of-sample forecast accuracy of the LSTHAR models is evaluated through the Diebold–Mariano test and the superior predictive ability test, in terms of the mean square error and the mean absolute error. Using high-frequency prices from the Australian New South Wales (NSW) electricity market as empirical data, we draw the following conclusions. 1) Introducing the logistic smooth transition structure with appropriate transition variable specification to the heterogeneous autoregressive models improves volatility forecasts. 2) Overall, the LSTHAR model that uses the sum of Beta function weighted past returns as the transition variable and includes past daily jumps as a predictor is the superior model for predicting volatility in the NSW market. This model significantly outperforms the others.