Probabilistic forecasting of electricity prices using an augmented LMARX-model

Abstract

In this paper, we study the performance of prediction intervals in situations applicable to electricity markets. In order to do so we first introduce an extension of the logistic mixture autoregressive with exogenous variables (LMARX) model, see (Wong, Li, 2001), where we allow for multiplicative seasonality and lagged mixture probabilities. The reason for using this model is the prevalence of spikes in electricity prices. This feature creates a quickly varying, and sometimes bimodal, forecast distribution. The model is fitted to the price data from the electricity market forecasting competition GEFCom2014. Additionally, we compare the outcomes of our presumably more accurate representation of reality, the LMARX model, with other widely utilized approaches that have been employed in the literature.