Online Multivariate Distributional Regression for Probabilistic Electricity Price Forecasting

Simon Hirsch, University of Duisburg-Essen / Statkraft Trading GmbH

Abstract: The increasing penetration of renewable energy sources in energy markets demands probabilistic forecast for prices to enable accurate decision-making. At the same time, the amount of data available for forecasting is increasing rapidly. To model the conditional volatility and tail behaviour of electricity prices, distributional regression methods have been successfully applied. However, their estimation is computationally expensive, making them impractical for high-dimensional data and streaming applications. Against this background, we present an online and regularized algorithm for multivariate distributional regression. We validate our approach in a forecasting study for the 24–dimensional vector of German hourly electricity prices. Our results show superior forecasting performance and can be estimated on standard laptops.