A1463
Title: Expectile regression averaging method in probabilistic forecasting of electricity prices
Authors: Joanna Janczura - Wroclaw University of Science and Technology (Poland) [presenting]
Abstract: A new approach for deriving probabilistic forecasts of electricity prices is proposed. To this end, the notion of expectiles is utilized, being a minimizer of an asymmetric least squares criterion. Since there exists a functional mapping between quantiles and expectiles, expectiles of a predicted distribution can be used for least squares estimation of the corresponding quantiles, i.e. the commonly used prediction intervals. Prediction intervals are calculated for the day-ahead electricity prices from different markets using the expectile regression averaging method built on individual point forecasts from different models. The results show that deriving prediction intervals from expectiles of the forecasted distribution of electricity prices outperform the quantile-based approaches if a variance stabilizing transformation is used.
