Probabilistic Forecasting of Reserve Power Prices in Germany using Quantile Regression

Optimization models for the bidding in the German reserve power markets need an assumption for the probability density of the market prices. Even though probabilistic forecasting is becoming increasingly popular in the field of energy forecasting, this application does not seem to be covered in the literature. This paper focuses on probabilistic forecasts developed specifically for this application. A quantile regression model with natural cubic splines is proposed and backtested with historical prices from the German secondary reserve market. In addition, problems such as quantile crossing as well as advantages and disadvantages, compared to other probabilistic forecasting models, are discussed. The assessment by the continuous ranked probability score shows that the quantile regression method outperforms the benchmark ARMA model. The significance is demonstrated with the Diebold-Mariano test.