Forecasting German day-ahead electricity prices using multivariate time series models

Abstract

Using a newly available dataset about the unavailability of power plants and the in-feed of renewable energies to forecast day-ahead electricity prices at the German Power Exchange, this work shows that the predictive power increases considerably when including exogenous variables. While a similar univariate approach based on the year 2001 yielded a Mean Absolute Percentage Error of 13.2%, the use of the presented variables improved the forecasting error to 8.3%. Other findings of this work include that a model based on 24 individual time series produces smaller forecasting errors than one time series which includes all consecutive hours, that the selection of the in-sample and out-of-sample periods varies greatly between different works and that the use of OLS seems to be underestimated in the existing forecasting literature for electricity prices.