Modeling and forecasting of the long-term seasonal component of the EEX and Nord Pool spot prices

Abstract

We present the results of a study on modeling and forecasting of the long-term seasonal component (LTSC) of electricity spot prices. We consider a vast array of models including linear regressions, monthly dummies, sinusoidal decompositions and wavelet smoothers. We find that in terms of forecasting EEX and Nord Pool spot prices up to a year ahead, wavelet-based models significantly outperform all considered piecewise constant and sine-based models. This result challenges the traditional approach to deseasonalize spot electricity prices by fitting monthly dummies or sinusoidal functions. We also find that extending the calibration window up to four years does not improve the results; two-and especially three-year windows lead to better spot price forecasts.

Multistage optimization filter for trend‐based short‐term forecasting

“…Lastly, there is a scarcity of well-performing trend estimation methods, which is a hurdle for trend-based forecasting scheme and is addressed in this paper. The best method suggested by the literature for trend estimation is wavelet decomposition (Nowotarski et al, 2013), but due to the complexity of the wavelets, as discussed in Section 1, new methods need to be explored for trend estimation that could improve the forecasts.…”