A Short-Term Electricity Price Forecasting on the Russian Market Using the SCARX Models Class

This research is focused on the approbation of Seasonal Component AutoRegressive with exogenous factors (SCARX) forecasting models class on two price area of the Russian electricity market. The SCARX model consists of extrapolation of long-term trend-seasonal component and independent forecasting of short-term seasonal-stochastic component of electricity price. The SCARX based on wavelet decomposition (SCARX-W) and Hodrick-Prescott filter (SCARX-HP) for the wide range of time-series smoothing parameters are compared with the usual autoregression model ARX and naive approach (based on assumption of the price similarity in the same weekday). The performance evaluation was carried out using weighted weekly and daily mean absolute errors, as well as the formal statistical procedure of the prediction ability comparison-Diebold-Mariano test (DM-test). The historical data of price and planed consumption in the Europe-Ural and Siberia price areas of the Russian electricity exchange were used for the numerical experiment, while testing period is 104 week or 728 days long. The study shows that in the Russian markets SCARX-W model exhibits more accurate forecast compare to SCARX-HP and ARX. The minimal weekly error achieved on Europe-Ural price area is 4,932%, daily error -4,997%. The same indicators for Siberia price area are 9,144% and 10,051%, correspondingly. The same results are proved by the formal DM-test carried for each hour in trading day. In order to overcome the problem of a priori selection of smoothing parameters, it is proposed to use various methods of forecast combinations.