Forecasting Electricity Demand in Russia Considering Spatial Interaction

This study is contributed to fill the gap associated with the lack of relevant long-term forecasts for regional electricity consumption in Russia. Electricity demand was modeled in terms of spatial econometrics approach. Different types of models are provided: fixed and random effect models for panel data, both with and without spatial lag. A deterministic forecasting approach is considered. A comparison of the predictive qualities of spatial and non-spatial models is provided through the predictive quality metrics RMSE, MAE, MAPE, sMAPE. The analysis is based on the regional data on electricity consumption in 2012-2017. It is shown that spatial models for panel data are superior to conventional models. The positive impact of the gross regional product and population on the amount of electricity consumed in the region and the negative impact of the average annual temperature was confirmed as well.