Probabilistic Forecasting of Electricity Prices Using Kernel Regression

Abstract

Electricity price forecasting has become crucial for energy companies due to its fundamental importance for decision making processes and operational management. Electricity price time series exhibit variable means, significant volatility and spikes, which places high demands on forecasting models. Moreover, in recent years researchers and practitioners have come to understand the limitations of point forecasts and require models to generate probabilistic forecasts. In contrast to point forecasts, the probabilistic forecasts takes the form of a predictive probability distribution over future quantities or events of interest. In the paper the probabilistic forecasting model based on Nadaraya- Watson estimator is proposed. The model generates the point forecasts as 24-component vectors representing day-ahead electricity prices. The probabilistic forecasts are calculated as quantiles based on the residual distribution for historical data forecasts. The performance of the proposed model is validated by testing on data from the Polish electricity market.