Enhancing Accuracy of Probabilistic Electricity Price Forecasting: A Comparative Study of Novel Quantile Regression Averaging Generalization

This paper presents a comparative study of two novel approaches for probabilistic electricity price forecasting, namely Smoothing Quantile Regression Averaging (SQRA) and LASSO Quantile Regression Averaging (LQRA). Both methods are generalizations of the well-established Quantile Regression Averaging (QRA) model. In this study, a novel LASSO Quantile Regression Machine (LQRM) is introduced, and the Bayesian Information Criterion (BIC) is proposed as a solution to determine the optimal tuning parameter value for the LQRM method. The models are evaluated on Polish power market datasets using 25 point forecasts obtained from calibration windows of varying lengths. The results show that both methods significantly improve forecast accuracy, with a slight advantage for the SQRA approach. Both generalizations offer promising potential for enhancing accuracy of electricity price forecasts, and as a result improves the risk management and supporting decision-making in electricity markets.