A0418
Title: Enhancing electricity price forecasting accuracy: A novel filtering strategy for improved out-of-sample predictions
Authors: Andrea Cerasa - European Commission - Joint Research Centre (Italy) [presenting]
Alessandro Zani - Politecnico di Milano (Italy)
Abstract: Electricity price forecasting (EPF) has become a crucial component in energy companies' operational strategy. An original filtering strategy is introduced, aimed at refining the accuracy of day-ahead EPF where outliers identification and replacement rely on a model-based procedure. Extreme spikes are identified through the standardized residuals from rolling window robust regressions of prices against a predefined set of regressors. They are then replaced by the values fitted by the model. This method offers several benefits, such as eliminating the need for prior decomposition of price series, reducing the number of choices typical of standard filtering procedures for outlier identification and replacement, and mitigating the issues of masking and swamping by robust methods. The filtering strategy is applied to open-access benchmark datasets of 5-day-ahead markets, using state-of-the-art models and accuracy metrics, and compared to a baseline no-filtering strategy. Empirical results demonstrate that the proposed filtering approach can significantly enhance the precision of EPF while maintaining reasonable computation times. The proposed method offers an efficient pre-processing tool that, through more accurate price forecasts, can significantly improve the optimization of operational and strategic decision-making in the energy sector. It can valuably support energy traders, companies, and generators in mitigating risks and enhancing profitability in day-ahead markets.
