A Mixed Time-Series & Machine Learning Approach for Price Forecasting in the Swedish Ancillary Market

Abstract: This study aims to forecast the Swedish FCR-D Down A2 market prices through a hybrid model combining a volatility model and a machine learning approach, and compares its performance with a standalone machine learning model. We further examine the impact of different lag orders (1-Hr vs. 24-Hr) on volatility estimates and forecast performance. Evaluation metrics include root of mean squared error (RMSE) and mean absolute error (MAE). Results suggest that the hybrid model incorporating day-to-day volatility changes effectively predicts market price spikes, providing a competitive edge for energy traders. However, this model falls short during docile market periods, indicating a need for supplementary models. The study concludes that the current hybrid model does not yield significant forecast performance increases compared to simpler machine learning models, prompting further investigation into the optimization of such hybrid models for effective market price forecasting.