A comparative study of Machine Learning methods for Power Futures Curves prediction

Accurate electricity price forecasting is a critical issue for policy makers and market players, especially during periods of market turmoil. Nevertheless, the development of robust forecasting methodologies still represents a challenging issue due to high volatility and irregular cycles in price dynamics that characterize international power markets. This work studies a Machine Learning (ML) approach to predict the term structure of power futures prices. In detail, four ML models, i.e. Generalized Regression (GR-NNs), Nonlinear Autoregressive (NAR–NNs), NeuralProphet (NP-NN) and Long Short-Term Memory Neural Networks (LSTM-NN) are examined for the 1 and 5 days-ahead pointwise forecast. To investigate the models predictive performances, the study has been run using daily settlement prices of electricity futures contracts traded on the European Energy Exchange (EEX) in the period spanning from March 2017 to October 2023.