Investigating Calibration Challenges in Probabilistic Electricity Price Forecasting
Hadeer Elashhab ⋅ Jan Niklas Lettner ⋅ Benjamin Schäfer
Abstract
Proper scoring rules are used to train probabilistic forecasts on the premise that they are minimized only by the true predictive distribution. We show this guarantee is weak in practice: across two model families for day-ahead electricity price forecasting, the objective alone changes calibration error by up to a factor of two, and a loss designed to improve calibration makes it markedly worse under temporal dependence.
Chat is not available.
Successful Page Load