Loss functions in regression models: Impact on profits and risk in day-ahead electricity trading

We study the impact of the loss function used to estimate the parameters of a regression-type model on profits and risk in day-ahead electricity trading. To provide practical insights, we consider a strategy that incorporates battery storage and includes realistic operating costs in the calculation of revenues. Using 2021-2024 data from the German market as the testing ground, we provide evidence that minimizing a loss function that combines absolute errors with a quadratic penalty for price spread predictions of the opposite sign is the preferred option. Forecasts based on the introduced directional loss function repeatedly and in the majority of cases yield trading decisions that outperform those based on predictions from models estimated using squared, absolute, percentage, or asymmetric losses, as measured by the Sharpe ratio and profits per trade. • Insights on how loss functions affect profits and risk in electricity trading. • The first study to consider realistic BESS operating costs in intraday trading. • A new loss function that penalizes price spread predictions of wrong sign. • This function leads to higher profits than MSE, MAE and sMAPE-based forecasts.