Combining predictive distributions of electricity prices: Does minimizing the CRPS lead to optimal decisions in day-ahead bidding?

In order to mitigate risks or increase profits from trading in day-ahead power markets, market participants use data-driven decision support techniques [12, 16, 17, 28]. For years, these have relied on point forecasts of the major variables of interest: loads (or demand for electricity), generation from renewable energy sources (RES), and electricity prices [10, 30]. However, as recently shown by Uniejewski and Weron [27], decisions based on probabilistic price forecasts, i.e., quantiles, prediction intervals or whole predictive distributions, can yield significantly higher profits. For the quantile-based bidding strategies considered in the Polish day-ahead power market, the profit obtained was from 5% to 19% higher than for the strategy based on point forecasts alone. Point forecasts are far more popular in the electricity price forecasting (EPF) literature, not only in a decision support context. As reported by Maciejowska et al. [18], probabilistic EPF was not part of the mainstream literature until the Global Energy Forecasting Competition in 2014 [9], and even now,