Probabilistic Forecasting for Day-ahead Electricity Prices, Battery Trading Strategies and the Economic Evaluation of Predictive Accuracy

Battle Count: 52

Highly relevant to quantitative trading in energy markets. The paper directly addresses battery storage arbitrage trading strategies, risk-averse optimization (CVaR), probabilistic forecasting for trading decisions, and the critical question of whether statistical forecast quality translates to economic performance. The theoretical results showing QBTS can be gamed by overdispersed forecasts and the low discriminatory power of battery optimization for model selection are directly applicable to trading strategy evaluation. The connection to Markowitz portfolio optimization and the development of proper scoring rules for decision quality are broadly relevant to quantitative finance.

The paper involves multiple complex components: distributional regression with GAMLSS, Gaussian Copula modeling, mixed-integer linear programming for battery optimization, dynamic programming approaches, and extensive simulation studies. The MILP formulation with binary variables for bid placement, charge/discharge constraints, and cycle limits adds computational complexity. The online learning algorithm for expanding window estimation and the scenario generation procedure with rank-reordering for dependence structure isolation are non-trivial. However, the authors use established tools (pyomo, HIGHS solver) and provide algorithm descriptions.

The paper provides detailed model equations, algorithm descriptions, and uses publicly available data from ENTSO-E and investing.com (provided by Lipiecki et al., 2024). The implementation uses pyomo and HIGHS solver. Training/test split is clearly defined (2015-2018 training, 2018-2023 testing). Supplementary material includes additional figures and algorithm descriptions. However, no explicit GitHub repository link is provided in the main text.