Wholesale electricity market pricing model design: Layering the transparency

The energy transition is challenging traditional wholesale electricity market pricing models. Divergences in modelling outcomes hinder comparisons across regional studies, often attributed to variations in model designs and data availability. This is further complicated by theoretical model designs that are insufficiently aligned with real-world constraints, such as limited access to market data, confidential information, or high-quality datasets. Rather than building a market model and then rummaging for suitable datasets, this study inverts the approach and layers data into the model design. The framework was applied in a longitudinal study across three study regions, Alberta (AESO), California (CAISO), and Western Australia (WEM), which have experienced high growth in renewable generation. The technical implementation employed a broad range of established machine learning methods (OLS, MARS, ANN, GAMLSS, XGBoost) alongside a structured probabilistic approach (Bayesian Belief Networks). The approach was found to be feasible as several models consistently provided the optimal design across regions and configurations implemented, as such supporting the assertion that models could be designed, configured and compared using this layered approach. Matching model complexity to all available data should serve as a guide, since higher-performing models do not necessarily require access to extensive datasets. Finally, cross-market comparisons showed that different market characteristics had a greater influence on model accuracy, illustrating that real-world data constraints need to guide model development.