Investigating empirical bidding curves in the electricity spot market: Expected patterns vs anomalies?
Jilles De Blauwe, Xiaobing Zhang, Dogan Keles
Abstract
In recent years, the dynamics of spot market bidding have evolved dramatically, driven by the increasing penetration of variable renewable energy sources, new market actors, and increased cross-border trading. Understanding and monitoring the market are crucial to ensure efficient operation, but traditional methods have not kept pace with the increasing complexity. This study introduces a machine learning framework to investigate empirical bidding patterns. The clustering of bidding curves is combined with an extensive explanatory variable dataset to define what bidding can be expected under different market conditions. Then, deviations from the expected patterns are investigated by leveraging a combination of hierarchical clustering, multinomial logit modeling, and random forest classification. As such, we identify outliers in the bidding data and detect instances where the observed patterns diverge from those predicted from the explanatory variables. This framework is applied to several years of empirical bidding data from the Belgian, Danish, and German spot markets to generate insights into the drivers behind expected and anomalous bidding. We observe that spot market bidding is driven by renewable electricity generation, particularly from wind, while outliers are caused by extremes in the available electricity supply, produced by extreme renewable events. Beyond domestic factors, cross-border effects, such as generation (un)availabilities in neighboring zones, also caused the observed patterns to deviate significantly from those expected. The findings reveal insights into market dynamics and highlight the importance of considering bidding behavior and its drivers in market design.