Learning short-term electricity market behaviours
Loading...
Files
Full Text E-thesis
Date
2025-12-31
Authors
Collins, Joseph
Journal Title
Journal ISSN
Volume Title
Publisher
University College Cork
Published Version
Abstract
Electricity markets across many jurisdictions have undergone substantial restructuring in recent decades. In some regions, this has been accompanied by significant shifts in the electricity generation capacity landscape, for example the large-scale deployment of variable renewable generation or the closure of nuclear and coal assets in parts of Europe. These developments have contributed to increasingly dynamic trading environments, particularly in short-term electricity markets. It is within this context that our research is situated.
Our work focuses on two distinct but related areas. The first involves forecasting the evolution of a short-term electricity market using granular participant-level datasets. The second presents empirical analyses within the same market environment, complementing and contextualising the forecasting strand. In the forecasting component, which is conducted in a Day-Ahead market setting, we apply statistical, machine learning and deep learning techniques to predict individual participant commercial behaviours; subsequently we develop an aggregate view of electricity demand and supply. This granular approach addresses challenges associated with bottom-up (i.e. fundamental) models, namely the reliance on and validity of expert-driven input assumptions and the associated model calibration efforts. As part of this work we also develop an explainability framework to interpret differences between forecast and actual outcomes in addition to algorithms to manage high-dimensional participant order data. Our findings show that while the bottom-up approach effectively captures the complexity and evolution of participant behaviour, and by extension the market structure, its price forecasting accuracy is lower than that of top-down models in the literature (which typically focus solely on price). Using the explainability framework, we trace this underperformance to a cohort of speculative units exhibiting highly dynamic and difficult-to-forecast behaviours. Building on this, the second strand of our research explores how speculator participation in the market has evolved over time. We analyse factors such as marginality rate (the proportion of time a speculator sets the marginal price) and overall market share. We observe a clear increase in speculator participation over the study period and find that, relative to their market share, these units are marginal in a large number of trading periods. We also present an intuitive metric for analysing Day-Ahead market price inertia (the market price’s ability to withstand small changes in demand or supply). We view the latter analysis as having conceptually similarities to order book analysis in continuous trading environments.
The relevance of the research is as follows. First, the modelling framework is, to our knowledge, the first of its kind applied at this scale offering a novel contribution to the literature and a new perspective on how participant-level data can be used (along with its limitations) in electricity market analysis. This is important given the continued use of bottom-up models by both regulatory authorities and market participants. Second, the empirical analysis provides concrete metrics on both speculator activity within a European market context and market price inertia further highlighting the susceptibility of market prices to jumps given small changes in demand or supply.
The research is informed by both practical industry experience and engagement with the broader academic literature. As such, it is written for two audiences: practitioners, who may benefit from its applied insights, and academics, to whom it offers methodological contributions and new perspectives.
Description
Keywords
Machine-learning , Neural Networks , Electricity price forecasting , Day-Ahead markets , Speculators
Citation
Collins, J. 2025. Learning short-term electricity market behaviours. PhD Thesis, University College Cork.