Causal machine learning reveals the impact of renewables on electricity prices

Researchers from Imperial College London and the UK National Energy System Operator have developed a causal analysis of how wind and solar generation relate to UK day-ahead and intraday electricity prices. Their study applies a causal machine learning framework to UK market data from 2018–2024 to estimate context-specific effects while accounting for confounding factors such as demand, fuel prices, seasonality, and time-of-day patterns.

 

“Our aim was to separate correlation from causation and to provide tools that the energy system operator and government stakeholders can use for system analysis and policy evaluation” said Dr. Davide Cacciarelli, lead author. “By residualizing both prices and renewable output with respect to a broad set of market drivers, we estimate how an additional unit of predicted wind or solar is associated with price movements at different penetration levels”.

 

They published their study on November 6, 2025, in iEnergy.

 

Non-linear price effects

The research group reports that price impacts vary with renewable penetration rather than remaining constant. For wind, the estimated effect on day-ahead prices follows a U-shaped pattern: additional predicted wind tends to reduce prices most at lower and higher penetration levels, with a comparatively smaller effect in the middle range. For solar, the analysis indicates consistently negative effects at low penetration, with diminishing marginal impacts as penetration rises. However, researchers expect similar patters to the ones observed for wind as solar installed capacity increases. Patterns observed for intraday prices are broadly consistent with the day-ahead results. The authors note that mid-range “price bumps” visible in descriptive statistics are largely attenuated once confounders such as demand and fuel prices are controlled, aligning the findings with the expected merit-order mechanism.

 

To obtain these estimates, the team used a local, partially linear Double Machine Learning approach that orthogonalizes both prices and predicted renewable generation against a rich set of covariates (including load, gas prices, carbon permits, hour, month, daylight hours, and installed capacity). Cross-fitting and bootstrap procedures were employed to reduce overfitting and quantify uncertainty. “These are observational data and results remain contingent on the available controls, but the framework helps separate correlation from plausible causal signals that are useful for analysis and planning” states Prof. Pierre Pinson.

 

Implications for market design

A time-resolved view suggests that the magnitude of renewable price effects has strengthened over recent years, consistent with rising penetration and evolving market fundamentals. According to the authors, this state dependence has practical consequences: support schemes, capacity planning, and pricing rules may benefit from designs that adapt to penetration levels and seasonal conditions rather than rely on a single average effect. The methodology and code are documented so that analysts can replicate or extend the pipeline to other price signals (e.g., intraday, balancing) and to other jurisdictions where comparable data are available.

 

The authors emphasize some limitations of the analysis. Unobserved confounding cannot be fully ruled out; estimates depend on the coverage and quality of inputs, particularly at very high penetration levels where observations are sparse. The team therefore frames the study as a measured, evidence-based contribution that can be scrutinized and iterated as data improve. “Our objective is practical,” Dr. Cacciarelli said, “provide transparent estimates that system operators, regulators and researchers can test, reuse and refine as the energy transition accelerates”.

 

The above research is published in iEnergy, which is a fully open access journal published by Tsinghua University Press. iEnergy publishes peer-reviewed high-quality research representing important advances of significance to emerging power systems. At its discretion, Tsinghua University Press will pay the open access fee for all published papers from 2022 to 2026.

 

About iEnergy

iEnergy is a quarterly journal launched on March 2022. It has published 4 volumes (13 issues). Authors come from 21 countries, including China, the United States, Australia, etc., and world’s top universities and research institutes, including University of Nebraska Lincoln, Columbia University, Imperial College of Science and Technology, Tsinghua University, etc. 12 published articles are written by academicians from various countries. The published papers have also attracted an overwhelming response and have been cited by 179 journals, including top journals in the field of power and energy like Nature Materials, Advanced Materials, Joule, Energy Environmental Science, etc., from 45 countries.

 

iEnergy publishes original research on exploring all aspects of power and energy, including any kind of technologies and applications from power generation, transmission, distribution, to conversion, utilization, and storage. iEnergy provides a platform for delivering cutting-edge advancements of sciences and technologies for the future-generation power and energy systems. It has been indexed by ESCI (Impact factor 5.1), Ei Compendex, Scopus (CiteScoreTracker 2024 7.4), Inspec, CAS, and DOAJ.

 

---

---