Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Evaluation of Electricity Market Clearing Mechanisms via Reinforcement Learning: Prices, Remuneration and Competitive Dynamics

Andrea Altamura, Fabrizio Lacalandra, Antonio Frangioni, Massimo La Scala

TL;DR

This paper evaluates Segmented Pay‑as‑Clear (SPaC) as a market design alternative to PaC and PaB in the European electricity market by combining bilevel optimization and reinforcement learning (Q‑Learning) to model strategic operator behavior. It analyzes two scenarios—PNIEC2030 and a 10‑operator portfolio derived from 2024 public offers—showing that SPaC reduces inframarginal rents and price volatility while maintaining participation incentives and displaying robustness to oligopolistic power. Under RL‑driven bidding, PaB becomes the most rent‑extractive and costly regime, PaC remains costly due to uniform pricing, and SPaC achieves a favorable middle ground with lower price spikes and improved allocation efficiency. The framework serves as a regulator‑oriented ex‑ante tool to assess market design reforms and shows SPaC’s potential to balance efficiency, investment signals, and equity in both Italian and broader European contexts.

Abstract

The Pay-as-Clear (PaC) mechanism currently used in the European electricity market can generate significant submarginal profits for renewable sources when the clearing price is determined by the marginal offers of gas-fired generation units and the cost of natural gas exceeds certain levels. This exposes consumers to high price volatility related to the cost of natural gas. This report analyzes the recently proposed Segmented Pay-as-Clear (SPaC) mechanism as a market alternative, evaluating its system cost-effectiveness through simulations based on Reinforcement Learning (Q-Learning) to model the strategic behavior of operators. Three market models are compared, the two classic Pay-as-Clear (PaC) and Pay-as-Bid (PaB) along with SPaC, under two scenarios: a simplified one based on the 2030 NECP objectives and one built on the portfolios of ten operators obtained from the GME's 2024 public offers. The results show that the SPaC market clearing mechanism reduces intramarginal profits and price volatility compared to PaC, while maintaining fair participation incentives for all operators, and is more robust than PaB to the exercise of market power in oligopolistic contexts. The developed framework can serve as a support tool for regulators and policymakers in the evaluation of proposals for market design reforms.

Evaluation of Electricity Market Clearing Mechanisms via Reinforcement Learning: Prices, Remuneration and Competitive Dynamics

TL;DR

This paper evaluates Segmented Pay‑as‑Clear (SPaC) as a market design alternative to PaC and PaB in the European electricity market by combining bilevel optimization and reinforcement learning (Q‑Learning) to model strategic operator behavior. It analyzes two scenarios—PNIEC2030 and a 10‑operator portfolio derived from 2024 public offers—showing that SPaC reduces inframarginal rents and price volatility while maintaining participation incentives and displaying robustness to oligopolistic power. Under RL‑driven bidding, PaB becomes the most rent‑extractive and costly regime, PaC remains costly due to uniform pricing, and SPaC achieves a favorable middle ground with lower price spikes and improved allocation efficiency. The framework serves as a regulator‑oriented ex‑ante tool to assess market design reforms and shows SPaC’s potential to balance efficiency, investment signals, and equity in both Italian and broader European contexts.

Abstract

The Pay-as-Clear (PaC) mechanism currently used in the European electricity market can generate significant submarginal profits for renewable sources when the clearing price is determined by the marginal offers of gas-fired generation units and the cost of natural gas exceeds certain levels. This exposes consumers to high price volatility related to the cost of natural gas. This report analyzes the recently proposed Segmented Pay-as-Clear (SPaC) mechanism as a market alternative, evaluating its system cost-effectiveness through simulations based on Reinforcement Learning (Q-Learning) to model the strategic behavior of operators. Three market models are compared, the two classic Pay-as-Clear (PaC) and Pay-as-Bid (PaB) along with SPaC, under two scenarios: a simplified one based on the 2030 NECP objectives and one built on the portfolios of ten operators obtained from the GME's 2024 public offers. The results show that the SPaC market clearing mechanism reduces intramarginal profits and price volatility compared to PaC, while maintaining fair participation incentives for all operators, and is more robust than PaB to the exercise of market power in oligopolistic contexts. The developed framework can serve as a support tool for regulators and policymakers in the evaluation of proposals for market design reforms.
Paper Structure (47 sections, 21 equations, 17 figures, 24 tables, 1 algorithm)

This paper contains 47 sections, 21 equations, 17 figures, 24 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Objectives and contributions of the report
  3. Methodology and analysis scenarios
  4. Document structure
  5. Market Models
  6. Pay-as-Clear (PaC)
  7. Pay-as-Bid (PaB)
  8. Segmented Pay-as-Clear (SPaC)
  9. Bidding Strategy and Q-Learning
  10. Basic Elements of Reinforcement Learning
  11. Q-Learning
  12. Implementation for Electricity Markets
  13. State Space Definition
  14. Action Space Definition
  15. Reward Function
  16. ...and 32 more sections

Figures (17)

  • Figure 1: Illustrative example of the SPaC model frangioniBilevelProgrammingApproach2024.
  • Figure 2: Bilevel structure of the SPaC model: the leader minimizes total cost by choosing $d_r$, the followers determine the marginal prices $\pi_r, \pi_g$frangioniBilevelProgrammingApproach2024.
  • Figure 3: Supply curves and results of simulations with offers at marginal cost: (a) PaB, (b) PaC, (c) SPaC.
  • Figure 4: Results of the three markets with offers at random markups (Table \ref{['tab:markup-casuali']}): left PaB and PaC, right SPaC.
  • Figure 5: Monitoring of training for the three markets and demand 1.0 GW. Left: operator profits, right: PUN.
  • ...and 12 more figures