On the Design of Stochastic Electricity Auctions

This paper proposes enhancing day-ahead electricity auctions by conditioning contracts on specific states of the world to address renewable energy uncertainty, utilizing microeconomic equilibrium theory to define optimal state partitions and demonstrating their application through a North Sea offshore wind case study.

The Gist

Imagine you are running a massive, high-stakes game of "What's the Weather Tomorrow?" but instead of just guessing, you have to buy and sell electricity based on that guess.

This paper, written by Thomas Hübner, argues that our current way of trading electricity is like playing that game with a broken rulebook. It proposes a new, smarter way to play that could save money, reduce waste, and help green energy thrive.

Here is the breakdown in simple terms, using some everyday analogies.

1. The Problem: The "Blind" Auction

Right now, electricity markets work like a day-ahead auction. At noon today, everyone bets on how much electricity they will need or produce tomorrow.

• The Thermal Power Plant (The Reliable Baker): Imagine a bakery that needs to decide today whether to turn on its massive ovens for tomorrow. It needs to know if it will sell its bread to make a profit. So, it sells its bread (electricity) in the morning auction.
• The Wind Farm (The Rain-Dependent Farmer): Imagine a farmer who only grows crops when it rains. They don't know if it will rain tomorrow.
  • If they sell their "crop" (electricity) in the morning auction, they are guessing.
  • The Risk: If they guess "It will rain" and sell 100 units, but it turns out sunny, they have nothing to deliver. They have to buy electricity later at a crazy high price to cover their promise.
  • The Result: To avoid this risk, wind farms often sell less than they could or get paid less. It's inefficient. The system wastes potential green energy because the market is too rigid.

The Current Flaw: Today's auction only asks: "How much power do you want at 8 AM in London?" It does not ask: "How much power do you want at 8 AM in London IF it is windy?"

2. The Solution: "Weather-Contingent" Contracts

The author suggests we change the rules of the auction. Instead of just trading "Electricity," we should trade "Electricity IF the wind blows."

Think of it like buying umbrellas:

• Current System: You buy 100 umbrellas today, regardless of the weather. If it rains, you use them. If it's sunny, you have 100 useless umbrellas taking up space.
• The New System: You buy a contract that says, "I will give you 100 umbrellas only if it rains tomorrow."
  • If it rains, the deal happens.
  • If it's sunny, the deal is void, and nobody loses money or space.

In this new auction, a wind farm can say: "I promise to sell 100 MWh of power IF the wind speed is between 10 and 15 m/s."

• If the wind is that strong, they deliver.
• If the wind is calm, the contract simply doesn't trigger.

This removes the fear of guessing wrong. The wind farm can sell all its expected power without the risk of having to buy it back later.

3. The Hard Part: Defining "The Weather"

Here is the tricky part the paper solves. If we want to trade based on the weather, how do we define "The Weather"?

There are infinite ways the weather could be:

• Is it "Windy"?
• Is it "Windy and Cloudy"?
• Is it "Windy at 10:00 AM but calm at 2:00 PM"?
• Is it "Wind speed is exactly 12.43 m/s"?

If we try to make a contract for every single possible weather scenario, the auction becomes a nightmare of complexity. It would be like trying to sell a ticket for "Every possible combination of rain, sun, clouds, and wind speed." No one could understand it, and the computer couldn't calculate it.

The Paper's Magic Trick: The "Voronoi" Map
The author uses a mathematical concept called a Voronoi Partition (think of it like a territory map).

Imagine you have a map of wind speeds. You want to divide this map into a few distinct "zones" (states) so people can trade based on those zones.

• Zone A: "Generally Low Wind"
• Zone B: "Generally High Wind"
• Zone C: "High Wind in the North, Low in the South"

The paper proposes a smart algorithm to draw these zones. It asks: "How can we draw 3 or 4 zones on the map so that every point inside a zone is as close as possible to the 'center' of that zone?"

It's like placing fire stations in a city. You want to place 3 fire stations so that every house in the city is as close as possible to a station. The "zones" are the areas each station covers.

• The algorithm finds the perfect spots for these "centers" (representative wind speeds).
• It draws the lines around them so that if the wind speed falls in a specific area, everyone knows exactly which "State" (contract) applies.

4. Why This Matters

By using this method, the market becomes smarter:

1. Renewables Win: Wind and solar farms can sell more power because they aren't afraid of the "what if" scenarios.
2. System Efficiency: The grid operator can make better decisions about which power plants to turn on, knowing exactly how much wind power is likely to arrive under different weather conditions.
3. Fairness: The prices reflect the true value of electricity in different weather scenarios. If electricity is scarce because it's calm, the price for "calm-day power" goes up. If it's windy, the price for "windy-day power" goes down.

The Bottom Line

The paper argues that we are currently trading electricity like it's a static object (like a chair). But electricity from wind and solar is dynamic; it depends on the world around it.

We need to upgrade our auction from a "One-Size-Fits-All" model to a "Choose Your Own Adventure" model. We don't need to predict the future perfectly; we just need to group the possibilities into a few clear, manageable "states of the world" (like "Stormy," "Calm," "Windy") using a smart mathematical map.

This allows the market to handle uncertainty gracefully, turning the chaos of the weather into a structured, efficient game where everyone wins.

Technical Summary

1. Problem Statement

The paper addresses a fundamental inefficiency in current day-ahead electricity markets caused by the integration of renewable energy (wind and solar).

• The Core Issue: Current day-ahead auctions operate under a deterministic framework where contracts are contingent only on time ($t$) and location ($n$). However, renewable generators face significant uncertainty regarding their actual output (e.g., wind speed) at the time of delivery.
• The Consequence: Because uncertainty cannot be directly communicated in current markets, renewable producers face a dilemma:
  1. Sell too little: Risk curtailment if actual generation exceeds the contract.
  2. Sell too much: Risk having to repurchase electricity at high prices in balancing markets if generation falls short.
• Theoretical Gap: While microeconomic theory (Arrow-Debreu) establishes that efficient trade under uncertainty requires contracts contingent on the state of the world ($s$), current electricity markets lack a mechanism to define and trade these state-contingent contracts. Existing literature on "stochastic market clearing" often focuses on optimization models that suffer from computational intractability or lack desirable market properties (like strategy-proofness).

2. Methodology

The author proposes a market design based on the Theory of Equilibrium Under Uncertainty from microeconomics, extending it to the specific constraints of electricity markets.

A. Theoretical Framework

• State-Contingent Contracts: The paper defines electricity contracts that are contingent on a specific realization of a random variable $\xi$ (the "state of the world").
  • Definition: A state $s$ is a subset $\Omega_s$ of the sample space $\Omega$ of a random variable $\xi$ (e.g., wind speed).
  • Mechanism: Agents trade contracts at $t=0$ (day-ahead). Payments are made upfront, but physical delivery only occurs if state $s$ is realized.
• Equilibrium Analysis: The paper demonstrates that if agents trade these contracts in a Walrasian auction, the resulting allocation maximizes social welfare and induces optimal day-ahead decisions (e.g., unit commitment) that account for uncertainty. This extends the First and Second Welfare Theorems to stochastic environments.

B. Defining the "State of the World" (The Core Innovation)

The paper identifies the central practical challenge: How to select a finite set of states from an infinite sample space?

• Criteria for Selection: The author proposes three criteria for selecting states $\Omega_1, \dots, \Omega_S$:
  1. Cover: The states must cover the entire sample space ($\bigcup \Omega_s = \Omega$).
  2. Pairwise Disjointness: States must not overlap ($\Omega_s \cap \Omega_j = \emptyset$).
  3. Minimal Size: The partition should minimize a specific "size" metric.
• The Size Metric: The size of a state is defined as the variance-weighted probability mass:
  $$\text{Size}(\Omega_s) = \int_{\xi \in \Omega_s} \|\xi - \omega_s\|_2^2 \, dP$$
  Where $\omega_s$ is the barycenter (centroid) of the state, and $P$ is the probability measure. This metric penalizes states that are large, unlikely, or spatially dispersed.
• Optimal Partitioning: The problem of finding the optimal states is shown to be equivalent to finding a Centroidal Voronoi Partition.
  • The optimal states correspond to Voronoi cells generated by points $\omega_1, \dots, \omega_S$ that solve a Location Problem (minimizing the average squared distance to the nearest point).
  • This transforms the state definition problem into a Mixed-Integer Quadratic Program (MIQP), which is computationally tractable for low-dimensional uncertainty (e.g., 1D or 2D wind speed).

C. Auction Design

The paper outlines how to implement this in a centralized auction:

• Bidding: Agents submit bid functions for state-contingent contracts.
• Properties: The design preserves key properties of current deterministic auctions:
  • Strategy-proofness in the large: Agents are incentivized to bid truthfully in large markets.
  • Individual Rationality: Participation does not guarantee losses.
  • Budget Balance: The auctioneer breaks even.
• Handling Non-Convexity: Acknowledging that electricity markets have non-convexities (e.g., start-up costs), the paper notes that approximate equilibria can be computed using established methods, extending current practices to the stochastic case.

3. Key Contributions

1. Conceptual Shift: Moves the debate from "stochastic optimization" (which often leads to complex, non-transparent clearing models) to "state-contingent markets" (which preserve the transparency and desirable properties of Walrasian equilibrium).
2. State Definition Algorithm: Provides a rigorous, mathematically grounded method for defining states using Centroidal Voronoi Tessellations. This solves the "infinite states" problem by selecting a finite, interpretable set of states that minimize information loss.
3. Interpretability: The resulting states are intuitive. For example, a state is defined as "Wind speeds at Location A and B are closer to point $P$ than to any other defining point." This keeps transaction costs low by ensuring agents understand what they are trading.
4. Extension of Welfare Theorems: Proves that the desirable properties of deterministic day-ahead auctions (efficiency, strategy-proofness) naturally extend to this stochastic design.

4. Results (Case Study)

The paper validates the methodology using a case study of offshore wind farms in the European North Sea.

• Setup:
  • Random Variable ($\xi$): Wind speeds at 11 PM at two locations (Southern UK/France and Dutch/German/Danish coast).
  • Data: Probabilistic forecast for Feb 18, 2026, with 39 scenarios.
• Computation: The MIQP was solved for $S=2, 3, 4$ states using Gurobi.
• Findings:
  • $S=2$: Separates "Generally Low Wind" vs. "Generally High Wind."
  • $S=3, 4$: Captures spatial nuances, such as "High wind at Location 1 but low at Location 2."
  • Interpretation: Increasing the number of states allows the market to condition trades more precisely on the specific wind pattern, reducing the need for costly balancing actions later.
  • Scalability: The approach remains tractable for low-dimensional variables (1D or 2D). The paper notes that the "curse of dimensionality" limits the number of states if too many random variables are included, suggesting a focus on the most critical uncertainty drivers.

5. Significance and Implications

• Efficiency: By allowing renewable generators to hedge against specific wind/solar states, the market reduces curtailment and the need for expensive redispatch, leading to higher social welfare.
• Practical Implementation: The proposed design is a "simple extension" of current auctions. It does not require a complete overhaul of the trading platform but rather an expansion of the contract space to include state indices.
• Policy Relevance: As renewable penetration increases, the current deterministic market design becomes increasingly inefficient. This paper provides a theoretically sound and computationally feasible roadmap for transitioning to a stochastic market design that maintains market stability and transparency.
• Future Work: The author suggests that while the current method uses wind speed as the primary driver, future iterations could incorporate load, network constraints, and other variables, provided the dimensionality remains manageable for the auctioneer and participants.

In summary, Hübner bridges the gap between abstract economic theory (Arrow-Debreu equilibrium) and practical power system operations by providing a concrete algorithm (Centroidal Voronoi Partitioning) to define the "states of the world" necessary for efficient, stochastic electricity trading.