⚡ electrical engineering

Modeling Coincident Peak Pricing in Electricity Markets: Challenges and Peak Shaving Effectiveness

This paper presents a behavioral game-theoretic framework analyzing Coincident Peak pricing in electricity markets, demonstrating that while aggregate flexibility is crucial, the effectiveness of peak shaving depends significantly on the learning dynamics, action resolution, and the design of information signals used to coordinate consumer responses.

Original authors: Qian Zhang, Sadie Zhao, Lucy Diao, Conleigh Byers, Yiling Chen, Derya Cansever, Le Xie

Published 2026-05-19

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Qian Zhang, Sadie Zhao, Lucy Diao, Conleigh Byers, Yiling Chen, Derya Cansever, Le Xie

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the electricity grid as a massive, crowded highway. Usually, traffic flows smoothly, but sometimes, everyone tries to drive at the exact same time, causing a massive jam. This "jam" is called the Coincident Peak (CP).

In the U.S., the people who run the power grid (like the ISOs) need to build enough power plants and wires to handle these worst-case jams. To pay for this expensive infrastructure, they use a pricing rule called Coincident Peak Pricing.

Here's how it works in simple terms: If you are using a lot of electricity at the exact moment the entire system is at its busiest, you get billed a huge fee. It's like being charged a "congestion fee" for driving on the highway at the exact second the traffic is worst.

This paper asks a big question: Does this fee actually stop people from driving during rush hour, or does it just make the traffic worse?

The authors used a "game theory" approach—basically, a mathematical way to predict how people play against each other when they are trying to save money. Here is what they found, using some everyday analogies:

1. The Two Ways People React (The "Myopic" vs. The "Smart" Player)

The paper tested two different ways people might react to the threat of this fee:

The "Myopic" Player (Best-Response Dynamics): Imagine a group of drivers who all see a traffic alert saying, "Rush hour is at 5:00 PM!" They all immediately decide to leave at 4:55 PM to beat the rush. But because everyone does this, they all end up creating a new, even worse traffic jam at 4:55 PM.
- The Result: If everyone reacts too quickly and blindly to the same signal, they often make the peak worse instead of better. The paper found this approach is risky and can actually increase the peak load if people don't have enough flexibility.
The "Smart" Player (Fictitious-Play Dynamics): Imagine these drivers are a bit more patient. They look at what everyone else did yesterday and the day before. They realize, "Oh, everyone tried to leave at 4:55 PM yesterday and got stuck. Maybe I should leave at 4:45 PM or 5:15 PM instead." They learn from history and spread themselves out.
- The Result: This approach worked much better. By learning from past patterns rather than just reacting to the immediate moment, the "smart" players successfully smoothed out the traffic, lowering the peak.

2. The Importance of "Steering Wheel" Precision

The paper also looked at how much control people have over their electricity use.

Coarse Control: Imagine a light switch that is either ON or OFF. If you have to turn your whole factory off for an hour to save money, you might all choose the exact same hour. This creates a new jam.
Fine Control: Imagine a dimmer switch where you can turn the lights down by 10%, 20%, or 30%. With this finer control, people can make tiny adjustments. One person dims a little at 4:50, another at 4:52, another at 4:55.
- The Result: Having a "dimmer switch" (finer control) is crucial. It allows people to spread out their load so they don't all bunch up at the same time.

3. The "Weather Forecaster" Problem

In the real world, people don't guess when the peak will happen; they listen to Information Providers (companies that predict the peak).

The Naive Forecaster: This forecaster looks at the weather and says, "It's going to be hot at 5:00 PM, so the peak will be at 5:00 PM."
- The Problem: If everyone listens to this one forecaster, they all shift their load to 4:30 PM, creating a new peak there.
The "Response-Aware" Forecaster: This forecaster is smarter. They think, "If everyone listens to the Naive Forecaster and shifts to 4:30 PM, the peak will actually move to 4:30 PM. So, I will predict 4:30 PM as the peak."
- The Result: The paper found that having a mix of different forecasters (some naive, some smart) is actually better than everyone listening to just one. It prevents everyone from "herding" (following the crowd) into a single new jam.

4. Does More People Help?

You might think that if more people try to save money, the peak will drop more.

The Finding: Surprisingly, if the total amount of flexible power available stays the same, it doesn't matter much if you have 5 big factories or 100 small ones. The key isn't the number of players, but how much total "flexibility" they have and how well they coordinate.

The Bottom Line

The paper concludes that Coincident Peak pricing is a powerful tool, but it's tricky.

If people react blindly and quickly, they might accidentally make the problem worse.
If they learn from the past, have fine-grained control (dimmer switches), and listen to diverse forecasts, they can successfully shave off the peak.

For the Power Grid Operators: They should try to give "smoothed" signals (not just a single scary alert) and encourage users to have fine control over their power.
For the Consumers: The more flexible you are and the more granular your control (being able to adjust small amounts), the more money you save and the better you help the grid.

In short: Don't all rush for the exit at the same time just because someone shouted "Fire!"; look at where everyone else is going, and maybe take a slightly different path.

Technical Summary: Modeling Coincident Peak Pricing in Electricity Markets

1. Problem Statement

Coincident Peak (CP) pricing is a prevalent mechanism in U.S. electricity markets (e.g., ERCOT, ISO-NE) used to allocate transmission and capacity costs based on a customer's demand during the system's highest demand intervals. While transparent and easy to implement, CP pricing introduces complex strategic behaviors among consumers.

The core challenge addressed in this paper is the difficulty of modeling user behavior under CP pricing. Unlike short-term energy markets, CP involves:

Endogenous Peak Formation: The peak interval is not a fixed external parameter but an outcome of aggregate consumer behavior. As more users engage in strategic load shifting, the peak becomes a stochastic, endogenous variable.
Nonlinear Cost Allocation: A consumer's charge depends on their individual demand at the peak relative to the total system load at that same moment. This creates a nonlinear coupling between individual actions and aggregate outcomes.
Information Asymmetry and Herding: Consumers rely on forecasts (from ISOs or third parties) to predict peak times. Widespread reliance on identical forecasts can induce "herding," where many users shift load to the same intervals, potentially creating new, unintended peaks.

The paper seeks to answer whether CP pricing reliably incentivizes peak shaving and how factors like flexibility, action resolution, and information diversity influence these outcomes.

2. Methodology

A. Modeling Framework

The authors develop a behavioral game-theoretic framework coupling a nonlinear cost-allocation model with decision processes in both day-ahead and real-time markets.

Cost Allocation Model: The transmission charge $c_i$ for consumer $i$ is modeled as a fraction of the total system cost $C$ proportional to their demand at the coincident peak time $t^*$ . The model acknowledges the nonlinearity where individual demand affects the denominator (total system load). A linearized approximation is discussed but noted to overlook key strategic interactions in high-flexibility systems.
Consumer Constraints: Responsive consumers are subject to:
- Capacity Bounds: $X_i \le x_i(t) \le \bar{X}_i$ .
- Energy Balance: Total energy over a window $W$ is fixed ( $\sum x_i(t) = E_i$ ).
- Action Representation: Demand is modeled as a baseline profile plus a shifting action $a_i(t)$ , where $\sum a_i(t) = 0$ .

B. Decision Processes and Dynamics

The paper distinguishes between two market stages and analyzes two canonical learning dynamics:

Day-Ahead Market: Modeled as a one-shot game where users submit a single bid. The paper notes that computing a Nash equilibrium here is computationally intractable (PPAD-hard) and behaviorally unrealistic due to information asymmetry.
Real-Time Market: Modeled as a sequential learning process where users update strategies based on observed history. Two dynamics are examined:
- Best-Response Dynamics (BRD): Users myopically choose the best action against the current observed actions of others.
- Fictitious Play Dynamics (FPD): Users form beliefs about others' strategies based on the empirical frequency of past actions and best-respond to this belief distribution.

C. Evaluation

The framework is evaluated using ERCOT peak-day data (2023). The study simulates Large Flexible Loads (LFLs) under various scenarios:

Action Spaces: Continuous (real-valued dispatch) vs. Finite (discrete, menu-based options).
Parameters: Varying levels of flexibility (capacity margins), number of participants ( $N$ ), and information provider strategies.

3. Key Contributions

Formalization of Nonlinear CP Models: The paper formalizes a nonlinear CP cost-allocation model and a tractable behavioral load model for responsive demand under both continuous and finite action sets, explicitly capturing the mutual dependence between individual and aggregate behavior.
Linking Time Horizons: It connects day-ahead one-shot decisions with real-time sequential learning, analyzing the theoretical and computational challenges of both Best-Response and Fictitious Play dynamics.
Role of Information Providers: The study incorporates the role of third-party information providers, demonstrating how their forecasting signals (Naive vs. Response-Aware) affect system-level peak shaving and the potential for herding.
Quantitative Insights: Using ERCOT data, the paper quantifies how flexibility, action resolution, participation scale, and information diversity shape peak outcomes, moving beyond theoretical counterexamples to practical case studies.

4. Results

A. Dynamics: BRD vs. FPD

FPD Superiority: Fictitious Play Dynamics (FPD) reliably reduces system peaks and converges rapidly. By smoothing reactions based on historical averages, FPD avoids the synchronization issues seen in BRD.
BRD Volatility: Best-Response Dynamics (BRD) is more variable. Under tight capacity constraints, BRD can lead to synchronized myopic shifts, causing users to cluster in the same interval and increase the system peak (e.g., a 1.08% increase in worst-case scenarios).

B. Flexibility and Action Resolution

Flexibility: Greater flexibility (higher capacity margins) leads to more reliable peak reductions. When flexibility is scarce, even optimal strategies may fail or backfire.
Action Resolution: Finer action resolution (smaller step sizes in load shifting) significantly improves peak shaving. Coarse, "on/off" style controls tend to concentrate shifts into the same few intervals, limiting coordination gains.
Number of Participants: When aggregate flexibility is held constant, the number of participants has a largely neutral effect on peak shaving. Scaling from a few large loads to many small loads does not inherently improve or degrade performance if the response rule is identical.

C. Information Providers

Herding Risk: "Naive" information providers (ranking peaks based solely on baseline demand) can induce herding, creating secondary peaks near the predicted times.
Response-Aware Signals: Providers that anticipate user behavior (Response-Aware) and target intervals that would become peaks after initial responses improve system outcomes.
Heterogeneity: A mixed population of users following different information signals (some Naive, some Response-Aware) often yields better peak shaving than a homogeneous population, as the diversity acts as a stabilizing buffer against synchronization.

5. Significance and Practical Implications

The paper concludes that the potential of CP pricing to reduce peaks is real but limited by behavioral and structural factors. The findings offer specific guidance for stakeholders:

For System Operators (ISOs):
- Broadcasting smoothed CP-likelihood signals (e.g., rolling averages) can induce FPD-like behavior, reducing synchronized herding.
- Setting minimum controllable-capacity thresholds for enrollment ensures participants have sufficient flexibility to shift load without causing secondary peaks.
- Encouraging diverse forecasting signals can mitigate the risks of information-induced herding.
For Consumers (LFLs):
- Maintaining adequate flexible capacity is crucial; insufficient capacity can lead to counterproductive outcomes under myopic strategies.
- Adopting finer control resolution (granular load shifting) improves both cost savings and peak-shaving performance compared to coarse, binary controls.

The study emphasizes that while CP pricing provides a coordinating signal, its effectiveness relies heavily on the granularity of control available to consumers and the nature of the information they receive. The framework provides a foundation for understanding how behavioral dynamics interact with market mechanisms in the transition toward more flexible, demand-responsive power grids.