Competition Law and Governance of Simulation-Driven Competition

1. Introduction

Simulation-driven competition refers to markets in which undertakings use computer simulations, artificial intelligence, digital twins, predictive models, reinforcement learning, scenario analysis, or algorithmic forecasting to determine competitive strategies. Firms may simulate competitors' likely reactions before changing prices, output, capacity, product design, investment, advertising, or market entry.

Simulation itself is generally a legitimate business tool. Competition-law concerns arise when simulations become a mechanism for:

• coordinating prices or output;
• exchanging competitively sensitive information;
• predicting and accommodating rivals rather than competing independently;
• implementing cartel arrangements automatically;
• excluding competitors through predictive models;
• coordinating capacity or market allocation;
• facilitating algorithmic tacit collusion;
• enabling a dominant platform to use data-driven simulations to disadvantage rivals.

The important legal principle is that the technological sophistication of the mechanism does not determine legality. Competition law generally examines the underlying conduct, agreement, information exchange, market power, effects, and purpose.

The CMA has specifically identified three risks associated with algorithmic systems: facilitation of explicit coordination, common third-party algorithms creating hub-and-spoke structures, and autonomous tacit coordination.

2. Meaning of Simulation-Driven Competition

A simulation-driven market can operate through a cycle such as:

Market data → Simulation model → Predicted rival response → Strategic decision → Market outcome → New data → Updated simulation

For example, two competing airlines may independently simulate:

“If we increase fares by 5%, how will competitors respond?”

That is normally legitimate.

The competition-law risk increases if the firms:

exchange confidential pricing data → use a common simulation model → agree on predicted/target prices → implement the recommendations.

The distinction is therefore between independent predictive competition and coordinated predictive behaviour.

3. Major Competition-Law Issues

A. Algorithmic or Simulated Price Coordination

The most obvious concern is when simulations enable competitors to converge on prices.

Suppose five competing firms use the same simulation platform. The platform receives:

• current prices;
• planned price changes;
• capacity;
• inventory;
• demand forecasts;
• margins;
• promotional plans.

If the system generates recommendations that effectively cause firms to avoid undercutting one another, the simulation may become a mechanism for coordination.

The legal issue is not simply whether the computer made the decision. Authorities may examine whether humans supplied the information, designed the parameters, accepted the recommendations, or knowingly participated in the resulting coordination.

The U.S. Department of Justice has expressly argued in the RealPage litigation that automation does not immunize an otherwise unlawful pricing arrangement from antitrust scrutiny.

4. Simulation as a Hub-and-Spoke Mechanism

A simulation provider can become a hub connecting competing firms.

Traditional model

Competitor A ↔ Hub ↔ Competitor B

The hub receives sensitive information from both competitors and generates strategic recommendations.

The danger is particularly significant where the algorithm uses:

• non-public competitor prices;
• future pricing plans;
• capacity information;
• inventory;
• margins;
• demand forecasts;
• customer-specific information.

This can transform an apparently vertical software relationship into a mechanism facilitating horizontal coordination.

5. Autonomous Tacit Coordination

The most difficult problem is autonomous algorithmic coordination.

Consider two competing AI systems:

1. Algorithm A observes market prices.
2. Algorithm B observes market prices.
3. Both independently learn that aggressive price reductions reduce profits.
4. Both gradually increase prices.
5. Each algorithm responds to the other's behaviour.
6. The market reaches a stable high-price equilibrium.

There may be no email, meeting, telephone conversation, or express cartel agreement.

This creates a difficult question:

Can competition law intervene where algorithms independently learn a coordinated outcome?

Traditional cartel law is principally designed around agreements and concerted practices. Autonomous machine learning therefore creates an important boundary between unlawful coordination and independent intelligent adaptation.

The CMA has specifically recognised autonomous tacit collusion as a potential competition concern.

6. Simulation and Information Exchange

Simulation models are highly dependent upon data.

Information exchange becomes particularly problematic when competitors exchange information concerning:

• future prices;
• future output;
• investment plans;
• capacity;
• costs;
• customer allocation;
• discounts;
• strategic intentions.

Historical and publicly available data generally present less concern than current, granular, confidential information.

The risk is amplified when the same information is systematically fed into a common simulation environment used by competitors.

7. Simulation and Market Allocation

Simulation can also be used to divide markets.

For example, competitors could use predictive models to determine:

• which geographic areas each should target;
• which customers each should serve;
• which contracts each should bid for;
• which territories should be left uncontested.

A simulation that merely predicts market demand is ordinarily different from a simulation used to implement an agreement to allocate customers or territories.

8. Simulation and Bid Coordination

Procurement markets are particularly sensitive.

Competitors could theoretically simulate:

• expected rival bids;
• likely winning prices;
• procurement probabilities;
• tender timing;
• capacity constraints.

If competitors use such information to coordinate bids, the simulation can become an instrument of bid rigging.

Competition authorities therefore need to distinguish between:

legitimate bid forecasting

and

coordinated bidding facilitated by a simulation system.

9. Simulation and Dominance

Simulation-driven competition is not confined to cartels.

A dominant digital platform might use massive datasets and predictive models to determine:

• which competitors are likely to emerge;
• which products threaten its ecosystem;
• which rival should receive less visibility;
• which prices will make entry unprofitable;
• which APIs or interoperability arrangements should be restricted.

Possible theories of harm include:

• exclusionary conduct;
• discriminatory access;
• self-preferencing;
• tying;
• predatory strategies;
• refusal to deal;
• leveraging;
• foreclosure of emerging competitors.

Thus, simulation governance is also a unilateral-conduct issue under abuse-of-dominance rules.

10. Six Major Case Laws

1. Eturas UAB and Others v Lietuvos Respublikos konkurencijos taryba, Case C-74/14

Facts

Travel agencies used a common electronic booking system operated by Eturas. A message was sent through the system indicating that discounts available to customers would be restricted. The system was subsequently technically modified to impose the restriction.

Decision

The Court of Justice of the European Union held that awareness of the system administrator's communication, together with continued participation and market conduct, could support a presumption of participation in a concerted practice, subject to rebuttal. Mere receipt of the message was not automatically sufficient without considering the evidentiary circumstances.

Relevance to simulation-driven competition

Eturas is important because it demonstrates that electronic architecture can become evidence of concerted conduct.

A future simulation system could similarly create evidence through:

• system messages;
• common parameters;
• automated restrictions;
• algorithmic modifications;
• system logs;
• acceptance of common recommendations.

It establishes that competition law can look beyond traditional face-to-face communications.

2. Trod Ltd / GB eye Ltd – Online Sales of Posters and Frames

Facts

Two competing Amazon Marketplace sellers agreed not to undercut one another.

They implemented their arrangement using automated repricing software.

Decision

The UK Competition and Markets Authority found an infringement of competition law. Trod was fined £163,371, while GB eye received immunity after reporting the cartel and cooperating with the CMA.

Relevance

This is one of the clearest demonstrations that:

Automation does not eliminate cartel liability.

The software did not independently create the cartel. Humans created the agreement and configured software to implement it.

For simulation-driven markets, the case establishes an important distinction:

Simulation as a tool for implementing an existing agreement → traditional cartel problem.

3. United States v. Topkins

Facts

David Topkins and other online sellers allegedly agreed to fix prices for posters sold through Amazon Marketplace.

The participants agreed to use algorithmic pricing software to coordinate prices.

Decision

Topkins pleaded guilty to the antitrust conspiracy and agreed to pay a criminal fine.

Relevance

Topkins demonstrates the application of traditional price-fixing principles to algorithmic environments. The algorithm was merely the technological mechanism through which the human agreement was implemented.

Principle

A useful analytical formula is:

Human agreement + algorithmic implementation = potentially ordinary price fixing

The fact that the final price is generated by software does not change the underlying nature of the agreement.

4. In re RealPage, Inc. Rental Software Antitrust Litigation

Facts

The litigation concerns allegations that competing landlords used RealPage's pricing software and supplied competitively sensitive information that was incorporated into algorithmic pricing recommendations.

The DOJ alleged that landlords provided information concerning rental rates and lease terms and that the software was used to generate pricing recommendations.

The U.S. government argued that competitors' use of common algorithms and combination of sensitive information could constitute unlawful price fixing.

Relevance

RealPage is particularly important to simulation-driven competition because it demonstrates the potential significance of:

• common predictive systems;
• shared confidential data;
• algorithmic recommendations;
• automated pricing;
• reduction of independent decision-making.

The litigation also illustrates that not every algorithmic pricing system is automatically unlawful; the factual question is whether the system facilitates an unlawful combination or materially removes independent competitive decision-making.

5. Gibson v. Cendyn Group, LLC, 148 F.4th 1069 (9th Cir. 2025)

Facts

Competing Las Vegas hotels used revenue-management software supplied by Cendyn. The software generated pricing recommendations using pricing and occupancy data.

The plaintiffs alleged that common use of the software resulted in higher hotel prices.

Decision

The Ninth Circuit affirmed dismissal of the Section 1 claims. It held that the allegations did not sufficiently establish an agreement among competing hotels or an anticompetitive restraint arising from their independent licensing of the software. The court noted the significance of allegations that the software did not share one hotel's confidential information with another.

Relevance

This case is extremely important because it illustrates the other side of the algorithmic-competition problem.

Using the same predictive or pricing technology does not automatically constitute a cartel.

The analysis may turn on whether:

• competitors actually agreed with one another;
• the provider imposed anticompetitive restraints;
• confidential competitor data were shared;
• firms delegated pricing decisions;
• the software merely supplied independent commercial information.

Thus:

Common software ≠ automatically unlawful coordination.

6. Samir Agarwal v. Competition Commission of India, Supreme Court of India, 2020

Facts

The case concerned allegations involving algorithmically determined fares on app-based cab platforms.

The Competition Appellate Tribunal had considered whether algorithmic fare determination could constitute a traditional hub-and-spoke arrangement.

Decision

The Tribunal's reasoning distinguished algorithmic pricing based on large datasets from a conventional hub-and-spoke arrangement involving exchange of competitively sensitive information through a third-party hub. The Supreme Court subsequently dealt with the appeal concerning the CCI's treatment of the allegations.

The discussion is particularly relevant because the algorithm could consider factors such as:

• time;
• traffic;
• demand and supply;
• events;
• day of week;
• personalised rider information. 

Relevance

For India, this case demonstrates that algorithmic price differentiation does not by itself establish a hub-and-spoke cartel.

The actual competitive mechanism must be examined.

This is particularly relevant for simulation-driven competition because an algorithm can produce similar or convergent outcomes without necessarily involving communication between competitors.

11. Comparative Principles From the Cases

12. Governance Framework for Simulation-Driven Competition

A sound competition-compliance system should examine the entire simulation lifecycle.

Stage 1 — Data collection

Ask:

• What data enters the model?
• Is the data public or confidential?
• Does it belong to competitors?
• Is it current or historical?
• Does it reveal future strategic intentions?

Stage 2 — Model design

Examine:

• objective functions;
• optimisation targets;
• constraints;
• competitor variables;
• pricing rules;
• market-allocation parameters.

Stage 3 — Model training

Maintain records showing:

• training datasets;
• data provenance;
• model versions;
• changes in parameters;
• personnel responsible for modifications.

Stage 4 — Simulation

Monitor whether the system:

• predicts rivals;
• recommends responses;
• systematically discourages competitive undercutting;
• converges toward coordinated outcomes;
• uses confidential competitor information.

Stage 5 — Human decision-making

The firm should preserve genuine independent decision-making.

Particular concern arises where employees simply accept automated recommendations without meaningful review.

Stage 6 — Deployment

Controls should exist over:

• automatic implementation;
• pricing;
• bidding;
• inventory;
• capacity;
• market allocation.

Stage 7 — Audit

Regular audits should test:

• unexpected price convergence;
• information flows;
• common parameters;
• model outputs;
• competitor data;
• communications between competitors.

13. Simulation Logs as Competition Evidence

An important emerging issue is the evidentiary value of simulation records.

Competition authorities may examine:

• source code;
• model architecture;
• training datasets;
• API records;
• system logs;
• version histories;
• prompts;
• configuration files;
• pricing recommendations;
• override records;
• communications concerning model outputs.

Accordingly, algorithmic governance becomes an evidentiary issue as well as a substantive competition-law issue.

A company should be able to demonstrate:

“The model made this recommendation from independently obtained data, and the company independently decided whether to accept it.”

That is substantially different from:

“Competitors jointly supplied the model with sensitive information and agreed to follow its predicted equilibrium.”

14. Simulation-Driven Competition and Market Definition

Simulation can also affect market-definition analysis.

Traditional tools such as the SSNIP test ask how consumers respond to a hypothetical price increase.

In highly digital markets, authorities may supplement traditional analysis with:

• demand simulations;
• diversion ratios;
• switching models;
• elasticity estimation;
• merger simulations;
• bidding models;
• consumer-choice models;
• network-effect analysis.

Simulation can therefore serve two opposite functions:

Pro-competitive function:
better understanding of consumer demand and competitive effects.

Anti-competitive function:
facilitating coordination or exclusion.

The same technology can therefore be lawful in one context and problematic in another.

15. Simulation in Merger Control

Simulation models are increasingly relevant to merger analysis.

A merger simulation may estimate:

• post-merger prices;
• diversion;
• unilateral effects;
• capacity reductions;
• product substitution;
• efficiencies.

Competition authorities can use such models to test whether a proposed concentration is likely to produce competitive effects.

At the same time, merging parties may use simulation to demonstrate:

• efficiencies;
• demand substitution;
• entry;
• customer switching;
• innovation effects.

Thus, simulation can become both an enforcement tool and a defence tool in merger control.

16. Competition Risks From Generative AI Simulations

Generative AI creates additional concerns because firms can ask an AI system to simulate:

• competitor reactions;
• market-entry scenarios;
• pricing strategies;
• bidding strategies;
• supply-chain responses;
• strategic retaliation.

For example:

“Simulate what Competitor X will do if we reduce our price by 7%.”

That is not necessarily unlawful.

However, if the system is trained on or receives confidential competitor information, or if competing firms coordinate their use of the same AI system, the competition-law analysis changes substantially.

The central issue remains how the information was obtained and how the model is used.

17. Key Legal Tests

A competition authority examining simulation-driven conduct should generally ask:

Test 1 — Is there an agreement?

Is there:

• an express agreement;
• tacit coordination supported by evidence;
• a hub-and-spoke arrangement;
• a concerted practice?

Test 2 — What information enters the system?

Is it:

• public;
• historical;
• aggregated;
• current;
• granular;
• confidential;
• forward-looking?

Test 3 — Who controls the simulation?

Is it controlled by:

• an individual firm;
• an independent vendor;
• competitors jointly;
• a dominant platform?

Test 4 — What does the model optimise?

Does it optimise:

• individual profits independently;
• consumer welfare;
• efficiency;
• joint industry profitability;
• coordinated pricing?

Test 5 — How autonomous is the system?

Did humans:

• design the rules;
• choose the data;
• approve the recommendation;
• implement the outcome?

Test 6 — What is the competitive effect?

Authorities may examine:

• price increases;
• output reductions;
• foreclosure;
• entry deterrence;
• market allocation;
• reduced innovation;
• reduced consumer choice.

18. Difference Between Legitimate and Problematic Simulation

19. Indian Competition-Law Perspective

Under the Competition Act, 2002, simulation-driven competition can potentially engage several provisions.

Section 3

Relevant where simulation facilitates:

• price fixing;
• output restriction;
• market allocation;
• bid rigging;
• information exchange forming part of an anticompetitive arrangement.

Section 4

Relevant where a dominant undertaking uses simulation technology to:

• exclude competitors;
• discriminate;
• deny access;
• leverage market power;
• impose unfair conditions;
• restrict innovation.

Sections 5 and 6

Simulation may also become relevant to merger analysis where predictive modelling is used to assess the competitive consequences of a combination.

The CCI has recognised that algorithmic systems can create competition concerns involving opacity, dynamic pricing, algorithmic collusion and data-driven decision-making.

20. Emerging Doctrine: Governance of the Algorithm Rather Than the Algorithm Alone

A major conceptual development is the movement from asking:

“Is the algorithm lawful?”

to asking:

“How is the algorithm governed, by whom, using what information, and for what competitive purpose?”

This is particularly important because the same simulation technology can have completely different competition consequences depending upon its governance.

For example:

Independent demand simulation → legitimate competition

Common simulation using confidential rival data → potential information-exchange problem

Simulation implementing an express cartel → price fixing

Dominant platform's simulation designed to exclude rivals → potential abuse of dominance

Independent algorithms independently learning competitive responses → difficult unresolved legal question

21. Compliance Measures

Businesses using competitive simulations should consider:

1. Data classification — identify competitively sensitive information.
2. Data segregation — prevent competitor-specific confidential data from being improperly pooled.
3. Algorithmic firewalls — restrict access to sensitive information.
4. Independent model development — avoid unnecessary joint development with competitors.
5. Human oversight — retain genuine independent pricing and strategic decisions.
6. Model documentation — maintain records of design and training.
7. Competition-law review — review models before deployment.
8. Periodic audits — test outputs for unexplained coordination.
9. Vendor controls — impose competition-law obligations on third-party providers.
10. Employee training — train employees not to exchange sensitive information.
11. Override mechanisms — preserve the ability to depart independently from algorithmic recommendations.
12. Incident reporting — investigate unexpected coordination or data leakage.

22. Six Core Principles for Examinations

Principle 1

Technology is not a defence to cartel conduct.

Trod/GB eye and Topkins illustrate this.

Principle 2

Common software does not automatically create a cartel.

Gibson v. Cendyn demonstrates the importance of proving an actual restraint or agreement.

Principle 3

Electronic communications can establish concerted practices.

Eturas is particularly important.

Principle 4

Competitively sensitive information is central to algorithmic coordination.

This is a major issue in RealPage.

Principle 5

Algorithmic price convergence is not automatically proof of collusion.

Independent algorithms may respond similarly to common market conditions.

Principle 6

The more autonomous the simulation becomes, the more difficult traditional competition-law concepts become to apply.

This is the principal unresolved frontier of simulation-driven competition.

23. Conclusion

Simulation-driven competition represents a transition from traditional human-centred competition to data-centred and machine-mediated competition.

Competition law does not need to prohibit simulations merely because they are sophisticated, predictive or AI-driven. The central questions are whether the simulation:

• preserves independent decision-making;
• relies upon legitimate information;
• facilitates communication between competitors;
• incorporates competitively sensitive information;
• implements an agreement;
• creates exclusionary effects;
• or is controlled by a dominant undertaking in a manner capable of harming competition.

The existing cases provide a developing legal framework. Eturas demonstrates how electronic systems can evidence concerted practices; Trod/GB eye and Topkins establish that algorithms cannot disguise conventional cartels; RealPage shows the significance of common pricing systems and sensitive data; Gibson v. Cendyn cautions that independent use of common software is not automatically unlawful; and Samir Agarwal illustrates the importance of distinguishing algorithmic pricing from a conventional hub-and-spoke arrangement.

The emerging governance model can therefore be expressed as:

Data governance → Algorithm governance → Simulation governance → Human oversight → Independent decision-making → Competition-law compliance

The fundamental challenge for future competition law will be determining when autonomous simulations remain legitimate competitive prediction and when they become instruments that replace independent rivalry with coordinated market behaviour.