Algorithmic Interdependence as a Substitute for Explicit Collusion in Competition Law

1. Introduction

Algorithmic interdependence describes a situation in which competing firms independently use algorithms that observe market conditions, competitors' prices, demand, inventory, or other signals and automatically adjust their conduct in response. The algorithms may produce coordinated or parallel market outcomes without a conventional cartel agreement or direct human communication.

The central competition-law problem is therefore:

Can competition law treat algorithmically generated interdependence as equivalent to explicit collusion when firms achieve a coordinated outcome without communicating a traditional agreement?

The answer under existing competition-law frameworks is generally not automatically. Mere parallel pricing, even when produced by sophisticated algorithms, does not ordinarily establish an unlawful agreement by itself. The critical legal question is whether there is evidence of an agreement, concerted practice, exchange of competitively sensitive information, common algorithmic intermediary, or other conduct that substitutes for independent competitive decision-making.

The CMA has expressly distinguished traditional collusion implemented through algorithms, hub-and-spoke arrangements, predictable-agent scenarios, and genuinely autonomous AI coordination. It has also noted that autonomous coordination without human communication remains an emerging legal issue.

2. Meaning of Algorithmic Interdependence

Traditional oligopolistic interdependence occurs where:

• Firm A observes Firm B's price;
• Firm B anticipates Firm A's response;
• both firms independently choose similar prices;
• neither firm communicates with the other.

Algorithms can make this process substantially faster and more systematic.

Example

Suppose three online retailers independently employ AI pricing systems.

The algorithms are programmed to:

1. monitor competitors' prices;
2. predict competitors' reactions;
3. avoid aggressive price competition;
4. raise prices when rivals raise prices;
5. immediately retaliate when a rival lowers its price.

No executive sends an email saying:

"Let us maintain the same price."

Nevertheless, the algorithms may gradually discover that high-price strategies produce greater profits when competitors respond similarly.

The resulting market may therefore resemble a cartel even though there is no conventional cartel communication.

3. Algorithmic Interdependence versus Explicit Collusion

4. The Legal Foundation: Independent Conduct versus Coordination

Competition law traditionally protects the principle that firms must determine their market behaviour independently.

The difficulty is that independence does not require firms to ignore their competitors.

A company is normally permitted to:

• observe competitors' prices;
• respond to market demand;
• anticipate competitors' reactions;
• use publicly available market information;
• employ pricing software.

The legal problem arises when technology changes ordinary interdependence into coordination.

The CMA has observed that algorithms can facilitate information exchange and coordination and that sophisticated systems may potentially create autonomous tacit coordination.

5. The "Substitute for Explicit Collusion" Theory

The theory can be represented as follows:

Competitor A's algorithm

↓ observes

Market signal / Competitor B's price

↓ processes information

Algorithmic prediction

↓ anticipates retaliation

Price adjustment

↓

Competitor B's algorithm responds

↓

Stable coordinated price

The important legal question is not merely:

"Did prices converge?"

Instead:

"What mechanism produced the convergence?"

This distinction is fundamental.

6. Six Major Case Laws

1. Eturas UAB v Lietuvos Respublikos Konkurencijos Taryba — CJEU

Case: C-74/14, Eturas UAB and Others v Lietuvos Respublikos Konkurencijos Taryba (2016).

This is one of the most important European cases involving technology-assisted coordination.

An online travel-booking platform operated a common system used by travel agencies. The system imposed a technical limitation concerning discounts that could be offered to customers.

The issue was whether businesses could be responsible for participating in a concerted practice where the coordination was facilitated through a technological platform rather than conventional bilateral communication.

Principle

The CJEU emphasised that participation in a concerted practice may be established through circumstances surrounding the use of the common technological system, including knowledge of the information communicated through it.

The case demonstrates that:

A digital platform can become the mechanism through which coordination is implemented.

It does not, however, establish that every common algorithmic outcome automatically constitutes collusion.

Relevance

Eturas is particularly important for algorithmic interdependence because it shifts attention from traditional face-to-face cartel meetings to:

• platform architecture;
• system-generated communications;
• knowledge;
• participation;
• technological implementation.

2. United States v. Topkins — U.S.

The Topkins prosecution concerned online retailers selling posters and frames through Amazon Marketplace.

The defendants agreed to coordinate prices and subsequently used pricing software to implement the arrangement.

Principle

The case is significant because the algorithm was not itself the origin of the unlawful agreement.

Instead:

Human agreement → algorithmic implementation → automated price coordination

The software made it easier to monitor competitors and maintain the agreed pricing strategy.

Importance for algorithmic interdependence

This establishes an important distinction:

An algorithm can be the instrument of a cartel even when it is not the source of the cartel.

That distinction is crucial when analysing newer forms of autonomous algorithmic coordination.

3. Trod Ltd / GB Eye Ltd — UK Competition Law

The UK poster-pricing case involved competing online sellers that agreed not to undercut one another on Amazon Marketplace.

They used automated repricing software to implement the arrangement.

The CMA has subsequently used this case as an example of how pricing algorithms can implement unlawful price-fixing.

Legal significance

The case illustrates:

• explicit agreement;
• algorithmic enforcement;
• rapid monitoring;
• automatic price adjustment;
• reduction of opportunities for unilateral deviation.

The algorithm effectively became a digital enforcement mechanism for the cartel.

Distinction

It is therefore different from genuinely autonomous algorithmic interdependence because the underlying coordination was human-created.

4. United States v. RealPage, Inc. — U.S.

The RealPage litigation represents the modern algorithmic hub-and-spoke problem.

The controversy concerns allegations that landlords used a common pricing platform and provided information to the platform that was used in generating rent recommendations.

The legal theory is substantially different from simple independent use of pricing software.

The relevant question is whether a common intermediary can facilitate the exchange or use of competitively sensitive information and thereby replace independent decision-making.

The contemporary debate surrounding RealPage has made the distinction between independent algorithmic pricing and algorithm-mediated coordination especially important.

Competition-law significance

The case illustrates a potential structure:

Competitor A →

Common algorithm / intermediary

← Competitor B

The intermediary may therefore function as a coordination hub.

The algorithm does not necessarily need to communicate the identity of the rival or disclose raw data directly. Its recommendations themselves can potentially reduce strategic uncertainty.

5. Theatre Enterprises, Inc. v. Paramount Film Distributing Corp. — U.S. Supreme Court

This classic U.S. antitrust case concerned parallel conduct among competitors.

The Supreme Court made an important distinction between:

• lawful parallel conduct; and
• unlawful concerted action.

Principle

Parallel behaviour alone does not necessarily establish a conspiracy.

Competitors may independently arrive at similar commercial decisions because the economic conditions confronting them are similar.

Importance for algorithmic interdependence

The principle becomes even more important when algorithms produce similar decisions.

Suppose ten retailers independently employ similar AI systems.

Their algorithms may all conclude:

"The optimal price is ₹999."

Identical pricing does not, standing alone, prove an agreement.

Therefore:

Algorithmic similarity ≠ automatic collusion.

The authority must investigate the mechanism generating the similarity.

6. Matsushita Electric Industrial Co. v. Zenith Radio Corp. — U.S. Supreme Court

This case is another foundational authority concerning proof of conspiracy.

The Supreme Court emphasised the importance of distinguishing lawful independent conduct from conduct resulting from an unlawful agreement.

Relevance

Algorithmic markets frequently exhibit:

• rapid price matching;
• identical prices;
• simultaneous price increases;
• retaliatory discounting;
• stable market prices.

Those facts can be economically suspicious but do not automatically establish an unlawful agreement.

The evidentiary question becomes particularly important where algorithms independently react to the same public information.

7. Additional Important Authority: Airtours v Commission

The EU decision in Airtours is highly relevant to the economic concept of collective dominance / coordinated effects.

The General Court examined whether market conditions were conducive to sustainable coordination.

Important factors included:

• market transparency;
• ability to monitor deviations;
• incentives to retaliate;
• predictability of competitors' behaviour.

These factors have obvious algorithmic parallels.

Algorithmic transformation

Traditional oligopoly:

High transparency → easy monitoring → retaliation → coordination.

Algorithmic oligopoly:

Real-time data → automated monitoring → instantaneous retaliation → potentially sustainable coordination.

Thus, algorithms may dramatically increase the speed and precision of monitoring.

8. Indian Position: Samir Agrawal v Competition Commission of India

The Indian competition-law context is particularly important.

In Samir Agrawal v Competition Commission of India, allegations were made that Ola and Uber's algorithmic pricing systems facilitated coordination among drivers.

The allegation was essentially that drivers who otherwise competed independently were subjected to fares calculated through the platforms' algorithms.

The CCI did not find a sufficient agreement or arrangement establishing the alleged price-fixing theory, and the Supreme Court ultimately upheld the closure of the matter.

Significance

The case demonstrates an important proposition:

The existence of algorithmically determined prices does not by itself establish an anti-competitive agreement.

The legal analysis must still establish the statutory elements of coordination.

For India, this is particularly relevant to Section 3 of the Competition Act, 2002, including the requirement of an agreement, arrangement or understanding in cases of concerted conduct.

9. Four Models of Algorithmic Coordination

The CMA's recent work identifies several conceptually distinct forms of algorithmic collusion.

A. Algorithm as Cartel Enforcement Tool

Human beings agree to collude.

The algorithm implements the agreement.

Example:

Trod/GB Eye and Topkins.

Legal position

Generally the least problematic category from an enforcement perspective because the underlying agreement already exists.

B. Algorithmic Hub-and-Spoke Coordination

Several competitors use:

One algorithm / common intermediary / common data source.

The algorithm may aggregate information and generate recommendations.

Risk

The intermediary may reduce uncertainty about:

• prices;
• inventory;
• demand;
• capacity;
• future strategies.

This can potentially substitute for direct communication.

C. Predictable-Agent Coordination

Each company independently uses an algorithm designed to respond predictably to competitors.

For example:

"If competitor increases price by 5%, immediately increase our price by 5%."

No direct communication is necessary.

But repeated predictable responses may make deviations from a coordinated price unattractive.

This resembles classical oligopolistic coordination but is intensified by automation.

D. Autonomous AI Coordination

This is the most difficult scenario.

Two independent AI systems are given objectives such as:

"Maximise long-term profit."

The systems interact repeatedly and independently learn that maintaining high prices produces greater expected returns.

No human:

• communicates;
• agrees;
• exchanges confidential information;
• instructs the competitor;
• explicitly establishes the coordinated strategy.

The algorithms nevertheless converge on coordinated conduct.

The CMA describes this as an emerging possibility and recognises that the legal treatment remains unsettled.

10. Why Algorithms Can Make Interdependence Stronger

10.1 Real-time monitoring

Algorithms can continuously monitor competitors.

Traditional human monitoring:

once a day/week.

Algorithmic monitoring:

potentially continuously.

This reduces the time available for a rival to deviate from a coordinated outcome.

10.2 Automatic retaliation

Suppose Firm A decreases its price.

The algorithm of Firm B can immediately:

1. detect the reduction;
2. classify it as competitive aggression;
3. reduce its own price;
4. trigger further responses.

This can create a retaliation loop.

10.3 Increased market transparency

Algorithms can observe:

• prices;
• inventory;
• promotions;
• delivery times;
• product availability;
• search rankings;
• consumer demand.

Greater transparency can reduce uncertainty between competitors.

The CMA has identified this as one of the ways algorithmic systems can facilitate coordination.

11. The "Meeting of Minds" Problem

Traditional competition law often looks for some form of:

meeting of minds

or

common understanding

between independent undertakings.

The difficulty with autonomous AI is that there may be:

No meeting of minds between humans.

Instead there may be:

Machine-to-market adaptation → machine-to-machine reaction → stable equilibrium.

This creates a fundamental attribution problem.

Traditional model

Human A
↓
Agreement
↓
Human B

Algorithmic model

Human A → Algorithm A
↓
Market interaction
↑
Human B → Algorithm B

The algorithms may interact without their owners ever communicating.

12. Can Tacit Collusion Be Punished?

Generally, mere conscious parallelism or tacit interdependence is not automatically equivalent to an unlawful cartel.

This follows from the fundamental distinction between:

Lawful interdependence

"I know my competitor will react if I lower prices, so I independently choose not to lower them."

and

Unlawful coordination

"We have established, directly or indirectly, a mechanism through which our prices will be coordinated."

Algorithms make this distinction more difficult, but they do not eliminate it.

13. Evidence in Algorithmic Cases

Competition authorities increasingly need evidence beyond emails and meeting records.

Important evidence may include:

A. Source code

Investigators may examine:

• pricing rules;
• objective functions;
• constraints;
• retaliation mechanisms;
• competitor-monitoring functions.

B. Training data

Questions include:

• What data trained the model?
• Was competitor data used?
• Was commercially sensitive information incorporated?

C. API architecture

An investigation may examine whether several competitors' systems receive information from the same intermediary.

D. Logs

Logs may establish:

• what the algorithm observed;
• what recommendation it produced;
• what the company did in response.

E. Version history

Changes to software may reveal whether the algorithm was deliberately modified to produce particular outcomes.

F. Human instructions

Internal instructions such as:

"Never undercut competitor X"

can be especially significant.

14. Economic Evidence

Algorithms also create unusually rich quantitative evidence.

Authorities can examine:

• price correlations;
• speed of price responses;
• frequency of retaliation;
• deviation patterns;
• price dispersion;
• margins;
• market shares;
• algorithmic recommendation histories.

However, correlation is not necessarily proof of collusion.

Two algorithms may produce similar prices simply because:

• they observe the same market data;
• they optimise against the same demand conditions;
• they use similar economic models.

15. The Counterfactual Test

A useful analytical question is:

What would the firms' conduct have looked like if the algorithm had not facilitated the coordination?

Authorities can compare:

Counterfactual A

Independent human pricing.

Counterfactual B

Algorithmic pricing without competitor information.

Counterfactual C

Algorithmic pricing incorporating competitor information.

Counterfactual D

Pricing through a common intermediary.

The greater the evidence that the algorithm deliberately reduces competitive uncertainty or coordinates reactions, the stronger the competition-law concern becomes.

16. Algorithmic Interdependence and Oligopoly

Algorithmic interdependence is especially significant in concentrated markets.

Suppose a market contains only three major firms.

If:

• prices are transparent;
• products are relatively homogeneous;
• algorithms monitor competitors;
• deviations are detected immediately;
• retaliation is automatic;

then coordinated outcomes may become more sustainable.

This resembles the economic conditions traditionally associated with oligopolistic coordination.

The important distinction is that algorithms can make:

monitoring + detection + retaliation

much cheaper and faster.

17. Competition-Law Risks

Algorithmic interdependence may produce:

1. Higher prices

Algorithms may converge on prices above competitive levels.

2. Reduced discounting

Automatic retaliation can discourage price reductions.

3. Reduced innovation

Firms may become less motivated to innovate when aggressive competition is automatically punished.

4. Market foreclosure

Algorithms can favour incumbents and react aggressively to entrants.

5. Reduced consumer choice

Coordinated pricing may make alternative suppliers less attractive.

6. Information-exchange risks

Common platforms may indirectly facilitate exchange of commercially sensitive information.

The CMA has identified all of these types of algorithmic competition concerns as areas requiring attention.

18. Difference Between Algorithmic Coordination and Ordinary Dynamic Pricing

Dynamic pricing itself is not inherently anti-competitive.

For example, an airline may independently change prices because:

• demand increases;
• seats become scarce;
• fuel costs change;
• a flight approaches departure;
• inventory falls.

That is ordinary commercial optimisation.

The concern increases when algorithms:

• use competitors' confidential information;
• are supplied by a common intermediary;
• intentionally implement an agreement;
• coordinate retaliation;
• exchange competitively sensitive information;
• deliberately avoid competitive deviation.

The CMA expressly recognises that algorithmic pricing can generate legitimate efficiency benefits while also creating competition risks.

19. Algorithmic Interdependence as a "Substitute" for Communication

The most important theoretical proposition is that technology can potentially substitute for some functions traditionally performed by cartel communication.

Traditional cartel

Communication performs four functions:

1. establish the target price;
2. communicate deviations;
3. monitor compliance;
4. punish deviations.

Algorithmic coordination

An algorithm can potentially perform all four functions:

1. infer the target price;
2. detect deviations;
3. monitor competitors continuously;
4. automatically respond to deviation.

Thus:

The economic function of communication may survive even when conventional communication disappears.

But from a legal perspective, economic equivalence does not automatically mean legal equivalence.

20. Attribution of Liability

One of the hardest questions is:

Who should be liable when an algorithm produces anti-competitive coordination?

Possible actors include:

A. The company using the algorithm

Liability may arise where the company:

• knowingly deploys the system;
• instructs it to coordinate;
• ignores obvious risks;
• provides competitively sensitive information.

B. Algorithm provider

Potential responsibility may arise where the provider:

• designs the system to coordinate rivals;
• knowingly incorporates confidential competitor data;
• facilitates anti-competitive information exchange.

C. Common intermediary

A platform may become particularly significant where it:

• aggregates competitor information;
• generates common recommendations;
• controls the information flow.

D. Autonomous algorithm itself

The algorithm itself generally does not possess independent legal personality merely because it makes autonomous decisions.

Consequently, legal systems generally have to attribute conduct to a natural or legal person.

21. Compliance Principles

Businesses using algorithmic pricing systems should consider:

1. Do not use competitors' confidential information.
2. Do not instruct algorithms to follow competitors' prices automatically where this creates coordination risks.
3. Audit third-party pricing software.
4. Understand the data sources used by the algorithm.
5. Maintain algorithmic audit trails.
6. Test whether the system retaliates against competitors' deviations.
7. Separate competitor information from pricing models.
8. Review common-platform arrangements carefully.
9. Document legitimate independent pricing objectives.
10. Conduct competition-law review before deploying autonomous pricing systems.

The CMA specifically advises businesses to understand how their pricing technology works and to ensure that pricing recommendations are not influenced by competitors' confidential information.

22. Emerging Legal Doctrine

A possible future doctrinal framework may distinguish three situations:

Category I — Explicit algorithmic collusion

Human agreement + algorithm

→ Existing cartel principles apply.

Category II — Facilitated algorithmic coordination

Common intermediary + information exchange + algorithm

→ Existing concerted-practice/hub-and-spoke principles may apply depending on evidence.

Category III — Autonomous algorithmic interdependence

Independent firms + independent algorithms + no communication

→ Existing competition law faces its greatest conceptual difficulty.

The third category raises the question whether competition law should remain dependent upon proof of human coordination or develop additional mechanisms for addressing algorithmically sustained coordinated outcomes.

The CMA itself has described this autonomous scenario as an emerging issue for which there is not yet an overall consensus.

23. Key Doctrinal Principles from the Cases

24. Conclusion

Algorithmic interdependence represents a significant challenge to the traditional architecture of competition law because algorithms can reproduce some economic functions of explicit collusion without necessarily requiring direct human communication.

The crucial distinction is between:

independent algorithmic adaptation

and

algorithmically facilitated coordination.

Existing case law strongly supports enforcement where an algorithm implements an existing agreement, facilitates information exchange, or operates as part of a coordinated arrangement. Eturas, Topkins, Trod/GB Eye, RealPage, and Samir Agrawal demonstrate different aspects of this problem.

The more difficult category is autonomous algorithmic interdependence, where independently designed systems learn that coordinated behaviour is profitable without any demonstrable communication or agreement between firms. Traditional competition law generally does not equate parallel outcomes alone with collusion. The emerging policy question is therefore whether existing concepts of agreement, concerted practice, communication, attribution and evidentiary proof are sufficient for markets in which machines can coordinate behaviour faster and more reliably than humans.

In short:

Algorithms may substitute economically for explicit collusion, but they do not automatically substitute legally for the agreement or concerted practice traditionally required to establish an infringement.

The future challenge for competition authorities is consequently not simply detecting identical algorithmic prices, but determining whether the architecture, data, instructions, intermediary relationships and resulting conduct demonstrate a loss of genuine competitive independence.