What Is Algorithmic Collusion?
Algorithmic collusion happens when pricing algorithms used by competing firms end up setting similar, higher prices without the firms talking to each other. The systems watch each other’s price moves and settle into patterns that reduce competition.
Classical collusion requires intent. Two firms meet, agree on prices, and act on the agreement. Algorithmic collusion challenges that framing. The pricing systems coordinate. The firms running them may not have agreed to anything.
Algorithmic collusion is often discussed in two forms.
- Explicit algorithmic collusion occurs when firms intentionally design or configure pricing systems to coordinate outcomes, or when a shared third-party system is used in a way that facilitates alignment between competitors. In these cases, the behavior can resemble traditional cartel activity if there is evidence of agreement or intent.
- Tacit algorithmic collusion, by contrast, emerges when independent pricing systems, each operating on its own logic, learn to match or stabilize prices through repeated market interaction. No direct communication or agreement is required. This form is significantly harder to classify under existing antitrust frameworks and poses challenges for enforcement.
Pricing software now updates prices constantly and reacts faster than any human team. In digital markets like e-commerce, travel, and ride-sharing, the conditions for this kind of coordination are everywhere.
Synonyms
- AI collusion
- Algorithmic coordination
- Automated price alignment
- Digital tacit collusion
- Parallel pricing behavior
- Pricing algorithm collusion
How Algorithms Enable Collusion in Pricing
Pricing software can now react faster than any human team could manage, and that speed shapes how coordination can emerge.
Role of Algorithms in Pricing Decisions
Manual pricing is slow. A pricing committee meets monthly or quarterly, reviews data, and adjusts prices.
Algorithmic pricing replaces that cycle with continuous updates. Some marketplace pricing systems update thousands of times per day, reacting to demand signals, competitor moves, and inventory changes within seconds.
Algorithmic Pricing Agents
Most firms rely on one of two setups:
- Rule-based systems follow fixed instructions, for e.g., match the lowest competitor price, maintain a 5% margin, never go below cost, etc.
- Adaptive systems use machine-learning models to adjust their behavior based on observed outcomes.
The adaptive systems are where the collusion question gets harder.
Reinforcement Learning and Adaptation
Reinforcement learning is the dominant approach in modern algorithmic pricing research. The system tries pricing actions, observes the results, typically measured as profit or revenue, and updates its strategy to favor actions that produced better outcomes. Run this loop millions of times, and the system develops a pricing strategy that responds to competitor behavior in stable, learned patterns.
The key academic finding came from Calvano, Calzolari, Denicolò, and Pastorello in their 2020 paper in the American Economic Review. They showed that Q-learning pricing algorithms consistently learn to charge prices above the competitive level without ever communicating. The algorithms even punished each other for undercutting, then returned to the higher price together. The same pattern economists have seen in human cartels for decades, produced by software with no intent to collude.
Feedback Loops Between Competitors
Once multiple algorithms operate in the same market, their outputs become each other’s inputs. Algorithm A raises a price. Algorithm B reads that as a signal and raises its own. Algorithm A holds. The loop settles at a higher price than either system would have set alone.
Speed makes it worse. Human teams can’t react fast enough to test coordination. Algorithms can. They can also drop a strategy instantly if it stops working and pick it back up later.
Data Inputs and Statistical Methods
Modern pricing algorithms ingest competitor prices, demand signals, inventory levels, time of day, day of week, historical pricing data, and increasingly, third-party signals like weather or events. The richer the input data, the more nuanced the coordination patterns the system can learn.
Economic Theory and Research on Algorithmic Collusion
Theory caught up to practice slowly. The economics of collusion was built around humans making strategic choices. Algorithms broke the assumptions.
Economic Foundations
At the core is repeated game theory. Firms in repeated competitive interaction have an incentive to maintain higher prices if they can sustain mutual cooperation. Traditional models assumed this required communication or at least clear signaling. Algorithms raised a new question: can coordination emerge from learning alone?
From Human to Algorithmic Models
Cartel models built around human decision-making missed two things. Algorithms operate at a different speed. Algorithms also operate without the moral, legal, and reputational considerations that constrain human collusive behavior. A human pricing manager who coordinates with a competitor risks prosecution. An algorithm has no risk awareness at all.
What the Research Shows
The Calvano et al. (2020) paper changed the conversation. By showing that Q-learning algorithms could reach above-competitive prices through pure learning, the researchers gave economists and regulators a concrete result. Later work by Klein (2021) and Asker, Fershtman, and Pakes (2022) refined when this happens and when it doesn’t.
The algorithms in these studies don’t settle at the competitive price. They reach a higher-price equilibrium that mirrors what colluding firms would set if they had agreed to coordinate. The equilibrium is stable, defended by punishment strategies (drop prices when a competitor undercuts), and resilient to small disruptions.
Impact on Market Efficiency
The trade-off is uncomfortable. Algorithmic pricing improves operational efficiency. But it also reduces price competition. Markets get more responsive but less competitive. For consumers, the net effect is usually higher prices.
Human vs. Algorithmic Collusion
| Feature | Human Collusion | Algorithmic Collusion |
|---|---|---|
| Communication | Direct | None required |
| Speed | Slow | Instant |
| Detection | Easier | Harder |
| Adaptability | Limited | Continuous |
| Legal precedent | Established | Limited |
Hub-and-Spoke vs. Autonomous Algorithmic Collusion
Two structurally different forms of algorithmic collusion sit under the same label, but the legal treatment of each is different.
Hub-and-Spoke
Multiple competitors use the same third-party pricing software. The shared tool is the hub. The competing firms are the spokes. Prices align because the same algorithm is making decisions across all of them. Legally, this is closer to a traditional cartel because the shared software acts like a shared decision-maker.
Autonomous
Each firm runs its own pricing system. The systems learn from competitor behavior and end up coordinating without any shared software, communication, or intent. This is the form Calvano et al. demonstrated in simulation. It’s the hardest form for regulators to address.
Why It Matters
Hub-and-spoke fits inside familiar legal categories like agreement and conspiracy. Autonomous collusion forces a harder question: what counts as coordination when no one decided to coordinate?
| Dimension | Hub-and-Spoke | Autonomous |
|---|---|---|
| Shared infrastructure | Yes | No |
| Direct coordination | Through software | None |
| Human intent | Often present | Often absent |
| Legal precedent | Stronger | Limited |
| Detection | Easier | Very Difficult |
Real-World Cases and Investigations
The pattern over the last decade shows us that enforcement is going after shared-software cases first. Autonomous learning cases are still ahead.
US v. Topkins (2015)
This was the first US criminal case involving algorithmic price coordination. David Topkins and other Amazon poster sellers programmed pricing software to coordinate prices. The coordination was intentional. The algorithm executed the agreement, but did not create it.
The DOJ secured a guilty plea and the case established that automation doesn’t shield firms from antitrust liability when the coordination is intentional.
RealPage Litigation (2022 to 2025)
The DOJ filed its civil suit against RealPage in August 2024, joined by attorneys general from eight states. The complaint said competing landlords shared nonpublic data through RealPage’s pricing software, which then produced coordinated rent recommendations.
The DOJ cited a striking detail: one landlord told RealPage it began raising rents within a week of adopting the software and raised them more than 25% within 11 months.
In November 2025, the DOJ and RealPage settled. RealPage agreed to stop using competitors’ nonpublic data and limit model training to historical data at least 12 months old. The company also accepted a court-appointed monitor.
Lufthansa Fare Investigation (2017 to 2018)
After the collapse of Air Berlin, fares on certain routes increased sharply. Regulators asked whether pricing algorithms played a role. The German Federal Cartel Office investigated and closed the case without finding a violation. The investigation drew public attention to algorithmic pricing in concentrated markets.
Eturas (2016, EU Court of Justice)
This case involved a shared booking system used by travel agencies. Travel agencies in Lithuania used a shared booking platform that capped discounts. The court ruled that knowledge of the system could imply tacit agreement. The decision set EU precedent for shared-software cases.
Risks to Competition and Market Dynamics
Empirical work now shows measurable changes in pricing behavior and market outcomes.
Impact on Pricing
The most consistent finding is a shift in how prices move:
- Less aggressive price competition
- More stable pricing over time
- Higher average price levels in some settings
A 2024 study by Assad, Clark, Ershov, and Xu examined Germany’s retail gasoline market after pricing software became widely adopted. Margins increased by about 9% in competitive markets. In duopoly markets where both firms used pricing algorithms, margins rose by up to 28%.
These results suggest that when multiple firms adopt similar systems, pricing pressure can weaken.
Consumer Effects
Price differences across sellers tend to narrow as AI-powered pricing optimization systems react to the same signals. This reduces the chance of finding lower-priced options at any given time. Discounts still appear, but they are less frequent and often short-lived. The benefits consumers usually gain from competition, especially the ability to shop across a wide price range, weaken as systems converge under similar market conditions.
Business Implications
Firms using pricing algorithms often see more predictable margins and fewer sudden price swings. As dynamic pricing becomes standard, the focus shifts away from competing on price toward managing system performance and inputs. Over time, competition moves toward product features, service, and positioning. That shift can support differentiation, but it changes how firms compete and reduces the role of price as a primary lever.
Market Structure
As more firms adopt similar pricing systems, outcomes can begin to align. This can make markets appear competitive while limiting actual price pressure. Higher and more stable pricing can raise barriers for new entrants, especially those trying to compete on price. Findings from economic research suggest that these effects build gradually, shaping structure over time rather than through sudden changes.
Legal, Regulatory, and Accountability Challenges
Antitrust law was built for human actors making intentional decisions. Algorithmic pricing does not fit neatly into that frame.
What Existing Law Covers
US, EU, and most national antitrust laws focus on agreements and intent. The Sherman Act, Article 101 of the Treaty on the Functioning of the European Union, and similar rules need evidence of an agreement or coordinated practice.
Why Enforcement Is Hard
There is often no direct communication, no shared plan, and no clear moment where coordination begins. Pricing decisions happen quickly and inside systems that are hard to interpret. The usual evidence, emails, meetings, or testimony, is often missing.
Who Is Responsible
Open question. The developer who built the model? The company that deployed it? The system itself, which has no legal status? Current law doesn’t have a clean answer.
The OECD Framework
The OECD’s 2017 report, “Algorithms and Collusion: Competition Policy in the Digital Age,” named four categories: monitoring, parallel, signaling, and self-learning. The report flagged self-learning algorithms as the biggest enforcement gap. It remains the most-cited framework in global competition policy on this topic.
Where Regulators Stand
Regulators are still shaping their approach. The European Commission’s 2017 e-commerce inquiry flagged algorithmic pricing as a watch area. The UK’s Competition and Markets Authority emphasizes monitoring rather than after-the-fact enforcement. The US Federal Trade Commission has held hearings signaling growing willingness to apply existing law to algorithmic conduct.
How Companies Can Reduce Algorithmic Collusion Risk
Firms using pricing systems are expected to manage how those systems behave in the market. Waiting for regulators to draw the line is not an ideal strategy.
Algorithm Audits
Review how pricing systems are built and how they perform. This includes logic, data inputs, and outputs. High-impact systems benefit from independent technical review. Clear documentation of design choices and constraints helps create an audit trail if questions arise later.
Output Monitoring
Watch how prices move over time, especially in relation to competitors. Consistent alignment or unusually stable pricing patterns can signal a problem. If your prices move in step with a key competitor over extended periods, it warrants closer review.
Pricing Guardrails
Set clear boundaries on automated decisions. This can include minimum and maximum price limits, as well as thresholds that trigger human review. Some firms also introduce controlled variation to avoid rigid pricing patterns.
Internal Governance
Oversight should involve more than one team. Legal, compliance, and pricing functions need visibility into how systems operate. Clear escalation paths help address unusual pricing behavior before it becomes a larger issue. Training ensures teams understand the risks tied to automated pricing.
Vendor and Third-Party Review
External pricing tools require the same level of scrutiny. Firms should understand how vendors handle data, especially whether competitor information is pooled or shared. Tools that aggregate sensitive data across competitors create higher exposure.
Risk Reduction Checklist
| Action | Frequency | Owner |
|---|---|---|
| Algorithm audit | Annual | Compliance + Engineering |
| Output monitoring | Continuous | Pricing team |
| Constraint review | Quarterly | Pricing leadership |
| Vendor assessment | Annual | Legal + Procurement |
| Team training | Annual | Compliance |
The Future of Algorithmic Collusion in AI-Driven Markets
Pricing systems are becoming more autonomous, and that shift is continuing across industries.
Algorithmic pricing is expanding beyond e-commerce and travel into sectors like logistics, manufacturing, and services. As adoption grows, competitive pricing decisions are increasingly handled by systems rather than teams.
These systems are also becoming more adaptive. They process larger datasets, update more frequently, and respond to competitor behavior in real time. At the same time, routine pricing decisions are moving away from human oversight. Teams set objectives and constraints, while systems manage execution.
This shift increases the need for transparency. As models become harder to interpret, firms and regulators are placing more emphasis on explainability, monitoring, and auditability.
Policymakers are working to balance efficiency gains with competitive outcomes. That balance remains unsettled.
What to Watch
- Greater regulatory focus on pricing systems
- More guidance on data use and system design
- Continued research on coordination dynamics
- Stronger expectations for internal controls
Pricing is becoming more automated and central to how firms compete. How that shapes competition is still being worked out.
People Also Ask
Can algorithmic collusion happen in competitive markets with many firms?
Yes, but it becomes less likely as the number of firms increases. Coordination is easier in concentrated markets where a small number of competitors interact repeatedly. In markets with many firms, pricing signals are noisier and harder for systems to interpret, which can disrupt stable coordination. Research shows that structure plays a major role in whether algorithms converge on higher pricing patterns.
Is algorithmic collusion illegal?
It sits in a legal gray area for now. Existing antitrust law focuses on agreements and intent, both of which are hard to prove when independent algorithms reach coordinated outcomes through artificial intelligence. Some cases like US v. Topkins (2015) and the RealPage litigation have been prosecuted because they involved shared software or programmed coordination.
Some policymakers are exploring whether persistent pricing patterns or market outcomes could be used as evidence instead. The challenge is separating coordinated behavior from normal competitive responses, especially in markets where firms react to the same signals.
How does uncertainty affect algorithmic collusion?
Uncertainty can disrupt coordination. When demand is volatile or competitor behavior is unpredictable, pricing systems receive weaker signals. This makes it harder for stable patterns to form and can push systems toward more competitive pricing.
Can firms design algorithms to avoid collusive outcomes?
Yes, to a degree. Firms can influence how pricing systems behave, even if they cannot fully control outcomes.
Objective design: Systems that prioritize short-term performance tend to react faster and are less likely to settle into stable price alignment.
Built-in constraints: Price floors, ceilings, and review thresholds limit how far automated decisions can go.
Update frequency: Slower adjustments or deliberate variation reduce predictable pricing patterns.
Data selection: Heavy reliance on competitor pricing increases the chance of aligned outcomes.
Ongoing monitoring: Regular review helps detect when pricing begins to move too closely with competitors.
These steps do not remove risk, but they can reduce the likelihood of sustained coordination.