Competition Law and Governance of Prediction-Based Economies

1. Introduction

A prediction-based economy is an economic environment in which firms use historical and real-time data, artificial intelligence (AI), machine learning, algorithms, behavioural profiles and automated decision systems to predict future consumer behaviour, demand, prices, risks, preferences, competitor responses and market conditions.

Examples include:

• predictive pricing and dynamic pricing;
• credit and insurance-risk prediction;
• demand forecasting;
• personalised advertising;
• recommendation and ranking systems;
• predictive search;
• predictive inventory management;
• ride-hailing and delivery allocation;
• algorithmic investment and trading;
• fraud detection;
• AI-driven recruitment and procurement;
• predictive healthcare and diagnostics;
• autonomous commercial decision-making.

Competition law traditionally examines conduct after it occurs. Prediction-based markets increasingly allow firms to anticipate conduct before it occurs. This changes the competitive process itself. OECD analysis notes that AI can increase market transparency and accelerate competitors' reactions, potentially facilitating coordination, while also creating exclusionary and data-access concerns.

The central competition-law question is therefore:

When does legitimate prediction and optimisation become a mechanism for collusion, exclusion, discrimination, exploitation or durable market power?

Importantly, the mere use of predictive algorithms is not unlawful. Prediction can lower costs, improve products, reduce waste and enable smaller firms to compete. The legal issue arises from how the predictive capability is designed, deployed and connected to market power or coordination.

2. Meaning and Characteristics

A prediction-based economy generally has five interconnected elements:

A. Data

Firms collect:

• transaction data;
• consumer-search data;
• location information;
• purchasing history;
• browsing behaviour;
• competitor information;
• prices;
• inventory data;
• credit information;
• social-network information.

Data may function as an important competitive input, particularly where its scale, quality or uniqueness creates advantages that rivals cannot readily reproduce. OECD research identifies access to quality data and computing power as potential competitive bottlenecks in AI markets.

B. Predictive models

Machine-learning models transform historical and real-time information into predictions concerning:

• willingness to pay;
• future demand;
• customer churn;
• competitor reactions;
• optimal prices;
• supply requirements;
• consumer preferences.

C. Automated decisions

Predictions can directly determine:

• price;
• ranking;
• advertising;
• access;
• recommendations;
• credit;
• discounts;
• delivery priority;
• product visibility.

D. Feedback loops

The decision itself produces new data.

For example:

Data → prediction → price → consumer reaction → new data → improved prediction → new price.

This creates a potentially self-reinforcing competitive advantage.

E. Network and ecosystem effects

A platform with millions of users may generate substantially more data than a smaller rival. Better data may produce better predictions, which attract more users, generating still more data.

This can create:

Data → Prediction → Better service → More users → More data

and potentially produce durable market power.

3. Competition-Law Framework

Prediction-based economies generally remain subject to ordinary competition law, although traditional concepts may have to be applied to technologically sophisticated conduct.

The principal areas are:

1. anti-competitive agreements and cartels;
2. exchange of competitively sensitive information;
3. algorithmic coordination;
4. abuse of dominance;
5. predatory or exclusionary pricing;
6. self-preferencing;
7. discriminatory access;
8. tying and bundling;
9. refusal to supply data or infrastructure;
10. anti-competitive mergers and acquisitions;
11. exploitation of data advantages;
12. consumer-facing personalised discrimination where connected with dominance.

China's 2026 Internet Platform Antimonopoly Compliance Guidelines expressly recognise risks associated with AI, big-data analysis and predictive algorithms, including their use to influence resale prices, coordinate conduct and produce discriminatory or excessive price adjustments. The Guidelines also recommend screening pricing algorithms, recommendation systems, ranking mechanisms and advertising algorithms.

4. Algorithmic Collusion

One of the most important problems is algorithmic collusion.

Suppose several competing firms employ predictive pricing systems. Each system continuously observes market prices and predicts competitors' responses.

The algorithms may learn:

"If I reduce my price, the rival will immediately match it."

Eventually, each algorithm may maintain a higher price because it predicts that aggressive competition will trigger retaliation.

The difficult legal question is whether this constitutes:

• an agreement;
• concerted practice;
• conscious parallelism;
• unilateral rational behaviour; or
• unlawful coordination facilitated by a third-party algorithm.

OECD analysis distinguishes between explicit algorithmic collusion, hub-and-spoke coordination and autonomous algorithmic coordination.

5. Information Exchange

Prediction becomes particularly problematic when competitors feed non-public strategic information into a common algorithm.

Potential information includes:

• future prices;
• inventories;
• capacity;
• discounts;
• occupancy;
• expected demand;
• strategic plans.

A common algorithmic intermediary can therefore become a coordination hub.

The fact that competitors do not communicate directly with one another does not necessarily eliminate competition-law concerns. The legal analysis focuses on the economic substance of the coordination and the evidence demonstrating the relationship between information sharing and market conduct.

6. Predictive Pricing

Predictive pricing can benefit consumers through:

• lower prices;
• better inventory allocation;
• reduced shortages;
• more efficient logistics.

However, risks may arise where prediction is used for:

1. Personalised exploitation

Prices are individually adjusted according to predicted willingness to pay.

2. Coordinated pricing

Competitors use common or interconnected systems that reduce competitive uncertainty.

3. Predatory pricing

A dominant undertaking predicts a rival's vulnerability and strategically prices below relevant cost to eliminate it.

4. Exclusion

The algorithm predicts which consumers or suppliers are most important and selectively disadvantages rivals.

5. Excessive price adjustment

The algorithm repeatedly increases prices beyond competitive levels.

China's current platform guidance specifically identifies "price over-adjustment," uniform pricing recommendations and strategic algorithm sharing as areas requiring algorithmic scrutiny.

7. Prediction and Abuse of Dominance

A dominant platform can use predictive systems to reinforce its position.

For example:

Dominant platform → observes rival → predicts rival's strategy → modifies ranking/price/access → rival loses users → platform gains data → prediction improves.

Possible Article 102 TFEU / national-law theories include:

• exclusionary discrimination;
• refusal of access;
• tying;
• self-preferencing;
• predatory pricing;
• leveraging;
• margin squeeze;
• discriminatory ranking.

The European Commission's 2026 Article 102 Guidelines expressly address exclusionary conduct by dominant undertakings, building on EU judicial decisions and enforcement experience.

8. Prediction as a Barrier to Entry

A new entrant may possess an excellent product but lack:

• historical datasets;
• computing resources;
• user behaviour data;
• training data;
• distribution channels;
• feedback data.

Consequently, the incumbent's predictive model may become difficult to replicate.

This can create a data-and-learning barrier to entry.

However, the existence of a superior predictive model alone does not establish an infringement. Competition authorities must generally establish relevant market power and an anti-competitive effect or exclusionary mechanism.

9. Feedback Loops and Market Tipping

Prediction can produce self-reinforcing advantages.

Example

A large platform:

1. collects enormous amounts of data;
2. trains a predictive model;
3. provides highly personalised recommendations;
4. attracts additional users;
5. receives additional behavioural data;
6. improves its predictive accuracy;
7. attracts still more users.

The result may be a data feedback loop.

Where rivals cannot obtain comparable data, this may contribute to:

• increasing concentration;
• high entry barriers;
• reduced multi-homing;
• dependence on the incumbent;
• reduced innovation incentives.

Recent OECD research describes AI markets as dynamic but notes that concentration in certain layers, particularly around hardware and data, can contribute to long-term entrenchment.

10. Prediction and Self-Preferencing

A platform may predict which products will perform well and then manipulate its own ranking system.

For example:

Third-party sellers → platform marketplace → predictive ranking algorithm → platform's own products receive preferred visibility.

The relevant competition question is whether the platform is merely improving search quality or using its predictive infrastructure to foreclose competitors.

This connects prediction-based competition to the broader jurisprudence concerning platform self-preferencing and search neutrality.

11. Prediction and Personalised Advertising

Advertising systems predict:

• consumer interests;
• purchasing probability;
• conversion probability;
• price sensitivity;
• likelihood of switching.

A dominant advertising platform may possess advantages because it has access to vast quantities of behavioural data.

Potential competition issues include:

• foreclosure of rival advertising exchanges;
• discriminatory access to data;
• tying advertising services to other platform services;
• self-preferencing;
• exclusion of competing ad-tech intermediaries;
• leveraging data from one market into another.

12. Prediction and Mergers

Prediction-based markets create special merger concerns.

Traditional merger analysis asks:

What happens to prices, output and competition after the merger?

In data-intensive markets, authorities may also ask:

What predictive capability will the combined firm acquire?

A transaction can combine:

• consumer data;
• behavioural datasets;
• cloud computing;
• AI models;
• distribution;
• advertising infrastructure.

Potential theories of harm include:

A. Data concentration

The merged firm obtains datasets that rivals cannot replicate.

B. Model concentration

The merger gives one firm superior predictive capabilities.

C. Vertical foreclosure

An AI-model provider may disadvantage downstream competitors.

D. Killer acquisition

An incumbent may acquire a potentially disruptive predictive-technology start-up.

E. Ecosystem expansion

The acquisition may allow the firm to connect previously separate datasets.

OECD's recent work identifies acquisition of AI start-ups by large incumbents as an area relevant to competitive dynamics.

13. Governance Mechanisms

Effective competition governance of prediction-based economies should include several layers.

A. Algorithmic auditing

Competition authorities and firms should examine:

• input data;
• model objectives;
• optimisation functions;
• output patterns;
• pricing changes;
• ranking changes;
• competitor responses.

B. Explainability

Authorities may need sufficient information to determine:

• why the algorithm produced an outcome;
• what variables were used;
• whether competitors' information was incorporated;
• whether discriminatory rules were embedded.

China's 2026 guidance encourages algorithmic screening, dynamic monitoring, explainability and audit records.

C. Data-access governance

Where data constitutes an essential competitive input, authorities may examine:

• refusal to provide access;
• discriminatory access;
• excessive restrictions;
• interoperability barriers;
• data portability.

D. Human oversight

Critical competitive decisions should not necessarily be left entirely to autonomous systems.

Human oversight can help identify:

• unlawful coordination;
• discriminatory rules;
• unintended exclusion;
• abnormal price movements.

E. Audit trails

Firms should preserve:

• model versions;
• training datasets;
• decision logs;
• pricing records;
• parameter changes;
• communications with algorithm providers.

These become particularly important in antitrust investigations.

14. Important Case Laws

Because prediction-based economies are a relatively new analytical category, there are comparatively few reported decisions expressly titled "prediction-based economy" cases. The following cases provide the principal doctrinal foundations for analysing predictive algorithms, automated pricing, platform data and algorithmically mediated competition.

Case 1: United States v. Topkins — Algorithmic Pricing

Jurisdiction: United States
Area: Algorithmic price-fixing

This is one of the clearest early examples of algorithm-assisted cartel conduct.

The Department of Justice prosecuted an online poster seller whose pricing system was used in connection with an agreement among competitors to maintain prices.

Competition-law significance

The important principle is:

An algorithm does not immunise an otherwise unlawful agreement.

A firm cannot convert cartel conduct into lawful conduct merely by implementing the agreed pricing strategy through software.

Relevance to prediction-based economies

Prediction systems can be used to automate an unlawful strategy. The legal inquiry therefore remains focused on:

• communication;
• agreement;
• implementation;
• competitive effect.

Case 2: Eturas v Lietuvos Respublikos konkurencijos taryba

Court: Court of Justice of the European Union
Case: C-74/14
Area: Online platform / algorithmic coordination

In Eturas, travel agencies used an electronic booking system. A centrally implemented system imposed a limitation on discounts.

The CJEU considered whether the participating businesses could be held responsible for concerted conduct through the platform.

Principle

Digital architecture can provide evidence of coordinated conduct.

The absence of traditional face-to-face cartel meetings does not automatically prevent competition-law liability.

Relevance

Eturas is highly relevant to prediction-based markets because the platform infrastructure itself can become the mechanism through which competitive behaviour is coordinated.

Case 3: RealPage Algorithmic Pricing Litigation

Jurisdiction: United States
Area: Rental housing / algorithmic pricing

The U.S. Department of Justice sued RealPage in 2024, alleging that competing landlords supplied non-public competitively sensitive information to RealPage for its algorithmic pricing system and that the arrangement reduced competition in rental pricing. The complaint invoked Sections 1 and 2 of the Sherman Act.

Competition-law significance

The case illustrates a modern theory:

competitor information → common pricing algorithm → predicted market conditions → coordinated pricing outcomes.

Importance

It demonstrates that competition authorities are examining not merely explicit price-fixing agreements but also the architecture through which competitors exchange information and make pricing decisions.

The allegations remain allegations unless and until established through adjudication or settlement.

Case 4: Google Shopping

Institution: European Commission / EU Courts
Area: Search algorithms and self-preferencing

The Google Shopping litigation concerned Google's treatment of comparison-shopping services within its search results.

Competition issue

The case illustrates how a dominant platform's ranking and search architecture can affect rivals' ability to compete.

Relevance to prediction-based economies

Predictive search and ranking systems can determine:

• visibility;
• traffic;
• consumer attention;
• conversion;
• competitive opportunities.

Thus, algorithmic ranking can become a competition-law instrument even without traditional price manipulation.

Case 5: Google Android

Institution: European Commission / EU Courts
Area: Platform ecosystem, tying and dominance

The Android case concerned Google's contractual arrangements involving Android devices, including applications and search services.

Competition-law relevance

The case illustrates how control over a technological ecosystem can be leveraged across connected markets.

Prediction-based economy connection

Predictive systems often operate across ecosystems rather than isolated markets.

A firm controlling:

• operating systems;
• data;
• applications;
• advertising;
• search;
• recommendation systems

can potentially use information obtained in one layer to strengthen another.

The broader lesson is that market power should sometimes be analysed across an ecosystem rather than through a single technological function.

Case 6: Google Search (AdSense)

Institution: European Commission / EU Courts
Area: Digital advertising / exclusion

The Google AdSense case concerned contractual restrictions affecting competing search-advertising intermediaries.

Relevance

Prediction-based advertising depends heavily upon:

• consumer data;
• search information;
• targeting;
• advertiser data;
• predictive matching.

Restrictions imposed by a dominant intermediary can therefore affect the development of competing predictive advertising systems.

Principle

Dominance in one digital layer can be used to affect competition in an adjacent market.

Case 7: Amazon Marketplace / European Commission Amazon Buy Box and Marketplace Concerns

Institution: European Commission
Area: Marketplace data and platform conduct

European competition authorities examined Amazon's use of non-public marketplace seller information and the treatment of sellers within its platform.

Prediction-economy relevance

A platform possessing data about:

• sales;
• demand;
• prices;
• inventory;
• seller performance

may obtain an informational advantage unavailable to independent sellers.

That information can improve predictive systems concerning:

• product demand;
• pricing;
• inventory;
• consumer behaviour.

The resulting competition issue is whether the platform uses its informational advantage to compete unfairly with dependent businesses.

15. Indian Competition-Law Relevance

Although the expression "prediction-based economy" is not a separate statutory category under Indian competition law, the Competition Act, 2002 can address conduct arising in predictive markets through established doctrines.

Relevant provisions include:

Section 3

Anti-competitive agreements, including:

• price fixing;
• information exchange;
• market allocation;
• resale-price restrictions.

Section 4

Abuse of dominant position, including:

• unfair or discriminatory conditions;
• unfair pricing;
• denial of market access;
• tying;
• leveraging.

Sections 5 and 6

Regulation of combinations and mergers.

Section 19

Investigation into:

• agreements;
• dominance;
• relevant market;
• competitive effects.

Section 26

Investigation procedure.

Section 27

Orders against anti-competitive conduct.

Section 32

Extra-territorial conduct affecting competition in India.

These provisions are technologically neutral and can therefore apply to algorithmically mediated conduct.

16. China: Particularly Important Development

China provides an especially relevant contemporary example because its 2026 Internet Platform Antimonopoly Compliance Guidelines expressly address algorithmic conduct.

The guidance identifies risks arising from:

• big-data analysis;
• AI;
• predictive algorithms;
• automated resale pricing;
• data sharing;
• platform rules;
• recommendation systems;
• ranking systems;
• advertising algorithms.

It recommends dynamic algorithm screening and monitoring and calls attention to discriminatory design, excessive price adjustments, uniform pricing recommendations and strategic algorithm sharing.

This represents a movement from traditional ex post enforcement toward ex ante algorithmic compliance governance.

17. Six Major Competition Concerns

18. Efficiency Defences

Prediction-based technologies can generate significant efficiencies.

Examples include:

• reduced logistics costs;
• better inventory management;
• lower transaction costs;
• improved fraud detection;
• reduced waste;
• personalised products;
• improved matching between buyers and sellers;
• improved forecasting;
• lower search costs.

Therefore, competition law should not treat algorithmic sophistication itself as suspicious.

The proper inquiry is whether the predictive system produces:

competition-enhancing efficiencies

or

market-power-enhancing exclusion or coordination.

OECD research emphasises that AI can lower entry barriers, reduce minimum efficient scale and facilitate innovation, while also producing data-access, model-restrictiveness and vertical-integration risks.

19. Evidentiary Problems

Prediction-based cases create unusual evidentiary challenges.

Authorities may need to obtain:

• source code;
• model architecture;
• training data;
• model outputs;
• system logs;
• API records;
• algorithmic instructions;
• communications with technology providers;
• parameter changes;
• A/B testing records.

A major issue is attribution.

If an autonomous algorithm produces anti-competitive results, the authority must determine:

Who made the relevant decision—the company, its programmers, its managers, the algorithm provider, or the algorithm itself?

Current competition law generally does not treat the algorithm as an independent legal person. Responsibility therefore normally has to be connected to the undertaking and its human or organisational conduct.

OECD research identifies attribution of liability as one of the emerging issues created by increasingly autonomous AI systems.

20. Proposed Governance Model

A useful governance framework can be expressed as:

DATA
↓
PREDICTIVE MODEL
↓
AUTOMATED DECISION
↓
MARKET EFFECT
↓
COMPETITION AUDIT

At each stage, regulators can ask:

Stage 1 — Data

Was competitively sensitive information improperly obtained or shared?

Stage 2 — Model

Does the model incorporate competitors' confidential information?

Stage 3 — Decision

Does the model produce discriminatory or exclusionary decisions?

Stage 4 — Market

Does the conduct foreclose rivals or facilitate coordination?

Stage 5 — Audit

Can the firm demonstrate legitimate objectives, compliance and corrective measures?

21. Key Legal Principles

The emerging law of prediction-based economies can therefore be summarised through ten principles:

1. Algorithms do not create an exemption from competition law.
2. Automated conduct can constitute commercially significant conduct of an undertaking.
3. Competitor data remains competitively sensitive even when processed by software.
4. A common algorithm may function as a coordination mechanism.
5. Predictive ranking can produce exclusionary effects.
6. Data advantages may contribute to barriers to entry.
7. Dominant platforms require particular scrutiny where predictive systems affect dependent businesses.
8. Algorithmic pricing may be pro-competitive or anti-competitive depending on its purpose, design and effects.
9. Mergers involving data and predictive capabilities require analysis beyond traditional price effects.
10. Competition compliance increasingly requires algorithmic auditing and governance.

22. Conclusion

Competition law and governance of prediction-based economies represent an evolution of traditional antitrust principles rather than an entirely separate branch of competition law.

The fundamental legal concepts—agreement, coordination, dominance, exclusion, discrimination, tying, refusal of access and merger control—remain applicable. What changes is the mechanism through which competitive harm can occur.

In a traditional market, competitors may react to each other's observable behaviour. In a prediction-based economy, algorithms can anticipate behaviour, learn from market reactions and automatically adjust commercial decisions.

Consequently, competition authorities increasingly need to examine not only what firms did, but also:

• what information their systems received;
• what their algorithms were designed to optimise;
• how algorithms interacted with competitors;
• whether predictive feedback loops reinforced dominance;
• whether data advantages foreclosed rivals; and
• whether automated systems facilitated coordination.

The emerging regulatory direction is therefore toward a combination of traditional antitrust enforcement, algorithmic auditing, data governance, merger scrutiny, transparency, record-keeping and ex-ante compliance. OECD's recent work similarly describes algorithmic competition as an area where traditional theories of harm remain important but may require adaptation to increasingly automated systems.