Competition Forecasting Using AI Models

1. Introduction

Competition forecasting using AI models means using artificial intelligence, machine learning, statistical learning, predictive analytics, and related computational techniques to anticipate future competitive conditions in a market.

In competition law, AI forecasting may be used to predict:

future market shares;

entry or exit of competitors;

price movements;

demand changes;

merger effects;

consumer switching;

potential foreclosure;

innovation effects;

network effects;

supply constraints;

coordinated behaviour;

effects of platform rules;

likely competitive impact of new technologies.

The concept is not itself a separate competition-law offence or established legal doctrine. Rather, it is a modern analytical technique that can assist competition authorities, courts, economists, companies and compliance teams.

The central legal question remains:

Can the AI model provide sufficiently reliable, transparent and relevant evidence for the particular competition-law inquiry?

2. Meaning of Competition Forecasting

Traditional competition analysis often examines historical information.

For example:

Company A had 45% market share during the previous three years.

AI forecasting attempts to answer:

What is likely to happen during the next three years if Company A acquires Company B?

The model may process:

historical prices;

transaction data;

customer behaviour;

product characteristics;

market shares;

search activity;

switching patterns;

entry data;

supply-chain information;

economic indicators;

competitor behaviour.

It can then produce a forecast.

3. Examples

Example 1 — Merger

AI predicts:

The proposed merger may increase prices by 8%.

The authority must then examine whether the model is:

appropriately specified;

based on reliable data;

sensitive to assumptions;

consistent with economic theory.

Example 2 — Platform Competition

An AI model predicts:

Removing third-party access to an API will cause 30% of users to remain within the dominant platform.

The competition analysis may examine:

switching costs;

network effects;

interoperability;

foreclosure;

consumer effects.

Example 3 — Entry Forecast

AI predicts:

A new competitor will enter within 18 months.

The authority must determine whether that prediction is actually supported by:

investment requirements;

regulatory barriers;

technology;

distribution;

customer switching;

historical entry patterns.

4. Main Types of AI Competition Forecasting

A. Price Forecasting

Predicts:

future prices;

price responses;

price elasticity;

promotional effects.

B. Demand Forecasting

Predicts:

consumer demand;

substitution;

switching;

product adoption.

C. Market-Share Forecasting

Predicts:

future market shares;

competitor growth;

market concentration.

D. Merger Simulation

Models the likely consequences of a merger.

Possible outputs:

price increases;

output reductions;

market-share changes;

innovation effects.

E. Entry/Exit Forecasting

Predicts:

likelihood of new entry;

competitor exit;

expansion.

F. Foreclosure Forecasting

Examines whether conduct could make it harder for rivals to:

access customers;

obtain inputs;

reach suppliers;

access data;

achieve sufficient scale.

G. Coordination-Risk Forecasting

AI can identify patterns potentially consistent with:

parallel pricing;

information exchange;

algorithmic coordination.

But:

Prediction of coordination is not proof of an unlawful agreement.

5. Legal Framework

AI forecasting can become relevant to several areas.

Article 101 TFEU

Agreements and concerted practices.

Article 102 TFEU

Abuse of dominance.

EU Merger Regulation

Prospective competitive effects of mergers.

US Sherman Act §1

Agreements and concerted action.

US Sherman Act §2

Monopolization and attempted monopolization.

US Clayton Act §7

Mergers and acquisitions that may substantially lessen competition.

6. Why AI Forecasting Is Attractive

AI can process very large datasets.

Traditional human analysis may struggle with:

millions of transactions;

high-frequency prices;

millions of consumer observations;

complex network effects;

rapidly changing digital markets.

AI can identify:

patterns;

correlations;

clusters;

anomalies;

nonlinear relationships.

This can make competition analysis more sophisticated.

7. But Prediction Is Not Proof

This is the most important legal principle.

Suppose an AI model predicts:

“Prices will increase by 15%.”

That does not automatically prove:

“The merger violates competition law.”

The legal decision requires additional questions:

What is the relevant market?

Does the transaction create or strengthen market power?

What competitive effects are legally relevant?

What efficiencies exist?

What alternatives exist?

How reliable is the forecast?

What is the counterfactual?

Therefore:

AI prediction = evidence/analytical input, not automatic legal conclusion.

8. Case Law 1 — United States v Philadelphia National Bank

374 U.S. 321 (1963)

Facts

The case concerned a proposed bank merger.

Principle

The Supreme Court emphasized the importance of market concentration in merger analysis.

The Court developed a structural approach in which concentration could provide important evidence regarding competitive effects.

Relevance to AI Forecasting

AI can improve structural analysis by forecasting:

future market shares;

concentration;

entry;

competitive constraints.

But historical market-share data and AI forecasts should not be confused.

Lesson

Forecasted concentration can supplement, but not replace, legal merger analysis.

9. Case Law 2 — United States v General Dynamics Corp.

415 U.S. 486 (1974)

Facts

The government challenged a merger involving coal companies.

The defendants argued that historical market shares did not accurately reflect future competitive conditions because existing coal reserves and other factors were important.

Principle

The Supreme Court recognized that historical market shares may not always accurately predict future competitive strength.

Importance for AI

This is highly relevant to competition forecasting.

AI may help identify:

future capacity;

resource constraints;

technological change;

likely entry;

future competitive conditions.

Lesson

Competition analysis may need to look forward, not merely backward.

10. Case Law 3 — Brown Shoe Co. v United States

370 U.S. 294 (1962)

Facts

The case concerned a merger in the shoe industry.

Principle

The Supreme Court examined:

market structure;

concentration;

vertical relationships;

potential competition.

The case is important for understanding how merger analysis considers the structure of competition.

AI relevance

An AI model could help identify:

changing market boundaries;

customer substitution;

supply relationships;

competitive overlaps.

But the court must still determine whether the model corresponds with the legally relevant market.

11. Case Law 4 — FTC v Staples, Inc.

970 F. Supp. 1066 (D.D.C. 1997)

Facts

The FTC challenged the proposed merger of Staples and Office Depot.

The case involved extensive economic analysis of pricing relationships among office-supply retailers.

Principle

The court relied significantly on empirical economic evidence to assess likely competitive effects.

The analysis examined whether competition between the merging firms constrained prices.

Importance

This is an excellent example of forward-looking competition analysis based on empirical evidence.

AI relevance

AI could potentially extend this type of analysis by processing:

transaction-level pricing;

geographic data;

customer switching;

local market effects;

product substitution.

But the model still must be validated and economically justified.

12. Case Law 5 — FTC v H.J. Heinz Co.

246 F.3d 708 (D.C. Cir. 2001)

Facts

The FTC challenged the Heinz/Beech-Nut baby-food merger.

Principle

The court considered market concentration and the potential effects of the merger on competition.

The case illustrates the importance of evaluating likely future competitive effects rather than waiting until harm has actually occurred.

AI relevance

Competition forecasting is inherently prospective.

Authorities may use models to estimate:

price effects;

product variety;

innovation;

entry;

competitive constraints.

Lesson

Competition law frequently asks:

What is likely to happen if the proposed conduct occurs?

AI can assist with that predictive question.

13. Case Law 6 — United States v Microsoft

253 F.3d 34 (D.C. Cir. 2001)

Facts

Microsoft held monopoly power in PC operating systems.

The government challenged several practices directed against competing technologies.

Principle

The court examined whether Microsoft had used exclusionary conduct to maintain its monopoly.

Importantly, the court considered the likely effects of Microsoft's conduct on emerging technologies and competition.

AI relevance

Modern AI models can attempt to forecast:

whether a competing technology would gain adoption;

whether interoperability restrictions would reduce adoption;

whether network effects strengthen incumbent power.

Lesson

Forecasting can be particularly important in technology markets where future competition depends on innovation and network effects.

14. Case Law 7 — Intel v Commission

C-413/14 P

Facts

Intel's rebate arrangements were assessed under Article 102 TFEU.

Principle

The Court of Justice emphasized the relevance of economic factors in evaluating potentially exclusionary rebates.

These include:

market coverage;

duration;

rebate amount;

competitors' position;

exclusionary capability;

as-efficient-competitor analysis where appropriate.

AI relevance

AI models can estimate:

customer switching;

competitor foreclosure;

price effects;

rebate coverage;

potential competitor viability.

But the model must correspond to the legal/economic test actually being applied.

Lesson

Advanced analytics can strengthen effects analysis, but it cannot replace the applicable legal standard.

15. Case Law 8 — Google Shopping

Google and Alphabet v Commission, C-48/22 P

Facts

Google's treatment of its own comparison-shopping service was challenged under Article 102.

Principle

The EU courts upheld the core finding concerning Google's conduct.

The case involved analysis of how a dominant search platform could use its position to disadvantage competing comparison-shopping services.

AI relevance

Modern competition authorities may use AI to examine:

search ranking;

click-through rates;

traffic allocation;

visibility;

consumer switching;

competitor traffic.

Lesson

In platform markets:

AI can help quantify competitive effects, but ranking data must be interpreted within the legal framework of Article 102.

16. Case Law 9 — Ohio v American Express

585 U.S. 529 (2018)

Facts

The case concerned rules imposed by American Express in a two-sided payment-card market.

Principle

The Supreme Court emphasized the importance of considering both sides of certain transaction platforms when assessing competitive effects.

AI relevance

AI models are particularly useful for multi-sided markets because they can model interactions between:

consumers;

merchants;

advertisers;

sellers;

developers.

Lesson

A forecasting model that examines only one side of a multi-sided platform may produce a misleading result.

17. Case Law 10 — Brooke Group v Brown & Williamson

509 U.S. 209 (1993)

Facts

The case concerned alleged predatory pricing.

Principle

The Supreme Court required proof involving:

pricing below an appropriate cost measure; and

a reasonable prospect of recoupment.

AI relevance

AI can forecast:

future prices;

competitor exit;

likely recoupment;

demand responses.

But the model must still satisfy the applicable legal requirements.

Lesson

Predictive economics must be connected to the governing legal test.

18. AI Forecasting in Merger Control

One of the most important applications is merger review.

A model can simulate:

Before merger

Firm A = 30%

Firm B = 20%

Firm C = 20%

Firm D = 15%

Others = 15%

Forecast after merger

A+B = 50%

The model can then estimate:

price effects;

output;

innovation;

entry;

consumer switching.

But market share alone does not establish the legal outcome.

19. Counterfactual Analysis

Competition forecasting requires a counterfactual.

The basic question is:

What would happen without the challenged conduct?

For a merger:

Scenario A

Merger occurs.

Scenario B

Merger does not occur.

The difference between A and B is the estimated competitive effect.

AI can generate sophisticated counterfactual models.

But the counterfactual itself depends upon assumptions.

20. The Problem of Model Bias

AI models may reproduce biases present in their data.

Suppose historical data shows:

Customers rarely switch from Platform A.

AI may predict:

Future customers will also rarely switch.

But perhaps customers did not switch historically because Platform A imposed high switching costs.

The AI could therefore transform:

historical exclusion → predicted future behaviour.

That would be problematic if the purpose of the analysis is to determine whether the conduct itself creates the switching barrier.

21. Training Data Problem

A model trained on historical competition conditions may be unreliable when:

technology changes;

new entrants emerge;

consumer preferences change;

regulation changes;

business models change.

This is known as a form of distribution shift.

Example

A model trained on smartphone competition from 2015–2020 may poorly predict competition after the introduction of new AI assistants.

22. Explainability

A competition authority or court may ask:

Why did the AI predict a 20% price increase?

If the model cannot provide a meaningful explanation, its evidentiary usefulness may be reduced.

Relevant information may include:

important variables;

assumptions;

model architecture;

validation;

confidence intervals;

sensitivity analysis.

23. Confidence Intervals

AI forecasting should ideally communicate uncertainty.

Instead of:

“Prices will increase by 10%.”

A better analytical statement might be:

“The central estimate is a 10% increase, with a specified confidence interval.”

This helps decision-makers understand:

uncertainty;

model error;

range of plausible outcomes.

24. Sensitivity Analysis

Suppose an AI model predicts:

12% price increase.

Change assumptions:

Scenario 1

10–14%

Scenario 2

5–8%

Scenario 3

1–3%

If the result changes dramatically, the court or authority should understand why.

25. Adversarial Testing

Competition authorities can test an AI model by asking:

What happens if the assumptions are deliberately changed?

For example:

higher entry probability;

lower switching costs;

greater customer elasticity;

faster innovation;

alternative market definition.

This can reveal whether the model is robust or fragile.

26. Human Oversight

AI should generally function as:

Decision-support technology

rather than:

Decision-maker.

A human competition expert should examine:

legal relevance;

assumptions;

data;

methodology;

model limitations;

alternative explanations.

27. AI Forecasting and Article 102

For dominance cases, AI may forecast:

Foreclosure

Will competitors lose sufficient access to customers?

Switching

Will consumers remain locked into the dominant ecosystem?

Pricing

Will the conduct increase prices?

Innovation

Will competitors reduce investment?

Network effects

Will the conduct strengthen the incumbent's network?

The compliance officer or authority should then connect those forecasts to the actual Article 102 legal analysis.

28. AI Forecasting and Article 101

Under Article 101, AI can assist in detecting:

unusual pricing patterns;

communication patterns;

coordinated bidding;

market allocation;

suspicious information exchange.

But:

Parallel behaviour is not automatically a cartel.

A predictive model identifying parallel pricing cannot by itself establish an agreement or concerted practice.

Human and documentary evidence may remain necessary.

29. AI Forecasting and US Antitrust Law

AI can assist in:

Section 1

Detecting possible coordination.

Section 2

Forecasting exclusionary effects.

Section 7

Predicting merger effects.

FTC Act

Analysing potentially unfair competitive conduct depending upon the legal theory.

Again, the model is an analytical tool, not the legal test itself.

30. Risks of AI-Based Competition Forecasting

1. Automation bias

Decision-makers may trust the AI simply because it appears objective.

2. Historical bias

Past market conditions may reproduce past exclusion.

3. Overfitting

A model may perform well on historical data but badly on new circumstances.

4. Underfitting

A model may be too simple to capture important competitive dynamics.

5. Data leakage

Information unavailable at the relevant historical point may accidentally enter the model.

6. Black-box predictions

The model may not provide adequate explanations.

7. False precision

Numerical outputs may appear more certain than the evidence justifies.

31. Judicial Evaluation Criteria

When a court receives competing AI forecasts, it should examine:

1. Relevance

Does the model answer the legal question?

2. Data quality

Is the dataset reliable?

3. Methodology

Is the model appropriate?

4. Validation

Was it tested?

5. Error rate

How often does it produce incorrect forecasts?

6. Robustness

Does the result survive alternative assumptions?

7. Transparency

Can the methodology be examined?

8. Reproducibility

Can another expert test it?

9. Causation

Does it establish causation rather than correlation?

10. Legal fit

Does the model correspond to the governing competition-law test?

32. Competition Forecasting Governance Framework

A large competition authority or company can use:

Raw Data   ↓ Data Validation   ↓ Market Definition   ↓ Economic/AI Model   ↓ Validation & Testing   ↓ Alternative Models   ↓ Sensitivity Analysis   ↓ Human Expert Review   ↓ Legal Analysis   ↓ Competition Decision   ↓ Continuous Monitoring

33. Competing AI Models

Suppose:

Model A

Predicts:

Merger causes 15% price increase.

Model B

Predicts:

Merger causes 3% price increase.

The authority should ask:

The court should not select the model merely because its prediction is more alarming or more favourable to one party.

The proper question is:

Which methodology is better supported for the legal and economic question being decided?

34. AI Forecasting and Precaution

Competition law sometimes operates before harm has actually occurred.

This makes forecasting important.

For example:

“If this acquisition proceeds, an emerging competitor may disappear.”

The authority cannot simply wait five years to observe the result.

However, predictive enforcement creates a corresponding danger:

An uncertain prediction should not automatically be treated as established fact.

Therefore, the quality of the prediction becomes critical.

35. AI and Potential Competition

Potential competition analysis may consider whether an undertaking is:

likely to enter;

capable of entering;

incentivized to enter.

AI may help estimate these factors.

But the authority must distinguish:

Genuine potential competitor

from

Hypothetical future competitor.

A model predicting entry is not itself proof that entry would actually occur.

36. AI and Innovation Competition

Traditional models often measure:

prices;

output;

market shares.

AI forecasting can potentially model:

R&D;

patents;

product launches;

technology adoption;

innovation races.

This is especially relevant in:

pharmaceuticals;

semiconductors;

AI;

cloud computing;

digital platforms.

37. AI and Network Effects

In digital markets:

Users → Data → Better service → More users → More data

creates a feedback loop.

AI can forecast how quickly this loop may operate.

But the analysis should distinguish:

Natural network effects

from

Artificially reinforced network effects created through exclusionary conduct.

Network effects themselves are not automatically unlawful.

38. AI and Algorithmic Collusion

AI pricing systems can potentially react rapidly to competitors' prices.

This creates a theoretical risk of:

tacit coordination;

algorithmic price alignment;

reduced price competition.

But competition law normally requires careful distinction between:

independent parallel conduct

and

coordination/agreement/concerted practice.

AI evidence may identify suspicious patterns, but additional legal analysis is necessary.

39. Compliance Officer's Duties Regarding AI Forecasting

A competition compliance officer should ensure:

Before deployment

legal purpose identified;

data source reviewed;

competition risks assessed.

During deployment

outputs monitored;

anomalies investigated;

model drift tested.

After deployment

periodic validation;

independent audit;

documentation;

human review.

40. Six Core Principles

Remember:

F-D-M-V-R-H

F — Fit the model to the legal question

D — Validate the data

M — Examine methodology

V — Validate predictions

R — Test robustness

H — Human oversight

41. Key Case-Law Lessons

42. Important Legal Distinctions

AI forecast ≠ legal conclusion

A model cannot itself determine infringement.

Prediction ≠ fact

A forecast represents an estimated future outcome.

Correlation ≠ causation

AI may detect patterns without establishing causal relationships.

Accuracy ≠ legal relevance

A highly accurate commercial forecast may still be irrelevant to the competition-law question.

Complexity ≠ reliability

A complicated neural network is not automatically superior to a transparent statistical model.

Historical data ≠ future reality

Technological and market conditions can change.

Algorithmic coordination ≠ automatically cartel

Parallel algorithmic behaviour does not by itself establish unlawful agreement.

43. Conclusion

Competition forecasting using AI models can significantly enhance modern antitrust analysis by allowing authorities, courts, companies and experts to process enormous quantities of economic and behavioural information.

Its strongest applications include:

merger simulation;

demand forecasting;

market-power analysis;

entry prediction;

foreclosure analysis;

price-effect analysis;

innovation forecasting;

platform competition;

algorithmic monitoring.

However, AI forecasting must remain subject to human economic and legal evaluation.

The central principle is:

AI may forecast competitive outcomes, but the legal decision must rest on a reliable methodology, appropriate data, transparent assumptions, validated results, consideration of uncertainty, and application of the governing competition-law standard.

The cases of General Dynamics, Staples, Heinz, Microsoft, Intel, Google Shopping, American Express and Brooke Group demonstrate that competition law already relies substantially on forward-looking economic reasoning. AI represents an evolution of that analytical process rather than a replacement for it.

Ultra-short revision formula

AI Data → Model → Forecast → Validation → Sensitivity → Human Review → Legal Test → Competition Decision

Keywords

AI Competition Forecasting – Predictive Analytics – Machine Learning – Merger Simulation – Market Power – Demand Forecasting – Price Effects – Counterfactual – Entry Prediction – Foreclosure – Network Effects – Algorithmic Coordination – Article 101 – Article 102 – Sherman Act – Clayton Act – Data Quality – Model Validation – Explainability – Robustness – Causation – Human Oversight – Digital Markets.