Goodhart’s Law in Competition Law Enforcement Systems

Introduction

Goodhart’s Law is commonly expressed as: “When a measure becomes a target, it ceases to be a good measure.” Although originating in economics and monetary-policy analysis, the principle has significant implications for competition-law enforcement.

Competition authorities necessarily use measurable indicators—market shares, prices, margins, diversion ratios, HHI, foreclosure rates, complaint numbers, investigation duration, settlement rates, fines, leniency applications, merger clearance rates and consumer-welfare effects—to decide where enforcement resources should be directed.

The problem arises when these indicators cease to be diagnostic tools and become institutional targets. Firms may then adapt their behaviour to the metric rather than to genuine competitive conditions, while enforcement agencies may optimize their performance around measurable outputs rather than the underlying objective of protecting competition.

Goodhart’s Law therefore presents a particular challenge in algorithmic, data-driven and AI-assisted competition enforcement, where increasingly sophisticated systems can transform legal standards into numerical proxies.

1. Meaning of Goodhart’s Law

The central idea is:

A metric that works as an indicator under normal conditions can become unreliable once participants have incentives to optimize specifically against that metric.

For example, suppose an authority informally evaluates enforcement effectiveness by the number of infringement decisions obtained each year.

Initially:

More successful infringement decisions → potentially stronger enforcement

But if decision numbers become an institutional performance target:

Target → pressure to increase case numbers → selection of easier cases → reduced attention to complex cases

The numerical indicator then becomes detached from the actual objective.

The same problem can occur with:

• market-share thresholds;
• HHI thresholds;
• price effects;
• consumer-surplus estimates;
• merger clearance rates;
• fine levels;
• number of investigations;
• investigation speed;
• settlement rates;
• number of dawn raids;
• number of leniency applications;
• number of complaints resolved;
• algorithmic risk scores.

2. Why Goodhart’s Law Matters to Competition Law

Competition law does not directly observe “competition” as a single measurable object.

Authorities instead use proxies.

Simplified enforcement model

Competition

↓

Observable indicators

↓

Market share
Price
Output
Margins
Entry
Innovation
Quality
Consumer switching
Capacity
Data access
Network effects

↓

Legal/economic assessment

↓

Enforcement decision

The difficulty is that the observable indicators are not identical to competition itself.

A firm can have:

• high market share without exercising substantial market power;
• low prices because of aggressive competition;
• high prices because of temporary cost shocks;
• large margins because of innovation;
• low margins because of investment;
• rapidly increasing market share because it is more efficient.

Consequently, converting competition into a single numerical target can create serious enforcement errors.

3. Goodhart’s Law and Market-Share Thresholds

Market share is one of the most important competition-law indicators.

However:

Market share ≠ market power.

A 60% market share may be concerning in one market but relatively harmless in another.

Relevant factors include:

• barriers to entry;
• countervailing buyer power;
• switching costs;
• network effects;
• innovation;
• potential competition;
• durability of market share;
• multi-homing;
• vertical integration.

If authorities treat a particular market-share percentage as an automatic enforcement trigger, companies may restructure transactions or markets around the threshold.

For example:

49% share → no automatic intervention

51% share → intervention

This creates a regulatory cliff that may have little relationship to actual competitive harm.

4. Goodhart’s Law and HHI

The Herfindahl-Hirschman Index (HHI) is another useful screening instrument.

It can help authorities identify potentially concentrated markets.

But if HHI becomes an enforcement target, parties can potentially structure transactions to remain below a numerical threshold while still increasing strategic control.

For example:

Transaction A + B produces an HHI below the screening threshold.

But the transaction may nevertheless eliminate an important potential competitor, increase data advantages or strengthen network effects.

The metric therefore fails to capture the complete competitive relationship.

5. Goodhart’s Law and Consumer Welfare

Consumer welfare is central to modern competition analysis.

But consumer welfare itself is difficult to reduce to a single number.

A short-term price reduction may appear beneficial while a platform simultaneously:

• reduces product quality;
• restricts interoperability;
• collects more data;
• eliminates rivals;
• increases switching costs;
• reduces innovation;
• creates long-term dependence.

Thus:

Short-term price ↓

does not necessarily mean:

Competition ↑

This is particularly important in digital markets where products may be supplied at a monetary price of zero.

6. Goodhart’s Law in Digital and AI Competition Enforcement

The risk becomes more serious when competition authorities employ automated systems.

An AI enforcement system might assign firms a risk score:

The system produces:

Competition-risk score = 82/100

If the authority automatically investigates firms above 80, the score effectively becomes a regulatory target.

Firms may then optimize their behaviour to reduce the score without reducing the underlying competitive risk.

This is sometimes called gaming the metric.

7. Goodhart’s Law and Algorithmic Collusion

Algorithmic pricing systems create a particularly important problem.

Suppose authorities monitor:

• identical prices;
• parallel price movements;
• price variance;
• frequency of algorithmic adjustments.

If these become the primary indicators of cartel risk, sophisticated firms may design algorithms that avoid obvious parallel pricing while maintaining coordination through other mechanisms.

Conversely, legitimate algorithms may produce parallel prices because they respond to the same market conditions.

Therefore:

Parallel pricing ≠ necessarily collusion

and

absence of obvious parallel pricing ≠ necessarily absence of coordination.

Goodhart’s Law warns against treating the observable signal as equivalent to the legal phenomenon.

8. Goodhart’s Law and Merger Enforcement

Merger control is particularly susceptible to metric-based decision-making.

Authorities may examine:

• market share;
• concentration;
• HHI;
• unilateral effects;
• diversion ratios;
• price-pressure indices;
• efficiencies.

These indicators are valuable but incomplete.

A transaction involving two firms with relatively modest shares could nevertheless eliminate:

• a disruptive innovator;
• a future competitor;
• an important source of technology;
• a potential entrant;
• a firm possessing strategically important data.

This is particularly relevant to nascent competition and technology acquisitions.

9. Goodhart’s Law and Enforcement Performance Metrics

Competition authorities themselves can become subject to Goodhart effects.

Suppose an agency evaluates performance using:

Metric A

Number of investigations.

Then:

More investigations = better performance

may incentivize excessive investigation.

Metric B

Number of infringement decisions.

This may incentivize easier cases rather than economically important ones.

Metric C

Amount of fines.

This may encourage large penalties even where behavioural remedies could better restore competition.

Metric D

Speed of investigations.

This may create pressure to conclude complex cases prematurely.

Metric E

Settlement rate.

This may incentivize settlements even where a fully reasoned infringement decision would produce greater legal clarity.

The institutional objective should instead be:

Effective restoration and protection of competitive conditions.

10. Goodhart’s Law and Case Selection

A sophisticated enforcement agency should distinguish:

Case-count optimization

from

harm-prevention optimization.

A single large digital-platform investigation may require years and enormous resources but have substantially greater competitive importance than dozens of small cases.

Therefore:

10 minor cases ≠ necessarily better enforcement than 1 systemic case.

Goodhart’s Law therefore supports using performance indicators as diagnostic information, not as rigid institutional objectives.

11. Goodhart’s Law and Fines

Fine levels can also become problematic metrics.

If an authority evaluates deterrence primarily by the amount of fines imposed, the system may become focused on punishment rather than restoration.

Competition law generally has several possible objectives:

• deterrence;
• compensation or consumer redress where applicable;
• cessation of unlawful conduct;
• restoration of competitive conditions;
• prevention of recurrence;
• institutional deterrence;
• protection of market access.

A €1 billion fine is not automatically more effective than a €100 million fine accompanied by a carefully designed structural or behavioural remedy.

12. Goodhart’s Law and Leniency

Leniency applications can also become a proxy for cartel-enforcement effectiveness.

Initially:

More leniency applications → potentially greater cartel detection

But if firms learn how the programme operates, the metric may become distorted.

For example, companies may file precautionary applications because they fear that competitors will apply first, even when the underlying evidence is weak.

Thus:

Number of leniency applications ≠ number of genuine cartels detected.

The quality of information remains more important than the numerical count.

13. Goodhart’s Law and Complaints

Competition authorities may use the number of consumer or competitor complaints as a prioritization mechanism.

But complaints can be strategically generated.

Competitors may submit large numbers of complaints against rivals for tactical reasons.

Conversely, serious exclusionary conduct may generate relatively few complaints because:

• victims are dependent on the dominant firm;
• customers fear retaliation;
• consumers do not recognize the harm;
• the market is technically complex;
• the conduct is hidden.

Therefore:

High complaint volume ≠ high competitive harm.

Low complaint volume ≠ low competitive harm.

14. Goodhart’s Law and Regulatory Gaming

The core risk can be represented as:

Legal standard

↓

Proxy

↓

Proxy becomes target

↓

Firms optimize around proxy

↓

Proxy loses informational value

↓

Enforcement error

This creates a feedback loop.

In digital markets the problem may become even more sophisticated because firms can use AI to predict the authority's enforcement model and optimize their conduct accordingly.

15. Six Important Case Laws

1. United States v. Philadelphia National Bank (1963)

The U.S. Supreme Court adopted a relatively strong structural presumption in bank mergers involving substantial concentration.

Relevance to Goodhart’s Law

The case demonstrates the usefulness—and potential danger—of relying on concentration measures.

Market concentration can be an important warning signal, but concentration alone does not perfectly describe competitive conditions.

Goodhart lesson:
A structural indicator is useful for screening, but should not automatically become the complete definition of competitive harm.

2. United States v. General Dynamics Corp. (1986)

The Supreme Court rejected an overly mechanical reliance on historical market-share data in merger analysis.

General Dynamics involved structural data that did not adequately reflect the competitive significance of the firms' future positions.

Goodhart significance

Historical market shares can become misleading when market conditions are changing.

The case illustrates the importance of considering:

• future competitive capacity;
• changing market conditions;
• actual competitive constraints.

Goodhart lesson:
A historical metric can cease to be a reliable proxy when the underlying economic environment changes.

3. FTC v. Heinz Co. (D.C. Cir. 2001)

The court examined the competitive effects of a proposed merger in the baby-food market, including concentration and the possibility of unilateral effects.

Goodhart significance

The case illustrates how concentration measures can provide an initial indication of competitive danger, but the legal assessment ultimately requires consideration of the competitive relationship between the merging firms.

Goodhart lesson:
The numerical concentration measure should support—not replace—the competitive-effects inquiry.

4. United States v. Baker Hughes Inc. (D.C. Cir. 1990)

The court emphasized the importance of considering the totality of competitive circumstances rather than treating structural presumptions as determinative.

The case is particularly important for understanding the relationship between:

• market concentration;
• presumptions;
• rebuttal evidence;
• competitive conditions.

Goodhart significance

A numerical presumption can be useful as an initial screen, but it cannot capture every relevant market characteristic.

Goodhart lesson:
Metrics should allocate the burden of inquiry rather than eliminate substantive inquiry.

5. United Brands Co. v Commission (Case 27/76)

The European Court of Justice considered whether United Brands occupied a dominant position and examined several factors, including market share and other economic circumstances.

The case is significant because dominance under EU competition law is not established simply by observing one numerical indicator.

Goodhart significance

The assessment required consideration of the broader competitive structure and the firm's ability to behave independently of competitors and customers.

Goodhart lesson:
Market share is evidence of dominance, not necessarily dominance itself.

6. AKZO Chemie BV v Commission (Case C-62/86)

The ECJ considered predatory pricing and developed important price-cost benchmarks for evaluating potentially abusive pricing.

The case demonstrates the usefulness of quantitative thresholds in competition law.

Goodhart significance

Price-cost tests can provide administrable evidence, but economic conduct must still be evaluated within the broader circumstances of the case.

Goodhart lesson:
A numerical threshold is powerful precisely because it simplifies analysis—but simplification can become dangerous if treated as exhaustive.

16. Additional Important Cases

7. Hoffmann-La Roche & Co AG v Commission (Case 85/76)

The Court's analysis of dominance and loyalty-inducing practices demonstrates that competition-law assessment involves the overall economic and commercial context.

Goodhart connection: dominance cannot be reduced to one numerical indicator.

8. Airtours plc v Commission (Case T-342/99)

The General Court's treatment of collective dominance emphasized the need to establish specific structural and behavioural conditions rather than relying on concentration alone.

Goodhart connection: a concentrated market is not automatically a coordinated market.

9. Intel Corp v Commission (Case C-413/14 P)

The later judicial treatment of loyalty rebates emphasized the importance of economic analysis concerning actual or potential foreclosure effects rather than relying solely upon the formal characterization of conduct.

Goodhart connection: a formal category or indicator should not substitute for an assessment of actual competitive effects where effects analysis is legally relevant.

17. Goodhart’s Law and the SSNIP Test

The SSNIP test asks whether a hypothetical monopolist could profitably impose a small but significant and non-transitory increase in price.

It is an important market-definition tool.

However, it becomes difficult where:

• products are supplied for zero monetary price;
• quality is the principal competitive variable;
• data is exchanged for services;
• innovation is rapidly changing;
• switching costs are substantial;
• multi-sided platforms are involved.

If authorities mechanically apply a price-based metric in a zero-price digital market, the test can become detached from actual competitive constraints.

Thus:

Price-based measurement → useful in appropriate markets

but:

Price-based measurement ≠ universal definition of competition.

18. Goodhart’s Law and Digital Platforms

Digital markets create particularly strong conditions for metric manipulation.

Platforms can be evaluated through:

• active users;
• downloads;
• engagement;
• click-through rates;
• advertising prices;
• API access;
• market share;
• time spent;
• switching rates.

Yet these metrics can obscure:

• ecosystem dependence;
• interoperability;
• default positioning;
• data accumulation;
• self-preferencing;
• ecosystem lock-in;
• algorithmic discrimination;
• technical barriers.

For example:

User share = 70%

does not itself explain whether that position results from:

• superior innovation;
• exclusionary conduct;
• network effects;
• contractual restrictions;
• defaults;
• acquisition of competitors.

19. Goodhart’s Law and AI Enforcement

AI systems can make the problem more severe because they encourage authorities to transform complex legal standards into numerical models.

An enforcement AI might calculate:

Probability of antitrust violation = 87%.

But the score may depend upon training data and selected variables.

If firms learn the variables, they can potentially engage in:

Metric evasion

Changing observable behaviour while preserving the underlying economic strategy.

For example:

• slightly altering prices;
• changing contractual language;
• restructuring subsidiaries;
• modifying algorithms;
• adjusting market-share allocations;
• changing reporting categories.

The firm may therefore become:

less detectable without becoming less anticompetitive.

20. Goodhart’s Law and Machine-Learned Enforcement

AI creates a three-level problem.

Level 1 — Measurement

The authority chooses variables.

Level 2 — Prediction

The AI predicts competition risk.

Level 3 — Strategic response

Firms adapt to the prediction system.

This creates an adversarial environment:

Authority model → Firm learns model → Firm changes behaviour → Model becomes less accurate → Authority retrains model → Firm adapts again.

Competition enforcement therefore becomes an ongoing strategic game rather than a one-time measurement exercise.

21. Goodhart’s Law and Regulatory Sandboxing

Regulatory sandboxes can also generate Goodhart problems.

If success is measured by:

• number of participating firms;
• number of innovations;
• number of approvals;
• time to authorization;

participants may optimize these metrics instead of producing genuinely competitive outcomes.

Competition regulators should therefore evaluate:

• market entry;
• durability of competition;
• consumer outcomes;
• innovation;
• contestability;
• switching;
• access to essential inputs.

22. Goodhart’s Law and Remedies

Remedies are particularly vulnerable to metric fixation.

A behavioural remedy may specify:

• maximum contractual duration;
• minimum access;
• response time;
• interoperability levels;
• price caps;
• data-access requirements.

These are measurable.

But firms may technically comply while defeating the remedy's purpose.

For example:

API access = provided

yet:

• documentation is poor;
• latency is excessive;
• access is technically available but economically unusable;
• functionality is restricted;
• security requirements are selectively imposed.

Therefore:

Formal compliance ≠ competitive effectiveness.

This is perhaps one of the most important applications of Goodhart's Law to modern competition remedies.

23. Goodhart’s Law and Essential Facilities

In an essential-facility or access dispute, an authority may monitor:

• number of access requests;
• percentage accepted;
• response times;
• number of refusals.

A platform could technically satisfy all three indicators while making access commercially ineffective.

Therefore the authority should also examine:

• quality;
• interoperability;
• functionality;
• discrimination;
• economic usability;
• actual downstream competition.

24. Goodhart’s Law and Competition-Law Institutions

Goodhart's Law affects three different institutional actors.

A. Competition authorities

Risk:

performance metrics distort enforcement priorities.

B. Regulated firms

Risk:

firms optimize conduct around regulatory indicators.

C. Courts

Risk:

judicial review becomes excessively focused on quantified evidence while underweighting qualitative competitive dynamics.

A resilient enforcement system therefore needs metric pluralism.

25. A Better Enforcement Model

Instead of:

One metric → one decision

competition authorities should employ:

Multi-dimensional assessment

Structural evidence

Behavioural evidence

Economic evidence

Qualitative evidence

Dynamic evidence

Counterfactual analysis

Market-participant evidence

Consumer evidence

=

Competition assessment

This reduces dependence on any single metric.

26. Goodhart-Resistant Competition Enforcement

A good enforcement architecture should follow seven principles.

1. Metrics should be diagnostic, not determinative

A threshold should trigger investigation, not automatically establish liability.

2. Use multiple indicators

No single metric should control the entire decision.

3. Periodically change indicators

If firms learn the enforcement model, indicators can lose predictive value.

4. Test for metric gaming

Authorities should expressly ask:

“Could a rational firm manipulate this indicator without changing the underlying competitive harm?”

5. Preserve qualitative judgment

Economic models should inform legal judgment rather than replace it.

6. Monitor outcomes

Remedies should be evaluated according to whether competition actually improves.

7. Use adversarial testing of AI

AI enforcement models should be stress-tested against firms attempting to evade detection.

27. Goodhart’s Law and the Future of Global Antitrust

The problem will become increasingly significant as competition authorities adopt:

• automated merger screening;
• AI cartel detection;
• algorithmic market monitoring;
• continuous platform monitoring;
• automated risk scoring;
• predictive enforcement;
• digital-market observatories;
• machine-readable compliance systems.

The regulatory environment could evolve into:

Law → metric → algorithm → corporate optimization → metric gaming → regulatory adaptation.

This means future competition law may require a new concept of algorithmic regulatory robustness.

Authorities will need to ask not merely:

“Is this metric accurate?”

but:

“Will this metric remain accurate after firms know that we use it?”

That is the central Goodhart problem.

28. Key Legal Implications

Goodhart's Law has several major implications for competition law:

1. Market share should not automatically equal market power.
2. Concentration indices should not automatically establish competitive harm.
3. Price effects should not be treated as the sole measure of consumer welfare.
4. AI risk scores should not automatically establish infringement.
5. Enforcement statistics should not become crude measures of regulatory success.
6. Formal compliance with remedies should not substitute for outcome-based monitoring.
7. Fines should not become the sole measure of deterrence.
8. Complaint numbers should not determine enforcement priorities automatically.
9. Merger-screening thresholds should remain screening devices rather than substantive legal rules.
10. Competition authorities must anticipate strategic responses to their measurement systems.

Conclusion

Goodhart’s Law exposes a fundamental weakness in metric-driven competition enforcement: the indicator can become the object being optimized.

Competition law inevitably requires measurement, but competition itself is multidimensional and dynamic. Market share, HHI, prices, margins, complaints, fines, investigation numbers and AI-generated risk scores are therefore best treated as evidence and screening mechanisms, not substitutes for the legal and economic inquiry.

The danger is especially acute in digital and AI markets because sophisticated firms can observe, model and strategically respond to regulatory indicators.

The strongest approach is therefore:

Metric → screening → contextual investigation → economic analysis → legal judgment → outcome monitoring

rather than:

Metric → automatic enforcement.

In modern competition law, the ultimate question should not be whether the regulated firm satisfies or violates the metric, but whether the metric continues to provide reliable evidence of the underlying competitive reality.