← All labs
Competition Law · LawLab

Cartel Room: Leniency and the Prisoner's Dilemma

Sit inside a fictional cement cartel, then take the regulator's chair. Why does a promise of lower penalties make cartels fall apart?
Quarter1/10
Profit ₹cr0
Best net-
Narmada Cement Ltd · Compliance head

Quarter 1

Last month the managing directors of four cement makers in the western region met at a hotel and agreed to hold prices at ₹420 a bag. You are the compliance head of Narmada Cement. Each quarter you advise the board on one move. Your three rivals each follow a fixed strategy, which you must infer from their behaviour.

The cartel is in place. The CCI has no file on the industry yet.

Investigation heat

Heat 0Dawn raid risk this quarter: 0%

Tip: rivals see price movements with a one-quarter lag. A leniency application ends the cartel at once and opens a case against every member.

The other three firms

Your payoff this quarter (₹ crore)

Other firms colludingYou colludeYou cheat
All 31016
2 of 3712
1 of 358
None34

Whatever the others do, cheating pays more this quarter. If everyone reasons that way, all earn 4 instead of 10: the prisoner's dilemma. A quarter in which anyone applies for leniency is competitive for all (4 each).

Quarter log

QYouRivalsYour profitHeatEvent
No quarters played yet.

Regulator's sandbox

Set the policy levers, then run 200 simulated cartels. Each run is repeated with leniency switched off so you can see what the programme changes.

Outcomes of 200 cartels

Survival: share of formed cartels still operating

With leniencyWithout leniency (dashed)

Policy levers

If on, firms race to apply after detection too, which cuts the expected fine.

Incentive-compatibility check

1 - 1/Nδ* without leniencyδ* with leniencyyour δ

In the repeated game each firm weighs a one-off gain from undercutting (taking the whole market for a period) against its share of the cartel profit in every future period. Collusion is self-enforcing only when the discount factor δ is at least a threshold δ*; with no enforcement δ* = 1 - 1/N. Leniency raises the value of leaving, because the firm that undercuts and reports first also escapes the fine, so δ* rises and fewer cartels can hold together.

Model assumptions
  • N identical firms; monopoly profit per period is 1, shared equally (π = 1/N) under collusion. Bertrand competition after collapse gives 0.
  • A deviating firm takes the whole market for one period, scaled by a private shock (1 + ε), ε uniform on ±0.15, drawn per firm per period. Each cartel draws firm discount factors δi = δ ± 0.03. These two sources of noise are what make cartels fail at different times.
  • Punishment is grim trigger: after any deviation the market is competitive forever.
  • Each period the cartel is detected with probability p. Every member then pays a fine K = F × π (F times one period's cartel profit). Real s.27(b) penalties grow with each year of the cartel; a stationary fine keeps the threshold in closed form.
  • Value of staying: VC = (π - pKe) / (1 - δ(1 - p)), where Ke is the expected fine after detection (K, or less if post-raid leniency is on).
  • Value of leaving: max of (1 + ε - pKe) for a plain cheat, and (1 + ε - K(1 - L1)) for cheat-and-report, the second only when leniency exists. The reporter is first in line; others then race for the second and third markers.
  • A cartel forms only if every firm prefers staying at ε = 0 and VC > 0. It ends by leniency (deviation plus report), collapse (deviation without report), detection, or survives to the 40-period horizon.
  • Post-raid leniency (toggle): every member races after detection and positions 1 to 3 get L1, L2, L3 in random order, so Ke = K(1 - (L1+L2+L3)/N). This lowers expected fines, a known effect in the economics literature (Motta and Polo) that can make generous leniency pro-collusive.
  • The model ignores reputation, communication, asymmetric firms, compliance programmes and criminal sanctions. It is a teaching model, not an estimate of Indian enforcement.
δ* = [1 - (π - pKe) / Vout] / (1 - p)

The law behind the game

Prohibition. Section 3(1) of the Competition Act, 2002 bars agreements that cause or are likely to cause an appreciable adverse effect on competition (AAEC) in India. Under s.3(3), an agreement between competitors that directly or indirectly fixes prices, limits output, shares markets or rigs bids is presumed to have an AAEC. A cartel of cement makers holding a common price falls squarely within s.3(3). Cheating on the cartel does not take a firm outside it; the firm remains a party to the agreement for the period it was in.

Penalty. Under s.27(b), the CCI may impose a penalty of up to 10% of the average turnover of the preceding three financial years. For cartels, the proviso allows a penalty on each member of up to three times its profit for each year of continuance of the cartel, or 10% of its turnover for each year of continuance, whichever is higher. These are ceilings; the CCI fixes the actual amount after considering aggravating and mitigating factors. This game assumes the ceiling is imposed so the stakes are visible.

Leniency. Section 46 lets the CCI impose a lesser penalty on a cartel member who makes a full, true and vital disclosure. The Competition Commission of India (Lesser Penalty) Regulations set priority by marker order: the first applicant may receive a reduction of up to 100%, the second up to 50%, and the third up to 30%. Each is an "up to" ceiling; the actual reduction depends on the timing of the application, the value its evidence adds and continuous cooperation. The game grants each ceiling in full for simplicity.

Lesser penalty plus. The Competition (Amendment) Act, 2023 introduced "lesser penalty plus": an applicant in one cartel investigation who discloses a different, previously undisclosed cartel may receive an additional reduction in the first case, besides leniency in the second.

Why leniency destabilises cartels. A cartel is a repeated prisoner's dilemma held together by the threat of future punishment. Leniency changes the pay-offs: the first firm through the regulator's door can escape most or all of the fine, while those who wait bear the full penalty. Each member now fears that another will report first, and the fear itself is enough to trigger a report. Distrust becomes self-fulfilling: a race to the regulator.

Teacher notes

Learning outcomes

  • Explain the cartel as a repeated prisoner's dilemma and identify the incentive to cheat.
  • State the s.3(3) presumption and the cartel penalty ceiling under the proviso to s.27(b).
  • Describe the marker system under s.46 and the Lesser Penalty Regulations, and why priority matters.
  • Use the incentive-compatibility condition (δ ≥ δ*) to predict how detection, fines and leniency change cartel stability.
  • Evaluate the design risk that generous post-investigation leniency lowers deterrence.

45-minute run-sheet

  • 0-5 min: Pose the question: "If you ran a cartel, whom would you fear most?" Collect answers.
  • 5-20 min: Students play Mode 1 in pairs, two rounds each. Ask them to guess each rival's strategy before the debrief reveals it.
  • 20-27 min: Board the net scores. Who applied first, and when? Draw the payoff table and locate the dilemma.
  • 27-40 min: Mode 2 on the projector. Start at the defaults, then (a) set leniency to zero, (b) raise detection, (c) increase N, (d) switch on post-raid leniency. Ask for a prediction before each run.
  • 40-45 min: Exit ticket: one paragraph on why a marker system with falling reductions (100/50/30) is better than a flat reduction for all applicants.

Discussion questions

  • Why does the law give the largest reduction to the first applicant rather than the one with the best evidence?
  • The profit-based ceiling in s.27(b) is three times profit per year. Why might a turnover-based cap alone under-deter a high-margin cartel?
  • Should a ringleader or coercer be eligible for leniency? What are the risks either way?
  • How might "lesser penalty plus" help the CCI detect cartels in adjacent markets?
  • In the sandbox, when does more generous leniency make collusion easier? What does this suggest about post-raid applications?

Teaching simulation with fictional firms, not legal advice. Figures are stylised. Built by Anshuman Sahoo, Law Matters Academy.