Litigation as a "win-win" game: A Theoretical Perspective and Empirical Evidence from Brazilian Labor Courts

Employers in Brazilian labor courts are ordered to pay something in roughly nine of every ten adjudicated claims, and a composition is available at any time at a fraction of what a judgment costs them. Yet large employers decline to compose, claim after claim, for years. We show that this is the equilibrium of a repeated game in which a long-lived defendant faces a continuing flow of short-lived claimants who differ in the personal cost of carrying a claim. The defendant is not buying a cheaper resolution of the present claim; it is buying a public record of refusing to compose, which lengthens the expected duration of any claim brought against it and thins the flow of claims it will face in future. The mechanism requires neither divergent beliefs about the merits nor private information. Three results carry the argument. A reputation for toughness deters entry if and only if the ratio of the two values at stake is smaller than the ratio of the two delays, each measured as a discounted annuity, a condition that is automatic when compositions price at the claimant’s reservation value. Because claimants differ, deterrence is always partial and the filing rate is bounded away from zero at every reputation the defendant can attain, so the stationary outcome is persistent litigation rather than its disappearance. And permanent toughness requires the defendant’s patience to exceed a threshold that falls with the responsiveness of entry and with the speed at which the market’s record forms. We take the model to approximately 130,000 decisions of Brazil’s largest regional labor court (TRT-2, São Paulo) for 2001–2017, obtained by text mining. The corpus establishes the institutional configuration the model assumes: court-fee waivers granted essentially automatically and held by 84.7% of claimants, no exposure of the losing claimant to the employer’s attorney fees, and contingent representation, which together produce a lottery with no losing ticket and guarantee participation at any degree of risk aversion. It also establishes the constraint that forces our definition of reputation, since an employer that disputes everything still loses most of what it disputes and cannot build a record of victories. We do not report a calibration: composition rates are not observable in a corpus of decisions, and we set out what would be needed to measure the model’s remaining objects.