📈 economics

Honesty, Stigma, and Cooperation in an Overlapping-Generations Game

This paper analyzes a two-period overlapping-generations game to demonstrate how honest agents' behavior and record-based stigma create multiple equilibria in strategic cooperation, revealing that while an interior tipping point exists where higher honesty paradoxically lowers cooperation, this branch is unstable and probabilistic record clearing ultimately eliminates strategic cooperation.

Original authors: David Li, Georgy Lukyanov

Published 2026-07-29

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: David Li, Georgy Lukyanov

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a world where you only get two chances to play a game with a stranger. In the first round, you are the "new kid," and in the second round, you are the "veteran." You never meet the same person twice, and there are no long-term friendships or repeated handshakes. This is the setup for a branch of science called Game Theory, which studies how people make decisions when their success depends on what others do. A key idea here is "reputation": if you are known to be a cheater, people will treat you poorly later. But what happens if your "cheating" is only recorded as a simple, one-word warning label that only the very next person you meet can see? Does having a few "good guys" in the crowd always make everyone cooperate more, or can it sometimes backfire? This paper asks exactly that question, exploring how a tiny bit of memory can keep a society honest, even when everyone is just passing through.

The authors, David Li and Georgy Lukyanov, set up a mathematical playground to test this. They imagine a population of agents who live for exactly two periods: "Young" and "Old." When they are Young, they meet an Old agent. If the Young agent cheats (defects) while the Old agent is being nice (cooperating), the Young agent gets a "stigma"—a permanent black mark on their record. When that agent grows up to be Old, their new partner (a new Young agent) can see this mark. If the mark is there, the new partner will definitely cheat back. If the record is "clear," the new partner might choose to be nice.

The cast of characters is split into two types. First, there are the Honest types. They are like robots programmed with a moral compass: if the opponent is clear, they cooperate; if the opponent has a stigma, they cheat. They don't calculate; they just follow the rule. Second, there are the Strategic types. These are the real-world players. They want to maximize their own points. They have a secret "pain level" (a private loss) that they feel if they cooperate and get cheated on. If this pain is too high, they won't risk being nice. But if the pain is low, they might cooperate when Young just to keep their record clean, hoping to be treated nicely when they become Old.

The big question is: How does the mix of these two types affect the whole system? Intuitively, you might think that having more Honest people is always a good thing. If there are more good guys, surely everyone will be nicer, right? The paper proves that this intuition is only half-right, and the other half is surprisingly tricky.

The researchers found that the system settles into a "cutoff" rule. Strategic agents decide to cooperate only if their secret pain level is below a certain number. If the pain is higher, they cheat. The paper shows that the "clear" record is actually a powerful filter. Because Honest agents never get a stigma (they only cheat against those who already have one), a clear record is a strong signal that an agent is likely Honest. Strategic agents, however, sometimes get stigmatized if they cheat on an Honest person. This means that when a Young agent sees a clear record, they are looking at a crowd that is "purer" than the general population.

Here is where the plot thickens. The paper identifies a specific scenario where the temptation to cheat is very high. In this high-stakes world, the math reveals three possible outcomes:

The "All Bad" Corner: Everyone cheats. No one trusts anyone.
The "All Good" Corner: Everyone cooperates. Trust is high.
The "Tipping Point" (The Interior Equilibrium): This is the weird middle ground. Here, the math shows a strange phenomenon: increasing the number of Honest people actually lowers the cooperation rate.

Wait, how can more good people lead to less cooperation? The authors explain that this middle ground is unstable. It's like a ball balanced on the very peak of a hill. If you add more Honest people, the "peak" (the tipping point) shifts lower. This means the threshold for entering the "All Good" zone becomes easier to cross, but the specific point where the system is currently stuck (the tipping point) actually sees less cooperation. It sounds contradictory, but the authors clarify that this isn't a stable state where society deteriorates. It's just a moving boundary. The presence of more Honest people makes it easier for the whole society to jump up to the high-cooperation state, even though the math at that specific unstable point looks worse.

The paper also tests how robust this system is by changing the rules slightly.

The "Worst-Case" Scenario: What if agents don't trust the math and only worry about the absolute worst thing that could happen? The paper finds that if agents play it safe and assume the worst, all cooperation disappears. The fear of being exploited outweighs the hope of a reward, and everyone defects.
The "Eraser" Scenario: What if the stigma can be wiped clean? If there is a chance that a bad record gets erased before the next match, the incentive to be good vanishes. The paper shows that if records are cleared often enough, strategic agents will always cheat, because the cost of cheating is too low. Even a little bit of "forgetting" weakens the system.

In the end, the paper suggests that a simple, one-bit record (a clear vs. stigmatized label) is surprisingly powerful. It can discipline behavior without needing long-term friendships or complex community monitoring. However, the system is delicate. It relies on the record staying dirty once it's earned and on agents trusting that the "clear" label means something real. If you try to be too forgiving by clearing records, or if agents are too scared to take risks, the whole mechanism of trust collapses. The lesson isn't that honesty is bad; it's that the design of the reputation system matters just as much as the number of honest people in it.

Technical Summary: Honesty, Stigma, and Cooperation in an Overlapping-Generations Game

Problem and Motivation
This paper investigates the capacity of a minimal, directed reputational mechanism to sustain cooperation in an environment where agents interact only once as young and once as old, never meeting the same partner twice. The central question is whether a single binary record (a "stigma" for defection) can discipline strategic agents who would otherwise exploit partners they will never see again, and how the presence of a share of "honest" commitment types affects this dynamic. The authors specifically challenge the intuition that a larger share of honest agents unconditionally improves cooperation rates, examining instead how honesty interacts with equilibrium selection and stability.

Methodology and Model Structure
The authors construct a two-period overlapping-generations (OLG) game with two populations ( $A$ and $B$ ) of unit mass.

Demographics and Matching: Agents are young at date $t$ and old at $t+1$ . Young agents from $A$ match with old agents from $B$ , and vice versa.
Types: Agents are either honest (probability $\pi$ $π$ ) or strategic (probability $1-\pi$ $1 - π$ ).
- Honest agents follow a commitment rule: they cooperate ( $C$ ) against clear opponents and defect ( $D$ ) against stigmatized ones.
- Strategic agents maximize expected payoffs. They possess a private "exploitation loss" $\ell_i \sim U[0,1]$ , representing the cost incurred if they cooperate against a defector.
Payoffs: The stage game is a Prisoner's Dilemma. If a strategic agent cooperates against a defector, they suffer loss $\ell_i$ . Defection ( $D$ ) yields a temptation payoff $b > 1$ against a cooperator.
Records: A binary record is created only if a young agent defects against a cooperating old opponent. This "stigma" is observed by the young agent's next partner (when they become old). Records are directed (seen by exactly one future partner) and do not broadcast to the community.
Equilibrium Concept: The authors analyze stationary symmetric Perfect Bayesian Equilibria. Crucially, they incorporate Bayesian selection: because honest agents never acquire a stigma while some strategic agents do, a clear record is informative. The posterior probability that a clear old agent is honest ( $h(x)$ ) is endogenously determined by the strategic cutoff $x$ .

Key Results and Characterization
The analysis reduces the equilibrium condition to a quadratic equation derived from the indifference condition of a strategic young agent facing a clear opponent.

Equilibrium Existence and Multiplicity:
- Uniqueness: If the temptation payoff $b$ is moderate ( $1 < b \leq \phi$ , where $\phi$ is the golden ratio), the equilibrium is unique.
- Multiplicity: If $b > 2$ $b > 2$ and the honest share $\pi$ $π$ falls within a specific intermediate range, three equilibria coexist:
  - A low-cooperation corner ( $x=0$ ): Strategic agents always defect.
  - A high-cooperation corner ( $x=1$ ): Strategic agents always cooperate.
  - An interior tipping equilibrium ( $x_I$ ): Strategic agents use a cutoff rule.
Comparative Statics on the Honest Share ( $\pi$ ):
- Along the interior branch, an increase in the honest share $\pi$ lowers the strategic cutoff $x_I$ , the conditional cooperation rate, and the unconditional cooperation rate.
- Interpretation: This counter-intuitive result arises because a higher $\pi$ increases the density of honest agents among the "clear" pool, making the "clear" label less informative about the opponent's type (diluting the signal).
- Stability: The interior equilibrium is locally unstable. It acts as a tipping point separating the basins of attraction for the two corner equilibria. Therefore, while a higher $\pi$ lowers the cooperation rate at the tipping point, it simultaneously lowers the threshold required to reach the high-cooperation corner. The result describes a shift in the coordination boundary, not a deterioration of a stable outcome.
Robustness and Extensions:
- Maximin Rule: If strategic agents evaluate actions based on the worst-case opponent type (rather than Bayesian expectations), strategic cooperation vanishes entirely ( $x=0$ ). The worst-case payoff of cooperation is strictly lower than defection for any positive loss $\ell$ .
- Record Clearing: Introducing a probability $\rho$ that a stigma is cleared before the next match weakens incentives. At stable interior equilibria, higher $\rho$ reduces cooperation. If records are always cleared ( $\rho=1$ ), the unique equilibrium is total defection.
- Discounting: Lower patience ( $\delta < 1$ ) weakens cooperation by reducing the future value of a clear record, mirroring the effect of record clearing.

Significance and Claims
The paper claims to provide a precise characterization of how endogenous record distributions and Bayesian updating shape community enforcement in a sparse environment. Its primary contributions are:

Selection Effects: Demonstrating that ignoring the selection of types induced by records (i.e., assuming the probability of honesty is constant regardless of the record) leads to incorrect equilibrium predictions.
The Role of Honesty: Refining the understanding of how honest types influence cooperation. The paper argues that while a larger honest share can reduce cooperation rates along an unstable interior branch, it ultimately facilitates the selection of the high-cooperation equilibrium by lowering the coordination hurdle.
Minimalist Discipline: Showing that even a single bit of directed memory, without repeated bilateral interaction or community-wide monitoring, can discipline behavior, provided the record is persistent and beliefs are updated correctly.

The authors conclude that design choices regarding record persistence and belief updating are critical; systems that clear records to appear forgiving or fail to account for type selection may inadvertently dissolve the incentives necessary for cooperation.

More like this