Peace Talk and Conflict Traps

Technical Summary: Peace Talk and Conflict Traps

Problem Statement
The paper addresses a central paradox in conflict resolution: costly pre-play messages can deter unnecessary wars by reassuring potential adversaries, yet the same messages can entrench stalemates once violence has begun. The authors investigate how communication interacts with persistent heterogeneity (the existence of "bad" types who always choose conflict) and thin, noisy private memory to produce these divergent outcomes. Specifically, the paper asks how one-sided, costly signaling affects the hazard of conflict onset and the duration of conflict spells in an overlapping-generations (OLG) security dilemma, and how the introduction of a small probability of publicity (leaks) transforms these dynamics into long-run "traps."

Methodology
The authors develop an overlapping-generations model of conflict between two groups, $G = \{A, B\}$.

• Population and Types: Agents live for two periods (young and old). Each group has a persistent type, $\theta \in \{n, b\}$ (normal or bad), drawn from a common prior. "Bad" types are behavioral and always choose conflict ($C$); "normal" types prefer peace ($N$) if sufficiently reassured.
• Stage Game: Interactions follow a symmetric security dilemma (stag-hunt) payoff structure with a first-strike advantage ($g$) and a second-strike disadvantage ($\ell$).
• Information and Memory: When young, agents privately observe a noisy signal of the opponent's action with error rate $\epsilon$. This creates a one-bit private state ($h \in \{g, b\}$) carried into the old age. Agents do not observe the opponent's type directly.
• Signaling Protocol: Before acting, the current old agent (sender) may send a costly private binary message ($m \in \{0, 1\}$) to the current young agent (receiver). Sending $m=1$ incurs a cost $k > 0$.
• Publicity Extension: The model is extended to include an exogenous probability $q$ that a costly message leaks, creating a public record ($R \in \{\emptyset, S, F\}$) of successful or failed reconciliation attempts.
• Equilibrium Concept: The authors characterize Stationary Perfect Bayesian Equilibria (SPBE), focusing on mixed-signaling strategies where senders randomize based on their private history.

Key Contributions

1. Integration of Signaling into OLG Security Dilemmas: The paper introduces pre-play communication into the Acemoglu-Wolitzky OLG framework. It characterizes a novel mixed-signaling equilibrium where normal senders send costly reassurance only after alarming private histories ($h=b$), while bad types strategically mimic this behavior to keep receiver beliefs at a critical cutoff.
2. Separation of Onset and Duration: The authors provide a sharp theoretical separation between the onset of conflict and its duration. They demonstrate that in a purely private environment, signaling strictly reduces the hazard of conflict onset among normal-normal matches but leaves the expected duration of conflict spells unchanged.
3. Publicity-Driven Bifurcation: The paper shows that even a small probability of publicity ($q > 0$) fundamentally alters the dynamics. Public leaks create a reputation state that leads to absorption in either a "peace trap" (permanent peace after a leaked success) or a "conflict trap" (permanent conflict after a leaked failure).
4. Cheap Talk Impossibility and Fix: The paper proves that zero-cost (cheap talk) messages are uninformative in this setting because all sender types prefer higher receiver cooperation. It offers a disciplined minimal fix: an arbitrarily small, single-crossing cost perturbation restores partial informativeness.

Key Results

• Mixed-Signaling Equilibrium (Private Model): For an intermediate range of signal costs $k$, a stationary equilibrium exists where:

  • Normal senders with a benign history ($h=g$) never signal.
  • Normal senders with an alarming history ($h=b$) mix between sending and not sending.
  • Bad senders mix to ensure the receiver's posterior belief after a signal remains exactly at the indifference cutoff ($\bar{p}$).
  • The sender's mixing probability is pinned by an indifference condition equating the expected gain from persuading a normal receiver to the cost $k$.

• Onset vs. Duration (Private Model):

  • Onset: Signaling strictly lowers the per-period probability of conflict onset among normal agents compared to a no-message benchmark. Reassurance defuses alarming histories that would otherwise lead to conflict.
  • Duration: Conditional on conflict onset, the expected duration of a conflict spell among normals remains $1/\epsilon$ (geometric), identical to the no-message case. This is because, in the private model, spells end only via private misperception (the young misperceiving $C$ as $N$), and signaling does not alter the continuation path once conflict has started.
  • Welfare: Due to reduced onset without increased duration, stationary welfare among normal agents strictly increases relative to the no-message benchmark.

• Publicity and Traps (Extended Model):

  • Absorption: With $q > 0$, the system generically absorbs into one of two states:
    • Peace Trap: A leaked success ($R=S$) coordinates future agents on permanent peace ($N$ forever).
    • Conflict Trap: A leaked failure ($R=F$) coordinates future agents on permanent conflict ($C$ forever).
  • Duration with Publicity: Unlike the private model, leaked failures on the conflict path ($R=F$) harden receivers, making them less responsive to future signals. This strictly increases the conditional expected duration of conflict spells on a non-empty parameter set.
  • Welfare Trade-off: Publicity improves welfare if the system absorbs into a peace trap (dominating all other regimes). However, if it absorbs into a conflict trap, welfare is minimized. In the transient phase (before absorption), welfare is ambiguous: it depends on whether the reduction in onset outweighs the increase in conditional duration caused by leaked failures.

• Cheap Talk: In the baseline with zero costs, no informative equilibrium exists (babbling). However, introducing a tiny cost asymmetry (where "dovish" messages are slightly cheaper for reassuring types than for bad types) allows for a partially separating equilibrium.

Significance and Policy Implications
The paper claims to clarify the mechanisms by which communication shapes conflict on two distinct margins: the initiation of violence and the persistence of stalemates.

• Private vs. Public Channels: The authors argue that the choice between private back-channels and public pledges depends on the audience's ability to coordinate on forgiveness.

  • Private Back-channels are preferable when institutions cannot coordinate on a "peace trap." They reduce the incidence of new conflicts without lengthening existing stalemates, thereby strictly improving welfare.
  • Public Pledges are desirable only if there is a mechanism (e.g., institutions or audience norms) to ensure that a single leaked success triggers a permanent shift to peace. Without such coordination, publicizing failed overtures can backfire by creating "conflict traps" that lengthen stalemates.

• Testable Predictions: The framework generates specific empirical predictions: back-channel reassurance should reduce the frequency of new conflict spells; publicized failed overtures should be followed by longer de-escalation periods; and publicized successful overtures should lead to durable peace.

The paper concludes that its simple, flexible framework disentangles how communication functions as a private persuasion device versus a public coordination device, explaining why some peace talks avert war while others, when exposed, entrench it.