Technical Summary: Money Burning Improves Mediated Communication

1. Problem Statement

The paper investigates whether a purely wasteful action—money burning—can strictly improve the Sender's equilibrium payoff in strategic communication. While money burning has long been viewed as a tool to enhance message credibility, the authors distinguish between credibility and payoff improvement. In standard cheap talk (no commitment), burning can only refine existing influential equilibria but cannot strictly improve the Sender's payoff because the cost of burning dissipates the gains from improved credibility. Conversely, under full commitment (Bayesian persuasion), burning is redundant as the Sender can directly commit to an information structure.

The central question is whether money burning is valuable in an intermediate commitment setting: specifically, in Mediated Communication with Money Burning (MDMB). In this setting, a Sender commits to a mediator who maps reports to messages and observable burning amounts. The Sender retains private information and chooses their report, but the mediator enforces the burning consequences. The paper asks if this intermediate environment allows burning to expand the Sender's equilibrium payoff set beyond what is achievable with standard mediated communication (MD) or cheap talk.

2. Methodology

Model Setup

The authors formalize a game between a Sender (with private type $\theta \in \Theta$) and a Receiver (choosing action $a \in A$).

• Transparent Motives: The Sender's payoff $v(a)$ depends only on the Receiver's action, not directly on the type $\theta$. This implies all Sender types rank induced actions identically.
• Protocol: The Sender commits to a mechanism $(M, S, \phi)$ where $M$ is the input set, $S$ is the output message set, and $\phi: M \to \Delta(S \times T)$ maps inputs to output messages and burning amounts $t \in [0, C]$, where $C$ is the budget.
• Equilibrium Concept: Perfect Bayesian Equilibrium (PBE).

Revelation Principle for MDMB

A key methodological contribution is the development of a tailored Revelation Principle for MDMB.

• Challenge: Standard revelation principles (Myerson, Forges) fail here because money burning is a payoff-relevant outcome observed before the Receiver acts, potentially conveying information about the Sender's type.
• Solution: The authors define Canonical MDMBs where the input is the type $\theta$, the output is a posterior belief $\nu \in \Delta(\Theta)$, and the burning amount $x(\nu)$ is a deterministic function of the posterior.
• Decomposition: Any equilibrium outcome can be decomposed into a signaling scheme $\pi$ (generating posteriors) and a deterministic burning scheme $x$. Crucially, in the canonical form, the burning amount contains no information beyond the posterior itself.

Optimization Program

The problem is reduced to a belief-based program maximizing the Sender's ex-ante payoff subject to:

1. Bayes Plausibility: The distribution of posteriors must average to the prior.
2. Obedience: The Receiver plays a best response to each posterior.
3. Incentive Compatibility (Type-Indifference): Under transparent motives, all Sender types must receive the same interim payoff to prevent deviation.
4. Budget Constraint: Burning $x(\mu) \le C$.

The authors introduce a generalized subjective payoff function $\hat{V}_{\lambda, C}(\mu)$, which incorporates Lagrange multipliers $\lambda$ from the type-indifference constraints. These multipliers create "distorted weights" $\omega_\lambda(\mu)$ on posteriors.

3. Key Results

Characterization of Optimal Equilibrium (Theorem 1)

The Sender's maximum equilibrium payoff $V^*_C(\mu_0)$ is characterized as:
$$V^*_C(\mu_0) = \min_{\lambda \in \text{aff}(\Theta)} \text{cav}(\hat{V}_{\lambda, C})(\mu_0)$$
where $\text{cav}$ denotes the concave envelope. The optimal mechanism partitions posteriors into two distinct groups based on the sign of the distorted weight $\omega_\lambda(\mu)$:

1. Persuasion Group ($\omega_\lambda(\mu) > 0$): The Sender burns zero money and induces the Receiver's best response most favorable to the Sender.
2. Credibility-Gaining Group ($\omega_\lambda(\mu) < 0$): The Sender burns the maximum budget $C$ and induces the Receiver's least favorable best response. These messages are not valuable for their direct persuasion but serve to discipline deviations, making the persuasion group incentive-compatible.

Strict Improvement of Communication (Theorem 2 & Proposition 3)

The paper establishes that money burning generically improves mediated communication:

• Theorem 2: Under a generic condition on the Receiver's payoff functions (Assumption 1), if the burning budget $C$ exceeds the range of action payoffs, increasing the budget strictly raises the Sender's payoff, unless the MDMB value collapses to the cheap-talk benchmark.
• Implication: If burning does not improve the payoff, then mediated communication with burning offers no advantage over cheap talk. Thus, burning is not merely a refinement; it expands the equilibrium payoff set.
• Proposition 3: If commitment has value (i.e., $V^*_0(\mu_0) > V^*_{CT}(\mu_0)$), then for almost all priors, money burning strictly improves the Sender's payoff once the budget is sufficiently large.

Value with Unbounded Budget (Proposition 4)

As the budget $C \to \infty$, the negative components of the Lagrange multipliers converge to zero. The limiting value of MDMB equals:
$$V^*(\mu_0) = \min_{\lambda \in \Delta(\Theta)} \text{cav}(\hat{V}_{\lambda})(\mu_0)$$
This is equivalent to the payoff of a cautious Sender with full commitment power who maximizes their minimum interim payoff across types (max-min persuasion). It also equals the Sender's payoff in Bayesian persuasion with heterogeneous priors under the worst subjective prior.

Application to Web 3.0 (Proposition 7 & Corollary 6)

The authors apply the framework to Web 3.0 environments (smart contracts, transparent algorithms).

• They define the refined value of commitment as the gap between Bayesian persuasion and MDMB ($V^*_{BP} - V^*$).
• Result: The refined value of commitment is positive if and only if the conventional value of commitment (gap between Bayesian persuasion and cheap talk) is positive.
• Conclusion: Algorithmic mediation and enforceable burning mitigate the loss from the absence of full commitment but generally do not eliminate it.

4. Significance and Claims

The paper claims to resolve the tension between the wastefulness of money burning and its potential to improve communication efficiency.

• Intermediate Commitment: The primary contribution is identifying that money burning is valuable specifically in the intermediate commitment regime. It requires enough commitment to be enforceable (via a mediator) but not enough to eliminate reporting incentives.
• Mechanism Design: The paper demonstrates that burning acts as an incentive instrument inside a mediated mechanism. By sacrificing resources on specific "credibility-gaining" messages, the Sender can discipline deviations and make "persuasion" messages credible without dissipating the entire surplus.
• Contrast with Cheap Talk: Unlike in cheap talk, where burning is a voluntary signal that cannot strictly improve payoffs under transparent motives, MDMB allows burning to be an enforceable consequence. This distinction allows burning to expand the equilibrium payoff set rather than just refining existing ones.
• Robustness: The results provide a micro-foundation for max-min persuasion objectives and connect mediated money burning to robust Bayesian persuasion with heterogeneous priors.

The authors remain modest regarding the scope, noting that their results rely on the transparent motives assumption. They acknowledge that in environments with state-dependent preferences, the interaction between burning and incentives may differ, and extending the analysis beyond transparent motives is a direction for future work.