How limiting endorsements can make unverifiable advice credible

The mechanism is scarcity. If an adviser limits favorable recommendations within a moving window of time, endorsing something today leaves less room to endorse something tomorrow. Even an adviser motivated entirely by personal gain may then pass over weaker opportunities to preserve capacity for stronger ones.

Wang’s preprint, Observable Self-Restraint in Repeated Cheap Talk, develops this argument mathematically. It also identifies a weakness: a sufficiently attractive opportunity can make breaking the rule worthwhile, even when the audience understands exactly what is happening.

A record without an accuracy score

The model pairs one long-lived adviser with a succession of receivers who care only about their current decision. Each period, the adviser privately observes a new situation and recommends one of two actions. Receivers can see past recommendations, but receive no feedback about the underlying situations—not even through realized payoffs.

“Cheap talk” means the recommendation itself carries no direct cost. The adviser cannot commit to telling the truth or obeying the announced restriction. Any discipline must come from the consequences of the recommendation record.

The situations are drawn independently across periods. That assumption matters: past endorsements do not predict whether today’s opportunity is good. Instead, the record changes the adviser’s incentives.

Wang represents those incentives as a finite discounted decision problem. For each possible recent record, the analysis compares the immediate benefit of an endorsement with the future opportunities it displaces. Advice becomes selective when that opportunity cost is high enough to deter recommendations the receiver would not want to follow.

Why the timing of advice matters

A rolling quota might allow at most two endorsements in any seven periods. An endorsement occupies capacity until it leaves that window. Two recent endorsements therefore constrain the adviser longer than two older ones.

The paper proves that removing an inherited endorsement, or moving it earlier in the record, weakly lowers the opportunity cost of a new endorsement under a rolling quota. Conversely, a more restrictive record can make favorable advice more selective.

A simple endorsement count misses this timing effect. Even a single score that discounts older endorsements can rank some records incorrectly: records with the same number of endorsements may release capacity on different schedules.

Quota size and window length pull in opposite directions. More capacity makes endorsing easier, potentially weakening credibility. A longer window makes endorsements more costly in terms of forgone future opportunities, potentially strengthening it.

A credible violation can end communication

The paper separates two questions. Would the adviser secretly misuse available capacity—for example, endorsing a mediocre opportunity while staying within the quota? And would the adviser openly exceed the quota?

Its hypothetical startup accelerator illustrates the distinction. Good and bad ventures arrive independently with equal probability. Funding a good venture gives both accelerator and investor a payoff of one. Funding a bad venture gives the accelerator one-quarter but costs the investor two. Without useful advice, the investor declines to fund.

Suppose the accelerator announces at most one endorsement in two consecutive periods. Saving that endorsement for a good venture can be worthwhile because using it on a bad venture blocks the next opportunity.

But what if two good ventures arrive consecutively? Within a particular range of patience, the accelerator endorses both. Investors infer that the second, rule-breaking endorsement also identifies a good venture and fund it. Informative communication then ends permanently in the equilibrium studied.

Under those assumptions, breakdown eventually occurs with probability one. The violation need not expose bad current advice; it can reveal a genuinely attractive opportunity while destroying the arrangement that made advice credible.

Restraint has limits

This creates a design trade-off. Restrictions must be tight enough to discourage indiscriminate endorsements, but not so tight that preserving the relationship becomes less valuable than breaking the rule.

Under a stated condition ensuring that increasingly selective recommendations remain acceptable to receivers, suitable rolling quotas approach the adviser’s theoretical payoff ceiling as future rewards receive greater weight. That ceiling still includes a cost of self-restraint; patience does not eliminate the conflict of interest.

These are theoretical results, not observations from accelerators or financial advisers. They depend on independent situations, publicly observable recommendations and the specified permanent loss of informative communication after a violation. The preprint explains how credibility could arise without accuracy feedback—not whether a particular endorsement policy would achieve it in practice.