📈 economics

Endogenous Vindication: Reputation and Effort in Expert Advice

This paper demonstrates that in a repeated expert-client relationship where evaluation depends on implementation effort, an expert's strategic control over whether a reputation-dependent test occurs—driven by diminishing career returns—can lead to withholding beneficial advice and result in less overall learning despite the availability of more informative tests.

Original authors: Georgy Lukyanov, Anna Vlasova

Published 2026-08-03

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Georgy Lukyanov, Anna Vlasova

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

The Science of Trust and the Hidden Cost of Being Famous

Imagine a world where you can't just say you are good at something; you have to prove it by doing it. This is the heart of reputation, a concept that lives at the intersection of economics and psychology. In this field, scientists study how people make decisions when they don't have all the facts. They look at signaling, which is how someone tries to show off their skills, and learning, which is how the rest of us figure out if those skills are real. Usually, we assume that if an expert gives good advice, we try it, and if it works, we trust them more. But what if the expert is afraid to give that advice in the first place? What if giving good advice is actually a risky gamble for their career? This paper asks a tricky question: Can the very act of trying to build a reputation actually stop people from learning the truth?

The Expert Who Was Too Afraid to Speak Up

In this study, the authors Georgy Lukyanov and Anna Vlasova set up a game between a long-lived Expert (like a doctor, a consultant, or a financial guru) and a stream of short-lived Clients (patients, companies, or investors). Here's the twist: neither the Expert nor the Clients know if the Expert is actually "High Ability" (a genius) or "Low Ability" (a bit of a fraud). They only have a guess, called a reputation score, which goes up and down based on what happens.

The game works like this:

The Expert gets a private hint about whether a risky project is a good idea.
The Expert tells the Client what to do.
The Client decides whether to do the project. If they do, they also decide how hard to work on it (effort).
The project succeeds or fails. If it fails, everyone updates their guess about the Expert's ability.

Here is the secret sauce: The Client works harder if they trust the Expert more. If the Expert has a high reputation, the Client puts in maximum effort. If the Expert has a low reputation, the Client barely tries.

This creates a strange feedback loop. When a Client works hard, a failure is a really bad sign for the Expert. It means the failure wasn't because the Client slacked off; it means the Expert's advice was probably wrong. But if the Client doesn't try hard, a failure is easy to blame on the Client's laziness. So, a high reputation actually makes the "test" of the Expert's ability much stricter and more dangerous.

The "Endogenous Vindication" Trap

The paper discovers a phenomenon the authors call Endogenous Vindication. This is a fancy way of saying: "You can only prove you are good by getting others to test you, but you might be too scared to let them test you."

The authors find that if an Expert cares too much about their career (reputation), they might start hiding good news.

The Scenario: Imagine a High-Ability Expert gets a signal that a project is a great idea. It would make money for the Client.
The Dilemma: If the Expert recommends it, the Client (trusting the Expert) will work super hard. If the project fails (which happens sometimes, even with good ideas), the Client will realize, "Oh no, the Expert must be a fraud!" and the Expert's reputation will tank.
The Result: To protect their reputation, the Expert decides to stay silent and recommend the "safe" option instead. They sacrifice the Client's potential profit to avoid the risk of a harsh test.

This leads to a Cutoff Rule. The paper proves that the Expert will only give a favorable recommendation if their reputation is already very high. If their reputation is in the middle—high enough to get the Client to try, but not high enough to be safe from a bad outcome—they will stay silent.

The Tragic Irony: Less Testing, Less Learning

The most surprising finding is that stronger career concerns can actually stop learning.

The Intensive Margin: When a test does happen, it is very high-quality because the Client is working hard.
The Extensive Margin: But because the Expert is so scared, they stop recommending projects sooner. They stop the testing process entirely.

The authors show that a High-Ability Expert can get "stuck" in a zone of False Distrust. If they have a few unlucky failures early on, their reputation drops below the "cutoff." Once it drops there, they stop giving good advice. The Client stops trying. No new data is generated. The Expert's reputation freezes at a low level, and the world never learns that they were actually a genius. They are permanently distrusted, not because they are bad, but because they were too cautious to let the world see their true talent.

The Numbers Behind the Story

The authors didn't just guess this; they built a mathematical model and ran simulations to prove it.

They found that if the Expert cares a lot about their career (represented by a ratio of career weight to project weight, $\beta/\alpha$ ), the "cutoff" for giving advice moves higher.
In their "canonical" example (a standard test case), when the career concern ratio was low ($0.10$), the Expert started giving advice when their reputation was about 0.67.
When the career concern ratio was high ($5.00$), the Expert waited until their reputation was 0.87 before speaking up.
This delay has a cost: In the high-concern scenario, a High-Ability Expert has a 52.68% chance of being permanently distrusted (absorbed into the "distrust region"). In the low-concern scenario, that chance drops to 16.40%.

What This Means for the Real World

The paper rules out the idea that "more career concern always leads to better advice." Instead, it suggests that when experts are too worried about their image, they might hide their best ideas to avoid the risk of a public failure.

This explains why, in fields like medicine or business consulting, we might see a paradox: The most trusted experts seem to be the ones who are most cautious, while the ones who are willing to take risks (and potentially fail) are the ones who actually help us learn the most. The paper concludes that trust is a double-edged sword. It makes people work harder, which makes the test of the expert's ability sharper, but that sharpness can scare the expert into silence, leaving us all in the dark about who is truly capable.

Technical Summary: Endogenous Vindication: Reputation and Effort in Expert Advice

1. Problem and Motivation

The paper investigates the strategic interaction between a long-lived expert and a sequence of short-lived receivers regarding the provision of advice on risky projects. The central problem is endogenous learning: the quality of the information generated to evaluate the expert's ability depends on the receiver's implementation effort, which in turn depends on the expert's current reputation.

Unlike standard models where performance tests are fixed or where the expert privately knows their own ability, this model features common uncertainty: neither the expert nor the public knows the expert's fixed ability ( $\theta \in \{H, L\}$ ). The expert observes a private binary signal about the suitability of a project but cannot fabricate favorable evidence; they can only suppress it. A favorable recommendation induces the receiver to implement the project and choose a costly effort level. Crucially, higher reputation leads to higher implementation effort. This effort increases the probability of success but, more importantly, makes a failure more diagnostic of low ability (as poor execution becomes a less plausible explanation for a bad outcome).

The expert faces a trade-off: recommending a suitable project yields immediate surplus but exposes the expert's uncertain reputation to a performance test whose precision is endogenously determined by the receiver's trust. The paper asks how this feedback loop affects the expert's advice, the production of information, and the long-run vindication of ability.

2. Methodology and Model Structure

Environment

Agents: A long-lived expert and a sequence of short-lived receivers.
State: A fixed, unknown ability $\theta \in \{H, L\}$ . A state $\omega_t \in \{0, 1\}$ indicates project suitability.
Signals: The expert observes a private signal $s_t$ with accuracy $q_\theta$ ( $q_L < q_H$ ). The signal structure is symmetric: $P(s_t=1|\theta) = 1/2$ for both types, meaning the signal itself does not reveal ability.
Actions:
- Expert: Recommends $R$ (risky project) or $S$ (safe alternative). Constraint: $R$ is only feasible if $s_t=1$ (favorable assessment).
- Receiver: Observes recommendation and public reputation $\pi_t$ . Decides whether to implement ( $x_t=1$ ) and chooses effort $e_t \in [0, 1]$ .
Outcomes: If implemented, $y_t=1$ (success) with probability $\omega_t e_t$ . Outcomes are public.
Payoffs: The expert maximizes a weighted sum of the receiver's project surplus ( $\alpha$ ) and a flow career return from reputation ( $\beta r(\pi_t)$ ), discounted by $\delta$ . The career return function $r(\pi)$ is increasing; concavity represents diminishing returns or risk aversion.

Key Mechanisms

Reputation-Dependent Effort: The receiver implements the project only if the expected surplus exceeds a fixed cost $\kappa$ . This creates a mechanical implementation threshold $\underline{\pi}$ . Above this threshold, effort $e(\pi)$ is strictly increasing in reputation.
Endogenous Diagnosticity: The experiment generated by a favorable recommendation is more informative (in the Blackwell sense) when effort is higher. Specifically, the likelihood ratio of failure becomes more unfavorable to high ability as effort increases, while the likelihood ratio of success remains constant.
Strategic Exposure: The expert chooses whether to trigger the test. If the career cost of a potential reputation drop (due to a failure) outweighs the project surplus, the expert may suppress a favorable signal, recommending $S$ even when $s_t=1$ .

3. Key Results

The Advice Cutoff and Monotone Exposure

The paper establishes conditions under which the expert's strategy follows a reputation cutoff.

Transformation: The infinite-horizon problem is reduced to an optimal-stopping recursion by subtracting the annuity value of a static reputation. The expert's decision depends on a "flow index" comprising the project surplus and a "Jensen cost" of reputational exposure (the cost of a mean-preserving spread in future reputation).
Monotone Exposure Condition: The paper proves that if the implementation threshold is sufficiently high, the Jensen cost of exposure is weakly decreasing in reputation. This occurs because near full reputation, the sensitivity of effort is bounded, and the posterior variance shrinks.
The Cutoff: Under this condition, there exists a unique cutoff $\pi^A \ge \underline{\pi}$ $π^{A} \geq \underline{π}$ .
- If $\pi_t < \pi^A$ , the expert suppresses favorable advice (recommends $S$ regardless of signal).
- If $\pi_t \ge \pi^A$ , the expert gives favorable advice.
- The cutoff $\pi^A$ is weakly increasing in the strength of career concerns ( $\beta/\alpha$ ). As career concerns dominate, the cutoff approaches 1.

Endogenous Vindication and Absorption

The strategic cutoff creates an absorbing distrust region $[0, \pi^A)$ .

Mechanical Absorption: Once reputation falls below $\pi^A$ , no projects are implemented, no outcomes are generated, and reputation remains fixed. Silence is permanent.
Low-Ability Experts: Enter the distrust region with probability 1 in finite expected time.
High-Ability Experts: Enter the distrust region with strictly positive probability due to an unlucky sequence of failures. If they avoid absorption, their reputation converges to 1 (full vindication).
False Distrust: The probability of a high-ability expert being permanently distrusted is characterized by an exact likelihood-ratio representation. Stronger career concerns raise the cutoff, thereby increasing the probability of false distrust and reducing the total information generated about ability.

Comparative Statics and Information

Intensive vs. Extensive Margins: Stronger career concerns lead to a higher cutoff. This has two opposing effects:
1. Intensive: Surviving tests occur at higher reputations, inducing higher effort and more diagnostic outcomes.
2. Extensive: Fewer tests are conducted overall because the process stops earlier.
Net Effect: The extensive margin dominates. Higher career concerns weakly reduce the total information about ability and the probability of vindication. The public record under a higher cutoff is a "garbling" of the record under a lower cutoff.

4. Contributions and Significance

Theoretical Contributions

Endogenous Vindication: The paper introduces a mechanism where an expert can only establish ability by repeatedly inducing others to undertake actions that reveal it. This contrasts with models where ability is revealed through fixed tests or private signals.
Strategic Cessation of Learning: Unlike standard career-concerns models where experts might "babble" or distort forecasts, here the expert stops the generation of evidence entirely. This leads to permanent learning breakdowns (absorption) rather than temporary distortions.
Role of Concavity: The paper clarifies that concave career returns (diminishing returns) are necessary for the distortion but not sufficient for a simple cutoff. It provides primitive conditions (smoothness, high implementation thresholds) under which the "monotone exposure" property holds, ensuring a clean cutoff strategy.
Common Uncertainty: By assuming the expert does not know their own ability, the model isolates the exposure decision from type-signaling motives. The expert's recommendation is a choice to expose uncertain talent to a public test, not a signal of known talent.

Significance Claims

The authors argue that the model explains why greater trust can be associated with less learning in the aggregate.

In environments like medical treatment, consulting, or organizational change, high reputation induces high effort, making failures highly diagnostic.
However, if the expert is risk-averse regarding career outcomes (concave returns), they may withhold favorable advice to avoid the risk of a diagnostic failure.
Consequently, data restricted to implemented recommendations may show high effort and diagnostic failures, but this selection bias hides the fact that many potential tests were suppressed. The "surviving" tests are more informative, but the total volume of learning is reduced.

The paper concludes that reputation acts as both an input to the production of performance evidence and an outcome of that evidence. Ability can only be vindicated while the expert continues to generate tests; once the expert enters the distrust region, the process of vindication halts permanently.