Reputational Conservatism in Expert Advice

Basic Information

• Paper ID: 2509.04036
• Title: Reputational Conservatism in Expert Advice
• Authors: Georgy Lukyanov (Toulouse School of Economics), Anna Vlasova (CREST—École Polytechnique)
• Classification: econ.TH (Economic Theory)
• Publication Date: October 14, 2025 (arXiv version v4)
• Paper Link: https://arxiv.org/abs/2509.04036

Abstract

This paper investigates an expert advice problem with career concerns, where an expert possesses a continuous private signal about binary outcomes. The principal always implements safe advice, while the probability of implementing risky advice increases with the expert's reputation; realized outcomes (if observable) update reputation. The study demonstrates that equilibrium advice follows a cutoff form of the signal, and under relatively diagnostic conditions, the cutoff value increases with reputation (reputational conservatism). Comparative statics analysis is transparent, and simple policy levers—success-contingent bonuses and gating—provide implementable tools for experimental control.

Research Background and Motivation

1. Core Problem: Organizations frequently rely on experts to provide advice between safe and risky action plans. A common phenomenon is that senior, high-status experts appear more conservative: they propose risky initiatives less frequently than junior experts, yet when they do take risks, their success rates appear higher.
2. Problem Significance: Understanding reputational effects in expert advice is crucial for organizational design, as it affects decision quality regarding innovation, experimentation, and risk-taking.
3. Limitations of Existing Approaches:
   • Political correctness models focus on perceived bias rather than ability-based reputation updating
   • Cheap talk models with continuous signals lack explicit implementation probability mechanisms
   • Dynamic career concerns environments are complex, making it difficult to isolate short-term selection effects
4. Research Motivation: Provide a simple, tractable framework explaining conservative behavior patterns of high-reputation experts and demonstrate how principals can modulate this pattern.

Core Contributions

1. Theoretical Framework: Establishes a single-period expert advice model where implementation probability increases with reputation, with outcomes updating ability-based reputation
2. Equilibrium Characterization: Proves that equilibrium advice adopts a unique signal cutoff form (Proposition 3.1)
3. Reputational Conservatism: Under transparent relatively diagnostic conditions, the experimentation threshold increases monotonically with reputation (Proposition 4.2)
4. Comparative Statics Analysis: Provides closed-form results for information precision, prior success probability, and career concern intensity
5. Policy Tools: Identifies two implementable levers—success-contingent bonuses and gating—to modulate experimentation behavior

Methodology Details

Task Definition

Investigates a single-period expert-principal advice problem where:

• Input: Expert observes a continuous private signal x∈ℝ about state t∈{0,1}
• Output: Expert recommends action a∈{R,S} (risky or safe)
• Constraint: Principal always implements S, implements R with probability λ(ρ), where ρ is the expert's current reputation

Model Architecture

Timeline Structure

1. Natural state t∈{0,1} is drawn with prior π := P(t=1)∈(0,1)
2. Market holds current reputation ρ∈0,1 about the expert
3. Expert observes private signal x∈ℝ about t and recommends a∈{R,S}
4. Principal implements S with probability 1, implements R with probability λ(ρ)∈0,1
5. If R is implemented, outcome y∈{0,1} realizes and is publicly observed; if S is implemented, outcome is uninformative
6. Market updates expert reputation to ρ⁺ based on public history

Key Technical Components

Information Structure:

• Posterior success probability: p(x) := P(t=1|x), strictly increasing and continuous
• Satisfies monotone likelihood ratio property (MLRP)

Payoff Functions:

• Safe advice: U_S(ρ) = W(ρ)
• Risky advice: U_R(p;ρ,b) = λ(ρ)p(W(ρ⁺{R,1}(ρ)) + b) + (1-p)W(ρ⁺{R,0}(ρ)) + (1-λ(ρ))W(ρ)

Incremental Value Function:

Δ(p;ρ,b) := U_R(p;ρ,b) - U_S(ρ)
           = λ(ρ)[p(Δ_s(ρ) + b) + (1-p)Δ_f(ρ)]

where Δ_s(ρ) := W(ρ⁺{R,1}(ρ)) - W(ρ) > 0, Δ_f(ρ) := W(ρ⁺{R,0}(ρ)) - W(ρ) < 0

Technical Innovations

1. Uniqueness of Cutoff Strategy: Guarantees unique cutoff point exists through strict monotonicity and continuity
2. Relatively Diagnostic Condition (RD): Formalizes the intuition that failure is more diagnostic than success at high reputation
3. Closed-Form Solutions: Explicit expression for posterior cutoff point:

   p̄(ρ;b) = -Δ_f(ρ)/(Δ_s(ρ) + b - Δ_f(ρ))
   

4. Policy Lever Separation: Bonuses affect recommendation threshold, gating affects implementation frequency without changing recommendation strategy

Experimental Setup

Theoretical Verification

As a theoretical economics paper, this work primarily verifies results through mathematical proofs and theoretical analysis rather than traditional empirical experiments.

Numerical Examples

Gaussian Environment:

• Ability types θ∈{H,L}, signal x~N(μ_t(θ), σ²)
• Parameter settings: σ=1, π=1/2, Δ_H=1.2, Δ_L=0.6, W(ρ)=ρ
• Verifies RD condition holds at reputation ρ∈{0.2,0.5,0.8}

Comparison Benchmarks

• Baseline model without reputation concerns
• Linear payoff function case
• Comparisons across different information precision parameters κ

Experimental Results

Main Theoretical Results

Proposition 3.1 (Cutoff Existence): For given (ρ,b), there exists a unique threshold p̄(ρ;b)∈(0,1) such that the expert recommends R if and only if p(x)≥p̄(ρ;b).

Proposition 4.2 (Reputational Conservatism): Under the relatively diagnostic condition, the experimentation cutoff increases monotonically with reputation:

ρ₁ ≤ ρ₂ ⟹ p̄(ρ₁;b) ≤ p̄(ρ₂;b)

Proposition 6.1 (Bonus Mapping): The recommended experimentation rate q(ρ;b) = P(p(x)≥p̄(ρ;b)) is continuous and strictly increasing in b, with a unique bonus b*(ρ;q) achieving target rate q.

Comparative Statics Results

1. Signal Informativeness: Higher precision κ lowers signal cutoff c(ρ;b;κ), posterior cutoff p̄(ρ;b) unchanged
2. Prior Success Probability: Higher π lowers signal cutoff, posterior cutoff unchanged
3. Career Concern Intensity: Stronger ϕ raises cutoff value (when b>0), cutoff unchanged when b=0

Policy Lever Effects

Success-Contingent Bonuses:

• Provide one-to-one mapping of experimentation rates
• Closed-form formula for target achievement
• Do not affect implementation probability

Gating Mechanism:

• Does not change recommendation threshold
• Proportionally scales implementation of experiments
• Slows reputation learning speed

Related Work

Career Concerns Literature

• Holmström (1999): Foundational framework for career concerns
• Scharfstein and Stein (1990): Herding behavior and investment
• Prendergast and Stole (1996): Age effects in reputation acquisition

Strategic Communication

• Crawford and Sobel (1982): Strategic information transmission
• Morris (2001): Political correctness model
• Ottaviani and Sørensen (2006): Cheap talk on reputation

Organizational Design

• Williamson (1996): Governance mechanisms
• Foss (2003): Selective intervention and internal hybrid organizations

Conclusions and Discussion

Main Conclusions

1. Mechanism Identification: High-reputation experts face more visible failures (because advice is more likely to be implemented), while success rewards do not increase proportionally
2. Policy Implications: Organizations can modulate proposal frequency through success bonuses and control risky implementation through gating
3. Empirical Predictions: As reputation increases, risky recommendation frequency decreases, conditional success rate increases

Limitations

1. Static Analysis: Does not fully account for long-term incentive effects of dynamic career concerns
2. RD Condition: Relatively diagnostic assumption may not hold in certain environments
3. Single Expert: Multi-expert competition and relative ranking effects insufficiently explored
4. Implementation Function: λ(ρ) treated as exogenous; endogenizing may alter results

Future Directions

1. Dynamic Extensions: Multi-period models examining early exploration vs. late conservatism tradeoffs
2. Endogenous Implementation: Principal's optimal acceptance strategy
3. Multi-Expert Competition: Relative reputation and differentiated incentives
4. Empirical Testing: Cross-level comparisons of recommendation frequency and success rates

In-Depth Evaluation

Strengths

1. Theoretical Contribution: Provides concise, tractable explanation of reputational conservatism
2. Policy Relevance: Identifies practical organizational design tools
3. Mathematical Rigor: Complete proofs of existence, uniqueness, and comparative statics
4. Clear Intuition: RD condition and mechanism explanation easily understood
5. Robustness: Results do not depend on specific signal distribution assumptions

Weaknesses

1. Assumption Limitations: Generality of RD condition requires further verification
2. Empirical Absence: Lacks empirical testing with real data
3. Dynamic Simplification: Static model may overlook important intertemporal effects
4. Passive Principal: Exogeneity of implementation function limits analytical depth

Impact

1. Theoretical Impact: Provides new analytical framework for career concerns literature
2. Practical Value: Offers concrete policy recommendations for organizational design
3. Methodological Contribution: Methodological value of cutoff strategy analysis and policy lever design
4. Cross-Disciplinary Application: Applicable to management, organizational behavior, and related fields

Applicable Scenarios

1. Consulting Industry: Advice patterns of senior vs. junior consultants
2. Medical Decision-Making: Treatment choices of senior physicians vs. residents
3. Financial Investment: Fund managers' investment strategy selection
4. Policy-Making: Reform propensity of officials at different levels
5. Academic Research: Research direction choices of established vs. emerging scholars

References

Core Theoretical Foundations:

• Holmström, B. (1999). Managerial incentive problems: A dynamic perspective. Review of Economic Studies, 66(1), 169-182.
• Crawford, V. P., & Sobel, J. (1982). Strategic information transmission. Econometrica, 50(6), 1431-1451.

Related Models:

• Morris, S. (2001). Political correctness. Journal of Political Economy, 109(2), 231-265.
• Ottaviani, M., & Sørensen, P. N. (2006). Professional advice. Journal of Economic Theory, 126(1), 120-142.

Organizational Design:

• Williamson, O. E. (1996). The Mechanisms of Governance. Oxford University Press.

This paper provides an important theoretical framework for understanding reputational effects in expert advice. Its parsimonious model and clear policy implications make it valuable in both theoretical economics and practical applications.