1 of 25

Would I lie to you? �Project Selection with Biased Advice

Miguel Martinez-Carrasco Univ. de Los Andes

Eric Schmidbauer U. Central FL

John Hamman FSU

ESA, Bogotá, June 2024

We are grateful to IFREE for funding

2 of 25

Our research question

How will competition among biased advisors affect the quality of a chosen project when the principal faces information and resource constraints?

3 of 25

A (stylized) motivating example

CEO of multidivisional firm with limited resources seeks to invest in a project
Division heads report profitability of their best project
CEO wants best project, but division heads want their project accepted
New projects available in future if none accepted now

See also: lobbyist pushing preferred agenda; construction firms bidding on project; agent seeking contract for client (sportsball!)

4 of 25

Conflict

Agent and Principal prefer high quality projects
Agent prefers own project, independent of quality

If facing competition: when might an agent truthfully report her project quality?

Contracting on outcome
Reputation
What if we get rid of these?

Endogenous lying cost

Inducing acceptance now consumes resources
Pre-empts better opportunities in future

5 of 25

Will agents truthfully report?

Stark modeling assumptions

Self-interested agents
No explicit cost of lying
No ability to contract on outcome
No role for reputation

6 of 25

Model

There is a principal and “n” advisors

Each advisor receives a project via iid draw 𝜃 (private information)

Two states: 0 ≤ l < h with Pr(l)=r.

Agents simultaneously report “Low” or “High” (cheap talk)

Principal picks one project to adopt, ending the game, or rejects all and the stage game repeats with iid draws.

Stage game payoffs

Payoff of advisor i = θ_i if adopted, 0 otherwise.
Payoff of principal = θ of adopted project, 0 if none is adopted

7 of 25

Example with 2 advisors

8 of 25

Proposition 1

9 of 25

Equilibrium selection

With two states, there are 2 qualitatively different equilibria.

L-advisor induces rejection, H-advisor induces acceptance (truth-telling).
L- and H-advisor induce acceptance (babbling).

10 of 25

Experimental design

Groups of three or four, with make-up {P, A1 , A2, A3 }.
Each A_i privately sees the randomly assigned project value, l with probability r or h with probability 1 − r.
Each A_isends the message “High” or “Low”.
P sees the messages and decides to accept one or none.
If none is accepted go back to step #2, with i.i.d draws.

Private known histories displayed.

11 of 25

Experimental design

We conducted all sessions in the xs/fs laboratory at Florida State University.
Each session involves between 18 and 24 subjects
45 minutes sessions.
Earnings: 15 dollars per subject on average.

Elicitation of beliefs on truthful reporting when agents received a low-quality project.
Questionnaire on demographics.
Eckel-Grossman risk aversion test.

12 of 25

18 to 24 subjects per session, with 297 participants total

Experiment 1

13 of 25

Results: Experiment 1

15 of 25

Experiment 2

Ten games of repeated play with random anonymous rematching.

16 of 25

Results: Experiment 2

18 of 25

Results: Experiment 2 –Number of bad projects accepted

19 of 25

Exp 1 (one-shot)

Truthful reporting increases in r
Decreases in l
Decreases in n, weakly
Participants hold surprisingly (to me) accurate beliefs

Exp 2 (repeated)

Truthful reporting increases in r
Decreases in l, weakly
Decreases in n
Participant beliefs become more pessimistic over time
Truthful reporting gradually declines over time

20 of 25

The role of beliefs

21 of 25

Results: Beliefs

22 of 25