Hidden Information and Self-Selection

This section is not intended for the public.

OPTIMAL CONTRACTING

Principal \(P\) / Agent \(A\) problem

\(x\) = output
\(t\) = payment
\(s\) = type

Agent maximises: \[ U(x,t,s) = t - C(x,s) \]

Principal maximises:

\[ V(x,t) = R(x) - t \]

Timing

  • \(A\) observes \(s\)
  • \(P\) believes \(s \sim D(\cdot)\)
  • \(P\) offers a contract
  • \(A\) accepts or rejects
  • \(A\) chooses output, then \(P\) pays \(A\)

Full information benchmark

Let the reservation utility be zero. Then, if \(P\) observes \(s\), the optimal contract solves:

\[ \max_{x,t} \; R(x) - t \]

s.t. no rent for \(A\):

\[ t - C(x,s) = 0 \]

\[ \Rightarrow R'(x^*) = C_x(x^*, s), \quad t^* = C(x^*, s) \]

Delegation approach

\(P\) commits to a contract \(t = T(x)\)

If \(s \in \{L,H\}\), then \(T(x)\) is defined at two values.

Revelation approach

\(P\) offers a DRM:

\[ \{x(\hat{s}), t(\hat{s})\} \]

Agent solves:

\[ \max_{\hat{s}} \; t(\hat{s}) - C(x(\hat{s}), s) \]

Truthful DRM:

\[ \hat{s} = s \]

Revelation principle

Given any contract \(t = T(x)\) or DRM, there exists a truthful DRM yielding the same \((x,t)\).

P-A model

\(s \in \{L,H\}\) where s enters the cost of production, with \(L < H\).

\[ P(s=H)=q, \quad P(s=L)=1-q \]

Principal solves:

\[ \max_{x_H,x_L,t_H,t_L} \; q(R(x_H)-t_H) + (1-q)(R(x_L)-t_L) \]

s.t.

\[ t_H - C(x_H,H) = 0 \qquad\text{(P for $H$)} \]

\[ t_L - C(x_L,L) \ge 0 \qquad\text{(P for $L$ - always satisfied)} \]

\[ t_H - C(x_H,H) \ge t_L - C(x_L,H) \qquad\text{(IC for $H$ - never bites)} \]

\[ t_L - C(x_L,L) = t_H - C(x_H,L) \qquad\text{(IC for $L$)} \]

Full information benchmark

\[ R'(x_s^*) = C_x(x_s^*, s), \quad t_s^* = C(x_s^*, s) \]

Hidden information solution

We have the following:

  • \(\hat{x}_L = x_L^*\) (L’s output is efficient)
  • \(\hat{t}_L - C(\hat{x}_L,L) \ge 0\) (L earns positive rent)
  • \(\hat{x}_H < x_H^*\) (H’s output is inefficient)
  • \(\hat{t}_H - C(\hat{x}_H,H) = 0\) (H does not earn rent)
  • \(\hat{x}_H < \hat{x}_L\) (H produces less than L)
  • \(\frac{\hat{t}_H}{\hat{x}_H} > \frac{\hat{t}_L}{\hat{x}_L}\) (H is paid more per unit of output than L)

Social planner

The social planner solves:

\[ \max_{x_H,x_L} \; q\big(R(x_H)-C(x_H,H)\big) + (1-q)\big(R(x_L)-C(x_L,L)\big) \]

s.t. participation and IC

\[ R'(x_s^*) = C_x(x_s^*, s) \]

SIGNALLING

One informed agent, multiple uninformed principals. The agent always moves first. Differences with respect to optimal contracting (OC):
- S: the type enters the P’s utility function - OC: one principal, one agent; S: multiple principals, one agent - OC: principal moves first; S: agent moves first - OC: unique equilibrium; S: multiple equilibria - OC: good type envies bad; S: bad type envies good

Agent:

\[ U(w,e,n) = w - C(e,n) \]

Principal:

\[ \pi = y(n,e) - w \]

where

\(w\) = wage
\(n\) = ability (type)
\(e\) = education
\(y\) = output

Timing

  • \(A\) observes \(n\)
  • \(P\) has prior \(P(n=H)=q\)
  • \(A\) chooses \(e\)
  • \(P\) offers \(w\)
  • \(A\) accepts or rejects

Strategies

\[ A: e(n), \quad P: w(e) \]

Beliefs:

\[ \mu(H|e) \]

Full information benchmark

As as consequence of perfect competition, \(\pi=0\) and \(w(e) = \mu(H|e) y(H,e) + (1-\mu(H|e)) y(L,e)\). Follows:

\[ w(e) = y(n,e) \]

The agent solves:

\[ \max_e \; y(n,e) - C(e,n) \]

Define envy as the situation in which the agent of type \(L\) prefers the full-information contract designed for type \(H\): \[ y(H,e^*(H)) - C(e^*(H),L) > y(L,e^*(L)) - C(e^*(L),L) \]

Pooling

By definition:

\[ e(H) = e(L) = e_p \]

\[ \mu(H|e_p) = q \]

The it must be:

\[ w(e_p) = q y(H,e_p) + (1-q) y(L,e_p) \]

usually pooling equilibria are enforced with strong off-patch beliefs.

Intuitive criterion

Set beliefs:

\[ \mu(H|e') = 0 \]

for deviations that are never profitable for \(H\)

Separating equilibria:

No envy: \(\mu(H|e) = 1\) for \(e \ge e^*(H)\), \(\mu(H|e) = 0\) for \(e < e^*(H)\)

\[ e(L) = e^*(L), \quad e(H) = e^*(H) \]

Envy: \(\mu(H|e) = 1\) for \(e \ge e^s\), \(\mu(H|e) = 0\) for \(e < e^s\): H overinvest in education to separate from L

\[ e(L) = e^*(L), \quad e(H) = e^s > e^*(H) \]

where \(e^s\) is such that L is indifferent between its full-information contract and the one designed for H:

\[ y(H,e^s) - C(e^s,L) = y(L,e^*(L)) - C(e^*(L),L) \]


SCREENING

Multiple competing uninformed principals move first, then one informed agent moves.

Timing

  • \(A\) observes \(n \in \{L,H\}\)
  • \(P\) has prior \(P(n=H)=q\)
  • \(P\) offers contract \((w,e)\)
  • \(A\) accepts or rejects
  • \(A\) acquires e and receives w

Strategies

\[ P_i: (w_i,e_i), \quad A: (w,e) \in \{(w_i,e_i)\} \]

Results

  • Any contract yields zero profits due to competition
  • No pooling is possible: let \(P_i\) offer a pooling contract \((w_p,e_p)\), then any \(P_j\) can offer a slightly better contract \((w_p+\epsilon,e_p)\) to attract only the high type and make positive profits (cream skimming).

Uniqueness

L always chooses \((w(L)^*, e(L)^*)\) since every contract of the type \((w(L)^*-\varepsilon, e(L)^*)\) could generate a profit for some \(P_i\).

No envy: H chooses \((w(H)^*, e(H)^*)\) for a similar reason.

Envy: H chooses \((y(H,e^s), e^s)\) since any contract of the type \((y(H,e^s)+\varepsilon, e^s)\) could generate a profit for some \(P_i\) while only attracting the high type.

Existence

No envy: PBE always exists

Envy: $ = $ such that

\[ q \le \hat{q} \Rightarrow \text{unique PBE} \]

\[ q > \hat{q} \Rightarrow \text{no PBE} \]