Problem Set 2 (Extended)

PhD Seminar in Corporate Finance

November 16, 2024

1 Rules versus Discretion (24 point)

Consider the problem of a bank endowed with a legacy asset that delivers stochastic returns. There
are 3 periods, t ∈ {0, 1, 2}. If the asset if liquidated at t = 2, it yields a random payoff v ∈ [v, v̄]
drawn from some distribution F (·). If liquidated at t = 1, it delivers a return of L < v. Additionally,
conditional on the asset not being liquidated until t = 2, the bank receives an extra return x ∈
{0, R} with probability q ∈ [0, 1],which is chosen by the bank at a cost c(q). Assume that c is non-
decreasing, convex, and satisifes c000 /c00 ≥ −1/R.
The bank’s short-term creditors do not observe v but observe a noisy signal s ∈ S, which is
drawn from the cdf G (s|v) with pdf g(s|v) log Supermodular. The bank has short and long term
liabilities in the amount of D and B, respectively. If short-term creditors withdraw at t = 1 they
are entitled to D. If instead, the roll-over until t = 2, they are entitled to D (1 + r ). We assume that
D > L.
Finally, the policy maker can choose to write down some of the bank’s long-term debt, a ∈ [0, B],
and convert it into equity to increase the amount of skin in the game for the bank. Importantly, the
policy maker does not have commitment power and cannot condition their policy to the realization
of v. Alternatevely, the asset return v is not verifiable.
The timing is as follows. First, the policy maker, observes v, and for each value of s publicly
announced the policy maker follows a decision rule α(v, s). Short-term creditors observe the choice
made by the policy maker and the realization of s and update their beliefs of v, accordingly. Finally,
the bank manager observes the policy maker’s decision and chooses q ∈ (0, 1). The solution concept
is PBE consistent with D1 criterion. We consider two different policies. One in which the policy
maker chooses a policy after observing the fundamentals v and public signal s and one in which the
policy maker makes her policy contingent on the realization of s, before knowing v, and commits to
it. We refer to these 2 cases as the discretionary and rules case, respectively.
We assume that E [min {(1 + r ) D, v + x} |s] > D so that short-term creditors do not run if they
expect the rest of short-term creditors not to run on the basis of public information alone.

1
1. Let K ( a, v) ≡ v + R − (1 + r ) D − ( B − a). Write down the expected payoff to the bank’s
shareholders as a function of K, q, and R.

2. Find the bank’s optimal choice of q∗ ( a, v). Is q∗ monotone in some of the variables?

3. Argue that the policy maker would choose a different level of q. What is the underlying
friction?

4. Argue that, if Eµ [min {(1 + r ) D, v + x}] < D, then under any equilibrium of the continuation
game, short-term creditors choose to run. Here, µ represents the distribution of x, and is such
that P [ x = R| a, v, s] = q∗ ( a, v).

5. Now, let’s turn to the policy maker’s problem. After she learns v, she chooses a anticipating
the bank’s manager will pick q∗ ( a, v). That is, she maximizes

W ( a, v) ≡ v + q ( a, v) R − c (q ( a, v)) − κa,

where κa represents the cost of intervention. Let

a∗ (v) ≡ arg max W ( a, v) .

a
What can you say about the monotonicity of a∗ ( v )?

6. Define v0 (s) as the lowest value v0 for which the regulator can (i) choose the first-best action
a∗ (v0 ), (ii) reveal to creditors that asset values are no better than v0 , and (iii) avoid a bank
run, in the sense that beliefs after this revelation are optimistic enough to violate the bank run
condition in (3). That is,

v0 (s) ≡ inf v0 : E min {(1 + r ) D, v + x} | a∗ v0 , s, v ≤ v0 = D .

 


Consider the discretionary case. That is, assume that the policy maker observes s and then
chooses her intervention. Prove that, in any equilibrium, we must have that the policy maker’s
optimal αd (v, s) = { a∗ (v) , a0 } where a0 ≤ a∗ (v0 (s)). Hint: Note that incentive compatibility
requires that the policy maker does not have incentives to choose αd (v0 , s) with v0 6= v when the true
value is v. In other words, it must be the case that
!
∂ 
 ∂ ∂ d
0 = 0 W αd v0 , s , v = W ( a, v) × α (v, s) .
∂v ∂a a=αd (v,s) ∂v

7. Consider now the rules case. Here, the policy pre-commits to write down the bank’s long-term
debt following a rule A (s) which does not depend on v. Show that, if A (s) > a∗ (v0 (s)), the
probability of insolvency is larger under discretion than under rules. What is the intuition of
this result?

8. Conclude that the expected welfare given public signal s is lower under discretion than under
a rule with A(s) > a∗ (v0 (s)), as long as the distribution of v given s = s places sufficient
weight on low realizations of v.

2
2 Investment and Coordination (20 points)

There is a continuum of investors over [0, 1]. Each investor has to choose an action a ∈ {0, 1}, where
a = 1 represents the action of investing in a given financial institution, and a = 0 the decision of not
investing/running from it. All investors have the same payoff function u( a, l, x ), where l represents
the fraction of investors who choose to invest, and x represents the investor’s private signal. To
analyze best responses, it is enough to know the payoff gain from choosing one action rather than
the other. Thus, the utility function is parameterized by a function π : [0, 1] × R → R with

π (l, x ) ≡ u(1, l, x ) − u(0, l, x ).

A state θ ∈ R is drawn according to the (improper) uniform density on the real line. Player
i observes a private signal xi = θ + σε i , where σ > 0. The noise terms ε i are distributed in the
population with continuous density f (·), with support on the real line.1 Assume that f is log-
concave.
We make the following assumptions:

• π (l, x ) is non-decreasing in l

• π (l, x ) has the single crossing from below property in θ. That is, for any l, if π (l, x ) > 0 for some
x, then π (l, x 0 ) > 0 for any x 0 > x.
R1
• There exists a unique value θ ∗ for which 0
π (l, θ ∗ )dl = 0.

• There exist θ, θ̄ ∈ R, such that [1] π (l, x ) < 0 for all l ∈ [0, 1] and x ≤ θ; and [2] π (l, x ) > 0 for
all l ∈ [0, 1] and x ≥ θ̄.

1. Let U ( x, k ) be the expected payoff gain to choosing action 1 for an investor who has observed
a signal x and knows that all other players will choose action 0 if they observe signals less
than k. Write down U ( x, k ) as an expected payoff.

2. Show that is U ( x, k ) non-increasing in k and has the single crossing from below property in x.

3. We now proceed according to the iterated deletion of interim strictly dominated strategies (IDISDS).
Suppose first that all investors run regardless of their private signals. Argue that there must
exist ξ̄ 1 satisfying U ξ̄ 1 , ∞ = 0. Use this fact to conclude that strategies with a ( xi ) = 0 for

xi > ξ̄ 1 are strictly dominated.

4. Proceed by induction and show that, if we know that strategies with a ( xi ) = 0 for xi > ξ̄ n
are iteratively dominated, then there must exist ξ̄ n+1 such that strategies with a ( xi ) = 0 for
xi > ξ̄ n+1 are iteratively dominated. What is the equation defining ξ̄ n+1 ? What do we know

about the sequence ξ̄ n n in terms of monotonicity?
1 The uniform prior on the real line is “improper” (i.e., has infinite probability mass), but the conditional probabilities
are well defined: a player observing signal xi puts density (1/σ) f (( xi − θ )/σ) on state θ.

3
 n o
5. Use a similar approach to show that there exists a sequence ξn such that all strategies
n
surviving n-rounds of the IDISDS have the property that a ( xi ) = 0 for all xi ≤ ξ n .

6. Finally, write Ψσ (l; x, k ) for the probability that a player assigns to proportion less than l of
the other investors observing a signal greater than k, if he has observed signal x. Observe that
if the true state is θ, the proportion of players observing a signal greater than k is 1 − F ((k −
θ )/σ ). This proportion is less than l if θ ≤ k − σF −1 (1 − l ). So,
Z k−σF−1 (1−l )
x−θ
 
1
Ψσ (l; x, k ) = f dθ.
−∞ σ σ
 
x −k
Show that Ψσ (l; x, k ) = 1 − F σ + F −1 (1 − l ) .

7. Use the following result to show that ξ¯n , ξ n → θ ∗ . Conclude that this investment game with
n→∞
heterogeneously informed investors has a unique equilibrium.

3 Challenging the Large Shareholder Intuition (20 points)

Consider the problem of a board and a manager. The board is composed by a large shareholder of
size α, and small (infinitesimal) shareholders of aggregate size (1 − α). The manager is in charge
of choosing one (if any) among three projects. Project 1 is a terrible project that delivers cashflows
y = −∞. Projects 2 and 3 deliver cashflows y ∈ R+ and personal perks to the manager B ∈ R+
y
according to Fi ∈ ∆ R2+ , i ∈ {2, 3}. We denote as Fi and FiB the marginals with respect to y and B,

respectively. We make the following assumptions:

(i) There exists π ∈ [0, 1] parametrizing the cashflow distribution of project 2 so that, for any
y y y
π0 > π ≥ 0, F2 [π 0 ] MLRP F2 [π ] MLRP F3 .
(ii) There exists λ ∈ [0, 1], F − , and F + , with F + MLRP F − and such that F2B [λ] = λF + +
(1 − λ) F − and F3B [λ] = (1 − λ) F + + λF − . In other words, with probability λ project 2 has larger
perks (in the MLRP sense) and with the complement probability it is project 3 (i.e. λ parametrizes
the congruence of interests between the board and the manager). The manager always has the
option of not pursuing any project.
Neither the board nor the manager are able to distinguish the projects. However, both of them
2
may choose to do research to tell them apart. Specifically, the manager can pay an effort cost e2
which allows him to distinguish the projects with probability e ∈ [0, 1]. Similarly, the board may
2
pay a cost k E2 which allows them to tell the projects apart with probability E, conditional on the
manager being able to distinguish the projects. That is, conditional on a given pair (e, E), the board
learns the projects with probability eE. To be clear, when the manager or the board distinguish
the projects they are able to tell not only whether the project in front of them is 1, 2, or 3, but also
whether F + or F − is the underlying distribution of perks. e and E are chosen simultaneously by

4
both type of agents. You may assume that the large shareholder makes the board’s decision and is
the one who pay the research cost.

1. Describe what the manager does if he does not learn to distinguish projects. What if the
manager learns but the board does not?

2. Write down an expression for the manager’s payoff and the large shareholder’s payoff as
functions of (e, E).

3. Let e ( E, π, λ, α) and E (e, π, λ, α) be the best response functions for both type of agents.

(a) Characterize e ( E, π, λ, α) and describe how the manager’s effort depends on ( E, π, λ, α).
What’s the relationship with the hold-up problem?
(b) Characterize E (e, π, λ, α) and describe how the large shareholder’s effort depend on
(e, π, λ, α). (Hint: Topkis theorem extends to the MLRP order2 )

4. Characterize e∗ (π, λ, α) and E∗ (π, λ, α). Describe how these functions behave as we change
(π, λ, α).

5. From an ex-ante point of view, what is the optimal level α that the large shareholder should
choose. Explain the different forces at play.

6. Which comparative statics do you think would yield empirical predictions that allow for the
sharpest test of this model? Why?

7. Summarize the findings of 3 empirical papers that support and/or contradict the predictions
you discuss in the previous subsection.

4 Activism through Exit (25 points)

It is usually argued that activism is costly and that investors, who are subject to the free-rider prob-
lem, may decide to engage by threatening to leave the firm, rather than engaging in activism activ-
ity, if the management does not act according to their objectives. Selling shares, the argument goes,
depresses market prices which negatively affect the manager’s compensation. In this question, we
explore whether these threats are a credible mechanism to impose discipline on management.
There are three periods, t ∈ {0, 1, 2}. In period 0, the manager of a corporation has the possi-
bility of taking an action that destroys the firm’s future cashflows but that induces private benefits.
Specifically, the cashflows of the firm are represented by v ∈ R+ . If the manager takes the action
(does not take the action), which we denote by a = 1 (a = 0), the firm’s cashflows are reduced by
 
δ ∼ F (·) over 0, δ̄ . The manager then earns a (deterministic) private benefit β > 0.
2 Instead ∂2 f ∂f
of using that a function f satisfies supermodularity≥ 0, you need to prove that ∂x increases as the
∂x∂y
distribution y improves in the FOSD (and hence MLRP) sense. If you use this result you need to prove it for full credit.

5
 The firms is owned by two types of risk-neutral investors: a large investor, denoted by L, and
a continuum of small investors. We assume that δ is only observed by the manager while β is
common knowledge. L is the only investor to observe the manager’s decision a ∈ {0, 1}.
In period 1, L may trade her shares with a competitive market at an endogenous price P1 . During
this period, L may suffer a liquidity shock, which transpires with probability θ ∈ (0, 1), that forces
her to sell all her shares. In the absence of the shock, L optimally chooses whether to sell or not.
In period 2, the market learns the manager’s action and the damage δ, and price the corpora-
tion’s shares competitively. The manager’s payoff is contingent on market valuation and is given
by ω1 P1 + ω2 P2 .

1. Show that, in the absence of L, the manager takes the action only if δ < β/ω2 .

2. Show that, in the presence of L, the manager takes the action only if

β − (1 − θ ) ω1 ( ES − ENS ) − ω2 δ > 0,

where ES and ENS represent the expected value of aδ conditional on L selling her shares and
not selling her shares, respectively.

3. Interpret the term − (1 − θ ) ω1 ( ES − ENS ).

4. Argue that, in any equilibria, it must be the case that the manager follows a cut-off strategy of
the form a (δ) = 1 {δ ≤ x } .
 
5. Suppose that the manager chooses a strategy a (δ) = 1 {δ ≤ x } for some x ∈ 0, δ̄ . Show that

E [δ|δ ≤ x ] F ( x )
ES ( x ) = .
θ + (1 − θ ) F ( x )

6. Describe the unique equilibrium of the game. Show uniqueness.

7. What happens to the equilibrium threshold x ∗ as θ increases? Explain the intuition.

8. Suppose next that δ > 0 represents a benefit to shareholders and that β > 0 represents a cost
the manager has to suffer to induce δ. Show that the manager takes the costly action only if

β − (1 − θ ) ω1 ( ES − ENS ) − ω2 δ < 0.

9. Characterize the unique equilibrium of the game and show uniqueness.

10. Show that the new equilibrium threshold x ∗∗ is smaller than x ∗ .

6
5 Challenging the "Control vs Liquidity" View (20 points)

A firm has future cashflows given by ṽ = L under the status quo and ṽ = H if a large investor F
monitors, with L < H. The monitoring cost is c < H − L. In order for monitoring to be effective,
F needs to have a fraction at least µ ∈ (0, 1) of the shares. There is a mass 1 of small investors
with shares normalized to 1. With probability 1/2, small investors suffer a liquidity shock which
forces a fraction φ of them to sell their shares. F is not subject to the liquidity shock. There is a
market maker that observes the order flow y and sets P1 = E [ṽ|y]. The timing of the game is as
follows. In period 0, F buys α shares from small investors at an equilibrium price P0 . In period 1,
F chooses the monitoring probability q and the number of shares x B to buy if monitoring, and the
number of shares xS to sell if not monitoring. Whether F monitors is F 0 s private information. F thus
chooses (q, x B , xS ).3 Small investors then may suffer the liquidity shock, in which case they sell a
mass (1 − α) φ of the shares. Finally the market maker observes y and sets price P1 .
1. Fix α ∈ (0, 1). We construct a (continuation) equilibrium where F confounds the market maker
by choosing x B and xS so that the latter cannot tell apart when F buys and small investors sell
from the situation where F sells but small investors are not hit by the liquidity shock. That is,
x B − φ (1 − α) = − xS . Suppose further that x B = xS .

(a) Find P1 as a function of whether F buys or sell, and whether there is or not a liquidity
shock. Compute E [ P1 | F buys] and E [ P1 | F sells].
(b) Write down F’s payoff when "monitoring and buying" and when "not monitoring and
selling".
(c) Find the value of q∗ that makes F indifferent between the two strategies. How does q∗
behave when (i) c increases, (ii) when α increases, and (iii) when φ increases? Interpret
these MCS.

2. We now find the optimal choice of α from the perspective of period 0.

(a) Compute the expected trade profit that F obtains, from the perspective of period 0, for
a given value of α. Hint: The expected trade profit is given by qx B ( H − E [ P1 | F buys]) +
(1 − q) xS ( L − E [ P1 | F sells]).
(b) Explain why
φ
q (1 − q ) ( H − L ) .
P0 = qH + (1 − q) L −
2
(c) Compute the expected profit that F obtains, from the perspective of period 0, for a given
value of α.
(d) Obtain α∗ , the amount of shares that maximize F’s payoff. How does q∗ as we increase
φ? Discuss why this result is somewhat surprising.
3 Obviously, if F does not monitor, then she does not buy as the market price will be larger than L. Similarly, when F
monitors, she does not sell as the price is smaller than H.

7
6 Information Orders (15 points)

Suppose that the state of nature is captured by ω ∈ Ω ⊂ R. A decision maker has a prior H ∈ ∆Ω
and observes a signal x ∈ X ⊂ R, with F : Ω × X → [0, 1] capturing the joint realization of (ω, x).
She makes a decision a ∈ A to maximize her utility u (ω, a) ∈ R. Let
Z Z 
V ( F, u) ≡ max u (ω, a) dF (ω | x ) dFX ( x ) ,
X a∈ A Ω

and define U ND as the set of all utility functions u satisfying that, for any a0 > a, u (ω, a0 ) − a (ω, a)
is nondecreasing.

1. Consider two signals (i.e., joint distributions), F, G : Ω × X → [0, 1]. Prove that V ( F, u) ≥
V ( G, u) for all u ∈ U ND , if, and only if,

F (·| FX ( x) > z)  ND G (·| GX ( x > z)) , ∀z ∈ [0, 1] ,

where  ND represents the following ordering: We say that P  ND Q if, for any u ∈ U ND ,
Z Z
u ω, a0 − u (ω, a) dQ (ω ) ≥ 0 ⇒ u ω, a0 − u (ω, a) dP (ω ) ≥ 0.





Ω Ω

2. [Optional] Next, consider two experiments F 00 ( x |ω ) and F 0 ( x |ω ). Let F00H and F0H be the in-
duced joint distributions when pairing each experiment with the prior H. We say that F 00 is
Lehmann more informative than F 0 about ω, which we write as F 00  F 0 , if for any x,F 00−1 ( F 0 ( x |ω ) |ω )
is nondecreasing in ω. Prove that, if F 00  F 0 , then for any supermodular function α (ω, x ),
Z Z
α (ω, x ) dF00H ( x, ω ) ≥ α (ω, x ) dF0H ( x, ω ) .
X ×Ω X ×Ω

3. Show that the former result implies that then, the marginal distributions on dimension x must
also coincide, i.e., margx F00H = margx F0H . Conclude about the appeal of using the Lehmann
order to rank informativeness.

4. Let z00 ≡ EF00H [ x|ω] and z0 ≡ EF0H [ x|ω], and denote by Q00 and Q0 the respective cumulative
distribution functions of z00 and z0 , respectively. Then, for any convex function γ : R+ → R,
Z Z
γ (z) dQ00 (z) ≥ γ (z) dQ0 (z) .

In other words, z00 dominates z0 in the convex order, which we write as EF00H [ x|ω] cvx
EF0H [ x|ω].