Nearsighted Justice


w o r k i n g

p a p e r

F E D E R A L  R E S E R V E  B A N K  O F  C L E V E L A N D

03  04

Nearsighted Justice

by Dan Bernhardt and Ed Nosal



Working papers of the Federal Reserve Bank of Cleveland are preliminary materials
circulated to stimulate discussion and critical comment on research in progress. They may not
have been subject to the formal editorial review accorded official Federal Reserve Bank of
Cleveland publications. The views stated herein are those of the authors and are not necessarily
those of the Federal Reserve Bank of Cleveland or of the Board of Governors of the Federal
Reserve System.

Working papers are now available electronically through the Cleveland Fed’s site on the World
Wide Web: www.clevelandfed.org.



Dan Bernhardt is at the University of Illinois.  Ed Nosal is at the Federal Reserve Bank of Cleveland and
may be reached at ed.nosal@clev.frb.org or (216) 579-2021. The authors are grateful to the SSHRC for
financial support.  Much of the initial work was done when Nosal was visiting the Department of
Economics at the National University of Singapore, when Bernhardt was at Queen’s University in
Canada.  Both authors appreciate the research environments provided.  The authors thank a referee, the
editor, Ron Giammarino, Burton Hollifield, Charlie Kahn and Steeve Mongrain for helpful comments.

Working Paper 03-04 June 2003
 

Nearsighted Justice

by Dan Bernhardt and Ed Nosal

Chapter 11 structures complex negotiations between creditors and debtors that are overseen by a
bankruptcy court. This paper identifies conditions under which it is optimal for the court to sometimes err
in determining whether a firm should be liquidated. Such errors can affect the optimal action choices by
both good and bad entrepreneurs. We first characterize the optimal error rate without renegotiation,
providing conditions under which it is optimal for the court both to sometimes mistakenly liquidate “good
firms,” but not “bad firms.” When creditors and debtors can renegotiate to circumvent an error-riven court
and creditors have all of the bargaining power, we show that for a broad class of action choices, a blind
court—one that ignores all information and hence is equally likely to liquidate a good firm as a bad one—
is optimal. For another class of action choices, the optimal court design places the burden of proof on the
entrepreneur. The robust feature is that in the optimal court design, the court sometimes errs in
determining whether a firm should be liquidated.

JEL Classification: G33, D80
Key Words: bankruptcy judge, moral hazard, adverse selection





“It may not be optimal for justice to be blind, but it can help if she’s near-sighted.”

“If parties can negotiate out of the way of blind justice, blind may be best.”

1 Introduction

Chapter 11 structures complex negotiations between creditors and debtors that are overseen by

a bankruptcy court. It is often unclear whether it is socially optimal for a bankrupt firm to be

re-organized as a continuing entity, or whether it should be liquidated and the proceeds from

liquidation distributed. If creditors and debtors cannot reach a settlement, then the bankruptcy

court may impose one. This paper asks the following questions: Is it always optimal for the court to

make the ‘right’ decision? If liquidation is efficient, should the court always mandate liquidation?

If the bankrupt firm could be a profitable entity, should the court always allow it to re-structure

under the existing management and continue? Or is it optimal for the court to sometimes make

mistakes? How does the possibility of negotiated settlements affect the optimal design of the court?

At one level the answers to these questions are straightforward. In a world where more accurate

appraisals of a firm’s quality are more costly, the optimal allocation by a court of resources to

evaluation of bankrupt firms will always reflect trade-offs between the marginal costs of better

appraisals and the marginal benefit of decision-making based on more accurate appraisals. In this

paper, we show that a court should also consider how the probability that it mis-identifies the

quality of a bankrupt firm affects the ex ante behavior of management, and hence the probability

that the firm becomes bankrupt.

We develop a simple model that abstracts completely from the increasing marginal costs of

more accurate appraisals in order to focus on how the probabilities that the court errs affect both

the actions taken by management and outcomes. In a sparsely-specified model of entrepreneurial

financewe showthat itmaybeoptimal for thecourt toerroccasionally. Weconsideranenvironment

in which some entrepreneurs are better than others, and their skills are private information. It is

efficient to liquidate an entrepreneur if and only if he is bad.

Entrepreneurs also cannot be trusted. An entrepreneur can take hidden actions that affect (a)

the probability that the firm cannot meet its loan obligations, (b) expected period entrepreneurial

profits, and (c) expected period project revenues. We consider a broad variety of action interpre-

tations. For example, the action choices could be investment choices; some investments might be

risky negative NPV investments that pay off in non-bankruptcy states, but increase the probability

of low revenue, bankruptcy outcomes. Alternatively, the action choices could correspond to effort,

the choice of whether to steal from the firm, the choice of whether to have a fire sale of inventory

1



that raises current revenues at the expense of lower future revenues, and so on.

There are at least three issues that might concern the court:

1. To identify which entrepreneurs are good and which ones are bad, so that bad firms, but not

good ones can be liquidated.

2. To discourage socially inefficient action choices prior to bankruptcy.

3. To internalize the effect of the court design on the incentives of entrepreneurs and creditors

to reach a settlement rather than leave the outcome to be decided by an error-prone court;

and the consequences for ex ante actions.

Unfortunately, these goals may conflict. One might like to use the threat of liquidation to dis-

courage a good entrepreneur from taking negative NPV gambles or under-exerting effort. However,

if a good entrepreneur is always identified as such by the bankruptcy court, then liquidation is not

time consistent and, the entrepreneur realizing this, may take actions that reduce total surplus.

Conversely, if a bad entrepreneur is always identified as such by the bankruptcy court, he

may want to take actions that reduce the probability of entering bankruptcy. For example, this

aversion to bankruptcy may cause bad entrepreneurs to over-exert, or to engage in fire sales that

inefficiently increase current revenues, reducing the probability of bankruptcy in the current period

at the expense of future revenues. It may be socially desirable to design the court so that it makes

some mistakes, thereby reducing a bad entrepreneur’s aversion to bankruptcy.

Thus, for both good entrepreneurs and bad, it may be optimal for the court to err occasionally:

1. The threat of mistakenly liquidating a good firm maybe sufficient to keep it from, for example,

taking negative NPV gambles that raise the probability of entering bankruptcy.

2. The possibility that a bad firm is not identified by the court as such, may encourage a bad

entrepreneur to take actions that land the firm in bankruptcy with a higher probability. The

direct total surplus associated with the action that lands the firm in bankruptcy can be either

positive (if it discourages fire sales), or negative (if it causes a bad entrepreneur to under-

exert, or to choose excessively safe, lower NPV projects). Even if the direct surplus effect of

the court design on action choice is negative, it may still be optimal for the court to err if, as

a result, more bad entrepreneurs are liquidated at early dates.

We characterize the optimal rate at which the court should make mistakes, both when creditors

and entrepreneurs cannot bargain prior to the court’s decision, and when they may be able to

2



reach a settlement that would obviate the need for the court to make a decision. The analyses

with and without renegotiation provide bounds on the benefits of a court that makes mistakes.

When renegotiation is infeasible, any liquidating mistakes by the court must be incurred, making

an error-riven court less attractive, although not necessarily sub-optimal.

In contrast, the possibility of renegotiation allows the court to be designed as a potentially

costly back-up threat, but one that, in equilibrium, need not be used: the threat that the court

may make the wrong decision may be enough to encourage the parties to reach a settlement rather

than take their chances with an unreliable court. The analysis with renegotiation is subtle, because

the creditor does not know the entrepreneur’s type. The creditor must design payments that

discourage a good entrepreneur from passing himself off as bad in order to receive a payment for

liquidation, rather than make a payment to avoid going to court; and discourage a bad entrepreneur

from passing himself off as good in order to circumvent the court and continue to operate.

Stark results obtain when an entrepreneur and creditor can always renegotiate to circumvent an

error-riven court. We characterize the optimal court design for two broad classes of action choices,

those where it is optimal to

1. Discourage good entrepreneurs from taking actions that raise bankruptcy probabilities (theft,

perk consumption, risky NPV investments, shirking), while discouraging bad entrepreneurs

from taking actions that lower bankruptcy probabilities (fire sales, working too hard, exces-

sively safe low NPV investments).

2. Discourage both good entrepreneurs and bad from taking actions that raise bankruptcy prob-

abilities.

For the first class of economies, a simple blind court design always dominates a court that never

errs. Indeed, a blind court can sometimes implement the social optimum. That is, not only should

the court always be near-sighted and err in the identification of entrepreneurs, but it should ignore

all information, essentially making liquidation decisions on the basis of a fair coin flip. Even when

a blind court cannot always induce ex ante entrepreneurs to take optimal actions, as long as it

is optimal to encourage bad entrepreneurs to take actions that increase the chance that they are

liquidated, then a blind court dominates a court that makes no errors. For the second class of

economies, the social optimum can be implemented by placing a very stringent burden of proof on

entrepreneurs. The burden of proof is stringent in the sense that, although a bad entrepreneur is

never able to pass himself off as being good, a good entrepreneur may sometimes fail to convince

that court that he is, in fact, good.

3



While the optimal court design differs depending on the nature of the action choice, the key

features that they share are (i) the optimal court designs are simple, and easily implemented, and

(ii) the optimal court design is one in which the court errs.

Although we pose the model in the context of Chapter 11 bankruptcy, the problem that we

analyze is more fundamental. The general formulation is a dynamic principal-agent costly state

verificationenvironment inwhichtheagent’s type isunknown and theprincipal and agent’s interests

are notaligned overaction choice. Theprincipal must choose thequalityof anevaluation technology

to employ to verify the agent’s type, and the outcome of the evaluation affects whether the agent

is retained. Sequential rationality implies that only the agent’s type is relevant for continuation

decisions, but type-contingent continuation probabilities affect the agent’s decisions about which

action to take.

We could have presented our analysis in the context of shareholders in a firm who must decide

whether to incur the costs of investigating whether or not to replace management. Shareholders

do not want to incur the costs of investigating good managers, but they do want to discourage

good managers from taking actions that lower the firm value (build personal empires, steal, etc.);

shareholders may also want bad managers to take actions that lead to signals (lower immediate

profits) indicating that the managers should be investigated and replaced. Shareholders may op-

timally choose a noisy evaluation technology for managers, both for reasons of costs and because

the possibility of mistakes may provide the right incentives for management: The noisy technology

may help deter inefficient action choices by good managers, and help identify bad managers.

So, too, the analysis could have been posed in a venture capital/equity finance context. Here,

ongoing projects require a second-period injection of capital, and a financier must decide whether

to provide the capital. The financier will set a revenue standard such that if revenues exceed that

level, it will provide financing; while for lesser levels, the financier will evaluate the entrepreneur,

and depending on the (noisy) signal received decide whether to provide the additional capital. Ex

ante, the financier commits to the quality of the data on an entrepreneur that he will collect.

Our paper is related to least four different literatures: the law and economics literature on

optimal court design; the law and economics literature on contracting in the face of an imperfect

and/or costly court; the bankruptcy literature; and the literature on the hold-up problem. We will

discuss, in turn, how our paper contributes to each of these areas.

Our paper contributes to the general law and economics literature on the optimal design of the

court. Existing research on court design focuses on how best to design the court in order to elicit

information from informed parties. This research builds on the “games of persuasion” literature

introduced by Milgrom (1981) and Milgrom and Roberts (1986). Here, an interested party who

4



has private information attempts to influence an asymmetrically-informed decision maker. In law

contexts, the interested parties are a plaintiff and a defendant in a trial and the decision maker is a

judge. The judge, who wants to make the socially optimal decision, has to rule for one party, and

his decision is based on the information that he receives. The issues that arise are:

• Should a trial be based on an adversarial system, where the plaintiff and defendant present
verifiable information to the judge and the judge makes his decision on the information

presented; or should the judge try to uncover the information himself (Shin (1998))?

• Should a judge restrict the information that can be presented (Fishman and Hagerty (1990))?

• Which interested party must provide the evidence in order to “win” the trial; that is, who
should bear the burden of proof (Hay and Spier (1997), Shin (1994), Sobel (1985))?

In this literature, the answers to these questions ultimately depend upon the quality of the

signal that the decision maker receives. For example, if restricting the information that the decision

maker receives increases the quality or the informativeness of the signal, then information should

be restricted, as this raises the probability of a correct decision. Most starkly, if, as in our model,

the decision maker has access to a costless technology that perfectly reveals the hidden information,

then in virtually all games of persuasion, the decision maker would use this technology.

What distinguishes our analysis from this literature is that in these games of persuasion the

analysis is ex post. That is, the actions that landed a plaintiff and defendant in front of a judge are

not examined: the analysis begins with the plaintiff and defendant in trial. It follows immediately

that better information is always preferred. In sharp contrast, our analysis explicitly models the

ex ante actions that could lead to a bankruptcy trial. Now the optimal signal that the judge

receives must strike a balance between ex ante and ex post considerations. Were the standard

literature to incorporate ex ante decision making into their persuasion games, then our analysis

strongly suggests that the most informative signal would be sub-optimal. One can interpret our

paper as contributing to this literature by modeling the ex ante decisions of the interested parties

and examining the implications for the optimal signal strength that the decision maker receives.

Our paper is also related to the law and economics literature on contracting in the face of

an imperfect or costly court. For example, Spier (1994) examines a situation where a victim can

experience either a mild or severe accident and the injurer can undertake costly precautions that can

reduce the probability and severity of an accident. If a court can observe the level of harm at “low”

cost, then it is optimal to set higher damage awards for higher levels of harm. In this situation, the

potential injurer invests in the first-best level of precaution. If, instead, it is sufficiently costly for

5



court to observeharm, thecourt sets aconstant damage award andall victims and injurers settle out

of court. Although the court uses no resources, the potential injurer under-invests in precaution. As

in our paper, court behavior influences ex ante behavior. But, in important contrast, the first-best

outcome would be achieved if the court costlessly observed the truth; in our model the first-best

is unattainable if the court observes the truth. Grossman and Katz (1983) consider a situation

where a court makes mistakes and a risk-averse defendant is either innocent or guilty. Here, plea

bargains–which are effective renegotiations–are optimal because they separate the innocent from

the guilty (guilty parties accept the plea bargain and innocent parties go to court). Thus, as in our

paper, allowing parties to renegotiate outside of court raises welfare; but again in Grossman and

Katz’ environment it is always optimal for the court not to err.

Our paper is closely related to the bankruptcy literature that explores how the optimal design

of bankruptcy law may introduce ex post distortions in resource allocations in order to alter the

ex ante behavior of management. Berkovitch et al. (1998) argues that “The structured bargaining

imposed by an optimal bankruptcy law provides the entrepreneur with optimal ex ante incentives

by placing him in a superior bargaining position in the negotiations triggered by financial distress.

The bankruptcy law serves as a commitment device and is required to enforce this re-balancing of

the relative bargaining strengths of the claimants ex post.” Giammarino and Nosal (1994) provide

conditions where Chapter 11, as an option, can increase efficiency by providing management with

additional ex post bargaining strength. The additional bargaining strength implies that manage-

ment now has an incentive to take actions that enhance social welfare. Bernhardt and Lu (1998)

model the dynamic features of Chapter 11, exploring how the time allotted management to make

restructuring offers affects both ex post outcomes such as the timing of liquidation and settlement

outcomes, as well as the ex ante effects on such managerial actions as effort and project selection.

Finally, our paper is related to the hold-up literature begun by Williamson (1985). In a typical

hold-up problem model, agents take actions, observe outcomes, possibly renegotiate contracts, or

go to court and make final exchanges. In the standard model, the court costlessly enforces only

those aspects of the contract that it can verify and some key attribute is unverifiable. When a

seller’s action is a costly investment that enhances the value of its good to the buyer, Che and

Hausch (1999) find that if the court cannot verify the investment, then even if the amount traded

is verifiable, under-investment necessarily results. However, in a recent working paper, Willingston

(2002) shows that when the court is “blind,” in the sense that it randomly decides whether the seller

breached, then it is possible to devise a contract in which the seller makes the optimal investment.

This result mirrors ours in number of ways. Because the court makes errors, the parties prefer

to renegotiate the original contract to eliminate ex post inefficiencies, and it is the fact the court

makes errors that promotes optimal ex ante behavior.

6



The next section presents the basic model. We identify the costs and benefits associated with

the courts making mistakes and develop the intuition for why courts may optimally choose to err.

By imposing only weak structure on the lending contract (in particular, the lending environment

need not be competitive), the nature of the possible action choices that an entrepreneur could take

and how they affect bankruptcy probabilities, the section provides insights into the nature of the

liquidation mistakes that the courts might make. The intuition underlying the analysis is made

more transparent by giving the court complete flexibility in the choice of monitoring technology

for each type of entrepreneur. In practice, however, the court generally does not have this kind

of flexibility in choosing type contingent monitoring technologies. We, therefore, consider how the

results are altered when the court’s choices are more constrained. We do this for two broad sets of

economies: In section 3 the action choice corresponds either to undertaking perk consumption or

having a fire sale, while in section 4 the action corresponds to theft or shirking. Section 5 concludes.

2 Unconstrained court evaluation technologies

A single risk-neutral entrepreneur requires external funding of one unit of capital to finance an ex

ante positive NPV project. The potentially two-period-lived project is financed by a risk-neutral

investor through the financing contract C. Neither entrepreneur nor investor discount payoffs. The

contract, which we describe below, specifies both the payments to the investor as a function of

observed project revenues, as well as a revenue cutoff such that for lower revenues the entrepreneur

is in default, cannot “meet” its contractual obligations and “enters bankruptcy.”

The project offers random payoffs in two periods. The project is either good or bad. The timing

of events is such that the entrepreneur does not know the project type when the financial contract

is negotiated with the investor, but does learn the type prior to taking any actions. The project’s

(or entrepreneur’s) type, ∈ {g,b}, is private information to the entrepreneur. Let p( ) be the
probability that the entrepreneur is type .

In the first period, entrepreneur chooses an action, a . Both the entrepreneur’s type, , and

his action choice, a , may affect:

• Total expected period project revenues, Rt( ,a ), t = 1,2.

• Entrepreneurial expected period profits, πt( ,a ), t = 1,2.

• The probability B( ,a ) that a firm defaults on its contractual obligations at date 1.

• The value z( ,a ) of a firm that is liquidated at the end of period 1.

7



The expected period payment to creditors is just the difference between expected period project

revenues and expected entrepreneurial profits: Rt( ,a )−πt( ,a ).
To simplify the presentation, we restrict attention to two action choices for each type of entre-

preneur, a ∈ {a1,a2}. We use “∆” to capture the difference in the variables from taking action
a1 instead of action a2. Thus, ∆Rt( ) ≡ Rt( ,a1)−Rt( ,a2) is the difference in expected period t
revenues from taking action a1 rather than a2, and so on.

The key assumption that we make is that if entrepreneur takes action a1 then he is more likely

to default and end up in bankruptcy than if he takes action a2:

A1: ∆B( ) ≡ B( ,a1)−B( ,a2) > 0.
Assumption A1 reduces to a normalization of the action choice if action choices have the same

signed impact on bankruptcy for both entrepreneur types.

Our formulation admits a broad variety of interpretations for the actions. For example, an

entrepreneur could be choosing whether or not to invest in a risky, negative NPV project that

raises expected period 1 profits at the expense of expected total period 1 project revenues and an

increased likelihood of bankruptcy. Then action a1 would correspond to making the risky, negative

NPV investment. Alternatively, a1 could correspond to stealing from period 1 investments, which

again raises expected profits at the expense of reducing expected period 1 project revenues and

raising the likelihood of bankruptcy. Another possibility is that the entrepreneur could be choosing

whether or not to have a “fire sale” of existing inventory at bargain prices. In this case action

a2 would correspond to having a fire sale, since a fire sale raises period 1 revenues and period 1

profits, and reduces the likelihood of immediate bankruptcy. However, a fire sale reduces future

revenues, profits and liquidation values. With a slight modification of the model (to allow for

non-pecuniary payoffs), action a1 could correspond to shirking (working less hard), while action a2

could correspond to working hard (working harder).

Indeed, the actions contemplated by the two types of entrepreneurs could even differ. For

example, suppose that the bad entrepreneur is in financial distress, but the good entrepreneur is

not. The bad entrepreneur might be choosing between a fire sale (a2b) and no fire sale (a
1
b), while

the good entrepreneur is choosing whether to steal some of the firm’s first period cash flows (a1g) or

not (a2g). Depending on the action interpretation, it could be socially optimal for the entrepreneur

to take the high default action a1 (if action a2 represents a fire sale, or working excessively hard);

or the low default action a2 (if action a1 represents a negative NPV gamble, theft or shirking).

If a project is liquidated after the first period, then it pays z( ,a ) to creditors and nothing to

the entrepreneur. We assume that it is efficient to liquidate a project run by a bad entrepreneur,

8



but inefficient to liquidate a project run by a good entrepreneur:

A2: R2(g,ag) > z(g,ag) and R2(b,ab) < z(b,ab).

We also assume that the potential liquidation value is not observed by the court when making

decisions, and that if renegotiation is feasible, then terms cannot depend on z( ,a ). These as-

sumptions simply ensure that the liquidation value does not reveal type in our sparsely specified

environment. We will also discuss the impact of intermediate divisions between the entrepreneur

and creditor of the liquidation proceeds later.

Observed first period project revenues are implicit in the bankruptcy probability specification.

We do not directly specify realized first period project revenues in order to be consistent with a

variety of possible forms of action choice, including, for example, theft–which leads to a distinction

between realized and observed first period revenues–and shirking or risky investments–where

realized and observed first period revenues correspond.

To be consistent with a variety of economic environments we impose little structure on the

financing contract between entrepreneur and investor. In equilibrium, the investor must expect to

get back at least his one unit of capital investment from the firm. The analysis is consistent with

standard limited liability loan contracts, but also holds for more general financing contracts.

If the firm “meets” its first period contractual obligations, then the entrepreneur retains control

of the assets. If the firm defaults, then at a cost of c, that is incurred by the creditor, the bankruptcy

court evaluates the firm’s future prospects. The court then determines whether (a) the firm should

be liquidated; or (b) the entrepreneur should retain control and continue to operate the firm in

the second period. We assume that the bankruptcy costs are not so high, as to cease to make

liquidation valuable: z( ,a ) > c > 0.

Initially, we preclude the possibility that the entrepreneur and creditors can renegotiate a set-

tlement to circumvent the bankruptcy court. As a result, the entrepreneur retains control of the

firm in the second period if and only if either (a) he meets his first period contractual obligations,

or (b) he fails to meet his first period contractual obligations, but the court chooses not to liquidate

the entrepreneur.

We model the design of the bankruptcy court as a pair (γ(g),γ(b)), where γ( ) is the probability

that the court concludes that the project has a negative NPV. If the entrepreneur defaults in period

1 and the court concludes that the project has a negative NPV, then the firm is liquidated and the

liquidated value, z( ,a ) > 0, is transferred to creditors. Assumption A2 implies that identifying

whether a project has a negative NPV amounts to identifying the entrepreneur’s type. Hence,

γ( ) can be interpreted as the probability that a type firm is liquidated if it defaults in the

9



first period. Society (or the court) is benevolent, choosing (γ(g),γ(b)) to maximize ex ante total

expected revenues net of bankruptcy costs.

We interpret the court design (γ(g),γ(b)), as the choice of a “monitoring technology,” where the

technology attempts to assess whether the project that has gone into default has a positive NPV.

This interpretation is sufficiently important to discuss in detail. The accuracy of the monitoring

technology’s assessment depends on the information provided. If, for example, all relevant infor-

mation regarding the project’s value is available and provided, then, in principle, the monitoring

technology could determine the project’s type with complete accuracy; if all relevant information

is not provided, then monitoring errors may be made. To highlight the intuition, we first place no

restrictions on the evaluation technologies to which the court has access.

In practice, owing to “natural” informational asymmetries that exist between creditors and

debtors, the court may be unable to gather all relevant information. As a result, a court must

sometimes make undesigned mistakes, imposing lower bounds on the error probabilities. Our

analysis is best interpreted within the context of asking how the court should choose its design–

the rules of evidence that restrict the information which creditors and firms can present and hence

the information to which the court has access–in order to structure the probability of errors in

such a way as to enhance social welfare. Our analysis can be interpreted as providing a rationale

for the imposition of these kinds of rules: such rules cause courts to make (more) errors and we

will show that increasing errors is often optimal.

To illustrate how the court might design the rules of evidence so as to generate the “right”

(γ(g),γ(b)), suppose that it is optimal for the court to liquidate all bad firms, but to err occa-

sionally in the identification of good firms. As a result, some good entrepreneurs who default are

(unfortunately) liquidated, but this possibility of liquidation may cause a good entrepreneur to

make better decisions when running the firm. The court could implement such rules of evidence

by placing the entire burden of proof on the entrepreneur: the firm is liquidated unless the entre-

preneur can prove that he is “good.” Since bad entrepreneurs can never do so, the choice of the

error-making probability amounts to restricting the rules of evidence about what evidence a good

entrepreneur can provide to attain the “right” error-making probability.

The court’s unrestricted evaluation technology provides a best case scenario for an error-riven

court. We later consider how outcomes are affected when the court has access only to a more

limited set of feasible monitoring technologies.

Let Π( ,a ,γ( )) be the expected lifetime entrepreneurial profits of a type entrepreneur who

10



takes action a ∈ {a1,a2} given γ( ) and C:

Π( ,a ,γ( )) = [π1( ,a )+(1−B( ,a )γ( ))π2( ,a )].

A type entrepreneur will choose the high default probability action, a = a1, if it is in his best

interest to do so, i.e. if Π( ,γ( ),a1)−Π( ,γ( ),a2) > 0. After some manipulation, we see that a
type entrepreneur will choose the high default action if

γ( ) <
∆π1( )+∆π2( )

B( ,a1)π2( ,a1)−B( ,a2)π2( ,a2)
≡ γ∗( ). (1)

Were the court’s liquidation probability to exceed γ∗( ), then the entrepreneur would choose the

low probability default action, a2; were γ( ) < γ∗( ), then the entrepreneur would always choose

the high probability default action a1. γ∗( ) can be loosely interpreted as the entrepreneur’s

private benefit-cost ratio associated with taking the high default probability action. The numerator

represents the potential increase in lifetime benefits to the entrepreneur from taking the high default

probability action, while the denominator represents the increased second period loss due to the

reduced probability of receiving a payoff in period 2 associated with taking the high default action.

Hence, given liquidation probability γ( ), a type entrepreneur will choose action

a∗ =
a1 if γ( ) ≤ γ∗( )
a2 if γ( ) > γ∗( )

∈ {g,b}. (2)

2.1 Optimal choices of γ( )

2.1.1 Good Entrepreneurs

To discourage a good entrepreneur from taking the high default probability action, i.e., from taking

the action ag = a
1
g, equation (1) implies that the court must choose a sufficiently high liquidation

probability γ(g) ≥ γ∗(g). It follows immediately that,
Lemma 1: If the court chooses to discourage the high default probability action by a good entrepre-

neur, then it optimally liquidates a good firm that enters bankruptcy with probability γ(g) = γ∗(g) .

If, instead, the court chooses not to discourage the high default action by a good entrepreneur, then

it optimally chooses γ(g) = 0.

Setting γ(g) > γ∗(g), i.e. mis-identifying a good firm as bad with a probability exceeding γ∗(g),

discourages a good entrepreneur from taking the high default probability action, but does so at an

unnecessarily higher cost–an excessively high proportion of good entrepreneurs is liquidated. The

associated total net expected revenues from discouraging the high default action are:

R1(g,a
2
g)+[1−B(g,a2g)]R2(g,a2g)+B(g,a2g)[(1−γ∗(g))R2(g,a2g)+ γ∗(g)z(g,a2g)−c]. (3)

11



If courts choose to encourage the high default probability action ag = a
1
g, then the liquidation

probability must be set so that γ(g) < γ∗(g). Clearly, in such a case it is optimal to set γ(g) = 0

since it is not optimal to liquidate good firms. In this case, total net expected revenues are

R1(g,a
1
g)+R2(g,a

1
g)−B(g,a1g)c. (4)

Proposition 1: It is optimal to encourage the low default probability action for good entrepreneurs,

ag = a
2
g, if and only if

−∆R1(g)−∆R2(g)+∆B(g)c > (R2(g,a2g)−z(g,a2g))B(g,a2g)γ∗(g). (5)

The left-hand side reflects thepotential gains fromdiscouraging a goodentrepreneur fromtaking

the high default probability action: expected lifetime project revenues are increased and the firm

is less likely to incur investigation costs in bankruptcy. The right-hand side reflects the costs: to

discourage a good entrepreneur from taking the high default action, the court must occasionally

mistakenly identifying bankrupt good entrepreneurs as bad, and liquidate them.

It is useful to rewrite equation (5) in terms of γ∗(g). Doing so reveals that it is optimal to

discourage a good entrepreneur from taking the high default probability action if and only if the

court does not have to make too many mistakes in order to induce a good entrepreneur to take

action a2g i.e., if and only if

γ∗(g) = t
∆πt(g)

B(g,a1g)π2(g,a
1
g)−B(g,a2g)π2(g,a2g)

<
− t∆Rt(g)+∆B(g)c

B(g,a2g)(R2(g,a
2
g)−z(g,a2g))

≡ γcourt(g). (6)

The left-hand side of the inequality reflects a good entrepreneur’s private trade-off from undertaking

the high default probability action, while the right-hand side of the inequality reflects the society’s

trade-off from discouraging this action. Good entrepreneurs refrain from taking the high default

action if, relatively speaking, the probability of liquidation is high enough–i.e., if B(g,a1g) is

sufficiently larger than B(g,a2g)–and/or if the potential profit gain from taking the high default

action, t∆πt(g), is small. It is socially optimal to discourage a good entrepreneur from taking

the high default action if the requisite liquidation probability, γ∗(g), is, relatively speaking, “small

enough.” The requisite liquidation probability will be (relatively) small if the changes in lifetime

revenues and changes in bankruptcy rates are large and if the expected costs associated with

liquidating a good entrepreneur, B(g,a2g)(R2(g,a
2
g)−z(g,a2g)), are not too large.

Note that for some interpretations of the action choice, it would never be optimal to design

a court that mis-identifies good entrepreneurs: for some action choice interpretations a good en-

trepreneur’s private interests over action choice are aligned with society’s (as long as bankruptcy

12



costs are not too high). For example, let action a2g correspond to having a fire sale and a
1
g cor-

respond to not having a fire sale. Although a fire sale decreases the probability of default in the

first period, it requires that the entrepreneur inefficiently liquidate resources which implies that

R1(g,a
2
g)+R2(g,a

2
g) < R1(g,a

1
g)+R2(g,a

1
g). Therefore, t∆Rt(g) > 0. But then the right-hand

side of (6) is negative provided that bankruptcy costs are not too high; it is, therefore, optimal for

the court to set γ(g) = 0. Intuitively, the court would not encourage the low default probability fire

sale action since it entails social costs with no offsetting benefits. So, too, if action a2g corresponds

to “working too hard,” then interests are aligned.

In contrast, if action a1g corresponds to theft or a risky negative NPV investment, then action

a1g is socially inefficient in the sense that R1(g,a
1
g) + R2(g,a

1
g) < R1(g,a

2
g) + R2(g,a

2
g). Since

t∆Rt(g) < 0, there are social benefits associated with discouraging the high default (theft)

action. If this benefit is sufficiently great, i.e., if inequality (6) holds, then by its choice of liquidation

probability, the court will encourage a good entrepreneur to take the low default (no theft) action.

These observations highlight that the optimal design of the court will depend crucially on the

nature of action choices that entrepreneurs are likely to undertake that have the biggest impact

on payoffs. In this paper, we do not take a stand on which action choices are most important,

but merely derive the consequences of the way in which action choices affect payoffs for the court’s

optimal design.

2.1.2 Bad Entrepreneurs

Lemma 2: If the court chooses to discourage bad entrepreneurs from taking the high default prob-

ability action, then it always liquidates bad firms that enter bankruptcy. If, instead, the court

chooses to encourage the high default action by bad entrepreneurs, then it liquidates bad firms with

probability γ(b) = γ∗(b).

Any liquidation probability γ(b) ∈ (γ∗(b),1] will discourage bad entrepreneurs from choosing
the high default probability action (e.g., taking a risky negative NPV project, stealing, shirking,

no fire sale). It follows immediately that the court should choose the liquidation probability that

induces ab = a
2
b at the minimum cost, i.e., it should choose γ(b) = 1. That is, given that bad

entrepreneurs do not choose the high default action, it is optimal to identify all bad entrepreneurs

who go bankrupt. In this case, total expected revenues are:

R1(b,a
2
b)+[1−B(b,a2b)]R2(b,a2b)+B(b,a2b)(z(b,a2b)−c). (7)

If, instead, courts wish to encourage bad entrepreneurs to take the high default action, then it is

optimal to set γ(b) = γ∗(b); i.e., it is optimal to select the largest liquidation probability consistent

13



with a bad entrepreneur choosing ab = a
1
b (e.g., preventing a fire sale that shifts cash flows ahead

at the expense of total expected revenues). In this case, total expected revenues are

R1(b,a
1
b)+(1−B(b,a1b))R2(b,a1b)+B(b,a1b)[(1−γ∗(b))R2(b,a1b)+ γ∗(b)z(b,a2b)−c]. (8)

Proposition 2: It is optimal for the court to encourage bad entrepreneurs to take the high default

probability action if and only if

−∆R1(b)−∆R2(b)+∆B(b)c < B(b,a1b)γ∗(b)(z(b,a1b)−R2(b,a1b))−B(b,a2b)(z(b,a2b)−R2(b,a2b)). (9)

The left-hand side of equation (9) reflects the loss incurred resulting from encouraging bad en-

trepreneurs to take the high default probability action. The right-hand side reflects the potential

gains: If γ∗(b)B(b,a1b) > B(b,a
2
b), then more bad entrepreneurs will be identified if they take the

high default action, even though some will slip through because the court makes mistakes in order

to have the opportunity to identify bad entrepreneurs. If the above inequality does not hold, then

it is always optimal to set γ(b) = 1.

Rewriting equation (9) in terms of γ∗(b), reveals that it is optimal to encourage the high default

action if and only if the court does not have to make too many mistakes in order to induce a bad

entrepreneur to take the low default action, i.e., if and only if

γ∗(b) = t
∆πt(b)

B(b,a1b)π2(b,a
1
b)−B(b,a2b)π2(b,a2b)

>
− t∆Rt(b)+∆B(b)c+B(b,a2b)(z(b,a2b)−R2(b,a2b))

B(b,a1b)(z(b,a
1
b)−R2(b,a1b))

≡ γcourt(b). (10)

Again, the right-hand side represents the social benefit-cost ratio from discouraging a bad entre-

preneur from taking the high default probability action, while the left-hand side represents a bad

entrepreneur’s private trade-off for undertaking the high default action. A bad entrepreneur will

take the high default action only if he is sufficiently likely to be mistaken by the court for a good

entrepreneur. Intuitively, encouraging a bad entrepreneur to take the high default action tends to

be optimal if

• The action is attractive to a bad entrepreneur (i.e., his potential profit gain, ∆π1(b)+∆π2(b),
is high, and profits foregone in a liquidation, B(b,a1b)π2(b,a

1
b)−B(b,a2b)π2(b,a2b), are low),

• The social benefits associated with getting a bad entrepreneur to take the low default proba-
bility action, −∆R1(b)−∆R2(b)+∆B(b)c+B(b,a2b)(z(b,a2b)−R2(b,a2b)) are low and the ex-
pected social cost associated with liquidating a bad entrepreneur, B(b,a1b)(z(b,a

1
b −R2(b,a1b))

is high.

14



Comparing γcourt(b) with γcourt(g), we see that there is an additional consideration in the

court design for the bad entrepreneur: The court may want to encourage bad entrepreneurs to

take the high default action in order to increase the probability that bad entrepreneurs go into

bankruptcy, and hence the probability that they are liquidated.

The interests of a bad entrepreneur and society are never aligned over bankruptcy, in the sense

that bad entrepreneurs prefer a lower bankruptcy probability, whereas society prefers a higher bank-

ruptcy probability. Excepting this consideration, however, whether a bad entrepreneur’s interests

are aligned with society’s interests depends on the action interpretation in ways similarly to those

discussed for good entrepreneurs.

3 Confiscation: Perk consumption and Fire Sales

We now focus on specific interpretations of the action choices. In particular, we assume that an

entrepreneur can confiscate some of the firm’s resources. The confiscated resources can be used

either to supplement first period revenues or for the entrepreneur’s personal consumption. In a fire

sale, an entrepreneur confiscates some of the firm’s infrastructure, liquidates it in the market and

then uses the liquidation proceeds to supplement first period revenues. A fire sale raises current

period revenues and profits, and lowers default probabilities at the expense of future period revenues

and profits, as well as total revenues. A bad entrepreneur may have an incentive to have a fire

sale to avoid default. Alternatively, an entrepreneur can consume the proceeds from the liquidated

infrastructure as well as some of the firm’s first period cash flow. As with a fire sale, an entrepreneur

who “consumes perks” reduces future revenues and profits, but doing so reduces rather than raises

first period revenues and default rates. A good entrepreneur might be tempted to engage in perk

consumption since he may be unlikely to be liquidated in default.

For the bad entrepreneur, action a2b corresponds to undertaking a fire sale. A fire sale entails

a socially inefficient liquidation of inventory or infrastructure by the entrepreneur, which means

that R1(b,a
1
b)+z(b,a

1
b) > R1(b,a

2
b) +z(b,a

2
b) and R1(b,a

1
b)+R2(b,a

1
b) > R1(b,a

2
b) +R2(b,a

2
b). A

bad entrepreneur may want to pursue a fire sale in order increase first period expected revenues

and profits, but it is at the expense of second period revenues, profits and liquidation values.

For a good entrepreneur, action a1g corresponds to consuming perks: “stealing” cash flows and/or

infrastructure and consuming them. To the extent that a good entrepreneur must conceal these

activities, real resources are used which means that R1(g,a
2
g) + R2(g,a

2
g) > R1(g,a

1
g) + R2(g,a

1
g)

and R1(g,a
2
g)+z(g,a

2
g) > R1(g,a

1
g)+z(g,a

1
g). A good entrepreneur may want to steal in order to

increase first period profits, albeit at the expense of first period revenues and second period profits,

15



revenues and liquidation values.

Section 2 considered a court that was unconstrained in its signal choices. In practice, the court

may be more limited both by what evidence is feasible to provide and in how it can design signal

qualities through its choice of rules of evidence, etc. In this section we consider how outcomes are

affected when the court has access only to a more limited class of signals. A natural choice for a

restricted class of signals is that where the court is as likely to mis-identify a bad entrepreneur as

good, as it is to mis-identify a good entrepreneur as bad: γ(b) = 1−γ(g). That is, the court makes
“symmetric” errors across different types. Not only might this be viewed as natural choice, but it

turns out that this choice has optimality properties for the types of action choices considered here.

3.1 No Renegotiation

When a court makes errors, both parties have an incentive to renegotiate the existing contract

to circumvent the error-riven court. The gains to renegotiation generally rise when the court

selects its monitoring policy from a more restricted class. To distinguish most sharply how each

feature–a court that errs and renegotiation–affects outcomes, we proceed in two steps. First,

in an environment with no renegotiation, we contrast outcomes when the court must adopt the

symmetric evaluation technology with those where the court faces no restrictions on the evaluation

technology. We then analyze how the possibility of renegotiation affects the outcomes.

Suppose first that γ∗(b) < 1−γ∗(g). Then the optimal error rate is one of

• γ(g) = 0 = 1−γ(b). Here all defaulting entrepreneurs are perfectly identified. However, good
entrepreneurs consume perks and bad entrepreneurs have fire sales.

• γ(g) = γ∗(g) < 1−γ∗(b). Good entrepreneursare induced to refrain fromconsumingperks but
bad entrepreneurs continue to engage in fire sales1 (and hence does not reduce the probability

that he is funded in the second period). Defaulting entrepreneurs of both types are no longer

perfectly identified.

• γ(g) = 1− γ∗(b) > γ∗(g). Good entrepreneurs now refrain from consuming perks, but at a
slightly higher error rate than when the court was unconstrained. As a result, more good

entrepreneurs are liquidated. This higher error rate may be justified since it prevents bad

entrepreneurs from having a fire sale.

Anyothererror rateunnecessarily introducesgreater ratesof error thanoneof thesealternatives,

while inducing the same behavior by entrepreneurs. The key observation is that to choose the error

1Since 1− γ(b) = γ(g) = γ∗(g), then 1− γ(b) < 1 − γ∗(b) or γ(b) > γ∗(b).

16



probability efficiently for one entrepreneur type one must introduce inefficiencies for the other

entrepreneur type. For example, if γ(g) = γ∗(g), then a good type is efficiently induced not to

consume perks, but a bad entrepreneur’s behavior is unaffected by the introduction of court error.

As a result, the bad entrepreneur may be mis-identified and hence not liquidated following a default.

This reduces the value of a error-prone court, perhaps to the point that it is optimal to have a

court that makes no errors.

If, instead γ∗(b) > 1−γ∗(g), then the optimal error rate is one of

• γ(g) = 0 = 1− γ(b). Again, all defaulting entrepreneurs are perfectly identified, but good
entrepreneurs consume perks and bad entrepreneurs have fire sales.

• γ(g) = 1−γ∗(b) < γ∗(g). Bad entrepreneurs cease to conduct fire sales. Good entrepreneurs
continue to consume perks and in default are sometimes liquidated.

• γ(g) = γ∗(g) > 1− γ∗(b). Good entrepreneur are induced not to consume perks and bad
entrepreneurs continue to refrain from fire sales. However, the error rate now exceeds that

required to induce bad entrepreneurs from having fire sales.

The more limited evaluation technology reduces the value of designing a court that makes

errors, relative to an environment in which the court is unconstrained in its evaluation technology.

Qualitatively, the value of noisy evaluation is reduced because to alter the behavior of one type but

not the other, the court also must err for the second type; and to affect the behavior of both types,

generically, the error rate must be unnecessarily high for one type. The costs of such mis-designs

depend on the specification of the economy in the expected ways (e.g. upon the productivities of

each entrepreneurial type if re-financed, the costs of the action choices, etc.). Generally, however,

the optimal design of the court’s monitoring technology should not be error free, because with an

error-free court, entrepreneurs take actions that are socially sub-optimal. Indeed, the optimal error

rate may be increased because, with an unconstrained evaluation technology, it may be optimal

to err for one type of entrepreneur, but not the other;2 but with the more limited evaluation

technology, the court sometimes errs in identifying both types.

We now explore how the analysis is altered by renegotiation. Incorporating a single signal

quality generally reduces the value of an error-riven court. In contrast, if the parties can sometimes

renegotiate and reach a settlement, this circumvents the error-making by the court, leaving only

the incentive effects for entrepreneurs and raising the value of a court that errs. In what follows, we

characterize conditions under which if creditors and debtors can always renegotiate and creditors

2Recall that this asymmetric technology can be implemented by placing the entire burden of proof on one party.

17



have all of the bargaining power then a blind court design can implement the social optimum.

Further, we identify situations where a completely uninformed court can implement the social

optimum and where it cannot.

Qualitatively, both factors are relevant: while the court may have limited flexibility in condition-

ing signal quality on type, the parties can probably only sometimes renegotiate to reach a private

settlement ( e.g. because a bankrupt entrepreneur may lack access to the funds required to make

the payment to the creditor that would circumvent the bankruptcy court). In such environments,

it would be optimal for the court only to be “near-sighted,” but not completely “blind.”

3.2 Renegotiation

Chapter 11 bankruptcy is characterized by a prolonged and involved bargaining session between

the firm and creditors. Creditors and entrepreneur may be able to renegotiate in bankruptcy and

reach a superior outcome by agreeing on the efficient liquidation decision. In particular, a bad

entrepreneur and creditor may agree to liquidate the firm and provide the entrepreneur with some

of the liquidation payoffs; or a good entrepreneur may offer the creditor a greater share of the firm’s

proceeds if the creditor lets the firmoperate. In both cases, agents circumvent the error-prone court.

If an entrepreneur defaults, a cost of c is incurred (which represents the cost associated with

renegotiation and/or using the court). If the entrepreneur and creditor fail to resolve their dispute

via renegotiation, the court then evaluates the entrepreneur and liquidates according to (γ(b),γ(g)),

where γ(b) = 1−γ(g). We assume that while the creditor does not observe the entrepreneur’s type
or the action taken, and that the creditor has all of the bargaining power in any renegotiations.

We do not place any restrictions on when renegotiation can occur. For example, if the entre-

preneur does not default, then the creditor can propose always a renegotiation offer, which the

entrepreneur can accept or reject.

We assume that 0 < γ∗(g) < γ∗(b) < 1. The fact that both critical values are strictly between

zero and one implies that the court can influence the behavior of entrepreneurs through its choice

of γ(g). Since γ∗(g) < γ∗(b), good entrepreneurs should be liquidated less frequently than bad if

there are no constraints on the evaluation technology.

In renegotiation, a bad entrepreneur would want to liquidate rather than take his chances in

court if the creditor pays him at least (1−γ(b))π2(b,ab) in return, where ab is the action that the
bad entrepreneur took in period 1. This payment corresponds to the bad entrepreneur’s expected

payment if he takes his chances in court. Similarly, a good entrepreneur would pay the creditor up

to γ(g)π2(g,ag), in exchange for avoiding the court and operating the firm in period 2. Consider

18



the following renegotiation offer that the creditor might make following a default in an equilibrium

where, at date 1, good entrepreneurs take action ag and bad entrepreneurs take action ab.

Renegotiation Offer: “If you (the entrepreneur) agree to liquidate, then you will receive P(b,ab);

if you do not want to liquidate and continue into period 2, then to avoid running the risk of the

court erring, you must pay me (the creditor) an additional γ(g)π2(g,ag) in period 2. If you reject

this offer, you will go to court.”

Suppose that the creditor makes this renegotiation offer and P(b,ab) is such that neither type

of entrepreneur rejects it in order to take their chances in court. A good entrepreneur would prefer

to pay γ(g)π2(g) in exchange for running the firm in period 2 to taking the liquidation contract

meant for a bad entrepreneur if the liquidation payment, P(b,ab), is not “too high,” i.e., if

A. (1−γ(g))π2(g,ag) ≥ P(b,ab).

A bad entrepreneur would agree to liquidate, rather than make a payment that would enable

him to circumvent the court and run the firm in period 2, if the liquidation payment, P(b,ab), is

“sufficiently high,” i.e., if

B. P(b,ab) ≥ π2(b,ab)−γ(g)π2(g,ag).

Together, conditions A and B imply that (1−γ(g))π2(g,ag) ≥ P(b) ≥ π2(b,ab)−γ(g)π2(g,ag).
Note that if π2(g,ag) > π2(b,ab), then there always exists a P(b,ab) that satisfies conditions A and

B. This “efficient liquidation” renegotiation offer is incentive compatible if

1. It satisfies conditions A and B, and

2. The creditor is willing to make an offer that satisfies conditions A and B.

It is incentivecompatible fora badentrepreneur to accept the renegotiation offerand liquidate if and

only if doing so dominates both (i) not settling and taking its chances in court, which has expected

payoff (1− γ(b))π2(b,ab); and (ii) passing himself off as a good entrepreneur, and circumventing
the court in return for the payment of γ(g)π2(g,ag). If the creditor makes an incentive compatible

renegotiation offer, then the creditor will make the smallest payment necessary to induce a bad

entrepreneur to liquidate, setting

P(b) = max{π2(b,ab)−γ(g)π2(g,ag),(1−γ(b))π2(b,ab)}.

19



However, it may not be in the creditor’s best interest to make this “efficient liquidation” rene-

gotiation offer. For example, if the payment to the bad entrepreneur necessary to ensure incentive

compatibility is too high, the creditor may prefer to make a renegotiation offer in which P(b,ab) = 0.

In this situation, a good entrepreneur would pay γ(g))π2(g,ag) in order to continue operating the

firm in period 2, but a bad entrepreneur would reject the offer and take his chances in court.

The question then arises: When would the creditor prefer to make an efficient liquidation

renegotiation offer that both entrepreneur types would prefer to taking their chances in court?

Lemma 3: Suppose that π2(g,ag) > π2(b,ab). Then, if bad entrepreneurs prefer taking their

chances in court to mimicking good entrepreneurs, i.e. if

(1−γ(b))π2(b,ab) ≥ π2(b,ab)−γ(g)π2(g,ag), (11)

then the creditor always makes an efficient liquidation renegotiation offer, setting P(b,ab) = (1−
γ(b))π2(b,ab). If, instead, bad entrepreneurs prefer to mimic good entrepreneurs to taking their

chances in court, so that (11) does not hold, then the creditor makes an efficient liquidation rene-

gotiation offer with P(b,ab) = π2(b,ab)−γ(g)π2(g,ag), only if

γ(g)(z(b,ab)−R2(b,ab)+ π2(b,ab)+ π2(g,ag)) ≥ π2(b,ab). (12)

Proof: If (11) holds, the creditor’s payoff from renegotiation offer P(b,ab) = (1−γ(b))π2(b,ab) is

z(b,ab)− (1−γ(b))π2(b,ab). (13)

If he instead offers P(b,ab) = 0, he expects

γ(b)z(b,ab)+(1−γ(b))(R2(b,ab)−π2(b,ab)). (14)

Since z(b,ab) > R2(b,ab) it follows that (13) exceeds (14). The bad entrepreneur is indifferent

between accepting P(b,ab) and going to court, so he accepts.

If, instead, (11) does not hold, then the payoff from making an incentive compatible efficient

liquidation renegotiation offer is

z(b,ab)− (π2(b,ab)−γ(g)π2(g,ag)),

while the payoff from making the renegotiation offer with P(b,ab) = 0 is

γ(b)z(b,ab)+(1−γ(b))(R2(b,ab)−π2(b,ab)).

The payoffs from the efficient liquidation renegotiation offer are greater only if (12) holds.

20



Definition: “Blind justice” is a court design where γ(g) = γ(b) = 0.5.

If the court design is blind, then the court ignores all information, essentially deciding whether

or not to liquidate an entrepreneur on the basis of a fair coin flip.

Corollary 1: If the court design is blind and if π2(g,ag) > π2(b,ab), then creditors will make

entrepreneurs an efficient liquidation renegotiation offer.

Proof: If γ(g) = γ(b) = 0.5, then condition (12) is satisfied.

Corollary 1 implies the following key result:

Proposition 3: It is never optimal to have an error-free court.

Proof: Here we prove that “blind justice” dominates an error-free court. An error-free court makes

efficient liquidation decisions, but both entrepreneurs choose socially inefficient actions, ag = a
1
g and

ab = a
2
b. When justice is blind, Corollary 1 ensures that if π2(g,ag) > π2(b,ab) then the creditor will

make an efficient liquidation renegotiation offer, so that again all liquidation outcomes are socially

efficient. The actions that entrepreneurs take depend upon the magnitudes of the critical values

γ∗(g) and γ∗(b). If γ∗(g) ≤ 0.5 ≤ γ∗(b), then the good entrepreneur chooses not to steal, ag = a2g,
and the bad entrepreneur does not have a fire sale, ab = a

1
b, since γ

∗(g) ≤ γ(g) and γ∗(b) ≥ γ(b).
These are the socially optimal actions and, therefore, they strictly dominate the actions induced

by an error-free court. (Note that since π2(g,a
2
g) > π2(b,a

1
b), the creditor will, in fact, make an

efficient liquidation renegotiation offer.) If γ∗(g) < γ∗(b) < 0.5, then the good entrepreneur does

not steal, ag = a
2
g, and the bad entrepreneur has a fire sale, ab = a

2
b. This outcome strictly

dominates the outcome induced by an error-free court, since here, the good entrepreneur takes

the socially optimal action. The creditor makes an efficient liquidation renegotiation offer because

π2(g,a
2
g) > π2(b,a

2
b). Finally, if 0.5 < γ

∗(g) < γ∗(b), then the good entrepreneur steals, ag = a1g,

and the bad entrepreneur does not have a fire sale, ab = a
1
b, and π2(g,a

1
g) > π2(b,a

1
b), which implies

that the creditor makes an efficient liquidation renegotiation offer. Therefore, blind justice always

performs strictly better than an error-free court.

Suppose now that the entrepreneur does not default at date 1. Even in the absence of a default,

it is still optimal for the bad entrepreneur to be liquidated. It is possible to achieve this outcome

if the creditor and entrepreneur can renegotiate their initial contract at the end of period 1. For

example, suppose that after achieving a level of first period revenues results in “no default,” the

creditor offers the entrepreneur π2(b,ab) in exchange for liquidating his firm, where ab represents the

21



equilibrium action taken by the bad entrepreneur. The bad entrepreneur will be indifferent between

accepting this offer and producing in the second period, so the bad entrepreneur will accept this

offer. For all of the possible equilibrium configurations that might emerge, it will always be the

case that π2(g,ag) > π2(b,ab) implying, the good entrepreneur will always reject the renegotiation

offer. Therefore, when the court is blind, the socially optimal liquidation decisions will always be

implemented, via renegotiation, both in and out of default.

Renegotiation introduces a subtle issue, albeit one that does not affect the above analysis: Rene-

gotiation may alter the “critical values,” γ∗(g) and γ∗(b), that determine whether each entrepreneur

type takes the high default probability action. Recall that those values were defined for a contract

that is not subject to renegotiation (equation (4)). However, when renegotiation is possible, the

initial contract between the entrepreneur and the creditor is altered. If, for example, credit markets

are competitive at the time that the initial contract is negotiated, then the entrepreneur extracts

all of the surplus generated by renegotiation.3

In fact, when credit markets are competitive, the incentives when renegotiation is feasible

correspond to those where it is infeasible: γ∗(g) and γ∗(b) are the appropriate critical values

under renegotiation. To see this, let CN represent that contract that maximizes the entrepreneur’s

expected payoff conditional on: (1) satisfying the creditor’s participation constraint; (2) the court

using γ(g) = γ(b) = γ = 0.5; and (3) no renegotiation. Let the initial contract that the entrepreneur

offers when renegotiation is possible, CR, take the form CR = (α∗;CN) , where α∗ is an up front

lump sum payment from the creditor to the entrepreneur. Given competitive credit markets, the

lump sum payment effectively transfers all gains from renegotiation to the entrepreneur. The lump

sum payment (see the appendix) has the form

α∗ = p(g)α∗(g,ag)+p(b)α∗(b,ab),

where p( ) is the ex ante probability that the entrepreneur is type and α∗( ,a ) represents the

lump sum transfer of gains that flows to the entrepreneur of type when he is induced to take

action a . Since contract CR only differs from contract CN by a lump sum payment, CR provides

the same entrepreneurial incentives in terms of action choice as CN.

The equilibrium path is characterized by:

1. The court (or society) chooses (γ(g) = 0.5,γ(b) = 0.5).

2. The entrepreneur offers contract CR = (α∗;CN), which is accepted by the creditor.
3If credit markets are not competitive, then the entrepreneur cannot extract all of the surplus associated with

renegotiation. Qualitatively, the results presented below are not sensitive to the assumed nature of credit markets,
although the precise form of the initial contract is.

22



3. The entrepreneur takes actions

(a) ag = a
2
g and ab = a

1
b if γ

∗(g) ≤ 0.5 ≤ γ∗(b), i.e., the good entrepreneur does not steal
and the bad entrepreneur does not have a fire sale.

(b) ag = a
2
g and ab = a

2
b if γ

∗(g) < γ∗(b) < 0.5, i.e., the good entrepreneur does not steal

and the bad entrepreneur has a fire sale.

(c) ag = a
1
g and ab = a

1
b if 0.5 < γ

∗(g) < γ∗(b), i.e., the good entrepreneur steals and the

bad entrepreneur does not have a fire sale.

4. If the entrepreneur does not default, then the creditor offers the entrepreneur π2(b,ab) in

exchange for liquidating the project: The good entrepreneur will not accept this offer (since

π2(g,ag) > π2(b,ab)) and the bad entrepreneur will accept.

5. If the entrepreneur defaults, then the creditor makes the above mentioned renegotiation of-

fer resulting in the liquidation of the bad entrepreneur and the continuation of the good

entrepreneur.

The equilibrium is characterized by efficient liquidations and continuations: Along the equilib-

rium path, the bad entrepreneur is always liquidated and the good entrepreneur is never liquidated.

Note that, depending upon model parameters, there can be equilibria where bad entrepreneurs have

fire sales and equilibria where they do not; there can be equilibria where good entrepreneurs steal

and equilibria where they do not.

In summary, if the entrepreneur and creditor can always renegotiate, and it is optimal (i)

to encourage bad entrepreneurs to take the high default action in order to reduce their survival

probability, and (ii) to encourage good entrepreneurs to take the low default action, then a blind

court design cannot be improved upon if creditors and entrepreneurs can always renegotiate their

way around the court. Note that our model predicts that even though the firm does not default,

renegotiations and liquidations may occur and, in the event of default, parties will always settle

out of court. We typcially do not observe liquidations outside of default and do observe defaulting

parties using the court to reshedule their debts. It is the simple and stylized structure of our

model that delivers these predictions. We now consider, in turn, how each of these predictions are

modified when our model is generalized in rather natural directions.

As we discussed earlier, renegotiation both in and out of bankruptcy may break down. For

example, suppose that with some relatively small probability, good entrepreneurs have an alter-

native opportunity that would pay w, where it is the case that (1) it is socially inefficient for a

23



good entrepreneur to take action w, w < R2(g,ag) − z(g,ag), and (2) in the absence of a rene-
gotiation offer, it is not attractive for a good entrepreneur to pursue the alternative opportunity,

π2(g,ag) > w, but (3) if the good entrepreneur can obtain the payment meant to induce the bad

entrepreneur to default, then the opportunity becomes attractive, w +P(b,ab) > π2(g,ag). Then,

if most entrepreneurs who do not go into default are good, it may not be optimal for creditors to

negotiate with entrepreneurs outside of bankruptcy. That is, even if only a few good entrepreneurs

have such an outside opportunity, the cost of those good entrepreneurs with outside opportunities

passing themselves off as bad in order to obtain the payment meant for a bad entrepreneur may

exceed the gain from inducing bad entrepreneurs to liquidate. Casual observation suggests that

such renegotiations are infrequent. Conversely, it may be that only a small fraction of entrepreneurs

in bankruptcy are both good and have outside alternatives, and hence would take offer meant for

the bad entrepreneur. Then there is only a small cost to making the above renegotiation offer in

bankruptcy. In such a circumstance, precisely because renegotiation outside of bankruptcy becomes

infeasible, the value of a blind court design may rise relative to that of a court that does not err:

more bad entrepreneurs are liquidated if the court design is blind.

Our model predicts that defaulting parties will always settle out of court. This stark result is

an artifact of our assumption that there are only two types of entrepreneurs. Suppose, instead,

that there are many type of entrepreneurs. In the event of a default, as above, the creditor makes

a renegotiation offer to the entrepreneur that specifies: (i) a payment to the entrepreneur for

liquidating the firm; (ii) a payment from the entrepreneur to in order to continue operating into

period 2; or (iii) going to court. With many types of entrepreneurs, only the very high quality

entrepreneurs agree to pay in order to continue; only the very low quality entrepreneurs accept

a payment and liquidate; and entrepreneurs of intermediate quality go to court. It follows that

entrepreneurs who reorganize out of court are more successful, on average, in the future than

entrepreneurs who survive and reorganize under the supervision of a court. This prediction is

consistent with the evidence reported by Gilson, John and Lang (1990).

4 Stealing

In this section we assume that the entrepreneur’s action choice is either “steal” or “do not steal,”

and examine the implications for burden of proof and court design. Here, stealing corresponds to

the high default probability action, a1, and not stealing corresponds to the low default probability

action, a2. Stealing isnot sociallyoptimalof either typeof entrepreneur. Wesimplify theanalysisby

assuming that the liquidation value, second period revenues and second period profits are unaffected

by first period action choices, and that the liquidation value is the same for both entrepreneurial

24



types. As a result, the second period profit, second period revenue and the liquidation value

simplify to π2( ), R2( ) and z, respectively. Since it is optimal to liquidate a bad entrepreneur

and to let a good entrepreneur continue in the second period, we have R2(b) < z < R2(g). Many

principal-agent models interpret the agent’s action as being an effort level and that the agent can

either “work hard” or “shirk;” it is assumed that it is socially optimal for agents to work hard. By

slightly modifying our model to allow for the non-pecuniary payoff associated with effort one could

interpret action a1 as shirking and action a2 as working hard.

If society were unconstrained in its choice of liquidation probabilities and were only concerned

about getting liquidation decisions right, then it would choose γ(g) = 0 and γ(b) = 1.4 That is, the

court would never liquidate a good entrepreneur and would always liquidate a bad entrepreneur.

Provided that 0 < γ∗( ) < 1 for ∈ {b,g}, these liquidation probabilities deliver the socially
optimal outcome for the bad entrepreneur, but would induce the good entrepreneur to steal. To

induce the good entrepreneur to not to steal, the court must choose a liquidation probability for

the good entrepreneur of at least γ∗(g).

Consider the followingburdenofproof rule: An entrepreneur mustdemonstrate to thecourt that

he is good, or else he will be liquidated. The burden of proof must be sufficiently stringent that a

bad entrepreneur is unable to prove to the court that he is good and that a good entrepreneur cannot

not always convince the court that he is good. Occasionally a good entrepreneur will be liquidated

by the court. Such a burden of proof rule implies that the court liquidates a good entrepreneur

with probability γ(g), where γ(g) > 0 and liquidates a bad entrepreneur with probability γ(b) = 1.

Suppose that an entrepreneur defaults. As in Section 3.2, the creditor has an incentive to rene-

gotiate the existing contract in order to eliminate the risk of surplus loss that occurs when a good

entrepreneur is liquidated. In contrast to the previous section, however, creditors do not have an

incentive to renegotiate with a bad entrepreneur–if they could identify –since a bad entrepreneur

will be liquidated with probability one by the court. Unfortunately for the creditor, however, he

cannot distinguish between good entrepreneurs and bad. Consider the following renegotiation offer

that a creditor might make to an entrepreneur following a default:

Renegotiation Offer: “If you (the entrepreneur) want to avoid running the risk of the court

erring, you must pay me (the creditor) an additional γ(g)π2(g) in period 2. If you reject this offer,

then you will go to court.”

Good entrepreneurs will accept the renegotiation offer and agree to pay the creditor an addi-

tional γ(g)π2(g) in period 2. Note, however, if γ(g)π2(g) is not “sufficiently large,” then bad entre-

4Note that these probabilities satisfy the single signal quality restriction.

25



preneurs will also accept this offer. Bad entrepreneurs will not accept offer if π2(b)−γ(g)π2(g) ≤ 0.
Thus, if

γ(g) ≥ π2(b)
π2(g)

,

the bad entrepreneur will reject the renegotiation offer, go to court, and be liquidated with prob-

ability one. If the courts adopt a “stringent” burden of proof rule, stringent in the sense that

it always liquidates bad entrepreneurs and liquidates the good entrepreneur with probability of

(at least) π2(b)/π2(g), then the liquidation probabilities will result in socially optimal liquidation

decisions. But what about the actions that the entrepreneur takes at date 1?

Since the burden of proof rule implies that γ(b) = 1, the bad entrepreneur will always work

hard, i.e., ab = a
1
b, since γ

∗(b) < γ(b) = 1. The good entrepreneur will work hard if γ(g) ≥ γ∗(g).
Hence, society can induce the good entrepreneur to work hard and at the same time generate a

socially optimal liquidation decision in the event of default if

γ(g) ≥ max{π2(b)
π2(g)

,γ∗(g)}. (15)

If the court’s liquidation probability for the good entrepreneur satisfies (15), then the good entre-

preneur will have an incentive to work hard in the first period–since γ(g) ≥ γ∗(g)–and a bad
entrepreneur who defaults will submit to liquidation–since γ(g) ≥ π2(b)/π2(g).

Note that if the court liquidates any entrepreneur that it faces with probability one, i.e., γ(g) =

γ(b) = 1, then at date 1, no entrepreneur steals/all work hard. In addition, a defaulting bad

entrepreneur will be liquidated and a defaulting good entrepreneur will, via renegotiation, continue

to operate at date 2. That is, if the burden of proof is so stringent that no one can demonstrate

that he is good, then the first-best outcome can be achieved. This is a rather odd result because

it implies that the court simply liquidates anyone who approaches it. Suppose, however, that

there is a some chance that renegotiation between the creditor and the defaulting entrepreneur

can break down–due to, say, a coordination failure–in which case the entrepreneur faces the

court upon default. If the entrepreneur happens to be bad, then there is no social loss associated

with the breakdown in renegotiation. But, if the entrepreneur is good, then there is a social loss

because he is always liquidated. To minimize the probability of liquidating the good entrepreneur

if renegotiation breaks down and, at the same time, provide incentives for a good entrepreneur to

work hard/not steal, the court’s liquidation probability should be set equal to the right-hand side

of (15).

The equilibrium path for this economic environment is characterized by:

1. The court (or society) imposes a burden of proof rule that implies that γ(b) = 1 and γ(g) =

max{π2(b)/π2(g),γ∗(g)}.

26



2. The entrepreneur offers contract CR = (α∗∗,CN), which is accepted by the creditor.5

3. The good entrepreneur chooses action ag = a
2
g and the bad entrepreneur chooses action

ab = a
2
b.

4. If there is no default, the creditor offers the entrepreneur π2(b) in exchange for liquidation:

the bad entrepreneur will accept the offer and the good entrepreneur will not.

5. If there is a default, the creditor asks for an additional payment of γ(g)π2(g) in period 2 or

else the entrepreneur goes to court: the good entrepreneur accepts this offer and the bad

entrepreneur does not and is ultimately liquidated by the court.

5 Conclusion

This paper shows that the optimal design of a bankruptcy court is generally one in which the

court occasionally makes errs, sometimes mistakenly liquidating good entrepreneurs, and failing to

liquidated bad ones. A mistake-prone court

1. May discourage good entrepreneurs from taking actions that lower total firm value (theft,

shirking, risky negative NPV investments) by raising the cost to them of entering bankruptcy.

2. May encourage ‘bad’ entrepreneurs to take actions that increase the probability that they

will enter bankruptcy (desist from fire sales) and hence liquidated.

We first provide conditions under which even where renegotiation is not possible, an error-

prone court is preferred to an error-free court. The optimality of the error-riven court depends on

the benefits of inducing entrepreneurs to take the socially optimal action relative to the costs of

sometimes making incorrect liquidation decisions.

We then consider the optimal court design when creditors and debtors can renegotiation to

circumvent an error-riven court and creditors have all of the bargaining power. Very generally,

we illustrate that the optimal court design is one where the court makes mistakes. The crucial

caveat to this is that there does not exist a “one type fits all” burden of proof rule, that is optimal

independent of the types of actions that agents may take.

If it is optimal to discourage good entrepreneurs from taking actions that raise bankruptcy

probabilities (theft, perk consumption, risky NPV investments, shirking), while discouraging bad

5As in the previous section, CN represents the contract where renegotiation is not possible and γ(b) = 1 and
γ(g) = max{π2(b)/π2(g),γ∗(g)}. The parameter α∗∗ represents an up-front lump sum payment that transfers all of
the gains from renegotiation to the entrepreneur.

27



entrepreneurs from taking actions that lower bankruptcy probabilities (fire sales, working exces-

sively hard, excessively safe investments), then we show that a blind court, which does not use any

information, dominates an error-free court. Facing a blind court, creditors and debtors negotiate

the ‘correct’ liquidation decision following a default, so that a blind court design induces the same

liquidation decisions as an error-free court. In addition, the blind court design induces entrepre-

neurs to take better actions. Thus, for this class of action choices, the same simple blind court

design leads to better outcomes.

However, if it is instead optimal to discourage bad entrepreneurs from taking actions that

raise bankruptcy probabilities, then we show that the optimal court design places the burden of

proof on the entrepreneur. As a result, the court would sometimes mistakenly liquidating good

entrepreneurs who are unable to document their quality. While the optimal court design is again

simple to implement, it differs from that where the goal is to discourage bad entrepreneurs from

liquidating some of the firm to avoid bankruptcy.

28



References

Berkovitch, E., Israel, R. and J. Zender, (1998), “The design of bankruptcy law: A case for man-

agement bias in bankruptcy reorganizations," Journal of Financial and Quantitative Analysis, 33,

441-64.

Bernhardt, D. and D. Lu (1998) “The Dynamics of Chapter 11 Bankruptcy’, mimeo.

Giammarino, R. and E. Nosal (1994) “The efficiency of judicial discretion in bankruptcy law”,

mimeo.

Che, Y. and Hausch, D., (1999), “Contract Damages and Cooperative Investment,” American

Economic Review, 89, 125-147.

Fishman, M. and K. Hagerty, (1990), “The Optimal Amount of Discretion to Allow in Disclosure,”

Quarterly Journal of Economics, 428-444.

Gilson, S., John, K. and Lang, T., (1990), “Troubled Debt Restructuring: An Empirical ...,”

Journal of Financial Economics, 27, 315-353.

Grossman, G. and Katz, M., (1983), “Plea Bargaining and Social Welfare,” American Economic

Review, 73, 749-757.

Hay, B. and Spier, K., (1997), “Burden of Proof in Civil Litigation: An Economic Perspective,”

Journal of Legal Studies, 413-431.

Milgrom, P., (1981), “Good New and Bad News: Representation Theorems and Applications,” Bell

Journal of Economics, 380-391.

Milgrom, P. and J. Roberts, (1986), “Relying on the Information of Interested Parties,” Rand

Journal of Economics, 17, 18-32.

Shin, H. S., (1994), “The Burden of Proof in a Game of Persuasion,” Journal of Economic Theory,

64, 253-264.

Shin, H. S., (1998), “Adversarial and Inquisitorial Procedures in Arbitration,” Rand Journal of

Economics, 29, 378-405.

Sobel, J. (1985), “Disclosure of Evidence and Resolution of Disputes: Who should bear the burden

of proof?”, in Game Theoretic Models of Bargaining, 341-361, Cambridge University Press.

29



Spier, K., (1994), “Settlement Bargaining and the Design of Damage Awards,” Journal of Law,

Economics and Organization, 84-94.

Williamson, O., (1985), The Economic Institutions of Capitalism (New York: Free Press.)

Willington, M., (2002), “Hold-Up under Costly Litigation and Imperfect Courts of Law,” mimeo.

30



Appendix

To understand the expressions for the components of α∗, suppose the court chooses γ(g) =

γ(b) = γ = .5 when renegotiation is not possible and the contract that is in place is CN. With

contract CN in place and a blind court, the entrepreneur will choose action a2g if he is good and

action a1b if he is bad. If renegotiation is possible when the court is blind, then bad entrepreneurs

will be liquidated and good entrepreneurs will continue. Consider the following lump sum payment

that is attributable to the good entrepreneur,

α∗(g,a2g) = B(g,a
2
g))(π2(g;a

2
g)+R2(g,a

2
g)−z),

This lump sum payment has two components: (1) the surplus generated through renegotiation with

the good entrepreneur, γB(g,a2g)(R2(g,a
2
g)−z), and (2) the payment that the good entrepreneur

gives the creditor if he defaults, γB(g,a2g)π2(g;a
2
g). Now consider a lump sum payment that is

attributable to the bad entrepreneur,

α∗(b,a1b) = ((1−γ)B(b,a1b)+(1−B(b,a1b))+(z−R2(b)−π2(b,a1b).

To understand this lump sum payment, note that following a default, renegotiation generates

surplus (1−γ)(z−R2(b)), and the entrepreneur receive portion (1−γ)π2(b,a1b) of the total surplus;
when the entrepreneur does not default, renegotiation generates a surplus of (z −R2(b)) and the
entrepreneur receives a portion π2(b,a

1
b).

When the court is blind and renegotiation is possible suppose that the creditor and entrepreneur

negotiate the contract CN with an additional up front lump sum payment of

α∗ = p(g)α∗(g,a2g)+p(b)α
∗(b,a1b).

Since α∗ is an up-front lump sum payment, contract CR = (α∗,CN) provides the entrepreneur with

the same incentives when renegotiation is possible as does contract CN when renegotiation is not

possible. By construction, α∗ transfers all gains from renegotiation to the entrepreneur.

31



Federal Reserve Bank

of Cleveland

Research Department

P.O. Box 6387

Cleveland, OH 44101

Address Correction Requested:

Please send corrected mailing label to the

Federal Reserve Bank of Cleveland

Research Department

P.O. Box 6387

Cleveland, OH 44101

PRST STD

U.S. Postage Paid

Cleveland, OH

Permit No. 385


	Abstract
	1 Introduction
	2 Unconstrained court evaluation technologies
	2.1 Optimal choices of γ( )
	2.1.1 Good Entrepreneurs
	2.1.2 Bad Entrepreneurs


	3 Confiscation: Perk consumption and Fire Sales
	3.1 No Renegotiation
	3.2 Renegotiation

	4 Stealing
	5 Conclusion
	References
	Appendix