Managerial Overconfidence and Moral Hazard: Why Try Harder?

Managerial Overconfidence and Moral Hazard:

Why Try Harder?

MASTER THESIS

MSc Business Administration – Finance

Corporate Financial Management

Groningen, 11/05/2009 By

Bouke de Boer

S1322133 bfdeboer@gmail.com

Supervisors: Dr. Lammertjan Dam Dept. of Finance

Faculty of Economics and Business

Prof. dr. Frans Tempelaar

Dept. of Finance

Managerial Overconfidence and Moral Hazard:

Why Try Harder?

11/05/2009

Bouke de Boer

S1322133

bfdeboer@gmail.com

This thesis examines the effects of overconfidence on optimal contracting, using the agency theory framework developed by Tirole (2006). I find that it is beneficial for managers to keep overconfidence hidden, as symmetric knowledge about differing priors will lead to more credit rationing. In a risk neutral setting, I also find that overconfidence leads to only a small interval of efficient investment outcomes, otherwise resulting in overinvestment. These results are robust to introducing control right allocation and in contingent control right models. Overconfidence, leading to overoptimism about expected project payoff, amplifies the agency problem and leads to increased moral hazard.

1. Introduction

Managers are not the logically thinking, rational beings that economic theory supposes them to be. A manager is a human being, with human concerns and human ambitions. Burdened with the leadership of a company, a manager does not necessarily act solely on behalf of investors. However, this is not why empirical outcomes are sometimes inconsistent with outcomes predicted by economic models. These inconsistencies occur because of the nature of a manager’s behavior. There is a difference between behavior of a manager that can be explained in a rational framework and behavior stemming from bounded rationality, based on psychological biases. The first type of behavior mainly involves issues of empire building, enjoying perks, or simply a lack of effort. These issues of moral hazard and adverse selection are governed by an incentive contract that is set up along the principles of agency theory, since they are regarded as rational issues. However, the second type of behavior is not considered to be fully rational. For example, if a manager believes he is more skillful than he actually is, he believes his actions are in the best interest of the company. Thus, if a manager is overconfident, he still firmly believes that his course of action is the right one. A standard incentive contract will not solve all issues concerning a manager’s actions, only the rational ones. His overconfidence needs to be addressed differently.

There is ample evidence suggesting overconfident management. For example, mergers and acquisitions are numerous, but many of them fail to result in higher profits, especially after accounting for the staggering sums paid for a takeover. In other cases, companies follow strategies that destroy value for the firm but are pursued nonetheless, because of a false belief that things will turn around. Standard economic theory cannot fully explain these findings. In a rational setting, the different players would have acted along guidelines and principles that result in economically efficient outcomes. Behavioral finance can extend the theory, but is still rooted in sound economic reasoning. Inconsistencies with empirical outcomes can be dealt with by including behavioral biases. In doing so, the framework for explaining and predicting investor and managerial behavior is enriched.

assets the manager can supply and his level of overconfidence determine whether the outcome is efficient or leads to overinvestment.

The reasoning behind this basic contracting model can be extended to other models. My thesis also looks at control right allocation and models of contingent control rights, to study whether introduction of overconfidence in these models leads to the same results as those for the basic contracting model. When I extend my research to multiple models, I can check my results for robustness and add to the explanatory power of these economic models. Again, initial assets and level of overconfidence lead to either an efficient outcome or overinvestment. Following these findings, and since overconfidence has been shown to be present in many managers, an interesting question is whether the existence of managerial overconfidence can be used to refrain from transferring control rights to investors.

My main contribution to existing literature is a negative consequence for project financing under managerial overconfidence, because it often leads to overinvestment. A number of previous studies have indicated that a slightly overconfident manager is beneficial to project financing, as this overconfidence mitigates moral hazard issues. In my analyses, I do not find this effect. In a setting of risk neutrality, managerial overconfidence leads to only a small interval of economically efficient outcomes of the contracting process between a rational investor and an overconfident manager. Otherwise, managerial overconfidence leads to overinvestment. In a situation with asymmetric knowledge about priors, in which the manager keeps his prior hidden from the investor, the agency problem is amplified and overconfidence leads to increased moral hazard.

In the next section, I review previous studies that have been done in the field of agency theory and behavioral biases. In the following sections, I present my analyses and findings concerning the adaptation of the contracting model and a number of extensions to this basic model. Section 6 concludes and discusses the results.

2. Literature Review

As researchers recognized that psychological biases of economic agents such as investors and managers affect the outcomes of decision-making, a strand of literature on this topic emerged. Early work mainly focused on investor behavior and problems of market efficiency. Reviews of this early work can be found in Shleifer (2000), Daniel et al. (2001), and Barberis and Thaler (2002).

The notion that managers are also influenced by these biases gradually took hold. Such a bias can take different forms. The most common biases are optimism, overconfidence and the illusion of control. All three are closely related. The illusion of control is defined by Langer (1975) as the expectancy of a personal success probability that is inappropriately higher than the objective probability would warrant, due to a false belief that chance events can be controlled. If managers believe they are in control, they are also prone to optimism. According to Weinstein (1980), optimism is enhanced by a feeling of commitment, and since most managers are professionally committed, it is likely that they are subject to such biases.

The other two biases, optimism and overconfidence, are introduced in various studies to explain managerial behavior. However, the definitions of these two phenomena vary, and they are used in different settings. Most research distinguishes the terms optimism and overconfidence, but there are studies that do not even separate optimism and overconfidence and use the terms interchangeably. An important issue is that in many studies, optimism and overconfidence are viewed as a belief relative to the beliefs of the rest of the market. In this sense, the market is viewed as having the rational and correct beliefs and the manager as having an optimistic or overconfident belief. In my research, I also assume that the market, i.e. the investor, holds the correct beliefs and that the manager has different, overconfident beliefs.

To illustrate, Baker et al. (2005) define optimism as an overestimate of mean and overconfidence as an underestimate of variance. In their review of behavioral corporate finance, they explicitly separate the two subjects. A study by Landier and Thesmar (2009) also defines optimism as an overestimate of mean or a higher perceived probability of success. Goel and Thakor (2006) and van de Venter and Michayluk (2008) use the term overconfidence for an underestimate of variance. This is different in the study by Gervais, Heaton and Odean (2007), that uses the two terms interchangeably and defines overconfidence/optimism as an overestimate of a manager’s own skills. Other authors share this view, such as Camerer and Lovallo (1999) and Moore and Small (2007), and define the term as a belief that the manager is better than average. These latter two studies use the term overconfidence only.

overestimate expected net present value of projects or firm performance. Interestingly, Fairchild (2007) cites this research in his overview but uses the term overconfidence. In his inaugural address, de Jong (2006) defines overoptimism as an overestimate of the probability of a favorable outcome and overconfidence as the overestimate of the payoff of a chosen course of action.

Thus, the use of these two terms is not uniform. In this regard, there is no single definition that covers this spectrum of managerial behavior. Many aspects can be defined as either overconfidence and/or optimism. However, two recent studies take a useful position to decide what term to use for which phenomenon. Malmendier and Tate (2005b) define overconfidence as overestimating expected returns of a manager’s corporate decision. They use this definition to align managerial behavior with the “better-than-average” effect discussed above. Thus, they distinguish between manager beliefs resulting from overconfidence and general optimism about exogenous events. Malmendier and Tate (2005a, 2008) use two measures for CEO overconfidence. In their 2005 article (Malmendier and Tate, 2005a), they study overconfidence in a revealed belief approach, where overconfidence is measured by a manager’s private investment decision to keep in-the-money stock options. The second article, Malmendier and Tate (2008), uses a perception of outsiders approach, tracking press publications about manager overconfidence. These studies are very useful in determining empirical evidence on overconfidence. De la Rosa (2007) shares the definition of overconfidence on the basis of endogenous factors with Malmendier and Tate (2005b). He also uses the term overconfidence to describe a manager overestimating the value of his forecasts. The manager overestimates the probability of favorable outcomes, following the agent’s actions. De la Rosa (2007) states that overconfidence is a more appropriate term for this behavior than optimism, because optimism suggests a more passive role when it relates to outcomes. I consider this a very useful definition of overconfidence and will use a similar one in my models. It relates to the topic of my research, where it is the choice of the manager to exert effort or not, that drives the model. The managerial bias is linked to action (or lack thereof) by the manager, and not to exogenous outcomes. As such, this definition of overconfidence can be embedded in agency models of moral hazard in a natural way, since moral hazard also commonly deals with action undertaken by the manager.

Just as the definition of the psychological biases differs and is used for different phenomena, so is the application of these biases in corporate finance widespread. To give some examples: there are studies that focus on the election of a CEO (Goel and Thakor (2006)), other studies use overconfidence as an explanatory factor for market entry (Koellinger, Minitti and Schade (2007), Moore, Oesch and Zietsma (2007)). Project termination is also reviewed in combination with overconfidence (Muir (2007)).

entrepreneurs. Contracting as a research topic started with the study by Alchian and Demsetz (1972), and subsequent research is neatly summarized by Eisenhardt (1989). She categorizes the development of agency theory along two complementary lines: positivist agency theory and principal-agent theory. Positivist agent theory focuses more on governance mechanisms, discussed in three seminal articles: Jensen and Meckling (1976), Fama (1980), and Fama and Jensen (1983).

Principal-agent theory is more abstract and mathematical in nature and usually focuses on designing an optimal contract. The analysis is commonly performed in the context of asymmetric information about two issues, moral hazard and adverse selection. Information about the actions taken by the manager is hidden from the investor in moral hazard scenarios and information about the type of manager is hidden in adverse selection scenarios. Tirole (2006) offers an overview of the main issues in corporate finance that are analyzed using agency theory. His models deal with both moral hazard and adverse selection. They are based on previous research concerning moral hazard and contracting, such as studies by Holmström (1979) and Aghion and Bolton (1992). My research focuses on moral hazard models. Next, I review studies that combine overconfidence or optimism with this type of model.

In behavioral corporate finance, the assumption of perfect rationality is loosened. In this field, there are a number of studies that deal with moral hazard and contracting issues. For example, Landier and Thesmar (2009) study the effects of optimism on financial contracting outcomes between rational investors and optimistic entrepreneurs. In a setting where entrepreneurs are risk averse, they find that optimism leads to more debt financing. The use of short-term debt results in the shift of control to investors when outcomes are bad. Heaton (2002) builds a model that incorporates higher perceived probabilities of success into an investment decision model for managers. Optimistic managers believe that the market undervalues risky projects. Thus, when a project requires external financing, the manager thinks the cost of capital is too steep, resulting in underinvestment. In terms of incentives and contracts, this means that an optimistic manager will decline an investment contract because he believes that the reward for the investor is too high.

A number of studies analyze the symmetric behavioral information scenario. Gervais, Heaton and Odean (2007) find that overconfident managers make capital budgeting decisions that a rational manager would not. They find that this has no negative effect on their welfare. In their model, in which both the investor and the manager know that the manager is overconfident, the manager is content with a flatter compensation structure due to overvaluation of information acquisition efforts. This reduces moral hazard problems and makes an overconfident manager hirable. The authors also find that too much overconfidence is detrimental to the manager’s welfare. Goel and Thakor (2006) study CEO appointment under managerial overconfidence. In addition, however, they review incentive contracts and find that overconfidence mitigates underinvestment, because the manager overestimates private information. Keiber (2005) also studies overconfidence in a symmetric behavioral information setting. He finds that overconfidence reduces agency costs and enhances the incentive component of a contract and thus effort. His study finds that overconfidence plays a crucial role in compensation contract setting. One of its conclusions about signals is that overconfidence about good signals is beneficial to shareholders. In contrast, overconfidence about bad signals is bad for shareholders. In relation to my research on incentive contracts and the use of signals, this would mean that an overestimate of the value of a good signal benefits the investor, but the underestimate of a bad signal harms the return for the investor. I discuss this issue in my analysis of control right models.

Malmendier and Tate (2005a, 2008) show that overconfidence of managers increases the sensitivity to cash flow. While their research does not focus specifically on contracting issues, their results are relevant to my research, as they find that an overconfident manager will refrain from outside financing. In contracting terms, this means that the manager thinks the necessary remuneration for the investor is too high. His own reward will not induce him to exert effort, should he undertake the project. De la Rosa (2007) studies the effect of manager overconfidence on incentive contracts in a moral hazard framework where principal and agent knowingly hold different beliefs regarding probability of success, corresponding with what I call symmetric behavioral knowledge. He finds that if a manager is overconfident about how his actions affect the probability of success of a project, he needs less incentive to exert effort on the project. Overconfidence and effort increase simultaneously.

they run. This leads to conclusions about overconfidence being beneficial for financing, as moral hazard issues are mitigated. Risk neutrality alters these results.

3. The Fixed-Investment Model

To study the effect of overconfidence on optimal contracting and financing decisions, I adapt the modeling framework by Tirole (2006). I allow for a difference in beliefs about the probability of success, that is, the investor and the manager do not necessarily use the same perceived probabilities of success of the project in their decisions. This approach is largely consistent with the approach taken by Heaton (2002), with an added focus on managerial effort.

Introducing overconfidence in the Fixed-Investment Model (FI Model) with moral hazard of Tirole (2006) affects the setting and outcomes of the model. Overconfidence leads a manager to believe his chances of success are higher than what the investor expects. This affects the working of the model and will ultimately result in an outcome that may lead to overinvestment. The model can be seen as a game between two players, who have a non-common prior about the prospects of an investment opportunity. In this setting, the parties either know that they have an uncommon prior and agree to disagree or this prior is hidden. I refer to this as symmetric or asymmetric behavioral knowledge, to distinguish it from the regular information asymmetry that is the cause of moral hazard. To make a statement about efficiency and welfare, I assume that the prior of the principal, the investor, is correct and the prior of the agent, the manager, is incorrect. Hence, I label this phenomenon as managerial overconfidence. I view overconfidence as a belief that the manager thinks himself to be more skillful and influential than he is in reality. This ‘better-than-average’ belief leads to higher expected probabilities of success. As such, the higher expected payoff can be viewed as overoptimism, and is a direct effect of managerial overconfidence. This definition is in line with the studies cited in section 2, for example Gervais, Heaton and Odean (2007). I view managerial overconfidence and subsequent choice of effort as the driving force behind my models, because this reflects the influence of the actions taken by the manager. This is closely linked to a sense of overoptimism about outcomes.

3.1 Basic FI Model – the symmetric behavioral knowledge scenario

can decide to shirk his responsibilities and exert less effort, giving him a private benefit, for example less stress and more leisure. Alternatively, this private benefit can take different forms. It might mean that the manager enjoys perks, using company funds for his own benefit by flying on private jet planes and driving a company car with chauffeur. It can also mean that the manager has to choose between a project with a high probability of success and a different project which he prefers (because it is easier, benefits a friend, or is more glamorous) but has a lower probability of success. The amount of effort is expressed in the monetary equivalent

ε

_H or

ε

_L, where the subscripts H and L indicate high and low effort respectively. Effort is subtracted from the manager’s utility. I make the natural assumption that

ε

_H

>

ε

_Land I define

B

=

ε

_H

−

ε

_L. The difference in levels of effort is the net gain in utility for the manager, and I label this net gain as the private benefit B. The project has a probability of success, defined as p, which takes value

p

H if the manager exerts all his effort, and

p

L if he shirks. I assume that

p

L

<

p

H and define

L H

p

p

p

=

−

∆

.

Both the investor and the manager are risk neutral. This means that we have linear utility functions, which gives us the opportunity to separate the effects of risk attitude and moral hazard. Risk attitude can be modified to either risk loving or risk averse – what matters is the difference in risk attitude. A difference in risk attitude will lead to slightly different results, but this is an extension which is beyond the scope of this thesis.

overconfidence of the manager as

ω

:

pˆ

_L

=

p

_L

+

ω

, so that

ω

increases the difference between the manager’s beliefs and the correct beliefs.

If the project is successful, the payoff R is divided between the manager and the investor. Out of the total payoff of the project, the manager receives

R

_M and the investor receives

R

_I, thus

R

=

R

_M

+

R

_I. I assume that a project is only viable if the manager behaves, even if one accounts for the private benefit:

0

>

− I

R

p

_H and

p

_L

R

−

I

+

B

<

0

. However, an overconfident manager believes that the net present value from the project is always positive and that

p

ˆ

_L

R

−

I

+

B

>

0

. This reflects the overoptimism about expected payoffs discussed earlier, as a direct result of managerial overconfidence. Because of this belief, the manager will not hesitate to shirk if he expects this to yield him a higher utility. Of course, the manager can also shirk under the standard FI Model, yet the assumption that I add here strengthens the idea that the manager sees no harm in exerting less effort, because even if there had been no moral hazard, i.e. when the investor is able to observe the manager’s actions, the manager would still choose the lesser project or low effort because of this perceived positive value in the case of shirking.

The manager and the investor set up a contract for the financing of the project that divides the project gains. The investor will only agree to finance the project if the manager behaves. Therefore, the necessary condition to be met is

p

H

R

M

−

ε

H

≥

p

ˆ

L

R

M

−

ε

L or:

B

R

p

_M

≥

_H

−

_L

=

∆

ˆ

ε

ε

(3.1)

This is the Incentive Compatibility Constraint for the manager (ICC). The investor will not participate if he believes the manager is not going to exert effort.

Given that the manager needs to be rewarded

R

_M, the maximum income that can be offered to the investor is

p

B

R

R

R

−

M

=

−

_∆

_ˆ

. For an overconfident manager, this pledgable income is lower, since

p

p

<

∆

∆ˆ

. For the investor, the expected pledgable income is       ∆ − p B R

p_{H ˆ} . The investor will only

agree to finance the project if this expected income is higher than the amount needed:

This is the Participation Constraint for the Lender (PCL). Financing of the project is thus dependent on the amount of assets the manager has available. The PCL requires the manager to supply at least the following:

(

p R I

)

p B p A A≥ = H _∆ˆ− H − ˆ (3.3)

A represents the minimum amount of assets the manager needs to supply to get the investor on board. An overconfident manager will need more assets to attract investors to his project, Aˆ > A. Thus, when the investor knows about the manager’s beliefs, there is more credit rationing when overconfidence is involved. If the manager can somehow keep his beliefs hidden, he might convince the investor to participate. This is detailed in the next section.

3.2 Basic overconfidence FI Model – the asymmetric behavioral knowledge scenario

A more interesting situation to study, is when the manager is overconfident, but the investor is not aware of this. Hence, we have asymmetric behavioral knowledge about probabilities of project success. The manager knows the investor’s beliefs, and the contract is set up along these lines. However, the manager then makes his own decision whether to exert effort or not, while the investor believes that he has forced the manager to behave, namely by assuming that the contract is incentive compatible. The choice made by the manager depends on the level of his overconfidence and the amount of assets he has available. If overconfidence is too high or assets are too low, this will result in overinvestment. The manager shirks his responsibilities, and the investor has financed a project that has, in expectation, a negative net present value.

Since the manager and the investor have different beliefs regarding probabilities of success, we derive two different ICCs. One according to the investor’s rational beliefs and the “true” ICC which induces the desired behavior by incorporating the overconfidence of the manager. In particular, the manager uses the following inequality,

p

B

R

ICC

M

:

M

≥

_∆

_ˆ

, while the investor uses

ICC

I

:

R

M

≥

B

_∆

_p

.

ICCI is less restrictive compared to ICCM. Therefore, ICCI is dominant: intuitively, the investor thinks the

manager is induced to behave for a lower amount of remuneration. ICCI will be used in setting up the

contract.

PCL is just met:

(

)

H

I

_p

A

I

R

=

−

. By pledging this amount of income, the manager ensures that the investor is willing to finance, given that level of assets A is sufficiently high. The investor believes that by accepting

R

_I, the remaining

R

_M satisfies the ICC, which will induce the manager to exert effort on the project. Yet to behave, the manager needs at least

p

B

R

M

=

_∆

_ˆ

, which means that the ICCM is not

necessarily met, since the “true” ICC for the manager is not in line with the investor’s beliefs. Depending on variables A and

pˆ

_L, the manager will choose to exert effort or not. This depends on the utility that the manager receives from the payoff of the project. If the utility for the manager when he exerts effort,

E

U

, is higher than the utility when he shirks, U_S, the manager will behave and exert effort. If U_S is larger than

U

_E, the manager shirks, and the investor has invested in a project that has a negative value in expectation. In this case, there is overinvestment. Again, this tradeoff is characterized by ICCM. In the

competitive equilibrium we have the following equations for U_i, i=E,S, that are defined as the expected payoff of the project minus the expected reward for the investor and also minus the assets that the manager invests himself:

(

)

_A

_p

_R

_I

p

A

I

p

R

p

U

H H H H H H E

−

=

−

−

−

−

−

=

ε

ε

, (3.4)

(

)

_A

p

A

I

p

R

p

U

H L L L S

−

−

−

−

=

ˆ

ε

ˆ

. (3.5)

These are expectations as perceived by the manager, i.e. based on his beliefs regarding the probabilities of success. The correct expectations are different. To prevent overinvestment, we require that U_E ≥U_S.

Solving this inequality for

pˆ

L results in a measure of acceptable overconfidence,

ω

, where acceptable

means that if overconfidence is lower than this threshold, it will not lead to overinvestment:

(

)

H H L L

p

A

I

R

B

p

p

p

−

−

−

≤

+

=

ω

ˆ

(3.6)

available for this project. If this amount is equal to the minimum amount needed to secure financing, A= A , any level of overconfidence leads to overinvestment (

ω

=0). Higher levels of

A permit higher levels of

pˆ

_L , so

ω

is rising in A . Thus, depending on the variables

pˆ

_L and A , there will be overinvestment, financing with acceptable overconfidence, or no financing. Figure 1 depicts these three states.

Now, reviewing the analysis above, and using

p

ˆ

_H

>

p

_H, we see that this will lead to a higher expected payoff when the manager exerts effort. Therefore, he will always choose to behave, which is exactly what the investor wants.

There is a large difference in outcomes for the symmetric and asymmetric behavioral knowledge cases. The symmetric knowledge case results in more credit rationing. Thus, the manager will want to keep his prior hidden. This is the most interesting scenario, and in the following sections, I will focus only on the asymmetric behavioral knowledge case. Then, we can identify the room for acceptable overconfidence, measured in

ω

.

4. Allocating control rights in the FI Model with overconfidence

We extend the basic Fixed-Investment Model with overconfidence of section 3 to address the question of how to allocate control rights. We look at the effects of overconfidence in this extended model and its consequences for the optimal contract.

Initially, a contract is set up between a manager and an investor to govern financing of the project and the behavior of the manager. However, as time passes and the project develops, new information accrues and circumstances may change. The original outlook on which financing was based may no longer be valid. During the course of the project, there is a need to gather new information and to make short- and long-term decisions. In this section, I focus on a situation that deals with this information gathering and decision making, in particular on the question whether the investor or the manager should

Three panels show the outcomes for different levels of assets A compared to the minimum level required A. The solid line denotes the maximum payoff for the manager,

R

_M, up to which point the investor is willing to finance. The dotted line denotes the minimum payoff the manager needs to behave. The dashed line shows the minimum payoff the manager needs to behave when he is overconfident.

Figure 1:

be in control to make these decisions. The modeling approach is in line with the adapted Aghion and Bolton (1992) model as presented in Tirole (2006).

Introducing overconfidence in this model leads to intervals of levels of initial assets A along which there is room for overconfidence. Because control rights can be used to raise pledgable income, I establish a pecking order of control right allocation, with differing levels of acceptable overconfidence, the threshold where overconfidence does not lead to overinvestment. That is, there is a relation between the pecking order, the amount of initial capital and the level of acceptable overconfidence.

4.1 The basics of the Aghion-Bolton Model

Here, I present the Aghion-Bolton Model (AB Model) as an extension of the FI Model of section 3. The setup is similar: there is a project with return R, that requires investment I . The manager has assets available,A. The project succeeds with probability

p

=

p

H if the manager exerts effort, and with

probability

p

=

p

L if he shirks.

The basic FI Model is extended by introducing a possibility to take an interim action, before the project is completed. Taking this action raises the probability of success uniformly by

τ

>0. The probabilities of success become

p

_H

+

τ

and

p

_L

+

τ

. This interim action comes at a cost,

γ

>0, for the manager. For example, raising the probability of success may mean that the manager has to switch to a more routine and less exciting line of production, divesting a division that is the manager’s favorite, or severing a relationship with a friendly supplier or client. There are two possible scenarios: either the interim action reduces utility,

τ

R<

γ

, or it is beneficial to utility,

τ

R>

γ

.

If the interim action is beneficial to utility, both the investor and the manager will not have a problem taking the interim action, as this also raises pledgable income, which facilitates financing. In short, both the investor’s and manager’s interests are aligned, yielding trivial outcomes. Therefore, I look into the more interesting cases for which there is a conflict of interest, such that there is actually a role for control right allocation.

extra costs. Pledgable income rises to

(

)

      ∆ − + p B R

pH

τ

. However, utility for the manager falls to

(

+

τ

)

−

ε

−

−

γ

=

p

R

I

U

_{E H H} .

If the manager retains the control rights, he will be induced not to take the interim action, since this will only harm his utility. The exception to this situation is when the manager has insufficient funds to secure financing, but will receive funding if the interim action is taken:

(

)

      ∆ − + < − <       ∆ − p B R p A I p B R

p_{H H}

τ

. Then, taking the interim action raises pledgable income so that the project can be financed. Imperfections in the capital market, in the form of credit rationing, can therefore result in a situation where the manager is forced to take an action that is suboptimal, yet better compared to the alternative of not receiving financing. The only credible way to secure financing in this situation is by giving control rights to the investor.

4.2 Overconfidence in the AB Model

I define overconfidence in the same way as I did in the previous section. The manager is overconfident and believes that his skills and probabilities of success are higher than what the investor gives him credit for, leading to overoptimism about project payoff. I focus on the asymmetric behavioral knowledge scenario, where the investor holds the correct beliefs and the manager is overconfident but keeps his prior hidden from the investor. Furthermore, asymmetric information about moral hazard is still present. Again, the most interesting scenario arises when

p

ˆ

L

>

p

L

1

. Now, given the level of assets A, either the manager retains control rights, or they shift to the investor. When the manager remains in control, the outcomes are the same as in the FI Model, since the manager will not take the interim action. The threshold for acceptable overconfidence rises as the level of assets increases.

When control transfers to the investor, probabilities of success increase by

τ

and the manager incurs cost

γ

. Since the investor knows that the interim action is taken, the minimum required pledgable income falls to

(

)

(

+

τ

)

−

=

H I

_p

A

I

R

. All other things equal, the investor believes that this will force the manager to behave, as in the FI Model. Under these circumstances, utility equations for the manager become:

1

Figure 2: Initial assets and acceptable overconfidence

This graph depicts the minimum level of initial assets that the manager requires, and also what his level of assets means for the room for acceptable overconfidence. Initial assets establish a pecking order for control right allocation, and more assets leave more room for acceptable overconfidence.

(

+

τ

)

−

ε

−

−

γ

=

p

R

I

U

_{E H H} (4.1)

(

)

(

)(

)

₍

₎

γ

τ

τ

ε

τ

−

−

+

−

+

−

−

+

=

A

p

A

I

p

R

p

U

H L L L S

ˆ

ˆ

(4.2)

As before, to ensure full effort of the manager, we require that U_E ≥U_S, and solving for

pˆ

_L yields:

(

)

(

τ

)

ω

+

−

−

−

≤

+

=

H H L L

p

A

I

R

B

p

p

pˆ

(4.3)

This equation is essentially the same as equation (3.6). However, the lower minimum amount of initial assets is realized by the addition of the

τ

on the right-hand side. Since room for acceptable overconfidence at asset level A is 0,

τ

is the only variable that lets A be lower, all the more so because the other variables are exogenous.

Considering the different levels of A, acceptable overconfidence is only possible along certain intervals. By giving the investor the control rights, the minimum amount of assets the manager has to supply falls to A', which is defined as

(

)

(

(

p

)

R

I

)

p

B

p

A

_H

−

_H

+

−

∆

+

=

τ

τ

'

. Because

R

_M

p

B

_τ

τ

=

∆

Therefore, if

A

'

≤

A

≤

A

'

+

τ

(

R

−

R

_M

)

,

ω

>0 and there is room for overconfidence. As A reaches level

(

R

R

M

)

A

'

+

τ

−

, acceptable overconfidence drops to zero again, because this is the level of assets where the manager has no need to relinquish his control rights: A. Subsequently, acceptable overconfidence is rising in A, equal to the FI Model. See also Figure 2 on page 17. This figure also shows another interesting characteristic of these inequalities. The slope of acceptable overconfidence

ω

is higher to the right of the A point. This means that the marginal increase of

ω

from an increase in A is higher when the control rights remain in the hands of the manager. When the manager remains in control, and if the assets the manager has available increase, the level of acceptable overconfidence increases even faster.

4.3 Exploring other channels of overconfidence

If the manager is overconfident about the raised probability of success,

τ

, the consequences are limited. The analysis of

pˆ

L already revealed that

τ

drops out of the equation for acceptable overconfidence, so

τ

ˆ also drops out. There is, however, one situation where the beliefs of an overconfident manager result in an outcome that is suboptimal for the manager.

This is the case when

τ

R<

γ

, but the manager believes that

τ

ˆR>

γ

. The added expected income is not sufficient to cover the extra cost, so utility is lower when the interim action is taken. However, the manager believes that taking the action will result in a higher utility, as he thinks that the extra revenue is larger than the incurred cost. This leads to a situation where the manager will take the interim action, while he should not have done so, given that A> A. Overconfidence causes the manager to choose a suboptimal course.

5. Contingent control rights in the FI Model with overconfidence

A different approach to control rights is to make the allocation contingent on an observable event. In order to study this scenario, the AB Model is adapted to include an intermediate stage where a signal accrues that is informative about the final outcome of the project. This signal can be either high (H) or low (L). Control rights remain with the manager when the signal is high and transfer to the investor in case of a low signal.

Introducing the use of a signal changes the setup of the model, because the reward for the manager is no longer based on the actual outcome of the project, but on an informative signal. As in the AB Model, transferring control rights to the investor raises pledgable income. Using contingent control rights, the minimum necessary level of assets A falls even more.

5.1 The use of signals for monitoring

In this section, I first explain the characteristics of this model, without applying managerial overconfidence. The next section analyzes the use of signals to allocate control rights. Lastly, overconfidence is introduced. The AB Model is changed if we make use of a signal, but the basics remain the same. The project has the same characteristics, and there is still moral hazard involved, such that the manager can choose to exert effort or to shirk his responsibilities. However, after the manager has chosen his effort, information can be acquired that is indicative for the final outcome of the project.

This is organized as follows. First, the manager chooses his effort, i, to be high or low. After this, a signal, j , accrues, that is either high or low. The probability of signal j is denoted as

σ

ij : the

probability of signal

j

∈

{

H

,

L

}

, conditional on effort

i

∈

{

H

,

L

}

. Furthermore,

σ

_iH +

σ

_iL =1 for all i. After this, the project succeeds or fails with probability of success

ν

_j, conditional on signal j. See also Figure 3.

Figure 3: Effort and signals

These probabilities of success are ex ante equal to

p

_H and

p

_L, such that the following equations must hold: L LH H HH H

p

=

σ

ν

+

σ

ν

(5.1) L LL H LH L

p

=

σ

ν

+

σ

ν

(5.2)

Equations (5.1) and (5.2) have an impact on the equations for the ICC and the PCL. The reward for the manager can be made dependent on the signal or the final outcome. Manager compensation theory states that the optimal contract makes compensation contingent on outcomes that the manager can control. Since the manager can influence the signal by exerting effort, but has no control over the probability of success given that signal, it is best for the manager to be rewarded according to the signal, and not the final outcome. The manager will receive a reward in case the signal is high, since this is good news about the final outcome, and no reward in case the signal is low. Thus, the manager receives

R

M with probability

HH

σ

if he behaves, and with probability

σ

_LH if he shirks. As before, the reward should be sufficient to induce the manager to behave, resulting in the following ICC:

(

HH LH

)

M B R

σ

σ

− ≥ (5.3)

By the same reasoning as in the previous sections, this is introduced in the PCL, resulting in the following inequality to ensure that the payoff for the investor is high enough to cover his investment:

A

I

B

R

p

LH HH HH H

≥

−

−

−

σ

σ

σ

(5.4)

In Appendix III, I show that

LH HH HH L H H

p

p

p

σ

σ

σ

−

>

−

. This means that by using signals to monitor the

5.2 Signals and control rights

Signals can also be used as a means to govern the allocation of control rights. The control rights remain with the manager if the signal is high, and transfer to the investor if the signal is low. Then the decision whether to take the interim action or not is made by the party in control. If the signal is high, the manager will refrain from taking action, but if the signal is low, the investor will force the manager to take action.

Three scenarios arise with the use of a signal. The control rights can be allocated non-contingently to either the manager or the investor or they can be made contingent on the signal. Each scenario has a different outcome for the minimum level of assets the manager needs to invest to secure financing.

As in the AB Model, it is important whether the interim action is beneficial or harmful to utility. In the beneficial situation, the manager will also take the interim action, and managerial and investor interests are aligned. In the harmful situation, there is a conflict of interests and outcomes depend on who is in control. Hence, I present only the case where

τ

R<

γ

. In this section, I present the outcomes for a rational manager. Thus, there is only one ICC, which leads to a single PCL. In the next section, I will introduce overconfidence to this scenario. Then, the differences between symmetric and asymmetric behavioral knowledge will be more apparent.

5.2.1 Non-contingent manager control

Since the manager will not take the interim action if control is left to him, this scenario is identical to the one described in section 5.1. Equations (5.3) and (5.4) hold as the ICC for the manager and the PCL for the investor, resulting in the following minimum level of assets:

(

p R I

)

A B A H LH HH HH = − − − ≥

σ

σ

σ

(5.5)

Equation (5.5) is equal to equation (5.4) but rewritten in terms of A. Analogue to the scenario in section 5.1, the use of signals lowers the necessary minimum amount of capital the manager needs to supply compared to the situation without signals, thereby facilitating finance.

5.2.2 Non-contingent investor control

The investor will force the manager to take the interim action. Probabilities of success are raised uniformly by

τ

. The resulting PCL is defined as

(

p

)

R B I A

(

)

(

p R I

)

A B A _H LH HH HH = − + − − ≥

τ

σ

σ

σ

(5.6)

Adding the

τ

term lowers equation (5.6) compared to equation (5.5). Transferring control rights to the investor facilitates financing, but just as in the AB Model, this comes at a private cost for the manager. The private cost

γ

does not enter these equations, but does affect the utility derived by the manager. Again, the choice of taking the interim action is a tradeoff between financing under suboptimal circumstances or no financing at all.

5.2.3 Contingent control

In this scenario, control rights remain with the manager in case of a high signal, and shift to the investor in case of a low signal. When

τ

R<

γ

, the manager will not take action, while the investor will change course.

Since the investor will only finance if he believes the manager to behave, the interim action will only be taken if the signal is low despite the manager’s good behavior. Therefore,

τ

enters the equation with probability

σ

_HL. The manager will only incur the private cost when the signal is low, so now

γ

must also be included. If the manager retains control, he avoids this private cost, so

γ

is included as an extra reward in case of a high signal,

R

_M

+

γ

.

This results in the following inequality for the PCL:

(

p

)

R

B

I

A

LH HH HH HL H



≥

−











−

−

−

+

γ

σ

σ

σ

τ

σ

(5.7)

Solving for A yields:

(

)

(

p R I

)

A B A HH H HL LH HH HH − − + − = − ≥

σ

γ

σ

τ

σ

σ

σ

(5.8)

Comparing equations (5.5), (5.6), and (5.8) shows that the minimum required amount of assets the manager needs to supply falls in each of these three scenarios2. Thus, a pecking order similar to the one I

2

established in section 4 also holds in this scenario. Depending on the level of initial assets the manager has available, he will either retain control, hand over control to the investor, or make the allocation of control rights contingent on the signal.

5.3 Signals, control rights, and overconfidence

The question arises how overconfidence affects the outcomes of this model. Overconfidence becomes somewhat more subtle in the presence of informative signals. I need to adapt the definition of overconfidence that I used in the previous sections to incorporate the new probabilities of success. An overconfident manager believes his skills are superior,

p

ˆ

_L

>

p

_L. In the situation discussed here,

p

_L is defined as

p

_L

=

σ

_LH

ν

_H

+

σ

_LL

ν

_L. Along the same lines as the previous sections, overconfidence is expressed as a higher perceived probability of success. This means that the manager believes that those probabilities are higher which he can influence by his behavior and actions. Here, that is the probability of a certain signal, given the manager’s effort, defined by

σ

_ij. An overconfident manager believes the probability of a high signal is higher, even when he exerts less effort:

σ

ˆ

_LH

>

σ

_LH. Due to the definition of

σ

_ij, this implicitly means that

σ

ˆ

_LL

<

σ

_LL.

Overconfidence can now be determined in terms of

σ

_LH. These inequalities are different from the ones in section 4, but have the same properties. Acceptable overconfidence is rising in A , and relinquishing control facilitates financing.

5.3.1 Non-contingent manager control

The derivation of

σ

ˆ

_LH is similar to the derivation of

pˆ

_L. Again, the most interesting scenario is when the manager succeeds in keeping his prior hidden from the investor and

τ

R<

γ

. The following results all assume this asymmetric behavioral knowledge. Investor and manager interests are not aligned and the manager uses a different ICC than the “true” ICC that the investor follows. The investor believes the ICC for the manager to be equal to equation (5.3), while the manager acts according to the following ICC:

(

HH LH

)

M B

R

_σ

_σ

_ˆ

−

≥ (5.9)

Keeping his prior hidden is beneficial to the manager since ICCI is less restrictive compared to ICCM.

Therefore, ICCI is used in the PCL, which is equal to equation (5.4). This leads to the following equations

project only yields a return if it is successful, the equations for utility are principally the same as in previous sections. They are comprised of the expected total return of the project minus the expected reward for the investor minus the cost of effort by the manager. The investor is still offered the same competitive reward, which he believes will satisfy the manager’s ICC. Thus, the utility equations are:

I

R

p

U

_E

=

_H

−

ε

_H

−

(5.10)

(

)

(

)

(

I

A

)

A

p

v

v

R

v

v

U

H L LL H LH L L LL H LH S

−

−

+

−

−

+

=

σ

ˆ

σ

ˆ

ε

σ

ˆ

σ

ˆ

(5.11)

In the second utility equation, I already replaced

pˆ

_L with

σ

ˆ

_LH

ν

_H

+

σ

ˆ

_LL

ν

_L. In order to arrive at the overconfidence inequality, we require that U_E ≥U_S . The manager will then exert effort and in equilibrium, the investment yields an expected positive net present value. This results in the following equation for acceptable overconfidence. Again, this is expressed in terms of

ω

, the threshold where overconfidence does not lead to overinvestment:

(

)

         − −       − − ≤ + = H L H HH LH LH p A I R B v v 1 ˆ

σ

ω

σ

σ

(5.12)

For a detailed derivation, I refer to Appendix IV. Because overconfidence is now linked to the outcome of the signal, this inequality is slightly different from the ones derived in section 3 and 4. However, the setup is similar. The probability of a high signal under low effort depends on the ratio of the private benefit to the expected payoff of the project for the manager. Overconfidence, as measured by

ω

, is rising in A. If the manager has more initial assets, room for acceptable overconfidence is larger.

5.3.2 Non-contingent investor control

(

)

(

)

          + − −       − − ≤ + =

τ

σ

ω

σ

σ

H L H HH LH LH p A I R B v v 1 ˆ (5.13)

Analogue to the case in section 4, the added

τ

reduces the minimum level of initial assets. If initial assets are above this level, the threshold

ω

>0, which means there is room for acceptable overconfidence. In addition, the slope of

ω

is lower in this case, as the added

τ

lowers the impact an increase in A has on the level of acceptable overconfidence

ω

.

5.3.3 Contingent control

In this scenario, the outcome is different. This is due to the fact that both

τ

and

γ

enter the PCL equation. However, they enter the subsequent utility equations with different probabilities, depending on whether the manager exerts effort or shirks. Therefore, they do not drop out of the inequality for overconfidence. This becomes apparent in the utility equations:

(

H

σ

HL

τ

)

ε

H

σ

HL

γ

E

p

R

I

U

=

+

−

−

−

(5.14)

(

)

(

)

(

)

σ

γ

τ

σ

τ

σ

σ

σ

ε

τ

σ

σ

σ

LL HL H LL L LL H LH L LL L LL H LH S

I

A

A

p

v

v

R

v

v

U

ˆ

ˆ

ˆ

ˆ

ˆ

ˆ

−

−

−

ˆ

+

+

+

−

−

+

+

=

(5.15)

The necessary condition for managerial effort UE ≥US results in:

(

)

(

)

(

)

(

)

          − − + − − − +       − − ≤ + =

σ

σ

τ

τ

σ

γ

σ

σ

σ

ω

σ

σ

HL LL HL H LL HL L H HH LH LH p A I R B v v ˆ ˆ 1 ˆ (5.16)

6. Conclusions and discussion

In this study, I review general moral hazard models concerning financial contracting. In light of developments in the field of behavioral corporate finance, I try to adapt these models to allow for overconfidence and overoptimism, psychological biases found to be present in many individuals, which includes managers. I study the effects of overconfidence by adapting the moral hazard contracting model developed by Tirole (2006). I introduce overconfidence in this model by raising the perceived probability of success for the manager. I define overconfidence as overestimating skill, so that the manager believes the project will succeed even if he exerts less effort. This leads to a feeling of overoptimism on the part of the manager relative to the beliefs of the market, represented by the investor. This overoptimism results in a higher perceived payoff if the manager exerts less effort.

Under symmetric behavioral knowledge, when the investor knows about managerial overconfidence, these different priors lead to more credit rationing. It is therefore beneficial for the manager to keep his prior hidden. In this asymmetric knowledge scenario, the investor risks investing in a project that has a negative value in expectation. I have found that, depending on the level of initial assets the manager has available, overinvestment might occur. The room for acceptable overconfidence increases as the level of initials assets increase. Because more initial assets lower the amount of the loan, risk decreases for the investor. This leaves the manager with room for overestimating his probability of success, while he still chooses to behave. It is this choice that leads to an acceptable outcome.

This result also holds if the basic model is extended. I have reviewed the Aghion-Bolton model (Aghion and Bolton, 1992) concerning the use of control rights as presented in Tirole (2006), and the use of signals, making the allocation of control rights contingent on this signal. While these extensions raise pledgable income and facilitate financing, the effects on the threshold for acceptable overconfidence are the same. As the level of initial assets increase, there is more room for overconfidence that does not lead to overinvestment. While many studies on agency theory and overconfidence arrive at the conclusion that a moderate level of overconfidence in a manager mitigates moral hazard, my research shows that in a risk neutral setting, such an effect does not exist and that overconfidence leads to a small interval of efficient outcomes. Otherwise, overconfidence leads to overinvestment. I have shown that overconfidence causes the agency problem to amplify due to asymmetric behavioral knowledge, leading to increased moral hazard.

take the action that the investor would want him to take, and the investor has no need to force the manager by exercising control rights.

Now, overconfidence might affect the decision-making process of the manager in two ways. His overconfidence affects his perceived chances of success, and it affects the amount of assets he thinks he will need. My analysis shows that while both do have an effect on the decision made by the manager, this is not beneficial to either the manager’s utility or his decision to behave or shirk.

If overconfidence is to prevent control rights to shift to the investor, the following scenario would have to unfold: the manager lacks the necessary amount of assets to be able to retain the control rights, so in a rational setting, he would be forced to relinquish these rights. However, because the manager is overconfident, he feels that he should not be forced to grant the control rights to the investor. Also, due to his overconfidence, the manager will decide on a course of action that will lead to an acceptable outcome in equilibrium, without interference by the investor. This scenario does not seem very likely. An investor will want to take the interim action to increase the probability of success, thereby increasing the expected revenue in equilibrium. Increased expected returns will allow the investor to lend more funds, so the manager requires less assets of his own. Why then should the investor agree to let control rights remain with the manager and let the probability of success remain at the basic level? And is it even true that the manager will feel the same way? I think that this reasoning should go the other way around.

A manager believes his chances of success in case of shirking are higher than what the investor believes. According to the manager, he would need more assets to procure financing. His overconfidence leads him to believe that he needs more funds. If this is the case, the manager will be even more willing to transfer control to the investor, believing this is the only way to get his project financed.

An increased probability of success raises expected revenue. Thus, the overconfident manager believes that he will earn a higher revenue when shirking than the expected revenue based of fully rational beliefs. This will lead him to shirk more easily. Again, this contradicts the intuition that the manager will behave more according to the investor’s wishes because he is overconfident.

where it is not harmful, but in others it leads to outcomes that are downright unprofitable for all parties involved.

My results depend crucially on the assumption of risk neutrality of both parties. In the literature review section, I noted that many previous studies model their research in a setting with risk averse managers. This is motivated by the fact that a manager has his human capital invested in the firm, and would therefore rationally be risk averse. In my research, I have chosen risk neutrality to be able to separate the effect of risk aversion from moral hazard. For intuitional purposes, this is a sound approach that yields straightforward results. Alternatively, one could argue that managerial overconfidence adjusts risk preferences from risk aversion towards risk neutrality. Now, it should be possible to extend my research to a setting in which the manager, while being overconfident, is also risk averse. It is most likely that such an adaptation will lead to a larger difference between the expected utility from the two options, exerting effort or not. In such a scenario, a risk averse overconfident manager values his effort higher and will be more easily induced to behave. In a way, being risk averse cancels the effect of overconfidence and overoptimism. The higher value of behaving also leaves more room for acceptable overconfidence. This outcome, while at this stage only intuitive, is more in line with other research that presupposes risk averse managers. This is a possible avenue for future research in this area.

Another point which is yet unanswered, is whether the pecking order from sections 4 and 5 can be combined to form a single pecking order that the manager will pursue. Unfortunately, it cannot be said whether the scenario with manager control without signals yields a higher or lower necessary amount of initial assets than the scenario with investor control with signals. Because the choice whether to use signals or not leads to two completely different scenarios, a useful comparison cannot be made. I cannot say whether one yields a higher necessary level of initial assets than the other. The two pecking orders must be viewed separately, and can only be implemented after the choice for the use of signals has been made.