Optimal Capital Allocation Using RAROCTM and EVA r

Neal M. Stoughton Josef Zechner

UC Irvine University of Vienna

January, 1999

0
RAROC is a trademark of Bankers Trust and EVA is a registered trademark of Stern Stewart & Co. This
paper has been presented at seminars at the European Finance Association, Amsterdam, Frankfurt, Hong Kong
University, Houston, Konstanz, Mainz, NYU, UC Irvine, the Stockholm School of Economics and Wisconsin. We
appreciate the comments of participants at these seminars. We especially acknowledge the helpful comments of
Wolfgang Bühler and Bruce Grundy.
 Abstract

This paper analyzes financial institutions’ capital allocation decisions when their required equity capital

depends on the risk of their projects chosen. We discuss the relevance of strict position limits against

discretionary trading through the use of an optimal compensation function. We show that (under

full information) the first-best investment decision can be delegated through an economic value added

(EVA) compensation contract and solve for the optimal capital allocation rules. We demonstrate how

the concept of incremental Value at Risk must be used to deal with the multidivisional firm. The results

are extended to deal with asymmetric information on the part of the trading division(s). The analysis

defines precisely the notion of risk adjusted return on capital (RAROC) and how it can be used as a

performance measure.
1 Introduction

Governmental regulatory authorities as well as international financial organizations focus significant

attention on the risk exposures, capital adequacy and the risk management practices of banking and

financial service industries. This contrasts with the traditional corporate finance paradigm, based on

frictionless markets, which leaves little scope for risk management considerations.

The banking sector does not fit well in this paradigm, however. The choice of capital structure

of financial institutions merits special consideration. There is widespread agreement that banks favor

debt-financing. Banks, by their very nature, are in the business of accepting deposits, which implies a

bias in favor of debt financing. The provision of services such as liquidity through demand deposits,

letters of credit etc. imply that depositors are willing to loan their funds at rates below those prevailing

in money markets. By contrast, in order to raise equity capital, banks must compete with other non-

financial firms in the security markets. Thus, in the absence of governmental or self-regulation, banks

will prefer the maximum use of debt finance.

Therefore, regulating bank’s capital structure and/or risk-management may be desirable for a num-

ber of reasons. First, banks’ role as financial intermediaries requires that their debt is held by a large

number of small depositors. This in turn would make it very costly for each depositor to monitor

the risk of a particular bank. It may therefore be optimal to regulate banks’ risk exposures. Second,

the presence of deposit insurance may create incentives for excessive risk taking of the bank. Third,

negative externalities imposed on the rest of the economy by a risky and highly levered banking sector

may require regulation. Many countries have therefore implemented the recommendations of the Basle

Banking Committee and define capital requirements on the basis of various risk measures. In addition

to monitoring risks to comply with regulation, many banks and financial institutions have also adopted

their own internal risk management systems designed to measure and limit their risks in accordance

with their equity capital.

Although in such an environment equity capital requirements will be determined at the firm level,

it is important to allocate capital at the divisional or project level. The key question involved in capital

allocation is how much equity capital to charge each unit with so that the effect of individual investment

decisions on the overall level of risk is internalized. In this way the capital charge to the division may be

1
different from the actual amount of capital required by this division’s investments. The most dramatic

example is a division entirely engaged in derivatives trading for which no literal cash investment may

be required.

In standard capital budgeting contexts the need to assess project-specific systematic risks for the

purpose of deciding on an appropriate cost of capital is well-documented. Less well-understood, however,

is the need to incorporate project-specific capital structure weights. Stulz (1998) discusses these issues in

the context of risk management and demonstrates that the capital budgeting process must incorporate

an adjustment to net present value to represent the impact of project-specific unsystematic risk on the

capital of the firm.

Our paper addresses several central issues related to the risk management of financial institutions.

There is considerable interest from the financial community in utilizing Economic Value Added (EVA)

as a compensation mechanism.1 A related measure, risk adjusted return on capital (RAROC) is also

applied to assign performance rankings across divisions. The main purpose of our paper is to derive

an optimal capital allocation mechanism so that the utilization of these techniques is appropriate to

achieving overall value maximization from the perspective of shareholders. Essentially the optimal

capital allocation mechanism we derive requires the institution to compute an economic capital, rather

than a book capital for use in the EVA and RAROC computations. Moreover, we show that the

economic capital is related to the widespread use of value-at-risk (VaR).

Despite these important developments, a puzzle has remained unresolved. In a multidivisional

setting, where risk management activities need to be coordinated, how can the institution decentralize

the use of EVA compensation at the divisional level? If divisional VaR is utilized for economic capital,

then the sum of divisional economic capital exceeds firm-wide economic capital due to the correlation

structure of investments. Ad hoc procedures, such as those in Kimball (1997) have been proposed, but

these may create significant distortions in the decisions taken.

Froot and Stein (1998) discuss this problem of divisional interdependence in a model in which risk

management arises endogenously from the need to avoid an adverse selection problem with respect to

external finance. They indicate the set of conditions a RAROC measure would have to satisfy in order to
1
See, for example, Uyemura, Kantor and Pettit (1996), in the banking context.

2
produce optimal decisions using a bank-wide hurdle rate criterion. We extend this research by analyzing

specific forms of capital allocation and RAROC and their effects on delegated investment decisions. We

also model explicitly the influence of asymmetric information on optimal risk adjusted performance

measures in an agency framework. If the institution uses an EVA-based incentive compensation system

with an appropriately designed economic capital allocation, then we demonstrate that the firm-wide

optimal risk decision can be implemented through local control. RAROC is more appropriately used in

our view as an ex post performance measure, as opposed to a bank-wide hurdle rate determination.

The method we propose involves the central authority of the institution specifying a mechanism

under which divisions are allocated economic capital equal to a division-specific price multiplied by

their own value-at-risk. The price is chosen appropriately so that the economic capital is equal in

realization to the incremental value at risk (IVaR), which takes into account the externality that one

division’s risk imposes on that of the other divisions within the institution. Moreover, this mechanism

has the virtue that the sum of divisional economic capital is exactly equal to the overall VaR of the firm.

In some ways our results are therefore reminiscent of the literature on internal capital markets (Stein,

1997). However an important distinction is that the “price” must be personalized for the division’s own

investment opportunities. The use of this risk pricing mechanism “separates” the investment decisions

in a way that allows each division to act independently of the others.

We model a bank whose equity capital must exceed an amount defined by its VaR, that is, by a

maximum acceptable bankruptcy probability. Each of the bank’s business units must decide among

various investment projects with different expected returns and standard deviations. Since investment

decisions are delegated, the bank’s center must induce an optimal investment decision consistent with

the capital structure constraint. Initially we consider a situation in which there is full information

within the firm about investment opportunities. This, of course, is not the realistic case. However it

allows for a relatively simple exposition of the main ideas, which are then subsequently extended to deal

with asymmetric information.

In two relevant papers, James (1996) and Zaik, Walter, Kelling and James (1996) analyze the cap-

ital budgeting process at Bank of America where equity capital is defined by a value at risk measure.

Milbourn and Thakor (1996) consider an asymmetric information model of capital allocation and com-

3
pensation. They focus on the moral hazard problem of managerial effort as well as enhanced bargaining

opportunities for the central authority.

One of the original papers to look at capital budgeting under asymmetric information and solve for

the amount of inefficiency induced was Harris, Kriebel and Raviv (1982). Harris and Raviv (1996) and

Harris and Raviv (1998) analyze capital budgeting decisions in the presence of asymmetric information

about project quality and empire building preferences by divisional managers. At a cost, headquarters

can obtain information about a division’s investment opportunity set. The paper demonstrates under

which circumstances headquarters will delegate the decision how to allocate capital across projects and

what form this delegation may take. Our paper focuses on capital budgeting decisions of banks which

are facing capital adequacy restrictions based on the risk of the projects chosen by the different business

units. In particular we consider capital allocations for investment decisions which do not require any

initial investment.

The remainder of the paper is structured as follows. Section 2 discusses the economic and in-

stitutional environment and develops the model. Section 3 provides the analysis for the case of full

information; whereas section 4 extends the model to the case of asymmetric information. The most

general setup appears in section 5. Section 6 concludes.

2 Model Development

Financial institutions and banks, in particular, face market imperfections such as costs of financial

distress costs, transactions costs in accessing capital markets, or simply regulatory constraints. These

frictions imply that risk management, capital structure and capital budgeting are interdependent. We

begin by discussing the nature of investment opportunities faced by the financial institution.

2.1 Investment Opportunities

We first specify how a business unit’s investment opportunities are modeled. A financial institution con-

sists of n divisions, each of which may choose investment projects, defined by their standard deviations

4
of cash flows to equity holders. Expected cash flows at the end of the period are given by

µi = µi (σi , θi ) (1)

where σi is the standard deviation of cash flows of division i and θi defines the investment technology

specified below.2

We assume that, ceteris paribus, more “aggressive” risk taking by a division translates into higher

expected returns, i.e.

∂µi
> 0. (2)
∂σi

To ensure interior solutions we also assume that the investment technology is concave, ∂ 2 µi /∂σi2 < 0,

and that lim(σi →0) ∂µi /∂σi = ∞ and lim(σi →∞) ∂µi /∂σi = 0.

The parameter θi defines the functional relationship between risk and expected return. Specifically,

we assume that the following cross-partial derivative condition applies:

∂ 2 µi
≥ 0. (3)
∂σi ∂θi

This is the well-known Spence (1973) - Mirrlees (1971) sorting condition, that is critical in the analysis

under asymmetric information. Condition (3) implies that a higher θi makes “risk” more productive

and makes a higher σi more desirable.

2.1.1 Examples

Equation (1) is a reduced form, representing the composition of a number of concurrent operating

activities of financial institutions. We provide two specific descriptions of activities that underlie the

specification in equation (1). The first example is that of a bank which engages in deposit-financed

lending activity. We hereby assume that deposit rates are below the expected return on loans. This

may be due to imperfect competition between banks or due to fixed costs in setting up a bank. We

normalize the interest rate paid to depositors to zero.

2
In order to generate the functional representation of investment opportunities embodied in (1) it may be necessary to
apply an efficiency criterion to a more general investment opportunity set.

5
 Let L denote the amount of lending undertaken. The marginal loan quality is decreasing in the

aggregate amount of lending and so we can represent the expected net cash flows from lending activities

by µ = f (L) with f being an increasing concave function. Suppose, for instance, that the aggregate

risk is increasing linearly in aggregate lending, so that σ = sL with s > 0. We can therefore rewrite the

expected cash flows in the following way

σ
µ = f ( ) ≡ µ(σ).
s

Therefore equation (1) characterizes the deposit-financed lending activity described above.

The next situation we consider is that of a trading division engaged in self-financing derivatives

trading. That is zero net investment positions are taken in financial instruments such as forwards,

futures, swaps etc. These activities may either be thought of as arising from offering structured products

to corporate customers or from trading in the capital markets based on proprietary valuation models.

Let σ represent the risk of the cash flows generated by the derivatives exposure. Similar to the previous

example, suppose that σ is linear in the face value and that the expected cash flow is an increasing and

concave function of the face value of the derivative position. This concavity may be due to a number

of sources, such as price impact, or estimation risk in the valuation model used to derive expected cash

flows.

Although we have given two examples of activities that financial institutions are involved in, our

model can be interpreted to reflect a variety of other investment activities. These investment activities

may be interpreted in our model, without loss of generality, as requiring zero initial cash outlay. Positive

initial cash investments can be modeled by assuming that they are partly debt financed. Cash flows are

always defined after appropriate interest costs.

2.2 Capital Structure Regulation

Financial institutions frequently find it advantageous to minimize their use of equity capital. This occurs

for a number of reasons. First, due to liquidity and security offered to depositors, bank deposits are

considered to be a cheap source of capital. Second, as in other industries, debt offers a tax subsidy due

6
to deductibility of interest. Third, as emphasized by Merton and Perold (1993), a bank’s customers are

frequently also its debtholders so that the nature of banking requires leverage. The more “customers”

a bank wants to attract, the more debt it must be prepared to accept in its capital structure. In our

model, we capture this effect by assuming that the after-tax cost of deposits is zero.

On the other hand, depositors and other debtholders are only willing to lend their capital to a

bank if it has a sufficient amount of equity capital to ensure its solvency. In addition regulators specify

minimum equity standards which are based on various risk measures. Thus, a bank’s capital structure

is determined on the one hand by several advantages of debt, and on the other hand either by the

bank’s own desire to limit costs of financial distress or by regulatory constraints. If the equity capital

required is greater than the investment outlay, it is invested in the riskless asset. This is assumed to

incur deadweight costs which, as in Froot and Stein (1998) may be interpreted as a tax. To simplify

the exposition, we normalize the bank’s after-tax return on the riskless asset to zero.

The cost of equity capital raised by the institution, r > 0, will generally depend on the exposure

of cash flows to various risk factors. While the question of the pricing of risk and the determination of

the cost of capital is an interesting one, we abstract from this issue in this paper and assume that r is

constant.

To enable the bank to realize the maximum benefits of leverage without violating regulatory con-

straints or incurring excessive transactions costs, most financial institutions have adopted sophisticated

risk management systems to monitor and limit their risks in accordance to their equity capital. The

basic concept is to specify a maximum loss, which may only be exceeded with a specified probability.

This loss is referred to as Value at Risk (VaR). The value at risk is a function of the entire cash flow

distribution and is measured by looking at the point on the lower tail where the probability is equal to a

specified threshold value. When cash flows are normally distributed, the Value at Risk can be expressed

as

VaR = ασ, (4)

where α is a factor defined by the probability with which the actual loss may not exceed the VaR;

7
σ is the standard deviation of cash flows on the institutions’s portfolio over a specified time period.3

Note that the definition of VaR in equation (4) does not take the expected return on investments into

account.4 There are several reasons why VaR is generally defined ignoring the mean return. First, since

VaR calculations are usually made for short periods of time, i.e., for one to ten day periods, the first

moments are of second order. Second, most VaR models used in practice do not take the mean return

into account when calculating risk and, finally, regulators are conservative with respect to allowing

banks’ capital standards to be reduced by modeling the mean returns.

Bank management as well as regulators increasingly define capital requirements in terms of its VaR.

More precisely, the equity capital of the bank, C, is given by the constraint5

C ≥ VaR = ασ. (5)

According to equation (5) a bank is limited in its risk taking such that the resulting VaR does not exceed

the amount of equity capital. This capital structure constraint ultimately determines the required capital

allocation to investment projects or divisions.6

2.3 The Institution’s Objective and Managerial Compensation

Based on the investment and regulatory environment described above, the net present value of the

financial institution is
P
i µi + C
NPV = − C, (6)
1+r

where r is the cost of capital. Maximizing NPV is equivalent to maximizing the economic value added

(EVA), where
X
EV A = µi − rC. (7)
i
3
For a detailed analysis of the concept of Value at Risk, see Jorion (1997). Stulz (1998) demonstrates it’s applicability
in risk management contexts.
4
Alternatively, VaR can be defined as VaR = ασ − µ, where µ denotes the expected cash flow on the bank’s investments.
5
Our formulation can be easily extended to the case where a non-regulated firm wishes to satisfy a constraint on its
debt rating; here one may wish to redefine the VaR as mean-adjusted.
6
The Basle Accord has been amended as of January 1, 1998 to allow banks to use their own internal models to assess
risk. In this case a standard of 99% for ten days times a multiplier of between 3 and 4 is used for the bank’s trading
activities. That is, α is between 6.9 and 9.2, and σ is defined as the ten day standard deviation.

8
EVA represents the contribution to shareholder value.7

In our model the central authority of the financial institution wishes to maximize EVA after com-

pensating the various divisions. Hence the institution’s objective function is

X
I = EV A − Ui , (8)
i

where Ui denotes the transfer or compensation of business unit i. The transfer to the divisional managers

must achieve two objectives. First, it must create incentives for the manager to expend some minimum

effort to ensure a strictly positive θi . Second, the transfer to the division must induce optimal investment

decisions for given θi .

Consistent with the overall objective of the firm, we consider the divisional compensation to be

determined by divisional EVA as in

Ui = γ(µi (σi , θi ) − rTi ), (9)

where Ti is a function of certain variables that can be contracted upon. We refer to Ti as the capital

allocation function. Thus, the manager receives a fraction of the economic value added by his division

given the specified capital allocation rule. It is important to recognize that the sum of allocated capital,
P
i Ti , does not necessarily add up to the bank’s total capital, C. The optimal functional form of Ti is

determined in the subsequent sections.

In the next section we solve for the optimal investment decision under full information from the

standpoint of the institution. Then we show how this decision can be delegated at the divisional level.

3 The Full Information Optimization

In this section we develop the model under the conditions of full information, where the central authority

of the financial institution and the trading division have the ability to contract fully on outcomes. The
7
The correspondence between EVA and net present value (NPV) was perhaps first discussed by Preinreich (1937). The
precise specifications of EVA and the corresponding depreciation schedule are discussed in Rogerson (1997) and Reichel-
stein (1997a), who showed that EVA-type measures represent the unique optimum among the class of linear measures.
Unfortunately this uniqueness result depends critically on an assumption that cash flows are certain. The results are
extended to the case of moral hazard in Reichelstein (1997b).

9
main purpose of this section is to derive the first-best outcome in order to facilitate comparison with

the second-best mechanisms considered in the next section.

3.1 The First-best Problem

In the first-best situation, the institution makes all the decisions and compensates the trading division

according to some minimum utility level. Given the information θ, the institution solves:

max E[µ(σ, θ) − rC] (10)

σ,C

subject to

C ≥ ασ. (11)

Equation (11) represents the regulatory constraint on total capital such that the VaR is limited at the

appropriate level. The cost of capital, r, is applied to this amount to yield the EVA objective (10).

Note that this objective function assumes that the institution is risk-neutral with respect to θ.

At this point we distinguish between two forms of capital constraints. We call flexible capital the

situation in which the institution can adjust the amount of capital in response to information. That is

C becomes a function of θ in this case. The second regime, which we call inflexible capital occurs when

C must be chosen ex ante, before θ is known.

First, consider the inflexible capital case. C is raised and then may be used to satisfy the VaR

constraint. If there are no alternative uses of capital other than for the set of investments represented

by µ, then once C is fixed, the optimal decision is to select the highest possible σ subject to (11).

This implies that the decision on risk will be independent of the information embodied in θ. It is

straightforward to see that the optimal decision is given (implicitly) by the following condition:

E[µσ (σ, θ)] = rα, (12)

where µσ is the partial derivative with respect to the level of risk.

Second, consider the flexible capital case. Now the amount of total capital can be tailored to reflect

10
the expected returns on the investment decision. Substituting C = ασ from equation (11) into (10) and

maximizing pointwise over σ gives the following implicit expression for the optimal investment decision:

µσ (σ, θ) = rα. (13)

For instance, suppose µ(σ, θ) = θ log σ. Then the first order condition given by (13) yields σ =

θ/(rα). Thus, the optimal allocation of risk is increasing in the information variable and decreasing

with respect to the cost of capital and the severity of the VaR requirement.

3.2 Delegation

We now consider what happens when the trading decision is delegated to the manager of a single

division. The manager of the division begins with an expenditure of effort. If a minimum level of effort,

e∗ , is provided then the manager learns the information parameter θ. If this minimum level of effort is

not expended, then θ = 0.

The manager of the division is compensated according to some share, γ, of the divisional EVA,

µ(σ, θ) − rT (σ), where T (σ) represents the capital allocated to the division. We assume throughout

that it is optimal for the minimum level of effort to be provided by the manager. When effort is

unobservable, the center must provide incentives through γ so that effort is undertaken. That is, γ

must be set equal to the value such that

E[γ(µ(σ, θ) − rT (σ)) − e∗ ] ≥ E[γ(µ(σ, 0) − rT (σ))]. (14)

Throughout the analysis, we assume that the headquarters sets γ > 0 to the value that solves (14).

Therefore we focus on the subproblem that ensues once θ has been revealed. Figure 1 illustrates the

sequence of decisions in the delegation problem.

3.2.1 The Institution’s Problem

The financial institution maximizes its share, (1 − γ), of the economic value added subject to the

regulatory requirement that the total capital, C is at least equal to the amount prescribed by the value

11
 Figure 1: Sequence of Events

Institution Division Mgr. Division Mgr. Portfolio Payouts

chooses expends makes choice according to
mechanism effort investment observed EVA
Ti = capital learns θi decision risk measured
allocation
as a function
of σi = risk

at risk, VaR. Given that the financial institution delegates the decision on the risk level to the trading

division, the overall problem to be solved is as follows:

max I = E[µ(σ, θ) − rC − U ] (15)

σ,σ̄,T (σ),C

subject to

σ ∈ arg max γ(µ(σ, θ) − rT (σ)), (16)

σ≤σ̄

U = γ(µ(σ, θ) − rT (σ)) ≥ U (17)

C ≥ ασ. (18)

Equation (18) is the regulatory constraint that assures that the financial institution must raise sufficient

capital from the external market so that the VaR condition is satisfied. This capital is evaluated at

the cost r in the objective function (15), in which the institution must subtract the amount of EVA

allocated to the division. The delegation constraint is represented by (16) implying that the optimal

investment decision is optimal from the standpoint of the divisional manager relative to EVA given the

capital allocated to him, T . This formulation of the delegation problem allows the institution to set a

position limit, σ̄. Further the manager must be given some minimum (reservation) utility in order to

be willing to commit the effort required to make the observation. This is embodied in (17).

Suppose that capital is flexible so that it can be chosen after information is known. Further, suppose

that T (σ) is differentiable and weakly convex and consider the first-order condition of (16). The solution

12
to this problem then gives:

µσ (σ, θ) = rTσ .

Notice that this is independent of γ. By contrasting this expression with the first-best solution (13),

it is clear that if Tσ = α then the first-best risk allocation will be achieved in the delegation problem.

Integrating this over σ then yields the following capital allocation function:

T (σ) = ασ + ν, (19)

where ν is some constant. If σ ∗ denotes the first-best risk decision, then the constant is determined by

the reservation utility constraint. Substituting into (17) yields

rν = µ(σ ∗ , θ) − rασ ∗ − (U /γ). (20)

We have shown that with flexible capital, the optimal capital allocation scheme (19) consists of

two parts. The first part, which we term economic capital is VaR and is linear in the risk of the

division. The second part is a constant adjustment factor, ν. In the full information problem this

adjustment is negatively related to the amount of reservation utility required to keep the manager.

Holding the reservation utility constant and varying the productivity of investment, this adjustment

factor is positively related to the EVA (µ − rασ) evaluated at the optimal risk level.

3.3 RAROC

We next explore the implications of the RAROC performance measure. As discussed in the introduction,

many banks use the concept of RAROC for capital allocation and performance evaluation.8 While there

are numerous different definitions of RAROC used in practice, it is most commonly used in the following

way:
(expected) return minus risk adjustment
RAROC = .
economic capital
8
See Matten (1996) for a very complete discussion of the various procedures for risk adjusted performance measurement.

13
One of the major contributions of the paper is in precisely characterizing the nature of this RAROC

measure. Specifically we shall illustrate exactly how the risk adjustment should be determined as well

as how economic capital should be computed.

Consider the following implementation of RAROC:

µ(σ, θ) − rν − rασ
RAROC = . (21)
ασ ∗

In this case, it is clear that net income is adjusted for realized risk in the numerator of this RAROC

measure. The denominator is exactly equal to the economic capital term from equation (19) evaluated

at the optimal risk level. Using this definition, it is easy to see that the institution’s EVA is equal to

EV A = (RAROC)ασ ∗ .

That is, EVA is the risk adjusted return on capital as applied to the allocated economic capital.

To study how this performance measure is related to the amount of optimal trading risk, we consider

the expected value of RAROC. Under full information, using the above capital allocation schedule, the

optimal risk level, σ ∗ will always be attained. To study the behavior of RAROC in this section, we will

consider the possibility that somehow the wrong investment decision was made. Substituting (20) into

(21) yields the following equivalent RAROC measure:

µ(σ, θ) − rασ − (µ(σ ∗ , θ) − rασ ∗ ) + (U /γ)

RAROC = . (22)
ασ ∗

It is clear from equation (22) and the fact that σ ∗ is optimal that this measure of RAROC attains its

maximum when the optimal decision is taken. At this point,

U /γ
RAROC = .
ασ ∗

We summarize our result for the case of flexible capital in Proposition 1.

Proposition 1 The optimal flexible capital allocation scheme, (19) consists of value at risk plus an

14
adjustment factor related to the potential economic value added. Using the optimal flexible capital

allocation scheme, the expected risk adjusted return on capital (RAROC) is maximized at the optimal

level of risk, σ ∗ .

Proposition 1 shows that RAROC is a perfect performance measure under the assumptions of full

information. In general it will be positive at the optimal decision, with the magnitude of this value

being related to the compensation necessary to induce the manager to become informed. For suboptimal

levels of risk it is entirely possible that RAROC becomes negative.

Now consider what happens with inflexible capital. Again, supposing that capital cannot be put

to alternative uses once it is raised, the first-best analysis has shown that the institution desires to

implement some (constant) risk decision, σ0 , independent of θ. It is easy to see that if the institution

uses a position limit σ̄ = σ0 coupled with a constant capital allocation T (σ) = ν, then the institution

can implement the first-best in the delegation case with inflexible capital. This follows because the

divisional manager will have the incentive to increase σ up to its maximum limit given in the constraint

to (16).

We thus obtain the following proposition.

Proposition 2 In the case of inflexible capital, the financial institution optimally delegates the in-

vestment decision to the division through a pure position limit coupled with a capital allocation that is

independent of the risk attained. Using the optimal inflexible capital allocation scheme, the expected risk

adjusted return on capital (RAROC) is maximized at the optimal level of risk, σ0 .

3.4 Multidivisional Capital Allocation

We now consider the problem of a multidivisional firm with a central authority under complete infor-

mation. As before we derive the first-best solution and show how it can be implemented in a delegation

environment. The basic aspects of the single divisional model are preserved in this environment.

First consider the institution’s problem when all decisions are centralized. With two divisions,

let µi (σi , θi ) represent the expected cash flows from taking risk level σi for division i; i = 1, 2 given

information θi . The overall risk of the portfolio of investments from the two divisions is defined to be

15
σp where

σp (σ1 , σ2 )2 = σ12 + 2ρσ1 σ2 + σ22 . (23)

For simplicity, we assume that information θi pertains only to trading division i and that it is indepen-

dent across divisions.

The first-best problem therefore becomes:

max E[µ1 (σ1 , θ1 ) + µ2 (σ2 , θ2 )] − rC, (24)

σ1 ,σ2

where

C ≥ ασp (σ1 , σ2 ). (25)

We focus in this subsection on the flexible capital case; the inflexible capital situation would be handled

analogously. As before, constraint (25) obviously holds with equality; substituting into (24) gives

max µ1 (σ1 , θ1 ) + µ2 (σ1 , θ2 ) − rασp (σ1 , σ2 ).

σ1 ,σ2

The necessary condition for optimality then becomes:

∂µi (σi , θi ) ∂σp

= rα .
∂σi ∂σi

Simplifying this expression using (23) gives the following result:

" #
∂µi (σi , θi ) σi + ρσj
= rα , (26)
∂σi σp

for i = 1, 2; j 6= i.

To interpret this result we use the concept of incremental value at risk (IVaR), as defined here:

Definition The incremental value at risk, vi (σi , σj ) for division i is defined as

σi2 + ρσi σj
vi (σi , σj ) = α . (27)
σp

16
It can be shown that the incremental value at risk for each division is [∂(ασp )/∂σi ]σi . The incremental

value at risk is proportional to the regression coefficient from a regression of the cash flows of division

i on the institution’s overall portfolio. Specifically, if βip is the regression coefficient, then vi = αβip σp .

Not surprisingly therefore, the incremental value at risk has the property that ασp = v1 + v2 , i.e., the

sum of the IVaRs equals the institution’s overall VaR.

Substituting this definition into the first-order condition above gives the following representation for

the optimal investment decision of each division:

∂µi (σi , θi ) vi
=r . (28)
∂σi σi

That is, investment occurs at the point where the marginal increment in expected returns is balanced

by the cost of the ratio of the incremental value at risk to the risk level chosen.

3.4.1 Delegation in the Multidivisional Problem

Given the perfect observability of the center, it is straightforward to find a delegation contract that

implements the first-best. Indeed, suppose that the central authority establishes capital allocation rules,

Ti (σ1 , σ2 ) for each division. Notice that this assumes that the capital allocation rules are interdependent.

The need for this derives from the possible correlation of returns across trading divisions.

In the case of flexible capital, the formal delegation problem solves:

max I = µ1 (σ1 , θ1 ) + µ2 (σ2 , θ2 ) − rC − U1 − U2 (29)

σi ,σ̄i ,Ti (σ1 ,σ2 ),C

subject to

σi ∈ arg max γ(µi (σi , θi ) − rTi (σ1 , σ2 )), (30)

σi ≤σ̄i

Ui = γ(µi (σi , θi ) − rTi (σ1 , σ2 )) ≥ U i (31)

C ≥ ασp . (32)

As in the single division case, the key constraint governing the incentives provided to the division

17
is present through (30). As long as the position limit constraint σ̄i is not operative, the first order

condition for optimal incentives on the part of the division is

∂µi ∂Ti
−r = 0. (33)
∂σi ∂σi

Therefore (from (28) and (33)) the implementation condition requires that at the first-best risk levels,

σi∗ and σj∗

∂Ti σi∗ + ρσj∗
=α . (34)
∂σi σp∗

It therefore follows by integrating (34) that the optimal capital allocation functions are of the following

form:
" #
σi∗ + ρσj∗
Ti (σ1 , σ2 ) = α σi + νi (σj ),
σp∗

where νi (σj ) denotes a function that depends only on the risk taken by the other division.

Expressing this result in terms of the IVaR, we obtain the following characterization result:

Proposition 3 In the multidivisional capital allocation problem under complete information, the opti-

mal capital allocation scheme to each division satisfies

Ti (σ1 , σ2 ) = νi (σj ) + λi (σi∗ , σj∗ )σi , (35)

where
vi (σi∗ , σj∗ ) vi∗
λi (σi∗ , σj∗ ) = ≡ (36)
σi∗ σi∗

and
Ui
rνi (σj ) = µi (σi∗ , θi ) − rαvi (σi∗ , σj ) − . (37)
γ

Notice that according to the the mechanism derived above, a division’s capital allocation is a constant

adjustment factor plus an amount of economic capital that is proportional only to its own standard

deviation, σi . The proportionality factor, λ, however, depends on its marginal contribution to portfolio

risk, ∂σp /∂σi , evaluated at the optimal risk levels. In the multidivisional model, λ, can be interpreted

as the “price” of division-specific risk. In solving the delegation problem, the divisional manager sees

18
the impact of the divisional risk on other divisions only through the price itself. There is no need for

one division to know the investment opportunities of other divisions.

Note that in the mechanism (35) above the amount of capital allocated to division i at the optimum

risk level is equal to vi + νi . Since the IVaRs of the two division add up to total institution-wide VaR,

the total capital allocation at the optimal risk level is given by

T1 (σi∗ , σ2∗ ) + T2 (σi∗ , σj∗ ) = ασp (σ1∗ , σ2∗ ) + ν1 (σ2∗ ) + ν2 (σ1∗ ). (38)

As in the single-division case the capital allocation levels are clearly related to the reservation utilities

of the trading divisions. However, the key difference is that the variable capital allocation depends both

on the standard deviation chosen by the division and on the covariance with the other division, evaluated

at the optimal risk levels.

Once again, it is possible to characterize the RAROC performance measure. The definition consistent

with the single division analysis is:

µi (σi ) − rνi (σj ) − rλi σi

RAROC = . (39)
λi σi∗

This definition provides the result that the EVA of division i at the optimal risk level is equal to RAROC

times the amount of economic capital as given by the optimal IVaR:

EV A = (RAROC)vi∗ , (40)

where vi∗ is the optimal incremental value at risk of division i. While the characterization of RAROC in

(39) is similar to the single trading division, here we notice that the denominator weights the standard

deviation of the division by its marginal contribution to portfolio risk. The performance measurement

properties of RAROC are as in the single trading division case, as indicated in the following proposition.

Proposition 4 In the multiple trading divisions case the expected risk adjusted return on capital for

each division as defined in (39) is maximized at the joint optimal set of investment decisions.

These results show that the appropriate use of capital allocation can be accomplished via a form of

19
internal capital markets where each division faces its own price of risk and multiplies its own divisional

risk (VaR) by this factor to determine its economic capital. The price internalizes the externality across

divisions. This procedure stands in strong contrast to the case where a division is given some allocation

of firm-wide VaR as economic capital and therefore is impacted by the decisions of other divisions.

It is quite possible that for some (but not all) division(s) within the institution, the IVaR will be

negative, as it is derived as a regression coefficient on overall firm cash flows. What does the optimal

capital allocation scheme prescribe in such as case? Here, rather than using up capital, the division is

essentially freeing up capital for use by other divisions. This means that if we preserve our definition

(39) of RAROC above, then divisional EVA and therefore divisional compensation should be redefined

as

EV A = (−RAROC)vi∗ .

This is analogous to what happens in translating between internal rate of return and net present value

in corporate finance. The usual case is that internal rates of return above costs of capital indicate

value-creating projects. But it is also possible to have situations where projects that contribute capital

create value when their internal rates of return are less than the cost of capital.

4 Asymmetric Information

Although the previous section derived an explicit functional form for the optimal capital allocation

process, the major deficiency is that the mechanism depends on information that would likely be private

to the trading division(s). In the case of equation (19) the adjustment to returns, rν is a function of

the expected return at the optimal risk level. The situation is even more sensitive in the multidivisional

case. Equations (35) and (36) show that the dependence on information is critical. The IVaR of

division i must be evaluated at the optimal portfolio decision of division j as well as that of the overall

institution, σp . Further the deviation from IVaR is also a function of the overall portfolio risk at the

optimum. Lastly, the adjustment function, νi depends on expected returns as in the single divisional

case. Therefore it is of considerable interest to explore the optimal mechanism in the presence of private

information of the divisions. The case studied in this section features a single trading division.

20
 As in the previous section, we assume that the trading division must expend some unobservable

effort to learn the information parameter, θ. This requires, as before that the division manager is

allocated a fraction, γ of the divisional EVA. As before, we assume that it is optimal for the institution

to encourage informational collection. Information is modeled as a single-dimensional draw, θ ∈ [θ, θ̄]

given distribution function F (θ).

As is well-known in the theory of private information and agency (Fudenberg and Tirole, 1992),

the optimal mechanism may be derived by using the revelation principle (Myerson, 1979). The direct

revelation mechanism is σ(θ̂) and T (θ̂), where θ̂ represents the division’s report of information or “type”.

The Bayesian Nash equilibria of the asymmetric information game are only those that can be supported

by the incentive compatibility condition that θ̂ = θ for all θ ∈ [θ, θ̄].

Formally under asymmetric information, the problem to be solved by the financial institution is as

follows:

max I = E[µ(σ(θ), θ) − rT C(θ) − U (θ)] (41)

σ(θ),T (θ),T C(θ)

subject to

θ ∈ arg max γ(µ(σ(θ̂), θ) − rT (θ̂)), (42)

θ̂

U (θ) = γ(µ(σ(θ), θ) − rT (θ)) ≥ U (43)

T C(θ) ≥ ασ(θ). (44)

The basic structure of this problem is analogous to the full information problem. The institution’s

objective is still to maximize total EVA of the institution less the compensation awarded to the trading

division. The institutions’s objective reflects the cost of the total capital raised as embodied in the reg-

ulatory restriction (44). Clearly this constraint will be met with equality. In designing the mechanism,

the central authority must satisfy the incentive compatibility condition on truth-telling by the trading

division (42) and must also provide a minimum level of utility in order to get the manager to participate

(equation (43)).

21
4.1 Optimal Delegated Capital

As before, there are two possibilities where capital may be flexible or inflexible. As the flexible allocation

scheme is more interesting, we focus entirely on that scenario here.

Consider the general problem in which the institution can select capital to respond to private infor-

mation. The optimal mechanism involves a trade-off between investment efficiency and informational

rent given to the trading division. This is a familiar situation given that this is a screening problem.

The capital allocation used in the first-best case (19) is a increasing function of the private information

of the division. As a result divisions with better information (higher θ) would intentionally misreport

their information as lower than it really is in order to have a higher EVA. This creates information rent

for the higher types that the institution would like to extract in the optimal second-best mechanism.

The major step in solving problem (41) is to first-convert the global incentive-compatibility condition

(42) into a local representation. Following Fudenberg and Tirole (1992, p. 264) it can be shown that a

necessary and sufficient condition for (42) to hold is that

Z θ
U (θ) = U + γµθ (σ(θ̂), θ̂)dθ̂, (45)
θ

and σ(θ) non-decreasing.9 Using this representation, the optimal risk of the institution solves the

following problem:
Z θ̄
max [µ(σ(θ), θ) − rασ(θ) − U (θ)]dF (θ). (46)
σ(θ) θ

subject to (45)

The solution to this problem is straightforward and is relegated to the appendix. Proposition 5 gives

the solution to the problem.

Proposition 5 The optimal risk taken by the division when capital is flexible is given by

γ(1 − F (θ))
µσ (σ(θ), θ) − rα = µθσ (σ(θ), θ). (47)
F 0 (θ)

Proposition 5 shows that when there is asymmetric information, optimal risk is reduced relative to the
9
The sorting condition (3) is the critical assumption that is necessary to obtain this representation of the incentive
compatibility conditions.

22
first-best level. This can be seen from (47) by noting that the right-hand side is positive. Therefore, the

marginal benefit of taking on increased risk at the second-best optimum is greater than the marginal

cost of increased VaR. The extent of this deviation is greater for lower types. This is intentional on the

part of the institution as it strives to make it more costly for the better trading divisions to misreport.

As an illustration of the optimal mechanism, we find that when µ(σ, θ) = θ log σ and F is uniformly

distributed,
(1 + γ)θ − γ θ̄
σ(θ) = .
rα

Since our interest is primarily in deriving the optimal capital allocation rules, we now consider a

transformation of the direct revelation mechanism. Consider the implementation of the optimal σ(θ)

via the following incentive schedule:

T̂ (θ̂, σ) = ν(θ̂) + λ(θ̂)σ. (48)

Suppose now that the division has “reported” θ̂, thereby determining the functional form of (48).

Consider the sub-problem where the institution now allows the division to select the risk level according

to:

max γ(µ(σ, θ) − r T̂ (θ̂, σ)). (49)

σ

Definition The indirect mechanism T̂ (θ̂, σ) implements the direct mechanism hσ(θ̂), T (θ̂)i whenever

the solution to (49), σ̂(θ̂) = σ(θ̂) and T̂ (θ̂, σ̂(θ̂)) = T (θ̂) for all θ̂.

Using this definition, we now see that a necessary condition for implementation of the optimal

second-best mechanism is that

µσ − r T̂σ = 0

yields the optimal decision according to (47). Substituting for the second-best optimality condition and

the definition of T̂ from (48) we arrive at

γ(1 − F (θ))
λ(θ) = α + µθσ (σ, θ). (50)
rF 0 (θ)

23
This result leads to the next proposition.

Proposition 6 Under asymmetric information, the optimal second-best mechanism may be imple-

mented via a modified VaR capital allocation schedule such that

 
γ(1 − F (θ))
T̂ (θ, σ) = ν(θ) + α + µθσ σ, (51)
rF 0 (θ)

where
Z θ
rν(θ) = − rσ(θ̂)dλ(θ̂) + µ(θ) − rλ(θ)σ(θ) − (U /γ). (52)
θ

The proof of proposition 6 is in the appendix.

Proposition 6 therefore shows that the division is made marginally more sensitive to risk than at the

first-best optimum according to the degree of information asymmetry and the share of divisional EVA

going to the manager. That is, the institution uses a more conservative capital allocation rule than that

under full information. Given the monotone hazard rate property, (1 − F (θ))/F 0 (θ), decreasing in θ,

it follows that generally λ is decreasing in the type of the trading division. This implies that trading

divisions with better investment opportunities face less of a deviation from the VaR factor as faced by

the institution itself.

The second part of the proposition illustrates the dependency on the fixed part of the capital

allocation function, ν, on the VaR factor λ. As λ(θ) is a decreasing function, the integral is negative.

Apart from the reservation utility, U , this shows that ν is positive as long as µ − rλσ is positive at the

lower endpoint. Further, the above relation shows that ν is increasing in θ; therefore there is an inverse

relationship between this fixed part of the capital allocation function and the VaR factor.

RAROC may be implemented under asymmetric information in a manner consistent with the full

information analyses. Define expected RAROC as follows:

µ(σ(θ), θ) − rν(θ) − rλ(θ)σ(θ)

RAROC(θ) = . (53)
λ(θ)σ(θ)

24
Now consider how RAROC behaves. At the lower endpoint, θ, substituting (52) into (53) shows that

U /γ
RAROC = .
λ(θ)σ(θ)

analogous to the full information case.

The performance measure is monotonically increasing in the information of the trading division so

that better divisions are correctly identified ex post. Of course, given that the capital allocation schedule

has already identified them ex ante, there is actually no reason in this single-period model to utilize

performance measurement.10

5 Multidivisional Capital Allocation under Asymmetric Information

This section features the most general set of conditions in the paper. We consider a multidivisional

financial institution where the divisions have private information about their expected cash flows from

trading activity. The problem is first formulated as a direct revelation game in which divisions each

report the value of their private information subject to a Bayesian Nash incentive compatibility condi-

tion.11

The direct revelation mechanism with two divisions is defined by the functions σi (θ1 , θ2 ); i = 1, 2,

denoting the risk level as a function of the information of the two divisions, and Ti (θ1 , θ2 ); i = 1, 2, the

capital allocation function. Information is represented by the joint distribution function, F (θ1 , θ2 ) ∈

[θ1 , θ̄1 ] × [θ 2 , θ̄2 ], exhibits independence with respect to (θ1 , θ2 ). Given the Bayesian Nash structure of

the problem, we use the following notation: a bar over a variable (e.g., µ̄i ) indicates that expectations

are taken by division i with respect to the private information of division j. With these features, the
10
For a multiperiod model of portfolio management in which performance measurement plays an important role see
Heinkel and Stoughton (1994).
11
Darrough and Stoughton (1989) discuss a joint venture mechanism design problem in which there is mutivariate private
information. They derive the functional form of optimal screening mechanisms. Mookherjee and Reichelstein (1992) discuss
the formulation of a multi-agent screening problem and the conditions under which implementation via dominant strategies
is possible. Unfortunately due to the joint dependence through the portfolio effects their “condensation” condition does
not hold. As a result we consider Nash implementation.

25
multidivisional problem for two divisions under joint asymmetric information is:

max I = E[µ̄1 (σ1 (θ1 , θ2 ), θ1 ) + µ̄2 (σ2 (θ1 , θ2 ), θ2 ) − rT C − Ū1 − Ū2 ] (54)
σi (θ1 ,θ2 ),Ti (θ1 ,θ2 ),T C

subject to

θi ∈ arg max γ(µ̄i (σi (θ̂i , θj ), θi ) − r T̄i (θ̂i , θj )), (55)

θ̂i

Ūi = γ(µ̄i (σi (θ1 , θ2 ), θi ) − r T̄i (θ1 , θ2 )) ≥ U i (56)

T C ≥ ασp (σ1 (θ1 , θ2 ), σ2 (θ1 , θ2 )). (57)

The objective function of the institution, (54) reflects the need to take expectations over the joint

distribution of divisional types, F (θ1 , θ2 ). The set of equations embodied in (55) indicates that each

division reports the “true” value of private information while taking the truthful report of the other

division as given. Similarly in (56) each division must take expectations over the other division’s

information parameter. The total capital regulatory constraint, (57) is as before.

Given the structure of the problem, it is fairly straightforward to apply analysis similar to that of

section 4, except with respect to the two divisions. The incentive compatible representation condition

for equations (55) is

Z θi
Ūi (θi ) = U + γ µ̄θ (σi (θ̂i , θj ), θ̂i )dθ̂i , (58)
θi

and the expectation (over θj ) of the designated risk level for division i, σ̄i (θi ), is non-decreasing in θi .

By analogy with equation (47), we therefore find that the optimal joint levels of risk obtained in the

multidivisional mechanism satisfies:

" #
vi 1 − Fi (θi )
µiσ (σi , θi ) − r − γµiθσ = 0, (59)
σi Fi0 (θi )

where vi (σi , σj ) is defined as before in (27), Fi denotes the marginal distribution of θi and σi (θi , θj )

depends jointly on the two divisions’ private information.

The optimal solution to the multidivisional problem (59) is similar to the single divisional case, and

exhibits less risk-taking than first-best. The degree of risk-taking is complicated by the interactive effect

26
of both divisions in each of the three terms of (59). However the qualitative behavior of σi as a function

of θi as θj is fixed is essentially the same as the single divisional analysis.

5.1 Implementation

In order to operationalize the above mechanism through an indirect mechanism where the risk level is

delegated to each divisional manager, we propose the following multivariate mechanism: delegate the

decision, σi to each division i so as to solve

max γ(µi (σi , θi ) − r T̂i (θ̂i , θj , σi )), (60)

σi

where

T̂i (θ̂i , θj , σi ) = ν̄i (θ̂i ) + λi (θ̂i , θj )σi . (61)

That is, each division is presented with a menu of capital allocation schedules with a fixed component,

ν̄ that does not depend on risk taken or the action of the other division. The risk factor, λ, however is

impacted by the choice of the other division. Generally, with better private information of division j,

the risk factor for division i will be greater. This induces a kind of internal capital market within the

financial institution.

Proposition 7 establishes the functional form of this optimal indirect mechanism. It’s proof is

essentially the same as that of proposition 6.

Proposition 7 The optimal multidivisional capital allocation mechanism can be implemented by a mod-

ified IVaR schedule such that

vi γµiθσ (1 − Fi (θi ))
λ(θi , θj ) = + , (62)
σi r Fi0 (θi )

and
Z θi
rν̄i (θi ) = − rσ̄i (θ̂i )dλ̄i (θ̂i ) + µ̄i (θ i ) − r λ̄i (θ i )σ̄(θ i ) − (U /γ), (63)
θi

where λ̄i denotes expectations of λi with respect to θj .

Proposition 7 is the key to the representation of RAROC in the multidivisional firm under asym-

27
metric information. We define
µi − rν̄i − rλi σi
RAROC = . (64)
λi σ i

This risk-adjusted return on capital exhibits the same properties with respect to each division as in the

single division case: it perfectly indicates the divisions with more favorable investment opportunities.

6 Conclusion

Discussions with practitioners support the view that corporate CFOs as well as their counterparts at

financial institutions view capital budgeting, risk management and capital structure as being inextrica-

bly linked. They seem less concerned with the determination and maintenance of an optimal debt to

equity ratio in response to capital budgeting decisions. Rather, the main issue is to optimally allocate

capital to various business units or investments and manage the resulting risks for a given financial

structure. Although firms focus considerable attention on this internal capital allocation process, very

little research has been done to provide management with robust normative rules. This paper attempts

to make a contribution in this direction.

This paper analyzed decentralized capital budgeting decisions of financial institutions in the presence

of capital adequacy restrictions. The results provide the capital allocation mechanism which induces

optimal investment behavior by various business units. It is shown that optimal equity capital allocation

is based on a business unit’s contribution to the institution’s total capital requirement, and can be

decomposed into two parts. The economic capital term is measured by the division’s incremental VaR.

In the case of a single-division firm, this is the same as the firm’s overall VaR. However, in the case of

multiple business units, the economic capital is equal to a price of risk multiplied by the division’s own

standard deviation. The risk adjustment term from the capital allocation function is a constant related

to the economic value added at the optimal investment level.

When there is asymmetric information about the business units’ investment opportunity set, then

the optimal capital allocation is changed. It is shown that asymmetric information makes the capital

allocation more sensitive to risk taking. Thus, in the presence of asymmetric information, the bank

allocates more capital for a given increase in volatility than in the presence of symmetric information.

28
 The analysis also demonstrates how banks should calculate the risk adjusted return on capital,

or RAROC to induce optimal investment decisions. It is found that a rigorous definition of RAROC

involves in the denominator an economic capital amount equal to the unit’s incremental VaR and in

the numerator the business unit’s expected or realized return minus an adjustment factor less a risk

adjustment comprised of economic capital . This adjustment reflects both the economic value added

at the optimal risk level and risk externalities imposed by one business unit on the incremental VaR

of other business units. The mechanism derived has the property that RAROC is larger for division’s

that make optimal investment decisions and for those with more favorable investment opportunities.

Furthermore, a business unit’s economic value added equals RAROC times the unit’s incremental VaR.

We believe that optimal capital allocation in the presence of capital adequacy regulation is a fruitful

area for future research. For instance it would be interesting to consider multiperiod settings in which

RAROC can be applied as a true performance measure. This extension would also more adequately ad-

dress the situation where capital cannot respond immediately to changes to the investment opportunity

set. Although tremendous advances have been made recently in the area of measuring and analyzing

risk of financial institutions, questions have arisen about application of these modern methodologies.

This paper represents a first step in developing a normative theory of the specialized nature of risk

management in financial institutions.

29
A Appendix

A.1 Proof of Proposition 5

The solution to this problem is standard within the screening literature (Guesnerie and Laffont, 1984).

With flexible capital, there is no need to exceed the capital regulatory constraint given by (44). Using

the representation of the incentive compatibility constraints (45), it is also clear that the reservation

utility constraint is only binding at the lower endpoint θ. The institution’s overall objective function

can be written as the total EVA minus the utility of the division, µ − rασ − U (θ).

Therefore the problem of (41) can be represented equivalently as:

Z θ̄
max I= [µ(σ(θ), θ) − rασ(θ) − U (θ)]dF (θ), (65)
σ(θ) θ

subject to (45). Using integration-by-parts, we find that

Z θ̄ Z θ̄
U (θ)dF (θ) = U (θ̄) − F (θ)dU (θ)
θ θ
Z θ̄
= U (θ̄) − γµθ (σ, θ)F (θ)dθ
θ
Z θ̄
= U (θ) + γµθ (σ, θ)(1 − F (θ))dθ,
θ

using (45). Substituting this expression into (65), gives the following maximization problem:

Z θ̄ Z θ̄
max I = [µ(σ, θ) − rασ]dF (θ) − γµθ (σ, θ)(1 − F (θ))dθ − U (θ)
σ(θ) θ θ
Z θ̄  
(1 − F (θ))
= µ(σ, θ) − rασ − γµθ (σ, θ) dF (θ) − U (θ).
θ F 0 (θ)

The determination of the optimal σ(θ) can now be accomplished in a pointwise manner. This gives

the following first order condition:

(1 − F (θ))
µσ (σ, θ) − rα − γµσθ (σ, θ)
F 0 (θ)
= 0. 2

30
A.2 Proof of Proposition 6

The first statement of the proposition (equation (51)) follows from the definition of T̂ and equation

(50).

To derive the optimal ν, substitute into the definition of the objective of the division to get

U (θ) = γ(µ(σ(θ), θ) − rν(θ) − rλ(θ)σ(θ)).

Substituting the representation of utility under incentive compatibility, (45),

Z θ
U+ γµθ (σ(θ̂), θ̂)dθ̂ = γµ(θ) − γrν(θ) − γrλσ(θ).
θ

Rearranging provides the following sequence of expressions:

Z µ Z θ
γrν(θ) = γ dµ0 − γrλσ(θ) − γµθ dθ̂ + γµ(θ) − U
µ(θ) θ
Z σ(θ)
= γµσ dσ 0 − γrλσ + γµ(θ) − U
σ(θ)
Zσ(θ)
= γrλ(σ −1 (σ 0 ))dσ 0 − γrλ(θ)σ(θ) + γµ(θ) − U
σ(θ)
Z θ
= −γrλ(θ)σ(θ) − γrσ(θ̂)dλ(θ̂) + γµ(θ) − U .
θ

Equation (52) is derived from the above equation. 2

31
References
Darrough, M. and N. Stoughton (1989), “A Bargaining Approach to Profit Sharing in Joint Ventures”,
Journal of Business, 62:237–270.

Froot, K. and J. Stein (1998), “Risk Management, Capital Budgeting and Capital Structure Policy for
Financial Institutions: An Integrated Approach”, Journal of Financial Economics, 47:55–82.

Fudenberg, D. and J. Tirole (1992), Game Theory, The MIT Press, Cambridge, Massachusetts.

Guesnerie, R. and J.-J. Laffont (1984), “A Complete Solution to a Class of Principal-agent Problems
with an Application to the Control of a Self-managed Firm”, Journal of Public Economics, 25:329–
369.

Harris, M., C. Kriebel and A. Raviv (1982), “Asymmetric Information, Incentives, and Intrafirm Re-
source Allocation”, Management Science, 28:604–620.

Harris, M. and A. Raviv (1996), “The Capital Budgeting Process, Incentives and Information”, Journal
of Finance, 51:1139–1174.

— (1998), “Capital Budgeting and Delegation”, Journal of Financial Economics, 50:259–289.

Heinkel, R. and N. Stoughton (1994), “The Dynamics of Portfolio Management Contracts”, Review of
Financial Studies, 7:351–387.

James, C. (1996), “RAROC Based Capital Budgeting and Performance Evaluation: A Case Study of
Bank Capital Allocation”, Working paper, Wharton Financial Institutions Center.

Jorion, P. (1997), Value at Risk: The New Benchmark for Controlling Market Risk, IRWIN, Chicago,
London, Singapore.

Kimball, R. (1997), “Innovations in Performance Measurement in Banking”, New England Economic

Review, May/Jun:23–38.

Matten, C. (1996), Managing Bank Capital: Capital Allocation and Performance Measurment, John
Wiley.

Merton, R. and A. Perold (1993), “Theory of Risk Capital in Financial Firms”, Journal of Applied
Corporate Finance, 6:16–32.

Milbourn, T. and A. Thakor (1996), “Intrafirm Capital Allocation and Managerial Compensation”,
Working paper, London Business School.

Mirrlees, J. (1971), “An Exploration in the Theory of Optimum Income Taxation”, Review of Economic
Studies, 38:175–208.

Mookherjee, D. and S. Reichelstein (1992), “Dominant Strategy Implementation of Bayesian Incentive

Compatible Allocation Rules”, Journal of Economic Theory, 56:378–399.

Myerson, R. (1979), “Incentive Compatibility and the Bargaining Problem”, Econometrica, 47:61–73.

Preinreich (1937), “Valuation and Amortization”, The Accounting Review, 12:209–226.

Reichelstein, S. (1997a), “Investment Decisions and Managerial Performance Evaluation”, Review of

Accounting Studies, 2:157–180.

32
— (1997b), “Providing Managerial Incentives: Cash Flows versus Accrual Accounting”, Working paper,
Haas School of Business, UC Berkeley.

Rogerson, W. (1997), “Intertemporal Cost Allocation and Managerial Investment Incentives: A The-
ory Explaining the Use of Economic Value Added as a Performance Measure”, Journal of Political
Economy, 105:770–795.

Spence, M. (1973), “Job Market Signalling”, Quarterly Journal of Economics, 87:355–379.

Stein, J. C. (1997), “Internal Capital Markets and the Competition for Corporate Resources”, Journal
of Finance, 52:111–133.

Stulz, R. (1998), Derivatives, Risk management, and Financial Engineering, Southwestern Publishing.

Uyemura, D., C. Kantor and J. Pettit (1996), “EVA for Banks: Value Creation, Risk Management and
Profitability Measurement”, Journal of Applied Corporate Finance, 9:94–113.

Zaik, E., J. Walter, G. Kelling and C. James (1996), “RAROC at Bank of America: From Theory to
Practice”, Journal of Applied Corporate Finance, 9:83–93.

33