Bund PDF

THE JOURNAL OF FINANCE VOL. LVIII, NO.
3 JUNE 2003
Internal versus External Financing: An Optimal

Contracting Approach n
ROMAN INDERST and HOLGER M. MLLER
Abstract
We study optimal ¢nancial contracting for centralized and decentralized ¢rms.
Under centralized contracting, headquarters raises funds on behalf of multiple
projects. Under decentralized contracting, each project raises funds separately
on the external capital market. The bene¢t of centralization is that headquar-
ters can use excess liquidity from high cash-£ow projects to buy continuation
rights for low cash-£ow projects. The cost is that headquarters may pool cash
£ows from several projects and self-¢nance follow-up investments without hav-
ing to return to the capital market. Absent any capital market discipline, it is
more di⁄cult to force headquarters to make repayments, which tightens ¢nan-
cing constraints ex ante. Cross-sectionally, our model implies that conglomer-
ates should have a lower average productivity than stand-alone ¢rms.
BEGINNING WITH FAZZARI, HUBBARD, AND PETERSEN (1988), several studies document
that the investment behavior of ¢rms is a¡ected by ¢nancing constraints.1 In this
paper, we explore the relation between ¢nancing constraints and organizational
structure. Speci¢cally, we ask whether centralized ¢rms where headquarters
raises funds on behalf of multiple projects face tighter or looser ¢nancing con-
straints than stand-alone ¢rms.
The internal capital markets literature provides some hints on the role of orga-
nizational structure for ¢nancing constraints. There, headquarters either adds
or destroys value, for example, by engaging in winner-picking (Stein (1997)),
n
Inderst is at the London School of Economics and Political Science and the CEPR and
Mˇller is at the Leonard N. Stern School of Business, New York University and the CEPR.
We are indebted to Patrick Bolton, Rick Green (the editor), and an anonymous referee for
helpful comments and suggestions. Thanks also to Yakov Amihud, Ulf Axelson, Mike Burkart,
Doug Diamond, James Dow, Zsuzsanna Fluck, Paolo Fulghieri, Martin Hellwig, Owen La-
mont, Christian Laux, Anthony Lynch, Alan Morrison, Lasse Pedersen, Per Str˛mberg, Elu
von Thadden, Je¡ Wurgler, Je¡ Zwiebel, and seminar participants at Berkeley, Chicago,
Frankfurt, INSEAD, Lausanne, LSE, Mannheim, NYU, Penn, Princeton, Saarbrˇcken, Stan-
ford, the Stockholm School of Economics, the European Summer Symposium in Economic
Theory (ESSET) in Gerzensee (2000), the European Summer Symposium in Financial Mar-
kets (ESSFM) in Gerzensee (2000), and the TMR Meeting on Financial Market E⁄ciency,
Corporate Finance and Regulation in Barcelona (2000) for comments and discussions. Earlier
versions of this paper circulated under the titles ‘‘Project Bundling, Liquidity Spillovers, and
Capital Market Discipline’’ and ‘‘Corporate Borrowing and Financing Constraints.’’All errors
are our own.
1
See Hubbard (1998) for an overview of the literature.
1033
1034 The Journal of Finance
redeploying assets across divisions (Gertner, Scharfstein, and Stein (1994)), or

weakening managerial incentives (Stein (2002)). Naturally, these activities will
a¡ect the return to capital and hence also the ¢rm’s ¢nancing constraint.
As none of these papers adopts an optimal contracting approach, the precise
nature and magnitude of the e¡ect remains unclear, however. On the other
hand, optimal contracting models, while deriving ¢nancing constraints and the
associated underinvestment problem from ¢rst principles, typically consider an
entrepreneurial ¢rm where the entrepreneur raises funds for a single project. In
this setting, questions of organizational structure and multiple projects cannot
be addressed.
This paper adopts an optimal contracting approach to examine the role of
headquarters for ¢nancing constraints, thus tying together internal and exter-
nal capital markets. We compare optimal contracting between (a) outside inves-
tors and individual project managers (‘‘decentralized borrowing’’), and (b)
outside investors and headquarters, who borrows on behalf of multiple projects
and subsequently allocates the funds to the various projects on the ¢rm’s internal
capital market (‘‘centralized borrowing’’).The term ‘‘borrowing’’ refers to the fact
that in our setting, like in related multiperiod settings by Bolton and Scharfstein
(1990), Hart and Moore (1998), and DeMarzo and Fishman (2000), the optimal
contract is a debt contract. Financing constraints arise endogenously in our mod-
el as we assume that part of the project cash £ow is nonveri¢able. The problem is
then to provide the ¢rm (i.e., the project manager or headquarters) with incen-
tives to pay out funds rather than to divert them.
The bene¢t of centralization is that ¢nancial contracts with centralized ¢rms
are more e⁄cient. To make the ¢rm reveal its true cash £ow, investors must o¡er
it a bribe. In a two-period model like ours, bribes can come in two forms. Either
the ¢rm makes a lower repayment in the ¢rst period, or it is o¡ered a higher con-
tinuation bene¢t. The continuation bene¢t is the expected rent captured by the
¢rm in the second period if the ¢nancing relationship is continued. Under centra-
lized borrowing, a greater fraction of the bribe comes in the form of continuation
bene¢ts, which is e⁄cient as it involves undertaking positive NPV second-period
investments that would have not been undertaken otherwise. E¡ectively, head-
quarters uses excess liquidity from high-cash-£ow projects to buy continuation
rights for low-cash-£ow projects.This allows headquarters to make greater repay-
ments, which eases ¢nancing constraints ex ante.
The cost of centralization is that headquarters can accumulate internal funds
by pooling cash £ows from di¡erent projects and self-¢nance second-period in-
vestments without returning to the capital market. Absent any capital market
discipline, however, it is more di⁄cult for investors to force the ¢rm to pay out
funds, which tightens ¢nancing constraints.This last point is reminiscent of Jen-
sen (1986), where the problem is also that ¢rms can undertake investments with-
out revisiting the capital market. Our model adopts an ex ante perspective:
Anticipating that a free cash-£ow problem might arise, investors are reluctant
to provide ¢nancing in the ¢rst place.
Based on these costs and bene¢ts of centralization, we trace out the bound-
aries of the ¢rm. Ceteris paribus, centralization is optimal for projects with a
Internal versus External Financing 1035
low expected return, or productivity, while decentralization is optimal for pro-

jects with a high expected return. Cross-sectionally, this implies that conglomer-
ates should have a lower average productivity than stand-alone ¢rms. Our model
provides testable implications linking ¢nancing constraints to operating produc-
tivity, the degree of ¢rm diversi¢cation, and the composition of a ¢rm’s invest-
ment portfolio.
An important question is to what extent the value created in an internal capi-
tal market will eventually be shared with investors. Our paper suggests that the
same ability that allows headquarters to create value, namely, the ability to pool
cash £ows, also protects it from fully relinquishing these gains to investors. As
the centralized ¢rm might not need the investors for follow-up ¢nancing, the in-
vestors’ability to extract some of the gains created in an internal capital market
is limited. Taking into account this con£ict, we provide conditions under which
conglomeration is optimal. We believe this is a central contribution of the paper,
and a possible answer to the question why the value created in an internal capital
market might not be fully passed on to investors.
Several papers have analyzed the costs and bene¢ts of internal capital
markets, notably Gertner et al. (1994), Stein (1997), Rajan, Servaes, and
Zingales (2000), and Scharfstein and Stein (2000). None of these papers
addresses the issue examined here, namely, contracting between headquarters
and outside investors, and the implications of this for the optimality of con-
glomerates. In terms of empirical predictions, a key feature of our model is
the relation between productivity or operating performance and the mode of
incorporation. None of the above papers studies this relation. Moreover, some
of these papers make predictions relating divisional investment to the divisions’
investment opportunities. In our model, opportunities are the same across divi-
sions, which is an assumption we make to abstract from winner-picking e¡ects.
Instead, our model makes predictions relating divisional investment to past divi-
sion cash £ows.
Berkovitch, Israel, and Tolkowsky (2000) and Matsusaka and Nanda (2002) also
explore the relation between internal capital markets and ¢rm boundaries. In the
paper by Berkovitch et al., headquarters has unlimited funds, which
implies ¢nancing constraints do not matter. Matsusaka and Nanda assume
that external ¢nance entails a deadweight loss that is equally great for con-
glomerates and stand-alone ¢rms. This assumption is precisely what we
question in our paper. Finally, we are not the ¢rst to note that cash-£ow pooling
can a¡ect agency problems between the ¢rm and investors. Papers in this genre
include, for example, Diamond (1984), Li and Li (1996), and Fluck and Lynch
(1999). Unlike these papers, cash-£ow pooling in our paper has both endogenous
costs and bene¢ts.
The rest of the paper is organized as follows. Section I derives the costs
and bene¢ts of centralization in an optimal contracting framework.
Section II discusses robustness issues. Section III presents various extensions
of the basic model. Section IV summarizes the empirical implications
and compares them to the evidence. Section V concludes. All proofs are in the
Appendix.
I. Centralized versus Decentralized Borrowing

A. The Model
The model is a multiperiod contracting model with partially nonveri¢able cash
£ows in the spirit of Bolton and Scharfstein (1990), DeMarzo and Fishman (2000),
Gertner et al. (1994), and Hart and Moore (1998). For the same reason as in these
models, the optimal contract in our model is a debt contract. Anticipating this
result, we use the term ‘‘borrowing’’ to denote the act of raising external ¢nance.
While the basic formulation here follows Bolton and Scharfstein, none of the re-
sults in this paper depend on the speci¢cs of their model. In Section II.C, we show
that the same trade-o¡ also obtains in a Hart^Moore type framework.
Suppose a project lasts for two periods. In each period, it requires an invest-
ment outlay I40 and yields an end-of-period cash £ow ploI with probability
p40 and ph4I with probability 1 p, where ph4pl. Cash £ows are uncorrelated
across periods. Instead of assuming that a project lasts for two periods, we could
equally imagine two separate, but technologically identical (sub)projects that
are carried out one after the other. The expected per-period cash £ow net of in-
vestment costs is strictly positive, that is, p :¼ ppl þ ð1 pÞph 4I:
Suppose a ¢rm has two such two-period projects. For the moment, we shall as-
sume that cash £ows are uncorrelated across projects. In Section III.A, we relax
this assumption. As the ¢rm has no funds, it must raise funds from outside inves-
tors. For convenience, we assume there is a single investor who makes a take-it-
or-leave-it o¡er to the ¢rm. While the assumption that there is a single investor
may seem unrealistic, it is inconsequential for our results. The only reason for
making this assumption is that it simpli¢es the contracting problem. See Section
II.E and footnote 3 for a discussion of this issue.
The ¢rm’s founder can choose between two organizational structures, which
di¡er in their assignment of projects to managers. Under centralized borrowing,
a single manager called headquarters is in charge of both projects. Under decen-
tralized borrowing, a separate project manager is in charge of each project. We use
the standard assumption that agents in charge of projects maximize the cash
proceeds from projects under their control, for instance, because they derive pri-
vate bene¢ts that are proportional to these proceeds. As projects require no mon-
itoring or managerial e¡ort, but only capital, the question is therefore whether
the founder should form one ¢rm where headquarters borrows on behalf of both
projects or two separate ¢rms that borrow independently on the external capital
market. While the problem is framed as an organizational design problem, it
could be equally framed as a divestiture problem where a conglomerate contem-
plates spinning o¡ one of its divisions. The model and results would be the same.
Neither cash £ows nor investment decisions are veri¢able, which implies con-
tracts can only condition on payments to and from the investor as well as public
messages. The assumption that cash £ows are nonveri¢able is standard and cap-
tures the notion that outsiders such as courts frequently have less information
than the parties to a contract. Since we adopt a message-game approach, it actu-
ally makes no di¡erence whether cash £ows and investment decisions are obser-
vable but nonveri¢able, or whether they can be observed by project managers and
headquarters but not by investors. We can therefore equally assume that project
managers and headquarters have private information about cash £ows and in-
vestment decisions. The assumption that investment decisions are nonveri¢able
simpli¢es the analysis, but is not needed. In Section II.C we show that the same
kind of trade-o¡ obtains in a setting where investment decisions are veri¢able.
Finally, even though courts cannot observe actual cash £ows, it is commonly
known that the lowest possible cash £ow is pl. Hence, we can equally assume
that a fraction pl of the cash £ow is veri¢able and only the di¡erence ph pl is
nonveri¢able.
Under both centralized and decentralized borrowing, the partial nonveri¢abil-
ity of cash £ows creates an incentive problem between the ¢rm and the investor.
Under centralized borrowing, the problem is to provide headquarters with incen-
tives to pay out funds. Under decentralized borrowing, the problem is to provide
individual project managers with incentives to pay out funds. Under centralized
borrowing, there are two subcases, depending on whether a high-cash-£ow ¢rm
can partly self-¢nance second-period investment or not.We shall label these sub-
cases ‘‘self-¢nancing’’and ‘‘no self-¢nancing,’’ respectively.
B. Decentralized Borrowing
The model of decentralized borrowing is adapted from Bolton and Scharfstein
(1990). Under decentralized borrowing, each of the two project managers borrows
separately on the external capital market. As the contracting problem is the same
for each manager, we will henceforth speak of the manager and the project. The
standard way to deal with nonveri¢ability of cash £ows is to adopt a message-
game approach. In the present context, this means that after the cash £ow is rea-
lized, the manager makes a publicly veri¢able announcement stating that the
cash £ow is either low or high. The sequence of events is as follows:
Date 0: The investor pays I and the manager (optimally) invests.

Date 1: The manager announces that the ¢rst-period cash £ow is ^s 2 fl; hg:
Based on this announcement, the manager makes a ¢rst repayment
R1 ð^sÞ; and the investor ¢nances second-period investment, that is, he
pays I a second time, with probability bð^sÞ: If the manager receives I,
he again (optimally) invests.
Date 2: Based on the date 1 announcement, the manager makes a second re-
payment R2 ð^sÞ:
Two comments are in order. Like most ¢nancial contracting models, we allow
for probabilistic (re)¢nancing schemes to permit nontrivial solutions. If the con-
tinuation probability can be either zero or one, the qualitative results are the
same but the bene¢ts from centralization are smaller. Second, while it is theore-
tically possible to have the manager also announce the second-period cash
£ow (in case he receives funding at date 1), this is pointless as he will always claim
that the second-period cash £ow is low. By contrast, it is possible to induce the
manager to truthfully reveal the ¢rst-period cash £ow by threatening him not to
provide second-period ¢nancing. An implicit assumption herein is that, if the

manager of a high-cash-£ow ¢rm claims that the cash £ow is low, he cannot use
the remaining cash to self-¢nance second-period investment. If he could, the in-
vestor’s threat to terminate funding would be empty and ¢nancing would break
down completely. Formally, the assumption is
ASSUMPTION 1: ph ploI.
Recall that the investor can always extract pl. An immediate implication of As-
sumption 1 is that pl40, or else the assumption that ph4I is violated.The investor
solves the following maximization problem:
max I þ p½R1 ðlÞ þ bðlÞðR2 ðlÞ IÞ

bðsÞ; R1 ðsÞ; R2 ðsÞ
þ ð1 pÞ½R1 ðhÞ þ bðhÞðR2 ðhÞ IÞ
s.t.
rðsÞ R1 ðsÞ þ bðsÞ½

p R2 ðsÞ
rðsÞ R1 ð^sÞ þ bð^sÞ½
p R2 ð^sÞ for all s; ^s 2 fl; hg; ð1Þ
R1 ðsÞ rðsÞ for all s 2 fl; hg;
and
R2 ðsÞ rðsÞ R1 ðsÞ þ pl for all s 2 fl; hg; ð2Þ
where r(l): 5 pl and r(h): 5 ph.

The ¢rst constraint is the manager’s incentive compatibility (or truth-telling)
constraint. The remaining two constraints are limited liability constraints. The
¢rst states that the ¢rst-period repayment must not exceed the ¢rst-period cash
£ow, while the second states that the total repayment must not exceed the sum
of ¢rst- and second-period cash £ows. Whenever (1) and (2) are satis¢ed, the
manager’s individual rationality constraint is also satis¢ed, which is why it can
be omitted.
From Bolton and Scharfstein (1990), we know that the solution to this problem
is b(l) 5 0, b(h) 5 1, R1(l) 5 R2(h) 5 pl, and R1 ðhÞ ¼ p: If the manager announces
that the ¢rst-period cash £ow is high, he receives second-period ¢nancing
for sure. If he announces that the cash £ow is low, he receives no second-period
¢nancing.
The optimal contract involves two types of ine⁄ciencies. First, with probabil-
ity p, the second-period investment is not undertaken. Despite this ine⁄ciency,
however, there will be no renegotiation on the equilibrium path as the
maximum which the investor can assure in the second period is ploI. Second,
if we insert the optimal contract in the investor’s objective function and
solve for the value of I at which he breaks even, we have that the investor
invests at date 0 if and only if

p pl
I p : ð3Þ
2p
Projects that cost less than p but more than the right-hand side in (3) receive no
funding at date 0 even though they have a strictly positive NPV. In other words,
the ¢rm is ¢nancially constrained.
C. Centralized Borrowing: No Self-Financing

Under centralized borrowing, headquarters borrows against the combined cash
£ow of the two projects. The relevant cash £ow is therefore r(l,l) :5 2pl with prob-
ability p2, r(l,h) :5 pl1ph with probability 2p(1 p), and r(h,h) :5 2ph with prob-
ability (1 p)2. The sequence of events is the same as under decentralized
borrowing.
As a contract now encompasses two projects, the potential contracting space
becomes richer. In particular, the investor may use separate re¢nancing prob-
abilities b1 ð^sÞ and b2 ð^sÞ for each of the two second-period investments, which im-
plies he will end up re¢nancing either zero, one, or two projects. It can be shown,
however, that any such contract is equivalent to a contract where the investor
uses a common re¢nancing probability bð^sÞ for both second-period investments.2
Without loss of generality, we can thus assume that the investor pays 2I with
probability bð^sÞ at date 1.
We ¢nally need to specify what the ¢rm’s self-¢nancing possibilities are if a
¢rm with two high cash £ows falsely claims that its cash £ows are low. Given As-
sumption 1, there are only two possibilities: (a) 2(ph pl)oI, in which case the
¢rm cannot self-¢nance at all, and (b) 2I42(ph pl)4I, in which case a ¢rm with
two high cash £ows can make one, but only one, second-period investment with-
out returning to the capital market. If a high cash-£ow ¢rm could self-¢nance
both second-period projects, Assumption 1 would be violated, that is, the inves-
tor’s threat would be empty and ¢nancing would break down.
In our model, centralization has costs and bene¢ts. We derive the basic trade-
o¡ in two steps.We ¢rst consider the case where self-¢nancing is not possible. In
this case, centralization has only bene¢ts.We then introduce the possibility that
a high cash-£ow ¢rm can self-¢nance one second-period investment with internal
cash. In this case, the bene¢ts derived earlier are still there, but there are also
costs.
Consider ¢rst the case where the ¢rm cannot self-¢nance second-period invest-
ment. Formally, we have the assumption that self-¢nancing is not possible.
ASSUMPTION 2: 2(ph pl)oI.
In what follows, we assume that Assumptions 1 and 2 hold.

2
A proof of this statement is contained in a previous working paper version of this paper
(Inderst and Mˇller (2000)). The proof is available from the authors upon request.
The problem under centralized borrowing is to provide headquarters with in-

centives to reveal the true cash £ow. Denote the set of possible cash £ows by S
:5 {(l,l), (l,h), (h,h)}. The investor maximizes
max 2I þ p2 ½R1 ðl; lÞ þ bðl; lÞðR2 ðl; lÞ 2IÞ

bðsÞ; R1 ðsÞ; R2 ðsÞ
þ 2pð1 pÞ½R1 ðh; lÞ þ bðh; lÞðR2 ðh; lÞ 2IÞ ð4Þ

2
þ ð1 pÞ ½R1 ðh; hÞ þ bðh; hÞðR2 ðh; hÞ 2IÞ
s.t.
rðsÞ R1 ðsÞ þ bðsÞ½2

p R2 ðsÞ
ð5Þ
rðsÞ R1 ð^sÞ þ bð^sÞ½2
p R2 ð^sÞ for all s; ^s 2 S;
R1 ðsÞ rðsÞ for all s 2 S: ð6Þ
and
R2 ðsÞ rðsÞ R1 ðsÞ þ 2pl for all s 2 S: ð7Þ
The individual rationality constraint can be again omitted as it is implied by

the stronger limited liability constraints (6) and (7).
The optimal contract is derived in the Appendix in the proof of Proposition 1.
In the low- and high-cash-£ow states, the optimal contract is the same as
under decentralized borrowing, except that all payments are multiplied by
two. We thus have b(l,l) 5 0, R1(l,l) 5 2pl, b(h,h) 5 1, R1 ðh; hÞ ¼ 2p; and
R2(h,h) 5 2pl. If both ¢rst-period cash £ows are low, the ¢rm obtains no
second-period ¢nancing, while if both ¢rst-period cash £ows are high, the ¢rm
obtains second-period ¢nancing for sure. In the intermediate case where one
cash £ow is low and the other is high, the optimal contract is either identical to
that of the high-cash-£ow ¢rm (if pZ1/2), or it has b(h,l) 5 1/[2(1 p)]41/2,
R1(h,l) 5 ph1pl, and R2(h,l) 5 2pl (if pr1/2).This case distinction is due to the fact
that if pZ1/2 the limited liability constraint (7) is slack. By contrast, if po1/2 the
constraint binds, which implies that the investor can extract the maximum pos-
sible date 1 repayment.
The only cash-£ow state where centralization makes a di¡erence is thus the
intermediate state where one cash £ow is low and the other is high. In this state,
the re¢nancing probability is strictly greater than one-half. By contrast, the aver-
age re¢nancing probability under decentralized borrowing is [b(h)1b(l)]/2 5 1/2.
We can therefore conclude that the ¢rst type of ine⁄ciency, namely, that e⁄cient
second-period investments are not undertaken, is less severe if borrowing is cen-
tralized.
If we insert the optimal contract in the investor’s objective function (4) and
solve for the value of I at which he breaks even, we obtain that the investor invests
at date 0 if and only if

p pl
I p ð8Þ
2 p þ p2
if pr1/2, and
p pl
I p ð9Þ
2 p2
if pZ1/2. Comparing (8) to (9) with the corresponding value under decentralized
borrowing, (3), we have that the second type of ine⁄ciency, namely, that positive
NPV projects are not ¢nanced at date 0, is also less severe under centralized bor-
rowing.This holds for any value of p.The following proposition summarizes these
results.
PROPOSITION 1: If Assumptions 1 and 2 hold, centralized borrowing is optimal for all p.

That is, it is optimal to have headquarters borrow on behalf of both projects and sub-
sequently allocate the funds on the ¢rm’s internal capital market rather than have each
project borrow separately on the external capital market.
The superiority of centralization vis-a'-vis decentralization is not based on a

superior allocation of funds to projects inside the ¢rm. At date 1, the two projects
are identical in every respect. Hence, there is no scope for winner picking.
Rather, the superiority of centralization derives from the fact that incentives for
revealing the true date 1 cash £ow can be provided more e⁄ciently.
The argument why centralized borrowing is optimal proceeds in two steps.
If both projects have a high cash £ow, the optimal repayment and re¢nancing
probability under centralized and decentralized borrowing are identical. Simi-
larly, if both projects have a low cash £ow, the optimal contracts under centra-
lized and decentralized borrowing are the same. In what follows, we can thus
focus on the intermediate cash-£ow state where one project has a high and the
other has a low cash £ow. To facilitate the exposition, we ¢rst derive some preli-
minary results. Incentive compatibility requires that a ¢rm with two high cash
£ows or one high and one low cash £ow has no incentive to mimic a ¢rm with two
low cash £ows. To make mimicking a low-cash-£ow ¢rm as costly as possible, the
investor sets b(l) 5 0 and R1(l) 5 pl (under decentralized borrowing) and b(l,l) 5 0
and R1(l,l) 5 2pl (under centralized borrowing). Moreover, it is evidently optimal
to set R2(h) 5 pl and R2(h,l) 5 R2(h,h) 5 2pl, which means the ¢rm must pay out its
entire veri¢able date 2 cash £ow.
Under decentralized borrowing, one ¢rm has a high and the other has a low
cash £ow. Consider the high-cash-£ow ¢rm’s incentive compatibility constraint:
ph R1 ðhÞ þ bðhÞ½ p pl ph pl : ð10Þ

|fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
first-period rent continuation benefit
The right-hand side depicts the payo¡ from mimicking the low-cash-£ow ¢rm.
Accordingly, to induce the high-cash-£ow ¢rm to reveal its cash £ow, the investor
must leave the ¢rm a rent of ph pl if it announces h. (This rent is called informa-
tion rent.) The investor can provide this rent in two ways: (a) he can demand a low-
er date 1 repayment R1(h), or (b) he can grant the ¢rm a higher continuation
bene¢t bðhÞ½p pl by increasing the continuation probability b(h). Consider the
cost to the investor under either option. Reducing the date 1 repayment is a zero-
sum transaction, that is, a dollar left in the ¢rm’s pocket is a dollar less in the
investor’s pocket. By contrast, granting the ¢rm a continuation bene¢t of
bðhÞ½p pl costs the investor only b(h) [I pl], which is less than bðhÞ½p pl :
The di¡erence bðhÞ½p I is the (expected) e⁄ciency gain from second-period in-
vestment. As the investor makes the contract o¡er, he can siphon o¡ this e⁄-
ciency gain. The solution is thus to provide as much rent as possible in the form
of continuation bene¢ts.3 As the maximum continuation bene¢t under decentra-
lized borrowing is p pl ; the remainder ph p must come in the form of ¢rst-
period rent, that is, in the form of a lower date 1 repayment.
Consider now the intermediate-cash-£ow ¢rm under centralized borrowing. As
we show in the Appendix in the proof of Proposition 1, it su⁄ces to consider the
incentive constraint ensuring that the intermediate-cash-£ow ¢rm has no incen-
tive to mimic the low-cash-£ow ¢rm:
ph þ pl R1 ðh; lÞ þ bðh; lÞ2½ p pl ph pl : ð11Þ

|fflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
The information rent that must be granted in the intermediate-cash-£ow state

is again ph pl. Unlike above, however, the investor can now provide a continua-
tion bene¢t of up to 2½p pl .The continuation bene¢t actually provided is either
2½p pl (if pZ1/2) or ph pl (if po1/2), which is both strictly greater than the
corresponding value p pl under decentralized borrowing.This is what constitu-
tes the fundamental advantage of centralized over decentralized borrowing. While
the total information rent is the same under centralized and decentralized bor-
rowing, its composition is di¡erent. Under decentralized borrowing, the conti-
nuation decision concerns a single project, which means that a relatively large
fraction of the rent must come in the form of ¢rst-period rent. By contrast, under
centralized borrowing the continuation decision concerns two projects, which
means that most (if pZ1/2), or even all (if po1/2) of the information rent can be
provided in the form of continuation bene¢ts.This is e⁄cient, as it involves under-
taking positive NPV investments that would have not been undertaken other-
wise. Our result that centralization improves contract e⁄ciency is robust in
3
If the ¢rm made the contract o¡er, it could siphon o¡ the e⁄ciency gain. Hence regardless
of who makes the contract o¡er, he or she will want to provide as much information rent as
possible in the form of continuation bene¢ts. This is important, as it underscores the fact that
the following argument (viz., that centralization is superior as it o¡ers more scope for provid-
ing rents in the form of continuation bene¢ts) applies irrespective of who makes the contract
o¡er. In particular, this means that the argument also holds under a competitive credit mar-
ket.
various ways. It holds if the state space is continuous (Section II.A), if invest-
ments are divisible (Sec. II.B), if credit markets are competitive (footnote 3 and
Sec. II.E), and if cash £ows are correlated (Sec. III.A).
Another way to view the trade-o¡ between ¢rst- and second-period rents is in
terms of ¢nancial slack. Any nonveri¢able date 1 cash £ow retained in the ¢rm
represents unused liquidity: E⁄ciency could be improved by trading it in for con-
tinuation rights. Consider the high-cash-£ow ¢rm under decentralized borrow-
ing. After paying out p, the high-cash-£ow ¢rm has remaining liquidity of
ph p: If the high-cash-£ow ¢rm were to share this liquidity with the low-cash-
£ow ¢rm, the latter could trade it in for continuation rights. But as each ¢rm
cares only about its own continuation decision, this does not happen.4
Under centralized borrowing, this externality problem does not arise. As head-
quarters adopts a ¢rm-wide perspective, it does not care which of the two projects
produces the cash £ow. E¡ectively, headquarters uses liquidity produced by the
high-cash-£ow project to buy continuation rights for the low-cash-£ow project.
(This also explains why the bene¢ts of centralization arise only in the intermedi-
ate-cash-£ow state.) Note that a ¢nancial intermediary, such as, for example, a
bank, cannot do this as it does not have direct access to the ¢rms’cash £ow. Much
like the investor under decentralized borrowing, a bank would have to provide
the high-cash-£ow ¢rm with incentives to disgorge cash.
Finally, consider again the inequality in Assumption 2 and replace 2 by n,
which implies that even n projects cannot provide su⁄cient cash to ¢nance a sin-
gle second-period investment. Replacing 2 by n, we obtain n(ph pl)oI. If n be-
comes su⁄ciently large, this inequality can no longer be satis¢ed. In other
words, the more projects there are under one roof, the more likely it is that the
¢rm can ¢nance at least one second-period investment without returning to the
capital market.Therefore, let us next consider the case where (partial) self-¢nan-
cing is possible.
D. Centralized Borrowing: Self-Financing

In the context of this model, self-¢nancing means that if both date 1 cash £ows
are high, headquarters can use its nonveri¢able cash £ow of 2(ph pl) and ¢nance
one second-period investment without having to return to the capital market.We
replace Assumption 2 by Assumption 3.
ASSUMPTION 3: 2I42(ph pl)ZI.
In what follows, we assume that Assumptions 1 and 3 hold.

The fact that a high-cash-£ow ¢rm can partly self-¢nance second-period invest-
ment tightens the ¢rm’s incentive compatibility constraint. In the absence of self-
4
What if the two ¢rms write an insurance contract at date 0? Due to the nonveri¢ability of
cash £ows, the high-cash-£ow ¢rm must earn an information rent of ph pl. If the insurance
contract were to oblige the high-cash-£ow ¢rm to share this rent with the low-cash-£ow ¢rm,
the former would not reveal its true cash £ow in the ¢rst place. Hence any incentive compa-
tible contract between the two ¢rms must lead to exactly the same allocation as here.
¢nancing, the payo¡ from mimicking a low-cash-£ow ¢rm is 2(ph pl). By con-
trast, if the ¢rm can self-¢nance second-period investment, the payo¡ from mi-
micking a low-cash-£ow ¢rm (and subsequently reinvesting the retained cash) is
2ðph pl Þ þ p I: To induce the high-cash-£ow ¢rm not to mimic a low-cash-£ow
¢rm the investor must now additionally compensate the ¢rm for the forgone prof-
it of p I, that is, he must pay the ¢rm a higher information rent. The general
idea is that, by pooling cash £ows from several projects, centralized ¢rms may
accumulate internal funds and make follow-up investments without having to re-
turn to the capital market. This weakens the investors’ threat to withhold future
¢nancing, which in turn tightens ¢nancing constraints at date 0.
If we solve the investor’s expected pro¢t for the value of I at which he breaks
even, we have that the investor invests at date 0 if and only if
pl
p
Ip
2 ; ð12Þ
1 þ p þ ð1pÞ
2
if p 1=2; and
p pl
I p 2 ð13Þ
1 þ 2pð1 pÞ þ ð1pÞ
2
if pZ1/2. Comparing (12) to (13) with the corresponding value under decentra-
lized borrowing, (3), we obtain the following result.
PROPOSITION 2: If Assumptions 1 and 3 hold, centralized borrowing

pffiffiffi where head-
quarters borrows
pffiffiffi on behalf of both projects is optimal if p 2 1: By contrast,
if p 2 1 it is optimal to have each project borrow separately on the external
capital market.
Self-¢nancing makes it more costly for the investor to induce the ¢rm to reveal its
true cash £ow, which is captured by the additional ‘‘bribe,’’ or information rent,
p I in the high-cash-£ow state. The cost of centralization thus depends on the
distribution of cash £ows in two ways. First, the probability of the high-cash-£ow
state is decreasing in p. Second, the ‘‘bribe’’ p I is also decreasing in p. Proposi-
tion 2 shows that if p is su⁄ciently small, that is, if the projects’ productivity is
su⁄ciently high, the costs of centralization outweigh the bene¢ts. To relax ¢nan-
cing constraints, the ¢rm should decentralize, or what is equivalent, disinte-
grate. As a single-project ¢rm does not generate enough cash to self-¢nance
second-period investment, it must necessarily revisit the capital market. Hence,
decentralization acts as a commitment vis-a'-vis investors to stay on a tight leash.5
The notion that ¢rms may bene¢t from committing to a policy of revisiting the
5
If both ¢rst-period cash £ows are high, the two ¢rms have a strong incentive to remerge at
date 1. To commit not to merge again, the ¢rms may write a covenant into their debt contract
restricting merger activity. Such covenants are common. For instance, Smith and Warner
(1979) ¢nd that 39.1 percent of the public debt issues in their sample include covenants re-
stricting merger activity.
capital market is not new and has been used as an explanation, for example, for
why ¢rms pay dividends (Easterbrook (1984)) or issue debt (Jensen (1986)). In
showing that a ¢rm’s organizational structure may act as a commitment to revisit
the capital market, our argument complements these arguments.
The investor cannot legally prevent the ¢rm from self-¢nancing as both cash
£ows and investment decisions are nonveri¢able. While the assumption that in-
vestment decisions are nonveri¢able may be realistic in some cases, in particular
if the ¢rm consists of a complex bundle of investments where it is di⁄cult for out-
siders to ascertain whether the i-th investment has been undertaken or not, it
may be less realistic in other cases. In Section II.C, we show that the assumption
that investment decisions are nonveri¢able is not needed if the parties can rene-
gotiate after default.
Proposition 2 admits an alternative interpretation, which goes beyond the is-
sue of ¢nancing constraints. It applies to situations where managers can with-
hold cash from both investors and the ¢rm’s owner(s). This may be because
managers are better informed or ownership is dispersed, implying that the own-
ers, while having formal control rights, have insu⁄cient incentives to enforce
their claims. Under this scenario, the ¢rm’s owners are in the same boat as the
investor: Unless management can be incentivized to pay out cash, neither the
investor nor the owners will see any of it. The optimal contract underlying
Proposition 2 is also optimal in this setting as it maximizes the cash £ow ex-
tracted by outsiders. The p-threshold characterizing the ¢rm boundaries also re-
mains optimal.
II. Discussion
A. Continuous Cash-Flow Distribution
The argument that one project does not generate enough cash to allow self-¢nan-
cing but two projects do is evidently independent of the cash-£ow distribution.
This is not so obvious with regard to the bene¢ts of centralization, that is, the
argument that ¢nancial contracts with centralized ¢rms are more e⁄cient.
Suppose cash £ows are continuously distributed with support [pl,ph]. Consider
¢rst the case where borrowing is decentralized. It can be shown that the optimal
contract is b(p) 5 1 and R1 ðpÞ ¼ p if p p; and bðpÞ ¼ ðp pl Þ=ðp pl Þ and
R1(p) 5 p if pop (e.g., Bolton and Scharfstein (1990) and DeMarzo and Fishman
(2000)). The optimal contract thus resembles a standard debt contract with face
value p and liquidation probability 1 b (p). Consider next centralized borrow-
ing. The ¢rm’s ‘‘type’’ is fully characterized by the sum o : 5 p11p2, where p1 and
p2 are the two ¢rst-period cash £ows. Again, it can be shown that the optimal
contract is b(o) 5 1 and R1 ðoÞ ¼ 2p if o 2p; and bðoÞ ¼ ðo 2pl Þ=2ðp pl Þ
and R1(o) 5 o if oo2p: Hence, the optimal contract is again a standard debt con-
tract, now with face value 2p and liquidation probability 1 b(o).
Consider the above optimal contracts. If either p1 p and p2 p or if p1 p
and p2 p; that is, if either both cash £ows are low or high, the re¢nancing
probability under centralized borrowing is identical to the average re¢nancing
probability under decentralized borrowing, [b(p1)1b(p2)]/2. In all other (i.e., in-

termediate) cash-£ow states, the re¢nancing probability is strictly greater under
centralized borrowing.The intuition for why centralized borrowing is superior is
the same as in the basic model: In intermediate-cash-£ow states, headquarters
can use excess liquidity from low-cash-£ow projects to buy continuation rights
for high-cash-£ow projects. In high- and low-cash-£ow states, such cross-subsidies
are either not necessary or not possible, respectively. By contrast, under decen-
tralized borrowing, cross-subsidies never occur. Accordingly, the ¢rst type of in-
e⁄ciency, namely, that e⁄cient second-period investments are not undertaken, is
strictly lower under centralized borrowing.
We can again solve for the value of I at which the investor breaks even. Again,
we ¢nd that the second type of ine⁄ciency, namely, that positive NPV projects
are not ¢nanced at date 0, is less severe under centralized borrowing if and only
if the expected re¢nancing probability under centralized borrowing is higher. By
the above argument, this is the case if and only if
Prðpi opjpj pÞ40 for i 6¼ j; ð14Þ
that is, if there is a nonzero probability that one cash £ow is below and the other
is above the mean. Proposition 1 thus extends to arbitrary continuous cash-£ow
distributions satisfying (14). If (14) does not hold, the organizational structure is
irrelevant. Condition (14) holds, for instance, if the joint distribution of p1,p2 has
full support. Conversely, if p1 and p2 are perfectly positively correlated, (14) does
not hold. Since both distributions have the same mean, the probability that one
cash £ow is below the mean and the other is above the mean is then zero.
B. Indivisibility of Investments
What is the role of investment indivisibilities in this model? In particular, sup-
pose at date 1 the ¢rm could invest a fraction ar1 of I in a project technology with
constant returns to scale.Would this a¡ect our results? We address this question
in three parts, pertaining to the role of indivisibilities for (a) the threat to with-
hold ¢nancing (‘‘termination threat’’), (b) the bene¢ts of centralization, and (c)
the costs of centralization.
B.1. TerminationThreat
Consider, for instance, the benchmark model of decentralized borrowing. There,
the incentives for a high-cash-£ow ¢rm to repay p instead of pl are that in return
the investor ¢nances second-period investment. Investment indivisibilities are
not essential in providing these incentives.What is essential is that the ¢rm can-
not capture all, or most, of the second-period e⁄ciency gains by self-¢nancing the
investment. Suppose the high-cash-£ow ¢rm can use its nonveri¢able cash £ow of
ph pl to ¢nance a fraction a : 5 (ph pl)/I of the second-period investment.With
constant returns to scale, the expected return on this investment is ap: By the
same logic as in Section I.D (‘‘self-¢nancing’’), the investor must then grant the
high-cash-£ow ¢rm an additional information rent of aðp I Þ to preserve incen-
tive compatibility. Solving the investor’s expected pro¢t for the value of I at which
he breaks even, we have that the investor invests at date 0 if and only if
p pl
I p : ð15Þ
ð1 pÞð1 aÞ þ 1
If a 5 0, (15) coincides with the corresponding threshold in the basic model, (3).
By inspection, the right-hand side in (15) is strictly decreasing in a, positive for
low a, and negative for high a. If a is high, the ¢rm can realize most of the second-
period e⁄ciency gains without borrowing additional funds from the investor. In
this case, the investor’s threat to withhold ¢nancing at date 1 is ine¡ective, and
¢nancing at date 0 breaks down. By contrast, if a is low, the ¢rm can only realize a
small fraction of the e⁄ciency gains by itself. In this case, the incentives to make
a high repayment and turn to the investor for additional funding are strong, and
date 0 ¢nancing becomes feasible. Moreover, the lower a, the greater is the right-
hand side in (15), and the looser is the ¢rm’s date 0 ¢nancing constraint. Hence,
the e¡ectiveness of the threat to withhold future ¢nancing does not depend on
investment indivisibilities. What is key, however, is that the ¢rm cannot exploit
all, or most, of the e⁄ciency gains through self-¢nancing.
B.2. Bene¢ts of Centralization

In Section I.C (‘‘no self-¢nancing’’), we elucidated the bene¢ts of cash-£ow pool-
ing by comparing the incentive compatibility constraint of the high cash-£ow
¢rm under decentralized borrowing, (10), with that of the intermediate cash-£ow
¢rm under centralized borrowing, (11). Consider again these two incentive con-
straints, now assuming that the ¢rm can use its nonveri¢able cash £ow to invest
in a project technology with constant returns to scale. The two incentive con-
straints become
ph R1 ðhÞ þ bðhÞ½ p pl ph pl þ að

p IÞ ð16Þ
|fflfflfflfflfflfflffl
ffl{zfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
and
ph þ pl R1 ðh; lÞ þ bðh; lÞ2½ p pl ph pl þ að

p IÞ; ð17Þ
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
respectively. The high-cash-£ow ¢rm under decentralized borrowing and the in-
termediate-cash-£ow ¢rm under centralized borrowing have the same nonveri¢-
able cash £ow, ph pl. Hence, both ¢rms can invest the same fraction a and, by the
argument in (a), earn the same additional information rent aðp I Þ.The bene¢ts
of cash-£ow pooling thus remain unchanged: Both ¢rms continue to earn the
same total information rent (right-hand side in (16) and (17)), but under centra-
lized borrowing, a greater fraction of this rent can be provided in the form of
continuation bene¢ts.
B.3. Costs of Centralization

In Section I.C (‘‘self-¢nancing’’), the costs of centralization were that in the high-
cash-£ow state, the centralized ¢rm can extract a higher information rent than
the two decentralized ¢rms together. These costs do not arise if the ¢rm can in-
vest a fraction a of I in a project technology with constant returns to scale at date
1. Note that in all three cash-£ow states the centralized ¢rm and the two stand-
alone ¢rms together have the same nonveri¢able cash £ow, namely, 2(ph pl)
(high-cash-£ow state), ph pl (intermediate-cash-£ow state), and zero (low-cash-
£ow state). By the argument in (B.2), the information rent collected by the centra-
lized ¢rm is therefore the same as that collected by the two decentralized ¢rms
together in each of the three cash-£ow states.
For centralization to be costly in our model, it is essential that in the high-cash-
£ow state the centralized ¢rm, after investing its nonveri¢able cash £ow of
2(ph pl), earns a higher second-period return than the two decentralized ¢rms
together after investing ph pl each.This is true if and only if the project technol-
ogy exhibits increasing returns to scale. The ¢xed-cost technology in Section I is
not the sole, but a practically very important example of this technology class.
C. Nonveri¢ability of Investment Decisions

In Section I.D, the nonveri¢ability of investment decisions allowed the ¢rm to
self-¢nance second-period investment without interference by the investor pro-
vided it had enough (nonveri¢able) cash £ow. The investor cannot simply add a
clause in the contract stating that the ¢rm cannot invest unless it makes a su⁄-
ciently high repayment. As investment decisions are nonveri¢able, the only way
for the investor to get the ¢rm to make a high repayment is to pay it a higher
bribe, that is, information rent. We argued earlier that the assumption that in-
vestment decisions are nonveri¢able may be realistic in some cases, but less rea-
listic in others. We shall now relax this assumption, instead assuming that the
parties can renegotiate after default.The basic insight remains the same: Centra-
lization is costly because the centralized ¢rm can realize a higher payo¡ in the
renegotiation than two decentralized ¢rms together. Consequently, the investor
must o¡er the centralized ¢rm a higher bribe to prevent default.
Our model of renegotiation is adapted from Hart and Moore (1998). In this set-
ting, if the ¢rm defaults, the investor seizes the asset underlying the project. To
bring their story in line with our model, we assume that the asset value corre-
sponds to the veri¢able cash-£ow component pl. The investor then has the choice
between selling the asset on the market or renegotiating ownership. If the inves-
tor sells the asset on the market, he receives pl. If he sells the asset back to the
¢rm, he receives P, which is the price obtained in the renegotiation. If the asset is
sold back to the ¢rm, it may be used for another period, where it generates a non-
veri¢able return of pl1D. At the end of the second period, the asset’s liquidation
value is zero. We shall assume that D40, that is, the asset is worth more to the
¢rm than to the market, which implies that date 1 liquidation is ine⁄cient. As
Hart and Moore point out, however, the ¢rm may not have enough cash to com-
pensate the investor for not liquidating the asset.
Consider the high-cash-£ow state, and suppose that 2(ph pl)4pl4(ph pl),
where the ¢rst inequality follows from Assumption 3. In this case, neither of the
two stand-alone ¢rms has su⁄cient cash to buy back the asset. However, the cen-
tralized ¢rmFafter earning 2ph but claiming that it has only 2plFhas the neces-
sary cash, which means there is scope for renegotiation. Depending on the
distribution of bargaining powers and the ¢rm’s liquidity, the outcome of the re-
negotiation is that the centralized ¢rm makes an additional net gain of
pl1D PZ0.6 In a renegotiation-proof contract, the investor must therefore pay
the centralized ¢rm a higher information rent than the two decentralized ¢rms
together. Even though investment decisions are veri¢able (the use of the asset in
the second period is observable) and the investor can prevent the ¢rm from con-
tinuing by liquidiating the asset, we have again that centralization lowers the
investor’s pro¢t in the high-cash-£ow state, which is all we need for Proposition
2 to hold.
D. Renegotiation
While the optimal contract under both centralized and decentralized borrowing
entails ine⁄ciencies, there will be no renegotiation on the equilibrium path, as
the maximum which the investor can assure in the second period is ploI. As the
discussion in the preceding subsection suggests, the situation is di¡erent if the
high-cash-£ow ¢rm falsely claims that its cash £ow is low (i.e., o¡ the equilibrium
path). Consider, for instance, the high-cash-£ow ¢rm under decentralized borrow-
ing. Upon claiming that its cash £ow is low, the ¢rm pays out pl, which implies its
remaining (i.e., nonveri¢able) cash £ow is ph pl.While this is not enough to self-
¢nance second-period investment, the ¢rm could renegotiate and ask the inves-
tor for additional funds of I (ph pl)opl. As the investor can assure a date 2
repayment of pl, he will provide these funds. A similar reasoning holds for the
high- and intermediate-cash-£ow ¢rms under centralized borrowing. In a renego-
tiation-proof contract, the investor would therefore have to pay all but the lowest
cash-£ow ¢rms an additional information rent. Besides, however, nothing
changes. In particular, as long as the investor has su⁄cient bargaining power
in the renegotiation, the costs and bene¢ts of centralization are the same as in
the commitment case.7
E. Competitive Credit Markets

Introducing competitive credit markets mitigates the underinvestment problem
but does not eliminate it. For instance, the re¢nancing probability of the low-
cash-£ow ¢rm under decentralized borrowing is then no longer zero but strictly
between zero and one. The fact that b(l) 5 0 facilitated many of our results. If
6
The inequality may be strict even if the investor has all the bargaining power in the rene-
gotiation. For instance, suppose 2(ph pl) 5 pl1D/2. In this case, the investor can extract, at
most, half of the surplus, since pl1D/2 is the most that headquarters can pay.
7
If the ¢rm has all the bargaining power, the investor’s payo¡ from each project is
I1plo0. Hence, ¢nancing breaks down completely, much like when there is only a single
period.
b(l)40, these results will be more complex, but their qualitative nature remains
the same. In particular, both the underinvestment problem and the basic trade-
o¡ analyzed here remain the same. One minor change is that the contract be-
tween the ¢rm and the investor must be augmented by a seniority provision (both
under centralized and decentralized borrowing). To see this, suppose the high-
cash-£ow ¢rm under decentralized borrowing defaults and approaches a new in-
vestor. As the ¢rm needs only I (ph pl)opl to ¢nance second-period invest-
ment, the new investor is willing to help out. But this means that the original
investor will make a loss. A seniority provision stating that the ¢rm cannot make
a repayment to a new investor unless it has fully settled its debt with the original
investor avoids this problem. Since payments to and from investors are veri¢able,
such a seniority provision is enforceable.
III.Which Projects Should Be Pooled?

In this section, we examine the decision to pool projects from various angles.The
empirical implications following from this are discussed in Section IV.
A. Correlation
We now relax our assumption that cash £ows in any given period are uncorre-
lated.We do, however, retain the assumption that cash £ows are serially uncorre-
lated. Denote the correlation coe⁄cient by r. While the optimal contract under
both self-¢nancing and ‘‘no self-¢nancing’’ remains unchanged, allowing for cor-
relation alters the probabilities of the three cash-£ow states and therefore the
investor’s expected payo¡. (Intuitively, the optimal contract remains unchanged,
as incentive compatibility and limited liability are both ex post constraints that
do not depend on the ex ante probabilities.) The new probabilities are
p[1 (1 r)(1 p)] for the low-cash-£ow state, 2(1 r)p(1 p) for the intermedi-
ate-cash-£ow state, and (1 p)[1 p(1 r)] for the high-cash-£ow state. A deriva-
tion of these probabilities can be found in the proof of Proposition 3 in the
Appendix.
If self-¢nancing is not possible, the result is evident. As centralization has ben-
e¢ts but no costs, centralized borrowing is optimal except if r 5 1. If r 5 1, the
probability of the intermediate-cash-£ow state is zero. But the intermediate-
cash-£ow state is the only cash-£ow state where centralized and decentralized
borrowing di¡er. If self-¢nancing is possible, the result is more complex.
PROPOSITION 3: Suppose Assumptions 1 and 3 hold. Decentralized borrowing is opti-

mal if rZ2/3 while centralized borrowing is optimal if rr 1/2. If rA( 1/2,2/3),
there exists a strictly increasing function pðrÞ such that decentralized borrowing is
optimal if p pðrÞ and centralized borrowing is optimal if p pðrÞ:
If r-1, the probability of the intermediate-cash-£ow state goes to zero while the
probability of the high-cash-£ow state remains positive. Hence centralization has
costs but no bene¢ts. Conversely, if r- 1 the probability of the high-cash-£ow
state goes to zero while the probability of the intermediate-cash-£ow state ap-
proaches one.8 Hence, centralization has bene¢ts but no costs. For intermediate
values of r, we have the same picture as before:While neither organizational form
dominates the other, there exists a critical value pðrÞ such that centralized bor-
rowing is optimal if p pðrÞ and decentralized
pffiffiffi borrowing is optimal if p pðrÞ.
When r 5 0, this critical value equals pð0Þ ¼ 2 1; which coincides with the re-
sult in Proposition 2.
B. High versus Low Future Pro¢tability

Is conglomeration more bene¢cial in industries where expected future pro¢ts are
low or high? To answer this question, we introduce separate probabilitiesFand
hence pro¢tabilitiesFfor each period. Denote the probability of the low cash £ow
in period t by pt and the corresponding expected cash £ow by pt.While we no long-
er assume that the two periods are the same, we retain the assumption that the
two projects are identical. Heterogeneous project bundles are considered below.
We have the following result.
PROPOSITION 4: Suppose Assumptions 1 and 3 hold. If p2r1/2 centralized borrowing is

optimal if and only if p1Z(1 p2)/(11p2). By contrast, if p2Z1/2, centralized borrow-
ing is optimal if and only if p1Z1/3.
If the ¢rst-period pro¢tability is su⁄ciently high, centralized borrowing is never

optimal. For all other values of p1, there exists a critical p2 -value such that decen-
tralized borrowing is optimal if p2 is below that value and centralized borrowing
is optimal if p2 is above that value. Intuitively, if the follow-up investment is un-
attractive, the incentives to engage in self-¢nancing are small, which implies that
centralized ¢rms can be disciplined at a relatively low cost. In this case, the ben-
e¢ts of centralization outweigh the costs. By contrast, if the follow-up investment
is attractive, the incentives to engage in self-¢nancingFand thus the informa-
tion rent that must be granted to the high-cash-£ow ¢rmFare high. In this case,
the costs of centralization outweigh the bene¢ts.
C. Cash-Flow Balancing
The termination threat is based on an intertemporal exchange: The ¢rm ex-
changes ¢rst-period cash £ow (thereby giving up ¢rst-period rents) for second-
period continuation rights. The termination threat is thus most e¡ective if there
is a balance between ¢rst-period cash £ow and continuation rights. If the conti-
nuation value is high but the ¢rst-period cash £ow is low the ¢rm can only buy a
small fraction of the continuation rights. Similarly, if the ¢rst-period cash £ow is
high but the continuation value is low the ¢rm is only willing to give up a fraction
8
Due to the two-point distribution, not all (r, p)-combinations are feasible. In particular, if
r 5 1, the only feasible p-value is p 5 1/2, which explains why the probability of the high-
cash-£ow state goes to zero as r- 1. The set of feasible (r, p)-combinations is derived in
the proof of Proposition 3 in the Appendix.
of its cash £ow equal to the continuation value. (Centralization mitigates this
problem by raising the continuation value.)
The above argument suggests that if projects are strongly front- (high p1 but
low p2 ) and backloaded (low p1 but high p2 ), it may be better to pool one front-
and one backloaded project instead of two front- or two backloaded projects.
The idea is that the high cash £ow generated by the frontloaded project can be
used to buy continuation rights for the (valuable) second tranche of the back-
loaded project. This intuition can be formalized. Suppose the probability of the
low cash £ow can take two values: pH and pL, where pH4pL. Frontloaded projects
have p1 5 pL and p2 5 pH, implying that p1 ¼ pL 4pH ¼ p2 : Backloaded projects
have p1 5 pH and p2 5 pL, implying that p1 ¼ pH opL ¼ p2 : The expected two-per-
iod cash £ow is the same for both projects.We obtain the following result.
PROPOSITION 5: Suppose Assumptions 1 and 3 hold. If pL pH is su⁄ciently large, im-

plying that projects are su⁄ciently front- and backloaded, it is optimal to pool one
front- and one backloaded project rather than two front- or two backloaded projects.
Proposition 5 has implications for investment policy. To maintain a balanced

portfolio, ¢rms might optimally want to forgo pro¢table investments in favor of
investments that are less pro¢table but have a more favorable cash-£ow pattern.
To give an extreme example of cash-£ow balancing, suppose there are two kinds
of projects: (a) Frontloaded projects generating an expected cash £ow of p in
the ¢rst period and zero in the second period, and (b) backloaded projects gener-
ating zero in the ¢rst period and an expected cash £ow of p in the second period.
For simplicity, suppose in a period where no cash £ow is generated, the invest-
ment cost is zero. Both projects are thus e¡ectively one-period projects. From
Bolton and Scharfstein (1990), we know that neither the front- nor the back-
loaded project aloneFnor a bundle consisting of two front- or two backloaded
projectsFcan raise external ¢nance. By contrast, a bundle consisting of one
front- and one backloaded project can raise external ¢nance if the investment
condition (3) holds.
IV. Empirical Implications

This section summarizes the empirical implications.The ¢rst implication follows
directly from the optimal contract. Consider a low-cash-£ow project (or division).
If the cash £ow of the other project is also low, the re¢nancing probability is zero.
By contrast, if the cash £ow of the other project is high, the re¢nancing probabil-
ity is between 0.5 and 1.The argument for the high-cash-£ow project is analogous.
IMPLICATION 1: Divisional investment is positively related to the past cash £ow of other
divisions.
Lamont (1997) and Shin and Stulz (1998) provide supporting evidence. Lamont
studies the reaction of U.S. oil companies to the 1986 oil price decline. He ¢nds
that a lower cash £ow in the companies’core business leads to investment cuts in
non-oil-related divisions. Similarly, Shin and Stulz ¢nd that the investment of
smaller divisions is positively related to the cash £ow of other divisions.
In our model, the fraction of nonveri¢able cash £ow ph pl measures the mag-
nitude of the agency problem between the ¢rm and the investor. If all cash £ow is
veri¢able, there is no value to project pooling. If some, but not too much cash £ow
can be diverted, project pooling is unambiguously valuable. Finally, if enough
cash £ow can be diverted to allow self-¢nancing of follow-up investments, the
value of project pooling is again low.
IMPLICATION 2: Internal capital markets are most valuable if agency problems between
¢rms and investors are small (but positive), and less valuable if they are large.
The result contrasts with Stein’s (1997) result that internal capital markets are
most valuable if agency problems are severe.The empirical evidence on this issue
is mixed. Consistent with Stein’s argument, Hubbard and Palia (1999) ¢nd that
the highest bidder returns in diversifying acquisitions in the 1960s were earned
when ¢nancially unconstrained buyers acquired constrained target ¢rms. Hub-
bard and Palia interpret this as evidence that capital markets viewed the forma-
tion of conglomerates in the 1960s as an e⁄cient response to the information
de¢ciencies of external capital markets, which were arguably greater at that
time. On the other hand, Servaes (1996) ¢nds that conglomerates in the 1960s
traded at a substantial discount, which is di⁄cult to reconcile with Hubbard
and Palia’s interpretation. Similarly, Claessens et al. (1999) document diversi¢ca-
tion patterns for corporations in the United States, Japan, and eight Asian coun-
tries, some of which do not have well-developed external capital markets. The
authors do not ¢nd a clear pattern of di¡erent degrees of diversi¢cation across
countries at di¡erent levels of development, concluding that ‘‘this contrasts with
the internal capital markets hypothesis ... which would suggest that ¢rms in less-
developed countries diversify more to reap the bene¢ts of internal markets’’
(p. 8).9
Direct support for Implication 2 is provided by Lins and Servaes (2001). Using
data from Asian emerging markets countries, Lins and Servaes ¢nd that the di-
versi¢cation discount is greatest when management control rights substantially
exceed their cash £ow rights. To the extent that the di¡erence between manage-
rial control rights and cash-£ow rights proxies for the severity of the agency pro-
blem between management and investors, this suggests an inverse relation
between the value of internal capital markets and the extent of the managerial
agency problem, as suggested in this paper.
The next two implications relate a ¢rm’s propensity to access external ¢nance
to exogenous characteristics such as operating productivity and the degree of
¢rm diversi¢cation.
9
Khanna and Palepu (2000) ¢nd that Indian ¢rms a⁄liated with highly diversifed business
groups outperform other ¢rms. The authors point out, however, that internal capital markets
have nothing to do with this. Unlike, for example, Japanese keiretsu, Indian business groups
have no common internal capital market.
IMPLICATION 3: Low-productivity conglomerates should have a higher, and high-pro-

ductivity conglomerates should have a lower propensity to access external ¢nance than
comparable stand-alone ¢rms.
IMPLICATION 4: The propensity of conglomerates to access external ¢nance should be

positively related to their degree of diversi¢cation.
Implication 3 follows from the analysis leading to Proposition 2. It follows from a

comparison of the investment threshold under decentralized and centralized bor-
rowing. Implication 4 follows from the analysis leading to Proposition 3. It does
not compare stand-alone ¢rms and conglomerates, but conglomerates with di¡er-
ent project correlations.
Implication 3 has, to our knowledge, not yet been tested. While Comment and
Jarrell (1995) and Peyer (2001) both ¢nd that conglomerates and stand-alones
have di¡erent propensities to access external ¢nance, neither paper compares
low- and high-productivity (or low- and high-performance) ¢rms separately.10
Implication 4 appears to be consistent with the empirical evidence. While Com-
ment and Jarrell ¢nd that highly and less diversi¢ed conglomerates have similar
propensities to access external capital markets, their analysis does not control
for internal capital market e⁄ciency. Peyer re¢nes Comment and Jarrell’s analy-
sis by discriminating between ¢rms with e⁄cient and ine⁄cient internal capital
markets. He ¢nds thatFprovided the internal capital allocation is e⁄cient
(which is the case in our model)Fthe propensity of conglomerates to access ex-
ternal ¢nance increases with their degree of diversi¢cation.
Implications 1 to 4 are general statements that hold regardless of whether the
organizational form is chosen optimally. The next two implications rest on the
assumption that the organizational form is chosen optimally. Implication 5 is a
direct corollary to Proposition 2, which is the central result of our paper.
IMPLICATION 5: Conglomerates should have a lower average productivity than stand-

alone ¢rms.
Implication 5 suggests that the diversi¢cation decision is endogenous: Low-pro-

ductivity ¢rms diversify while high-productivity ¢rms do not. Using plant-level
data, Maksimovic and Phillips (2002) ¢nd thatFfor all but the smallest ¢rms in
their sampleFconglomerate ¢rms in the United States are indeed less produc-
tive than single-segment ¢rms. (The smallest size category constitutes 3.3 per-
cent of their sample.) Similarly, Berger and Ofek (1995) (for the United States)
and Lins and Servaes (2001) (for emerging markets countries) ¢nd that diversi-
¢ed ¢rms have a smaller operating pro¢tability than stand-alone ¢rms, and Lang
10
Comment and Jarrell (1995) ¢nd that conglomerates use less external ¢nance than single-
segment ¢rms, although the di¡erence is small. Peyer (2001) ¢nds that conglomerates with
e⁄cient internal capital markets use more external ¢nance than single-segment ¢rms. Our
model would suggest that di¡erences in productivity might explain some of the cross-sec-
tional variation in these studies.
and Stulz (1994) ¢nd that diversifying ¢rms are poor performers relative to ¢rms
that do not diversify. Finally, Campa and Kedia (1999) and Graham, Lemmon, and
Wolf (2002) ¢nd that diversifying ¢rms already traded at a discount prior to the
diversi¢cation, and that targets in diversifying acquisitions had already been dis-
counted before they were acquired, respectively. Contrary evidence is provided
by Schoar (2002), who ¢nds that plants of diversi¢ed ¢rms are more productive
than plants of single-segment ¢rms.
The next statement follows from Proposition 4. It rests on the notion that the
incentives to engage in self-¢nancing, and hence the costs of centralization, are
lower if the follow-up investment is relatively unattractive.
IMPLICATION 6: Compared to stand-alone ¢rms, conglomerates should be more preva-

lent in slow-growing or declining industries.
Few studies have examined the relation between conglomeration and industry
growth. Consistent with our hypothesis, Lang and Stulz (1994) ¢nd that diversi-
¢ed ¢rms tend to be concentrated in industries with fewer growth opportunities.
Similarly, Burch, Nanda, and Narayanan (2000) report a negative correlation be-
tween industry conglomeration and investment opportunities as measured by in-
dustry market-to-book ratios.
The last implication follows from Proposition 5. It compares di¡erent invest-
ment policies for conglomerates.
IMPLICATION 7: Conglomerates operating both in growing and mature/declining indus-

tries should have a higher propensity to access external ¢nance than conglomerates
operating only in growing or only in mature/declining industries.
While based on a di¡erent logic than models of internal cash-£ow recycling, the
implications for investment policy are similar: Firms should hold a balanced
portfolio of projects generating immediate cash (‘‘cash cows’’) and projects gen-
erating cash in the future (‘‘growth projects’’).We are not aware of empirical work
examining the relation between ¢nancing constraints and the composition of a
¢rm’s investment portfolio.
V. Concluding Remarks
Financial contracting models typically consider an entrepreneur who raises
funds for a single project. In this setting, questions regarding organizational
structure or the role of internal capital markets cannot be addressed. On the
other hand, internal capital markets models, while analyzing the choice between
centralization and decentralization, do not consider optimal contracts between
headquarters and outside investors.This paper links both literatures, thereby ty-
ing together internal and external capital markets.
We derive the optimal contract for both centralized ¢rms where headquarters
borrows on behalf of multiple projects and decentralized, or stand-alone, ¢rms
where individual project managers borrow separately on the external capital
market. Centralization has bene¢ts and costs. On the bene¢t side, headquarters
uses excess liquidity from high-cash-£ow projects to buy continuation rights for
low-cash-£ow projects. This, in turn, allows headquarters to make greater repay-
ments, which relaxes ¢nancing constraints ex ante. On the cost side, headquar-
tersFby pooling cash £ows from several projectsFmight pursue follow-up
investments without returning to the capital market.This makes it more di⁄cult
for investors to discipline the ¢rm, which tightens ¢nancing constraints.
We believe our model yields insights which are applicable to other areas of
economics and ¢nance. By showing that cash-£ow pooling can strengthen
a ¢rm’s ability to expropriate investors, the paper is one of few papers emphasiz-
ing the potential costs of cash-£ow pooling.11 Other models, especially in the
¢nancial intermediation literature, rest largely on the bene¢ts of cash-£ow
pooling (e.g., Diamond (1984)). Introducing costs in these models might yield
additional insights. Second, internal capital marketsFvia their e¡ect on ¢nan-
cing constraintsFmight a¡ect the strategic behavior of ¢rms in the product mar-
ket. For instance, in Bolton and Scharfstein (1990), the presence of ¢nancing
constraints creates incentives for deep-pocket ¢rms to lower the pro¢ts of ¢nan-
cially constrained rivals. In this paper, we show that grouping several ¢nancially
constrained ¢rms together can reduce ¢nancing constraints, and therefore the
incentives of competitors to prey. Third, internal capital markets might play an
important role for the credit channel and monetary transmission mechanism. In
particular, to the extent that they alleviate credit constraints, internal capital
markets can damp the e¡ect of shocks on business lending and hence stabilize
production and economic growth.12
Appendix
Proof of Proposition 1: It remains to derive the optimal contract under centralized

borrowing given Assumptions 1 and 2. The rest follows from the argument in the
text.
Instead of solving the problem (4) to (7), we solve a relaxed problem where the
global incentive compatibility constraint (5) is replaced with the downward con-
straints that neither type (h,h) nor type (h,l) has an incentive to mimic type (l,l).
We subsequently show that the solution to this relaxed problem also solves the
original problem. In the relaxed problem, the investor solves (4) subject to the
limited liability constraints (6) and (7) and the downwards incentive compatibil-
ity constraints
rðsÞ R1 ðsÞ þ bðsÞ½2p R2 ðsÞ rðsÞ R1 ðl; l Þ þ bðl; l Þ½2p R2 ðl; l Þ ðA1Þ
11
Another paper pointing out that cash-£ow pooling may entail costs is Axelson (1999), who
analyzes the costs and bene¢ts of pooling assets in a common value auction. If the number of
bidders is large relative to the number of assets, pooling is costly, as it decreases bidding
competition in the upper tail of the signal distribution, and, hence, seller revenues.
12
For a discussion of the macroeconomic implications of credit constraints see Bernanke,
Gertler, and Gilchrist (2000).
where sA{(h,h), (h,l )}. Denote these constraints by C(h,h) and C(h,l ), respec-
tively. The following two lemmas considerably simplify the analysis.
LEMMA A1: At any optimum, it must hold that b(l,l ) 5 0 and R1(l,l ) 5 2pl.
Proof of Lemma A1: We argue to a contradiction. Suppose b(l,l )40 and de¢ne
1 ðl; l Þ :¼ 2pl and R
R 2 ðl; l Þ :¼ R2 ðl; l Þ 2pl þ R1 ðl; lÞ. If b(l,l )o1 replacing R1(l,l )
2 1
and R (l,l) with R ðl; l Þ and R 2 ðl; l Þ strictly increases the investor’s expected prof-
it, whereas if b(l,l) 5 1 replacing R1(l,l ) and R2(l,l ) with R 1 ðl; l Þ and R
2 ðl; l Þ leaves
the investor’s expected pro¢t unchanged. Moreover, if C(h,h), C(h,l ), and the two
limited liability constraints are satis¢ed under R1(l,l ) and R2(l,l ), they are also
satis¢ed under R 1 ðl; lÞ and R 2 ðl; lÞ:
From the second-period limited liability constraint for type (l,l ), it follows that
2 ðl; l Þ 2Io0. On the other hand, since p
R I40 and R 2 ðl; l Þ 2pl , it must be
true that 2 2
p R ðl; l Þ40. Accordingly, reducing b(l,l ) strictly improves the in-
vestor’s expected pro¢t without violating any of the incentive compatibility con-
straints, which contradicts the optimality of b(l,l )40. Given that b(l,l ) 5 0 is
optimal, the fact that R1(l,l) 5 0 is also optimal is obvious. Q.E.D.
LEMMA A2: At any optimum, the constraints C(h,l) and C(h,h) must bind.
Proof of Lemma A2: We argue again to a contradiction. Suppose C(h,h) is slack. If
b(h,h) 5 0 then C(h,h) implies that the ¢rst-period limited liability constraint for
type (h,h) is also slack. But this implies that the investor can improve his ex-
pected pro¢t by raising R1(h,h) without violating any constraint, contradiction.
If b(h,h)A(0,1), the unique optimal payments for type (h,h) are R1(h,h) 5 pl1ph
and R2(h,h) 5 2pl. Since we showed above that R1(l,l ) 5 2pl and b(l,l ) 5 0, this vio-
lates C(h,h), contradiction. Finally, if b(h,h) 5 1, any optimal contract must satis-
fy R1(h,h)1R2(h,h) 5 2ph12pl. Since 2ðph pl Þ42ð p pl Þ, this violates C(h,h),
contradiction.
Next, suppose C(h,l ) is slack. If b(h,l ) 5 0, the argument is the same as above. If
b(h,l )A(0,1), the unique optimal payments for type (h,l) are R1(h,l ) 5 ph1pl and
R2(h,l ) 5 2pl. Observe that if 2bðh; l Þð
p pl Þ ph pl , this contract is indeed in-
centive compatible. Since 2(pl I)o0, however, the investor is strictly better o¡ by
reducing b(h,l ), contradiction. Finally, if b(h,l ) 5 1, any optimal contract must
satisfy R1(h,l )1R2(h,l) 5 ph1pl12pl. In particular, this implies that any optimal
contract yields the same pro¢t to the investor as a contract where R1(h,l ) 5 ph1pl
and R2(h,l ) 5 2pl. As we showed above, however, the investor would then want to
decrease b(h,l ), contradiction. Q.E.D.
The ¢rst of the above lemmas implies that the lowest type (l,l ) receives no rent in
equilibrium. The second lemma is a standard feature of contracting problems of
this sort. Equipped with these two lemmas, we can now derive the optimal contract.
LEMMA A3: The following contract is optimal:
(1) Type (l,l ):b(l,l ) 5 0 and R1(l,l ) 5 2pl.
(2) Type (l,h): b(h,l ) 5 1/[2(1 p)], R1(h,l ) 5 ph1pl, and R2(h,l ) 5 2pl if pr1/2,
and b(h,l ) 5 1, R1 ðh; l Þ ¼ 2p; and R2(h,l) 5 2pl if pZ1/2.
(3) Type (h,h): b(h,h) 5 1, R1 ðh; hÞ ¼ 2p, and R2(h,h) 5 2pl.
Proof of Lemma A3: Setting b(l,l) 5 0 and R1(l,l) 5 2pl and inserting the binding
C(h,l) and C(h,h) constraints in (4) we can rewrite the objective function as
2ðpl IÞ þ 2pl þ 4pð1 pÞbðh; lÞðp IÞ þ 2ð1 pÞ2 bðh; hÞðp IÞ: ðA2Þ
By inspection, (A2) is strictly increasing in both b(h,l) and b(h,h), implying

p ph þ pl , setting b(h,l) 5
that the solution is b(h,l) 5 b(h,h) 5 1, if feasible. If 2
b(h,h) 5 1 is indeed feasible. The optimal payments R1(h,l), R2(h,l), R1(h,h), and
R2(h,h) then follow from C(h,l), C(h,h), and the respective limited liability con-
straints.
p4ph þ pl , setting b(h,l ) 5 1 violates either C(h,l) or the second-period lim-
If 2
ited liability constraint for type (h,l ). Accordingly, we must have b(h,l )o1. Next,
observe that 2p4R2 ðh; lÞ: To see this, suppose to the contrary that 2p R2 ðh; l Þ.
Subtracting the binding C(h,l) constraint from the second-period limited liabili-
ty constraint for type (h,l) gives
ph þ pl R2 ðh; lÞ þ bðh; lÞ½2p R2 ðh; lÞ: ðA3Þ
If 2p ¼ R2 ðh; lÞ this violates 2 p4ph þ pl ; contradiction. Suppose, there-

2
fore,
that 2poR ðh; lÞ: Solving (A3) for b(h,l) we have bðh; lÞ
ph þ pl R2 ðh; lÞ = 2p R2 ðh; lÞ , which is strictly greater than one since
2poR2 ðh; lÞ and 2 p4ph þ pl together imply that ph1ploR2(h,l), contradiction.
Solving the binding C(h,l ) constraint for b(h,l), we obtain bðh; lÞ ¼
½R1 ðh; lÞ 2pl = 2p R2 ðh; l Þ : Moreover, since we have 2p4R2 ðh; lÞ, it must
hold that @b(h,l )/@R1(h,2)4@b(h,l )/@R2(h,l )40, implying that both the ¢rst-
and second-period limited liability constraint for type (h,l) must bind. Solving
the binding limited liability constraints for R1(h,l) and R2(h,l), we have
R1(h,l ) 5 ph1pl and R2(h,l) 5 2pl. Inserting these values in bðh; lÞ ¼
½R1 ðh; l Þ 2pl = 2
p R2 ðh; l Þ yields
ph pl 1
bðh; lÞ ¼ ¼ ; ðA4Þ
2ðp pl Þ 2ð1 pÞ
where the second equality follows from the de¢nition of p:
It remains to show that the solution to the relaxed problem also solves the ori-
ginal problem (4) to (7). Since C(h,l) and C(h,h) are both binding, all other incen-
tive compatibility constraints must bind as well, which implies that the solution
is globally incentive compatible. Q.E.D.
Proof of Proposition 2: It remains to derive the optimal contract under centralized

borrowing given Assumptions 1 and 3. The rest follows from the argument in the
text.
LEMMA A4: The following contract is optimal:
(1) Type (l,l ):b(l,l ) 5 0 and R1(l,l ) 5 2pl.
(2) Type (h,l ): b(h,l ) 5 1/[2(1 p)], R1(h,l ) 5 ph1pl, and R2(h,l ) 5 2pl if pr1/2,
and b(h,l ) 5 1, R1 ðh; l Þ ¼ 2p; and R2(h,l ) 5 2pl if pZ1/2.
(3) Type (h,h): b(h,h) 5 1, R1(h,h) 5 2ph, and R2 ðh; hÞ ¼ p 2ðph pl Þ þ I:
Proof of Lemma A4: As in the proof of Proposition 1, we solve again a relaxed

problem. The corresponding incentive compatibility constraint for type (h,h),
which explicitly takes into account the possibility that type (h,h) can ¢nance
one or more second-period projects with internal funds by mimicking type (l,l),
is denoted by C ðh; hÞ: Type (h,h)’s payo¡ from deviating and mimicking type (l,l)
is then as follows:
U D ðh; hÞ :
8
>
> 2ph R1 ðl; lÞ þ bðl; lÞ½2
p R2 ðl; lÞ
>
>
>
> þ½1 bðl; lÞð
p IÞ if I 2ph R1 ðl; lÞo2I
<
¼
>
> 2ph R1 ðl; lÞ þ bðl; lÞ½2
p R2 ðl; lÞ
>
>
>
> þ½1 bðl; lÞ2ð
p IÞ if 2ph R1 ðl; lÞ 2I:
:
ðA5Þ
Since R1(l,l)r2pl, the case where 2ph R1(l,l)oI can be safely ignored as it vio-
lates Assumption 3. Moreover, Lemmas A1 to A2 in the proof of Proposition 1 con-
tinue to hold (with C(h,h) being replaced by C ðh; hÞ). Since b(l,l) 5 0 and
1 D
R (l,l) 5 2pl, Assumption 3 implies that U ðh; hÞ ¼ 2ðph pl Þ þ p I. Similar to
the proof of Lemma A1, the investor’s objective function can then be rewritten as
p IÞ þ ð1 pÞ2 ð2bðh; hÞ 1Þð
2ðpl IÞ þ 2pð1 pÞbðh; lÞ2ð p IÞ: ðA6Þ
Given that (A6) is strictly increasing in both b(h,l) and b(h,h), the arguments in
the proof of Proposition 1 extend to the current proof. In particular, the optimal
contracts for types (l,l) and (h,l) are the same as in Proposition 1. Furthermore,
we have that b(h,h) 5 1, which, together with C ðh; hÞ; implies that R1(h,h) 5 2ph
2
and R ðh; hÞ ¼ p þ I 2ðph pl Þ. To verify that the neglected incentive compat-
ibility constraints hold, note that it is impossible for type (h,l) to make a repay-
ment of R1(h,h) 5 2ph at date 1. Q.E.D.
Proof of Proposition 3:We ¢rst derive the joint probabilities for types (l,l), (h,l), and
(h,h) for arbitrary correlation coe⁄cients. Denote the random variables asso-
ciated with the two project cash £ows by X and Y, respectively.The joint probabil-
ities are o: 5 Pr(x 5 pl, y 5 ph) 5 Pr(x 5 ph, y 5 pl), Pr(x 5 y 5 pl) 5 p o, and
Pr(x 5 y 5 ph) 5 1 p o. Since r: 5 Cov(X,Y)/sXsY and sX 5 sY we have
r 5 1 o/p(1 p). Solving for o we obtain the probabilities given in the text.
Moreover, since ormin[p,1 p], it follows that the correlation coe⁄cient is
bounded from below by r :¼ 1 ðmin½p; 1 pÞ=½pð1 pÞ. (This function charac-
terizes the set of feasible (r,p)-combinations.)
While the optimal contract under centralized borrowing is the same as that
derived in the proof of Proposition 2, the investor’s expected pro¢t has changed
as the probabilities for types (l,l), (h,l), and (h,h) have changed. Inserting the
terms of the optimal contract in the investor’s objective function while taking
into account the new probabilities, we then have that the investor’s expected prof-
it equals 2ðpl I Þ þ ½1 p þ pð1 rÞð1 þ pÞðp I Þ if pr1/2 and 2ðpl I Þ þ
½1 þ 3pð1 rÞð1 pÞðp I Þ if pX1/2. Comparing these values with the investor’s
expected pro¢t under decentralized borrowing, 2ðpl I Þ þ ð1 pÞ2ðp I Þ, we
obtain the following result.
LEMMA A5: If Assumptions 1 and 3 hold and projects are arbitrarily correlated, the
comparison between centralized and decentralized borrowing is as follows.
(1) rA(2/3, 1]: Decentralized borrowing is optimal.
(2) rA(1/3, 2/3]: If pr1/[3(1 r)], decentralized borrowing is optimal, whereas if
pZ1/[3(1 r)], centralized borrowing
h is optimal.
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffii
(3) rA( 1/2,1/3]: If p pðrÞ :¼ r 2 þ 8 þ r2 8r =½2ð1 rÞ, decentra-
lized borrowing is optimal, whereas if p p, centralized borrowing is optimal.
(4) rA[ 1, 1/2]: Centralized borrowing is optimal.
It is easy to check that the functions 1/[3(1 r)] and pðrÞ are both strictly in-
creasing and intersect at r 5 1/3, which completes the proof. Q.E.D.
Proof of Proposition 4: The proof is analogous to that of Proposition 2.The optimal

contract under decentralized borrowing is the same as in Section I.B, except that
R1 ðhÞ ¼ p2 : The optimal contract under centralized borrowing given Assump-
tions 1 and 3 is the same as in Section I.D, except that R1 ðh; lÞ ¼ 2p2 if p2Z1/2,
b(h,l) 5 1/[2(1 p2)] if p2o1/2, and R2 ðh; hÞ ¼ p2 2ðph pl Þ þ I: Inserting the
optimal contract in the investor’s objective function, we have that under decen-
tralized borrowing the investor invests at date 0 if and only if
p2 pl
I p2 : ðA7Þ
2 p1
By contrast, under centralized borrowing given Assumptions 1 and 3, he in-
vests at date 0 if and only if
p2 pl
I p2 2 ðA8Þ
1 þ 2ð1 p1 Þp1 þ ð1p
2
1Þ
if p2Z1/2, and
p2 pl
I p2 2 ðA9Þ
ð1p1 Þp1
1þ ð1p2 Þ þ ð1p
2
1Þ
if p2r1/2. Comparing these expressions yields the result. Q.E.D.
Proof of Proposition 5: If pL pH is large, we have that pLo1/2opH. Consider ¢rst

the investment threshold if either two front- or two backloaded projects are
pooled. From the proof of Proposition 4, we have that
4ð1 pL ÞpL þ ð1 pL Þ2
I pl þ ðph pl Þð1 pH Þ ðA10Þ
2 þ 4ð1 pL ÞpL þ ð1 pL Þ2
if two front-loaded projects are pooled, and

2pH þ ð1 pH Þð1 pL Þ
I pl þ ðph pl Þð1 pH Þ2 ðA11Þ
2ð1 pL Þ þ 2ð1 pH ÞpH þ ð1 pH Þ2 ð1 pL Þ
if two backloaded projects projects are pooled. As pH pL-1, the spread pL
pH ¼ ðph pl ÞðpH pL Þ widens, and both (A10) and (A11) converge to pl.
Consider next the investment threshold if one front- and one backloaded pro-
ject are pooled.We ¢rst characterize the optimal contract, where we can build on
arguments in the proof of Proposition 2. The contract with type (l,l) is identical
to that in the proof of Proposition 2. Regarding type (h,l), we can treat the
state where the frontloaded project has a high cash £ow and the backloaded
project has a low cash £ow equivalently to the state where the backloaded
project has a high cash £ow and the frontloaded project has a low cash £ow.
Under the optimal contract, the investor pays I with probability one at date 1,
which ensures that the ¢rm can continue the pro¢table backloaded project. The
optimal repayment is pL at date 1 and pl at date 2. As for type (h,h), the investor
pays again I with probability one at date 1. Due to the additional self-¢nancing
constraint, however, the investor can extract at most 2pl at date 1 and zero at date
2. Substituting these speci¢cations into the investor’s pro¢t function yields the
investment threshold
pH ð1 pL Þ þ pL ð1 pH Þ
I pl þ ðph pl Þð1 pL Þ ; ðA12Þ
2 þ pH ð1 pL Þ þ pL ð1 pH Þ
which converges to pl1(ph pl)/34pl as pH pL-1. Q.E.D.
References
Axelson, Ulf, 1999, Pooling, splitting, and security design in the auctioning of ¢nancial assets, Mimeo,
University of Chicago.
Berger, Philip G., and Eli Ofek, 1995, Diversi¢cation’s e¡ect on ¢rm value, Journal of Financial Econom-
ics 37, 39 ^ 65.
Berkovitch, Elazar, Ronen Israel, and Efrat Tolkowsky, 2000, The boundaries of the ¢rm: The choice
between stand-alone and integrated ¢rms, Mimeo, Tel Aviv University.
Bernanke, Ben S., Mark Gertler, and Simon Gilchrist, 2000,The ¢nancial accelerator in a quantitative
business cycle framework, in John B. Taylor and Michael Woodford, eds.: Handbook of Macroeco-
nomics (North-Holland, Amsterdam).
Bolton, Patrick, and David S. Scharfstein, 1990, A theory of predation based on agency problems in
¢nancial contracting, American Economic Review 80, 93^106.
Burch, Timothy R.,Vikram Nanda, and M.P. Narayanan, 2000, Industry structure and the conglomer-
ate ‘‘discount’’: Theory and evidence, Mimeo, University of Miami.
Campa, Jose M., and Simi Kedia, 1999, Explaining the diversi¢cation discount, Mimeo, Harvard
University.
Claessens, Stijn, Simeon Djankov, Joseph P.H. Fan, and Larry H.P. Lang, 1999, The pattern and valua-
tion e¡ects of corporate diversi¢cation: A comparison of the United States, Japan, and other east
Asian economies, Mimeo, Hong Kong University of Science and Technology.
Comment, Robert, and Gregg A. Jarrell, 1995, Corporate focus and stock returns, Journal of Financial
Economics 37, 67^87.
DeMarzo, Peter M., and Michael J. Fishman, 2000, Optimal long-term ¢nancial contracting with
privately observed cash £ows, Mimeo, Stanford University.
Diamond, Douglas, 1984, Financial intermediation and delegated monitoring, Review of Economic
Studies 51, 393 ^ 414.
Easterbrook, Frank H., 1984, Two agency-cost explanations of dividends, American Economic Review
74, 650 ^ 659.
Fazzari, Steven M., R. Glenn Hubbard, and Bruce C. Petersen, 1988, Financing constraints and corpo-
rate investment, Brookings Papers on Economic Activity 1, 141^195.
Fluck, Zsuzsanna, and AnthonyW. Lynch, 1999,Why do ¢rms merge and then divest? A theory of ¢nan-
cial synergy, Journal of Business 72, 319^346.
Gertner, Robert H., David S. Scharfstein, and Jeremy C. Stein, 1994, Internal versus external capital
markets, Quarterly Journal of Economics 109, 1211^1230.
Graham, John R., Michael L. Lemmon, and Jack G.Wolf, 2002, Does corporate diversi¢cation destroy
value? Journal of Finance 57, 695^720.
Hart, Oliver D., and John H. Moore, 1998, Default and renegotiation: A dynamic model of debt,
Quarterly Journal of Economics 113, 1^ 41.
Hubbard, R. Glenn, 1998, Capital-market imperfections and investment, Journal of Economic Literature
36, 193^225.
Hubbard, R. Glenn, and Darius Palia, 1999, A reexamination of the conglomerate merger wave in the
1960s: An internal capital markets view, Journal of Finance 54, 1131^1152.
Inderst, Roman, and Holger M. Mˇller, 2000, Project bundling, liquidity spillovers, and capital market
discipline, Mimeo, University of Mannheim.
Jensen, Michael C., 1986, Agency costs of free cash £ow, corporate ¢nance, and takeovers, American
Economic Review Papers and Proceedings 76, 323 ^329.
Khanna, Tarun, and Krishna Palepu, 2000, Is group a⁄liation pro¢table in emerging markets? An
analysis of diversi¢ed Indian business groups, Journal of Finance 55, 867^891.
Lamont, Owen A., 1997, Cash £ow and investment: Evidence from internal capital markets, Journal of
Finance 52, 83^109.
Lang, Larry H.P., and Rene M. Stulz, 1994, Tobin’s q, corporate diversi¢cation, and ¢rm performance,
Journal of Political Economy 102, 1248^1280.
Li, David D., and Shan Li, 1996, A theory of corporate scope and ¢nancial structure, Journal of Finance
51, 691^709.
Lins, Karl, and Henri Servaes, 2001, Is corporate diversi¢cation bene¢cial in emerging markets? Mi-
meo, University of Utah.
Maksimovic, Vojislav, and Gordon Phillips, 2002, Do conglomerate ¢rms allocate resources ine⁄-
ciently across industries? Theory and evidence, Journal of Finance 57, 721^767.
Matsusaka, John G., and Vikram Nanda, 2002, Internal capital markets and corporate refocusing,
Journal of Financial Intermediation 11, 176 ^211.
Peyer, Urs C., 2001, Internal and external capital markets, Mimeo, INSEAD.
Rajan, Raghuram, Henri Servaes, and Luigi Zingales, 2000, The cost of diversity: The diversi¢cation
discount and ine⁄cient investment, Journal of Finance 55, 35^80.
Scharfstein, David S., and Jeremy C. Stein, 2000,The dark side of internal capital markets: Divisional
rent-seeking and ine⁄cient investment, Journal of Finance 55, 2537^2564.
Schoar, Antoinette S., 2002, E¡ects of corporate diversi¢cation on productivity, Journal of Finance 57,
2379^2403.
Servaes, Henri, 1996, The value of diversi¢cation during the conglomerate merger wave, Journal of
Finance 51, 1201^1225.
Shin, Hyun-Han, and Rene M. Stulz, 1998, Are internal capital markets e⁄cient? Quarterly Journal of
Economics 113, 531^552.
Smith, Cli¡ord W., and Jerold B. Warner, 1979, On ¢nancial contracting: An analysis of bond cove-
nants, Journal of Financial Economics 7, 117^161.
Stein, Jeremy C., 1997, Internal capital markets and the competition for corporate resources, Journal
of Finance 52, 111^133.
Stein, Jeremy C., 2002, Information production and capital allocation: Decentralized vs. hierarchical
¢rms, Journal of Finance 57, 1891^1921.

Bund PDF

Uploaded by

Copyright:

Available Formats

Bund PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Bund PDF

Uploaded by

Copyright:

Available Formats

THE JOURNAL OF FINANCE VOL. LVIII, NO.

Internal versus External Financing: An Optimal

ROMAN INDERST and HOLGER M. MLLER

redeploying assets across divisions (Gertner, Scharfstein, and Stein (1994)), or

low expected return, or productivity, while decentralization is optimal for pro-

I. Centralized versus Decentralized Borrowing

Date 0: The investor pays I and the manager (optimally) invests.

provide second-period ¢nancing. An implicit assumption herein is that, if the

max I þ p½R1 ðlÞ þ bðlÞðR2 ðlÞ IÞ

þ ð1 pÞ½R1 ðhÞ þ bðhÞðR2 ðhÞ IÞ

rðsÞ R1 ðsÞ þ bðsÞ½

R2 ðsÞ  rðsÞ R1 ðsÞ þ pl for all s 2 fl; hg; ð2Þ

where r(l): 5 pl and r(h): 5 ph.

invests at date 0 if and only if

C. Centralized Borrowing: No Self-Financing

ASSUMPTION 2: 2(ph pl)oI.

In what follows, we assume that Assumptions 1 and 2 hold.

The problem under centralized borrowing is to provide headquarters with in-

max 2I þ p2 ½R1 ðl; lÞ þ bðl; lÞðR2 ðl; lÞ 2IÞ

þ 2pð1 pÞ½R1 ðh; lÞ þ bðh; lÞðR2 ðh; lÞ 2IÞ ð4Þ

rðsÞ R1 ðsÞ þ bðsÞ½2

R1 ðsÞ  rðsÞ for all s 2 S: ð6Þ

R2 ðsÞ  rðsÞ R1 ðsÞ þ 2pl for all s 2 S: ð7Þ

The individual rationality constraint can be again omitted as it is implied by

at date 0 if and only if

PROPOSITION 1: If Assumptions 1 and 2 hold, centralized borrowing is optimal for all p.

The superiority of centralization vis-a'-vis decentralization is not based on a

ph R1 ðhÞ þ bðhÞ½ p pl ph pl : ð10Þ

ph þ pl R1 ðh; lÞ þ bðh; lÞ2½ p pl ph pl : ð11Þ

The information rent that must be granted in the intermediate-cash-£ow state

D. Centralized Borrowing: Self-Financing

ASSUMPTION 3: 2I42(ph pl)ZI.

In what follows, we assume that Assumptions 1 and 3 hold.

PROPOSITION 2: If Assumptions 1 and 3 hold, centralized borrowing

probability under decentralized borrowing, [b(p1)1b(p2)]/2. In all other (i.e., in-

B.2. Bene¢ts of Centralization

ph R1 ðhÞ þ bðhÞ½ p pl ph pl þ að

ph þ pl R1 ðh; lÞ þ bðh; lÞ2½ p pl ph pl þ að

B.3. Costs of Centralization

C. Nonveri¢ability of Investment Decisions

E. Competitive Credit Markets

III.Which Projects Should Be Pooled?

PROPOSITION 3: Suppose Assumptions 1 and 3 hold. Decentralized borrowing is opti-

B. High versus Low Future Pro¢tability

PROPOSITION 4: Suppose Assumptions 1 and 3 hold. If p2r1/2 centralized borrowing is

If the ¢rst-period pro¢tability is su⁄ciently high, centralized borrowing is never

PROPOSITION 5: Suppose Assumptions 1 and 3 hold. If pL pH is su⁄ciently large, im-

Proposition 5 has implications for investment policy. To maintain a balanced

IV. Empirical Implications

IMPLICATION 3: Low-productivity conglomerates should have a higher, and high-pro-

IMPLICATION 4: The propensity of conglomerates to access external ¢nance should be

Implication 3 follows from the analysis leading to Proposition 2. It follows from a

IMPLICATION 5: Conglomerates should have a lower average productivity than stand-

Implication 5 suggests that the diversi¢cation decision is endogenous: Low-pro-

IMPLICATION 6: Compared to stand-alone ¢rms, conglomerates should be more preva-

IMPLICATION 7: Conglomerates operating both in growing and mature/declining indus-

Proof of Proposition 1: It remains to derive the optimal contract under centralized

By inspection, (A2) is strictly increasing in both b(h,l) and b(h,h), implying

If 2p ¼ R2 ðh; lÞ this violates 2 p4ph þ pl ; contradiction. Suppose, there-

R2 ðsÞ rðsÞ R1 ðsÞ þ pl for all s 2 fl; hg; ð2Þ

R1 ðsÞ rðsÞ for all s 2 S: ð6Þ

R2 ðsÞ rðsÞ R1 ðsÞ þ 2pl for all s 2 S: ð7Þ