Available online at www.sciencedirect.

com

International Journal of Industrial Organization 26 (2008) 393 411

www.elsevier.com/locate/econbase

To build or not to build: Normative and positive theories

of publicprivate partnerships
David Martimort a,, Jerome Pouyet b,1
a

University of Toulouse (IDEI, GREMAQ) and Institut Universitaire de France. Address: IDEI,
Manufacture des Tabacs, Bt. F, 21 Alle de Brienne, 31000 Toulouse, France
b
Ecole Polytechnique and University of Toulouse (IDEI). Address: Department of Economics,
Ecole Polytechnique, 91128 Palaiseau Cedex, France
Available online 5 December 2006

Abstract
This paper analyzes whether the two tasks of building infrastructures which are socially useful in providing public services and
managing these assets should be bundled or not. When performance contracts can be written, both tasks should be performed
altogether by the same firm if a better design of the infrastructure helps also to save on operating costs. Otherwise, tasks should be
kept apart and undertaken by different units. In incomplete contracting environments we isolate conditions under which either the
traditional form of public provision of services or the more fashionable publicprivate partnership emerges optimally. The latter
dominates when there is a positive externality and the private benefits from owning assets are small enough. Finally, we take a
political economy perspective and study how incentive schemes are modified under the threat of capture of the decision-makers.
Much of the gains from bundling may be lost in this case.
2006 Elsevier B.V. All rights reserved.
JEL classification: H11
Keywords: Publicprivate partnership; Bundling/unbundling; Agency costs; Capture

1. Introduction
One of the most intriguing issues in modern industrial organization consists of delineating the optimal

We are grateful to several members of Veolia Institute for helpful

discussions which have motivated this research. We also thank
Elisabetta Iossa, Marc Ivaldi, Markus Ksoll, Jan-Eric Nilsson, Patrick
Rey, Eric Strobl, Jean Tirole, and, especially, Bernard Caillaud and two
referees for helpful comments on a previous version.
Corresponding author. Tel.: +33 0 561128614; fax: +33 0 561128637.
E-mail addresses: [email protected] (D. Martimort),
[email protected] (J. Pouyet).
1
Tel.: +33 0 169332646; fax: +33 0 169333858.

0167-7187/$ - see front matter 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.ijindorg.2006.10.004

division of labor between the public and the private

spheres. In this respect, the recent privatization wave
which took place over the eighties and nineties in most
industrial countries and which was also advocated by
international agencies for developing countries certainly
testifies that this question is at the heart of most major
reforms. Even though defenders of full privatization
schemes can still be found nowadays in the most liberal
spheres, an unequivocal commitment to privatization is
often viewed as an excessive response to the inefficiency of the public sector (if any) even when
privatization is accompanied by a convenient regulatory
environment. Most scholars and public decision-makers
advocate thus for a more pragmatic approach which

394

D. Martimort, J. Pouyet / Int. J. Ind. Organ. 26 (2008) 393411

consists of promoting efficient (or at least as efficient as

possible) partnerships between the public and the
private sectors for the provision of major services and
public goods.2 Only tasks where the private sector has a
comparative advantage should thus really be delegated
to the private sphere.
To understand the optimal pattern of delegation it is
useful to keep in mind that most public services (like
water management, waste disposal services, sanitation,
public transportation, prison management) involve a
complex array of tasks. Those activities necessitate
indeed, first, to build infrastructures and, second, to
operate these assets as efficiently as possible. Delegation to the private sector thus takes place de facto in a
multi-task environment.3
The traditional form of public procurement used in
most industrial countries has so far relied on some kind of
unbundling of these tasks. First, a government designs the
characteristics and quality attributes of the project.
Second, the government chooses a private builder to
build assets but retains ownership. Finally, the government chooses an operator, who may be either public or
private, to manage these assets and provide the service.
More recently, several initiatives around the world4 and
various legal reforms5 have proposed an alternative form
of procurement, the so-called PublicPrivate Partnerships (PPPs henceforth). With this procurement mode the
government takes a more minimalist stance: it only
chooses a private consortium which is in charge of both
designing the quality attributes of the infrastructure,
building these assets and, finally, managing them as
efficiently as possible. Compared with the more traditional form of procurement, the PPP alternative is thus
characterized by two important features. First, the two
tasks of building and managing assets are now bundled.
Second, the ownership pattern is also quite different.
Taking first a normative point of view, the first
objective of this paper is to understand why and under
which circumstances those two alternative forms of
procurement are optimal. Of course, this issue is really
2

See the 1998 United Nations Development Programme.

See Holmstrm and Milgrom (1991) (moral hazard) and Laffont
and Tirole (1993, Chapter 3) (adverse selection) for general analysis
of the multi-task problem.
4
Berger (1985) traces the references to partnerships between the
public and the private sectors in the U.S. to the Carter administration
and its willingness to include private actors in the development of
urban projects in areas of very costly public funds and huge public
deficits. Daniels and Trebilcock (2002) offer a nice overview of some
issues raised by publicprivate partnerships in Canada.
5
See the June 2004 text prepared by the Raffarin government in
France, for instance.
3

only relevant in a framework where delegation of tasks to

the private sector also comes with some agency
problem.6 To make the analysis interesting we will
thus envision the case where efforts in building and
managing assets are non-verifiable and delegation comes
with moral hazard. We then ask whether or not agency
costs exhibit some kind of economies of scope when
tasks are bundled. The analysis shows that ownership
and its impact on incentives is not the key to understanding the optimal form of procurement. Instead, the
key reason for bundling is to be found in technology and
in the impact of a good design on operating costs, not on
the ownership issue which is only secondary. This result
is quite robust to the space of compensation schemes that
can be used by the local government to delegate the
services and to the exact organizational form taken by the
merger of two firms when tasks are bundled.
Two cases are a priori feasible and are documented by
practitioners. First, a better design of the infrastructure
may help to save on operating costs, the case of a positive
externality. The prison sector serves as a prime example
because the design of a prison may significantly affect the
cost of implementing a given security level.7 Second, a
better design may also require learning new procedures
for managing assets and thus increase operating costs, the
case of a negative externality. As an illustration, the report
made by the French Cour des Comptes following the
Roissy Airport Terminal E2 crash argued that an
important issue was that Aroport de Paris cumulated
several hats as an owner of the infrastructure, a designer
and a builder. It argued that this bundling of tasks induced
a sacrifice in terms of the quality of the infrastructure.
With a positive externality both tasks should be
performed by the same firm which is better able to
internalize the impact that a better infrastructure design
has on operating costs. Intuitively, under moral hazard,
there is a trade-off between providing incentives to the
builder to improve the quality of the infrastructure and
giving him insurance against adverse shocks on the
realized quality. This trade-off calls for reducing the
power of incentives so that the builder exerts less than the
first-best effort. This decreased quality of the assets may
excessively increase the operating costs and thus exerts a
negative externality on the operator if building and
managing assets are unbundled. The builder and the
operator should thus be merged into a single entity. The
optimal organizational form exhibits thus an important
6

Otherwise the first-best could be achieved with simple forcing

contracts, thereby making the organizational issue of whether to
bundle the two tasks irrelevant.
7
See Schneider (2000).

D. Martimort, J. Pouyet / Int. J. Ind. Organ. 26 (2008) 393411

feature found in publicprivate partnerships. For a

negative externality, the two tasks should be split
because solving the agency problem on one task
exacerbates the incentive problem on the other. This is
reminiscent of the tasks separation occurring under
standard procurement practices.
The previous argument behind the optimal organizational form is thus unrelated to the ownership issue. In
practice, performance contracts are, however, not
always feasible and ownership matters. For instance,
the quality attributes of an infrastructure may be hard to
specify in advance so that complete contracting with a
builder may be difficult or even impossible to write.
Ownership provides then incentives to improve quality.
The allocation of ownership can thus be viewed as a
specific form of contracts with imperfect incentive
alignment and imperfect insurance properties.8
When incentives for building can only be provided
by allocating ownership, the decision whether to bundle
the two tasks may help to improve the quality-enhancing
effort. For instance, when the private owner does not
have enough private incentives to improve the quality of
the assets, making him also responsible for the
management of these assets fosters incentives in the
case of a positive externality. In such incomplete
contracting environment the modern form of public
private partnerships emerges when private owners have
rather weak incentives to enhance assets quality
compared with what would be socially optimal. On
the other hand, the traditional form of procurement
emerges when the externality is negative and uncertainty on the realized quality of the assets is too large to
let private owners bear such risks.
Although the normative arguments above have
certainly some appeal, they do not explain the fierce
opposition to the modern form of publicprivate
partnerships that is sometimes found among practitioners and political decision-makers. Opponents often
argue that this organizational form may increase the
scope for capture of the decision-maker so that the
possible efficiency gains from bundling may be offset
by influence costs.9 In fact, as a decision-maker may
find both bundling and unbundling optimal depending
on the kind of externality between tasks, he may exert
his discretion to favor the industry by this organizational
8
Because assets are privately owned, the owner may not internalize
the full social value of her investment in enhancing the quality of the
infrastructure.
9
In Libration dated June 21st 2004, Arnaud Montebourg, an
impetuous young leader of the French socialist party argued that PPPs
had a caractre opaque et corrupteur (a feature of opacity and
corruption).

395

choice. To analyze these issues we must significantly

extend our model. First, the decision-maker must have
private information on the sign of externality so that
manipulations of his decision can be made at the
expense of the general public. Second, the operator
willing to integrate backwards into infrastructure
building must also obtain some rent from doing so
and, here again, some sort of private information is
needed.10 Now, the political economy drawback from
the bundling decision becomes clearer. Because bundling is called for in the case of a positive externality, it
raises also incentives to improve operating costs. Under
adverse selection, this is a source of a greater
information rent. 11 Even when the externality is
negative and unbundling is socially optimal, the
operator has an incentive to bribe a (non-benevolent)
decision-maker to integrate backwards and also build
the infrastructure by herself. When the social cost of
such collusion is taken into account, bundling may not
be as attractive.
Let us now turn to a brief review of the literature. Two
papers address issues close to ours: Bennett and Iossa
(2002) and Hart (2003). Both papers lie in the realm of the
property rights literature la Grossman and Hart (1986):
Inefficiencies in assets quality-enhancing and costreducing efforts stem from the hold-up problem that
arises when no contract can be written and only ex post
negotiation between the government and the operator and/
or builder is feasible. Although ex post efficient, this
negotiation generates payoffs which depend on the threat
points defined by the ownership structure.12 By a
reasoning close to the one we will make in our more
complete contracting environment, a positive externality
somewhat weakens the hold-up problem on both tasks
and calls thus for integration. Even though they are similar
in spirit, our findings should nevertheless be distinguished
and contrasted. First, even though, we are quite
sympathetic to the idea that the quality of assets may be
hard to describe in advance, so that complete contracts
with a builder may be difficult to enforce,13 one may be
more skeptical of the use of this paradigm when it comes
to analyzing the relationship between the government and
the operator. Operating costs are readily observable and
often used in practice to contract for service provision.
This suggests that the role of ownership might have been
overemphasized so far. More basic agency problems may
actually explain much of the organizational forms which
10
The pure moral hazard model analyzed in the first part of the paper
does not generate any rent to the builder and operator.
11
See Laffont and Tirole (1993, Chapter 1) for instance.
12
See also Hart, Shleifer and Vishny (1997) for such an analysis.
13
Indeed we use this idea in Section 6 below.

396

D. Martimort, J. Pouyet / Int. J. Ind. Organ. 26 (2008) 393411

emerge, even though the distortions due to ownership

allocations can be superimposed. Second, because the
property rights approach de-emphasizes informational
issues, it cannot endogenize the stake for capture and
address the political economy issues which are crucial to
getting any positive theory of publicprivate partnerships.
This is where a second important insight available within
our framework lies.
This paper belongs also to a broader theoretical
literature which investigates task assignments in organizations in the presence of agency problems. In pure
moral hazard environments, Holmstrm and Milgrom
(1991) showed that incentives in one task may destroy
incentives in another when tasks are substitutes in the
agent's cost function; a result which suggests that tasks
should be split when there is a negative production
externality. Although the result that complementary
tasks should be bundled altogether can also be found in
Holmstrm and Milgrom (1990), Macho-Stadler and
Perez-Castrillo (1993), Itoh (1994), and Ramakrishnan
and Thakor (1991) under various forms, the specific
context of publicprivate partnerships and, most
specifically, the sequentiality of tasks imposes some
specific assumptions on contracts under unbundling and
a more thorough discussion of what is cooperation
between separated entities than what the existing
literature provides.14 In particular, we will distinguish
below between the case where the two tasks are bundled
altogether and performed by the same agent, keeping her
risk tolerance as given, and the case of a consortium
where the two tasks are jointly performed. The first of
these organizational choices focuses on the incentive
effect of bundling tasks, whereas the second one
introduces risk-sharing benefits which are already
well-known from the literature.15 From a methodological perspective, when considering the bundling of tasks,
our analysis allows one to clearly disentangle the impact
on incentives from the benefit associated with improved
risk-sharing. Schmitz (2005) investigates a sequential
moral hazard model with limited liability as the source
of the agency problem, no production externality but
with the added twist that the outcome of the first project
affects the cost of incentives on the second one. Finally,

14

More generally, the impact of production externalities on the kind

of incentive schemes used in multi-agent contexts (most notably
relative performance evaluations versus joint performance evaluations), keeping this separation between agents as given, has also been
investigated in Choi (1993) and Che and Yoo (2001).
15
This result is already well-known from Holmstrm and Milgrom
(1990), Macho-Stadler and Perez-Castrillo (1993), Itoh (1994), and
Ramakrishnan and Thakor (1991).

in pure adverse selection frameworks, Baron and

Besanko (1992, 1999), Dana (1993), Gilbert and
Riordan (1995), Laffont and Martimort (1998), Mc
Afee and Mc Millan (1995), Mookherjee and Tsumagari
(2004), and Dequiedt and Martimort (2004) have also
discussed whether bundling tasks and having a single
agent privately informed on cost parameters related to
each task dominates unbundling when tasks are perfect
complements.
The paper is organized as follows. Section 2 presents
the model. Section 3 addresses the respective optimality
of bundling and unbundling tasks when both the builder
and the operator receive a compensation scheme which
depends only on their own performance. This means
that, although the operator's cost may later on reveal
some information on the builder's effort, costs are not
used to compensate the builder. Under bundling the two
tasks are undertaken by a unique firm, the merger of the
builder, and the operator. Section 4 generalizes our
findings to the case where the cost realizations can also
be used to compensate the builder and delayed payments
are feasible. Section 5 enters in more detail into the
process of merger formation. Within a consortium two
otherwise identical risk-averse firms perfectly coordinate their decisions and share risk. Again bundling is
optimal for a positive externality but may also be so for a
negative one thanks to a coinsurance motive. Section 6
tackles the ownership issue and isolates conditions
under which either the more traditional form of
procurement or the more novel form of publicprivate
partnerships dominates. Section 7 discusses the political
economy of the model. Section 8 briefly concludes by
presenting alleys for further research. Proofs are
relegated to an Appendix.
2. The model
Consider the two tasks of building a key infrastructure and managing these assets: Should these tasks
be bundled and performed by the same firm or by two
different entities (sometimes referred to as the agents)
contracting independently with a common local government (the principal).16 Let the builder of this infrastructure (resp. the operator) be denoted by B (resp. O).
A merger of those two firms is denoted accordingly as
BO.

16

Our focus on two tasks only is made in order to capture the

essence of the argument. In the real word one often has to distinguish
between designing a project, getting outside financing, building the
corresponding infrastructure, and managing these assets.

D. Martimort, J. Pouyet / Int. J. Ind. Organ. 26 (2008) 393411

Both firms are symmetric and have the same CARA

utility function with risk-aversion r.17 A merger BO is
also assumed to have the same degree of risk aversion. A
merger corresponds thus to the case where both tasks are
a priori allocated to one of the two existing firms.18
Building and managing assets are two activities
which are both subject to moral hazard. Although the
builder exerts a non-verifiable effort e1 to improve the
intrinsic quality of the infrastructure, only a rough
quality index q can be used for contracting purposes
with:
f
q e1 f
e;
where is a random shock which is normally distributed
with zero mean and variance 2.19 Indeed, the intrinsic
quality of an infrastructure may not be fully observable.
In the case of water networks, although the quality of
produced water can be tested and specified in the
contract between the service provider and the municipality, it is only indicative of the effort incurred in
correctly designing and building these assets. The
intrinsic quality of the network (quantity of leakage,
quality of the tubes, etc.) remains by large not
observable. Other interpretations of our quality index
include the delay in building the infrastructure or the
adequacy between the agent's design of the infrastructure and the community needs.
The local government withdraws a benefit S q
(where S N 0) from building an infrastructure with
quality q. This benefit includes not only the private
value pocketed by owners, but also any externality that
this infrastructure may generate in terms of employment
or boosting economic activity. Society also withdraws a
benefit from the operations that we keep as fixed and
does not need to be modeled for simplicity. The local
government has all the bargaining power in designing
contracts with the builder and the operator of these
assets capturing thereby the ex ante competitive supply
of services in the production of key public services.
The operating costs c are observable and contractible.
However, they reflect only imperfectly the operator's
non-verifiable effort e2 in reducing these costs. We
postulate the following relationship:
f
cf
ge de ;
1
2

e2

Exerting effort ei on given task costs wei 2i to

the concerned agent.20 For a merger these disutility
functions are additive to avoid any systematic bias due
to either economies or diseconomies of scale in the
comparison of both organizational structures. For
simplicity, both firms have the same exogenous
reservation payoff which we normalize to zero.
Importantly, operating costs depend on the quality of
the infrastructure (see Eq. (1)). The sign of this
externality plays actually a major role in comparing
organizational structures as we will show below:
Positive externality, N 0: Building an infrastructure
of greater quality reduces operating costs. This
happens when, for instance, these infrastructures
facilitate operating tasks. Prisons provide an interesting example along these lines. A better design
certainly makes it easier to maintain safety.
Negative externality, b 0: Sometimes a quite novel
infrastructure design calls for innovating in some of the
most operating tasks, giving up routines and learning
new job processes. This certainly increases operating
costs at least in the short-run. Airports may be a case in
order here. A good design in view of facilitating
passengers access to terminals may nevertheless be
accompanied by an increase in the costs of providing
all other services required (food, shops, airplane
maintenance of the tarmac, etc). It is also a major
complaint that designing, building, or maintaining
existing assets are tasks which might involve large
dissonance between the different teams involved;
operating teams often report that the complex design
of an infrastructure coming out of the design and
construction team can hardly be operated without
cost overruns. This negative externality can then be
viewed as a short-cut for those dissonances which may
be due to cultural backlash between different teams or
to improper incentives for their members.
To avoid corner solutions at zero in the effort choices,
we will assume that is small enough in the case of a
negative externality.
2.1. Complete information benchmark

where is a random variable normally distributed with

mean 0 and variance 2.
17

397

The symmetry assumption is again made for simplicity only.

Section 5 will analyze the case where the two firms remain
independent entities coordinating their efforts in a consortium.
19
We denote Ex the expectation operator w.r.t a random variable x.

Suppose that efforts e1 and e2 are both verifiable.

The principal can thus use forcing contracts to
implement any such efforts to pair. Then, the first-best
efforts can be chosen and full insurance provided to both

18

20

The assumption of symmetry could again be relaxed at the cost of

increasing the notational burden.

398

D. Martimort, J. Pouyet / Int. J. Ind. Organ. 26 (2008) 393411

firms by offering them fixed-fees which cover their

respective costs of effort. The first-best effort levels {e1,
e2} maximizes the government's expected social
welfare and one finds:
e1 S d and e2 1:
The first equality, for instance, just says that the
expected marginal benefit of raising the quality of the
infrastructure, including the extra benefits in reducing
operating costs, is equal to the marginal cost of that effort.
Of course, the organizational structure is irrelevant in
this complete information context. Whether bundling or
unbundling is chosen yields the same first-best outcome.
3. Organizational forms with restricted schemes
Following Holmstrm and Milgrom (1987), we focus
on compensations which are linear in the agents'
realized performances.21
3.1. Unbundling
In this environment, general linear contracts should
be of the form t(q, c) = b + aq ac for the builder and
z(q, c) = c + q for the operator. Such schemes
may, however, be hard to implement in practice and
even sometimes useless. Indeed, first note that there is no
value in making the operator's compensation depend on
the observable quality of the infrastructure. Doing so (i.e.,
0) would only increase the risk borne by the riskaverse operator without any positive incentive effect on
his effort supply. The informativeness principle tells us
that this cannot be optimal.22
Second, the builder's payment is made just after his
task has been accomplished, i.e., it takes place before
operating costs are realized. General schemes t(q, c) of the
kind above may not be feasible when payments cannot be
delayed because of, for instance, the well-known limited
commitment ability of local governments.23 Delayed
payments may also be susceptible to ex post collusion and
accounting manipulations between operators and local
governments. For these reasons, but also because it
21

We are not, of course, in a pure Holmstrm and Milgrom (1987)

environment since there are two sequential tasks. We conjecture that
there exists a dynamic version la Holmstrm and Milgrom (1987)
the limit of which would justify the use of linear contracts.
22
Holmstrm (1979).
23
This assumption on non-delayed payments is standard in the
literature. See for instance Laffont and Tirole (1993, Chapter 8) for an
analysis of repeated auctions of franchise contracts which also
assumes that delayed payments are not feasible.

already highlights the key costs and benefits of merging

tasks, we will thus focus in this section on restricted
schemes having a = 0 and = 0. The more complex
contracting environments where delayed payments are
available are analyzed in Section 4 below.
When the payment of an agent depends only on his
realized performances and not on that of the other agent,
B and O receive, respectively, the following linear
contracts:
tq b aq and zc bac
where b and are fixed-fee payments, whereas a and
are piece-rate parameters.
Given the above restricted incentive schemes, the
builder and the operator's marginal incentives to exert
effort are given by the slopes of their respective
incentive schemes:
a w Ve1 e1 and a w Ve2 e2

Because compensations are based on individual performances only, the externality between the two tasks does
not affect the agent's efforts. Of course, this externality
still plays a role because it affects the average cost and
thus how much profit is extracted from the operator
through the fixed-fee .
Assuming that the agents' profit has no weight in the
social welfare function, the local government induces
the agents to choose efforts {e1u, e2u} which are less than
their first-best values:
eu1

Sd
1
be and eu2
be :
1 rr2e 1
1 rr2g 2

Indeed, for each risk-averse agent there is a trade-off

between providing the agent with enough incentives to
exert effort on his task and reducing the risk he bears for
insurance purposes.24 Because incentives expose agents
to risk, the principal chooses contracts which implement
too little effort, even if there is an externality that she
internalizes in designing individual schemes.
3.2. Bundling
The merged entity BO receives now a linear
scheme which depends on both its performances in
building assets and operating them:
tq; c B aqac;
where B is an aggregate fixed-fee.
24

See Holmstrm (1979).

D. Martimort, J. Pouyet / Int. J. Ind. Organ. 26 (2008) 393411

Now there is no problem in having the total payments

t() being delayed until the operating costs are realized
since this is the same entity which receives them.
Alternatively, this payment can be decomposed into two
different parts: one being offered after the realized
quality has been observed, the other being delayed until
costs are observed. To induce effort and participation of
the merged agent only the inter-temporal transfer
matters.25
The merged entity BO can better internalize the
impact of raising the quality of the infrastructure on the
operating costs. To see how, note that the merged entity
maximizing the certainty equivalent of his expected
profit chooses the efforts:
a ad e1 and a e2 :

When the externality is positive, a bonus on cutting

costs not only helps to reduce costs by exerting more
operating effort, but also improves incentives in better
designing the infrastructure. The reverse happens in the
case of a negative externality. Then, the principal
dealing with a single agent cannot use the costreimbursement rule to provide incentives in two efforts
which affect these costs in opposite directions.
Having determined the optimal payments and
incentives under both organizational structures, we
now turn to comparing bundling and unbundling.
Proposition 1. Assume that efforts are non-verifiable.
Bundling is the optimal organizational structure if and
only if N 0 (positive externality). Efforts are ranked as
follows:

399

social welfare viewpoint.27 If the local government

could increase both efforts, it would unambiguously
raise welfare. This is precisely what the joint offer of the
contract to the merger does in the case of a positive
externality. Then, the merger finds it worth to expand
the effort in designing and constructing the infrastructure because it finds some additional rewards in doing so
through the impact on operating costs that it fully
internalizes. This is not the case with a negative
externality since then the merger gets torn between his
incentives to do a better design and the negative impact
it has on operating costs. A better provision of
incentives can be obtained by simply separating the
two tasks. Then, the principal is no longer asking the
agents to perform well on two conflicting tasks.
Incentives are better designed by having agents being
focused on only one task. Of course, this argument can
be improved since the local government can actually do
better by redesigning the incentives of the merger to
improve even more his ability to internalize the positive
externality.
This discussion raises two issues. First, what are
really the gains of a merger in the case where more
general contracts are already allowed under unbundling?
Do we still improve effort and welfare when merging
tasks and by how much? Second, what are the benefits
of having separate agents sharing risks in a consortium
instead of single firm in charge of both activities? Are
they enough to expand the benefit of a merger beyond
the case of a positive externality? We answer these
questions in the next two sections.

eb1 Neu1 and eb2 Neu2 fdN0:

4. General schemes

To understand this result, let us suppose that the local

government decides to offer to the merger an aggregate
contract that would just consist of adding up the two
linear schemes offered to the builder and the operator if
they had been kept apart. In terms of risk tolerance,
doing so keeps unchanged the overall risk premium paid
by society to induce participation of this merger.26 If
there are any gains of merging tasks, they should be
found on the incentives side. Under unbundling,
remember that efforts are too low from a (first-best)

To test the robustness of our results we now assume

that more complex contracts can be implemented under
unbundling. The contract offered to any given agent is
still linear, but can now also depend on the other agent's
realized action. This possibility allows one of course to
achieve a weakly higher welfare under unbundling since
the space of contracts is enlarged. Hence, under a
negative externality, unbundling still dominates bundling. The only relevant issue is thus to compare these
organizational choices assuming instead a positive
externality.

25

One can also adjust fixed-fees in each period to make this intertemporal contract robust to the possibility that the agent leaves the
relationship after having built the infrastructure.
26
Things are different in Section 5 below, where we model the
bargaining process between two separate units to jointly decide on
effort levels.

27
It is well-known that the second-best level of effort in a pure moral
hazard environment may not always be below its first-best level (see
Laffont and Martimort (2002, Chapter 5) for instance). However, the
lessons of the linear-CARA model la Holmstrm and Milgrom
(1987, 1991) capture the Folklore of the profession.

400

D. Martimort, J. Pouyet / Int. J. Ind. Organ. 26 (2008) 393411

Payments to the builder and the operator can now be,

respectively, written as:
tq; c b aqa Vc and zq; c bac a Vq:
Since the externality is one-sided, making the
operator's payment depend on the realized quality is
useless for the local government. Doing so would again
only increase the risk faced by the operator, thereby
leading to an increase in the risk-premium needed to
ensure his participation without any incentive benefits.
In the following = 0 is still optimal.
By contrast, linking the builder's payment to the
operator's cost (through delayed payments for instance)
makes the builder internalize at least partly the onesided externality. The builder's marginal incentives to
enhance the infrastructure quality are now given by:
a aVd e1 :

The builder's effort is greater if he receives a share of

the gains in cost reduction that a better design would
permit.
To better understand the comparison between
organizational forms, it may be useful to optimize
expected social welfare first over the piece-rate parameter a to find:
aV

de1 r2e
;
r2g d2 r2e

which is positive under a positive externality. As

expected, the builder's payment decreases when costs
are higher since costs provide information on the fact
that the builder's effort may have been too low.
Moreover, since the risk on costs can now be used to
provide incentives, there is no need to let as much of the
quality risk to be borne by the builder. Overall, the tradeoff between incentives and insurance is relaxed and the
risk-premium that society has to pay to induce the
builder's participation is reduced accordingly. We show
rr2e r2g
e2 that
2r2g d2 r2e 1
rr2
risk-premium2 2e e21
rg

in the Appendix that the risk-premium

must now be paid is lower than the

paid with restricted contracts. The fact that r2 d2 r2 b1
g
e
captures therefore the informativeness gain from using
costs to improve the builder's incentives. This gain is
second-order in the size of the externality . Hence,
using costs to contract with the builder does not bring
much in the limit of a weak positive externality.
Let us now come back to the case of bundling. The
discussion after Proposition 1 shows that, by simply
merging incentive schemes, the quality enhancing effort
already increases in the size of the externality by first-

order magnitude. Although the use of a larger class of

contracts improves unbundling, it is not enough to
reverse our previous findings.
Proposition 2. Assume that efforts are non-verifiable
and that general contracts are available. Then there
exists 0 N 0 such that for all [0, 0] bundling is still
preferred to unbundling. Unbundling is still preferred
for negative externalities.
The second part of the proposition shows that, in
practice, the gains from unbundling with a negative
externality can already be achieved with restricted
schemes. Delayed payments to the builder do not bring
much to the local government. This points to the fact
that limited commitment ability on the government side
does not prevent one from achieving much of the gains
from splitting tasks.
5. Consortium
So far our modelling of a merger BO of the two firms
has been rather crude. By assuming that both tasks were
performed by a single agent, either the builder or the
operator, we have alluded to the question of how such a
coalition between the two entities might be formed in
practice. In this section we precisely investigate this issue.
When considering a detailed analysis of the formation of a consortium between the two otherwise identical
risk-averse agents, two problems should be kept in
mind. First, by merging these two agents may be better
able to share risk. A coalition improves risk-sharing
compared to the case of a single firm.28 Second, by
merging these two agents may be more or less able to
observe each other's effort. The benefits of a coordinated choice of efforts might be somewhat dissipated by
the internal agency problem that such a consortium may
have to solve. To model a consortium of two otherwise
identical risk-averse firms we will put aside this internal
agency problem (efforts are mutually observable and
coordination is perfect) and focus on the risk-sharing
issue.29 We assume that the two firms form a jointventure denoted by J V which is assumed to be
infinitely risk-averse. J V Vs reservation payoff is
exogenously normalized to zero.
J V receives the aggregate net transfer t(q, c) = B +
aq c from the government and then redistributes
this transfer between the two individual firms B and O.
28

This point is well-known from the collusion literature in multiagents environments. See Varian (1989) and Itoh (1993).
29
The consortium acts thus as a syndicate in the sense of Wilson
(1968).

D. Martimort, J. Pouyet / Int. J. Ind. Organ. 26 (2008) 393411

We denote by t i(q, c) = B i + a i q i c the share of the

overall revenue which accrues to firm iiafB; Og.
Because J V is infinitely risk-averse, it will transfer all
risk on the aggregate transfer t(q, c) to the builder and
the operator, so that necessarily aB aO a and
aB aO a. Assuming that J V has all bargaining
power in designing the individual compensations of
the builder and the operator, 30 J V maximizes the
certainty equivalent of the aggregate payoff of the
firms. Because they have the same risk tolerance, the
firms share equally the risk of their aggregate
compensation:
aB aO

a
a
and aB aO :
2
2

For a given incentive scheme offered by the government, optimal effort levels are thus still given by Eq.
(4). The consortium is efficient in the sense that it
perfectly internalizes the effort externality just like a
merger in Section 3. However, because the two firms
share risk equally, the aggregate risk-premium to be
paid to induce participation of such consortium is half
of what was paid in the case of the merger. Even when
there is no externality, a consortium strictly dominates
because it allows a better allocation of risk between
two otherwise identical risk-averse firms.
Proposition 3. There exists 0 N 0 such that an efficient
consortium dominates unbundling for N 0.
This result reinforces again our previous findings.
Certainly, bundling must be observed for a positive
externality. It also offers a justification for our earlier
assumption that a merger keeps the same degree of riskaversion as the agents. This assumption allows one in
fact to focus on the incentives benefits of bundling and
to disregard the issue of risk-sharing, which is another
advantage already well-known in the literature.31
6. Ownership and organizational forms
We have so far assumed that the perceived quality of
the infrastructure q was observable and verifiable and
could thus be used in any contract linking the
government and the builder. Let us now suppose that
this variable is itself non-verifiable. In this incomplete
contracting environment the only feasible way of
providing incentives consists of allocating ownership
rights of the assets to the builder. Of course, ex post,
30

This assumption somewhat simplifies the analysis but could easily

be relaxed.
31
See Holmstrm and Milgrom (1990), Macho-Stadler and PerezCastrillo (1993), Itoh (1993, 1994), and Ramakrishnan and Thakor (1991).

401

once the quality q is observed, the government and the

agent can bargain over the realized gains from trade.
Whoever owns the assets enjoys a return P q (with
P 0) by disposing of the assets in case the ex post
negotiation breaks down. This can be viewed as the resale
value of these assets. Because assets may have a greater
social value than what they are worth to owners, we have
S = E + P P where E 0 captures the positive externality
impact of the infrastructure. Several origins can be found
in this discrepancy between the social and the private
values of the assets. Indeed, once built, assets could be
redeployed to social uses other than initially intended.
Second, the infrastructure may have a positive impact on
employment and local economic activities, as it has been
stressed by the New Economic Geography literature.
For both the case of bundling and unbundling we
may wonder what is the optimal ownership structure.
Our goal in this section is thus to investigate whether the
incompleteness of the contracts modelled by assuming
the non-verifiability of the perceived quality q affects
the choice of bundling tasks or not and, if it does, in
which directions these distortions should go.
Whatever the organizational structure chosen, the only
feasible contracts with the builder consists now of allocating
asset ownership.32 Of course, on top of this allocation, the
government has to still decide of an ex ante price to be paid
to the builder to induce his participation. On the other hand,
contracts with the operator keep the general linear form
used above. By jointly making these two different
assumptions concerning the two tasks, we capture what
seems to be a major feature of most real-world partnerships:
the difficulty to verify quality of an infrastructure and the
fact that costs instead are readily observable, verifiable, and
used in cost-sharing agreements.33, 34
To understand the implications of ownership, it is
useful to see it as a simple contract fixing the marginal
incentives to improve the infrastructure quality to either
0 under government ownership or to P under builder
ownership. In doing so, we thus assume that the
government has all bargaining power in the ex post
negotiation that takes place with the builder once the
perceived quality q is realized. With this specification
in mind, it becomes easy to assess the quality-enhancing
effort of the builder under both ownership structures and
under both organizational forms.
32
We will assume that only deterministic ownership structures are
relevant, which is justified when ownership can be renegotiated.
33
Water management, waste disposals, transports, etc. are examples
in order there. This is contrary to what is assumed in Hart, Shleifer
and Vishny (1997) for instance.
34
See Hart, Shleifer and Vishny (1997) and Hart (2003) for similar
assumptions.

402

D. Martimort, J. Pouyet / Int. J. Ind. Organ. 26 (2008) 393411

6.1. Local government ownership

Let us first suppose that the government initially
approaches a constructor, who does not keep ownership,
before turning to a different operator. The builder has no
incentive to design a good infrastructure whatsoever and
thus exerts no effort, e1 = 0. Under bundling, instead, and
still assuming that the local government retains ownership,
two cases must be distinguished depending on the sign of
the externality. For a negative externality the merged entity
BO has still no incentive to make an efficient design
because this would only increase costs and reduce the only
verifiable performance on which he is rewarded. Again,
we have e1 = 0. Social welfare is unchanged and depends
only on the incentive for cost reduction.

restricted schemes where available for both tasks. The

sole role of ownership is to constrain the effort on the
first task to his marginal value for an owner not to the
slope of the optimal incentive scheme au found before.
Under bundling, the merged entity chooses a level of
quality-enhancing effort which takes into account the
impact on operating costs and we thus have:35
P ad e1 and e2 a:

The cost reimbursement rule has a role which is

complementary to ownership in boosting incentives in
the first task.
Proposition 6. Assume that N 0 and builder ownership. If P is small enough, bundling strictly dominates.

With a positive externality, the merged entity BO

may find it instead beneficial to increase the quality of the
infrastructure even though he does not own it, just because
this is another way of cutting operating costs. In fact, given
the positive slope N 0 of the cost-reimbursement rule, e1
is now raised so that =e1 N 0, whereas =e2.
Since, under government ownership, the merged
entity continues to bear no risk linked to the quality of
the infrastructure when it is not the owner, raising effort
e1 through bundling increases welfare at least to the first
order in when the externality is small.

The intuition behind this proposition is straightforward. When ownership by itself does not give enough
incentives to the builder to improve the quality of the
assets, bundling improves these incentives by making
the builder more eager to save also on operating costs.
Of course, for bundling to dominate, one wants to make
sure that ownership does not give too much incentive;
otherwise bundling would be suboptimal by worsening
an already suboptimal outcome.
When bundling dominates, one should note that
improving incentives on the first task requires also
pushing even further incentives on the second one.
Builder ownership also comes with cost reimbursement
rules which are closer to fixed-price contracts. Ownership and high-powered incentives go hand in hand.

Proposition 5. Assume that N 0 but small enough.

Under government ownership, bundling strictly dominates unbundling.

Proposition 7. Assume that N 0 and that P is small

enough. Then bundling and builder ownership is the
optimal organizational form.

Since explicit incentives to quality-enhancing and

implicit incentives through ownership are both absent,
the only way to induce quality-enhancing effort is to bundle
tasks so that the builder enjoys some benefit in exerting
effort e1 through the reduction of operating costs it induces.

As long as ownership provides enough incentives

(but still not too much) to the builder to improve asset
quality under bundling, the latter should not only own
the assets but also manage them. Proposition 7 highlights
conditions under which the most common form of
public-private partnership emerges. Bundling of tasks
helps to improve incentives in quality-enhancing effort
when ownership of the assets alone does not suffice.36
A contrario, let us find conditions under which public
ownership and separation dominates. This will correspond to the more traditional form of public procurement where two different agents are called for at the
building and operating stages and government retains

Proposition 4. Assume that b 0. Under government

ownership, bundling and unbundling yield the same
outcome.

6.2. Builder ownership

Under unbundling and when the builder owns the
assets, his quality-enhancing effort is given by:
P e1 :

As an owner enjoying the random private returns from

ownership, the builder bears also some risk and must be
compensated for doing so through an ex ante riskr
premium r2e P2 so that he prefers becoming an owner
2
than taking his outside option. Of course, social welfare
is still maximized by choosing the same effort e2u as if

We assume again that is small enough to ensure a positive effort.

In case the externality generated by the project E is negative, the value
of the private benefits P is already too high with respect to what is socially
optimal and private ownership gives too much incentives to improve
quality to make bundling optimal. We chose to disregard this case.
35
36

D. Martimort, J. Pouyet / Int. J. Ind. Organ. 26 (2008) 393411

ownership. We already know from Proposition 4 that

separation and integration are equivalent under government ownership and when the externality is negative:
No incentives on quality-enhancing can be provided.
The hope for unbundling to strictly dominate in this case
thus vanishes. Nevertheless, we have:
Proposition 8. Assume that b 0 and that P and 2 are
large enough, then public ownership and unbundling is
the optimal organizational form. This is more likely as
2 increases.
With a negative externality, the only way to incentivize
effort on the first task is to give ownership to the builder.
The cost of ownership is that the buyer must be
compensated for the risk associated to the realized quality
of these assets. When the private benefits of ownership are
too high, the buyer as an owner has too much incentives
and bears too much risk. Public ownership is preferred
when this insurance motive is too costly, i.e., when
uncertainty on the quality of these assets is too large. This
is so even if public ownership destroys any incentive to
improve this quality.
7. PPPs and capture
So far the decision whether to bundle tasks was taken
by a benevolent social welfare maximizer. Opponents to
publicprivate partnerships have often argued that this
form of procurement increases the scope for capture of
non-benevolent decision-makers by private interests.
We now turn to this issue by introducing some political
economy considerations in our model.
The bare-boned model analyzed in Section 3 already
provides some hints to understanding why an operator
may want to influence a policy-maker to favor bundling
even though that decision may not be socially optimal.
Indeed, we know from Proposition 1 that, under bundling
and with a positive externality, the optimal incentive
scheme offered to the operator is higher-powered than
under unbundling. Formally, b = e2b N u = e2u. In the pure
moral hazard model used so far, these higher-powered
incentives are not the source of any rent.37 From Laffont
and Tirole (1993) we know that high-powered incentives
may also give excessive information rents to the operator
in adverse selection contexts. These rents may provide
operator with an incentive to capture the decision-maker
to manipulate his decision so that bundling is always
chosen. Information rents are the engines of any capture
of such a decision-maker.

37

The operator's ex ante participation constraint is always binding.

403

Of course, for this story to hold, two extra ingredients

are needed. First, the decision-maker must be nonbenevolent and attracted by the prospects of withdrawing private benefits from conceding favors to the
operator. Second, the decision-maker and the operator
must share some piece of private information which is
not available to the general public and this piece of
information must give some rent to the operator.
In our context, this piece of information from which the
decision-maker gets discretion is the sign of the externality
between the two tasks. By hiding evidences on a negative
externality that would optimally call for unbundling, the
decision-maker may let the operator enjoy some extra
information rent associated with an inefficient choice of
bundling. Preventing such manipulation has a social cost
which must be taken into account at the time of evaluating
whether bundling is the most preferred organizational
form from a social welfare point of view.
To make the political economy model described
below more transparent, we depart from the ownership
considerations discussed in Section 6 completely. As
argued above, ownership problems arise in incomplete
contracting environments, but this incompleteness is not
really needed in order to understand the stake of the
operator in manipulating the public decision on whether
to bundle or not. What really matters is the link between
the information rent of the operator and the organizational structure.
To extend the scope of our previous model to a political
economy context, let us suppose that the level of the
externality is a random variable taking values in {, }
(where N 0) with respective probabilities and 1 .
We assume that is a piece of information learnt by both
the decision-maker and the operator. They may collude
to hide this piece of information from the general
public.
Let us also assume that the mean 0 of the shock on
operating costs is also a random variable taking values in
{ 0, 0} with respective probabilities p and 1p (with
0 = 0 0 N 0). The operator with type 0 is on

efficient. 0 is realized after has been

average more
learned by the decision-maker and the operator. This is
the operator's private information on 0 which gives him
an information rent.
7.1. Benevolent decision-maker
Suppose first that the decision-maker is benevolent
and reveals truthfully any information he may have on
so that the best organizational form is always chosen.
In this environment with asymmetric information on
0, an incentive mechanism is a menu {( 0), ( 0),

404

D. Martimort, J. Pouyet / Int. J. Ind. Organ. 26 (2008) 393411

a( 0), b( 0)} (resp. {( 0), a( 0), B( 0)}) under

unbundling (resp. bundling) where
0 is the operator's
report on 0. According to the Revelation Principle,38
there is no loss of generality in focusing on such truthful
mechanisms. Once the operator has picked the contract
corresponding to his type 0, effort levels are chosen
according to the organizational structure which prevails.
7.1.1. Unbundling
When the operator reports 0, he chooses an effort
e2 = ( 0) whereas the builder chooses e1 = a( 0). The
certainty equivalent of the operator's expected utility is:
U O g0 ; g0
bg0 ag0 g0 ag0 dag0

1 rr2e 2
a g0 :
2

Denoting U O g0 ; g0 UO g0 , the relevant adverse

selection incentive constraint of a low-cost operator can
be written as:
10
UO g0 zUO g0 ag0 Dg0 ;

whereas the participation constraint of a high-cost operator

is:
UO
g 0 z0:
11
These are the only binding constraints at the optimum.39 We
show in the Appendix that the socially optimal mechanism
implements the same effort level for an efficient operator
than when 0 is common knowledge. Only the power of the
inefficient operator's incentive scheme diminishes to reduce
the adverse selection information rent of a 0 operator:

au g0 ; d au g0 ; d eu1 ;
12

p
1 1p
Dg0
au g0 ; d eu2 Nau g0 ; d
:
13
1 rr2g

In the sequel we shall assume that 1N

p
Dg to maintain
m1p 0

a positive effort by both types of operators even under

adverse selection and the threat of capture. We also make
explicit the dependence of the solution on when needed.
Finally, note that the builder's incentive scheme is not
distorted because it plays no role in reducing the operator's
rent. It thus induces the same quality-enhancing effort as in
Section 3.
7.1.2. Bundling
The merged entity BO now internalizes the impact
of his quality-enhancing effort on cutting operational

costs. BO chooses effort levels on both tasks, which

are, respectively, given by e2 = (0) and e1 = a(0)
(0), when he reports having realized average costs
0. BO gets thus a certainty equivalent of his expected
utility which is worth:
U BO g0 ; g0 Bg0 ag0 g0 ag0 dag0
dag0 ag0 ag0 dag0

1 rr2g 2
a g0
2

1 rr2e
ag0 dag0 2 :
2
Denoting the merged entity's information rent by
UBO g0 U BO g0 ; g0 , the relevant incentive compatibility and participation constraints are still Eqs. (10)
and (11). Both constraints are, again, binding at the
social optimum and, again, only the bonus (0) is used
to extract the costly information rent of the 0-operator.

This leads to the solution:

14
ab g0 ; d eb1 Nab g0 ; dfdN0;

ab g0 ; d eb2 Nab
g0 ; d:
15

As previously, reducing the information rent of the

efficient operator calls for moving the cost-reimbursement rule towards a cost-plus contract. Depending on
the sign of the externality, this has also a negative impact
on the merged entity's incentives for enhancing the
quality of the infrastructure. When N 0, having more of
a cost-plus contract on operations also reduces these
incentives, whereas the reverse happens with b 0.
Gathering the results of the optimization both with
respect to bundling and unbundling, we observe that
the only role of adverse selection is to diminish the social
benefit of inducing a cost-reducing effort by
the inefficient operator. Instead of being equal to 1, as
before, this social benefit must be reduced in order to
take into account of the socially costly information
rent left to the most efficient operator. The corresponding
p
virtual social benefit becomes 1
Dg . These distor1p 0
tions are thus independent of the sign of the externality.
Since the optimal organizational choice does not depend
on the social benefits of both tasks, but only on the sign
of , we can directly import the results of Proposition 1.
Proposition 9. Assume that the operator has private
information on his average costs 0 and that the decisionmaker is benevolent. Then, the optimal organizational
form is still bundling (resp. unbundling) when N 0 (resp.
b 0).

38

See Laffont and Martimort (2002, Chapter 2).

This is a standard result of two-type adverse selection models. See
Laffont and Martimort (2002, Chapter 2), for instance.
39

Asymmetric information on 0 per se is not enough

to modify the basic insights of Section 3 when the

D. Martimort, J. Pouyet / Int. J. Ind. Organ. 26 (2008) 393411

decision-maker is benevolent. A positive externality still

calls for bundling.
7.2. Non-benevolent decision-maker
Suppose now that the decision-maker is nonbenevolent and may be captured by the privately
informed operator who withdraws some information
rent. This decision-maker is thus now viewed as a
strategic player with his own private incentives. In
particular, he must be induced to reveal to the public the
realized value of .
Let us suppose that a negative externality may be
manipulated and publicly reported as being a positive
one. In contrast, the reverse manipulation is assumed to
be not feasible.40
When the decision-maker hides the realized negative
externality from the general public and reports
instead a positive externality
, the decision of whether
to separate the two tasks is unduly modified into a
decision of whether to bundle them. Through this
modification, the operator increases then his expected
information rent by an amount:
pDg0 ab g0 ; dau
g 0 ; d :
This quantity represents the stake of capture. It is in fact
positive when evaluated at the optimal incentive schemes
found above. Indeed, it is proportional to the difference in
the efforts made by an inefficient operator between the
cases of bundling and unbundling, namely e2be2u, and this
quantity is positive at as one can see from Proposition 1.
We will assume that the non-benevolent decisionmaker has all of the bargaining power in the collusive
side-deal with the operator. Before the operator knows 0,
the decision-maker makes thus a take-it-or-leave-it offer,
asking for a bribe equal to p0[b(0, ) u(0, )]
and, in exchange, commits to report = when instead
= .
Following Tirole (1986), the decision-maker enjoys
an ex ante private benefit from his possible collusion
with the operator which is worth:
0g:
k1mpDg0 maxfab g0 ; dau
g 0 ; d;

16

The fraction 1 k represents the deadweight-loss of

capture associated to the fact that side-deals are
unofficial side-contracts which are enforced only by
repetition, words of honor, etc., or which entail nonmonetary transfers between the colluding partners.
The information structure is thus such that is partially verifiable
in the sense of Green and Laffont (1986).
40

405

Preventing capture of the decision-maker is socially

costly. To behave, the decision-maker should actually
receive at least his expected benefit from manipulating
information. Eq. (16) is thus an extra agency cost that
must be paid by society to prevent capture. This agency
cost is reduced by distorting downward b(0, ) and
by increasing u(0, ). This stake fully disappears if
b(0, ) is equal to u(0, ).
Keeping as fixed the decision rule to bundle tasks when
N 0 and to split them when b 0, one may investigate
how the threat of capture changes incentives. Note first
that one may always choose b(0, ) =u(0, ). With
such a policy, the decision-maker has no longer any
discretion over tailoring the power of the operator's
incentives to the sign of the externality. The stake of
capture disappears even though one may still benefit from
a positive externality under bundling in state in order to
boost the effort on the first task.
Proposition 10. There exists 0 N 0 such that for b 0
capture is never a concern but the cost reimbursement
rule does not depend on : b(0, ) = u(0, ). More
generally, b(0, ) (resp. u(0, )) is reduced (resp.
increased) under the threat of capture.
The possibility to manipulate the sign of the externality
gives discretionary power to the decision-maker. When
the externality is negative, he might instead reveal to the
public that it is positive thereby leading to bundling
whereas splitting tasks would have been optimal. To avoid
this issue, cost-reimbursement rules must be tilted
towards cost-plus contracts under bundling and towards
fixed-price contracts under unbundling. This means that
efforts will be somewhat misallocated under both
organizational forms with little internalization of the
positive externality under bundling. This convergence of
cost-reimbursement rules might go up to the point where
they are the same, irrespective of the organizational form.
Then the collusive stake fully disappears, but much of the
benefits of bundling are lost in the process.
8. Conclusion
The presence of a production externality between
building and operating assets raises the issue of the
optimal organization of such tasks. Bundling allows to
better internalize this externality and improves incentives
when the externality is positive, thereby increasing
welfare. In contrast, when the externality is negative,
unbundling reduces agency costs and is socially preferable. Hence, a simple and technology-driven reason is at
the heart of the decision to bundle or unbundle the
various activities. This result is obtained under a class of

406

D. Martimort, J. Pouyet / Int. J. Ind. Organ. 26 (2008) 393411

restricted schemes where the compensation of an agent

depends only on his own performances, but it holds more
generally with a larger class of compensations, whether
the decision-maker relies on a single agent or on a
consortium to perform both tasks and even under
asymmetric information and the risk of capture. In each
case, we showed how the basic lessons of our
framework should be modified and, in particular, the
directions towards which cost-reimbursement rules
should be distorted to account for various contractual
environments.
Even though this organizational issue seems a priori
somewhat orthogonal to the ownership issue which has
often been pushed forward in the PPP literature, it helps to
understand part of the debate about it. To the extent that
ownership of an infrastructure is a very crude contract, but
sometimes the only feasible one, leaving ownership to the
builder might be the only device to provide incentives to
building and designing it efficiently. The discrete contractual choice of whether to leave ownership does not allow to
fine tune incentives as efficiently as under more complete
contracts. Nevertheless, the insights obtained in our more
complete contracts framework might at least partly carry
over. Depending on the private benefits that the builder
withdraws from ownership and the risks he bears when
acquiring assets, publicprivate partnerships defined as
the joint bundling of tasks and allocation of ownership to
the builder might nevertheless outperform the more
traditional form of public procurement (where a local
government keeps ownership of the assets and chooses two
different contractors at each stage) in the case of a positive
externality. The reverse may happen for a negative
externality. Of course, all the extensions investigated in
this paper could also be possibly cast in an incomplete
contracts setting. We feel reasonably confident that the
results found in more complete contracting environments would carry over at least to some extent, but,
certainly, some more formal analysis is required to
qualify this assertion.
Coming back to our initial complete contracting
framework, several other extensions seem to us
particularly attractive. First, it could be worth coming
back to the maintained assumption that the firms' degree
of risk-aversion was kept constant as one changes
organizational modes. Section 5 has gone some way
towards endogenizing that degree but certainly more
could be done. More specifically, one may be interested
in tracing out the impact of organizational forms
(whether firms are multi-taskers or not) on their access
to the financial market and thus on the amount of risk
they should keep as a result of frictions on these markets.
The corporate finance literature (see Leland and Pyle

(1977), for instance) suggests reasons (related most

notably to asymmetric information and frictions in
getting access to the capital markets) why firms may
not be able to fully diversify. In a full-fledged model one
might want to analyze how the decision whether to
bundle activities affects these frictions on the capital
market. The implicit assumption that we maintained here
was that these frictions remain by and large independent
of the chosen organizational form but the validity of this
assumption should be further investigated.
Second, in our modelling of consortia between
builders and operators, we have assumed that efforts
of the member firms could be coordinated efficiently.
This assumption should be relaxed. Consortia may be
inefficient when they suffer from internal agency
problems. These problems may tilt the organizational
choice towards unbundling. In an incomplete contract
perspective this would make a stronger a case for the
more traditional form of procurement.
Third, our approach of the political economy of PPPs
has been very partial in stressing only one aspects of it:
the possibility of excessive manipulations by decisionmakers to favor the formation of large conglomerates.
We show that a robust form of procurement immune to
the risk of capture goes towards low-powered incentives
and cost-reimbursement rules that might not really help
firms to internalize the gains from bundling. This is
clearly reminiscent of a move towards traditional forms
of procurement. The extent to which the choice of
procurement modes responds to political forces remains
nevertheless by large to be investigated.
Finally, it could be worth investigating whether
competition between potential builders and between
operators may also change the incentives to form
consortia and the decision of whether to bundle activities.
Appendix A
A.1. Proof of Proposition 1
Given a linear restricted incentive scheme, the
builder wants to maximize the certainty-equivalent of
his expected utility, namely:
b ae1 we1

rr2e 2
a:
2

The builder's effort is thus given by Eq. (2). The fixedfee b chosen by the government extracts all the builder's
expected profit:
b

e21
rr2e 1:
2

A1

D. Martimort, J. Pouyet / Int. J. Ind. Organ. 26 (2008) 393411

Similar computations can be made for the operator who

wants to maximize:
bag0 e2 de1 we2

rr2g 2
a
2

and chooses an effort level given by again by Eq. (2).

The optimal fixed-fee which extracts all the operator's
expected profit is:
e2
b e2 g0 2 rr2g 1de1 e2 :
2

A2

407

and Wb(e1, e2, ), the welfare comparison between the

bundling and the unbundling scenarios only depends on
the comparison of effort levels. Simple manipulations
show that:
eb1 eu1 ~d1 r1 d2 dSr2e ; and
eb2 eu2 ~dS

rS dr2g :

The proposition follows.

A.2. Proof of Proposition 2

Taking into account the expression of these fixedfees, the local government optimizes under unbundling
the following expression of expected social welfare:

Under unbundling and general schemes the builder

maximizes:

Pu :

e1 a

max

e1 ;e2 a2

W u e1 ; e2 ; duS de1

e2
e2
e2 1 1 rr2e 2 1 rr2g :
2
2

A3

This leads to the agents' effort levels given by Eq. (3).

Under bundling the merger gets an expected payoff
worth:
Bag0 e2 de1 ae1 we1 we2

rr2e 2 rr2g 2
a
a
2
2

where the effort levels are given by Eq. (4). The fixedfee B is then used by the principal to extract B  O Vs
expected rent so that the government's problem can be
written as:
Pb :

W b e1 ; e2 ; duS de1

max

e1 ;e2 a2
e2 rr2
e2 1 e

e1 de2 2

e22
1 rr2g ;
2

A4

the optimum {e1b, e2b} of which is given by:41

eb1

eb2

S d1 rr2g d2 rr2e drr2e

1 rr2g 1 rr2e d2 rr2e
1 rr2e S ddrr2e
1 rr2g 1 rr2e d2 rr2e

and

A5

The builder's effort is now given by Eq. (5). As usual by

now, the principal sets the fixed-fee so as to extract all
the builder's expected rent, i.e.:

e

r
1
b r2e e1 aVd2 r2g aV2 e1
aVd
2
2
a Vg0 e2 de1 :
A6
Expected welfare under unbundling can thus be written
as:
W u e1 ; e2 ; a V; d S de1 e2

The second-order conditions are trivially satisfied.

e21 e22 rr2g 2

e
2 2
2 2

r
e1 da V2 r2e a V2 r2g :
2

A7
(A7)

To better understand the comparison between organizational forms, it may be useful to optimize first over the
piece-rate parameter a and, from there, get an indirect
welfare function Wu(e1, e2, ) which depends only on
the effort variables (e1, e2). Doing so yields Eq. (6) and a
new expression of the principal's maximization problem
as:42
fu
max W e1 ; e2 ; d
e1 ;e2 a2

To keep the analysis interesting, we will assume that

ekb 0 (for k {1, 2}) which necessarily holds when
N 0 (positive externality), but also when b 0 but small
compared with S (the case of a sufficiently weak
negative externality).
Remember now that agents get no expected profit in
each configuration. Moreover, looking at Wu(e1, e2, )
41

r
max b ae1 aVg0 e2 de1 we1 a2 r2e aV2 r2g :
2

S de1 e2

rr2e r2g
e21 e22 rr2g 2

e2
:
2 2
2
2r2g d2 r2e e21
A8

fu
fu
2W fu
2Wfu
2
We a l s o h a v e :
1 rr2 ,
2 1 rr ,
2
W
W
e 2
2 1 rrg2 ,
W e a l s o h a v e : e21e
e
2 1 rre ,
g
2 22
e
1
fu
fu
fu
fu
2
2
2
W
W
W
2W

0,
drr2e .
rd2 r2e r2g ,
0,
aV2
e1 e2
e2 aV
e1 aV
One can then check that the Hessian associated to the maximization
problem is negative semi-definite at the optimum.
42
42

408

D. Martimort, J. Pouyet / Int. J. Ind. Organ. 26 (2008) 393411

Solving for the first-order conditions associated with the

three free parameters e1, e2 and a, the optimal effort
levels under unbundling are given by:
eu1

S dd2 r2e r2g

r2g

r2e d2

rr2g

; eu2

1
:
1 rr2g

The optimal piece-rate parameters are denoted by u, au,

and au.
Let us now turn to the case of bundling. Remember
that expected welfare under bundling is given by Eq.
(A4). Comparing Eqs. (A8) and (A4) is then straightforward, at least in the limit of a small positive externality.
The gain from bundling is of a first-order magnitude in
since now a merged entity can better internalize the
choice of quality-enhancing and cost-reducing efforts.
Tedious, but straightforward, computations show that:




limdY0 W b eb1 ; eb2 W u eu1 ; eu2 ; aVu 0;


d  b b b
W e1 ; e2 W u eu1 ; eu2 ; aVu
limdY0
dd
rSr2e

N0:
1 rr2e 1 rr2g
A.3. Proof of Proposition 3
J V maximizes the certainty equivalent of the
aggregate payoff of the firms subject to constraints aB
aO a and aB aO a which captures the fact that
the aggregate compensation risk is shared between these
firms. J V Vs problem can thus be written as:
max

faB ;aB ;e1 ;e2 a2 g

B ae1 ag0 e2 de1

rr2g
e2 e2 rr2
aaB 2 a2B :
1 2 e aaB 2 a2B 
2 2
2
2
Because both firms have the same risk tolerance, they
share equally the risk of the aggregate compensation:
a
a
aB aO and aB aO :
2
2
For a given incentive scheme offered by the government,
optimal effort levels are thus the same as in Section 3:
e1 a da and e2 a:
The government's problem becomes thus:
max W c e1 ; e2 ; d

e1 ;e2 a2

e2 rr2
S de1 e2 1 e e1 de2 2
4
! 2
2
2
rr
e
g
2 1
:
2
2

Using Eq. (A5), but for a level of risk-aversion half as

high, the optimal effort levels with an efficient
consortium are now given by:



rr2
d2 rr2
drr2
S d 1 2g 2 e 2 e



and
ec1

rr2
rr2
d2 rr2
1 2 g 1 2e 2 e
ec2

rr2e
drr2
S d 2 e
2


rr2g
rr2
d2 rr2
1 2e 2 e
2

1
1

The comparison of Eqs. (A3) and (A9) is now

straightforward.
A.4. Proof of Proposition 4
Immediate.

A.5. Proof of Proposition 5

Following from the discussion in the text, social
welfare under unbundling can be written as:
e2
WGu e2 e2 2 1 rr2g ;
A10
2
where e2 is the operator's cost-reducing effort.
Since the merged entity bears no risk linked to the
realized quality of the infrastructure when it is not the
owner, social welfare can be written as:


d2 e22
WGb e2 ; d dS de2
2

e22
e2 1 rr2g :
A11
2
b
Denoting e2G
as the optimal effort, we have:

eb2G

1 dS d
:
1 rr2g d2

The first bracketed term is the social value of the

quality-enhancing effort when the incentives for doing
so come only from the willingness of the merged entity
to reduce his operating cost. Assuming that is small
enough, this is a positive term when evaluated at the
effort level e2u = 1 / (1 + r2 ) which maximizes Eq. (A10).
The second bracketed term is nothing else than the
u
expression for W G
(e 2). Henceforth, we get the
result.
A.6. Proof of Proposition 6

A9

(A9)

Under unbundling, the builder's payoff as an owner is:


e21
f
b
;
Efe;f
exp
r
Pe

1
g
2

D. Martimort, J. Pouyet / Int. J. Ind. Organ. 26 (2008) 393411

d
which is positive at eu2 when PVS d eu2 , i.e., P is small
2
enough.

whose certainty equivalent is worth:

Pe1

rr2e 2 e21
P b;
2
2

A.7. Proof of Proposition 7

where Eq. (8) holds and b is a fixed-fee paid by the builder

to become an owner.
Social welfare under unbundling expressed again as a
function of the operator's effort can be written as:
WBu e2 ; d E SPe1 f
e f
g e2 de1 bb
f
a g e de ;
2

S dP
e2

e22
2

P
1 rr2e
2

E f
f
e; g

We observe that:
WBb e2 ;
dWGb e2 ; d

1
2
P S dde2 1 rre P
2

b
when P is small enough.
which is positive at e2G

1 rr2g ;

We observe that:


e2
e2
g e2 de1 2
exp r Pe1 f
e 1 baf
;
2
2

WBu e2 ; dWGu e2 ; d S dP
which is negative when PN

Pe1

P2

e21
2

bag0 e2 de1

e22
2

rr2g
2

a2 ;

e b1af
g e2 de1 :
Efe;fg Ee1 f

1 rr2e 2
P
2
d2 e22
dPe2
dS de2
2


1 rr2e
Pde2
bP S d
2
b

Note that when the fixed-fee is used to extract all the

builder's expected payoff, social welfare becomes:
rr2e 2 e21
e2
P 1 rr2g 2 :
2
2
2

Or using Eq. (9) to eliminate e1:

WBb e2 ; d S dP de2
rr2e 2 1
P P de2 2
2
2
e22
e2 1 rr2g :
2

A12

Denoting eb2B as the corresponding optimal effort, we have:

eb2B

1 dS dP
:
1 rr2g d2

Simple manipulations show that:


d
WBb e2 ; dWBu e2 ; d de2 S d e2 P
2

2S d
. Hence, government
1 rr2e

WBb e2 ; dWGu e2 ; d S dP

where e1 and e2 satisfies Eq. (9).

The principal's expected payoff can be written as:

WBb e2 ; d S de1 e2

1rr2e 2
P
2

ownership is better than private ownership under

unbundling. Moreover, using Eq. (A13), we observe
also that when b 0:

whose certainty equivalent is worth:

rr2e

A13

A.8. Proof of Proposition 8

since, because he is not the owner, the principal enjoys

only E q from the infrastructure building. Of course,
this expression is still maximized for e2u.
Under bundling, the agent's payoff as an owner is:


409

which is a negative number at e 2 B when

2S ddeb2B 2S d
PN
N
. Since P b S, a necessary condi1 rr2e
1 rr2e
tion to have this last inequality, at least in the case where
2
is large enough, is that is large enough.
Therefore, government ownership and unbundling
dominates builder ownership whether bundling or
unbundling has been chosen. Note that these lower
2
bounds on P are more likely to hold when or ||
increases.
A.9. Proof of Proposition 9
Taking into account the binding incentive and
participation constraints, the socially optimal contract
under unbundling when the realized externality is
solves the reduced-form problem:

max pW u ag0; ag0; da g 0 Dg0 

fad;adg

1pW a g 0 ; a g 0 ; d:
u

The optimization is straightforward and yields Eqs. (12)

and (13).

410

D. Martimort, J. Pouyet / Int. J. Ind. Organ. 26 (2008) 393411

Similarly, the optimal contract under bundling

solves:
max pW ag0; ag0; dag 0 Dg0 
b

fad;adg

g 0 ; a
g 0 ; d:
1pW b a

ab g0; d eb1 Nab g 0 ; d

1 1p Dg0 1k u
f
g 0 ; d
Na g 0 ; d:
a u f
1 rr2g
p

A17

This is really the optimum when

b(0, ) N
u(0, ),
i.e., when there is a positive stake of capture. Let us
suppose that:


p
k1m
Dg0 1
1N
1p
m

which leads to
P

and:


p
S d1 rr2g d2 rr2e drr2e 1 1p
Dg0
1 rr2g 1 rr2e d2 rr2e

fdN0;

(A14)
ab g0; d eb2 Nab
g 0 ; d
P



p
1 rr2e 1 1p
Dg0 S ddrr2e

:
1 rr2g 1 rr2e d2 rr2e

A15

so that b(0, ) is a positive number for sure. Then,

note that as is small enough:



p
1 1p
Dg0 1 k1m
m
f
g0;
d f
bf
a u
g 0 ; d : A18
a b

1 rr2g
dY0
Thus for small enough, we have b(0, ) b
u(0,
) and a contradiction. This is more likely as k increases.
For small enough, the optimum is achieved when
(0, ) =
u(0, ).

A.10. Proof of Proposition 10

References
Taking into account that bundling (resp. unbundling)
is implemented when N 0 (resp. b 0), expected social
welfare can thus be written as:
mfpW b ab g0;
d dab g0;
d; ab g0;
d ;
d
P

g0;
d Dg0 1pW b ab
g0;
d
a
b

g0;
d; ab
g0;
d;
dg
dab
d ; d
1mfpW u au g0; d ; au g0;
P

g 0 ; dDg0
a
u

g 0 ;
d ; au
g 0 ; d ;
d g
1pW u au
g0;
d au
g0;
dg:
k1mpDg0 maxf0; ab
A 16

Let us first consider the case where the optimal bonuses

b(0, ) and
u(0, ) under the threat of capture are
b
such that
(0, ) N u(0, ) so that there is a positive
stake of capture. Optimizing yields:
f
a b g 0 ;
d


 
p
S
d d rr2e
1 rr2e 1 1p
Dg0 1 k 1m
m

bab
g;
d;

1 rr2g 1 rr2e
d rr2e

Baron, D., Besanko, D., 1992. Information, control and organizational

structure. Journal of Economics and Management Strategy 1, 237275.
Baron, D., Besanko, D., 1999. Informational alliance. Review of
Economic Studies 66, 743768.
Bennett, J., Iossa, E. (2002). Building and Managing Facilities for
Public Services. Mimeo Brunel University.
Berger, R., 1985. Private sector initiatives in the reagan era: new actors
rework on old theme? In: Salomon, L., Lund, M. (Eds.), The
Reagan Presidency and the Governing of America. The Urban
Institute Press, Washington D.S., pp. 181211.
Che, Y.K., Yoo, S.W., 2001. Optimal incentives for teams. American
Economic Review 91, 525541.
Choi, Y.K., 1993. Managerial incentive contracts with a production
externality. Economics Letters 42, 3742.
Dana, J., 1993. The organization and scope of agents: regulating multiproduct industries. Journal of Economic Theory 59, 288310.
Daniels, R., Trebilcock, M., 2002. An organizational analysis of the
publicprivate partnership in the provision of public infrastructure.
In: Vaillancourt Rosenau, P. (Ed.), PublicPrivate Policy Partnerships. MIT Press, Cambridge.
Dequiedt, V., Martimort, D., 2004. Delegated monitoring versus arm's
length contracting. International Journal of Industrial Organization
22, 951981.
Gilbert, R., Riordan, M., 1995. Regulating complementary products: a
comparative institutional analysis. Rand Journal of Economics 26,
243256.
Green, J., Laffont, J.-J., 1986. Partially verifiable information and
mechanism design. Review of Economic Studies 53, 447456.
Grossman, S., Hart, O., 1986. The costs and benefits of ownership: a
theory of vertical and lateral integration. Journal of Political
Economy 94, 691719.

D. Martimort, J. Pouyet / Int. J. Ind. Organ. 26 (2008) 393411

Hart, O., 2003. Incomplete contracts and public ownership: remarks
and an application to publicprivate partnership. Economic
Journal 113, 6976.
Hart, O., Shleifer, A., Vishny, R., 1997. The proper scope of
government: theory and applications to prisons. Quarterly Journal
of Economics 112, 11191158.
Holmstrm, B., 1979. Moral hazard and observability. The Bell
Journal of Economics 10, 7491.
Holmstrm, B., Milgrom, P., 1987. Aggregation and linearity in the
provision of intertemporal incentives. Econometrica 55, 303328.
Holmstrm, B., Milgrom, P., 1990. Regulating trade among agents.
Journal of Institutional and Theoretical Economics 146, 85105.
Holmstrm, B., Milgrom, P., 1991. Multitask principal-agent analyses:
incentive contracts, asset ownership and job design. Journal of
Law Economics and Organization 7, 2452.
Itoh, H., 1993. Coalitions, incentives and risk-sharing. Journal of
Economic Theory 60, 410427.
Itoh, H., 1994. Cooperation in hierarchical organizations: an incentive
perspective. Journal of Law, Economics, and Organization 60,
410427.
Laffont, J.-J., Martimort, D., 1998. Collusion and delegation. Rand
Journal of Economics 29, 280305.
Laffont, J.-J., Martimort, D., 2002. The Theory of Incentives: The
Principal-Agent Model. Princeton University Press.
Laffont, J.-J., Tirole, J., 1993. A Theory of Incentives in Regulation
and Procurement. MIT Press, Cambridge.
Leland, H., Pyle, D., 1977. Informational asymmetries, financial
structure and financial intermediation. Journal of Finance 32,
371387.

411

Macho-Stadler, I., Perez-Castrillo, D., 1993. Moral hazard with several

agents: the gains from cooperation. International Journal of
Industrial Organization 11, 73100.
Mc Afee, P., Mc Millan, J., 1995. Organizational diseconomies of
scope. Journal of Economics and Management Strategy 4,
399426.
Mookherjee, D., Tsumagari, M., 2004. The organization of supplier
networks: effects of delegation and intermediation. Econometrica
72, 11791219.
Ramakrishnan, R., Thakor, A., 1991. Cooperation versus competition
in agency. Journal of Law Economics and Organization 7,
248283.
Schmitz, P., 2005. Allocating control in agency problems with limited
liability and sequential hidden actions. CEPR Discussion Papers,
vol. 5145.
Schneider, A.L., 2000. Publicprivate partnerships in the U.S. prison
system. In: Vaillancourt Rosenau, P. (Ed.), PublicPrivate Policy
Partnerships. MIT Press, Cambridge Mass.
Tirole, J., 1986. Hierarchies and bureaucracies: on the role of collusion
in organizations. Journal of Law, Economics, and Organization 2,
181214.
Varian, H., 1989. Monitoring agents with other agents. Working Paper,
vol. 8918. U. Michigan, Center for Research on Economic and
Social Theory.
Wilson, R., 1968. The theory of syndicates. Econometrica 36,
119132.