Optimal Team Size and Monitoring in Organizations

∗
Optimal Team Size and Monitoring in Organizations

Pierre Jinghong Liang Madhav V. Rajan
Tepper School of Business Graduate School of Business
Carnegie Mellon University Stanford University
Tech and Frew Streets 518 Memorial Way
Pittsburgh, PA 15213-3890 Stanford, CA 94305-5015
Korok Ray
Graduate School of Business
University of Chicago
5807 S. Woodlawn
Chicago, IL 60637
Revised: March 2006
Abstract
We formulate and analyze a general model of team structure and monitoring. We in-
corporate three broad instruments in the internal design of an organization involving team
production: team size, monitoring activities, and incentive contracts. When workers and
managers are self-interested and subject to moral hazard, there are complex interactions
in the trade-offs among these instruments. We generate closed-form solutions to the firm’s
optimization problem and show that these interactions lead to surprisingly simple impli-
cations. One such result is that worker contracts are robust - the equilibrium level of
pay-for-performance is invariant to most environmental variables of interest. Another is
that the cost of managerial free-riding renders it almost never worthwhile to employ multi-
ple managers to supervise a given set of workers. Our work also demonstrates the presence
of complementarities between team size and monitoring, and between worker talent and
managerial monitoring ability. Finally, we derive unambiguous predictions about the im-
pact of environmental variables on the choice of optimal team size and structure, even in
the presence of an external marketplace for talent.
∗
We would like to thank Jeremy Bertomeu, Jon Glover, Carolyn Levine, Brian Mittendorf, Lin Nan, Richard
Saouma, Haresh Sapra, Jason Schloetzer, Xiaoyan Wen, and seminar participants at Carnegie Mellon, Chicago,
and the ”Issues in Managerial Accounting” conference at Texas-Austin for helpful comments and suggestions.
1 Introduction
Teams play a pivotal role in modern organizations. The study of teams has occupied a similarly
important place in modern economic thought, beginning with the work of Marschak and Radner
(1972). The primary focus of inquiry has been the provision of incentives to motivate team
members in order to overcome concerns regarding “free-riding.” Virtually all of this work,
following the lead of Holmstrom (1982), takes team structure and size as given. Yet, the broader
elements of organizational design also form a key component of modern firms (see Roberts
(2004)). Management consultants and the business press increasingly exhort executives to
view organizational design as yet another choice variable. The academic press has only recently
begun to explore such notions (Roberts (2004), Simons (2005)), and theoretical research on
the topic is still nascent (see Harris and Raviv (2002) and the references therein).
In this paper, we provide a general model of team formation and monitoring that enables
us to shed light on optimal incentives, team size, and the efficient matching of worker and
managerial abilities. We consider three broad instruments in the internal design of an organi-
zation involving team production: team size, monitoring activities, and the incentive contracts
offered to workers and managers. The complex interactions among the three instruments gen-
erate new insights into the optimal provision of incentives. One such insight is highlighted by
an invariance result. We find that when the error in the measurement of employee contribution
to firm value satisfies certain regularity conditions, incentives are optimally structured to be
invariant to noise, risk preferences and the cost of worker effort. Our invariance result offers
a novel explanation for the empirical “missing link” between risk and incentives (see, e.g.,
Bhattacharyya and Lafontaine (1995) or Prendergast (2002)).
A central feature of modern organizations is the careful design and the implementation of
internal and external monitoring activities. These activities influence (and are influenced by)
other organizational design choices, such as team size and incentive schemes. In this study, we
examine the optimal number of risk-averse workers and the nature of their incentives when the
principal hires a risk-averse manager to watch over the workers. Much of the previous literature
on monitoring has assumed that monitoring is either exogenous or is done by the principal
herself. In this paper, we deviate from this (unrealistic) assumption, and generate closed-
form solutions to the incentive problems among the principal, the self-interested manager, and
the workers being monitored. We identify conditions under which it is optimal to employ
a manager, and demonstrate the existence of endogenous complementarities across the skill
levels of the manager and workers. We also analyze the value of adding an additional manager
as a function of the planned division of labor between the multiple managers. We show that
the introduction of another self-interested manager induces additional agency costs (due to
1
free-riding on monitoring activities). These costs are overcome only if teams can be split into
sub-groups and there are significant monitoring synergies between managers; in all other cases,
there is no value to expanding the management team.
Some authors have suggested that monitoring, or the inherent loss of control from hiring
monitors, ultimately limits the size of the firm. One of the first papers to suggest such a ratio-
nale for finite firm size was Williamson (1967). Calvo and Wellisz (1978) formalize Williamson’s
argument and demonstrate that Williamson’s results require particular ad hoc assumptions on
the firm’s monitoring technology. For example, if agents are unaware of when they are being
monitored, then Calvo and Wellisz find that there is no hierarchical limit to firm size. In our
model, the active monitoring reduces the variance of the group performance measure. As such,
our managers can be interpreted as auditors whose job is to reduce measurement errors. The
reduction in noise allows the principal to offer stronger incentives to the workers without the
burden of paying additional risk premia. In our model, there is no loss of control with the
delegation of monitoring to a third party, as our entire model is cast in a moral hazard setting.
Because our monitors are separate from the principal, our framework differs from the
traditional role assigned to monitors in the accounting literature (see, e.g., Baiman and Demski
(1980)). More recently, the economics literature has expanded to allow for the monitor to be a
separate agent. Varian (1990) examines the possibility of having the working agents monitor
one another. In his paper, Varian shows that it may be optimal to accept only groups of
agents, because the agents will only select “good types” for their own group. In section 3, we
allow for heterogenous types of workers and managers and find that as risk increases, rather
than selecting a better monitor, the principal will want to decrease the team size, and hire a
lower quality monitor.
We utilize a multi-agent LEN framework to analyze issues related to optimal team in-
centives. Perhaps the earliest paper to employ this approach was McLaughlin (1994), which
looked at the determinants of the links between individual compensation and team perfor-
mance. Recent work has extended the analysis of team incentives to incorporate elements
related to optimal task assignments (Hughes et al. (2005)) and the role of productive synergies
in teams (Autrey (2005)). Hughes et al. (2005) find sufficient conditions for an owner to pre-
fer diverse task assignments as opposed to specific assignments. Autrey (2005) demonstrates
that an owner favors the measurement of individual output when there is little synergy across
workers’ efforts, but prefers to compensate employees based on team output once the level of
synergy is sufficiently high. None of the above papers incorporates managerial monitoring or
allows for the endogenous determination of team size and composition.
Qian (1994) is one of the few studies that allows for an endogenous number of workers.
Following Williamson (1967) and Calvo and Wellisz (1978), Qian examines the optimal size
2
of a hierarchy and compensation throughout all tiers of the hierarchy. In Qian’s paper, the
principal benefits from having fewer tiers in the hierarchy because it is assumed that there are
production losses at each tier, and in general, fewer tiers implies fewer managers to compensate.
On the other hand, if the firm has fewer tiers, then each manager’s span of control is enlarged,
and hence monitoring becomes less effective; the principal is thus forced to pay an higher
efficiency wage to ensure that the managers put forth costly effort. In our paper, we assume
that the managers serve to reduce the noise in measuring workers’ effort, and hence we only
consider a two-tier hierarchy: principal at the top, and workers accompanied by managers at
the bottom. In Qian’s model, the agents are assumed to be risk-neutral with limited liability
and are compensated with two-tier wages. Our model is different, because we do not allow the
principal to select the number of tiers in her organization; however, we assume that all workers
and managers are risk averse, and are compensated via the optimal linear contract.
From the standpoint of the accounting literature, our work is most closely related to Ziv
(2000), who studies the optimal number of layers of a firm’s hierarchy and the number of agents
in each layer. In his model, the principal can monitor the agents herself, which eliminates
the incentive problem with monitoring. Hiring supervisors to monitor the lower-layer agents,
on the other hand, introduces such an incentive problem. Thus, Ziv (2000) concludes that
”flatter” organizations are optimal in most cases. (See Ziv (1993) and Baldenius et al. (2002)
for related work along these lines.) In our paper, we assume that the absentee principal is
not able to engage in monitoring directly but must choose between a mechanical monitoring
system and hiring one or more self-interested managers. This allows us to clearly establish the
trade-offs inherent in hiring a manager (i.e., the gains from improving the worker’s incentive
versus the incentive problem associated with the manager’s monitoring). In addition, we
are able to characterize incentive contracts for the workers and managers, and are also able
to determine, in closed-form, the optimal number of workers. In turn, we generate insights
regarding the presence of organizational complementarities when manager and worker abilities
are exogenously given, as well as in cases where the firm must employees of varying abilities
(and associated reservation wages) in a marketplace for talent.
Huddart and Liang (2005) also study some related issues, but in the context of a part-
nership. In a partnership, the sharing of profits invites all partners to shirk their productive
efforts. While monitoring activities may improve the incentive situation, they may also lead
to additional shirking because the partners also act as monitors. A key difference in our paper
is the presence of a risk-neutral principal who acts as a ”sink” or budget-breaker and thereby
eliminates the pure risk-sharing considerations inherent in a partnership setting. In addition,
we model managers as agents specialized in monitoring, while not directly involved in actual
production, thereby capturing real-world settings beyond those in professional service firms in
3
partnerships form. Our setting allows us to analyze the nature of the tradeoff among design
instruments (incentive contracts, worker size, and monitoring assignments) as well as to derive
insights on the robustness of worker incentives (e.g., the invariance result) and the value of
adding additional managers under various assumptions regarding monitoring technologies.
The remainder of the paper is organized as follows. Section 2 lays out the basic model and
derives its solution under both the mechanical and managerial monitoring settings. Section
3 derives the existence of complementarities in the selection of managerial and worker talent,
in particular when external labor markets are present. Section 4 explores the desirability of
employing multiple managers under different monitoring technologies. The last section offers
concluding comments and suggests ways in which the analysis in this paper could be extended
or generalized.
2 The Model
Consider a principal (the firm) contracting with n workers. Each worker exerts effort ei at
ci 2
cost of effort Ci (ei ) = 2 ei . The cost of effort parameter ci measures the quality/ability of the
worker: higher quality workers have a lower ci . Let N = {1, 2, ..., n} denote the set of workers.
Joint output is given by:
X
x= ei + εx .
i∈N
We assume that x is either unverifiable or is realized too late to be used in contracting. Instead,
the principal can install a mechanical monitoring system at a price (normalized to zero), which
produces a signal y. This is the only variable on which contracts can be written. We specify
that:
X
y= ei + , where ∼ N (0, G(n)σ 2 ).
i∈N
Note that σ measures the uncertainty in performance measurement.1 We assume that G(n) is
increasing in n, so that measurement becomes noisier as the team size grows. This reflects a
basic assumption on the monitoring technology, namely that the performance of larger teams
are harder to measure because of problems with coordination, communication difficulties, etc.2
1
We emphasize that in this benchmark case, the monitoring system is mechanical in the sense that the
precision of the signal y is determined by team size n and the exogenous factor σ; it does not depend on any
monitoring effort by the principal or by some manager hired by the principal.
2
The function G(n) is key to our model and encompasses a large set of possible scenarios. One natural
interpretation of G(n) is the following. Suppose y is the sum of n individual signals about the n workers
in the team. Suppose each individual signal is an unbiased signal of the worker’s effort: yi = ei + i where
i ∼ N (0, Ĝ(n)σ 2 ). Function Ĝ(n) represents the team-size effect on the variance of an individual’s signal.
Every possible case of Ĝ(n) is covered in our model by choosing a corresponding G(n). In particular, if all i
4
The firm offers a linear contract on y to each worker, so each worker receives a wage
wi = ai + bi y. All n workers are equally risk averse with exponential utility and a common
coefficient of risk aversion parameter r. Thus, each worker’s preferences assume a mean-
variance representation. In certainty equivalent terms, each worker receives Ewi − 2r Var(wi ) −
Ci (ei ): expected wages less a risk premium less the cost of effort. Each worker solves
r
max ai + bi Ey − b2i G(n)σ 2 − Ci (ei ),
ei 2
yielding the standard incentive constraint3 ei = bi /ci . Suppose all workers have outside options,
denoted by ū, which is normalized to zero.4 The firm will set the salaries (ai ) such that
each worker’s individual rationality constraint binds, given that all other workers select their
equilibrium effort levels.5
Assume each unit of output (x) is sold in a competitive market at an exogenous price q. The
firm maximizes the expectation of revenues less wage payments, and substitutes the binding
(IR) constraints into its optimization. So the firm selects bonus coefficients (incentives) to
maximize total surplus:
Xh r i
max q ei − Ci (ei ) − b2i G(n)σ 2 .
bi 2
i∈N
This objective function reveals the two costs of production which the principal must pay
the workers. The first is the personal cost of providing effort: C(ei ). The second is the
r 2 2
risk-premium: 2 bi G(n)σ . This is the incentive cost, which is main focus of our study. After
inserting the worker’s incentive constraint, the first-order condition gives the optimal incentives
for an exogenous n:
q
b∗i (n) = , (1)
1 + rci G(n)σ 2
yielding the standard risk/incentive tradeoffs. Observe that a larger team leads to a higher
variance of the team performance measure, which leads to the firm cutting back on incentives.
b∗i
Also note that the equilibrium effort from each worker is ci , which is lower than the first-best
q
effort level of ci .
are independent, then G(n) = nĜ(n).

3
Notice that for each worker, the effort choice is also a dominant strategy response in the worker’s subgame.
4
We relax this assumption in Section 3 of the paper.
5
The outside option is zero in monetary terms but −e−rū = −e−r0 = −1 in utility terms. Each worker’s
salary is ai = −bi Ey + Ci (ei ) + r2 b2i G(n)σ 2 .
5
2.1 Invariant Incentives under Optimal Team Size
To explore optimal firm size, we consider a special case where all workers are identical, so
ci = c for all i ∈ N . Therefore, for a given n, all workers face the same incentives given by
q
b∗ (n) = . (2)
1 + rcG(n)σ 2
The firm’s equilibrium profit for exogenous n, denoted by Π̂(n), is

∗
qb (n) b∗ (n)2 r ∗ 2 nqb∗ (n)

2
Π̂(n) = n − − b (n) G(n)σ = . (3)
c 2c 2 2c
Observe that increasing the team size (n) increases the revenue from production, but also
increases the risk premium the firm must pay to each worker. The optimal n trades off revenue
from production against the cost of risk premia. Assuming G(n) = n2 , taking the first-order
condition with respect to n gives the closed form solution for the optimal team size, denoted
by n̂, as r
1
.
n̂ = (4)
rcσ 2
As uncertainty increases, the optimal team size shrinks. More precisely, an increase in
uncertainty raises the risk premium for each worker. The firm reacts by shrinking the team
size. Plugging the optimal team size (4) into the optimal incentives (2) gives the optimal
incentives at the optimal team size:
q
b∗ (n̂) = . (5)
2
In effect, the incentives exhibit an invariance property: the optimal incentives do not vary
with σ, r, or c. The firm responds to changes in uncertainty (σ) by shrinking the team size and
keeping the incentives fixed. In contrast to the standard agency model, the firm here has two
instruments: team size (n) and incentives (bi ). The inclusion of this additional instrument of
team size allows the firm to more effectively adjust to changes in the environment.
While the invariance property of the incentives may appear to be a consequence of the
special choice G(n) = n2 , it, in fact, holds more generally. To see this, suppose that G(n) is a
degree function: G(n) = Anγ for arbitrary parameters A > 0 and γ > 1. The parameters A, γ
measure the difficulty of team performance measurement.6
We now generalize the invariance result to any G(n) that is a degree function.
6
In particular, observe that as either A or γ increase, the variance of y rises. Moreover,
∂ Var(y)
= γAnγ−1 σ 2 .
∂n
In other words, the marginal effect of larger teams on performance measurement increases in γ and A. These
two variables are essentially the parameters of the performance measurement process.
6
Proposition 1 Suppose that G is differentiable and invertible. Then the optimal b does not
depend on parameter values (σ 2 , r, c) if and only if G is a degree function.7
To understand the intuition behind this result, suppose first that n is exogenous, and σ
increases. From equation (2), it is intuitive that b∗ may decrease, since the firm cuts incentives
as the risk premium rises. As the incentives fall, so do the profits, as shown in equation (3).
Now if the firm decreases the team size, a smaller n reduces the risk premium and hence lifts
the incentives, as apparent in equation (2). This rise in incentives offsets the fall in profits
from the smaller team size. In particular, the firm will decrease the team size exactly to offset
the increase in uncertainty, leaving the incentives constant. Essentially, team size is a more
flexible instrument than incentives.
Before ending our discussion of the benchmark model, we comment on the robustness of
the above finding. Two aspects of our formulation may invite special examination - the use
of a single group measure to compensate each worker, and the specification of an additively
separable production function, with no apparent reason for the existence of a team. We show, in
Appendix II, that versions of the invariance result can be derived even when these assumptions
are altered. In particular, the result continues to hold if each worker is compensated using his
own individual performance signal, as opposed to the common group measure y, and also when
the production technology exhibits complementarities among workers, as opposed to the linear
technology with uniform marginal productivity (q) in the benchmark model. More generally,
we also demonstrate that for a much larger set of (non-degree) G(n) functions, the ability to
vary the optimal team size results in an attenuation of the efficient level of incentives.
Returning to the benchmark model, we can incorporate our findings regarding the optimal
team size (4) and incentives (5) into the profit function (3) to compute the optimal level of
profit:
q2
Π̂(n̂) = √ .
4 rc3 σ 2
We end this section by collecting the main calculations thus far.
Lemma 1 Assume ci = c for all i ∈ N , and the principal uses the mechanical monitoring
technology. Then
(1) if firm size n is exogenous, we have
q nqb∗ (n)
b∗ (n) = Π̂(n) = ;
1 + rcG(n)σ 2 2c
(2) if G(n) = n2 and firm size n is endogenous, we have
q 1 n̂qb∗ (n̂)
b∗ (n̂) = n̂ = √ Π̂(n̂) = .
2 crσ 2 2c
7
Proofs appear in Appendix I.
7
2.2 Adding a Manager
Now we introduce a manager to the benchmark setting. The manager serves the role of
actively and endogenously supplying the monitoring activities (as opposed to a mechanical
monitoring system). Endogenous monitoring alters the economic tradeoff on team size and
worker incentives (and thus welfare) but we show that the invariance result continues to hold
when team size is endogenously chosen.
We model the manager as an effort- and risk-averse economic agent who can produce a
signal about the collective efforts of the team under his supervision. In particular, the manager
exerts monitoring effort m at personal cost C(m) = km, where k is the manager’s cost of
effort parameter. This parameter k measures the manager’s quality/ability: higher quality (or
lower cost) managers have a lower k. The manager’s effort reduces the variance on the team
performance measure.8 Specifically, we assume
G(n)σ 2
X
y= ei + where ∼ N 0, .
m
i∈N
The more the effort the manager exerts, the less noisy the performance measure y:
X G(n)σ 2
Ey = ei and Var(y) = .
m
i∈N
Observe that the task-specialization is such that the manager’s effort reduces the variance of y
while the worker’s effort increases its mean.9 Assume m is unobservable so the manager must
be motivated to supply monitoring efforts. The firm offers a linear contract to the manager
consisting of a salary and a bonus on the team performance measure,10 i.e., ω = α + βy. The
manager thus solves:
r G(n)σ 2
max α + βEy − β 2 − C(m).
m 2 m
Even though all workers and the manager are under linear contracts, the provision of incentives
to exert efforts come through in different ways. The manager’s effort does not affect the
8
By this modelling choice, we choose to emphasize the monitoring role of the manager, leaving their other
functions (such as fulfilling some direct productive tasks) as second-order effects. So, the term manager is
interchangeable with supervisor.
9
If the manager chooses to exert one unit of monitoring effort, i.e., m = 1, the signal variance is identical to
that of the mechanical monitoring system.
10
We believe this is a simple and natural way to model the compensation arrangement between the principal
and the manager. In practice, managers’ compensation is, in large part, determined by the performance of the
group under their supervision; see, for example, Bushman et al. (1995). Providing managerial incentives via
the signal produced by the manager represents a sharp contrast to the model in Ziv (2000), where the firm
provides incentives to managers via a signal about managerial effort that is generated by yet another manager
or, ultimately, the principal. While Ziv’s model potentially allows for the investigation of multi-layer hierarchies,
the imposed complexity is daunting and precludes development of a tractable framework.
8
expectation of his wage Eω, but such effort does decrease its variance. Given the manager
is risk-averse, he has an incentive to reduce the variance. The first-order condition yields the
incentive constraint for the manager
r
∗ G(n)r
m = βσ . (6)
2k
Notice that the manager’s problem is concave even when his personal cost C(m) is linear. This
2
is because his personal benefit of monitoring effort (- 2r β 2 G(n)σ
m ) is increasing and concave in
m.11
From (6), we learn that, for a given incentive coefficient β, if either the measurement
process worsens (σ 2 increases) or as the managers become more risk averse (r increases), the
risk premium grows and hence the manager works harder at monitoring. Moreover, as the
manager’s span of control grows (n increases) the manager also works harder. Adding workers
boosts the variance of y and hence increases the risk premium. Because the manager dislikes
variance, he spends more time monitoring to compensate for the additional risk cost. However,
the manager’s incentive coefficient β is endogenous, which may also depend on parameters σ
and r. We return to address the overall effect on equilibrium monitoring efforts later.
As before, the firm sets the salary levels (ai , α) such that the individual rationality con-
straint binds for every worker and the manager. Assume the opportunity wage for the
manager is ūM , which is also normalized to zero. The manager’s salary then satisfies α =
2
−β j ej + C(m) + 2r β 2 G(n)σ
P
m . The firm therefore maximizes total surplus, as before.
Xh r i r
max qei − Ci (ei ) − Var(wi ) − C(m) − Var(ω),
bi ,β 2 2
i∈N
where
G(n)σ 2 G(n)σ 2
Var(wi ) = b2i and Var(ω) = β 2 .
m m
Substituting in each worker’s incentive constraint gives
X qbi b2 r G(n)σ 2 r 2 G(n)σ 2

∗
max − i − b2i − km − β , (7)
β,bi ci 2ci 2 m∗ 2 m∗
i∈n
where m∗ is given by (6). The first-order conditions yield the optimal incentives for the worker:
q
b∗i (n) = rci G(n)σ 2
.
1+ m∗
Similar to the earlier setting (with a mechanical monitor), an increase in risk aversion (r),
cost of effort (ci ), or noise (σ 2 ) causes the firm to cut the worker’s incentives, delivering the
11
A benefit of linear personal cost is that it provides tractability. Other functional forms of C(m) provide
similar economic intuitions but much messier algebra.
9
standard trade-off. What is different now is that the manager’s effort (m) matters in choosing
optimal incentives for workers. Of course, the firm can affect m by choosing the manager
incentive β carefully.
Optimizing the firm’s objective function with respect to β results in a first-order condition,
from which we can see the principal’s tradeoff in choosing manager’s incentive:
n
" #
r G(n)σ 2 X 2 r G(n)σ 2 2 ∂ ∗ G(n)σ 2
b i − k + β m − rβ = 0. (8)
2 m∗2 2 m∗2 ∂β m∗
i=1
Providing more managerial incentive (i.e., increasing β) induces more managerial efforts (see
equation 6). The benefit is two-fold. First, it reduces the variance of the performance measure y
and (thus) the risk-premium needed for the manager because the manager’s contract is written
on y. Second, increasing β also reduces the needed risk-premium of all workers because the
worker contracts are also written on y. Notice here that the firm exploits a spill-over effect
from the managerial monitoring effort. The manager is self-interested and desires to exert
efforts to reduce the variation in his own compensation. But in doing so, he reduces the
wage variations of all workers. This is beneficial to the firm because it reduces salary costs
(a∗i and α∗ ). These two (marginal) benefits are shown as the two positive terms in (8). The
cost of providing managerial incentive is also two-fold. First, more incentives leads to more
personally costly managerial effort, for which the manager requires compensation. Second,
more incentives leads to a managerial pay more sensitive to the signal variations, which leads
to higher costly risk-premium. These two (marginal) costs are shown as the two negative terms
in (8).12 q
∂ ∗ G(n)r
Substituting in the manager’s incentive constraint (6) and (thus) ∂βj mj =σ 2k into
(8), and rearranging terms yields s
1X 2
β= bi .
2
i∈N
The manager’s incentives thus increase with each worker’s incentives. Intuitively, as the firm
increases bi , it induces more effort from the workers but also increases the risk premium. To
compensate, the firm increases β which induces the manager to spend more effort monitoring
the workers; in turn, this reduces the risk premium for everyone.
12
Consider a slight alteration of the model where the manager may exert a productive effort, say em , which,
similar to workers’ efforts ei , increases the output and the aggregate performance measure y. When choosing
the optimal β, the principal now must consider the marginal benefit and cost of inducing em , in addition to the
four effects consider in equation (8). However, so long as these additional marginal effects are well-behaved (i.e.,
smooth and bounded), their only impact is to make the optimal β quantitatively different. The fundamental
tradeoff in the managerial monitoring activities, which is the focus of our paper, persists and leads to the same
qualitative results.
10
Let B = (bi ) ∈ Rn be the vector of worker incentives. Note that
1
β = √ kBk
2
where k · k is the Euclidean norm operator. With identical agents, every vector B ∈ R2 lies
on the 45-degree line. The projection of B onto the x and y axes generates a right isosceles
triangle with B as the hypotenuse. By the Pythagorean Theorem, the length of the equal sides
√
of this isosceles triangle is kBk/ 2 ≡ β. In a precise mathematical sense, the strength of the
manager’s incentive thus reflects the strength of the workers’ incentives.
Combining the previous expressions for bi , β, and the manager’s incentive constraint yields
the following expression for b∗i as an implicit function of the model parameters:
q
b∗i (n) = . (9)
PrG(n)σb k
r
2
1 + 2ci ∗2
j∈N j
Despite the fact that equation (9) is an implicit function of bi , we are able to prove the following
intuitive comparative statics result.
Proposition 2 Incentives (bi ) fall as the cost of effort (ci ) increases.
2.3 Optimal Incentives and Team Size with a Manager
Assuming identical workers (ci = c) leads to b∗i = b∗ , which in turn allows us to solve (9) and
derive a closed-form solution for the equilibrium worker incentive (b∗ ) for a given n:
b∗ (n) = q − 2cσ
p
rkG(n)/n.
Recall that the equilibrium manager incentive (β ∗ ), monitoring effort (m∗ ), and worker effort
(e∗ ) are all functions of b∗ .13 From (7), we can now write the profit function with a manager
for a given n, denoted by Π(n), as follows:
  r
∗ ∗ 2 2 ∗σ
qb b r ∗2 G(n)σ b rnG(n)
Π(n) = n  − − b q  − 2k
c 2c 2 ∗
b σ rnG(n) 2 k
2 k
nb∗ 2
= . (10)
2c
Now, suppose that G(n) = n2 . Taking the first-order condition with respect to n and simplify-
ing yields a closed form solution for the optimal team size with managerial monitoring, which
13
The maintained assumption here is that b∗ > 0. We rule out combinations of parameters such that the
above equation gives a negative b∗ , because that implies equilibrium worker incentives would be set to zero,
effectively shutting down production.
11
we denote by n∗ :
q2
n∗ = .
16c2 σ 2 rk
As either c, k, r, σ increase, this inflates the risk premium. The firm responds by choosing
a smaller team, and thus deflating the risk premium somewhat. In addition, as the quality of
the manager improves (k shrinks), the optimal team size grows. Better managers are better
able to measure performance, and can thus handle larger teams.
The optimal incentives and labor supply of the worker are:
q q
b∗ (n∗ ) = and e∗ = . (11)
2 2c
The equilibrium incentives for the workers are exactly the same as in the benchmark case
without a manager. In particular, the invariance property continues to hold: the workers’
incentives are independent of r, c, σ, and here, even k. The logic is the same as illustrated
earlier: the firm takes advantage of the presence of an additional instrument (team size) to
adjust to changes in the risk premium, and it employs this instrument optimally while keeping
incentives fixed. Similarly, invariance holds if and only if G(n) is a degree function:
Proposition 3 The invariance property holds in the setting with a manager if and only if
G(n) is a degree function.
It is straightforward to show that this invariance result continues to hold for the case
when each worker is compensated using his own individual performance signal and when the
production technology exhibits complementarities among workers (see Appendix II for details).
The findings in Propositions 1 and 3 provide a possible explanation for the lack of strong
empirical support for the negative relation between risk and incentives suggested by traditional
agency models (see Prendergast (2002)). Prendergast suggests that firms may contract directly
on employee effort in some settings, and contract on output in less certain settings. Under
this construct, the strength of incentives given to employees need not vary monotonically
with the level of uncertainty. Our model differs in that we assume that employee effort is
unobservable and that contracts with employees can only be structured on observed joint
output. In effect, while the Prendergast result points to the omission of delegation as a choice
variable in traditional agency models, our finding is based on the ability of the principal to
control team size in addition to employee incentives.
The manager’s incentives and monitoring effort (in equilibrium) are given by
q2 q4
β∗ = √ and m∗ = . (12)
8cσ 2rk 256c3 σ 2 rk 2
Observe that the invariance property does not hold for the manager’s incentives. An increase
in any of the parameters that increase the risk premium (an increase in r, c, k, σ) causes the
12
firm to decrease incentives for the manager, and thus the manager works less. This seems
counterintuitive, as one might expect the firm to raise incentives for the manager to monitor the
workers more precisely if the measurement process worsens (σ rises). Instead, the firm shrinks
the size of the team to respond to the increase in risk premium. Smaller teams require less
monitoring, and hence the firm can afford to reduce the incentives for the manager. The firm
is once again using team size as the main instrument to respond to exogenous changes in the
environment. Managerial incentives, however, do vary with the parameters of the model. Our
model thus yields the testable prediction that we should empirically observe greater variation
in the incentives for supervisors than those for production workers.
Substituting n∗ into the profit function, we have the following form for the principal’s profit
at the optimal n:
n∗ b∗ 2 q4
Π(n∗ ) = = .
2c 128c3 σ 2 rk
Lemma 2 Assume ci = c for all i ∈ N and the principal hires a manager with ability k. Then
n[b∗ (n)]2
r
∗
p ∗ ∗ n
b (n) = q − 2cσ rkG(n)/n β = b (n) Π(n) = ;
2 2c
(2) if G(n) = n2 , and firm size n is endogenous, we have
q q2 q2 n∗ (b∗ (n∗ ))2

b∗ (n∗ ) = n∗ = β∗ = √ Π(n∗ ) = .
2 16c2 σ 2 rk 8cσ 2rk 2c
2.4 Value of the Manager
When is it profitable for the firm to hire a manager? Suppose the firm size is exogenously
given. Then for any given set of parameters (q, r, c, σ 2 , k), the firm compares the expected
profit under the mechanical monitor with the expected profit under a hired manager. From
Lemmas 1 and 2, the value of hiring a manager is:

n p 2 q
Π(n) − Π̂(n) = q − 2cσ rkG(n)/n − q .
2c 1 + rcG(n)σ 2
If the firm size is endogenous, then for any given set of parameters (q, r, c, σ 2 , k), the firm
compares the expected profit under the mechanical monitor at the optimal team size and the
expected profit under a hired manager at the (possibly different) optimal team size. Assuming
G(n) = n2 , the value of hiring a manager is
q4 q2
Π(n∗ ) − Π̂(n̂) = 3 2
− √ .
128c σ rk 4 rc3 σ 2
13
For the case where team size is endogenous, further manipulations show that
n∗ (r, c, σ, k) q2
= √ .
n̂(r, c, σ) 16k rc3 σ 2
Based on these calculations, we obtain results on the value of hiring a manager, which we
summarize in the following proposition.
Proposition 4 Assume ci = c for all i ∈ N and the principal chooses between using a me-
chanical monitor and hiring a manager with ability k. Then
q √
1+rcG(n)σ 2
> 4 k =⇒ Π(n) > Π̂(n);
√ 2
rG(n)nσ
(2) if G(n) = n2 , and firm size n is endogenous, we have

q2
√ > 1 ⇐⇒ Π(n∗ ) > Π̂(n̂) =⇒ n∗ (r, c, σ, k) > n̂(r, c, σ). (13)
3
32k rc σ 2
For both cases (exogenous or endogenous n), a manager is valuable if the quality of the
worker or the manager is sufficiently high (k, c sufficiently low). Moreover, a manager is
valuable if the uncertainty in monitoring technology is sufficiently low (σ 2 sufficiently small).
Intuitively, a low k or σ reflects fewer difficulties in motivating an effort- and risk-averse
manager to work. This results in lower incentive coefficients and, consequently, a lower risk
premium. Therefore, it is profitable for the firm to hire a manager because the manager’s work
has a spill-over effect which reduces the risk premium of workers in the team as well. For the
case where team size is endogenous, if it is profitable to hire a manager, the optimal team size
with a manager exceeds the optimal team size without the manager. Managers assist in team
performance measurement, and hence allow the formation of larger teams. That is, a valuable
manager leads to larger team sizes, which points to a synergy between team size and the use
of managers. What is interesting here is that the synergy (or complementarity) is not assumed
of the production technology, but is due to the inherent moral hazard problem associated with
team production.
To summarize, in this model, the firm has three design instruments to maximize its ex-
pected profit. They are (1) number of workers to hire (n), (2) the choice of whether to hire a
manager, and (3) the choices of incentive coefficients for the workers and the manager (b and
β). When the environment changes (q, r, σ, c, or k), all three design variables may change
in response. The special property of the responses, in our example of identical workers, is a
certain robustness in the worker-incentives (i.e., b∗ does not respond to any changes except a
change in q). For any changes in the environment other than q, the firm reacts by changing
team size or monitoring and leaves worker-incentives intact.
14
3 Complementarity between Managers and Agents
In the previous section, we demonstrated the presence of endogenous complementarities be-
tween team size and managerial monitoring. In this section, we explore more such aspects
of optimal incentives and monitoring in team settings. Of specific interest to us is the link
between worker and manager quality. Recall that the parameters c and k in our model reflect
the costs of effort for the agent and the manager (better agents and managers have lower c’s
and k’s). This section explores complementarity, in the sense of optimal matching, between
managers and agents. That is, should better managers be asked to supervise better workers
or worse ones? We show that it is always optimal to do the former, i.e., manager and worker
abilities are complements. We then explore how the selection of the optimal manager and
worker combination in a labor market varies with uncertainty and other parameters of our
model.
3.1 Positive Sorting
Recall that the profit function at the optimal team size is
q4
Π(n∗ ) = . (14)
128c3 σ 2 rk
Taking derivatives of the profit function (14) gives
∂Π 3q 4 ∂Π q4
=− <0 and =− < 0. (15)
∂c 128c4 σ 2 rk ∂k 128c3 σ 2 rk 2
So (15) shows that better agents and managers raise profits. What drives this result? It is
tempting to argue that because it is easier for better agents to deliver a given level of effort,
the firm will reduce their incentives, thus reducing the risk premium payments and hence
raising profits. Similarly, it seems plausible that because better managers are more effective
at reducing variance, this will also lead to lower risk premiums and higher profits. However,
this logic neglects the main point of this paper: that the firm is simultaneously adjusting
the team size, and this can reverse the usual comparative statics. To see this, note that in
equilibrium m∗ = n2 σ 2 cr. A better manager allows the firm to increase team size, which causes
the manager to work harder in equilibrium. Recall that
G(n)σ 2 G(n)σ 2
Var(wi ) = b2i and Var(ω) = β 2 , (16)
m m
so more managerial effort will pull wage variance downward. But the larger team size coun-
teracts this effect: in equilibrium, m∗ is of order n2 , so if G(n) grows at least as quickly as n2 ,
15
the net effect is that wage variance will increase. Indeed, for G(n) = n2 , combining equations
(16), (11), and (12) yields the equilibrium variance of the agent’s and manager’s wage
q2 q4
Var(wi ) = and Var(ω) = .
4cr 128c3 σ 2 r2 k
Better agents and managers (lower c’s and k’s) in fact increase wage variance. In particular,
observe that the risk premium as well increases with better agents and managers:
nr r 3q 4
RP = Var(w) + Var(ω) = .
2 2 256c3 σ 2 rk
But if the risk payments are increasing, how do profits rise? Remember that larger teams not
only inflate the risk premium, but also earn more production revenue for the firm. The firm
maximizes total surplus, which (for identical agents) is

T S = n qe − C(e) − C(m) + RP .
Suppose the firm hires a better manager. This does not change the agent’s effort choice,
so the term in braces is unchanged. As shown above, the manager works harder and the risk
premium rises, so the term in brackets rises, pulling down profits. Finally, the firm hires a
larger team, so the revenue from production lifts profits upward. The net effect is that the
benefits from higher production exceed the costs of more managerial and more risk payments,
so profits overall rise. A similar argument holds for better agents.
Observe that
∂2Π 3q 4
Πck ≡ = > 0, (17)
∂c∂k 128c4 σ 2 rk 2
which gives positive sorting: it is optimal to assign better managers to better workers. Precisely,
∂Π ∂Π ∂Π
the condition above shows that ∂c increases in k and ∂k increases in c. Now ∂c is the marginal
∂Π
return from hiring a better manager.14 So as the quality of the manager improves (k falls), ∂c
falls, so the marginal benefit from hiring a better agent rises. Said differently, better agents
raise profits, and in fact raise profits more under better managers.
====================
Insert Figure 1 Here
====================
To see this visually, refer to Figure 1. The two lines represent profit as a function of k for
different costs of worker effort. Observe that both lines are decreasing: as the quality of the
14
Actually it’s the marginal loss of hiring a worse manager, since worse managers have higher k’s. This is
equivalent to the marginal return of a better manager.
16
manager improves (k decreases), the firm earns more profit. Equation (17) says that Πk rises
in c. So as c2 rises to c1 , Πk (c2 , k) will rise to Πk (c1 , k). Since Πk (c2 ) < 0, this means Π(c1 , k)
is less negative, and hence flatter, than Π(c2 , k). Said differently, as k1 falls to k2 , profits will
rise. But the increase in profit under the better agent (∆2 ) exceeds the increase in profit under
the worse agent (∆1 ). So as the firm hires a better manager, the firm earns more profit by
hiring a better worker as well.
This shows that good managers and good agents are complements. The intuition once
again lies in the optimal team size. From (11), it is easy to see that the cross partial of n∗
with respect to c and k is positive. With better agents and better managers, the firm can
afford to increase team size; team size increases the most with good agents working under
good managers. A good manager can more easily reduce the variance for a larger team, and
even more so if the team consists of good workers. It is this complementarity between manager
and agent quality at the team level that translates to complementarity at the profit level.
3.2 Optimal Manager and Agent Quality
The previous section shows that the firm earns more profits by hiring better agents and better
managers. This begs the question: why not hire the best agent and monitor possible? The
reason, of course, is that employing higher quality talent is more expensive. We have, as
does virtually all of the compensation/agency literature, thus far assumed a partial equilib-
rium setting where the labor market is captured only via the assumed (exogenous) reservation
wage. In this section, we break from this paradigm and explicitly model a labor market for
talent.15 This enables us to draw inferences about the impact of firm-specific factors, as well
as worker and manager characteristics, on the optimal team size and labor mix employed by
organizations.
To model this in a parsimonious manner, we specify that the firm hires the manager and
agents in distinct external labor markets. In particular, suppose that the quality k > 0 of the
manager and the quality c > 0 of the agent are observable by both the firm and the market.
A manager of quality k receives a market wage ūM (k). Better managers (lower k) earn higher
market wages, so ūM (k) is a positive and strictly decreasing function of k. Similarly, each
identical agent receives a market wage of ū(c), a positive and decreasing function of c.
It is clear that the firm must, and will, pay exactly these market wages to acquire its labor.
As such, the firm sets the salaries (a,α) such that the agents’ and manager’s expected utilities
precisely equal their market wages. Given binding (IR) constraints, the firm’s profit function
15
Other papers that have modelled an outside labor market are Ricart I Costa (1988) and Sappington (1983).
17
for exogenous n is then given by:
nb2
Π(n, c, k|σ, r) = − nū(c) − ūM (k),
2c
q
rkG(n)
where b = b∗ = q − 2cσ n is the optimal incentive payment for each agent. Hiring a
better quality manager (decreasing k) allows the firm to increase the size of the team, and
hence generate more revenue from production. At the same time, better managers are more
expensive, since ūM (k) increases as k decreases. The optimal k trades off the increased revenue
from larger teams against the increased cost of a better quality manager, where this cost reflects
the manager’s market wage. A similar trade off holds for agent quality.
The firm now chooses the optimal agent and manager in addition to optimal team size:
max Π (n, c, k|σ, r) such that n, c, k > 0, and b ≥ 0. (18)

n, c, k
√
Consider the special case where G(n) = n2 . In this case, b = q − 2cσ rkn, and hence the
derivatives are:
∂b b−q ∂b b−q ∂b b−q
= , = , and = .
∂n 2n ∂c c ∂k 2k
The first order conditions of (18) with respect to n yields 2b2 − bq = 2cū(c). Similarly, the
first order condition with respect to c gives b2 − 2bq = 2c2 ū0 (c). From these two equations, it
is clear that at the optimum, b and c do not depend on σ or r.
We now show that this result holds much more generally. In particular, suppose G(n) is
p
a degree function. Then we can write G(n)/n = Anγ , where A, γ can be arbitrary, positive
real numbers. We have the following result:
Proposition 5 The optimal b and c for a fixed k do not depend on σ, r or k.
The invariance property thus holds within this setting as well. For exogenous changes in
uncertainty, the firm optimally adjusts the team size and the quality of the manager, but keeps
incentives for the agent and the quality of the agent fixed. So invariance applies not just to
incentives but to the quality of the agent. Including additional choice variables in the firm’s
problem allows the firm to use these instruments to adjust to changes in its environment.
Essentially the team size and manager quality are more flexible instruments in making the
adjustments, compared to the incentives and agent quality.
If the firm does not alter b or c after changes in σ, how does it change n and k? As
uncertainty increases, one might expect the firm to hire a better manager, since better managers
by definition are more effective at reducing output variance. Instead, we have the following
result:
18
Proposition 6 As uncertainty (σ) or risk aversion (r) increase, it is optimal to decrease the
team size, hire a lower quality manager, and keep the quality of the workers fixed.
Once again, it is important to consider that in equilibrium the firm adjusts the team size
optimally. As shown earlier, the firm will shrink the team in response to higher uncertainty.
Because managers are costly, it is a waste of resources to assign an expensive, high-quality
manager to a small team when a lower quality manager will suffice. Even though the expensive
manager by himself can more effectively reduce output variance than a cheaper manager, the
firm uses team size as the primary instrument to deal with the increase in uncertainty, and
then selects the quality of the manager to fit the team.
p
Because G(n)/n = Anγ , the parameter γ measures how quickly signal quality degrades
p
as team size increases. To see this, observe that Anγ = G(n)/n implies G(n) = A2 n2γ+1 . In
this case the output variance becomes
G(n) 2 A2 nσ 2
Var(y) = σ = .
m m/n2γ
Therefore, as γ increases, the manager’s effectiveness of monitoring (m/n2γ ) decreases, and

hence the output variance increases. The firm bears the cost of a high γ through higher risk
payments. So γ measures the quality of the performance technology. How will changes in γ
affect the firm’s choice of manager and agent quality? The next result provides insight into
this question.
Proposition 7 As the performance measurement technology improves (γ decreases), it is op-

timal to enlarge the team size, hire better quality managers, and hire lower quality agents.
Note that any given manager is more effective at reducing variance if γ is low. Naturally,
more effective managers can handle a larger number of agents, so the firm expands the team
size. But surprisingly, a change in γ affects the optimal quality of the agents and the manager in
opposite directions: the firm will hire lower quality agents but better quality managers. Again,
focus on the effects coming from the optimal team size. As the team size increases, the firm
hires a better manager to handle the larger team. In this sense, quality of the performance
measurement system and the quality of the manager are complementary. Better managers
are more effective at utilizing a high quality performance measurement system than less able
managers.
On top of this, the firm hires lower quality agents. This arises from the combination of
two forces: first, high-quality agents are now costly, and second, the firm is already earning
more revenue from larger teams. Therefore the firm expands the size of the team but lowers
the quality of each team member. So team size and agent quality are substitutes in the firm’s
19
profit function, whereas team size and manager quality are complements. This arises directly
from the nature of the monitoring production technology.
A remarkable feature of the previous comparative statics is that they require virtually no
restrictions on the outside option functions. It is necessary only to impose that the functions
are positive and strictly decreasing, but not necessary to specify their functional form. In
particular, even if agents are much cheaper than managers (ūc (k) ūM (k)) or if their market
wages are much more sensitive to quality (|ū0c (k)| |ū0M (k)|), the firm still keeps agent quality
fixed in response to changes in uncertainty. This shows that simply imposing a cost of agent
and manager quality is sufficient to generate rich comparative statics on the parameters of
the model. While prior empirical work on team size within firms is thin, this paper produces
several testable hypotheses to guide future empirical research in this area.
4 The Value of Management Teams

All of our modelling thus far has assumed either the presence of a mechanical monitor or a single
manager to assess the team’s performance. A natural question that arises is whether the owner
would benefit from employing multiple supervisors for a given team. Such analysis would also
lead us to consider the nature of interactions across managers, as well as help determine the
optimal span of managerial control in our setting. In this section, we expand our framework to
study these issues. We proceed in two steps. First, we examine the optimality of expanding the
size of the management team for a given team of workers. In particular, we analyze the value
of employing two managers who, either jointly or independently, supervise the entire worker
team. Subsequently, we consider the possibility of having multiple managers, where each is
given the responsibility for supervising a subset of the worker team. This second scenario is
analogous to analyzing the optimality of splitting a team into sub-groups. Taken together, our
findings in this section yield several implications for the value of employing multiple managers,
the nature of optimal managerial incentives in such settings and, finally, the optimal design of
worker teams.
4.1 Two Managers Supervising the Entire Team
To analyze the value of using a team of managers, we begin by simply adding a second manager
to the setting in Section 2. Both managers are entrusted with the role of supervising the entire
worker team. We consider two natural specifications of the interaction across the managers’
efforts. First, we assume that the managers work jointly to produce a single signal about the
efforts of the worker-team. We refer to this as Joint Monitoring (JM ). In the second scenario,
we allow for the managers to work separately and produce independent signals of the workers’
20
collective efforts. We denote this as Independent Monitoring (IM ). Below, we analyze each of
these settings in detail.
4.1.1 Joint Monitoring
We assume that the principal hires a second manager to work together with the first manager
in order to improve the precision of the group performance measure (y). The second manager
is also work- and risk-averse, with a personal cost function (k2 m) and an exponential utility
function. Letting m1 and m2 represent the monitoring efforts of manager 1 and 2, respectively,
the distribution of the realized signal is given by:
G(n)σ 2
X
y= ei + where ∼ N 0, ,
m1 + m2
i∈N
The firm offers each manager a linear contract, ωj = αj + βj y, for j = 1, 2. Each manager
maximizes the certainty equivalent of his wage, taking the other manager’s effort as given.
This yields the following pair of first-order conditions:
s s
G(n)r G(n)r
m∗1 = β1 σ − m2 and m∗2 = β2 σ − m1 . (19)
2k1 2k2
In turn, the firm chooses bi and βj to maximize total surplus:
n
b2 r G(n)σ 2 G(n)σ 2

X qbi r 2
max − i − b2i − (k1 m1 + k2 m2 ) − β1 + β22 ,
βj ,bi ci 2ci 2 m1 + m2 2 m1 + m2
i=1
yielding a first-order condition on bi analogous to that in the one-manager setting:

q
b∗i = rci G(n)σ 2
.
1+ m1 +m2
Similarly, the first-order condition on βj , for j = 1, 2, is given by:
" n
# s
r G(n)σ 2 X 2 r G(n)σ 2 G(n)r G(n)σ 2
β12 + β22 σ

∗ ∗ bi − kj + ∗ ∗ − rβj ∗ = 0. (20)
2 (m1 + m2 )2 2 (m1 + m2 )2 2kj m1 + m∗2
i=1
The difference here is that the firm must consider the behavior of both managers as opposed
to a single manager in choosing worker and manager incentives.
Assuming identical workers (ci = c for all i ∈ N ) and managers (kj = k for all j = 1, 2),
we conjecture a symmetric solution where bi ∗ = b∗ and βi∗ = β ∗ . The symmetry simplifies the
first-order conditions and helps solve for equilibrium incentive choices. In particular, we first
simplify the managers’ effort choice in equilibrium. Because m∗1 = m∗2 ,
r r
∗ ∗ G(n)r ∗ ∗ ∗ β ∗ σ G(n)r
m1 = β σ − m2 =⇒ m1 = m2 = , (21)
2k 2 2k
21
which leads to a simplification of worker incentive bi ,
q q
b∗ = rcG(n)σ 2
= √ .
1+ cσ 2krG(n)
2m∗ 1+ β∗
Substituting (21) and βi∗ = β ∗ into (20), we have

√
β ∗ = b∗ n.
Combine the last two equations and we have a closed-form for b∗ :
b∗ = q − cσ
p
2rkG(n)/n.
We can now calculate the equilibrium profit in this two-manager setting for a given team size
n, denoted by ΠJM (n), and compare this with the profit in the one-manager setting (equation
10).
Proposition 8 Assume firm size n is exogenous, ci = c for all i ∈ N , and kj = k for all
j = 1, 2. Then, assuming symmetric managerial contracts:
(1) If the principal hires two managers, we have
√ nb∗ (n) h i
b∗ (n) = q − cσ β ∗ = b∗ n
p p
2rkG(n)/n ΠJM (n) = q − 3cσ 2rkG(n)/n ;
2c
(2) The value of the second monitor is negative, i.e.,
ΠJM (n) < Π(n).
The result in (2) implies that, with joint monitoring, the efficient contracting arrangement
involves offering one manager the optimal contract in the single-manager setting, and offering
the other zero for all y. If effect, this reverts to the solution to the single-manager setting.16
Given the nature of managerial monitoring considered here, why would a firm hire a second
manager? The benefit is that a second manager may provide additional monitoring effort and
thus reduce the variance of the performance measure, which allows more incentives placed upon
workers. It is easy to see that this is indeed true (e.g., compare the optimal worker incentives
b∗ in both settings). However, there are costs of hiring another manager that, in this setting,
always outweigh the benefits.
16
Even though we conjectured a symmetric solution in presenting this result, it can be shown that if the two
managers have identical talent, k1 = k2 , there are only two equilibria in which non-trivial monitoring exists: (1)
the symmetric solution where β1 = β2 > 0; and (2) the single-manager solution where either β1 > β2 = 0 or
β2 > β1 = 0.
22
The basic reason is the hidden cost of joint monitoring production. Hiring a second manager
incurs usual costs such as the personal cost of monitoring activities and risk-premium due to
the second manager. In addition, the joint monitoring of production induces a hidden incentive
cost. That is, the monitoring technology in this two-manager case invites free-riding. Recall
in the manager’s optimization problem, each manager’s effort choice (19) reveals a shirking
incentive at the margin (i.e., manager 1’s monitoring effort is negatively related to manager
2’s effort). In other words, the precision of team performance signal is a public good and
each manager internalizes the marginal (personal) cost but not all the marginal benefit; this
leads to shirking. In turn, the manager’s incentive (β) is less effective in equilibrium. Indeed,
comparing (21) and (6) reveals that, for every unit of the manager incentive (β), the reaction
of each manager is half the monitoring intensity of the single-manager setting.
4.1.2 Independent Monitoring
A second, reasonable, specification of the multi-manager setting is that each manager works
independently to produce a separate, verifiable signal about the performance of the entire
worker team. With two managers, this results in a total of two contractible signals, denoted
by y1 and y2 respectively. The distribution of performance measures in this scenario can be
characterized as follows:
G(n)σ 2
X
y1 = ei + 1 where 1 ∼ N 0, , and
m1
i∈N
G(n)σ 2
X
y2 = ei + 2 where 2 ∼ N 0, ,
m2
i∈N
where 1 and 1 are independent.

With the availability of a second signal, worker and manager contracts are modified because
additional incentive coefficients are now at the disposal of the firm. In keeping with the LEN
framework, we assume that contracts remain linear. For worker i and manager j, their wages
are thus given by:
wi = ai + b1i y1 + b2i y2 ;
ωj = αj + β1j y1 + β2j y2 .
As before, we are interested in conditions under which a second manager is valuable. This
requires, for a given worker team size (n) solving the principal’s optimization problem under
IM and comparing the equilibrium profit with that of the signal-manager setting (i.e., Π(n) in
equation 10). We do so next and demonstrate that, under the IM assignment, the two-manager
profit is identical to the profit in the single-manager setting.
23
Proposition 9 Assume ci = c for all i ∈ N and kj = k for all j = 1, 2. For an exogenous
worker team size n:
(1) Under the IM monitoring assignment, we have, for i ∈ N , j, l ∈ {1, 2}:
b∗1i + b∗2i
e∗i =
cr
G(n)r
m∗j = m∗ = β∗σ
r 2k
∗ n
βjj = β∗ = b∗
2
∗
βjl = 0 f or j 6= l
b∗ji = b∗ = q/2 − cσ rkG(n)/n
p
n (2b∗ )2
ΠIM (n) = ;
2c
(2) The firm is indifferent between employing a single manager or hiring two managers
with the IM assignment, i.e.,
ΠIM (n) = Π(n).
Under the IM assignment, worker contracts involve two performance measures, compared
to one in the base model. The managers’ reaction to incentives is as in the base model. Since
each manager works independently, his pay is not linked to the signal produced by the other
manager (i.e., βjl = 0). For his own signal, each manager’s incentive coefficient is related to
his span of control (n) and is proportional to the worker incentive tied to that signal (in the
same way as in the single-manager case). For each signal, the firm chooses a worker incentive
that is lower than in the single-manager case.
Given that an additional informative signal (y2 ) is available for contracting, it may seem
this would lead to a solution closer to the “first-best.” However, hiring a second manager
increases the costs to the firm. In addition to the direct costs (i.e., personal cost of monitoring
and risk-premium due to the manager), they also include indirect costs of imposing the risk of
the second signal on each worker whose pay depends on the second signal. The principal takes
both the direct and indirect costs into account when designing worker and manager contracts.
In this case, any would-be benefits are exactly offset by the would-be costs.17
17
To see this, observe that the firm chooses to set the worker incentive on each signal at one half of the
worker incentive in the single-manager case, so the total worker incentives are identical. Thus the total team
production stays the same. Total risk-premium also stays the same. This is because each manager works half as
hard, resulting in twice as much variance for each signal. Each worker is exposed to two signals with twice the
variance, while the exposure rate (βi2 ), given the mean-variance preference, is one-fourth as before. Combined,
the total risk-premium due to workers is identical to that in the single-manager setting. A similar argument
applies to the managers’ risk-premia.
24
On the surface, the uselessness of an additional, informative signal appears to contradict
the informativeness criterion in Holmstrom (1979). However, the additional signal here (y2 ) is
costly because the production of the signal is subject to moral hazard (note that y2 is costless
in the Holmstrom setting). Moreover, because the quality of the signal is linked to the personal
and risk-premium cost of the second manager, both the benefit and cost are endogenous and,
in our specification, completely offset each other when another manager is already in place.
The above analysis of the JM and IM scenarios lends strong support for keeping a team
under a single manager’s control, whatever its size. In fact, if the presence of productive
complementarities favors larger worker teams, then the optimal span of control of a single
manager may be quite large. It is also clear that in order to identify an economic rationale
for splitting a team into smaller groups under different managerial supervision, we must focus
on settings in which monitoring synergies exist across multiple managers. In the next section,
we provide such a ”possibility” result. Specifically, we provide a parametric illustration of
conditions under which the synergy is large enough that the division of labor between two
managers involves splitting the worker-team into two smaller sub-groups.
4.2 Splitting the Worker-team into Two sub-groups
To better understand the economic trade-offs that might lead to a role for creating smaller
teams with dedicated managers, we employ a variant of the “Joint Monitoring” (JM ) setting.
The major difference is that we now specify that the two managers can work together to
produce two signals, where each signal is informative about the collective effort of a separate
sub-group of workers. We label this “Joint Specialized Monitoring” assignment or simply
JSM .
The benefit of having the managers work jointly is that their interaction allows them to
produce signals that are more focused (e.g., have lower variance). This can be achieved when
managers communicate with each other and are able to isolate the contribution of each sub-
group within the worker team. As a result, each signal is more precise than if the managers
had worked independently. In other words, a synergy in monitoring emerges. We have shown
earlier that the disadvantage of having managers work together is that it invites shirking on
their part. The key to the success of this monitoring assignment thus lies in the tradeoff
between the synergy benefit and the shirking incentive inherent in joint monitoring activities.
25
The monitoring technology behind the JSM assignment is characterized as follows:
n
2
H(n)σ 2
X
y1 = ei + 1 where 1 ∼ N 0, ;
m1 + m2
i=1
n
H(n)σ 2
X
y2 = ei + 2 where 2 ∼ N 0, .
m1 + m2
i= n
2
+1
Note that each signal is now focused on a sub-group of the worker team that is half in size. The
precision of each signal is affected by the monitoring effort by both managers, capturing the
idea of joint managerial activities. Function H(n) captures the synergy generated by asking
the two managers to work together to produce specialized signals. If H(n) ≤ G(n), a synergy
is present. In the following analysis, we sometimes assume H(n) = G( n2 ). This specification
represents substantial synergy, especially when G(.) is convex (such as G(n) = n2 ).18
Using the solution techniques developed earlier, we can solve the principal’s problem for the
JSM setting. We next show that under the JSM assignment, the profit with two managers
exceeds that with a single manager if synergy and the value of output (q) are high enough.
Proposition 10 Assume ci = c for all i in N and kj = k for j = 1, 2. For any exogenous

worker team size n:
(1) Under the JSM monitoring assignment, we have, for i ∈ N , j, l ∈ {1, 2}:
b∗1
e∗i = f or i ∈ {1, 2, ..., n/2}
c
b∗2
e∗i = f or i ∈ {n/2 + 1, n/2 + 2, ..., n}
c r
β∗ H(n)r
m∗j = m∗ = σ
2r k
∗ n
βjl = β∗ = b∗ ,
2
b∗ji = b∗ =
p
q − cσ 2rkH(n)/n
nb∗ h p i
ΠJS (n) = q − 3cσ 2rkH(n)/n ;
2c
(2) Let G(n) = n2 and H(n) = G( n2 ). If the firm chooses between hiring one manager and
hiring two managers with the JSM assignment, the firm will hire two managers if and only if
q is large enough.
Under the JSM assignment, a worker contract involves only the signal that is informative
about his own sub-group. However, worker incentives differ from those in the single-manager
18
A recent study by Marino and Zabojnik (2004) also considers the optimality of splitting a team into two
sub-groups. While we consider the resulting monitoring synergies, their key driving force is the incentive tradeoff
of holding a tournament between two groups of risk-neutral workers.
26
setting. First, even with no synergy (H(n) = G(n)), worker incentives are higher than in the
one-manager setting (and resemble those of the JM setting). With positive synergy (H(n) <
G(n)), worker incentives are even higher and result in even greater levels of production. The
managers’ reactions to incentives also differ in important ways. First, similar to the JM setting,
the shirking incentives due to joint monitoring reduces each manager’s response. Second, the
synergy (H(n)) matters here. Each manager’s pay depends on both signals because each
contributes to improving the precision of both signals.
In deciding whether to expand the number of managers, the usual direct and indirect cost
associated with hiring a second manager must be considered, as well as the hidden cost of
shirking between the two managers. Notice that with the presence of synergy, each signal is
now focused and thus more precise for a given managerial effort than the single-manager case.
When such synergy (indexed by the H(n) function) is substantial, hiring a second manager
with a JSM assignment overcomes not only the usual direct and indirect cost of the additional
monitoring activity but also the hidden cost due to the induced free-riding incentives associated
with multiple managers. In that case, the JSM assignment allows the firm higher profits than
the single-manager setting.
The analysis of the two-manager setting (Section 4) points to the desirability of concentrat-
ing monitoring activities within a small group of managers (e.g., a single manager in our stylized
model). The economic intuition is that placing seemingly substantial risk on a single manager
produces a strong incentive for the manager to work hard on monitoring (i.e., to reduce the
variance of performance measure).19 The desirability of hiring more managers lies in the syn-
ergy/complementarity that exists with group monitoring (such as the low-variance specialized
signals in the JSM setup). Given the free-riding problem associated with the expansion of the
size of the management team, it is essential for organizations to realize substantial synergy for
group monitoring to be viable.20
5 Conclusion
In this paper, we study three broad instruments in the internal design of an organization involv-
ing team production: team size, monitoring activities, and the incentive contracts offered to
workers and managers. When both types of agents (workers and managers) are self-interested
and subjected to moral hazard, there exist complex interactions in the trade-offs among these
instruments. We show, however, that such complex interactions produce rather simple and
19
See Huddart and Liang 2004 for a similar argument in a partnership setting.
20 G(n)
In fact, we can show in the JSM setting that if synergy exists but is not large, say H(n) = 2
while
2
G(n) = n , the JSM assignment is inferior to the single-monitor setting for any value of q.
27
stark implications. One such result is the robustness of worker contracts, i.e., the invariance
property, whereby the equilibrium pay-for-performance scheme offered to workers does not
vary with many environmental variables of interest. Another is our finding that the cost of
managerial free-riding is such that it is almost never worthwhile to employ multiple managers
to supervise a given set of workers. Our work also demonstrates the presence of complementar-
ities between team size and monitoring, and between worker talent and managerial monitoring
ability. Finally, we are able to derive unambiguous predictions about the impact of environ-
mental variables on the choice of optimal team size and structure, even in the presence of an
external marketplace for talent.
Our exploration may be extended in at least three broad ways. First, we take some steps
in the latter half of the paper to get at issues related to the assignment of managerial talent
across supervisory roles. This analysis could be expanded further, under diverse assumptions
regarding the synergies available under production (which argues for the formation of a single
team and a single manager) and complementarities involving monitoring (which pushes for
multiple teams and multiple managers). For instance, if it is possible to form more than two
sub-groups under the (suitably adjusted form of the) JSM assignment, under what conditions
would it be efficient to do so, and what is the optimal number of such sub-groups? Another
promising avenue for future study is the extension of our model to a multi-period setting.
This would allow for analysis of the tradeoffs inherent in team incentives and monitoring
in a dynamic setting, and under various scenarios involving the firm’s ability to commit to
long-term incentive structures. Finally, all of our work has been carried out in a multi-agent
LEN framework. While it would be virtually impossible to replicate our analysis in a general
agency model, it would be of interest to examine (perhaps via numerical simulations) whether
our comparative static results, for example, hold in a broader range of settings.
References
Romana Autrey. Team synergy, team composition and performance measures. HBS Working
Paper, 2005.
Stanley Baiman and Joel Demski. Economically optimal performance evaluation and control
systems. Journal of Accounting Research, 18:184–220, Supplement 1980.
Tim Baldenius, Nahum Melumad, and Amir Ziv. Monitoring in multiagent organizations.
Contemporary Accounting Research, 19(4):483–511, Winter 2002.
Sugato Bhattacharyya and Francine Lafontaine. Double-sided moral hazard and the nature of
share contracts. RAND Journal of Economics, 26(4):761–781, Winter 1995.
28
Robert Bushman, Raffi Indjejikian, and Abbie Smith. Aggregate performance measures in
business unit manager compensation: The role of intrafirm interdependencies. Journal of
Accounting Research, 33:101–128, Supplement 1995.
Guillermo Calvo and Stanislaw Wellisz. Supervision, loss of control, and the optimum size of
the firm. Journal of Political Economy, 86(5):943–952, October 1978.
Milton Harris and Artur Raviv. Organizational design. Management Science, 48(7):852–865,
July 2002.
Bengt Holmstrom. Moral hazard and observability. Bell Journal of Economics, 10(1):74–91,
Spring 1979.
Bengt Holmstrom. Moral hazard in teams. Bell Journal of Economics, 13(2):324–340, Autumn
1982.
Steve Huddart and Pierre Liang. Profit sharing and monitoring in partnerships. Journal of
Accounting and Economics, 40(1-3):153–187, December 2005.
John Hughes, Li Zhang, and Jai-Zheng Xie. Production externalities, congruity of aggregate
signals, and optimal task assignments. Contemporary Accounting Research, 61(2):398–408,
Summer 2005.
Anthony Marino and Jan Zabojnik. Internal competition for corporate resources and incentives
in teams. RAND Journal of Economics, 35(4):710–727, Winter 2004.
Jacob Marschak and Roy Radner. Economic Theory of Teams. Cowles Foundation. Yale
University Press, New Haven, 1972.
Kenneth McLaughlin. Individual compensation and firm performance: The economics of team
incentives. University of Chicago - George G. Stigler Center for Study of Economy and State
Working Paper, 1994.
Canice Prendergast. The tenuous trade-off between risk and incentives. Journal of Political
Economy, 110(5):1071–1102, 2002.
Yingyi Qian. Incentives and loss of control in an optimal hierarchy. The Review of Economic
Studies, 61(3):527–544, July 1994.
Joan E. Ricart I Costa. Managerial task assignment and promotions. Econometrica, 56(2):
449–466, Mar 1988.
29
John Roberts. The Modern Firm: Organizational Design for Performance and Growth. Claren-
don Lectures in Management Studies. Oxford University Press, Oxford, May 2004.
David Sappington. Limited liability contracts between principal and agent. Journal of Eco-
nomic Theory, 29(1):1–21, Feb 1983.
Robert Simons. Levers of Organizational Design. Harvard Business School Press, Boston, July
2005.
Hal Varian. Monitoring agents with other agents. Journal of Institutional and Theoretical
Economics, 146:153–174, 1990.
Oliver Williamson. Hierarchical control and optimum firm size. Journal of Political Economy,
75(2):123–138, 1967.
Amir Ziv. Performance measures and optimal organization. Journal of Law, Economics, and
Organization, 9(1):30–50, 1993.
Amir Ziv. Information technology and optimal firm structure. Journal of Accounting Research,
38(2):297–328, 2000.
6 Appendix I
Proof of Proposition 1. The optimal incentive weight is:
q
b= . (22)
1 + rcG(n)σ 2
Substituting this into the profit function in (3) gives
q2n
max . (23)
2c(1 + rcσ 2 G(n))
n>0
To explore the dependence of b on parameter values, we change variables, using b instead of n

in our maximization problem. Using (22), we can express n in terms of b:

q
n=H θ −1 ,
b
where θ = 1/(rcσ 2 ), and H is the inverse of G. Substituting into (23), and getting rid of
multiplicative constants, we see that our original problem (23) is equivalent to

q
max H θ − 1 b. (24)
b
0<b<q
30
Sufficiency. Observe that G is a degree function if and only if its inverse H is. Thus, we know
that H(x) = Axδ , for some A and δ. Substituting into (24) and getting rid of multiplicative
constants, we see that our initial problem is equivalent to
q δ
max − 1 b.
b
0<b<q
Now it is obvious that the optimal b, if it exists,21 does not depend on any of the parameters
of interest.
Necessity. It is easy to see that at the optimal n, G0 (n) 6= 0 (use FOC for the original
problem (23)). This means that H is also differentiable in a neighborhood of its optimal
argument. Consequently, the FOC holds for problem (24):

∗ q 0 q
b H θ ∗ −1 − θqH θ ∗ − 1 = 0. (25)
b b
The fact that the optimal b∗ does not depend on σ 2 , r and c, in terms of our modified prob-
lem (24), means that b∗ does not depend on θ.
Let us define
q
f (x) = H x ∗ − 1 .
b
Taking derivatives and rearranging,
b∗

0 q
H x ∗ −1 = f 0 (x).
b q − b∗
Substituting into (25) (with x = θ) and dividing through by b∗ 6= 0 we get
q
f (θ) − θf 0 (θ) = 0, (26)
q − b∗
which holds for all θ > 0. Let us solve this differential equation for f . Rearrange (26) as
f 0 (θ) (q − b∗ ) 1
=
f (θ) q θ
This is equivalent to
d ln(f (θ)) (q − b∗ ) d ln θ
=
dθ q dθ
Now, since b∗ is independent of θ, we can integrate both sides of this equation over θ > 0 and
(q−b∗ )
take q outside of the right-hand side integral. Thus
(q − b∗ )
ln f (θ) = ln θ + C1
q
21
Note that optimal b exists if and only if δ < 1.
31
Applying the exponential function to both sides of this equation yields
(q − b∗ )
f (θ) = Cθδ , where δ = and C = eC1 .
q
Finally, using the definition of function f , we get
δ
b∗

δ
H(x) = Ax , where A = C .
q − b∗
Again, notice that H is a degree function if and only if so is G, which completes the proof.
Proof of Lemma 1. All claims in the lemma are derived in the text.
Proof of Proposition 2. Totally differentiating this equation w.r.t. ci gives:
P
r
rnσ 2
 
2 2
√
∂bi b 2ci rnσ 2 −bi
j∈N j  ∂bi
= − 2 −

P P
2  3 
∂ci r
rnσ 2
r
rnσ 2
P
2 2
∂ci
1 + 2ci b2
1 + 2ci b2 j∈N bj
j∈N j j∈N j
Thus,
Prnσ b Prnσ b
 r  r
2 2
2ci 2
" # 2 2
 j∈N j bi  ∂b
 i j∈N j
1 − = −

P P
2 P 2 2
 ∂ci

j∈N bj
r r
rnσ 2 rnσ 2

1 + 2ci b2j
1 + 2ci b2j
j∈N j∈N
Using the fact that

1 2
2 = b i
P
r
rnσ 2
1 + 2ci b2j
j∈N
we can continue:
Prnσ b
r
2
s " #! 2 2
rnσ 2 b2i ∂bi j∈N j
1 − 2ci bi = −
Prnσ b
2 2 2
P P
j∈N bj j∈N bj ∂ci r
2
1 + 2ci 2
j∈N j
We know that Pb 2
i
j∈N b2j
< 1 and 0 < bi < 1. Hence,
s " # s
rnσ 2 b2i rnσ 2
1 − 2ci bi P 2
P 2 > 1 − 2ci bi P 2 = bi > 0
j∈N bj j∈N bj j∈N bj
32
∂bi
So the multiplicative term on ∂ci is positive. We also know that:
Prnσ b
r
2
2 2
j∈N j
− 2 < 0
P
r
rnσ 2
1 + 2ci b2j
j∈N
Thus,
∂bi
<0
∂ci

Proof of Proposition 3. One can verify the claim using a similar proof strategy to that
used in the proof of Proposition 1. In particular, the incentive weight is:
b∗ = q − 2cσ
p
rkG(n)/n.
Substituting this into the profit function in (10) gives

p
n(q − 2cσ rkG(n)/n)2
max .
2c
n>0
As before, we can express n in terms of b:

n = Ĥ θ (q − b) ,
√ p
where θ = 1/(2σc rk), and Ĥ is the inverse to G(n)/n. The problem is thus equivalent to
max Ĥ θ (q − b) b2 .

0<b<q
Sufficiency and necessity are then established following the same technique in the proof of
Proposition 1.
Proof of Lemma 2. All claims in the lemma are derived in the text.
33
Proof of Proposition 4. To prove claim (1), compare the profit functions in claim (1) of
Lemmas 1 and 2. We can compute the value of the manager as:
q
n(q − 2cσ rkG(n)/n)2 nq 1+rcG(n)σ2
p
Π(n) − Π̂(n) = −
2c 2c
n p 2 q
= (q − 2cσ rkG(n)/n) − q
2c 1 + rcG(n)σ 2
   
n  p q √  2 2
= qcσ rG(n)/n  1+rcG(n)σ2 − 4 k  + 4c rkσ G(n)/n ,
 
2c √ 2 rG(n)nσ
which leads immediately to claim (1) in the proposition.

To prove claim (2), compare the profit functions in claim (2) of Lemmas 1 and 2. The
value of the manager is:
q4 q2
Π(n∗ ) − Π̂(n̂) = − √
128c3 σ 2 rk 4 rc3 σ 2
q2 q2

= √ √ −1 .
4 rc3 σ 2 32k rc3 σ 2
Also, it is easy to verify that
q2 q2 n∗ (r, c, σ, k)
Π(n∗ ) − Π̂(n̂) > 0 ⇐⇒ √ > 1 =⇒ √ = > 1.
32k rc3 σ 2 16k rc3 σ 2 n̂(r, c, σ)
Combined, these findings verify claim (2) in the proposition.
p
Proof of Proposition 5. Under the assumption G(n)/n = Anγ , our objective function is
√ 2
n q − 2cσ rkAnγ
Π(n, c, k; σ, r) = − nū(c) − ūM (k). (27)
2c
√
If we use b = q − 2cσ rkAnγ instead of n (change of variables), then
1/γ
q − b 1/γ

1
n= √ .
Aσ rk 2c
Substituting into (27) and getting rid of n, we get our objective function
1/γ
q − b 1/γ b2

1
Π(b,
b c, k; σ, r) = √ − ū(c) − ūM (k).
Aσ rk 2c 2c
The optimal b and c are then determined from the following optimization:
max Π(b,
b c, k; σ, r) such that 0 ≤ b < q and c > 0.
b, c
34
Obviously, this problem is equivalent to
q − b 1/γ b2

max − ū(c) such that 0 ≤ b < q and c > 0,
b, c 2c 2c
which does not depend on σ, r or k.
Proof of Proposition 6. The objective function for maximization is

√ 2
n q − 2cσ rkAnγ
Π(n, c, k; σ, r) = − nū(c) − ūM (k)
2c
q2 √
= n − 2qσ rkAnγ+1 + 2cσ 2 rkA2 n2γ+1 − nū(c) − ūM (k).
2c
Our maximization problem now becomes
√
max Π(n, c, k|σ, r) such that n, c, k > 0 and q − 2cσ rkF (n) ≥ 0.
n, c, k
First, observe that the comparative statics with respect to σ and r always share the same
sign. The optimal value functions are Z ∗ (σ, r) where Z = n, c or k. Notice that parameters σ
√
and r enter our problem only through the expression σ r. This implies that
√
Z ∗ σ, r = Z ∗ σ r, 1 .

Consequently,
√
∂Z ∗ σ, r √ ∂Z ∗ σ r, 1

= r
∂σ ∂σ
∂Z ∗ σ, r ∗ σ √r, 1

σ ∂Z
= √ .
∂r 2 r ∂σ
Therefore, √
∂Z ∗ σ, r ∂Z ∗ σ, r ∂Z ∗ σ r, 1

sign = sign = sign .
∂σ ∂r ∂σ
Assumption 1 The optimum is reached in the interior.
First Order Conditions

q2 √
Πn = − 2q(1 + γ)σ rkAnγ + 2(1 + 2γ)cσ 2 rkA2 n2γ − ū(c) = 0; (28a)
2c
q2

0 2 2 2γ
Πc = n −u (c) − 2 + 2σ rkA n = 0; (28b)
2c
r
0 r
Πk = −uM (k) − qσ Anγ+1 + 2cσ 2 rA2 n2γ+1 = 0. (28c)
k
35
The optimal values n∗ (σ, r), c∗ (σ, r) and k ∗ (σ, r) solve system (28). The implicit function
proposition ensures the existence of differentiable implicit functions n∗ (σ, r), c∗ (σ, r) and
k ∗ (σ, r), provided the matrix
 
Πnn Πnc Πnk
D2 Π = 
 
 Πnc Πcc Πck 

Πnk Πck Πkk
is nondegenerate.
Assumption 2 Matrix D2 Π is negative definite at the optimal solution.
A matrix is negative definite if and only if the leading principal minors switch signs:
Πnn < 0 (29)

Πnn Πcc − Π2nc > 0 (30)
det D2 Π < 0 (31)
Equation (31) shows that we can apply the implicit function theorem to system (28). Differ-
entiating (28) w.r.t. σ, we get
    
Πnn Πnc Πnk ∂n∗ /∂σ Πnσ
    
 Πnc Πcc Πck   ∂c∗ /∂σ  +  Πcσ  = 0. (32)
    
Πnk Πck Πkk ∂k ∗ /∂σ Πkσ
In other words,    
∗
∂n /∂σ Πnσ
 ∂c∗ /∂σ  = − D2 Π −1  Πcσ  .
   
   
∂k ∗ /∂σ Πkσ
Second derivatives. Differentiating (28a), we get

√
Πnn = 4γ(1 + 2γ)cσ 2 rkA2 n2γ−1 − 2qγ(γ + 1)σ rkAnγ−1 . (33)
Differentiating (28b) w.r.t. n and using FOC for c, we get
q2
Πnc = −u0 (c) − + 2σ 2 rkA2 n2γ + 4γσ 2 rkA2 n2γ = 4γσ 2 rkA2 n2γ , (34)
2c2
q2
where −u0 (c) − 2c2
+ 2σ 2 rkA2 n2γ = 0 by (28b). Differentiating (28c) w.r.t. n, we get
r
r
Πnk = −q(γ + 1)σ Anγ + 2(2γ + 1)σ 2 crA2 n2γ .
k
36
In a similar manner,
q2

Πcc = n −u00 (c) + 3 ,
c
Πck = 2σ 2 rA2 n2γ+1 ,
1
Πkk = −u00M (k) + qσr1/2 k −3/2 Anγ+1 ,
2
√
Πnσ = −2q(1 + γ) rkAnγ + 4(1 + 2γ)cσrkA2 n2γ ,
Πcσ = 4σrkA2 n2γ+1 ,

r
r
Πkσ = −q Anγ+1 + 4cσrA2 n2γ+1 , (35)
k
r
k γ
Πnr = −q(1 + γ)σ An + 2(1 + 2γ)cσ 2 kA2 n2γ ,
r
Πcr = 2σ 2 kA2 n2γ+1 ,
1
Πkr = −qσ √ Anγ+1 + 2cσ 2 A2 n2γ+1 .
2 rk
Finally, by direct calculations, using formulas above, we can see that
n n
Πnσ = Πnn , Πnk = Πnn , (36)
γσ 2γk
Similarly,
n n
Πcσ = Πnc , Πck = Πnc . (37)
γσ 2γk
Behavior of c∗ (σ, r). Apply Cramer’s rule to the linear system (32) and get
 
Π nn Π nσ Π nk
∂c∗ 1  
=− det  Πnc Πcσ Πck  .
∂σ det D2 Π  
Πnk Πkσ Πkk
In other words,
∂c∗
− det D2 Π = −Πnσ Πnc Πkk − Πck Πnk
∂σ
+ Πcσ Πnn Πkk − (Πnk )2 − Πkσ Πnn Πck − Πnk Πnc .
Eliminate Πnk , Πnσ , Πcσ and Πck by substituting (36) and (37):
∂c∗ n2

2 n
− det D Π = − Πnn Πnc Πkk − 2 2 Πnc Πnn
∂σ γσ 4γ k
n2

n
+ Πnc Πnn Πkk − 2 2 (Πnn )2
γσ 4γ k

n n
−Πkσ Πnn Πnc − Πnn Πnc = 0.
2γk 2γk
37
Now det D2 Π 6= 0, from (31), so
∂c∗
= 0.
∂σ
Behavior of k ∗ (σ, r). Again using Cramer’s rule for the linear system (32), we get
 
Π Πnc Πnσ
∂k ∗ 1  nn 
=− det 
 Πnc Πcc Πcσ  .
∂σ det D2 Π
Πnk Πck Πkσ
In other words,
∂k ∗
− det D2 Π = Πnσ Πnc Πck − Πcc Πnk
∂σ
− Πcσ Πnn Πck − Πnc Πnk + Πkσ Πnn Πcc − (Πnc )2 .
We eliminate Πnk , Πnσ , Πcσ , and Πck by substituting for (36) and (37):
∂k ∗

2 n n 2 n
− det D Π = Πnn (Πnc ) − Πcc Πnn
∂σ γσ 2γk 2γk

n n n
− Πnc Πnn Πnc − Πnc Πnn + Πkσ Πnn Πcc − (Πnc )2
γσ 2γk 2γk
n2

2
= Πnn Πcc − (Πnc ) Πkσ − 2 Πnn . (38)
2γ kσ
The last factor in the expression above is
n2
r
r
Πkσ − 2 Πnn = −q Anγ+1 + 4cσrA2 n2γ+1
2γ kσ k
n2 √
− 2 −2qγ(γ + 1)σ rkAnγ−1 + 4γ(1 + 2γ)cσ 2 rkA2 n2γ−1
2γ kσ
r
r
= −q Anγ+1 + 4cσrA2 n2γ+1
k
r
1+γ r 1 + 2γ
q Anγ+1 − 2 cσrA2 n2γ+1
γ k γ
r
q r 2
= Anγ+1 − cσrA2 n2γ+1
γ k γ
γ+1 √
r
An r γ
= q − 2cσ rkAn
γ k
Anγ+1 r
r
= b. (39)
γ k
Substituting back into (38), we get
∂k ∗ γ+1
r
1
2 An
r
=− Π nn Π cc − (Πnc ) b > 0,
∂σ det D2 Π γ k
1 2
where − det D 2 Π > 0 and Πnn Πcc − (Πnc ) > 0, by (30) and (31).
38
Behavior of n∗ (σ, r). We use Cramer’s rule for the linear system (32) to obtain
 
Π Πnc Πnk
∂n∗ 1  nσ 
=− det  Πcσ
 Πcc Πck  .
∂σ det D2 Π
Πkσ Πck Πkk
In other words,
∂n∗
− det D2 Π = Πnσ Πcc Πkk − (Πck )2
∂σ
− Πcσ Πnc Πkk − Πnk Πck + Πkσ Πnc Πck − Πcc Πnk .
We eliminate Πnk , Πnσ , Πcσ and Πck by substituting (36) and (37):
∂n∗ n2

n
− det D2 Π = Πnn Πcc Πkk − 2 2 (Πnc )2
∂σ γσ 4γ k
n2

n n 2 n
− Πnc Πnc Πkk − 2 2 Πnn Πnc + Πkσ (Πnc ) − Πcc Πnn
γσ 4γ k 2γk 2γk
n n
= Πnn Πcc Πkk − Π2nc Πkk + Πkσ Π2nc − Πnn Πcc
γσ 2γk

n n
= Πkk − Πkσ Πnn Πcc − (Πnc )2 . (40)
γσ 2γk
Let us show that the first factor in the last expression above is negative. Since the ordering
of the rows and columns of the Hessian D2 Π is arbitrary, we can swap the second and third
columns and rows of the matrix. As before, the matrix is negative definite if and only if the
leading principal minors alternate in sign. In particular, this implies that Πnn Πkk −(Πnk )2 > 0.
Using (36),
n2

2
0 < Πnn Πkk − (Πnk ) = Πnn Πkk − 2 2 Πnn . (41)
4γ k
As Πnn < 0 from (29), inequality (41) implies
n2
Πkk < Πnn . (42)
4γ 2 k 2
Using this inequality to eliminate Πkk from (40), and using (39) to simplify, we get
n n n n2 n
Πkk − Πkσ < × 2 2 Πnn − Πkσ
γσ 2γk γσ 4γ k 2γk
n2

n
= Πnn − Πkσ
2γk 2γ 2 σk
n Anγ+1 r
r
= − b < 0. (43)
2γk γ k
39
Thus, from (40) we get
∂n∗

1 n n
=− Πkk − Πkσ Πnn Πcc − (Πnc )2 < 0,
∂σ det D2 Π γσ 2γk
1 n n
where − det D 2 Π > 0 by (31), γσ Πkk − 2γk Πkσ < 0 by (43), and Πnn Πcc − (Πnc )2 > 0 by (30).
If γ < 0, optimal Π will be negative. For the sake of completeness, it is easy to see that
in case γ < 0 the following discussion with slight changes still applies, yielding the opposite
results: ∂c∗ /∂σ = ∂c∗ /∂r = 0, ∂k ∗ /∂σ < 0, ∂k ∗ /∂r < 0, ∂n∗ /∂σ > 0, ∂n∗ /∂r > 0.
Proof of Proposition 7. Differentiate (28) with respect to γ:

√
Πnγ = 4cσ 2 rkA2 n2γ − 2qσ rkAnγ
√
+ 4(1 + 2γ)cσ 2 rkA2 n2γ − 2q(1 + γ)σ rkAnγ ln n
1 √ √
= 4(1 + 2γ)cσ 2 rkA2 n2γ − 2q(1 + γ)σ rkAnγ − 2qγσ rkAnγ
1 + 2γ
√
+ 4(1 + 2γ)cσ 2 rkA2 n2γ − 2q(1 + γ)σ rkAnγ ln n
√

2 2 2γ γ 1
= 4(1 + 2γ)cσ rkA n − 2q(1 + γ)σ rkAn + ln n
1 + 2γ
2γ √
− qσ rkAnγ
1 + 2γ
√
Since 4(1 + 2γ)cσ 2 rkA2 n2γ − 2q(1 + γ)σ rkAnγ = nγ Πnn by (33),
√

n 1 2γ
Πnγ = Πnn + ln n − qσ rkAnγ . (44)
γ 1 + 2γ 1 + 2γ
The two remaining equations are straightforward.

n
Πcγ = 4σ 2 rkA2 n2γ+1 ln n = Πnc ln n see (34). (45)
γ
r
r γ+1 2 2 2γ+1
Πkγ = −qσ An + 4cσ rA n ln n = Πkσ σ ln n see (35). (46)
k
Behavior of c∗ (γ). By applying Cramer’s rule to a system analogous to (32) (with γ in place
of σ), we get  
Π Πnγ Πnk
∂c∗ 1  nn 
=− det  Πnc Πcγ

.
Πck 
∂γ det D2 Π
Πnk Πkγ Πkk
40
In other words,
∂c∗
− det D2 Π

= −Πnγ Πnc Πkk − Πnk Πck
∂γ
+ Πcγ Πnn Πkk − (Πnk )2 − Πkγ Πnn Πck − Πnc Πnk . (47)

We use (36) and (37) to eliminate Πck and Πnk from (47):
∂c∗ n2

2 n n 2
− det D Π = −Πnγ Πnc Πkk − Πnn Πnc + Πcγ Πnn Πkk − 2 2 Πnn
∂γ 2γk 2γk 4γ k

n n
−Πkγ Πnn Πnc − Πnc Πnn
2γk 2γk
n2

= Πcγ Πnn − Πnc Πnγ Πkk − 2 2 Πnn . (48)
4γ k
Let us transform the first factor in equation (48). We use (44) and (45) to eliminate Πnγ and
Πcγ :

n n 1
Πcγ Πnn − Πnc Πnγ = Πnc ln n Πnn − Πnn + ln n
γ γ 1 + 2γ
√

2γ
+ qσ rkAnγ
1 + 2γ
√

n 1 2γ γ
= Πnc −Πnn + qσ rkAn > 0, (49)
γ 1 + 2γ 1 + 2γ
since Πnc > 0 by (34), and −Πnn > 0 by (29). Finally, we get
∂c∗ n2

1
=− Π Π
cγ nn − Π Π
nc nγ Πkk − Π nn < 0,
∂γ det D2 Π 4γ 2 k 2
1 n2
since − det D 2 Π > 0 by (31), Πcγ Πnn − Πnc Πnγ > 0 by (49), and Πkk − Π
4γ 2 k2 nn
< 0 by (42).
Behavior of n∗ (γ). By applying Cramer’s rule to a system analogous to (32) (with γ in

place of σ), we get  
Π Πnc Πnk
∂n∗ 1  nγ 
=− det 
 Πcγ Πcc .
Πck 
∂γ det D2 Π
Πkγ Πck Πkk
In other words,
∂n∗
− det D2 Π = Πnγ (Πcc Πkk − (Πck )2 )
∂γ
− Πcγ (Πnc Πkk − Πck Πnk ) + Πkγ (Πnc Πck − Πcc Πnk ) (50)
41
We use (36) and (37) to eliminate Πck and Πnk from (50):
∂n∗ n2 n2

2 2
− det D Π = Πnγ Πcc Πkk − 2 2 (Πnc ) − Πcγ Πnc Πkk − 2 2 Πnc Πnn
∂γ 4γ k 4γ k

n n
+Πkγ Πnc Πnc − Πcc Πnn
2γk 2γk
n2 n2 n2

2
= Πnγ Πcc 2 2 Πnn − 2 2 (Πnc ) + Πnγ Πcc Πkk − 2 2 Πnn
4γ k 4γ k 4γ k
2

n n 2
−Πcγ Πnc Πkk − 2 2 Πnn + Πkγ (Πnc ) − Πcc Πnn
4γ k 2γk

2 n n
= Πnn Πcc − (Πnc ) Πnγ − Πkγ
2γk 2γk
n2

+ Πnγ Πcc − Πcγ Πnc Πkk − 2 2 Πnn , (51)
4γ k
n
where Πnn Πcc − (Πnc )2 > 0 by (30), 2γk Πnγ − Πkγ = E1 , Πnγ Πcc − Πcγ Πnc = E2 , and
n2
Πkk − 4γ 2 k2
Πnn < 0 by (42).
Before we can calculate the signs of E1 and E2 , defined above, we must make an additional
assumption in order to fix the sign of ln n∗ .
Assumption 3 At the optimum, the value of n is greater than one, i.e., n∗ > 1.
Using (44) and (46) with this assumption, we get

n
E1 = Πnγ − Πkγ
2γk
n2 √

1 n
= Π nn + ln n − qσ rkAnγ − Πkσ σ ln n
2γ 2 k 1 + 2γ (1 + 2γ)k
n2 n2 qσAnγ+1 r
r
1
= −σ ln n Πkσ − 2 Πnn + 2 Πnn −
2γ kσ 2γ k 1 + 2γ (1 + 2γ) k
< 0, (52)
n2
where σ ln n > 0 by 39 by Assumption 3, Πkσ − Π
2γ 2 kσ nn
> 0 by (39), and Πnn < 0 by (29).
Using (44) and (45) similarly, we get
E2 = Πnγ Πcc − Πcγ Πnc

√

n 1 2γ γ n
= Πcc Πnn + ln n − qσ rkAn − Πnc ln n Πnc
γ 1 + 2γ 1 + 2γ γ
√

2 n n 1 2γ γ
= Πcc Πnn − (Πnc ) ln n + Πcc Πnn − qσ rkAn
γ γ 1 + 2γ 1 + 2γ
> 0, (53)
42
n
where Πcc Πnn − (Πnc )2 > 0 by (30), γ ln n > 0 by Assumption 3, Πcc < 0, and Πnn < 0 by
Assumption 2. Combining (51), (52), and (53), we see that
∂n∗
< 0.
∂γ
Behavior of k ∗ (γ). By applying Cramer’s rule to a system analogous to (32) (with γ in

place of σ), we get  
Π Π Π
∂k ∗ 1  nn nc nγ

=− det  Π nc Π cc Π cγ
.
∂γ det D2 Π  
Πnk Πck Πkγ
In other words,
∂k ∗
− det D2 Π

= Πnγ Πnc Πck − Πnk Πcc
∂γ
− Πcγ Πnn Πck − Πnc Πnk + Πkγ Πnn Πcc − (Πnc )2

To get rid of Πck and Πnk , we use (36) and (37).
∂k ∗

2 n 2 n
− det D Π = Πnγ (Πnc ) − Πnn Πcc
∂γ 2γk 2γk

n n 2
− Πcγ Πnn Πnc − Πnc Πnn + Πkγ Πnn Πcc − (Πnc )
2γk 2γk

2 n
= Πnn Πcc − (Πnc ) Πkγ − Πnγ > 0,
2γk
n
where Πnn Πcc − (Πnc )2 > 0 by (30), and Πkγ − 2γk Πnγ = −E1 > 0 by (52). Therefore,
∂k ∗
> 0.
∂γ

43
Proof of Proposition 8. The first two claims in part (1) are derived in the text. The
derivation of the profit function is as follows:
" #
qb∗ b∗ 2
p p
∗2 σ 2rkG(n)/n ∗ ∗2 σ 2rkG(n)/n
ΠJM (n) = n − −b − k(2m ) − 2β
c 2c 2b∗ 2b∗
r p
nb∗ h ∗
p i
∗√ G(n)r ∗2 σ 2rkG(n)/n
= 2q − b − cσ 2rkG(n)/n − kb nσ − 2nb
2c 2k 2b∗
nb ∗ h i
2q − b∗ − cσ 2rkG(n)/n − .5nb∗ σ 2rkG(n)/n − nb∗ σ 2rkG(n)/n
p p p
=
2c
nb∗ h i
2q − b∗ − cσ 2rkG(n)/n − 3cσ 2rkG(n)/n
p p
=
2c
nb∗ h p i
= q − 3cσ 2rkG(n)/n .
2c
As to claim (2), recall that with one manager, the profit for any exogenous n is given by
n
√ 2
2c q − 2cσ rnk . So, the value of a second manager is:
n h √ ih p i n p 2
q − cσ 2rnk q − 3cσ 2rkG(n)/n − q − 2cσ rkG(n)/n
2c 2c
n h p √ p i
= q − 2cσ rkG(n)/n + (2 − 2)cσ rkG(n)/n
2c
h p √ p i n p 2
q − 2cσ rkG(n)/n + (2 − 3 2)cσ rkG(n)/n − q − 2cσ rkG(n)/n
2c
√ √ √ p
2
n p p
= (4 − 4 2)cσ rkG(n)/n q − 2cσ rkG(n)/n + (2 − 2)(2 − 3 2) cσ rkG(n)/n .
2c

∗
p
Notice that b = q − 2cσ rkG(n)/n > 0 and all other parameters (c, r, k, G(n), and σ)
are positive, so the above quantity is always negative. The second manager thus has negative
value.
Proof of Proposition 9. To show claim (1), we start with each worker’s problem under
IM ,
X X ci 2 r 2 n2 σ 2 r 2 n2 σ 2
max ai + b1i ej + b2i ej − e − b1i − b2i
ei 2 i 2 m1 2 m2
j j
b1i +b2i
which leads to the first-order condition, e∗i = ci .
Now, each manager’s problem is
n
X r G(n)σ 2
max αj + βj ei − kmj − βj2 .
mj 2 mj
i
44
It is clear that each manager should be compensated based only on the signal he produces
∗ = 0, for l 6= j). Using the more intuitive notation β = β , the optimization program
(i.e., βlj jj j
q
∗ G(n)r
leads to the first-order condition, mj = βj σ 2kj .
Substituting these conditions, along with the IR constraints, the principal’s problem is:
     
n n
X qbi X X X X X
max − α1 + β1 ej  − α2 + β2 ej  − ai + b1i ej + b2i ej 
β1 ,β2 ,bi ci
i=1 j j i=1 j j
n
" #
X q (b1i + b2i ) (b1i + b2i ) 2 2 2
r G(n)σ r G(n)σ
or max − − b21i − b22i
β,bi ci 2ci 2 m1 2 m2
i=1
r G(n)σ 2 r 2 G(n)σ 2
−k1 m1 − k2 m2 − β12 − β2 .
2 m1 2 m2
This leads to the following first-order conditions:
q − b∗2i q − b∗1i
b∗1i = 2 and b∗2i = 2
1 + rci G(n)σ
m1 1 + rci G(n)σ
m2
n
" #
r G(n)σ 2 X ∂
and b2i − 2kj m∗ (βj ) = 0.
2 (mj )2 ∂βj j
i=1
Substituting mj and ∂ ∗
∂βj mj (βj ) into the last condition, we arrive at
v
u n
∗
u1 X
βj = t b2i .
2
i=1
Now simplifying these conditions with identical workers and managers (i.e., ci ≡ c and
ki ≡ k) implies b∗i ≡ b∗ , βj∗ ≡ β ∗ and m∗j ≡ m∗ . We use these identities to solve β ∗ and b∗ . In
equilibrium, we have
r
∗ ∗ n
β = b
2
r
G(n)nr
m∗ = b ∗ σ >1
4k
q − b∗ q − b∗
b∗ = 2 = √
1 + rcG(n)σ
m∗ 1+
2cσ rkG(n)/n
b∗
b∗
p
= q/2 − cσ rkG(n)/n
√
r G(n)σ 2 σ rkG(n)/n
2b∗
p
Finally, using the fact that = q − 2cσ rkG(n)/n and 2 m∗ = b∗ , total
45
profits can be computed as follows:
" #
q2b∗ (2b∗ )2 r ∗2 G(n)σ 2 ∗ r ∗2 G(n)σ
2
ΠIM (n) = n − − (2b ) − 2km − 2β
ci 2ci 2 m∗ 2 m∗
p
n2b∗ h ∗
p i
∗ ∗2 σ rkG(n)/n
= 2q − 2b − 2cσ rkG(n)/n − k(2m − 1) − nb
2c r b∗
n2b∗ h i G(n)nr
2q − 2b∗ − 2cσ rkG(n)/n − 2kb∗ σ − nb∗ σ rkG(n)/n
p p
=
2c 4k
n2b ∗ h i
2q − 2b∗ − 2cσ rkG(n)/n − nb∗ σ rkG(n)/n − nb∗ σ rkG(n)/n
p p p
=
2c
n2b∗ h i
2q − 2b∗ − 2cσ rkG(n)/n − 2cσ rkG(n)/n
p p
=
2c
n2b∗ h i
2q − 2b∗ − 4cσ rkG(n)/n
p
=
2c
n (2b∗ )2
= .
2c
∗ )2
To show claim (2), the profit function is ΠIM (n) = n(2b ∗
p
2c , where b = q/2−cσ rkG(n)/n.
Recall that with one monitor the profit for an exogenous team size of n is given by Π(n) =
2
n
p
2c q − 2cσ rkG(n)/n . The two profit functions are identical, implying that the value of a
second manager is zero.
Proof of Proposition 10. To show claim (1), we start with each worker’s problem under
JSM . If i ∈ {1, 2, ...n/2}, then we have
n/2
X ci 2 r 2 H(n)σ 2
max ai + b1i ej − e − b1i
ei 2 i 2 m1 + m2
j=1
b1i
leading to the first-order condition e∗i = ci . For j ∈ {n/2 + 1, n/2 + 2, ...n}, the analogous
b2j
condition is e∗j = cj . It is easy to see that each worker’s contract should depend only on the
signal about his own sub-group of the team, i.e., b∗2i = 0, for i ∈
/ {1, 2, ..., n/2} and b∗1i = 0, for
i∈
/ {n/2 + 1, n/2 + 2, ..., n}.
The manager’s problem is
n/2 n
X X r 2 2
H(n)σ 2
max α1 + β11 ei + β12 ei − kJ mj − β11 + β12 ,
mj 2 m1 + m2
i=1 i=n/2+1
q H(n)r
leading to the first-order condition m∗j = σ 2 + β2
β11 12 2kj − ml , for l 6= j.
46
Substituting these two conditions, along with the usual IR constraints, the principal’s
problem is now
n/2 n
" #
b21i r 2 H(n)σ 2 b22j H(n)σ 2

X qb1i X qb2j r
max − − b + − − b22j
βj l,bi ci 2ci 2 1i m1 + m2 cj 2cj 2 m1 + m2
i=1 j=n/2+1
r 2 2 2 2
H(n)σ 2
−k1 m1 − k2 m2 − β11 + β12 + β21 + β22
2 m1 + m2
leading to the first-order conditions
q q
b∗1i = rci H(n)σ 2
and b∗2j = rcj H(n)σ 2
.
1+ m1 +m2 1+ m1 +m2
For signal yl and manager j, we have

   
2 n/2 n 2
r H(n)σ X X r H(n)σ 2  ∂
b21i + b22i  − k + 2 2 2
m∗ (βlj )

β11 + β12 + β21 + β22
∂βlj j
 
2 (m1 + m2 ) 2 2 (m1 + m2 )2
i=1 i=n/2+1
H(n)σ 2
−rβlj = 0,
m1 + m2
where manager j’s first-order condition implies
s
∂ H(n)r βlj
m∗j (βlj ) = σ q .
∂βlj 2kj β2 + β2 1j 2j
Simplifying these conditions with identical workers and managers (i.e., ci ≡ c and ki ≡ k)
∗ ≡ β ∗ and m∗ ≡ m∗ . We use these identities to solve β ∗ and b∗ . In
implies b∗1i = b∗2i ≡ b∗ , βlj j
equilibrium, we then have
r
∗ ∗ H(n)r
2m = β σ ;
k
q q
b∗ (β ∗ ) = rcH(n)σ 2
= √ ;
1+ cσ H(n)kr
2m∗ 1+ β∗
and r
r H(n)σ 2 ∗2 r H(n)σ 2 ∗2 2

H(n)r ∗ H(n)σ
nb − k + 4β σ − rβ = 0.
2 4m∗2 2 4m∗2 4k 2m∗
47
q
β∗ H(n)r
Substituting m∗ (β ∗ ) = 2 σ k , we have
r r
k k H(n)r H(n)r
0 = ∗2
nb∗2 − k + ∗2 4β ∗2 σ − 2kσ
2β 2β 4k 4k
r
k H(n)r
0 = ∗2
nb∗2 − k + 2k − 2k σ
2β 4k
nb ∗2
0 = −1
2β ∗2
r
n
β∗ = b∗
2
We can incorporate this into the b∗ (β ∗ ) expression to obtain

q q
b∗ = √ = √
cσ H(n)kr cσ 2H(n)kr/n
1 + ∗√ 1+ b∗
b n/2
or
b∗ = q − cσ
p
2rkH(n)/n.
q q
Total Profits with exogenous n, using 2m∗ = β ∗ σ H(n)r k = b ∗σ H(n)nr r
2k , and 2 Var(y) =
√
r H(n)σ 2 σ 2rkH(n)/n
2 2m∗ = 2b∗ , is then given by:
" #
qb∗ b∗ 2
p p
∗2 σ 2rkH(n)/n ∗ ∗2 σ 2rkH(n)/n
ΠJS (n) = n − −b − k(2m ) − 4β
c 2c 2b∗ 2b∗
r r 2 p
nb∗ h ∗
i
∗ H(n)nr n σ 2rkH(n)/n
− 4 b∗
p
= 2q − b − cσ 2rkH(n)/n − kb σ
2c 2k 2 2b∗
nb∗ h i nb∗ p
2q − b∗ − cσ 2rkH(n)/n − σ 2rkH(n)/n − nb∗ σ 2rkH(n)/n
p p
=
2c 2
nb∗ h ∗
p p i
= 2q − b − cσ 2rkH(n)/n − 3cσ 2rkH(n)/n
2c
nb∗ h p i
= q − 3cσ 2rkH(n)/n
2c
To show claim (2), re-express ΠIM (n) as
nb∗ h p i
ΠJS (n) = q − 3cσ 2rkH(n)/n
2c
n h p ih p i
= q − cσ 2rkH(n)/n q − 3cσ 2rkH(n)/n
2c
Recall that with a single monitor, the profit function for exogenous n is given by Π(n) =
2
n
. If G(n) = n2 and H(n) = G( n2 ), the value of a second manager,
p
2c q − 2cσ rkG(n)/n
48
√
letting α = cσ rnk, is given by:
ΠJS (n) − Π(n)

n h ih i n
(q − 2α)2
p p
= q − 1/2α q − 9/2α −
2c h 2c
n p ih p i
= (q − 2α) + α(2 − 1/2) (q − 2α) + α(2 − 9/2)
2c
n
− (q − 2α)2
2c h
nα p p p p i
= (4 − 9/2 − 1/2) (q − 2α) + (2 − 1/2)(2 − 9/2) (α)
2c
5α√
which is greater than zero if and only if: q > 8−4 2
≈ 2.13α
49

Optimal Team Size and Monitoring in Organizations

Uploaded by

Copyright:

Available Formats

Optimal Team Size and Monitoring in Organizations

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

Optimal Team Size and Monitoring in Organizations

Uploaded by

Copyright:

Available Formats

∗

Optimal Team Size and Monitoring in Organizations

Revised: March 2006

are independent, then G(n) = nĜ(n).

The firm’s equilibrium profit for exogenous n, denoted by Π̂(n), is

Proposition 2 Incentives (bi ) fall as the cost of effort (ci ) increases.

2.3 Optimal Incentives and Team Size with a Manager

(2) if G(n) = n2 , and firm size n is endogenous, we have

q q2 q2 n∗ (b∗ (n∗ ))2

2.4 Value of the Manager

(2) if G(n) = n2 , and firm size n is endogenous, we have

3.1 Positive Sorting

Recall that the profit function at the optimal team size is

3.2 Optimal Manager and Agent Quality

max Π (n, c, k|σ, r) such that n, c, k > 0, and b ≥ 0. (18)

Proposition 5 The optimal b and c for a fixed k do not depend on σ, r or k.

Therefore, as γ increases, the manager’s effectiveness of monitoring (m/n2γ ) decreases, and

Proposition 7 As the performance measurement technology improves (γ decreases), it is op-

4 The Value of Management Teams

4.1 Two Managers Supervising the Entire Team

4.1.1 Joint Monitoring

yielding a first-order condition on bi analogous to that in the one-manager setting:

Similarly, the first-order condition on βj , for j = 1, 2, is given by:

Substituting (21) and βi∗ = β ∗ into (20), we have

Combine the last two equations and we have a closed-form for b∗ :

ΠJM (n) < Π(n).

4.1.2 Independent Monitoring

where 1 and 1 are independent.

4.2 Splitting the Worker-team into Two sub-groups

Proposition 10 Assume ci = c for all i in N and kj = k for j = 1, 2. For any exogenous

To explore the dependence of b on parameter values, we change variables, using b instead of n

Proof of Proposition 2. Totally differentiating this equation w.r.t. ci gives:

Using the fact that

Substituting this into the profit function in (10) gives

As before, we can express n in terms of b:

which leads immediately to claim (1) in the proposition.

Proof of Proposition 6. The objective function for maximization is

First Order Conditions

Assumption 2 Matrix D2 Π is negative definite at the optimal solution.

Πnn < 0 (29)

Second derivatives. Differentiating (28a), we get

Differentiating (28b) w.r.t. n and using FOC for c, we get

Πcσ = 4σrkA2 n2γ+1 ,

Proof of Proposition 7. Differentiate (28) with respect to γ:

The two remaining equations are straightforward.

Behavior of n∗ (γ). By applying Cramer’s rule to a system analogous to (32) (with γ in

Using (44) and (46) with this assumption, we get

E2 = Πnγ Πcc − Πcγ Πnc

Behavior of k ∗ (γ). By applying Cramer’s rule to a system analogous to (32) (with γ in

To get rid of Πck and Πnk , we use (36) and (37).

For signal yl and manager j, we have

We can incorporate this into the b∗ (β ∗ ) expression to obtain

ΠJS (n) − Π(n)

You might also like

where 1 and 1 are independent.