MIT LIBRARIES
IP'
DUPL
¥w\fm
'ii
'"iMiUMi
yiJM.,!!'!,
§!
3 9080 02246 2482
''•"""'"iiilil
--
7- l!f
nlmTrfmHff HnUTJmFl
(
:
i
mi<;ia'i!;;iim!i!i;i;'ft;ti:iii!'ui!i
;|[:n«it!J(iii.;
if'
|!i,ilU;;:i(!Mi:ii!i;iii|i|ii'i;i;K:iiiii!;;)iiiifi!;iiriii
ill
I'll'!
m
it.fiii,iMi
Digitized by the Internet Archive
in
2011 with funding from
Boston Library Consortium IVIember Libraries
http://www.archive.org/details/organizationaldeOOathe
DEWEY
Massachusetts
Technology
Department of Economics
Working Paper Series
Institute of
Organizational Design:
Decision Rights and Incentive Contracts
Susan Athey,
MIT and NBER
John Roberts, Stanford University
Working Paper
Room
01 -12
E52-251
50 Memorial Drive
Cambridge, MA 02142
This
paper can be downloaded
witliout cliarge from tine
Network Paper Collection at
http://papers.ssrn. com/paper.taf?abstract_id=261 773
Social Science Research
Technology
Department of Economics
Working Paper Series
Massachusetts
Institute of
Organizational Design:
Decision Riglits and Incentive Contracts
Susan Athey,
MIT
and
NBER
John Roberts, Stanford University
Working Paper
Room
01 -12
E52-251
50 Meimorial Drive
Cambridge, MA 02142
This
paper can be downloaded
witlnout cliarge from the
Paper Collection at
http://papers.ssrn. conn/paper.taf?abstractJd=26 1773
Social Science Research Network
Organizational Design: Decision Rights and Incentive Contracts
Susan Athey and John Roberts'^
Where should
decision rights be lodged in organizations? Michael C. Jensen and William H.
Meckling (1992) argue
that
moving
a decision
away from
involves costs in communication and garbling but
make good
incentives to
organizational design.
they
make
lodge
But generally we expect
decisions.
Why
may
the inherently best-informed party
it
with someone
who
has better
that incentives are part
of the
not just provide incentives to those with the best information so that
the right decisions?
One reason
that the available incentive instruments
is
must serve multiple purposes and
designing them to induce better decisions worsens perfonnance against other organizational
objectives.
means
Our experience suggests
this is a
common
situation in actual organizations;
The
available to affect one sort of behavior or decision mevitably affect the incentives
governing other choices. Then the design of incentive schemes and the allocation of decision
rights
become
interlinked.
This paper looks
agents
to
provide
investments.
Eugene
idea in the specific context of a principal's problem of inducing
unobservable
effort
while
also
motivating
the
Each of these problems has been extensively studied
effort, see, e.g.,
e.g.,
at this
F.
efficient
Bengt Holmstrom 1979, Holmstrom and Paul Milgrom 1991; on decisions
Fama and Jensen
1983,
motivating effort
on the
is
value created in the firm. At the same time,
right investment choices
may
see,
Milgrom and John Roberts 1990a, 1990b, Philippe
We
done best by rewarding agents on precise measures of
total
of
in isolation (on inducing
Aghion and Jean Tirole 1997, Tirole and Matthias Dewatripont 1999).
necessarily
selection
it
is
thus
know
that
their effort, not
clear that getting the
require that the decision makers' rewards be tied to total value
1
created.
The
difficulty is that the available
measures do not allow doing both. The only available
performance measures are aggregates whose component pieces cannot be disentangled, while
must be written
contracts
in
advance of learning about investment
contingencies in the contracts ex ante
eventuality
More
is
possibilities. Including
many
is
impossible, and on-going renegotiation in every ex post
is
not possible to contract on investment projects, nor can
prohibitively costly.
we assume
formally,
that
it
the principal bargain with the agents over the adoption of these projects once they are identified.
Instead, returns to projects are reflected in the performance measures available for use in the
effort-incentive contracting.
Then
together. In this framework,
we
of individuals
to tasks, the
the incentives for effort and for decisions are inextricably tied
explore the interactions
among
the design of jobs and assignment
shape and intensity of effort incentives, and the allocation of authority
over project selection.
We argue
that
it
best-informed party.
may
An
indeed be optimal to assign decisions rights to someone other than the
authority-based hierarchy then emerges endogenously, with
some
agents
being given the right to make organizational decisions over projects that others discovered.
Moreover, as
in
interested way.
Herbert Simon (1952), those in authority will
Simon emphasized
the decisions in a self-
that this will not be ex post efficient.
incentives appropriately, however, the inefficiency
1.
make
may be
By
designing the
mitigated.
The Model
Consider a situation with a risk-neutral principal
Holmstrom and Milgrom
Projects
may
or
may
returns (before any
who
has two tasks to be performed, as in
(1991). In addition, investment opportunities (or "projects") arise.
not be undertaken.
If a project is not undertaken, the principal's
payments she makes) are ni(ei) +
112(^2),
where
e, is
expected
the total effort devoted
and the n, functions are concave.
to task
/
n2(e2)
+
y]
measures,
with
x\
=
V,
+
where
yi.
=
>'i+V2 is the total return to
Tl]ie\) +>'i
+
and
<
var(e,)
If a project is undertaken, the returns are lliCei)
v\
+
eo
and xj =
£\
Assume
v^.
adopted projects. There are two performance
Tiiie^)
+ yi +
Co
+
£2,
where the
dimensions being identically and independently distributed with
distributions be symmetric about 0. Let
I,,
e,
are noise terms,
the project returns are random, with the
that
+
denote E(
yi
>>,
>
0):
=
E(_v,)
This
is
and
0,
a
let
two
the
measure of the
\
importance of project choice.
There are two agents,
rvar(w'),
and
/•
i
=
A, B,
where w'
A and
is
B, with mean-variance utility functions U'
the agent's wealth, the cost-of-effort function (c)
Each agent
parameterizes the agent's risk aversion.
expected
tasks: ej
utility
exceeds his reservation
= ef + ef,j =1,2, and
The
signals
literature,
x,
e
=
e\
=
e-l, i
that
payments
is
likely to
yi.
that exactly
do
so).
one of the two agents "discovers"
When an
These are not
is strictly
c(e')
convex,
commonly assumed
we
will focus
this project
agency
in the
of the signals.
on just one
project.
(and each agent
is
that adopted projects shift the
is
We
We
equally
agent "discovers" a project, he learns the corresponding payoffs
contractible, nor
-
allocate effort to either or both
to the agents are linear functions
think of the projects as being numerous, but for simplicity,
assume
-
A. B.
are observable and contractible, as
and we assume
E(h'')
willing to accept a contract if his
Each agent can
utility.
+
is
=
3^]
and
whether a project has or has not been implemented. Note
means of the two
signals.
Thus the dependence of the payments
the agents on the signals determines the agents' preferences over projects.
to
(From an ex ante
perspective, the decision rules about projects affect the variance of payoffs as well.)
In the spirit of the separation of ownership and control,
learn the returns to projects at
any
cost.
Then
we assume
that the principal cannot
the principal cannot participate in the investment
decision.
It
worth noting, however,
is
that the principal
does not automatically have the right
incentives for project selection in any case, because she will be concerned not with the total
returns but with returns net of payments to the agents.
We also
assume
be suppressed -
that projects cannot
other agent can, at cost
A',
observe
returns. This
its
if
seems
an agent discovers
to
be the appropriate modeling for
projects like choosing an advertising agency, a supplier, or a quality level,
must be made. For other
sorts,
it
a project then the
might be more appropriate to assume
where some decision
that
an agent can hide
projects he does not like, and the monitoring applies only to projects that the agent reveals.
Moreover,
we
ignore the problem of inducing the agents to bear the cost k to learn the value of
when
the other agent's projects, simply assuming that the agents incur k
The
analysis
is
similar but
more complicated when these assumptions
Susan Athey and Roberts (2001
The
maximizing
If there
thus permit
maximize her ex ante expected
transferable through the
were no investment choice
is
at least for
possible)
more
Thus, reward
agent
is
wage
which (due
to
same solution
as
profits,
contracts) yields the
-
say,
A
I'o
issue, the solution to the effort elicitation
A
and
problem would
not too small. Assign one agent to each task (even though
to task
1
and
B
to 2.
To improve
the accuracy of the signals and
intense incentives and increased effort, agent A's incentive contract places a
negative weight on xi to reduce the effect of the
Ofivo/(vo+V2)
are relaxed, as discussed in
total value.
be straightforward,
multitasking
is
it.
).
principal designs the contracts to
the assumption that utility
the regime calls for
with
fix
=
w'^
=ao +
oc\x\
-i32Vo/(vo+vi).
common
noise
+ aixj and B with w*
As
vo approaches 0, ai
rewarded solely on the direct signal of his
own
=j3b
and
j8i
task; but,
£b,
+
and likewise
P\X\
for agent B.
+ ^xj, where Ui = -
also go to zero, so that each
even then,
it
will typically
be
optimal for the agents to specialize on only one task.^ Then a\ and
^
are chosen subject to the
usual incentive constraint that the return to effort must equal the marginal cost, while Oo and ^q
are set so that the participation constraints are just met.
positive weight
We refer to incentives such as these, with
on one signal and negative weight on the
other, as
"comparative performance
evaluation."
Now
consider the
induced preferences over project selection. While
>
increased by a project if and only if vi
y\
>
B
-{a2/a\)y2, while
implement
will
A
will
and only
ifjK2
-yi, agent
if
wish
>
to
surplus
total
implement a project
-{P\iP2)y\-
Note
that this
if
is
and only
means
that
each agent has very bad incentives on project selection, because each puts negative weight on
one element of the
-vo/(vo+V|)
On
=
13]/ 132-
returns.
Thus,
Notice further that
/i's
a
>
v^i,
payment of y(xi+X2)
-vo/(vo+v2)
>
left to
y,
which would cause them
adopt whatever projects he sees and
to
likes.
consider several alternative allocations of decision rights between the two agents over
project selection. In the
first
regime, "empowerment," each agent can choose whether or not to
implement any projects he discovers.
In the
second regime,
''A
payoffs of B's potential projects (at cost k) and can overrule
if
and only
authority. Finally,
learn the payoffs
that all but the
the
=
were important, then the solution would give
for arbitrarily small but positive
value total wealth creation. Then each can be
adopted
^ and ailccx
then (X\>
decision-making incentives are less distorted than 5's.
the other hand, if only project selection
each agent
We
if V2
if
we
they benefit A. The third regime,
fi's decisions.
"5
authority,"
Thus
is
A
learns the
projects are
analogous to
A
consider collegial or "consensus" decision-making, where both agents
from the
others' projects
empowerment regime
two acting together
authority," agent
and both must agree for a project to go forward. Note
give authority over an agent's projects to the other agent or
as a committee.
Two
other possible allocations of authority are conceivable but do not
One
context of our model.
any
While
projects.
this
is
"prohibition," under
might be attractive
be enforced under the assumptions
make
the agents can
transfers
in
which the agents
some circumstances,
we have made. The second
it
is
hard to see
how
between
payments
agents,
it is
hard to see
how
it
might
bargaining and
If the
would be
attractive.
However,
could be enforced, and the problems created by allowing side
particularly
under
comparative
known (Holmstrom and Milgrom,
multitasking, are well
are not allowed to undertake
and decide jointly over project allocation.
this restriction
in the
a "bargaining" regime, where
is
transfers could be limited to the investments, allowing this option
seem relevant
1990).
performance
We
evaluation
and
discuss bargaining at greater
Athey and Roberts (2001).
length in
We now
in different
proceed to compare the returns
to different authority
parameter regions for our model. The model
is
regimes and reward schemes
sufficiently
complex
that
we do
not
attempt to characterize optima except in extreme cases, but instead focus on the nature of the
solutions, the trade-offs involved,
2.
and comparative
statics.
Motivating Effort and Decisions
We
begin with the case where decisions are not very important (so that providing effort
incentives dominates the determination of the reward schemes).
to V]
and
with
0:2
=
V2.
Then
-CC]
at least for
and
/?]
=
-)32
<
0,
and the best allocation of decision
is
because allowing
or the two agent-authority regimes has a
profits,
that vq
is
large relative
the optimal incentives will have strong comparative performance evaluation,
k not too large. This
empowerment
of overall
<
Suppose
but
ex ante total value
it
is
leads the agents to bear
some
new
mean
rights will
be consensus,
projects to go forward under the
effect that
is
close to zero in terms
extra risk. Since the agents are risk averse,
decreased by allowing project adoptions, and the ideal would be
However, since the agents
prohibition.
consensus essentially eliminates
(although the cost k
is
are profitable, so if A"
As
all
will
be
in
almost complete disagreement about projects,
and so largely
projects
replicates the effect of prohibition
incurred for each project). However, the few projects that are implemented
is
small then consensus will certainly be best.
vq falls for fixed
and
V|
vi,
the use of comparative performance evaluation decreases, and
so the distortion of the agents' decision-making decreases. At
large relative to the agents' risk aversion, a
recall that
^'s decisions are
authority, while
vo falls further,
also
become
On
less distorted than 5's,
however, the empowerment regime
the other hand, if k
evaluation diminishes as vq
B
approaches 72 >
0.
is
Thus,
the
A
if the
common
authority
Next,
k,
may become
that
falls,
which
The mean decision
the extra cost of A-
we
quality
to
the
is
is
this,
better than
prohibitively costly.
B
As
>
0,
while that
fewer projects and thus reduced variance in agent
on each project
is
vi
performance
same under consensus decision-making
is
not sufficient to offset this gain.
not great, the solution depends on the magnitude of
Empowerment, consensus or
depending on the parameters.
turn to analyze what happens as decisions
benchmark where
is
A approaches
noise term and the costs of reviewing project returns.
best,
authority
is
preferred, because 5's decisions
the acceptance region for agent
importance of the projects
may be
A
k
v\, if
be preferred. To see
as vq approaches zero. Since the use of comparative
optimal to modify the reward scheme to account for
the
projects,
all
may
>
small relative to the agents' risk aversion, consensus decision-
and empowerment, but consensus leads
payoffs. For small
which means
point, with vi
and empowerment saves the cost k on 5's projects.
making may be preferred even
for
regime of ^ authority
consensus involves incurring k on
less distorted,
some
vo is large
and k
is
its
grow more important, so
effect
on decision
that
quality. Starting
small, recall that consensus decision-making
it
is
from
may
be
grow more important
preferred. In that case, as decisions
commission
improve decision
rates to
quality. This will
weaker
less negative. In turn, this leads to
the optimal
scheme
will
moderate the
be implemented by making ai and ^\
(lower a\ and
effort incentives
/32).
because the
variance of payments mcreases.
we
Finally,
making
is
very important and et
is
B now must
decisions. Because agent
set
e\^ is large
low
levels of
enough
)S|
is
induced. Since
there
is
no reason
aivo/(v'o+V2).
to
^_,
1
B
and
be exposed to the risk
is
given authority over
in £o if effort is to
project
be induced,
it
will
2.
As
Thus B
is
apparently
and yet do not "work,"
of ^'s decisions will be reviewed by B,
all
be induced on task
e\^ will be zero.
that are quite costly
fare well if decision-
does not receive any comparative
effort will
< ^n2'(e2*), then
that ^\Y\\{e\'^)
all
=
/3i
and ^, and not much
receiving effort incentives for task
B
not too important: agent
performance evaluation, but instead has
be best to
may
consider an alternative organizational design that
who
is
in that
well, if
no
effort
a perfect decision-maker,
moderate the use of comparative performance evaluation for ^: aj = -
Thus, allocating authority to agent
B
allows the firm to provide agent A with
pointed, narrow incentives that are intense but poorly aligned with overall firm value. Agent 5's
incentives are moderated and his effort
is
reduced, but his low-powered incentives are aligned
with overall value creation, making him a better decision-maker.
This organizational design achieves perfect decision-making and second-best effort on task
1
at the cost
projects
of reduced
effort
must be reviewed
general, however,
important,
we
will
it
is
on task 2 and
at cost k,
/3i
< ^,
decisions but incurs risk, while
^
/3i
=
fh-
So long as
reflecting the fact that Pi
is
by B. Each of agent A's
while agent 5's projects don't need to be reviewed. In
not optimal to set
have
inefficient risk bearing
needed
to induce effort
8
is
effort
on task 2
is
somewhat
used only to induce balanced
on task
2.
Thus decision-making
is
biased: V2
making
3.
is
is
overvalued relative to
This organizational design will be optimal
vi.
important enough, effort on task 2
if
k
make
agents, to
agents
is
very important
small enough.
is
to tasks
1
and
agents
C
is,
not too large,
optimal. Assign
C
makes
perfect decisions, while
is
characterized by a
who
"CEO" who
receives a small share of
receive comparative performance evaluation.
specializes in project evaluation and investment decisions, while the division
exert high levels of effort on
Finally,
when
and give them the optimal incentives from the
hi this case, agent
overall performance and "division managers"
CEO
is
receive the effort incentives that are optimal in the absence of decision-making
considerations. Thus, the organization
The
of hiring an agent
to give all authority
comparative performance evaluation). Give authority to
and pay him y{x\+X2), where y>0.
A and B
can be useful
be optimal to use a third agent, C, identical to the other
2, respectively,
perspective of effort allocation (that
agent
may
if the cost
it
The following organizational design may then be
decisions.
A and B
it
small enough,
is
an agent with balanced incentives. Thus,
decision-making
to
not too important, and k
decision-
Hiring a Third Agent
This logic suggests that in general,
to
is
if
it is
managing
managers
their respective divisions.
fmds
natural to consider the possibility that such a third agent
who
monitor project returns than the agents
exert effort in tasks
k'>k to monitor projects. Consider a regime where ai>0,
decision rights are allocated as follows: Agents
1
^>Q,
and
2:
|o;2|<
more expensive
Agent
and
CC\
A and B can implement,
it
C incurs a cost
|jSi|
< ^, and
without review, any
projects that both agree on, and they can also individually call projects to the attention of agent
C. In other words,
Under
C reviews
all
these assumptions,
desirable effect:
Any
projects
if
where A and
k and
k'
B
disagree.
are small enough,
conflicting incentives have a
value-enhancing project that A opposes will be taken to
C by
agent B, and
vice versa. Thus,
good
all
projects
come
to the attention
of agent C, who correctly evaluates
them. This will prevent good projects from being missed.
incentives for
A and
5, however,
is
that too
many bad
review. However, anticipating rejection by agent C,
rejection of projects that increase A's
for
A and B
wage
to protest exactly the subset
hurt the other. Finally, observe that since
allowing
A and B
to
A
potential cost of unbalanced
projects might be brought forward for C's
A
will not see
any gain from appealing 5's
but decrease total value.
Then
it
is
weakly optimal
of value-maximizing projects that help one agent and
A and B
agree only if the projects increase total value,
implement projects when they have consensus economizes on C's review
costs.
Summarizing,
profitable for
is
it
A and B
where the agents
IBM
if
to
more
costly for agent
to evaluate decisions than
review one another's projects, and for agent
disagree.
A
scheme with these main
C
A and
5,
it
may be
to review only projects
features has been used, for example, at
Corporation, where an operating unit had to get the concurrence of other such units to
implement
projects.
If
concurrence was withheld, the decision was passed to a
hierarchic superior (Richard F. Vance, Arvind
4.
C
common
Bhambri and James Wilson, 1980).
Extensions and Conclusions
Our model has endogenized
of decision
rights, generating (in
however: to characterize
fially
the simultaneous design of incentive
some
schemes and allocation
cases) an authority-based hierarchy.
Major gaps remain,
the optima, to allow the agents to choose whether to incur the
costs of evaluating others' projects, to allow agents to suppress projects they discover
whose
adoption would harm them, and to introduce a cost of discovering projects. These are addressed
in
Athey and Roberts (2001).
10
References
Aghion, Philippe and Tirole, Jean. "Formal and Real Authority
Political
Economy, February 1997, 705(1),
Journal of
pp. 1-29.
Athey, Susan and Roberts, John. "Authority and hicentives
Stanford
in Organizations."
in
Organizations."
Mimeo,
GSB, 2001.
Dewatripont, Matthias and Tirole, Jean. "Advocates,"
Joz/r/ta/
of Political Economy, February
1999,707(1): 1-39.
Fama, Eugene
F.
and Jensen, Michael
Law and Economics,
Hart,
Oliver
C. "Separation of Ownership and Control." Journal of
June 1983, 26(2), pp. 301-25.
and Moore, John. "On
Specialization,"
Design
the
NBER Working Paper No.
of Hierarchies:
Coordination
versus
7388, 1999.
Holnistrom, Bengt. "Moral Hazard and Observability," The Bell Journal of Economics, Spring
1979, 70(1), pp. 74-91.
Holmstrom, Bengt and Milgrom, Paul. "Regulating Trade among Agents." Journal of
Institutional
,
and Theoretical Economics, March 1990,
146{\), pp. 85-105.
"Multitask Principal-Agent Analyses: Incentive Contracts, Asset Ownership, and Job
Design." Journal of Law, Economics and Organization, Special Issue 1991, 7(0), pp. 24-52.
Jensen, Michael C. and Meckling, William H. "Specific and General Knowledge, and
Organizational Structure," in L. Werin and H. Wijkander, eds.. Contract Economics. Oxford:
Blackwell Publishers, 1992, pp. 251-74.
MacDonald, Glenn and Marx,
Leslie.
"Adverse Specialization," Journal of Political Economy,
forthcoming.
11
Milgrom, Paul and Roberts, John. "Bargaining Costs, Influence Costs, and
Economic
Activity," in
J.
E. Alt
and K. A. Shepsle,
Economy. Cambridge: Cambridge University
,
"The Efficiency of Equity
Review,
May
eds.. Perspectives
the Organization of
on Positive Political
Press, 1990a, pp. 57-89.
in Organizational
Decision Processes," American Economic
1990b (Papers and Proceedings), 80(2), pp. 154-159.
Simon, Herbert. "A Formal Theory of the Employment Relationship," Econometrica, July
1951, 79(3), pp. 293-305.
Vance, Richard
F.,
Bhambri, Arvind and Wilson, James. "IBM
Incident" Harvard Business School case 9-180-042, 1980.
12
Corp.:
The Bubble-Memory
Notes
Athey: Department of Economics, Massachusetts Institute of Technology, E52-252C, 50
Memorial Drive, Cambridge,
MA
02142
(e-mail:
athey@mit.edu) and, for
AY
2000-2001,
National Fellow, Hoover Institution; Roberts: Graduate School of Business, Stanford University,
Stanford,
CA
94305-5015 (e-mail: roberts_john@gsb.stanford.edu).
MacLeod and Michael Riordan
'
it
may be
nonlinear contracts that induce a complementarity between x\ and
face the
^
model would change
same basic tradeoff as
in the current
.xt.
profitable to use
Although many of the
were allowed, we would
effort.
This might occur
if V2
were
B
is
premium) of giving agent incentives
given a positive P\ and
large, so that
it
is
We will
fi's
is
vo there is
induced to supply both sorts of
very expensive to induce a
with a low marginal cost of effort. In this case, however,
n, fiinctions are similar, so that
still
modeling.
second task using a second signal (Holmstrom and Milgrom, 1991). For small
another possible solution in which
B
if nonlinearities
Specialization eliminates the fixed cost (through the risk
for a
are gratefiil to Bentley
for their suggestions.
Following the logic of MacDonald and Marx (forthcoming),
specific predictions of the
We
we would
lot
actually need
of e^, leaving
P\>
^ if the
investment preferences over-weight the easily measured task.
largely ignore this possibility in
what follows.
13
56^5 n29
Date Due
MIT LIBRARIES
^^
^^^^
3 9080 02246 2482