Laffont and Tirole 1991

THE POLITICS OF GOVERNMENT DECISION-MAKING:
A THEORY OF REGULATORY CAPTURE*
Downloaded from https://academic.oup.com/qje/article/106/4/1089/1873396 by Queen Mary University of London user on 22 November 2023
JEAN-JACQUES LAFFONT AND JEAN TIROLE
The paper develops an agency-theoretic approach to interest-group politics and

shows the following: (1) the organizational response to the possibility of regulatory
agency politics is to reduce the stakes interest groups have in regulation. (2) The
threat of producer protection leads to low-powered incentive schemes for regulated
firms. (3) Consumer politics may induce uniform pricing by a multiproduct firm. (4)
An interest group has more power when its interest lies in inefficient rather than
efficient regulation, where inefficiency is measured by the degree of informational
asymmetry between the regulated industry and the political principal (Congress).
I. INTRODUCTION
A major task of economics and political science is to explain the
pattern of government intervention in industries. Two main
theories have been proposed to this effect. The "public interest"
theory emphasizes the government's role in correcting market
imperfections such as monopoly pricing and environmental exter-
nalities. While regulatory agencies may face informational con-
straints, they are viewed as benevolent maximizers of social
welfare. Almost all of the theoretical work on the regulation of
natural monopolies,' for instance, has embraced the public interest
paradigm. The "capture" or "interest group" theory emphasizes
the role of interest groups in the formation of public policy. Its
origin can be traced back to Marx's view that big business controls
institutions and to the early twentieth century political scientists.
Stigler's work [1971] considerably extended the paradigm by
noting that the regulatory process can be captured by small
business industries as well, and by using Olson's [1965] theory of
*The authors are grateful to Barry Weingast for helpful discussions, and to
Patrick Rey, Eric Maskin, the participants at the Harvard-M.I.T. political economy
seminar, and three referees for useful comments. The Guggenheim Foundation, the
Ford Foundation, the Pew Charitable Trust, the Center for Energy Policy Research
at M.I.T. and Harvard University provided financial support for this research.
1. See the literature on Ramsey pricing (e.g., Boiteux [1956] and Baumol and
Bradford [1970]), on contestable markets (e.g., Baumol, Panzar, and Willig [1982])
and on the agency approach to regulation (e.g., Baron and Myerson [1982] and
Laffont and Tirole [1986]). An exception is the agency approach in the Demski and
Sappington [1987] paper, in which an agency must be given incentives to exert effort
to acquire information about the industry. Yet the Demski and Sappington
contribution is in the vein of the public interest literature in that social welfare is
maximized conditionally on the agency's information—there is no regulatory
capture.
© 1991 by the President and Fellows of Harvard College and the Massachusetts Institute of
Technology.
The Quarterly Journal of Economics. November 1991
1090 QUARTERLY JOURNAL OF ECONOMICS
collective action' as a building block to explain how "regulation is
acquired by the industry and is designed and operated primarily for
its benefit" [p. 3]. 3 Olson's logic of collective action implies that, for
a given issue, the smaller the group, the higher the per capita stake,
and therefore the incentive of its members to affect the regulatory
outcomes. Stigler inferred that members of an industry have more
incentives than dispersed consumers with a low per capita stake to
organize to exercise political influence. The emergence of some
powerful consumer groups and the regulatory experience of the
seventies led Peltzman [1976] and the academic profession to take
a broader view of Stigler's contribution, that allows government
officials to arbitrate among competing interests and not always in
favor of business.'
The positive models developed in the last two decades by the
Chicago school (Stigler, Peltzman, and Becker) and the Virginia
school (Tollison and Tullock) suffer from two methodological
limitations. First, they are not agency theoretic in that they ignore
informational asymmetries. In the absence of such asymmetries,
regulated firms would be unable to extract rents and therefore
would have no incentive to influence regulatory outcomes. Simi-
larly, voters and legislators would have no difficulty controlling
their agents (members of committees and agencies) who thus could
not get away with policies favoring interest groups over the
common good. In contrast, an agency-theoretic framework can
explain why regulators have discretion and why interest groups
have stakes and power. Second, the Chicago and Virginia schools
have focused on the "demand side" in their study of political and
regulatory decision-making, in that all the action takes place on the
side of interest groups. By "blackboxing" the "supply side" (the
political and regulatory institutions), they have ignored a crucial
agency relationship between politicians and their delegates in the
bureaucracy. This paper brings together the demand and the
supply side in an agency-theoretic framework.
Interest groups try to capture government decision-making
because it affects the industry and the consumers' welfare. Interest
groups have means to influence public decision makers: (a) mone-
2. See also Buchanan [1965].

3. Stigler also offered the view that there is a market for regulation, in which
outcomes are determined by supply and demand. See Peltzman [1976] and Becker
[1983, 1985] for complete information voting models of regulatory behavior.
4. For reviews of the capture argument, see the second part of Moe [1986], Noll
[1983, 1985], Posner [1974], and Wilson [1980].
POLITICS OF GOVERNMENT DECISION-MAKING 1091
tary bribes are feasible, although not common.' (b) More pervasive
are the hoped-for future employment for commissioners and
agency staff with the regulated firms or their law firms or with
public-interest law firms.' (c) Personal relationships provide incen-
tives for government officials to treat their industry partners
kindly.' (d) The industry may cater to the agency's concern for
tranquility by refraining from criticizing publicly the agency's
management. (e) Last, but not least, the industry can also operate
indirect transfers through a few key elected officials who have
influence over the agency. These include monetary contributions to
political campaigns (Political Action Committees)," as well as the
votes and lobbying of the "Grass Roots" (employees, shareholders,
suppliers, and citizens of communities where plants are located).
Such attempts at capturing the supervisory decision-making
through collusive activities are likely to be only the "tip of the
iceberg" [Tirole, 1986]. That is, the hidden and bigger part of the
iceberg is the organizational response to prevent collusion, in this
case the rules and policies whose raison d'étre is the potential for
regulatory capture, and their effect on industry performance.
5. Note that some monetary bribes are legal, however. For instance, the U. S.
Defense Department directive 55007 allows gratuities when they are a part of a
"customary exchange of social amenities between personal friends and relatives
when motivated by such relationships and extended on a personal basis"
Adams, [1981 p. 177].
6. Breyer and Steward [1979, pp. 141 42] and Adams [1981] contain extensive
-
descriptions of the "revolving door" phenomenon. Two quotations from Adams [pp.
82-83] illustrate the point nicely: "The availability of jobs in industry can have a
subtle, but debilitating effect on an officer's performance during his tour of duty in
procurement management assignment. If he takes too strong a hand in controlling
contractor activity, he might be damaging his opportunity for a second career
following retirement. Positions are offered to those officers who have demonstrated
their appreciation for industry's particular problems and commitments" (former
Assistant Secretary of Defense J. Ronald Fox); and "The greatest public risks
arising from post-employment conduct may well occur during the period of
Government employment, through the dampening of aggressive administration of
Government policies" (New York Bar).
Post-employment restrictions are costly because of the tight market for
managerial expertise in industries [Breyer and Steward, pp. 142-44].
7. The full circle revolving door between government and industry is obviously
conducive to the development of such relationships. The 1978 U. S. Ethics in
Government Act aimed at restricting post-employment contacts between former
top-level administrators and their former agencies. But as Warren [1982, p. 205]
notes: "Conflicts of interest laws are virtually impossible to enforce unless
governmental employees flagrantly violate them." (On this, see also Adams [1982,
p. 79].) Contacts occur in various manners, including committees between govern-
ment and private sector representatives; for instance, there were 820—mainly
closed doors—committees in the defense sector in 1979 [Adams, 1981, p. 165].
8. See, e.g., Adams [1981, pp. 8, 9] for a list of political contributions by defense
contractors to the members of the Senate and House Defense Appropriations and
Armed Services Committees.
Our model depicts the regulation of a natural monopoly (or
alternatively a cohesive industry). The regulatory structure regu-
lates the firm's rate of return and price. The firm (the "agent") has
private information about a technological parameter and chooses
an unobserved level of cost reduction. Its private knowledge of
technology allows it to enjoy an informational rent. The regulatory
structure is two-tiered: agency (the "supervisor") and Congress
(the "principal"). In contrast to Congress, the agency has the time,
resources and expertise to obtain information about the firm's
technology. 9 Congress relies on information supplied by the agency.
The agency's expertise allows it to hide information from Congress
in order to identify either with the industry or with consumer
groups affected by the price (output) decision. That is, these
interest groups can bribe the agency to retain specific kinds of
information. To keep the model tractable, we assume that a
monetary equivalent of $1 received by the agency costs $(1 + X 1 ) to
interest group i. The shadow price of transfers X 1 has two facets:
first, it reflects the fact that transfers to an agency are not fully
efficient. (A monetary bribe exposes the parties to the possibility of
legal sanctions; government officials would prefer to receive the
monetary equivalent of entertainment expenses; catering to spe-
cific interests goes against the agency's concern for "public ser-
vice;" etc.). Second, it embodies organizational costs. While the
latter are likely to be small for a firm, they may be substantial for
consumers; following Olson, one would expect small consumer
groups with a high per capita stake to have a smaller cost of
organizing than the group of all taxpayers, for instance. The legal
environment (Ethics Acts, appropriations for intervenors pro-
grams) and other "exogenous" variables (rise of consumerism or of
environmental awareness) affect the transfer costs and the relative
influence of the interest groups. 1°
9. For instance, Barke and Riker [1982, p. 77] note that: "Administrators
within a particular system are, however, full-time employees, devoting all their
professional attention to the rules and cases before them. Their role renders them
better informed than legislators and at the same time wholly identifies their
interests with the condition of the regulatory scheme." This view is shared by
Warren [1982, p. 51]: "Bureaucracy, as Max Weber and other organizational
theorists have recognized, is able to maintain its power position, despite challenges,
because the bureaucrats are able to make themselves the real experts by keeping
and controlling virtually all of the information. . ." and by Breyer and Stewart
[1979, p. 144]: "At present, Congress usually gets only an agency's official view of its
activities—a view which may filter out unfavorable, though potentially important,
information."
10. Most of our results still hold when, more generally, the maximum amount
of resources than can be channelled to the agency by interest group i when the latter
has stake A, in the agency's decision, can be written as pi(XA ) with < 0, po > 0,
,
In order to use standard agency methodology, we assume that
side contracts between the agency and the interest groups are
enforceable. Side contracts should not in general be thought of as
being enforced by a court (and therefore might be best labeled
"quasi-enforceable"). Rather, enforcement comes from the parties'
willingness to abide by their promise to cooperate. This interpreta-
tion may cover a spectrum of cases. First, the parties may pledge
their word and be loath to cheat on agreements with other parties
even in a one-shot relationship ("word-of-honor" case). Second, a
variation on this theme (not formally equivalent to the word-of-
honor hypothesis, but having the same flavor) is that the agency
and the interest groups over time develop reputations for not
breaching side contrasts even if they have no aversion to cheating
on the agreements (see Tirole [1990] for examples). Third, one may
have in mind situations in which the benefits from a collusive
agreement accrue as a flow and in which adherence to the
agreement is insured by the threat that the flow of benefits and the
associated flow of side transfers will stop if anyone cheats on the
agreement. The enforceability-of-side-contracts hypothesis is a
good description of cases in which collusion works well (such as the
idealized word-of-honor and self-enforcement interpretations); it
does not do full justice to intermediate cases (such as the reputa-
tion interpretation) in which collusion is feasible but not fully
effective. However, we do think that our analysis sheds light on
such intermediate cases, because it focuses on when collusion is
likely to be an issue and on how agency discretion can be reduced to
prevent collusion.
Congress has the means to reward or punish the agency.' It
and < 0. Here pl(kA,) = / (1 + X) But one can think of other functional
forms. For instance, if there are n members in the interest group, and there is a fixed
per capita cost f of collecting funds, the resource function might be pi = — nf.
Defining X, = n yields a function that satisfies the assumptions above.
11. In the United States, Congress can abolish or reorganize an agency, change
its jurisdiction, cut its appropriations, and conduct embarrassing investigations.
Weingast [1984] and Weingast and Moran [1983] have shown in specific instances
that Congress has a substantial influence on agencies.
The focus on Congress as the external monitor may be a good first approxima-
tion in the United States. The President has theoretical, but small actual, control
over the bureaucracy [Fiorina, 1981], and Courts are often limited to the punish-
ment of clear deviations from vague legislative mandates and are also constrained to
taking universalistic decisions [Warren, 1982].
Note that there is no conflict between the observations that "monitoring and
sanctions do not comprise a perfect solution to the problem of bureaucratic
compliance" [McCubbins et al., 1987, p. 253], and studies showing that agencies
tend to be obedient to Congress (e.g., Barke and Riker [1982], Joskow [1972], and
McFadden [1976]). In our model Congress can dictate regulatory policy, but is
dependent on the agency for information.
maximizes a social welfare function that adds consumer, agency,
and producer surpluses. The assumption that Congress is a
benevolent maximizer of a social welfare function is clearly an
oversimplification, as its members are themselves subject to inter-
est-group influence. There are three justifications for making this
assumption. First, ignoring the politics of Congress and focusing
on the politics of the agency is a first step toward a more general
theory of regulatory politics; yet it allows the derivation of a rich
set of insights. Second, the model may admit alternative interpreta-
tions; in particular, the "agency" in the model might represent the
coalition of a government agency and the members of the relevant
congressional oversight committee, and "Congress" the rest of the
legislature. Third, and most important, our methodology can be
straightforwardly applied to cases in which Congress does not
maximize social welfare but tries to control the regulatory out-
come. Our model is thus mainly one of control of agencies by their
political principals. 12
Because interest groups have a stake in the agency's behavior,
Congressional oversight of the agency and the industry must thus
respond to the potential for collusion between the agency and the
interest groups.
This simple model permits the study of several central issues
in the theory of regulation: (a) the determinants of interest-group
power (an interest group has power if its potential for organizing
triggers a regulatory response. As we shall show, because of the
latter response, an interest group may be hurt by its own power);
(b) the effect of regulatory politics on the agency's incentive
structure and discretion (discretion is measured by the sensitivity
of regulatory decision to agency reports); (c) the effect of regulatory
politics on the regulated firm's incentives and rent, and on pricing;
(d) the dependency of these effects on the power of interest groups
and on the amount of resources appropriated to the agency; (e)
whether interest groups' pressures offset or add up, and how
interest groups affect each other's welfare (does an improved
organization of consumers hurt or benefit the industry?).
Section II introduces the model. Sections III through V
consider the case in which production is essential (the firm cannot
12. It cannot, however, explain rules that constrain the regulatory process and
decision-making (such as the definition of the scope of regulation, the limitations on
transfers to the industry, etc.). Restraining the choice set of a benevolent Congress
can only reduce welfare in our setup. In contrast, in the absence of any benevolent
party, it may pay to design regulatory institutions so as to limit the regulatory
structure's scope of authority. See Laffont and Tirole [1990b].
be shut down), and solve the model in an increasing order of
generality: Section III considers the benchmark in which interest
groups are powerless, Section IV studies "producer protection,"
and Section V allows multiple interest groups. Section VI discusses
the case in which the firm can be shut down when it has an
inefficient technology. Section VII proposes a political theory of
cross-subsidization in the spirit of the previous sections, and
Section VIII summarizes the main economic insights.
II. THE MODEL

We consider a three-tier hierarchy: firm /agency/ Congress.'
All parties are risk neutral.
A. Firm
The firm produces output q at cost,
(1) C = (13 — e)q.
The cost or technology parameter p can take one of two values:
"low" or "efficient" (p) with probability v and "high" or
"inefficient" (130 with probability (1 — v). The firm knows the
realization of 13. Let A p --,-- 173 - A > 0. The firm's managers incur an
increasing and convex (monetary) disutility ce) (41' > 0,i" > 0)
by exerting effort e to reduce cost. For technical reasons, we
assume that kr ' , 0. 14
The gross consumer surplus is denoted by S (q), an increasing
and concave function. Let P (q) E-- S' (q) be the inverse demand
-
function. The revenue is thus P (q )q. Let

dq I q
11(P) a — dpI p
. —
describe the elasticity of demand. Our accounting convention is

that Congress pays the firm's cost and receives the revenue.
Letting t denote the (net) transfer from Congress to the firm, the
firm's utility or rent is
(2) U = t — ige).
We normalize the firm's reservation utility at 0, so that the
13. The model builds on Laffont and Tirole [1986] and Tirole [1986].
14. This assumption implies that the optimal incentive schemes are nonstochas-
tic. It is not used in the derivations of comparative statics exercises.
firm's participation or individual rationality constraint is
(3) U O.
B. Agency
The agency receives income s from Congress and derives
utility from its relationship with Congress: V(s) = s s* . That is, —
its reservation income (that is, the income under which its
employees refuse to participate) is s* . For simplicity, we assume
that the agency is indispensable (that is, Congress needs the agency
to regulate the firm's price and cost). Thus, Congress must pay at
least s* to the agency in each state of nature:
(4) V(s) = s — s* _. O.
(The remark below discusses this ex post formalization of the
agency's individual rationality constraint.)
The agency obtains information (a signal cr) about the firm's
technology. With probability the agency learns the true p (0. = 13);
with probability (1 — 0 the agency learns "nothing" (cr = (1)).
There are thus four states of nature: with probability v the
technology and the signal are A; with probability (1 — )v the
technology is A, but the agency does not know it, and therefore still
puts probability v on the firm's being efficient; etc. The signal is
hard evidence in the sense that the agency is able to reveal the true
technology to Congress if if = 13. For simplicity, we assume that the
interest groups (firm, consumer groups) learn what signal the
agency receives.' Note also that is exogenous (so in particular we
take the agency's effort to discover the technology as given); can
be thought of as entirely determined by the agency's budget for
investigation.
The agency reports r E {cr,(1)} to Congress. That is, if it has
learned nothing (o- = (I)), it can only say so (r = 4)). If it has learned
the truth (o = (3), it can either tell the truth (r = 1) or claim its
search for information was unfruitful (r = .43.).
C. Congress
As discussed in the introduction, Congress's utility is the sum
of producer, agency, and consumer surpluses:
(5) W = U + V + [S (q) — P (q)q
— (1 + X)(s + t + (i3 - e)q — P (q)q)1,
15. Alternatively, one could assume that, when the agency has an incentive to
collude with an interest group, it can go to this interest group and disclose the signal
it has received.
where X is the shadow cost of raising public funds through
distortionary taxation. Using (2) and (4) to eliminate t and s in (5)
yields
(6) W = [S (q) + XP (q)q1 — (1 + X)
x (s* + (13 — e)q + i(e)) — XU — XV.
That is, from the "generalized consumer surplus" (S (q) + XP (q)q)
must be subtracted (1 + X) times the total cost of the project
(s * + (13 — e)q + ii(e)) and X times the rents left to the firm and the
agency. The important property of (6) for our analysis is that the
Congress dislikes leaving a rent to the firm and to the agency. Note
also that W does not incorporate any deadweight-loss associated
with side transfers. It turns out that (except in Section VII) optimal
contracts can be designed so as to leave no scope for side transfers:
see Appendix 1.
Congress observes neither 13 nor ff. It observes the cost C and
the output q (or the price p = P (q)) and receives the agency's
report r. Congress designs incentive schemes s (C ,q ,r) and t (C ,q ,r)
for the agency and the firm so as to maximize expected social
welfare EW (where expectations are taken over the four states of
nature). 16
The timing is as follows: at date 0 all parties learn their
information simultaneously. That is, they all learn the nature of
the project; Congress learns that 13 belongs to 1A,1.1; the agency
learns ff, and the firm learns 13. The probability distributions are
common knowledge. Then Congress designs incentives schemes for
the agency and the firm. The agency can then sign side contracts
(see below) with the interest groups. Next, the agency makes its
report, and the firm chooses its effort and price (the exact timing in
this stage turns out to be irrelevant). Last, transfers are operated
as specified in the contracts.
REMARK. The formulation implicitly assumes that the project is
too ill-defined before date 0 for the parties to be able to sign
relevant contracts before that date. Alternatively, we could
assume that the project is well defined before date 0 so that the
parties can sign complete contracts before obtaining their
information, as in Tirole [1986]. Most results (on the effect of
collusion on incentive schemes, on pricing, and on the circum-
stances under which an interest group has power) are qualita-
16. More complex mechanisms (including, e.g., announcements by the firm

and the agency) would not raise welfare in this model with collusion. See Appendix
1.
tively unaffected if the firm and the agency are risk averse and
attention is restricted to deterministic contracts; the differ-
ence is that the agency and the firm then have no ex ante rent,
unless an ex post no-slavery or limited-liability constraint is
imposed. The analysis is a bit more cumbersome than in the
case in which there is no contract prior to date 0, except when
the agency and the firm are infinitely risk averse (see Section
IV).
D. Consumer Groups
When consumers cannot organize (Sections III and IV), it does
not matter how the net surplus (S (q) — P (q )q) and the taxes
(( 1 + X)(s + t + C — P (q )q)) are allocated among consumers. In
contrast, when they can influence policy decisions (Section V), the
distribution of costs and benefits among consumers becomes
important, as consumers have different marginal rates of substitu-
tion between consumption of the good and taxes. To simplify
computations without losing insights, we shall assume in Section V
that there are two groups of consumers: one that pays all taxes and
another that receives the entire net surplus. Let us give three
examples: (1) q is the output of an intermediate good used by
another industry, or else the output of a final good consumed by a
small group of consumers; taxes are paid by the general taxpayer.
(2) q is the level of welfare benefits enjoyed by the poor; taxes are
paid by the rich. (3) q is the level of pollution or pollution
abatement that affects local residents; taxes are paid by the federal
taxpayers. (In these last two examples the objective functions must
be changed slightly as the good is nonmarketed, but this is
inconsequential.)
III. COLLUSION-FREE REGULATION

In this section we analyze the benchmark in which interest
groups have no influence on the agency (their transfer costs are
infinite). We sketch the solution, and summarize the relevant
points for subsequent analysis.
Congress optimally offers the agency a constant income equal
to its reservation income: s = s* . The agency then has no incentive
to misreport the signal. Hence, at social cost (1 + X)s *, Congress
has the same information structure as the agency.
Next we consider optimal regulation of the firm when Con-
gress has full information (FI) and asymmetric information (AI).
A. Full Information (a = (3)
Congress, who knows the firm's technology parameter, can
deprive it of its rent (we index variables by an asterisk to indicate
the socially optimal policy under full information). For all 13,
(7) U(13) = UN3) = 0.
The effort e*(r3) and output q* ((3) or price p* ((3) (which we
shall write as (e* ,q* ,p* ) for the efficient type and (e*,4 *,13 *) for the
- -
inefficient type) are set so as to maximize the full information

welfare, ((S (q) + XP (q )q) — (1 + X)(s* + (13 e)q + tiff (e))). Hence
—
for all 13,(eN3),p*((3)) solves'

(8) tifi' (e) = q
and
p — (0 — e) X 1
(9) — , orp ----- R(13 — e) .
P 1 + X Ti(p)
Equation (8) states that the marginal cost and benefit of effort
are equal. Equation (9) shows that price is given by a simple
Ramsey formula. The Lerner index (price-marginal cost margin) is
inversely proportional to the elasticity of demand (because public
funds are costly, revenue is socially valuable, so pricing is interme-
diate between marginal cost and monopoly pricing). R (c) is called
the Ramsey price for marginal cost c. The formula giving the price
as a function of marginal cost turns out to be independent of
informational asymmetries (see subsection B) below). 18 We let q* (e)
and q* (e) denote the Ramsey outputs given by (9) contingent on
marginal cost being § — e or 13 e. It is easy to show that q*() and
—
4 *() are nondecreasing functions of e. 19

--
Below we show that, under asymmetric information, the

efficient type's allocation is unchanged relative to symmetric
information (it is equal to (e* ,q*)). The inefficient type's output 4 is -
still conditionally optimal given the inefficient type's effort 6 (i.e., -
4 = 4*(0). The focus of the analysis will thus be how e differs from
- -
the full information level è* . This suggest singling out the ineffi-
-
17. In the whole paper we shall assume that optimization programs have
interior solutions.
18. This is a special instance of the pricing-incentives dichotomy (see Laffont
and Tirole [1990a]).
19. The proof of this is the same as the proof showing that a monopoly price is a
nondecreasing function of the monopolist's marginal cost, and uses a "revealed
preference argument."
cient type's effort for the purpose of the analysis. Let WF/ (e) denote
the expected social welfare (that is, social welfare when Congress
has not yet learned 13, but knows that it will do so before regulating)
given that the efficient type's allocation is at its full information
level, the inefficient type's output is conditionally (Ramsey) opti-
mal, but the inefficient type's effort is an arbitrary e:
wFI (e\ =
(10) ) v[(S(q*) + XP(q*)q*)
- - -
—(1 + X)(s* + (p. — e*)q* + tge*)))]

+ (1 — v)RS (q*(0) + XP (4*(e)).4*(e))
—(1 + X)(s* + (0 — 0 4*(e) + ic(e))].
-
We assume that W" (•) is strictly concave.' The expected social

welfare WF/ is
wFI , wFI (e-*)
(11) •
B. Asymmetric Information (cr = 4))

We let (e,q,t) and (e ,4,t) denote the efforts, output levels and
--
transfers for types § and 13 under the optimal incentive scheme

when the firm has an informational advantage over Congress. As is
easily seen, the regulatory issue is to prevent the efficient type from
claiming it is inefficient. That is, we must add an incentive
constraint to the full information program:
(12) t — ige) I — *(F — 6,13).
(The efficient type can produce at cost 1 — ë by exerting effort e —

643 and obtain transfer t.)
The inefficient type obtains no rent (t = 116)). The efficient
type's rent under asymmetric information will be denoted by U.
Because (12) is binding at the optimum, we have
(13) U t — tii(e) = 1 — *Ce — A13) = tO) — tI6 — Al3)
or
(14) U = (KO,
where
(15) (I)(e) a 4)(e) — 4)(e — 6,43).
-
20. Sufficient conditions are X small or decreasing marginal revenue. (See

Laffont and Tirole [1986].)
Under our assumptions the function (I) (which will play a crucial
role below) is increasing and convex. Note that when e increases
(i.e., when the inefficient type is given "more incentives"), the
efficient type's rent increases.
Congress maximizes expected social welfare:
(16) (qp,T,E)
max IvRS(q) + XP(q)q) — (1 + X)
——
x (S * ((3 — e)q + kji(e)) — X(I)(F)] + (1 — v)
x [S(4) + XP(4)4) — (1 + X)(s* + ([2. - 0 4 + iii(e))11.
-
A simple inspection reveals that q =q*, e = e* and q = ei*(e) as

announced. The absence of distortion of (q,e) and of conditional
distortion of 47 is not surprising: the incentive constraint (14) tells
us that only e should be distorted:
(17) (e) = Cr(e) — (Xv/(1 + X)(1 — v))(1)'(e),
with"
(18) <
Thus, the inefficient type's effort e is distorted downwards in order
to reduce the efficient type's rent (intuitively, if the inefficient type
were given a cost-plus contract, the efficient type would not get any
rent by mimicking the inefficient type).
Because the Ramsey output 4'0 is increasing, (18) implies
that
(19) q < 4*.
Let Wm (e) denote the expected welfare under asymmetric
information when the inefficient type's effort is exogenously fixed
at e (and (16) is maximized with respect to the other variables).
Note that
(20) WM(e) = wFr(e\) — v(I)(e).
(•) is strictly concave when VIM() is strictly concave.
The expected social welfare under asymmetric information can
21. The proof of (18) is a standard revealed preference argument. Recalling

} - 7
that q = q* and e = e* {e,4*(e)lyields a higher maximand in (16) than le*,4 *(e*)1.
I
Conversely, in the full information pro Tam (which is the same as (16) except that
there is no (—vX(130(e)) term, te*,-41*(e* yields a higher maximand than {e,C7*(e)}.
Adding up these two inequalities yields vX((13(e*) — (1)(e)) > 0 (the inequality is
strict, as maximands are strictly convex and (17) implies that ë e*). Because (1) (.)
is strictly increasing, e * > e.
be rewritten as
(21) wAl = max i wn (e,) — Xv(1)(e)} = wn (e)) Xv).
'
Let us summarize the relevant points for what follows. Congress

obtains the agency's information by giving it a constant income s *.
The expected social welfare under full and asymmetric information
can be written as strictly concave functions of the inefficient type's
effort e: W' (e) and Wm (e) = wFl (e)) — Xv1(e), where (1)(e) is the
efficient type's rent under asymmetric information and is an
increasing function. The optimization with respect to e therefore
implies that the inefficient type is given a less powerful incentive
scheme under asymmetric information (e < e*) in order to extract
some of the efficient type's rent. The corresponding market price is
higher than under full information.
IV. PRODUCER PROTECTION

In this section we allow the firm to collude with the agency.
More precisely, the firm can give a transfers to the agency (so that
the agency's income equivalent becomes s + §) at cost (1 + X f )§,
where X f > 0 denotes the shadow cost of transfers for the firm
(equivalently the agency attributes monetary value 1/(1 + X1 ) per
dollar of the firm's collusive activity): see the introduction for a
general discussion of transfer costs. We here content ourselves
with a heuristic derivation of the equilibrium outcome under
collusion with the firm. Appendix 1 offers a complete proof. In
particular, it shows that (a) Congress can without loss of generality
restrict attention to `collusion-proof"' schemes, i.e., that do not
induce the agency and the firm to collude and that lead the agency
to report truthfully. Hence, there is no welfare loss in requiring
that there be no equilibrium bribes; 22 (b) the agency's income
depends only on its report. We let s „-§„ and s o denote the agency's
income when r = j, r = 13, and r = 4), respectively.
Collusion occurs when the agency has an incentive to hide
22. Equilibrium collusion may be unavoidable in situations in which Congress

has incomplete information about the agency's cost of colluding. Suppose, for
instance, that X 1 = +co with probability 0.99 and X 1 = 0 with probability 0.01 (the
firm and the agency know Xp but Congress does not). Then Congress may find it
cheaper to let collusion occur with probability 0.01 (i.e., when X f = 0) than to ensure
that the coalition-incentive constraint (equation (22) below) is satisfied when Xf. = 0,
which yields the rent given by (24) to the agency even when the agency would not
have colluded anyway (X1 = + co). A similar analysis can be found in Kofman and
Lawarthe [1990].
information from Congress. The analysis in Section III suggests
the following intuition: collusion can arise only if the retention of
information benefits the firm. If the signal is IA, the firm has no
stake in the agency's report as it obtains no rent under either full
information or asymmetric information. In contrast, when the
signal is g, the firm has a stake, as the revelation of the truth
lowers its rent from t(e) (where e is the inefficient type's effort
under asymmetric information) to zero. To prevent the firm from
bribing the agency, the cost to the firm of compensating the agency
for the income (s 1 — s o ) lost by not reporting must exceed its stake:
(22) (1 + Xf )(s i — s o ) 1(e).
From the agency's individual rationality constraint, we know
that s i § i , s o all exceed s *. Because revelation is not an issue when
Q = 13 or u = szt), and because income given to the agency is socially
costly, we have -s- 1 = s o = s *. We can thus rewrite (22) as
(23) (1 + Xf )(s, — s*) 1(e).
Because income given to the agency is socially costly, (23) holds
with equality at the optimal policy:
(24) si = s* + (1)(e)/(1 + Xd.
Equation (24), which depends only on e and s 1 , suggests that
Congress should give lower incentives to an inefficient firm under
asymmetric information, but that it should leave the other vari-
ables (except s 1 ) unchanged; that is, the efficient type's allocations
under full and asymmetric information and the inefficient type's
allocation under symmetric information are still the socially opti-
mal ones (e* ,q*) and (e*, 4*). Furthermore, under asymmetric
-
information the inefficient type's output is the Ramsey level 4* (e)

relative to effort e. That these properties indeed hold is verified in
Appendix 1.
Congress chooses e so as to maximize expected social welfare:
(I)(e) 1
(25) EW = max
e 047
' + (1 — )147itt(e) — OA
1 + Xf '
where the last term reflects the fact that the agency's rent has
social cost X from (6).
Using the fact that the objective function in (25) is strictly
concave, the envelope theorem and the first-order condition in (25)
yield
PROPOSITION 1. Under producer protection:
(a) Collusion reduces social welfare (a(Ew) / ax f > 0).
(b) The firm is given a low-powered incentive scheme (e < e).
(c) Output is still Ramsey-optimal, but is lowered from 4. (e)
to rec (e) under asymmetric information for the inefficient
type.
(d) The agency is given an incentive scheme (s 1 > s = s 0 ).
(e) The efficient firm enjoys a lower rent than in the absence
of collusion (4)(e) < ((0).
(f) e (and therefore (1)(e) and q * (e)) increase with Xf .
To prevent collusion, Congress reduces the stakes, i.e., the
efficient type's rent under asymmetric information. To this pur-
pose, the inefficient type is given an incentive scheme under
asymmetric information that is even less powerful than the
corresponding scheme in the absence of collusion. Because the
other states of nature are unaffected, producer protection can only
reduce incentives in the potential advocate regime. Note also that
as 4* (•) is increasing, the price is higher, and the transfer to the
-
firm is lower than in _ the absence of collusion, under asymmetric

information and type 13.
REMARK. What happens if the project is sufficiently well defined
before date 0 so that contracts can be signed before date 0?
(Appendix 1 contains some of the details of the following
discussion.) As mentioned above, the results are quite similar
if the agency and the firm are risk averse and attention is
restricted to deterministic contracts. Suppose, for instance,
that before date 0 the firm and the agency have objective
functions min U and min V (they are infinitely risk averse).
Thus, U and V must be nonnegative for any realization of
uncertainty at date 0. The coalition incentive constraint is still
(22). What is modified relative to our analysis is the expression
of the social welfare function. At the optimum, min U = min
V = 0. Furthermore, ex post rents (U > 0, V > 0) have no ex
ante social value as they are not "enjoyed" by the parties.
Therefore, they have unit cost (1 + X) instead of X. So (25) is
replaced by
(26) EW = max { twFI + ( 1 (e) — tv(1 + X)

(1)(e)
x — (1 — t)v(1)(e)},
1 + Xf
where W"1 is defined by (11) and Wfle) is defined by (20).
Clearly, the results are qualitatively similar. The main differ-
ence is that the agency and the firm may enjoy an ex post rent,
but do not have any ex ante rent.'
V. MULTIPLE INTEREST GROUPS

To illustrate the effects of multiple interest groups, we now
consider a specification of our model in which the firm's output
affects the environment. Suppose that the total gross surplus
associated with a level of production q is equal to the gross
consumer surplus S (q), associated with consumption of the good,
minus an increasing and convex pollution damage D (q) borne by a
fraction of the population called "environmentalists" or "local
residents":
(27) S (q) = S (q) – D (q).
-
Suppose also that the environmentalists do not purchase the good

and do not pay the taxes associated with the regulation of the
industry (as discussed in Section II, this assumption simplifies
computations and does not affect qualitative results; what matters
for our theory is that the environmentalists' marginal rate of
substitution between output and taxes exceeds that of the rest of
the public). In a first step we shall assume that, among the
nonindustry groups, only environmentalists can organize. They
can transfers to the agency at cost (1 + X e ),q (so that the agency's
income equivalent becomes s + s + s, where s is the firm's transfer
to the agency). We assume that public collection of funds is more
efficient than the private collection of funds (where the latter cost
takes account of the inefficiency of transfers to the agency; see the
introduction): X e > X. This assumption allows us to focus on
collusion-proof incentive schemes. (The intuition for this property-
23. If the agency is very risk averse, and the firm is less risk averse, it may pay
for Congress to commit to leave an ex post rent U > 0 when the agency announces
r = J. The reason for this is that leaving such a rent relaxes the coalition-incentive
constraint, which becomes (1 + X f. )(s 1 — s o ) (I)(e) — U, which allows Congress to
reduce s 1 . Reducing s 1 is socially important because Congress cannot reduce the
agency's utility much in other states of nature if s 1 is large and the agency is very
risk averse. On the other hand, leaving a rent to tile firm is costly. But this cost is
small if the firm is not too risk averse, because Congress can reduce the firm's utility
in other states of nature while keeping the firm's individual rationality constraint
satisfied in expectation. So it may pay for Congress to somewhat renounce the
pursuit of the extraction of the firm's ex post rent. In this case, the agency acts not
only as a potential advocate for the firm, but also as an effective advocate.
which is proved in Appendix 2—is that if the optimal allocation
involved actual transfers from the consumers to the agency, it
would be socially cheaper to have Congress substitute for the
environmentalists and give these transfers to the agency. If X e < X.,
it may be optimal to let the agency be rewarded by bribes, as private
collection is more efficient than public collection. We believe that
the assumption X e > X is reasonable for developed economies,
where X is relatively small—of the order of 0.3 for the United
States from econometric studies.)
Again, we give an informal treatment. Complete proofs are
relegated to Appendix 2.
Congress must ensure that the agency colludes neither with
the firm nor with the environmentalists. Because Congress's
optimization program has more coalition-incentive constraints
than when X = + 00 social welfare cannot exceed the level obtained
,
for Xe = +x. 24 Wew show that environmentalists affect the regula-

tory outcome.
For intuition about which coalition-incentive constraints are
binding, it is useful to go back to Proposition 1.
When if = A, the firm has a stake in regulation. To preserve its
rent, it is willing to bribe the agency up to the level t(e), where e
denotes the inefficient firm's effort if r = 4). In contrast, the
environmentalists have no stake in the agency's report as output is
the same for both reports. Thus, the only coalition-incentive
constraint when o- = { is
(28) (1 + X1 ) (s 1 – s o ) (1)(e).
When if = (3, the firm enjoys no rent and has no stake in the
agency's report. In contrast, the agency's hiding its information
induces asymmetric information and reduces output (Proposition
1). Let (e,q) and (J4 ) denote the inefficient type's effort and output
when r = (I) and r = 13, respectively. We must add a second
coalition-incentive constraint:'
(29) (1 + Xe )(§ 1 – s o ) D(4) - D(q)•
The optimal policy implies that s o = s*. Therefore, (28) and
24. In our model a reduction in X e—through a better organization, the advent

of consumer activism, or government subsidies—always reduces welfare. If consum-
ers play a substantial role as watchdogs, i.e., if they bring information about the
industry and products and check the agency, a reduction in their organization costs
may improve social welfare: see Laffont and Tirole [1990b].
25. Note that this constraint does not in general define a convex set.
(29)(which hold with equality at the optimum) can be rewritten as
(30) s l = s* + 1(e)/(1 + Xt.)
and
(31) sl = s* + (D (4) — D (q)) I (1 + Xe ).
This suggests (and it can be verified) that e, q, and q- are distorted at
the optimal allocation so as to reduce the agency costs. More
precisely, let wn(4,0 and Wm (q ,e) denote the expected welfares
under full information and under asymmetric information when
the inefficient type's alloction is (4,e) and (q ,e) , respectively, and
when the efficient type's allocation is undistorted (q = q* ,e = e*).
Using (30) and (31), Congress maximizes expected social welfare:
(32) EW = max{tWn (4 ,ê) + (1 — OW' (q ,e)

{4,ê,q,e}
(1)(e)D(4) — D(q)1
—
1 + Xf
— — v)x i + . j.
We shall assume that the maximand in (32) is strictly concave (for
this, it suffices that X be small or that X e be large). A straightfor-
ward analysis of (32) yields
PROPOSITION 2.
(a) The environmentalists have an influence on regulation.
(b) s 1 and, when q > q, g i strictly exceed s o = s* .
(c) A decrease in Xe raises e, and therefore raises the firm's
rent (I)(e) . It lowers e. And it lowers q and raises q, and
therefore it reduces (4 — q) > 0 toward 0. 26
(d) A decrease in X1 decreases q, and therefore raises the
environmentalists' welfare.
The intuition behind Proposition 2 is simple. To relax con-
straint (29), Congress lowers q and raises q, so that the environmen-
talists' stake (D(4) — D(q)) in regulation is reduced. Because q
increases, marginal cost reduction becomes more valuable when
Q = sf• and 13 = p. Hence, e increases. The striking conclusion is that
the more powerful the environmentalists, the higher the firm's
rent! This is not altogether surprising. In this economy the firm
and the environmentalists are "objective accomplices" in that they
26. For X small enough, it may be the case that q = q (a corner solution).
both have a stake in making regulation inefficient. The firm wants
Congress to be uninformed to enjoy a rent. The environmentalists
want Congress to be uninformed to reduce output and thus
pollution. We shall see in Section VI that this coincidence of
interests heavily relies on the assumption that production is
essential. An increase in the environmentalists' power may well
hurt the firm if shutdown is a relevant option.
As we mentioned, environmentalists are powerful here be-
cause their interest lies in inefficient regulation. Note also that the
effects of multiple interest groups do not cancel, but rather add up.
Furthermore, as in Section IV the agency must be rewarded
for cooperating with Congress.
We have assumed that neither the consumers of the good nor
the taxpayers can organize (presumably because their per capita
stakes are too small). Let us now show that even if the consumers
of the good (henceforth, the "consumers") could organize, they
would have no influence on the regulatory outcome. Without loss of
generality we assume that the consumers enjoy net surplus S"(q)
S(q) — P(q)q and that they do not pay the taxes or bear the
pollution cost associated with the project. They have a cost of
transfer X, > X. We can now state
PROPOSITION 3. Whether the environmentalists can organize or
not, the consumers have no political power. That is, the
regulatory outcome is the same as if X were infinite (as given
by Proposition 2).
The proof of Proposition 3 is straightforward. Introducing the
possibility of collusion between consumers and the agency cannot
raise welfare, as the number of constraints facing Congress
increases. Conversely, suppose that Congress adopts the regula-
tory policy that is optimal when consumers cannot organize. When
o- = §, the output is at its socially efficient level q* , regardless of
whether the agency reports the truth (r = §) or not (r = 41)) . Hence,
the consumers have no stake in the report. When if = i3, the
consumers do have a stake. The output is 4 if the agency reports
the truth (r = 13) and q _^ 4 if the agency lies (r = 44)). Hence, by
bribing the agency to hide its information, the consumers can only
raise the price. Therefore, they have no incentive to bribe the
agency. 27
27. One might conjecture that the agency could extract a bribe from the
consumers by threatening them to hide the information g = 13. However, such a
The intuition behind Proposition 3 is that consumers favor
high outputs. Because asymmetric information between Congress
and the firm leads to low-powered incentives and hence to low
quantities (Section III), a high output requires full information.
But the potential power of consumers (as well as of other interest
groups) lies in inducing the agency to hide information from
Congress.
Last, we can consider what happens when taxpayers (who
want to minimize taxes (1 + X) (s + t + C — P(q)q)) can organize,
although their high cost of organization in many situations makes
this analysis irrelevant. We are unable to give a general character-
ization of whether taxpayers have influence on regulation. How-
ever, there is a case of interest in which the answer is straightfor-
ward. Suppose that the taxpayers and the consumers are the same
people so that they form a single group (with objective function
S"(q) — (1 + X)(s + t + C — P(q)q)). When a = .g, this group's
interest lies in rent extraction, i.e., in the truth being reported.'
Hence the group has _ no incentive to bribe the agency to misreport.
Similarly, when a = p, it can be shown that the group prefers that
the agency report the truth.' Hence, the taxpayers-consumers
group has no political power in this model.
Discussion
An important principle emerging from this section is that the
power of an interest group depends not only on its stake and its
transfer cost, but also on what kind of influence it wants to exert. An
interest group has more political power when its interest lies in
inefficient rather than efficient regulation, because the agency's
threat is not "subgame-perfect": when the day comes at which the agency must
report to Congress, the agency has an incentive to tell the truth, as "g i > s o from
Proposition 3.
Only in the case in which the agency can develop a reputation for being tough
(lose income to hurt consumers) can such a threat be effective. Such a reputation
might develop in organizations where the supervisor monitors a large number of
subordinates.
28. Consider the solution described in Proposition 2 (which includes that
described in Proposition 1 as a special case). As 4(e) = (1 + X1 s 1 s o ), s 1 s o <
)( — —
(I)(e) so that the total wage bill (s + t) is lower when the report is r = J (the cost and
the output are independent of the report).
29. Again, consider the solution described in Proposition 2. Because Congress
can always duplicate the outcome for r = 4 and p = 13 when r = 13, social welfare is at
least as high in the latter case as in the former. But the firm has no rent in either
case and the environmentalists prefer the former case to the latter from Proposition
2. Hence, the remaining group (consumers plus taxpayers) strictly prefers the latter
to the former.
discretion lies in hiding information from Congress and asymme-
tries in information makes regulation less efficient.
This principle can be transposed to other examples:
Pollution Abatement versus Production - Embodied Pollution.
Proposition 2 shows that environmentalists are powerful when
pollution is tied to production. We now show that they may have no
power in other circumstances. Let the firm's output be fixed at
some level q o . The firm can reduce its pollution level by an amount
q at abatement cost C = (i3 - e)q (which comes on top of a given
cost of producing q 0 ). C can be thought of as the cost of buying and
installing a new pollution-reducing technology. 0 here denotes a
technology parameter that affects the marginal cost of pollution
abatement, and e the effort to reduce the abatement cost. The
reduction in pollution yields benefits B(q) to the environment-
alists (B• is assumed increasing and concave). Ignoring the
constant cost of producing q o and the generalized consumer surplus
(S(q 0 ) + XP(q 0 )q 0 ), the social welfare function is
(33) W = B(q) — (1 + X)(s + C + ili(e)) — XU.
Replacing [S(q) + XP(q)q] by B(q), the analysis of Sections III and
IV can be directly transposed to the pollution abatement model.
However, the environmentalists have no power here, as they
resemble the "consumers" of the production-embodied pollution
model: their interest lies in high pollution abatements (high qs).
That the environmentalists have power in one case and not in
the other is not surprising. They favor inefficient regulation in the
production-embodied pollution model and efficient regulation in
the pollution-abatement model.
Welfare Benefits. Consider a two-class economy (rich-poor),
and suppose that the poor are the recipients of a quantity q of
welfare benefits financed by taxes on the rich. The poor have an
interest in efficient regulation, as the latter is conducive to higher
benefits, and therefore have less power than the rich, who save on
taxes when inefficient regulation limits the level of welfare benefits.
VI. SHUTDOWN OF THE REGULATED FIRM
The analysis in Sections III—V proceeded under the assump-

tion that the firm is essential. That is, it must produce even if it is
inefficient (has type (3). This is the case if the consumer surplus is
sufficiently large, so that Congress cannot run the risk of forgoing
production (shutting the firm down when it has type 13 and
allowing production by type (3 only). This sounds like a reasonable
assumption for many regulated firms. In some instances, however,
shutdown is a relevant option.
Shutting down type 13 is a simple policy in our two-type model.
Type A has now no rent because mimicking type 13 brings none.
Congress has full information on the technology conditionally on
the firm's choosing to produce. This implies that the optimal policy
in the collusion-free environment with the shutdown option is still
collusion-proof when the interest groups can organize.'
This, together with the results in Sections IV and V, implies
that the more powerful the interest groups (i.e., the lower their
transfer costs), the more attractive the shutdown policy is relative
to the no-shutdown policy.
The possibility of shutdown reinforces most of our insights.
For instance, in our model it corresponds to an extreme absence of
agency discretion. Furthermore, the shutdown of the firm can be
viewed as an extreme case of low-powered incentive scheme.
There is a result, however, that relies heavily on the essential-
ity of production. In Section V we observed that the better
organized the environmentalists, the higher the firm's rent. This
may not be so when shutdown is a relevant option. A decrease in
the environmentalists' transfer cost reduces the welfare associated
with the no-shutdown policy. So it may induce Congress to switch
to the shutdown policy, which annihilates the firm's rent.
VII. A POLITICAL THEORY OF CROSS-SUBSIDIZATION

Our methodology can be applied to study whether interest
group pressure may lead to cross-subsidization by a multiproduct
firm. To this purpose, we consider a variant of the model of Section
II, in which none of the types of cross-subsidizations listed in
Laffont and Tirole [1990a] applies. In this variant, cross-
subsidization may emerge as an optimal response to the political
activities of some customers of the regulated firm. (One may, for
30. The optimal shutdown policy consists in requiring that the firm produce
q = q* at cost C* = ([3 e* )q* and in giving transfer t * = kli(e* ) (the efficient firm
—
has no rent).
Note that the agency has no role in the two-type model under the shutdown
policy. With more than two types the agency would bring information that helps
Congress to distinguish those types that are not shut down. The features discussed
in this section would still be relevant in the many-type model as long as the
shutdown option is a relevant one.
instance, think of the captive coal shippers' successful fight against
unrestrained price discrimination by railroad monopolies.)
Suppose that there are two classes of consumers, i = 1, 2, with
identical demands. Let S(q,) and Sn(q,) = S(q,) P(q,)q, denote the
- - —
gross and net surpluses of class i, where PO is the inverse demand

function. We letii(p,) denote the elasticity of demand at pricep,.
The regulated firm's cost is
(34) C = ( v. - e) (q1 + q2) + d(q 2 — q i )x,
where x = 1 or —1 with equal probabilities and d > 0. The
parameter x indicates which category of consumers is cheaper to
serve (that is, the marginal cost of serving one category is 2d lower
than the marginal cost of serving the other category). The "cost-
differential parameter" d is common knowledge.
To simplify the analysis, we assume that the agency does not
learn x and therefore has no role (this involves no loss in insight:
see below). So Congress regulates the firm directly. To focus on
cross-subsidization, we assume that 13 is known to Congress in this
section.
If Congress knows that x = 1 (without loss of generality), from
Section III, the optimal regulation specifies Ramsey pricing:
p i — (3 - e — d) X 1
(35) Li a- — R i =--- ,
P1 1 + Xl(p 1)
p2 — (13 e + d)
— X 1
(36) — Ro —
L2— P2 ' 1 + X iri(p 2 )
Furthermore, the marginal disutility of effort is equal to marginal
cost savings:
(37) kiii(e) = q1 + q2*
R 1 and R 2 are the Ramsey terms. When we allow collusion, we shall
say that there is cross-subsidization of good 2 by good 1 if L 1 > R 1
and L 2 < R 2 . Note that (35) and (36) imply thatp 1 < P2) ql > (h and
L i = R 1 < L2 = R 2 .
From now on we assume that the firm, but not Congress,
knows x. Furthermore, x is "soft information." That is, the firm
cannot "prove" to Congress that x is equal to 1 or —1, but only
announce it (V; in other words, Congress knows that the firm
knows x, but cannot subpoena the firm to supply evidence that
substantiates its announcement X. (The softness of information is
not crucial to the analysis of cross-subsidization, but as we shall see
in Proposition 5, it introduces the possibility that actual bribes are
desirable in equilibrium.) Last, we shall assume for simplicity that
the consumers also know x. 31
If Congress does not know x and there is no collusion between
the firm and any group of consumers, the solution is unchanged, as
the firm has no incentive to misreport X. Indeed, lying about x
would only lead Congress to switch the roles of good 1 and good 2
and increase the firm's cost by 2 (q 1 — q 2 )d and therefore the firm's
effort by (2(q 1 — q 2 )d) I (q, + q 2 ) without any gain. In contrast,
suppose, for instance, that type 2 consumers can organize and
make a take-it-or-leave-it offer to the firm, when x = 1, to induce
the firm to announce that 5Z = —1. This in turn leads Congress to
quote a low price for good 2 and a high price for good 1, which
benefits type 2 consumers and hurts type 1 consumers.
Let us assume that the two classes of consumers have transfer
costs X, with X, > X 32 In a first step we assume that it is optimal for
Congress to structure incentives so as to prevent collusion. We
shall then relax this assumption and show that collusion proofness
is optimal only for a subset of parameters. To avoid collusion with
type 2 consumers when x = +1, the gain for type 2 consumers of a
misreport of x, Sn(q,) — Sn(q 2 ), must be lower than the extra
disutility of effort, iji(e + (2(q 1 — q 2 )d)I (q, + q 2 )) — tP(e), valued at
the transfer cost between the type 2 consumers and the firm; the
coalition incentive constraint is thus 33
2d(q 1 — q 2 ))
(38) (1 + X,)He + — kii(e)i Sn(qi) — Sn(q 2 )•
q1 + q2
Using the symmetry of the model, we look for a symmetric

solution. It is easily seen that the firm, which has no private
information about 13, enjoys no rent. Congress's optimization
program is
(39) m,a), IS(q i ) + S(q 2 ) — (1 + X)[(13 — e) (q1 + q 2 )
4
+ d(q 2 — q 1 ) + Ip(e)] + X[P(q1)q1 + P(q 2 )q 2 ]}
subject to (38).
31. The analysis is qualitatively the same when the consumers do not know x;
the main difference is that there is less incentive to collude, and therefore a lower
likelihood of cross-subsidization when the consumers have incomplete information
about x.
32. As before, assuming that X, _. Xis meant to rule out the possibility that side
transfers occur only because an interest group is a better collector of funds than
Congress.
33. This constraint does not define a convex set.
We can now state
PROPOSITION 4. For the solution to the collusion-proof program
(39) (assuming that x = 1; for x = —1, indices are permuted),
there exist d 1 > 0 and d,> d i (d, _^ +00) such that
(a) If d < d i , pricing is uniform (p 1 = p2 = p; q1 = q2 = q. The
values of p and q are intermediate between the ones that
prevail under symmetric information about x). Cross-subsidi-
zation occurs (L 1 < R 1 and L 2 > R 2 ).
(b) If d 1 < d < d,, price discrimination occurs, and the threat
of collusion is socially costly.
(c) If d > d 2 , the threat of collusion is socially costless (i.e., the
solution is given by equation (35)—(37)).
The proof of Proposition 4 is relegated to Appendix 3. An
interesting conclusion is that, for small d, the stakes in collusion
(Sn(q 1 ) — Sn(q 2 )) are not only reduced at the optimum, but totally
disappear. Congress imposes uniform pricing, an extreme form of
cross-subsidization. The intuition for this result is as follows. The
welfare loss due to collusion is at most of order d when d is small,
because Congress can adopt uniform pricing, which is collusion
proof and involves only a loss of order at most d. Hence, a policy in
which (q 1 — q 2 ) is not of order at most d is suboptimal, as it involves
a distortion relative to the full-information case that does not
converge to zero at rate d or faster. Now, consider the collusion-
incentive constraint (38). As a first approximation the left-hand
side is proportional to (q 1 — q 2 ) d, and the right hand side is of
-
order (q 1 — q 2 ). Hence the constraint cannot be satisfied unless

q1 = q2.
For a large d the firm's cost of lying is very large, and
constraint (38) is satisfied by the solution to (35) — (37) in the case of
d2 < +00. For an intermediate d the analysis is complex, and we
were not able to get specific results. This is due to the fact that
lowering the differential (q 1 — q 2 ) reduces the consumers' stake
(Sn(q 1 ) — Sn (q,)), but also makes it more costly for the firm to lie.
We now investigate the possibility that bribes be socially
optimal in equilibrium. (The following analysis has benefited from
discussions with Bengt Holmstriim, who, in another context,
suggested to us that it may be socially optimal to allow bribes
between two members of an organization who share soft
information.)
To see why bribes may be optimal, suppose that x = 1 and
there is price discrimination: q 1 > q 2 . The type 2 consumers are
willing to pay Sn(q 1 ) — Sn(q 2 ) to the firm. Let
Sn(q 1 ) — Sn(q 2 ) [( 2d(q, — q 2 ) )
(40) 6, I.—. — *e+ — li(e)1
1 + X, q1+ q2
denote the bribe that the firm must receive from type 1 consumers
to tell the truth if equation (38) is not satisfied (A > 0). It is in the
interest of type 1 consumers to bribe the firm to tell the truth if and
only if 34
(41) Sn(q1) — Sn(q 2 ) (1 + X)A,
or
1 e
(2d(qi — q 2 ) )
(42) + — *(e) 0.
(q1 + q2)
Note that for q 1 q 2 , (42) is satisfied. So, in particular, price
discrimination is feasible even for a small d. But there is a cost of
having type 1 consumers transfer A to the agency, equal to
(X — X)A. There is thus a trade-off between relaxing the collusion-
proofness constraint by having the type 1 consumers bribe the firm
and creating costly side transfers.
Congress must then choose between two regimes. The "no-side-
transfer regime" corresponds to A ^ 0, and has already been
studied. The "side-transfer regime" corresponds to A > 0. There is
no collusion-proofness constraint, and the social welfare function is
given by
(43)To IS(Q) + S(q ) — (1 + XX(13 — e)(q1 + q )+ d(q — q )

2 2 2 1
+ tge)) + X[P(q 1 )q, + P(q 2 )q 2 ]

(Sn(q i ) — Sn(q 2 ) ( 2d(q, — q 2 ) \
— (X, — X) — 41 e + + iii(e))},
1 + X, qi + q2
where the last term takes into account the reduction in Congress's
transfer to the firm in amount equal to the bribe received.
When X is close to X, Congress can reach almost the collusion-
free welfare in program (43), while it cannot in program (39).
Hence the side-transfer regime is optimal.
34. We are here envisioning an auction between the two groups of consumers.
The firm announces that $Z which is favorable to the highest bidder, where the bid of
the expensive-to-serve consumers is deflated by the extra disutility of effort
engendered by lying.
Fixing X, > X, when d tends to zero, the no-side-transfer
regime (which we know from Proposition 4, involves uniform
pricing) is optimal. To show this, it suffices to take the derivatives
of (43) with respect to q 1 and q 2 and to note that (q 1 — q 2 ) becomes
negative when d tends to 0+, while q 1 q 2 for constraint (41) to be
relevant. The intuition for this result can be obtained from (43):
choosing (q 1 — q 2 ) positive and of order d yields two gains that are
second order in d: cost savings (1 + X)d(q, — q 2 ) and cost for the
firm of lying = tiii(e)2d(q i — q 2 )I(q 1 + q 2 ), and imposes a first-order
loss = (X, — X)p(q i — q,)I (1 + X) due to an inefficient side-transfer.
PROPOSITION 5. Assume that X, > X.
(a) When X is close to X, it is socially optimal to practice price
discrimination and to let the consumers who are cheap to serve
bribe the firm.
(b) When d is small, uniform pricing and the absence of
side-transfers are optimal.
A striking conclusion is that equilibrium side transfers may
arise. Type i consumers are then used as a countervailing force to
type j consumers. Recall the collusion-proofness principle obtained
for the hard information model of Sections II—VI. The interest
groups could bribe the agency to report or misreport its piece of
hard information, but Congress could duplicate this bribe at a
lower transfer cost. Here, Congress does not know whether 5(' = 1 is
a true report because of the softness of information, while the
consumers are able to base their transfers on both the announce-
ment and the truth.'
Last, we have assumed that there was no agency. Alterna-
tively, one could assume that the agency colludes with the firm.
Suppose, for instance, that the agency learns x (soft information)
and announces it. While the outcome is similar to the one obtained
above, this more complex framework allows the possibility that the
consumers' side-transfers be directed to the agency rather than
directly to the firm.
VIII. CONCLUSION
This paper has shown that interest-group politics can be

comprehended in a tractable agency framework. Its general in-
35. This shows that the possibility of equilibrium bribes is linked with our
assumption that consumers know the true value of x. If consumers do not know x,
then the collusion-proofness principle holds as Congress can duplicate the consum-
ers' side-transfers.
sights are
1. The organizational response to the possibility of agency
politics is to reduce the stakes interest groups have in regulation.
2. The threat of producer protection leads to low-powered
incentive schemes. That is, the theory predicts contracts that are
somewhat closer to cost-plus contracts than a theory ignoring the
possibility of producer protection.
3. The agency's discretion to choose among price levels,
pollution levels, and more generally variables affecting the other
interest groups than the regulated industry is reduced when the
latter become better organized.
4. Our approach refines the view that there is a market for
regulatory decisions. First, the regulatory inefficiencies associated
with the pressures of several interest groups may compound rather
than cancel. For instance, an industry (eager to extract a rent) and
an environmental group (eager to limit production to curb pollu-
tion) may have a common interest in Congress's not being in-
formed about the production technology. Second, the power of an
interest group does not depend only on its willingness to pay, i.e.,
on the combination of its stake in the regulatory decision and of its
cost of organizing and of influencing government, but also on the
kind of influence it wants to exert. The group has more power when
its interest lies in inefficient rather than efficient regulation, where
inefficiency is measured by the degree of informational asymmetry
between the regulated industry and Congress.
5. In contrast with the conventional wisdom on interest-
group politics, an interest group may be hurt by its own power.
6. Congress must reward the agency for "cooperating," i.e.,
for supplying information.
The more specific insights are
7. In our production-embodied pollution model (Section V),
the better organized the environmentalists, the higher the firm's
rent, if the firm's production is essential. In contrast, if production
is not essential so that shutdown is a relevant policy, the environ-
mentalists' pressure may hurt the firm.
8. The methodology developed in this paper is extended to
yield a political theory of cross-subsidization. Interest-group poli-
tics may yield uniform pricing by regulated multiproduct firms.
9. The optimal allocation can be implemented without side-
transfers when the supervisory information is hard. Soft supervi-
sory information may make equilibrium side-transfers desirable;
that is, Congress may use one interest group as a countervailing
force to another interest group.
Due to their complexity, hierarchical models such as the one in
this paper are necessarily highly stylized. It is therefore worth
commenting on which results are likely to extend. The strong
intuition behind the general insights gives us confidence in their
robustness. Insights (1)—(3), on the desirability of reducing the
interest groups' stakes in regulation, are likely to carry over to
other models in which interest groups only try to capture, but do
not contribute positively to the regulatory process. In contrast, if
interest groups may bring new information about the agency's
activity, it may be socially desirable to increase their stakes in
regulatory decisions so as to induce them to acquire information
and make regulation more efficient [Laffont and Tirole, 1990b].
Insight (2), on the optimality of low-powered incentive schemes,
will carry over to the other situations in which the agency brings
technical expertise to the political principal; high-powered incen-
tive schemes are bound to leave high potential rents to the industry
and thus to create high payoffs to collusion. A different situation
arises when the agency performs the role of an accounting office.
Then stakes in collusion may be reduced by the use of high-
powered incentive schemes; in particular, fixed-price contracts
remove agency discretion by suppressing cost accounting by the
agency (Laffont and Tirole [1990c1 examine the validity of this
intuition). Insight (4), linking interest group power and gain from
inefficient regulation, is a natural conclusion in models of hard
information, in which the agency's degree of freedom is necessarily
to hide information from Congress. Insight (5) is akin to the classic
point in game theory that a player may lose from having more
options in a multiplayer situation. Insight (6) conforms to standard
agency theory. While in this paper agency incentives are provided
by rewards, they might alternatively be provided by punishments
inflicted when the agency is caught colluding with interest groups
(as in Laffont and Tirole [199013]). Last, insights (7) through (9),
while making sense, are quite special, and their extension to more
general frameworks is an important line for future research.
APPENDIX 1: PRODUCER PROTECTION

Let us index the four states of nature in the following way:
state 1: {13 = A,Cr = §}, state 2: 1p = §,Q= 43.1 state 3: {13 =
;
[3,Q= 4,1; state 4: 1p = 5,0-= 51. We let x, denote the probability of

state i (for instance x 1 = Cy). A pair of contracts offered by Congress
leads to a side contract between the agency and the firm, and to
some equilibrium allocation. We index the final incomes and
utilities (which include the equilibrium bribes, if any) by a hat: Ii i ,
S'„ U,171_ 1 . The actual transfers from Congress to the agency and
to the firm are denoted s, and t, in state i. Letting s, denote the
firm's bribe to the agency, we have
(A.1) §i = Si + §i
(A.2) ti = ti — (1 + xd§i
(A.3) si > 0
-
(A.4) üi = ti — kli(e i )
(A.5) IATi = s' i — s *.
(We shall assume that the final allocation is deterministic. The

reasoning is easily extended to random final allocations.)
We want to prove that there is no loss of generality in
assuming that (a) the agency reports if truthfully; (b) transfers are
based on (q ,C ,r) only (in particular, more complex mechanisms,
like announcement games, do not raise welfare); (c) the agency's
income depends only on its report; (d) there is no side-transfer in
equilibrium (§, = 0 for all i). The strategy of proof is the following:
first, we derive an upper bound on expected welfare. To do so, we
derive a couple of necessary conditions that must be satisfied by the
final allocation in any equilibrium. We then write welfare as a
function of the final allocation and equilibrium bribes and maxi-
mize it subject to this limited set of constraints. We find in
particular that optimal bribes are equal to zero. Second, we show
that this upper bound can indeed be reached by an incentive
scheme that satisfies properties (a)—(d) (and in particular is
collusion-proof). It is then straightforward to check that the
optimization program is equivalent to (25).
First, we claim that all i,
(A.6) §, > s* .
(A.7) U, 0 .
If either of these inequalities is violated, one of the parties refuses

to participate in the regulatory process because it rationally
anticipates that its final utility will be lower than its reservation
utility. Next, we claim that
(A.8) (72 > U3 + (I) (e 3 ) .
36. We allow only positive bribes for simplicity. Negative bribes (bribes i, from
the agency to the firm, which would cost (1 + X a )t), can be shown to be suboptimal
as long as X., 0. (The reasoning is the same as the reasoning below.)
Because, in state 2 the firm is the only one to know that 13 = 13, it
can mimic the behavior of type 13 and get utility U3 + (1)(e 3 ). Last,
(A.9) (1 + Xf )(§ 1 — S 2 ) — (71 .
If (A.9) were violated, the agency and the firm would be better off
signing a different side-contract in state of nature 1. The crucial
point here is that any messages m 2 that are sent by both parties in
state of nature 2 can also be sent in state 1 (the converse is not true,
as in state 2 the agency cannot substantiate a report if = 13). So the
two parties can agree to send the messages m 2 and specify a large
side-transfer from a party that defects from these messages to the
other party."
The expected social welfare is
(A.10) 7= ± x i 1[S(q i ) + XP(q i )q i ]

i=i
— (1 + X) (s, + t, + (13, — e, )q,) + Ui + 1^7,1,
or, using (A.1), (A.2), (A.4), and (A.5),
4
(A.11) Et=i x,{[S(q,) + XP(q,)q,1 — (1 + X)(s* + X §, + 4 (e ) 1 1
+ (Ni — e,)q,) — XUi — X(§, — s*)}.

We now find an upper bound 7illax for 7. when the constraints
(A.3) and (A.6)—(A.9) are imposed on the control variables
e,, §„U,} 4,,. That is, we ignore other potential constraints for
{sq„A
the moment.
Because rents are costly, the solution must satisfy
(A.12) si = s* for i = 2,3,4,
(A.13) = U4 =0.
Furthermore, (A.8) and (A.9) are satisfied with equality. Next, be-
cause the problem is separable between bribes and other variables,
(A.14) si = 0 for all i.
Last, we must show that U1 = 0. To do so, note that (A.8) and (A.9)
imply that
(A.15) (1 + Xf )(S' i — s * ) = (1)(e 3 ) —
37. We are here assuming that, to be enforceable, the transfers from Congress
to the agency and the firm, the price level and the cost targets are based on
observable messages. But the analysis can be extended to cases in which the
messages are not observed by all parties (under risk neutrality the parties can
design side-transfers based on the observable transfers, price, and cost target that
deter any party from deviating from m, ).
Thus, maximizing with respect to t JOs equivalent to maximizing
A
I— X U1 — X( — U1 1(1 + X1 )} subject to U 1 0. Thus, U 1 = 0. The

maximizations with respect to q, and e, are as announced in Section
IV: output q, is Ramsey optimal given marginal cost (13, — ed:
state 3, in which
(A.16) tli'(e 3 ) = q 3 1_
q, = R(0, - ed. And effort is socially optimal (jr' (e,) = (id except in
x2 + x1
— 1 + X x 3 x 3 (1 + Xf )
(1)'(e3)•
using (A.8) and (A.9). (That is, e 3 is the arg max e of (25).)
The second step of the proof consists in showing that the upper
bound can be reached (that is, in the notation of the text,
EW = 7 max To do so, suppose that Congress offers the following
. ) .
incentive schemes: the agency makes a report r, and the firm

announces its type p (equivalently the firm could be rewarded on
the basis of r and Clq). Letting i = 1 denote the state in which r = A
and p = p, etc., Congress gives transfers
(A.17) t, = II, + tge, )
(A.18) s, = §,
and imposes cost target
(A.19) C, = ( N edgi
and price
(A.20) p i = R(ii — e,),
where lq„e,A, Or 1 are the solutions to the maximization (A.10)
and p i a P(q,). (If the agency's report and the firm's announcement
are inconsistent, or if the cost target is not reached, Congress
imposes a large penalty on the other two parties.) Now, it is
straightforward to check that in no state of nature do the agency
and the firm have an incentive to collude against this scheme, or to
individually misreport or lie. Thus, the upper bound can be reached
by a pair of contracts that satisfy (a)—(d), as claimed above.
Last, we say a few words about the case in which the parties
can sign a contract before the agency and the firm (simultaneously)
get their information. We assume that both the agency and the
firm are infinitely risk averse, so that they care only about their
worst payoff. It is clear that for the optimal contract, both the
agency and the firm are put at their reservation utilities, s* and 0
(otherwise, transfers could be reduced uniformly for at least one
party, without any incentive effect). We focus on deterministic
contracts. The social welfare function is slightly different from T ,.
as ex post rents no longer have a social value:
4
(A.21) k= Ex l[S(q) + XP(q,)q,] — (1 + X) (s, + t, + (13, — e,)q,)}

i=1
i
or
4
(A.22)T.= E
1=1
x l[S(T) + XP(q,)q ]
i i
— (1 + X)(§, + Xi s, + CI, + ((3, — e i )q, + tge,)).

To show that the analysis is (qualitatively) identical to that of the
no-prior-contract case, it suffices to note that the minimal set of
constraints ((A.3) and (A.6)—(A.9)) is still a set of necessary
conditions when the parties are infinitely risk averse and contracts
are signed prior to the revelation of information.
APPENDIX 2: MULTIPLE INTEREST GROUPS
The reasoning is the same as in Appendix 1, so we shall skip

the details. Let S- , denote the environmentalists' transfer to the
agency in state of nature i. To show that i, = 0 for all i in
equilibrium, one proceeds as in Appendix 1, by writing the social
welfare function as a function of the final allocations and bribes,
and showing that one has a corner solution for bribes.
An intuitive argument is the following: a bribe S-, allows
Congress to reduce s, by §, for a final income s, (as in Appendix 1, the
optimization problem is separable between bribes and the final
allocations; so we must hold the final allocation as fixed). The social gain
is (1 + X)si . But, the environmentalists' welfare is reduced by (1 + k)s,.
So, the net welfare gain is equal to (X — X )§, ^ 0 if X, X.
e
The social welfare function is (using the notation of Appendix

1):
4
(A.23) = E
t=i
x {[S(q,) + XP(T)T1
j
— (1 + X) (s, + ti + (13 i — e, )q, ) + t1i + 17j.

Using (30), (31) and the fact that S-, = §, = 0 for all i,
4
(A.24) 7=. E
t=i
x l[S(q,) + XP(q )q i
i i i
— (1 + X)(s* + 111(e,) + ((3, — e i )q,))

Ce 3 ) D(q4) D(q3)
— +24)(e 3) + xi
1+ f
+x
4 1+XI .e
(In the notation of Section V, q 4 = 4, e 4 = e, q 3 = q, e 3 = e.) We thus
obtain
(A.25) (1 + X)(P(q,) — (Ni - e 1 )) + Xq,P' (q,)) = 0 i = 1,2
(1 + x)(p( q3 ) — Cif — e 3 )) + Xq 3 P'(q 3 )) = — X X4D '(q 3 )
(A.26) xo
+ fie)
D' (q 4 )
(A.27) (1 + X)(P(q 4 ) — (0 - e 4 )) + Xq 4P (( 1 4) = X 1 + xe
(A.28) 41' (e,) = q, i = 1,2,4

X x2 x1
(A.29) (e3)q3 1 + X X3 4- X3(1 + xf ) (1)'(e3).
That is, the price is distorted away from the Ramsey price in
states 3 and 4, and effort is distorted downward in state 3. Now,
using our assumption that the program is concave (as we men-
tioned, a sufficient condition for this is that X not be too big), (A.26)
and (A.27) imply that
dq 3 dq4
dX,< 0, c1)T.
,> 0.
Note that it may happen that the solution above satisfies q 4 <
q 3 . In this case the solution is a corner solution: q 4 = q 3 (because
q 4 = 4 and q 3 = q in the coalition-incentive constraint (29), (29) is
no longer binding if q 4 ^ q 3 . Hence the term in D(q 4 ) — D(q 3 ) must
be omitted in the expression of which yields q 4 > q 3 , a
contradiction).
APPENDIX 3: PROOF OF PROPOSITION 4
Let K be the Kuhn-Tucker multiplier associated with (38). The

first-order condition for program (39) for an interior solution is
p 1 — ((3 - e — d) X 1
(A.30) L1=
— 11++ X
A *p i )
K I 1 1 + xe 4dq 2 ( 2d(q i — q 2 )\1
+
1 + X [ 11( P 1) — Pi (q1 + q 2 ) 2 e+ (11 + q2 ) .1
P2 — (13 — e + d) X 1
(A.31) L 2 = -
P2 1 + x. 1(p2)
K 1 1 1 + X, 441 i 2C1(qi - q2 ) \ 1
+
1 + X [ 1( P 2 ) P2 (q1 + q2 ) 2 4I' k e + (11 + q2 ) i
+ K(1 + X, )( 2d(q i — q 2 ))
(A.32) Ce) = q 1 + q 2 e+ — (e).
1+x q1 q2
(a) Let us first assume that d is small. We see that {q 1 = q 2 = q,
P1 = P2 = p ,e,K1 satisfy (A.30)—(A.32) if
p — (i3 - e) X 1
(A.33)
p 1 + X p)
(A.34) (e) = 2q
(1 + X)d
(A.35) K- (—P ' (q)q) — 2(1 + X, )d
Note that p, q, and e are independent of d. The proof that uniform
pricing is optimal proceeds in two steps. First, q, < q 2 is dominated
by uniform pricing from the concavity of the social welfare function
in a collusion-free world. Furthermore, uniform pricing is collusion
free. Second, for any E > 0, there exists d o > 0 such that for any d
< d o , the optimal q 1 and q 2 satisfy max (kg, — q , fq 2 — q I) < E,
where q is the solution to (A.33) and (A.34) and is the optimal
output for each category of consumers when do = 0. If this property
were not satisfied, for some E > 0, there would exist a sequence d —>
0 with optimal (q ,q2) such that for all n, max (1q7 - ql - ql) >
E. Because in the absence of collusion, outputs must tend to q when
d tends to zero, welfare along this sequence would be bounded
away from the collusion-free welfare. But the difference between
the collusion-free welfare and the welfare obtained under uniform
pricing tends to zero as d tends to zero. Hence, (q ,q2) must be
strictly dominated by uniform pricing for n large, a contradiction.
Last, now that we have established that (q 1 — q 2 ) tends to zero
when do tends to zero, which implies that K tends to zero, we can
make a first-order Taylor expansion of the coalition-incentive
constraint. Using (A.32), the left-hand side of (38) is equal to
(1 + X, )2d(q, — q 2 ) to the first order, and the right-hand side of
(38) is equal to (-1) 1 (q)q)(q, — q 2 ) to the first order. Hence if q 1 >
q 2 and d (-13 ' (q)q) I (1 + X, ), (38) is not satisfied. If q i < q 2 , (38) is
not binding, implying that K = 0 and from (A.30) and (A.31) q 1 >
q *2 , a contradiction.
(b) Consider the case in which the solution to (35)—(37) does
not satisfy (38) for any d (so that d2 = +00). We want to show that
uniform pricing is not optimal for a large d (note that large d's raise
the possibility that marginal costs become negative. We shall
assume that f3 is large enough so that this does not occur).
To this purpose, suppose that optimal pricing at p, yielding
demand q, is optimal. Consider a small deviation around uniform
pricing: q 1 — q 2 = E > 0. The left-hand side of (38) is equal to
(1 + X, )4i'(e)(2c/E/2q) = 2(1 + X.,)d€ to the first approximation,
where use is made of (A.34). Similarly the right-hand side of (38) is
[— P ' (q)q] and is independent of d from (A.33) and (A.34). Hence,
for d large enough, (38) is satisfied for small amounts of price
discrimination. From the concavity of the social welfare function in
the collusion-free world, a small amount of price discrimination,
which, we just saw, is feasible, is preferable to uniform pricing.
Next consider the case in which there exist d's such that (38) is
not binding for the collusion-free solution (given by (35)—(37)). Let
d 2 denote the smallest such d. We claim that for d = d 2 — E (where E
is positive and small) pricing is discriminatory. We know that at d 2 ,
(38) is just binding. Because kIJ is convex, one can increase e to e +
where q is small such that
2(d 2 — €)(q 1 — q 2 ))
(e + +— tge +
qi q2
(e 2d 2 (q 1 — q 2 )\
qi + q2
Equation (38) is still satisfied for the collusion-free levels q 1 and q 2 .
This implies that Congress can obtain almost the collusion-free
level of welfare when d is close to d 2 , which obviously is impossible
under uniform pricing.
Next, we observe that if (38) is satisfied for the collusion-free
levels and parameter d, it is also satisfied for the collusion-free
levels and parameter d' > d. This means that the set of parameters
for which the collusion-free solution obtains is indeed the open
interval [d 2 , + co).
Last, we want to show that there exists d 1 such that uniform
pricing obtains on [0,d 1 ] and not elsewhere. To this purpose, note
that the welfare under uniform pricing is independent of d. More
generally, the envelope theorem shows that the derivative of the
social welfare function with respect to d is equal to
2d(q 1 — q2) 2(q 1 — q 2 )
— (1 + X)(q 2 — q 1 ) + K(1 + (e +> 0
q1 + q2 ql q2
for price discrimination (q 1 > q 2 ). Hence, the region with discrimi-
natory pricing and binding collusion is exactly an interval (d l ,d 2 ).
GREMAQ, TOULOUSE
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
REFERENCES
Adams, G., The Politics of Defense Contracting—The Iron Triangle (New Brun-
swick, NJ: Transaction Books, 1981).
Barke, R., and W. Riker, "A Political Theory of Regulation with Some Observations
on Railway Abandonments," Public Choice, XXXIX (1982), 73-106.
Baron, David, and Roger Myerson, "Regulating a Monopolist with Unknown
Costs," Econometrica, L (1982), 911-30.
Baumol, William, and David Bradford, "Optimal Departures from Marginal Cost
Pricing," American Economic Review, LX (1970), 265-83.
Baumol, William, John Panzar, and Robert Willig, Contestable Markets and the
Theory of Industry Structure (New York: Harcourt, Brace, Jovanovich, 1982).
Becker, Gary, "A Theory of Competition Among Pressure Groups for Political
Influence, " Quarterly Journal of Economics, XCVIII (1983), 371-400.
, "Public Policies, Pressure Groups, and Deadweight Costs," Journal of Public
Economics, XXVIII (1985), 329-47.
Boiteux, Marcel, "Sur la Gestion des Monopoles Publics Astreints a l'Equilibre
Budgetaire," Econometrica, XXIV (1956), 22-40.
Breyer, S., and R. Steward, Administrative Law and Regulatory Policy (Boston:
Little, Brown and Company, 1979).
Buchanan, John, "An Economic Theory of Clubs," Economica, XXXIII (1965),
1-14.
Demski, Joel, and David Sappington, "Hierarchical Regulatory Control," Rand
Journal of Economics, XVIII (1987), 369-83.
Fiorina, M., "Congressional Control of the Bureaucracy: A Mismatch of Incentives
and Capabilities," in Congress Reconsidered, L. Dodd and B. Oppenheimer,
eds. 2nd edition (Washington, DC: Congressional Quarterly Press, 1981).
Joskow, Paul, "The Determination of the Allowed Rate of Return in a Formal
Regulatory Hearing," Bell Journal of Economics, III (1972), 632-44.
Kofman, Fred, and Jacques Lawarree, "Collusion in Hierarchical Agency," mimeo,
University of California, Berkeley, 1990.
Laffont, Jean Jacques, and Jean Tirole, "Using Cost Observation to Regulate
Firms," Journal of Political Economy, XCIV (1986), 614-41.
, and , "The Regulation of Multiproduct Firms, Part I: Theory,"Journal of
Public Economics, XLIII (1990a), 1-36.
, and , "The Politics of Government Decision Making: Regulatory Institu-
tions,"Journal of Law, Economics and Organization, VI (1990b), 1-32.
, and , "Accounting and Collusion," mimeo, M.I.T., 1990c.
McCubbins, Matthew, Roger Noll, and Barry Weingast, "Administrative Proce-
dures as Instruments of Political Control," Journal of Law, Economics, and
Organization, III (1987), 243-77.
McFadden, Dan, "The Revealed Preferences of a Government Bureaucracy: Empir-
ical Evidence," Bell Journal of Economics, VI (1976), 55-72.
Moe, Terry, "Interests, Institutions, and Positive Theory: The Policies of the
NLRB," mimeo, Stanford University, 1986.
Noll, Roger, "The Political Foundations of Regulatory Politics," Journal of
Institutional and Theoretical Economics, CXXXIX (1983) 377-404; reprinted
in Congress, M. McCubbins and T. Sullivan, eds. (Cambridge: Cambridge
University Press, 1987).
, "Behavior of Administrative Agencies," in Noll, ed., Regulation and the
Social Sciences (Berkeley: University of California Press, 1985).
Olson, M., The Logic of Collective Action (Cambridge, MA: Harvard University
Press, 1965).
Peltzman, Samuel, "Toward a More General Theory of Regulation," Journal of
Law and Economics, XIX (1976), 211-40.
Posner, Richard, "Theories of Economic Regulation," Bell Journal of Economics, V
(1974), 335-58.
Stigler, George, "The Economic Theory of Regulation," Bell Journal of Economics,
II (1971), 3-21.
Tirole, Jean, "Hierarchies and Bureaucracies," Journal of Law, Economics, and
Organization, II (1986), 181-214.
"Collusion and the Theory of Organizations," forthcoming in Jean Jacques
Laffont, ed., Proceedings of the 6th World Congress of the Econometric Society,
(Cambridge University Press, 1990).
Warren, K., Administrative Law in the American Political System (St. Paul, MN:
West Publishing Company, 1982).
Weingast, Barry, "The Congressional-Bureaucratic System: A Principal-Agent
Perspective," Public Choice, XLIV (1984), 147-92.
, and M. Moran, "Bureaucratic Discretion on Congressional Control: Regula-
tory Policymaking by the Federal Trade Commission," Journal of Political
Economy, XCI (1983), 765-800.
Wilson, J., The Politics of Regulation (New York: Basic Books, 1980).

Laffont and Tirole 1991

Uploaded by

Copyright:

Available Formats

Laffont and Tirole 1991

Uploaded by

Document Information

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

Laffont and Tirole 1991

Uploaded by

Copyright:

Available Formats

THE POLITICS OF GOVERNMENT DECISION-MAKING:

A THEORY OF REGULATORY CAPTURE*

The paper develops an agency-theoretic approach to interest-group politics and

collective action' as a building block to explain how "regulation is

2. See also Buchanan [1965].

Our model depicts the regulation of a natural monopoly (or

In order to use standard agency methodology, we assume that

maximizes a social welfare function that adds consumer, agency,

be shut down), and solve the model in an increasing order of

II. THE MODEL

function. The revenue is thus P (q )q. Let

describe the elasticity of demand. Our accounting convention is

firm's participation or individual rationality constraint is

where X is the shadow cost of raising public funds through

16. More complex mechanisms (including, e.g., announcements by the firm

III. COLLUSION-FREE REGULATION

A. Full Information (a = (3)

inefficient type) are set so as to maximize the full information

for all 13,(eN3),p*((3)) solves'

4 *() are nondecreasing functions of e. 19

Below we show that, under asymmetric information, the

still conditionally optimal given the inefficient type's effort 6 (i.e., -

—(1 + X)(s* + (p. — e*)q* + tge*)))]

We assume that W" (•) is strictly concave.' The expected social

B. Asymmetric Information (cr = 4))

transfers for types § and 13 under the optimal incentive scheme

(The efficient type can produce at cost 1 — ë by exerting effort e —

20. Sufficient conditions are X small or decreasing marginal revenue. (See

A simple inspection reveals that q =q*, e = e* and q = ei*(e) as

21. The proof of (18) is a standard revealed preference argument. Recalling

Let us summarize the relevant points for what follows. Congress

IV. PRODUCER PROTECTION

22. Equilibrium collusion may be unavoidable in situations in which Congress

information from Congress. The analysis in Section III suggests

information the inefficient type's output is the Ramsey level 4* (e)

PROPOSITION 1. Under producer protection:

firm is lower than in _ the absence of collusion, under asymmetric

(26) EW = max { twFI + ( 1 (e) — tv(1 + X)

where W"1 is defined by (11) and Wfle) is defined by (20).

V. MULTIPLE INTEREST GROUPS

Suppose also that the environmentalists do not purchase the good

which is proved in Appendix 2—is that if the optimal allocation

for Xe = +x. 24 Wew show that environmentalists affect the regula-

24. In our model a reduction in X e—through a better organization, the advent

(32) EW = max{tWn (4 ,ê) + (1 — OW' (q ,e)

both have a stake in making regulation inefficient. The firm wants

The intuition behind Proposition 3 is that consumers favor

discretion lies in hiding information from Congress and asymme-

VI. SHUTDOWN OF THE REGULATED FIRM

The analysis in Sections III—V proceeded under the assump-

allowing production by type (3 only). This sounds like a reasonable

VII. A POLITICAL THEORY OF CROSS-SUBSIDIZATION

instance, think of the captive coal shippers' successful fight against

gross and net surpluses of class i, where PO is the inverse demand

desirable in equilibrium.) Last, we shall assume for simplicity that

Using the symmetry of the model, we look for a symmetric

+ d(q 2 — q 1 ) + Ip(e)] + X[P(q1)q1 + P(q 2 )q 2 ]}

We can now state

order (q 1 — q 2 ). Hence the constraint cannot be satisfied unless

willing to pay Sn(q 1 ) — Sn(q 2 ) to the firm. Let

—(1 + X)(s* + (p. — e)q + tge*)))]

A simple inspection reveals that q =q, e = e and q = ei*(e) as

I— X U1 — X( — U1 1(1 + X1 )} subject to U 1 0. Thus, U 1 = 0. The