Documente Academic
Documente Profesional
Documente Cultură
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend access to The
Journal of Business.
http://www.jstor.org
Lawrence R. Glosten
NorthwesternUniversity
The purpose of this article is to investigate a rationalefor the specialist system on the New York
Stock Exchange (NYSE). A key characteristicof
the specialist system is that there is one individual in each stock, the specialist, that has sole
access to informationabout the tradingprocess,
and this information provides some monopoly
power. It is shown that this monopoly power
may, in some environments,reduce some of the
social costs associated with trading on private
information.
Promotionalliteraturepublished by the New
York Stock Exchange points to the "specialist
system" as a superiorform of marketorganization, citing the "increased liquidity" and the
maintenanceof a "fair and orderly market" attributableto the specialist. On the face of it, the
claim seems preposterous. One's intuitionmight
suggest that any model of rationalmarketmaking
* Many of the ideas in this article were developed in discussions with Paul Milgrom.I also benefittedfrom discussions with Anat Admati,DouglasDiamond,MichaelDothan,
MichaelFishman,KathleenHagerty, Pat Hess, Ravi Jagannathan, Juan Ketterrer, Ananth Madhavan, Steven Matthews, Ailsa Roell, WilliamRogerson, and Chester Spatt. I
also appreciatethe commentsof seminarparticipantsat the
University of Minnesota,the London School of Economics
conference on market making, Standford University, and
Yale University. The usual disclaimerapplies. Fundingfrom
the BankingResearch Centerat NorthwesternUniversityis
gratefullyacknowledged.Partof this researchwas performed
while visiting the CurtisL. CarlsonSchool of Management,
Universityof Minnesota.
212
Journal of Business
would show that a monopolistwill optimallyprovidea less liquidmarket than competing marketmakers. In fact, Ho and Stoll (1981) show
that competingmarketmakerswill lead to a more liquidmarketin the
sense that the average bid/ask spread will be smallerwith competing
marketmakers.
A possible response is that the NYSE monitorsthe performanceof
specialists, and therefore, in fear of retribution,specialists provide a
more liquid marketthan they would if unregulated.However, if this is
the only argument,the obvious rebuttalis, If a regulatedmonopolistcan
supply what would be suppliedby competitors,why shouldregulatory
resources be expended when the seemingly more efficient alternative
wouldbe to replacethe specialistsystemwith competingmarketmakers?
The possible advantage of the specialist system was suggested in
Glosten and Milgrom(1985),and the intuitionis providedby the observation that a monopolist specialist can average profitsover time. For
example, the analysis of Glosten and Milgromsuggests that there may
be instances in which competingmarketmakerswill not quote, bid, or
ask prices. If the adverse selection problem is too extreme, then no
matterwhat bid and ask are set, each marketmakerwill expect to lose
money on a trade. The consequence is that the marketshuts down until
enough public information arrives to relieve the adverse selection
problem. Now imagine a situation in which no public informationarrives until the informationalevent occurs. In this case, after the informedtradersreceive their informationthe marketwill remainclosed
until the informationis revealed.
A monopolist specialist may also close the marketin such a situation, but he need not. By keepingthe marketopen, the specialistlearns
some of the informationof the informed.This may reduce the adverse
selection problem, making subsequent trades more profitable.With a
protected position, the specialist is able to realize the greater profits
that offset the losses earned early on in the trading.The result is that
both liquiditytradersand informedtradersare made better off relative
to the competing marketmaker system.
This article extends this intuitionby consideringan environmentin
which trades are not restrictedto unit amountsas in Glosten and Milgrom(1985). In this case, the marketmaker(monopolistor competitor)
quotes a price schedule that specifies the price per share as a function
of the quantity to be traded. While competing market makers are
forced to set a price schedule that leads to a conditionalexpected profit
of zero (conditionedon the quantitytraded),the monopolistspecialist
maximizes expected profitsover all possible quantities.This extra degree of freedom, the abilityto averageprofitsacross tradesat a point in
time, reinforces the above observationthat the specialist can average
his profits across time. The result is that in some environmentsthe
monopolistwill provide a more liquid market.This reinforcementmay
213
214
Journal of Business
215
216
Journal of Business
217
Q = Qp(S,Wo,W),
(2)
218
Journal of Business
(4)
The specification of the objective function in (4) deserves further
comment. It is conceivable (indeedin the exampleconsideredit will be
shown to be the case) that for some Q the expected profit, conditional
on that Q, will be negative. In specifying the objective function as the
expected profit across possible quantities,the tacit assumptionis that
the specialist quotes a price schedule and then is requiredto honor
those quotes for the trade, no matterwhat Q is requested.Of course, in
reality there is no posting of a price schedule-the only prices posted
are a bid and ask for small trades. If a traderwishes to trade a larger
quantity, then quotes must be obtained for larger quantities. In any
event, once the specialist supplies quotes for various quantities,he is
in fact bound to honor those quotes, whereas the trader is free to
choose the quantityhe wishes to tradebased on those quotes. The fact
that the specialist "moves first," in the sense that he must honor his
quotes, is the motivationfor choosing the objective function in (4).
Price functions satisfying(2) and (3) or (4) need not exist. In fact, the
subsequent analysis will show that there are situations in which the
competitivepricingfunctiondoes not exist. This happenswhen there is
very little uncertaintysurroundingthe endowmentsandrelativelygreat
uncertaintysurroundingthe informationaboutX. In this instance, the
adverse selection problem that market makers face is so severe that
there is marketbreakdown-all marketmakersrefuse to make a market.
This may or may not happenin the case of the monopoly specialist,
and in the example considered here it does not. Since the specialist
maximizes expected profits, he may expect to lose for some Q and
profit on other trades and still expect a positive profit. His monopoly
position allows him to average his profits cross-sectionally, that is,
across possible trades.
The specificationof the competitive and monopoly environmentsin
(1)-(4) are so general as to preclude all but the weakest analysis summarized in the above two paragraphs.To obtain more results, more
3. Of course, schedules satisfying(3) may not be ordered,in which case there is no
basis for choosing among the equilibria.This will not be the case with the example
considered.
219
S = X+
R=
?, ,
- N(O, lIiTrs),independentof X,
1.
(5a)
(5b)
(5c)
(5d)
(5e)
(6)
+ (W + Q)E[XIS]
- .5p(W + Q)2var(XIS).
4. This specificationof the liquidity motivationfor trade is also used in Diamond
(1985).Also notice that the specificationof a zero meanimpliesthatthe riskrepresented
by the randomvariableX is in zero net supply. Thus, the effects of the risk-neutral
marketmakersas insurersis minimized.Thatis, the marketmakersfacilitaterisksharing
between agents and they do not expect to end up with a positive (or negative)inventory
of shares. Furthermore,the risk neutralityof the marketmakersdoes not affect the
averageprice of the security.
220
Journal of Business
(8)
=
-
P"(Q)Q
2P'(Q)
m,
rsT)]<
[P/('ITx+
0.
(10)
X +
ii,
+ -r,).
E[XIZ] = m + TrZ/Q(rrx
(11)
rwrs(rs
rx),
a),
K unrestricted,or
P(Q)
(x =
ko unrestricted,
ki = p/(lTx + ws).
221
If
OX= P2ITr/[P2Trr +
Tr1Trs(Trr +
= P2iTx + 1TwITs(TS5+
1)N]}Q,
(13)
Tx).
The Q chosen by an arrivinginvestor is derived from (9). The equilibrium Q, QJ(Z)is given by
QJ(Z) = (Z - m)irs(2a - 1)/p.
(14)
Journal of Business
222
W2),
Trw
(15)
223
is strictly less than the ex ante maximizedutility withoutprivateinformation,E[U(Wm)].The Appendixcontainsthe prooffor proposition2.
The proposition is true, not because of resources wasted in the acquisitionof information(informationis costless in this model) but because of the imperfect risk sharing due to the existence of private
information.The result is thus analogousto the results of Hirschleiffer
(1971) and Marshall(1974) that private informationcan frustrateoptimal risk-sharing.If informationwere costly, the result would continue to hold and the interpretationwould be similarto thatin Diamond
(1985). The advantageof reconsideringthis result in this frameworkis
that the source of the inefficiencyis obvious and (at least in principle)
observable. Private informationdecreases the liquidityof the market,
and it is this lack of liquiditythat provides the result.
B. Monopolist Specialist Equilibrium
Given that any law restrictinginformationacquisitioncan only be imperfectly enforced, the next question to ask is whetheror not the costs
of informedtradingare a functionof institutionaldesign. In this section
it is shown that, in certain circumstances,a specialist with an institutionally protected position can increase the liquidity of the market.
Given the past literatureon the topic, the environmentsin which this
should hold must be those in which asymmetricinformationis thought
to be a significantproblem.Otherwise,the usual inefficiencyattributed
to a monopolist will hold. A monopolist specialist in an environment
with symmetric informationwill restrict trade, leading to inefficient
risk sharing.
It is convenient to recast the investor's choice and the monopolist's
maximizationproblemin the followingway. Notice from(9) that, given
any price schedule, the investor's optimaltradewill be a functionof Z
and observables, say Qp(Z).The specialist'sproblemis to finda sched-
224
Journal of Business
ule P() that will maximizehis expected profits.For a particularschedule, P(), define the functions Q() and R() by
Q(z) = Qp(urzz + m),
[p2 Sx
wws(x
2
2lS)]iT"iT
and z is Z normalizedto have mean zero and unit variance. Then the
specialist's optimalchoice of P() amountsto a choice of functionsQ()
and R('), which maximize his expected profit subject to an incentive
compatibilityconstraint. Formally, his problemis
max E[R(z) - XQ(z)],
(17)
subject to
CE[W,S,Q(z),R(z)]- CE[W,S,Q(z'),R(z')],
for all z', z = [S - (pWI-rrs)- m]/uz, and CE() is given in (7).
The strategy for solving this problem is to first assume that the
constraintcan be summarizedby the local conditionfor maximization.
This leads to functions R() and Q() that are optimal given the firstordercondition. However, this solutionis not feasible because there is
a discontinuity that implies that the local representationof the constraintin (17) does not, in fact, characterizethe constraint.A modification of the solution to the first problem is optimal and feasible, however.
Assuming that R(') and Q( ) are differentiablewith derivativesR'(Q)
and Q'(.), and definingD(z) by D(z) = R(z) - mQ(z), the local representation of the constraintin (17) is
D'(z) - Q'(z)[Isruzz - pQ(z)]I(as +
(18)
- Q'(z)V[z,Q(z)].
(19)
225
Q(z)e(z)],
u(z)V[z,Q(z)].
(20)
subject to
Q'(z) = u(z); D'(z)
The proof of proposition3, in the Appendix, establishes that the solution to this problem is
Q(z) = (asTz/p)(otz - {[1 - F(z)]If(z)}) for z
=
Ox =
OrrUzIp)[az + F(z)If(z)]
p2 rr/X[p22Trx + Trw
Trsf(x +
>
for z < 0,
(21)
s)] .
+ mrs)- pQ(z)I[2(QTx
+ ws)]
Q(z){JiTuzzI(iTx
(23)
ls(rzu(irx+ Trs)fQ(t)dtIQ(z)},
Journal of Business
226
- .5z*2/Z
{1- [1
- F(t)]Iflt)}
F(z)]I[oazf(z)]}
(1/aZ)f[I
[Trrsuz(l -
ot)z]/2(Tr
Tr,)
m + .5(ua1TrwTs2z)/[p2Trx
+ TwWs(,Tx+
IrS)].
24zz)/(p21Tx
(1TwT
TrwTrsTrx +
ITiT 2),
and hence for large z, R(z)/Q(z) < E[Xlz], and the specialist loses
money on large trades.
Extreme z's are relatively unlikely, and hence the specialist still
expects to profiton average. In fact, the specialist's expected profitis
given by
{[P2ITx+
Tw1Tss(1Tx+ 1Ts)]/1TrwTx(1Tx
1TS)}
f(t){at
[1
F(t)]If(t)}2.
227
Journal of Business
228
less than or equal to .5, the adverse selection problemis so severe that
the competitive market closes down. Of course, the monopolist specialist will not dominatein all environments.In particular,if there is no
private information, then everyone would prefer competing market
makers. This can be seen by noting that when there is no private
information, competitive market makers will set a zero spread,
whereas the monopolistwill set a positive spread.Thus, in the competitive case, there will be complete risk sharing,whereasit will be limited
in the monopolist-specialistcase.
The next proposition shows that after a reparameterizationof the
model, there is an cx> .5 such that, ex ante, the monopolistspecialistis
preferred when a < 'a, whereas competing market makers are preferred if a > 'a. This verifies the intuition that the superiorityof the
monopolist position derives solely from informationalproblems. If
those problems are not great then the monopolist-specialistsystem is
inferiorto a system with competing marketmakers.
5. Let Vm denote the ex ante expected utility of a
PROPOSITION
229
230
Journal of Business
231
P(Q) =
7rx
7rz
7rx
7rs
7s
7rz
Define g(Q) by
g(Q) = P(Q) - m -
Qpw,wsID,
where
D = p27rX
rwwws(n,+ ws)
if D
7O O.
After substitutingfor
s =
112/(p2
nwgs),
we have
p27T,/Q = iwTs(w.( + 1TS)g'(Q)Ig(Q).
ForQ > 0,
g(Q) = KQ, -y = P21T./[1Tw,TS(1Tx + 1TS)]I
with K unrestricted.A similaranalysis holds for Q < 0. If D
expression above can be written
(7x + 7S)[g'(Q)Q
_ g(Q)]1Q2 + pIQ = 0;
0, then the
g(Q) = p(Q) - m.
This has the solution given in the statementof the lemma. Q.E.D.
Journalof Business
232
Proof of Proposition I
First, the uniquenessof the linearpricefunctionis shownwhenD > 0. Assume
that a pricing schedule is given as in lemma 1 with K > 0. Now consider a
continuous,differentiableprice schedulethat is equalto this price functionfor
Q C 1, and linearfor Q> 1. To ensure differentiabilitymake the slope of the
linearportionequal to the slope of the originalscheduleat Q = 1. Evaluation
of the expected profit will show that this second price schedule will yield
positive profitsfor trades largerthan 1. On the other hand, K cannotbe taken
to be negative, for then the second-orderconditionwill be violated. Thus, the
only undominatedpricing schedule when D > 0 is linear.
Suppose D < 0. Then the derivativesof the solutiongiven in lemma 1 are
P(Q) =
(p'n?1TnID) + KylQlj-1,
IQI-1 >
{P[P2ITx + lTwlTS(lTx +
1TS)]}I[Ky(y+ 1)(-D)].
(A1)
Since D < 0, and -y< 1, the second-orderconditionis satisfiedfor - Q' < Q<
Q' where Q' satisfies (Al) with equality. Considerthe expected profitto the
marketmakerson trades of Q'. A traderwill trade Q' if Z is largerthan some
Z'. But then, E[X|Q'] > E[XIZ'] - P(Q'), and the marketmakers'profit is
negativeon tradesof Q contraryto the hypothesis.Thus, if D < 0, thereis no
schedule where the market maker breaks even. The analysis of D = 0 is
similar.Q.E.D.
Proof of Lemma 2
If the marketcloses down (D s 0), then Q, = 0 andA = 0 in the statementof
analysisof
the lemma.If D > 0, then the resultis obtainedby a straightforward
the expressionsin proposition1, assumingthatS and Ware independentandX
and W are independent.Q.E.D.
Proof of Proposition 2
First note that if there is no private information(i.e., ws = 0), competing
marketmakerswill set P(Q) = m for all Q. In that case, the optimaltrade is
- W and the ex ante expected utility is E[U(Wm)].Assume this expectation
exists. When there is privateinformation,the optimaltradeis Qand the price
scheduleis PC(-).The realizedwealthof a typicaltraderis Wm+ W(X - m) +
Q[X - P,(Q)], and the ex ante utility is E[U(Wm + W[X - m] + Q[XP'(Q)])]. This expectation may not exist. If it does not, the proof is complete.
Suppose it does. A sufficientconditionfor the ex ante utility without private
informationto exceed the ex ante utility with privateinformationis
E(U'(Wm){W(X- m) + Q[X- PC(Q)]})' O.
Given the independenceof X and W,this can be rewrittenE[U'(Wm)e(W)]' 0.
Note that
E[e(W)] = E[Q{X - P(Q)}] = E[Q{E[XIQ] - Pj(Q)}] = 0.
233
e(w)fw(w)[U'(wm)+ U'(-wm)] +
{
_00
e(w)f(w)[U'(wm) + U'(-wm)].
Since U"'() 2 0, U"(x)is increasingin x, and hence U'(wm) + U'(- wm) is
decreasing in w. Thus, for w c -w', U'(wm) + U'(-wm) 2 U'(w'm) +
U'(-w'm)
< w
0, U'(wm) + U'(-wm)
'
U'(w'm) +
The Hamiltonianassociated with (20) is-Q(z) and D(z) are state variables,
Q'(z) = u(z) is the controlf(z)[D(z) - e(z)Q(z)] + X1(z)u(z)+ X2(z)u(z)V[z,Q(z)].
The conditions the maximummust satisfy are
i. X1(z) + X2(z)V[z,Q(z)]
= 0;
= (ursquIp)q(z),
where the competitive and monopolist q's, qc and qm, respectively, are given
by qc(z) = (2a - 1)z, qm(z) = oz - [1 - F(z)]If(z) for z > z*, 0 otherwise. The
'Ts)
IT2oMz/[p(w
ws)]
+
qi(t)dt,
Journal of Business
234
where i = c,m. Denote the coefficient on the integralby Yi/Pand the integral
by Gj(z,a). Let CEc and CEmdenote, respectively, the equilibriumcertainty
equivalentsfor the competitiveand monopolycases. Ourintentionis to evaluate.
E[exp(- pCEm) - exp(- pCEc)],
where K dependson all the parameters,and the y's can be chosen independent
of a. The differencein expected utilities can thus be written
KE[H(z,y,a)exp{jy1[Gc(z,6) - GC(z,o)]} (exp{jy[Gc(z,a) - Gm(z,aY)]}- 1)].
235
Ketterrer,Juan. 1987. Market games with inside information.Workingpaper. Minneapolis, Minn.: Universityof Minnesota,Economics Department.
Kyle, Albert. 1985.Continuousauctionsand insidertrading.Econometrica53 (November): 1315-35.
Leffler,George L., and Farwell, LoringC. 1963. The Stock Market.3d ed. New York:
RonaldPress.
Madhavan,AnanthN. 1987.Tradingmechanismsin securitiesmarkets.Workingpaper.
Ithaca, N.Y.: CornellUniversity.
Manne,H. 1966.Insider Tradingand the Stock Market.New York: Free Press.
Marshall,John M. 1974.Privateincentivesand publicinformation.AmericanEconomic
Review 64 (June):373-90.
Rothschild,M., and Stiglitz, J. 1978.Equilibriumin competitiveinsurancemarkets:An
essay on the economics of imperfectinformation.In P. Diamondand M. Rothschild
(eds.), Uncertaintyin Economics. New York: AcademicPress.
Spence, Michael. 1978. Productdifferentiationand performancein insurancemarkets.
Journalof Public Economics 10 (December):427-47.
Stiglitz,J. 1977.Monopoly,non-linearpricingand imperfectinformation:The insurance
market.Review of Economic Studies 44 (October)407-30.
Stoll, Hans R. 1984. Alternativeviews of market-making.Workingpaper. Nashville,
Tenn.: VanderbiltUniversity, Owen GraduateSchool of Management.