THE JOURNAL OF FINANCE VOL. XXXIII, NO.

4 SEPTEMBER 1978

THE SUPPLYOF DEALER SERVICESIN SECURITIESMARKETS

HANS R. STOLL

THERE HAS BEEN much discussionof a policy nature about the function of dealers
in equity securitiesmarkets,the efficiencyof differentmethodsof providingdealer
servicesand the regulatoryconstraintsunder which dealersshould operate. Some
empiricalworkhas been carriedout to determinewhat factorsunderliedealercosts
and to assess the efficiency of different market organizationsand regulatory
constraints, but that work has not been based on a very explicit theoretical
foundation [see Demsetz (1968), Institutional Investor Study, Ch. 12 (1971), Tinic
(1972),Tinic and West (1972),and Benstonand Hagerman(1974)].The purposeof
this paperis to developa more explicitand rigorousmodel of the individualdealer
and to discussthe implicationsfor the cost of tradingof differentmarketorganiza-
tions of dealers. It is hoped this model will provide a better framework for
empirical work and for discussion of the policy issues involved. The paper is
restrictedto the supplyside. For steps in the directionof specifyingthe demandfor
dealer servicessee Copeland(1976) and Epps (1976). Dealers facilitate tradingby
investorsbecausethey are willingto tradefor theirown accountas principalswhen
investors'agents(or investorsacting for themselves)cannot immediatelyfind other
investorswith whom to trade.' Following Demsetz (1968) one can, therefore,think
of dealersas providingthe serviceof immediatetradingor immediacy.The cost of
immediacydevelopedin this paper is the sum of: (1) holding costs, the price risk
and opportunitycost of holding securities;(2) order costs, the costs of arranging
trades,recordingand clearinga transaction;and (3) informationcosts which arise
if investorstradeon the basis of superiorinformation.
Dealers are to be distinguishedfrom brokers who act strictly as agents for
investorsand do not assumerisk.This paperemphasizesholding costs, which have
also been the emphasisof earlierstudieson dealers.Ordercosts, which are incurred
both by dealers and brokers, are included in the analysis but do not receive
extensivedevelopment.Informationcosts facing dealersare also modeled, but see
Jaffe and Winkler(1976)for a more extendedtreatmentof that issue. Dealers exist

*IUniversityof Pennsylvania.This is a revision of an earlier paper written at the University of

Pennsylvania.I benefitedgreatlyfrom the commentsof my colleaguesat the Universityof Chicago
while a visitortherein 1975-76,from the commentsof my colleaguesat Ecole Superieuredes Sciences
Economiqueset Commercialeswhile a visitor there in 1976-77, as well as from the commentsof my
colleaguesat the Universityof Pennsylvania.Specialthanksare due MarshallBlume,BernardDumas,
ThomasEpps, RobertHamada,JonathanIngersoll,RobertLitzenberger,Ron Masulis,MertonMiller,
JamesMorris,Roland Portait,RandolphWesterfieldand JosephWilliams.All are absolvedfrom the
remainingerrors.The financialsupportof the Universityof PennsylvaniaLaw School Centerfor the
Studyof FinancialInstitutionsis gratefullyacknowledged.
1. For the purposesof this paperinvestorsincludeindividualsand institutionalinvestorswho in other
contextsmightbe consideredto be intermediaries.
1133
1134 The Journal of Finance

because they provideimmediacymore cheaply than investorscould provide it for

themselves(by placinglimit orderswith brokersand borrowingor lendinguntil the
limit order is executed, for example).Not everyone is a dealer because there are
fixed costs (office, phone, etc) to the dealerbusiness.
Section I presentsan intuitiveapproachto holding costs. Section II of the paper
discusses the principalassumptionsunderlyingthe analysis. Section III develops
the holding cost function in a one period context. Implicationsof the model for
determinantsof bid-ask spreadsand for the role of diversificationby dealers are
discussedin Section IV. Ordercosts and informationcosts are discussedin Section
V. In Section VI certain policy issues related to the industrial organizationof
dealersare discussedin the context of the model. In Section VII modificationsare
made to put the model in a multiperiodcontext. The conclusionsare summarized
in Section VIII.

I. A FAMILIAR PICTURE
The dealer can be viewed as any investor who has a desired portfolio (his
investment account) based on the opportunitieshe sees and on his preferences.
Supplying immediacy to other investors means moving away from this desired
portfolioin orderto accommodatethe desiresof investorsto buy or sell a stock in
which the dealer specializes.As a result the dealer assumes unnecessaryrisk and
moves to a level of risk and returnwhich may be inconsistentwith his personal
preferences.
These points are best illustratedby consideringFigure 1. Line RfE is the dealer's
efficient frontier,which representsthe possible combinationsof his efficient port-
folio of risky assets (point E) and the risk free asset (yielding Rf). Assume the
dealer'sdesired position-his investmentaccount-is point N. By taking nonop-
timal portfoliopositions the dealer moves away from N and indifferencecurve U0
to a lower indifferencecurve such as Ul. The dealer is forced off the efficient
frontierbecauseit is assumedthat he cannot initiatetradesin his efficientportfolio
of stocks, E. The portfolio acquiredin the process of acting as a dealer is termed
the trading account. Long or short positions in his trading account may de-
diversifythe dealer'stotal portfolio and cause the frontierof his new portfolio set
(trading plus investment account) to be described by a line such as ANB, the
positionof which dependson the riskcharacteristicsof the tradingaccount.
A movement upward along NA means the dealer has an undiversifiedlong
position financed by borrowingat Rf. At point A he has, for example,borrowed
100%of his wealth. This imposes a total percentageholding cost of g on him. In
other words,his customersmust pay him g per cent on his currentwealth in order
to keep the dealer at his initial indifferencecurve. Part of the cost is due to the
de-diversificationcausedby dealingin few stocksand partis due to the assumption
of a level of risk not consistentwith the dealer'spreferences.Even if a dealerwere
to trade a fully diversifiedportfolio and therebymaintainRfE as his frontier,he
incurs costs because he moves to a lower indifferencecurve. A movement along
NB means the dealer has an undiversifiedshort position in the tradingaccount.
Once the dealeris at a nonoptimalpoint like A, the cost of anothertransactionis
the difference between the total percentagecost at A and at the new position
 The Supply of Dealer Services in Securities Markets 1135
U0

fR

FIGURE 1

resultingfrom the transaction.The cost may be negativeif the dealerdecreaseshis

position and trades a different stock so as to increase diversificationand reduce
risk. The case of a decrease in position accomplishedby selling another stock is
illustratedin Figure 1 by the move fromA to A1. Sinceg, Kg, the cost to the dealer
of the transactionsis negative,i.e., he is willing to pay customersto take him from
A to AI1.The cost would be positive were the transactionsto increasethe dealer's
position and reduce diversification.If the dealer specializes in one stock, his
movementis along a curve like ANB. If he makes a marketin many stocks, many
paths are possible.
The task now is to make more explicit the holding cost incurredby a dealerand
its relation to the size of the transactionand other variables.Before doing so the
assumptionsunderlyingthe model are set out.

II. AsSUMPTIONS

1. Dealer inventorypositions acquiredin the process of providingimmediacyare

financed solely at the risk free rate of interest,Rf. Thus dealerpurchasesof shares
are financedsolely by borrowingat Rf, and the proceedsof shortsales are invested
solely in the risk free asset. The dealer'spersonalwealth (his investmentaccount)
and the positionin the tradingaccount serveas collateralfor the borrowingof cash
1136 The Journal of Finance

or of shares. Ruled out under this assumptionis the possibility that dealers can
acquirefunds by sellingother stocks or that they use the proceedsof short sales to
buy other stocks. To permit this would simply transfer costs of immediacy to
anotherdealer.
However,by appropriatechangesin his bid-askquotationsthe dealerencourages
transactionsby the public that will rebalance his portfolio. In other words, the
dealeracts passivelysettingprices and letting the public choose which stocksit will
purchase from him and which stocks it will sell to him. Portfolio adjustments
arisingfrom his dealeractivity are thereforerestrictedto those stocks in which he
makes a market,i.e. his tradingaccount. The assumptionof constant Rf could be
relaxed at the cost of complicating the model, and the possibility is discussed
below.2
2. The dealer is assumed to have a utility function over terminalwealth. Since
the vast majority of dealers are proprietorships,partnershipsor closely held
corporations,ascribinga utilityfunction to the dealeris realistic.3Alternativelyone
can assume the utilityfunctionbelongs to the ownersof a dealerfirm and that the
ownerscannot offset on personnelaccount the positionstakenby the firm (because
they are unawareof the positionsor becauseit is too costly for them).
3. The dealermakes estimatesof the "true"price and "true"rates of returnthat
would exist in the absence of transactioncosts. This "true"price is the discounted
value of the dealer'sexpectedequilibriumprice one period hence. This estimateis
derivedfrom the fundamentalcharacteristicsof the stock and need not be the same
as the estimatesof other dealers or investors.All rates of returnin the paper are
"true"rates of return.The analysis is a partial equilibriumanalysis of the dealer
industry and the determinantsof "true"equilibriumprices are for example not
treated.
4. The dealer makes one transaction per trading interval during which the
stock's price does not change. Prices may change between tradingintervals.In a
one period world, the dealer buys or sells shares in the first tradinginterval and
becomes subject to one period of uncertainty.The period is assumed to be very
short,and certainapproximationsto be specifiedlater may be justified.The world
ends in the second tradingintervalwhen the dealer'sinventoryis liquidatedat the
equilibriumprice of the second tradinginterval.
Certainother assumptionsof lesser importanceare detailed in the development
of the model. These involve constraints on the utility function and certain
approximations.

2. This assumptionis realisticfor dealersthat also do a brokeragebusiness(as in the OTC market)

and have,on deposit,customersecurities,that can be borrowedat no cost. It is also realisticif aggregate
borrowingsof stockare smallsince competitionby lenders(owners)of stock would tend to drivedown
the cost to the borrowerof stock. Discussionswith nonbrokerdealerssuggestthat thereis at presenta
sharingbetweenborrowerand lenderof the intereston the proceedsof short sales. Recognitionof this
fact changessomewhatthe resultsof the model and the implicationsare discussedat the appropriate
point in the paper.
3. A possible explanationfor the observedtendencyto find noncorporateforms of organizationin
the securitiesindustryis the frequencyof verbal commitments,the fulfillmentof which depends on
personalintegrityand the threatof personalbankruptcy.
 The Supply of Dealer Services in Securities Markets 1137

III. THE HOLDING COST FUNCTION

The dealer tradesa stock if the dollar compensationpaid him is enough to offset
the loss of utility caused by deviatingfrom his initial portfolio. In other words he
will requirethat expectedutility of terminalwealthof the initial and new portfolios
be the same:
EU(W*)= EU(W) (1)

where W*= terminalwealth of the initial portfolio.

W= terminalwealth of the new portfolioafter the transaction.
indicatesrandomvariable.
The initial portfoliois a combinationof the dealer'sinvestmentaccount (repre-
sented by point N in Figure 1) and his initial tradingaccount.Thus

w*=WO[1+kRe+ Q RP+ 1-k-- (2)

where WO= initial wealth.

k = optimal fraction of the dealer's wealth invested in portfolio E, a
constantbecause of assumption1.
Re= returnon portfolio E-the optimal efficient portfolio of risky assets.
(Under homogeneousexpectationsthis would be the so-called market
portfoliothat includesall risky assets.)
QP= "true"dollar value of stocks in tradingaccount. Although one period
remains, the dealer is allowed to enter the trading interval with a
non-zero trading account, acquired in prior periods. If Q= 0, the
initial portfolio is the desired portfolio-point N in Figure 1. Qp 0
according to whether the dealer has long or short position in the
tradingaccount.
R = rate of returnon the tradingaccount.
Rf= risk free rate.
To simplifyexposition,(2) may also be writtenas follows:
W*=Wo(I+R*) (3)
where RP*is the rate of returnon the initial portfolio and is defined by the last
three termsinside the bracketsof (2).
Terminalwealth under the new portfoliois given by:
W= WO(I+ R*) + (I + Rj)Q - (I + Rf)(Qi -C ) (4)
where Qi= "true"dollar value of the transactionin stock i, the stock in which
immediacyis being provided.Negative values indicate a sale; positive
values, a purchase.
Ri = rate of return on stock i.
Ci presentdollarcost to the dealerof tradingthe amount Qi. Ci is positive
1138 The Journal of Finance

or negative accordingas the transactionin stock i raises or lowers the

costs of holding the inventoryQP.
The dealer'scost, Ci,is incorporatedas in (4) because,undercurrentinstitutional
arrangements,Ci, is not paid explicitly to the dealer at the time he provides
immediacy.Instead the dealer trades at the bid or ask price different from the
"true"price of the stock. Thus he need borrowonly Qi- Ci to finance a purchase
the "true" value of which is Qi. On a short sale he earns interest on Qi+ C1
althoughthe present"true"value of the shortsale is Qij4
Approximating(1) by expanding each side in a Taylor series around the
respectivemeansand droppingtermsof orderhigherthan two yields5:

E[ U( W*) + U'( W*)( W* - W*) + I U"( W*)( W*-W*) X

=E[U(W)+U'(W)(W-W)+ IU(W)(W-W)] (5)

where the bar () over a variableindicates expected value and where tildes have
been dropped when the meaning is clear. Writing W and W* in terms of initial
wealth and rates of returnand takingexpectationsyields:

U(W*)+O+ U"( W*)W6a*

U(W)+O+ 2U"(W)[ W + Qi2a2+ 2 WoQicov(R *, Ri)] (6)

where a* and a are the varianceof rate of return of the initial portfolio and of

4. The dollarvalueof borrowingor lendingdiffersfromthe "true"valueof the transactionexactlyby

the cost, Ci, only in perfectcompetition.When competitionis less than perfect,the dealercan price
above cost, and thereforethe discountfrom "true"value exceedscost.
5. Droppingthesehigherordertermscan be justifiedby assumingthat the pricedynamicsfor stocks
is given by
R,= jliAt+ a,V@Z
WhereRi=rate of returnduringthe intervalAt.
- expectedrate of returnduringAt.
a,= standarddeviationof returnduringAt.
Z = normalrandomvariablewith zero mean and unit variance.
vt= time interval.
Droppingtermsof orderhigherthan two implies that termsin At raisedto powersof 2 or more are
dropped,which is justifiedsince At is assumedto be very small in this analysis:let W= WO(l+ R). In
termsof the above process
Wi= W0(I+,IAt+aKt Z)
Then in (5)
(W- W)=-W0aV?IZ
(W- W*)= Woa*.VKZ
Thereforetermsof orderthreewould involve(At)2.
 The Supply of Dealer Services in Securities Markets 1139

stock i. The followingapproximationswhich simplifythe problemcan be made:

U"(W*)= U"(W). (A. 1)

U(W)- U(W*)
U(-)-U(W_ ) = w- W* (A.2)
U'(W*)
The approximationsare legitimatein termsof the processdescribedin footnote 5.6
Using (A.1) and (A.2), (6) can now be writtenas:
I z [Q a2+2WoQ cov(R*,R.)[ WW*]O (7)

where z= - U"(W*) Wo/ U'(W*);the Pratt index of relative risk aversion.

Note from (3) and (4) that
W W = Qi(Ri -Rf ) + Ci(I + Rf )(8)

and, using the definitionof R*, that

cov(R*,Ri)= kaJie+ W (9)

where o.e is the covariancebetween the rate of return on stock i and the rate of
returnon portfolioE, and a.s is the covariancebetween the returnon stock i and
the return on the initial trading account. Furthermorek, the desired holding of
portfolioE (representedby point N in Figure 1) can be eliminatedfrom (9) since it
dependson the utilityfunctionand the knowndesiredopportunityset (the line RfE
in Figure 1). By setting the slope of the dealer'sindifferencecurve equal to the
slope of the desiredopportunityset, it can be shown that7

Re-Rf
k= e (10)
ZoJe

6. On the basis of the processin the precedingfootnote,(5) can be writtenas

U(W*)+ 2 U"( W*)W2G2At= U(W)+ U"(W)W02A2t (F6.1)
A.1 and expand U"(W*)aroundW by one term:
Now consi~der
U"(W*)= U''(W)+(W*- Ww"(W). (F6.2)
Substitutingfor (W*- W) on the basis of the processin fn. 5 and substituting(F6.2)in (F6.1) will show
that neglectingthe second term in (F6.2) neglectsa term in (At)2which is small. Followinga similar
procedurefor A.2, one gets
I (F6.3)
U(W )-U( W*) = (W_ W*)U,( W*) + (W_ W*)2U"( W*)
Or
I
U(W) _ U( W*) = (W W*) U'( W*) + W0( t- _*)2(At)2U"( W*) (F6.4)
The last termin (F6.4) involves(At)2and can thus be neglected.
7. For Qp 0, terminalwealthis:
W*= Wo[h+R*= Wo[I +kRe+( -k)Rf].
1140 The Journal of Finance

Substituting(10) in (9) and (9) and (8) in (7) yields:

2Z Q22
+ z p;,JC(l+Rft)-Q[(Ri-RfJ)-(Re-RfJ Cie _- (II)
2 o WO'4 J/ 2L

and

z Q~~~~~~~~~~~~~~~[0
WO iP i2 W0 i(i-t-R-g i2e
ci = l+Rf (12)

Portfolioequilibriumfor the dealerrequiresthat:8

R-Rf =j(R Rf) 2a

-Rf (13)
Ge

This result depends on Assumption1-that the dealer can borrowand lend at Rf

and assumessecurityi is in his investmentaccount.Given (13), the last termin the
numeratorof (12) is zero. Note also that letting Rf= 0 in the denominatorof (12)
has minimal effect on (12). These modifications yield the dollar holding cost
functionwhich is to be viewedas an incrementalcost functionsince it refersto the
cost of a single additionaltransactionundertakenin the tradinginterval:

Ci WO
p 2W1 l (14)

The percentage cost is:

i
= ci= aipQp + 2 Qi (15)

The holding cost of takinga position in stock i depends on:

(1) Dealer characteristics-relativerisk aversionz, and dealerequity W0.Of two
dealersin the same stock, the one with largerz and/or smallerW0chargesa higher
fee for taking a position of given size; or, at the same fee, would take smaller
positions.

Using-thisdefinitionof W*,the differentialof the L.H.S.of (6) is:

dEU( W*)= aU( W*) a( W*)dk* + Utt(W*)WJ2.dau.

aW* ak*
Note that aW*/ *= WOand a= kae,set the differentialequal to zero and solve for the slope of the
indifferencecurve:
dR* u",(W*)Wke
do. U'(W*)
Settingthis equal to the slope of the opportunityset, (Re- Rf)/tae,yields (10).
8. Under homogeneousexpectationsthis would be the Sharpesecuritymarketline, an equilibrium
relationship,with portfolioE being the marketportfolioand aie/ae2 the well-known"Beta"coefficient.
 The Supply of Dealer Services in Securities Markets 1141

(2) Size of the transactionin stock i, Qi.Total cost rises as the squareof Qi, and
percentagecost rises linearlywith Qi.
(3) Characteristicsof the stock-variance of returnand the covariancebetween
the returnon stock i and the returnon the initial tradingaccount portfolio.9Note
that the covariancewith the investmentaccountdoes not enter.
(4) Size of the initial position in the tradingaccount, Qp.If Qpis positive (and
a. > 0), the cost of buying stock i is largerthan if there were no initial position.
donverselythe cost of sellingstock i is smallerthan if therewere no initialposition.
In Figure2, Ci is plottedas a functionof Qi using some reasonablevalues for the
remaining variables. Placement of the curves depends on the dealer's initial
position, QP,and the size of aip.If Qp= 0 or if sip= 0, dollarcost has a minimumof
Qi= 0. If Qp# 0 and sip> 0, the minimumis at Qit 0 accordingas Qp O. 0 A notable
aspect of (14) illustratedby Figure2 is its symmetry-a sale of given size costs the
dealer the same amount as a purchaseof given size. Assumingthat the dealer has
no initial holding in the trading account and is thereforeat N in Figure 1, this
result implies that points A and B in Figure 1, which representlong and short
positions of the same amount, lie on the same indifference curve, as shown.
AlthoughB is much fartherinside the efficient frontierthan A, it is closer to the
level of risk desired by the dealer and these two factors are offsetting. If the
probabilityof purchasesby the dealerequals the probabilityof sales by the dealer,

C1 Ci I Q, = 50,0ooo0

1 000

._~~~~~~~~~~ 500
pR

-150,000 -100,000' -50,000 50,000 100,000 150,000

FIGURE 2

9. Note that ai2and aipare not directlyobservablesince they dependon variabilityin "true"returns.
Observedvariabilityof returndependsas well on the cost of immediacywhich is reflectedin bid and
ask pricesand whichin turndependson the volumeof tradingand othervariables.
 1142 The Journal of Finance

the symmetriccost function implies that the optimal inventory in the dealer's
tradingaccountis zero; or, in otherwords, that the optimaloverallportfolioof the
dealer is the same as that of any nondealer with the same preferences and
expectations.Thus even afterbecominga dealerthe desiredportfolioremainspoint
N in Figure 1.
Inability of the dealer to borrow and lend at the same rate of interest can
eliminate the symmetryof the cost function and could complicate the problem
slightly. In particularsuppose the dealer can borrow at Rf but can lend the
proceedsof shortsales only at a fraction,E(,of Rf. This affects the thirdtermin the
numeratorof (12) because eRf replaces Rf. Given (13), the term does not go to
zero but to Q'Rf(E)- 1),where Q. <0 is the dollarvalue of short sales requiredand
e)=I if Qi7=0 and e <I if Q <0.'0 The effect is to raise Ci by the amount of
interest not earned on the proceeds of a short sale. Dollar costs are therefore
greaterfor Qi<0 than for Qi>0. Given a symmetricprobabilitydistributionon
purchasesand sales, the dealer tends, in this case, to keep a positive inventoryin
the tradingaccountto avoid the extracost of shortselling."
It should be pointed out that 0 can be made a more complicatedfunction of
other variables.For exampleif the bank is concernedabout default risk on dealer
borrowings,the logical variableis the dealer'sdebt-equityratio, L = (Qi+ QP)/WO,
where dE)(L)/dL>O. Such a modificationis straightforwardand would tend to
raise dealer costs above those in (14). However, since the theoreticalform and
empiricaljustificationfor these modificationsare not clear and since these modifi-
cations would not alter the character of the final model while adding to its
complexity,the modificationsare omitted from furtherconsideration.

IV. BID AND ASK PRICES

The dealeris compensatedby purchasingsharesat the bid price, pb, usuallybelow

the "true"price, P*, and by selling shares at the ask price, pa, usually above the
"true"price. Consider a dealer who specializesin a single stock i. Curve cio in
Figure 3 representsthe percentageholding cost function of the dealer with no
initialpositionin the stock. Supposehe standsready to buy Qb > O and sell Q,. < 0,
and quotes bid and ask pricesthatjust cover his costs of doing so.'2 Then the price
of immediacy of a sale of Q b to the dealer is set at (P.O_
- Pbo)/
P.Os= cio(Qbo),and the
price of immediacy of buying from the dealer is (Pio- P1)/ Pio= cio(Qao).If a seller
appearsand trades Qb at P.bo,the dealer sets the new ask price in the second (and
final) trading interval at the new equilibrium price, Pia = Pi',+,, and the position

10. Since the dealermay have stock i in his tradingaccount,shortsales may not be necessarywhen
the dealersells stock i.
11. The observedtendencyto find positiveinventorymay be due to a numberof other factors.For
example dealers may be able to anticipatebuying by the public better than selling by the public.
Second, and probablymost important,there are often tax benefits to carryingstock on one's trading
accountratherthan one's investmentaccount.If the dealeris taxed as a corporation,the corporatetax
may be lowerthanhis individualtax. Long termlosses can receiveordinaryincometax treatmentin the
tradingaccount.
12. It is assumedthat the dealeris in a competitiveenvironmentfor the purposesof this illustration.
 The Supply of Dealer Services in Securities Markets 1143

P*- p

CC

b 11 ii c~~~~~~~~~~

io
Ci i)
(qb Ptl -- -
b -o i,^
io

iO iO) P*= C

Qio ~ ~ ~ Q
QiG i
1i 10 P

(Qa) -o
P*o ai
/~~~~~~~~~i |o

FIGURE 3

is sold. If a buyer appears and trades Q,Oat P,.O,the dealer sets Pb= pe in the
second tradinginterval,and the shortpositionis covered.
A dealerfaced with a tradeof Q,would like in the same tradingintervalto make
an offsetting transaction,in the same stock or some equivalent combination of
stocks, that perfectlyhedges his portfolio. Real world constraintsimposed in this
model are that the dealer cannot actively and immediatelytake such offsetting
positionsin the same stock or other stocks. However,the bid-askquotationin any
stock is set so as to encouragetransactionswhich reduce the risk of holding the
initial portfolio.This point is illustratedin Figure 3 by the line cil, the percentage
cost function for Qp>0, qip>0. Since in this case the dealer already has a long
position the returnon which is positively correlatedwith stock i's return,the bid
and ask prices are set so as to encourage sales by the dealer of stock i and to
1144 The Journal of Finance

discouragepurchasesby the dealerof stock i. Thus the bid price,P, l, is lower than
if there were no initial position and the ask price, Pial,is lower. If QP= Q,
= - 2(z/ WO)aiQi;and in this case the dealeris willing to pay customersto take
him back to his optimumportfolio (N in Figure 1) an amount equivalentto the
cost of holding the risky position. If it is assumed he has been paid off for the
immediacy costs of trading Qp, the dealer just breaks even and is properly
compensatedduringeach time intervalfor bearingrisk.
The percentagespreadis the percentagedifferencebetweenbid and ask price, or
just the verticaldistancebetweenthe two percentagepricesin Figure3. The spread
function correspondingto (15) for Qib = Qa = IQil is:
pa - pb

Si = ' (Q,b)C ,(Q,a)= W a7IQI (16)

whichis independentof the initialinventoryof the dealerand does not involve any
covarianceterm. Thus, if the dealerpricesjust to cover the costs of each transac-
tion, the spread(but not the bid or ask price)is independentof the initialinventory
and thereforeholds for the dealer in many stocks as well as for a dealer in one
stock.This resultdependson the assumptionsthat previousinventoryholdingcosts
are sunk costs and that only one stock is tradedper tradinginterval.
V. OTHERCOSTS
In additionto holdingcosts the dealerincursothercosts that shall receiverelatively
brief treatmenthere. First the dealer incurs certain explicit costs-called order
costs-in carryingout a transaction.These costs which are incurredby brokersas
well as by dealers include the cost of labor, the cost of communicating,and the
cost of clearingand recordkeeping.The simplestassumptionis that ordercosts are
a constant dollar amount, M, per transactionand thereforea declining propor-
tional amountM/ Qi per dollar traded.
A second cost arises if some investorstrade with the dealer because they have
superiorinformation.[See Bagehot(1971).Jaffe & Winkler(1976)on this point.] In
organizedmarkets,a dealerquotes a bid price at which he is willingto buy and an
ask price at which he is willing to sell withoutknowingwhetherthe next tradewill
be a purchaseor sale. Even if the dealer possesses inside information(because of
knowledgeof the book of limit orders,for example), such informationwould be
reflectedin bid and ask prices;and relativeto such knowledgehe is still subjectto
losses from investorswho have informationhe does not possess.On the assumption
that the dealercannot distinguishinformationtradersfrom otherswithoutinforma-
tion (liquidity traders),the dealer must increase his bid-ask spread vis-a-vis all
traders to protect himself against possible losses of dealing with information
traders.He widens the spreadso as to recoverfrom liquiditytraderswhat he loses
to informationtraders.
For the purposesof this paper,informationtradingcan be incorporatedinto the
dealer's cost function by recognizing that (13) does not hold in such a case.
Instead:
Ri Rf =(Re Rf ) 2Yai (17)
ge
 The Supply of Dealer Services in Securities Markets 1145

where a, is the expected returnon the informationpossessed by those that trade

with the dealer, assumed the same for buyers and sellers and independent of
transactionsize, and
fY 1 if dealerpurchasesshares.
-1I if dealersells shares.
In other words the dealer expects to earn less than he would in the absence of
informationaltrading, and his bid-ask spread must be wider as a result."3In
additioninformationaltradingmay cause ai2and cipto.be different,but this is not
assumedhere. The effect of (17) is to make the last term in the numeratorof (12)
nonzero and equal to - yak.
Modifyingthe percentagecost function (15) to include ordercosts and informa-
tion costs yields:
=
Ci= Q + 2~-j-a7Q
ipQ+ WoQI + i-a1 +- Qi (18)
(18)
WO
The adverse informationcost (at) and order cost M/Q, could undoubtedly be
treatedin a more complexway. For examplethe likelihoodof adverseinformation
may be a functionof the size of the transaction(becausethe rich are smarter).This
would result in higher bid-ask spreadsas Qiincreasesnot only because of larger
holding costs but also because of larger informationcosts. In principle the two
costs could be distinguishedbecauseinformationcosts are not a function of dealer
inventorylevel whereasholding costs are. Similarlyorder costs may in part be a
function of transactionsize. This would resultin a less steeply decliningper dollar
ordercost as a function of transactionsize.
Such modificationsare unlikely to change the basic shape of the cost function
(18) which is plotted on Figure 4. The differencebetween Figures4 and 3 is that
thereis now a discontinuityat Qi= 0 due to switchesin sign of y and the presence
of M.
A second more importantdifferenceis that thereis now an optimal-scalefor the
dealer because falling order costs are offset by rising holding costs. Adverse
information costs do not affect the scale decision because they are assumed
independentof transactionsize. If the dealer operates at minimumaverage cost,
[the minimumof (18)] output is at:

Qi* 2W
=+ ZG2 (19)

also shown in Figure4. Whetherthe dealer actually operateshere depends on the

numberof otherfirmsmakinga marketin the stock and on the existenceof special
skills of the dealer,etc.
A spreadfunction independentof Qpcorrespondsto (18):

Si= vI QiM+2ai+ (20)

The spreadfunction is a " U" shaped function of IQil

13. a, is analogous to the Jensen (1968) measure of mutual fund performance.
1146 The Journal of Finance

c.

_ 0 ~ ~ ~ ~ _Q

C.~~~~~~~~~~~~~~~~~~~~~~~~~~~C

1~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~0l

FIGURE 4

VI. ORGANIZATION OF D)EALERS

An importantpolicy issue is the organizationof dealers that regulatorypolicy

should foster -whether a monopolisticspecialistsystem such as has existed on the
New York Stock Exchangewhere no stock has had more than one dealer or a
competitivedealersystemsuch as has been permittedin Over-the-Counter stocks.14
Probablythe principalundesirableaspect of monopoly dealersis pricingabove
marginal cost, the extent of which depends on the elasticity of demand for
immediacy,a subjectnot consideredhere. Howeverthe organizationof dealersalso
affects the cost of dealerservices.
First consider the effects of changing from monopoly dealers to competitive
dealers in the short run where the numberof dealers is fixed. In a free market a

14. As of the Spring 1976 competingdealers were permittedon the NYSE, but few have as yet
appeared.
 The Supply of Dealer Services in Securities Markets 1147

given numberof dealersallocate themselvesamong stocks such that the cost of the
marginal transaction depends only on characteristics of stocks and not on
characteristicsof dealers.Wealthierdealers take larger positions (in the form of
holding more stocks or more per stock) than less wealthy dealersso that marginal
cost in any particularstock is equal across dealers.Similarlydealersless averse to
risk take larger positions than more risk averse dealers. To the extent that the
allocation of stocks to dealers under a monopoly dealer system is not optimal,
reorganizationinto a competitiVesystemproducesa net welfaregain to society by
equalizingmarginalcosts across dealers.Reorganizationinto a competitivesystem
benefits marketsin stocks with previouslyinadequatecapital or high z specialists
and harmsmarketsin stockswith previouslyexcessivecapitalor low z specialists.15
Second, if entry restrictionscause monopoly dealers to operate at too large a
scale (to the right of Q,*in Figure 4), competitionfrom new entrantswould push
costs to the minimum.The equilibriumnumberof dealersin a stock can be derived
using (19) if one is willing to assume that dealershave identicalcost functionsand
operateat the minimumaveragecost. Let IDl denote the absolutedollar value of
investors'tradingwith dealersin stock i at the price of immediacycorrespondingto
minimumaveragecost. Let I= absolute dollar value of minimumcost output.
The equilibriumnumberof dealers,d*, is the numberthat demand can supportif
all are operatingat minimumaveragecost. From (19) this is:

di*=1IQ*1IDil 2M (21)

For given long-rundemand,the numberof dealersis greaterin riskierstocks and

stocks in which individual dealers are more risk averse; the number is less the
greater the order costs incurred by dealers and the greater the wealth of the
individualdealer.
It is interesting to note a few policy implications of (21). It is sometimes
suggestedthat undercompetition,dealersin riskystockswould not be forthcoming.
However, (21) suggests that, ceterisparibus, more dealers will operate in risky
stocks.This is due to the fact that each dealerwill take a smallerposition.Another
policy question concerns dealer capital requirements(our WO).Regulatorsargue
that dealers should be requiredto maintain a minimum capital in each stock.'6
Under monopoly dealerships such requirements may be useful in ensuring
adequatecapital since entry of dealerswith additionalcapital is prohibited.(The
appropriatelevel of capital depends on the characteristicsof the stock and the
dealer and would be quite difficult to set precisely.)Under competitivemarkets,

15. However,the total gain of shiftingfrom a monopolyto a competitivesystemmay not be as great

as implied here. For example,monopoly profits may make it possible for regulatoryauthoritiesto
requiredealersto act as if they had more capitalor a smallerz.
16. On the NYSE, the specialistmust be able to carry2,000 sharesof each stock (e.g., $80,000for
sharespricedat $40). However,most of the fundsfor carryinginventorycan be borrowed.On the OTC
a dealermust have net capitalof $2,000.
1148 The Journal of Finance

minimum capital requirementsmay be counterproductive;if set too high, they

reduce the numberof dealersin a stock accordingto (21) and therebyreduce the
beneficial effects of competition. (Furthermore,minimum capital requirements
may have little effect in practice because it is difficult to compel utilization of
capital.)
A final issue concernsthe ease of entry and exit undera competitivesystem.To
the extent that entry and exit costs are zero, dealerswould fluctuateso that supply
is perfectlyelastic at the price of liquiditycorrespondingto IQi* even in the short
run. If short-runsupply is derivedfor a period in which the numberof dealersin
each stock is fixed, it is upwardsloping.This is realisticbecausethereare probably
entrycosts and becauseregulationsoften requiredealersto maintaina marketfor a
minimumperiodof time (6 monthsin the O.T.C.).Such regulationsare undesirable
in that they lengthenthe short-runand raise costs on average.

VII. MULTIPERIOD
CONSIDERATIONS
There is no guaranteethat the dealer can readily liquidate his inventory in the
second trading interval.Assume there is a cost, Di, of liquidatingthe inventory
should it still exist. This cost, a randomvariablein the first tradinginterval,can be
thought of as the paymentnecessaryto cause someone to accept the inventoryat
the new "true"price, or alternativelyas the differencebetween trade price (bid or
ask price)and the new "true"price that is necessaryto createsufficientincentiveto
othersto purchasethe dealer'sinventory.Under certainassumptionsone can view
D, as the (implicit)payment by the dealer to himself that makes him willing to
continueto hold the inventory.
It is helpful to think of events unwindingin the following time sequence.In the
first trading interval the dealer takes a position, Qi. He enters a period of price
uncertaintyin which thereis no trading.The second tradingintervalis dividedinto
two parts.The dealerentersthe second tradingintervalby pricinghis inventoryat
Pt*+, the new "true"price. Thus if he initially bought the stock (at P,tb), he sets
p,a +=P*t+1; if he initially sold the stock (at Pa) he sets t = Pt+ . In the
model of SectionII, the world ends here, and the dealeris assumedto liquidatehis
position at P,*t+ . In fact he may not be able to liquidate his inventory at Pi,t+. If
it is not liquidated,he sets at new concessionprice which deviatesfrom P,*t+, and
whichis sufficientto liquidatehis position.The dollaramountof the concessionon
all his sharesis Di, a randomvariablein the first tradinginterval.
The one period frameworkof section II can be maintainedbut modified simply
by noting that terminal wealth in (4) will be reduced by Di. The development
proceeds exactly as before except that some of the expressionsare more compli-
cated. The primarycomplicationarises in going from (5) to (6), which involves
writingE( W- W)2= a2(W) in termsof its components.With Di this is
a2( W)= WOa+ Qi +& (Di )+ 2 WoQicov(R*,Ri)
2 WOcov(R *, Di )-2 Qi cov(Ri, Di) (22)

Recall that Di is the concessionrelative to the "true"price and that R and R* are
 The Supply of Dealer Services in Securities Markets 1149

returnsbased on "true"prices. There is no reason to believe that D should be

correlatedwith these returns.Thus:
cov(R*, Di) = cov(Ri, Di) =0 (A.3)

With these assumptionsand otherwisemaking the same assumptionas in section

II, the simplifiedcost functionbecomes:
z z
C= alp QpQ,+Di + wa 2(Di)+ 2 WOQi (23)

Considernow a more precisespecificationof Di that is based on the assumption

that there is never partialliquidationof any prior transaction.This implies that D.
1 and the value Ci if
is a Bernoullivariatetakingthe value zero if Qi is sold at P*,t+
the position is not sold. The correspondingprobabilitiesare (1 - 7i) and 'Wi.Then:

D==rg C, (24a)

192(D ) =isi(I _gi)ci (24b)

Now suppose Ci = Ci, that the price concessionnecessaryto eliminatea position
equalsthe cost of assumingthe position in the first place. SubstitutingCi for Ci in
(24) and substitutingfor D and a2(D) in (23) yields
z lIz
aq,pQQ.+-
Q UQ2
I7QI
w0 ' W0(25)
2 WO ](5
WO

If the probabilityof a forced liquidationis zero, the cost functionis the same as in
(14).
The solution(25) can be statedin multiperiodterminologyby letting Ci represent
the cost to the dealerof continuingto hold the position acquiredin the first period.
Under this interpretation,the dealer does not liquidatehis inventoryat a conces-
sion price, but he continues to price it at the "true"price and to wait until an
investortradesin the opposite direction.This is not a true multiperiodframework
in which intermediatedecisionswould be allowed. It is the case that C,= C, under
the assumptionthat characteristicsof the stock (a72, cip)and the dealer (z, WO)are
unchanged over time and that the probability(,g) of holding the stock is un-
changed. Independence and stationarity of the distribution of returns and of
trading volume will lead to unchanged ai, ai , gi for given Qp.It is not unreasonable
to assume constantz. However WOchanges,and we must argue that the change is
too small to have a significanteffect.17
When viewed as the cost of continuingto hold the stock, it is naturalto specify
the dealer'scost in termsof the numberof periodshe expectsto hold the inventory

17. On average Wo increasesover time, which would reduce costs. This partly offsets the pre-
ponderenceof factorswhichlead to increasedcosts.
1150 The Journal of Finance

when it is pricedat P*. Under the assumptionof the stationarityand independence

of the distributionof volume, there is a simple relationshipbetween the expected
holding period i and gi:
T
T = 1I+ E ihh(I - gT) (26)
h=1

whereh = numberof periods,inventoryis held

T= total numberof possible periods.
Letting T-*oo (26) can be shown to be:
= 1 (27)

Then, from (25)

Z Z2l
,,Q, Q.+12 W
C WO W1q (28)

where
1 z 1 Ti-1 _

*i 2 WO Ti

This differs from (14), only in that the expected holding period multiplies the
per-periodvariance and covarianceand in the second term in the denominator,
which is small. If Ti = 1, (14) results.

VIII. SUMMARYAND CONCLUSIONS

A dealer cost function composed of holding costs, order costs and information
costs is developed.The emphasisis on the holdingcost componentwhich is derived
on the assumptionthat the cost is an amountwhich maintainsthe dealer'slevel of
expected utility of terminalwealth in responseto transactionsimposed upon him
by the public that tend to move him away from his optimalportfolio.The holding
cost depends on the dollar size of the transaction,the variance of return of the
stock-being traded, the size of the initial holdings of all stocks in the dealer's
tradingaccount, the covariancebetween the returnon the stock being traded and
the returnon the tradingaccount,the wealth of the dealer,and his attitudetoward
risk. Dollar holding cost for the incrementaltrade is a quadraticfunction of the
size of the position acquired,and percentageholding cost is thus linear.Under the
assumptionthat dealersare able to earn full intereston the proceedsof short sales,
the cost function is symmetric-that is, the cost of going short equals the cost of
going long. Under certain simplifyingassumptions,the multiperiodholding cost
function is shown to be quite similarto the one period function, differingonly in
that the holding period enters the function (or, equivalentlyin this model, the
 The Supply of Dealer Services in Securities Markets 1151

probabilitythat the dealer is unable to dispose of his inventoryat the equilibrium

price after one period).
The order cost is a minimumcost per transactionwhich thereforedeclines per
dollaras the size of the transactionincreases.Falling ordercosts and risingholding
costs determinean optimumscale of operationby the dealerin each of his stocks.
Informationcosts arise when investorstrade on the basis of superiorinformation,
which adverselyaffects the dealer'sexpectedreturnon his inventory.
The paper also examines some policy issues related to the organization of
dealers. Monopoly dealers are undesirablenot only for the standardreason that
they price above marginalcost but also becausethe assignmentof stocks to dealers
may be arbitraryand can result in a nonoptimumdistributionof dealer wealths
and risk attitudesacross stocks and because limits on entry may cause dealers to
operateat a nonoptimalscale. The equilibriumnumberof dealersin a competitive
systemis also considered.Given demand,the numberof dealersincreaseswith the
risk of the stock and the risk aversionof the "representativedealer".The desirabil-
ity of regulatingminimumcapital requirementsand imposing other entry condi-
tions is questioned.
REFERENCES
1. W. Bagehot(pseud.)."TheOnly Game in Town,"Financial Analysts Journal (March/April 1971).
2. George J. Benston and Robert L. Hagerman."Determinantsof Bid-Asked Spreads in the
Over-the-Counter Market,"Journal of Financial Economics, 1 (Dec. 1974),pp. 353-364.
3. ThomasE. Copeland."A Model of Asset TradingUnder the Assumptionof SequentialInforma-
tion Arrival,"Journal of Finance (September1976).
4. HaroldDemsetz."TheCost of Transacting,"QuarterlyJournal of Economics, 82, No. 1 (February
1968).
5. ThomasW. Epps. "The Demand for Brokers'Services:The Relation BetweenSecurityTrading
Volumeand TransactionCost,"Bell Journal of Economics, 7 (Spring1976).
6. Institutional Investor Study Report of the Securities and Exchange Commission, 92nd Congress,1st
SessionHouse DocumentNo. 92-64,March 12, 1971.Washington:G.P.O., 1971.Part4, Ch. 12.
7. J. Jaffeand R. Winkler."OptimalSpeculationAgainstan EfficientMarket,"Journal of Finance, 31
(March 1976).
8. Michael C. Jensen. "The Performanceof Mutual Funds in the Period 1945-1964,"Journal of
Finance, 23 (May 1968).
9. SEC,"PolicyStatementof the Securitiesand ExchangeCommissionon the Structureof a Central
MarketSystem,"in Securities Regulations and Law Report, No. 196,April4, 1973,publishedby
Bureauof NationalAffairs,Inc.
10. H. R. Stoll. "The Pricingof SecurityDealer Services:An EmpiricalStudy of NASDAQ Stocks,"
Journal of Finance, this issue.
11. Seha M. Tinic. "The Economics of Liquidity Servites," Quarterly Journal of Economics, 86
(February1972).
12. Seha M. Tinic and Richard West. "Competitionand the Pricing of Dealer Services in the
Over-the-Counter Market,"Journal of Financial and QuantitativeAnalysis, 7 (June 1972).