• Non ci sono risultati.

Advertising as a Signal of Quality, A New Explanation

N/A
N/A
Protected

Academic year: 2021

Condividi "Advertising as a Signal of Quality, A New Explanation"

Copied!
27
0
0

Testo completo

(1)

p

a

p

a

p

a

Figure 4 Figure 5

(2)

dH

pH

dL

pL

pH

pL

p

a

p

a

Figure 1 Figure 2 Figure 3

(3)

thatconfrontsalowqualit yopp onent quotingpricec. This pro tis(p H

0c)(1+

c0p H

)0a . Thereforesincethe playofthe equilibrium recommendation implies

anull pro t tob oth rms, the condition for nodeviationto  H writes as (p H 0c)(1+c0p H )0a0: (7.1)

It follo wsthataseparating equilibrium existsifthe threeconditions (A),(B)

and (7.1) are met by H

.

Noteforfurtherreferencethatthe crossingof (7.1)andofa (p)o ccursexactly

for a value of p H

= 2(1 + c)=3  p 3 H

, the full information price for rm H.

Therefore the full information price mayb e used as a signal if coupled with an

advertising campaign costing a (p 3 H

). In factalso some sligh tly lower prices may

b elong to the admissible region for  H

. Recalling formProp osition 2 that when

c< 1=2 the region for whic h signaling with a p ositiv eadvertising exists is

non-empt yand it implies a price in [2c;1], one can easily c heck that condition (7.1)

issatis ed for somecouples (p H

;a ) whenp 3 H

2[2c;1], namely when c<2.

The issue of existence is thereforeresolv ed.

One ma y note, to complete the discussion, that there can exist many (in

general an in nite numb er of) equilibria when the regions as described in the

preceding paragraph are non-empt y. While we do not address here the issue

of selecting among equilibria, it is in teresting to note that solely based on the

c haracterization of equilibria, one can conclude that in some cases equilibria

withoutadvertising donot exist.

Of course, lik e in the case of a monopoly, one can construct b elief systems

thatsustainonlyp o olingequilibria,but sincethefo cus hereisonseparationand

notontheissueofp o oling v ersusseparationwedonot pursueadetailedanalysis

here.

Thesystemofb eliefsin(i)-(iii)sharesafeaturethatismostgeneralingames

with twosignal senders,namely that b eliefs dep end up on what b oth playersdo,

or, inshort, b eliefs are "correlated".

It is p ossible to generate separating equilibria of this game also by using

systems with "uncorrelated" b eliefs. For instance, b eliefs of this typ e assign

probabilit y one of b eing H to any rm playing a strategy that satis es some

conditions(inparticular(A)and(B)),whatevertheotherplayerdo es. Obviously

this impliesthat the observationof a strategy couplelik e( H ; H ) impliesb elief b ( H j  H

) =1. This is valid only if ! comprises the state (H,H). It is p ossible

to show that there exist systems of non correlated b eliefs sustaining separating

(4)

H

equilibria m ust) the c haracterization in Section 3 ab ov e based up on conditions

(A) and (B). To eliminate the in uence of the value taken by b ( H

j  H

) up on

the result that advertising ma y b e necessary, we assume that b ( H

j  H

) = 0,

so as to render as large as p ossible the set of situations in whic h a price alone

strategy ofthe typ e H

=(p;0) can work asa signal.

Assume that consumers b eliefs have the follo wing structure. (i) If they

ob-serv ethat no rm uses advertising and that prices are equal to c at b oth rms

consumers b elieve that the state is (H,H). (ii) If only one rm plays  H

and if

this satis es (A) and (B) ab ov e, then they b eliev e that this rm is H and the

otherisL, unlessthe other rm also playseither  H

, or(0;0)inwhic hcase they

b eliev e that state is (L,L). (iii) An y other strategy couple di eren t from those

contemplatedby (i)and (ii) inducesthe b elief that the state is(L,L).

Formally: (i) (b i ( i j j );b j ( i j j ))=(1;1)if ( i ; j )=((c;0);(c;0)); (ii) (b i ( i j  j );b j ( i j  j )) = (1;0) if ( i ; j ) = ( H ;x ) with x2= f H ;(0;0)g and with  H

satisfying conditions (A)and (B);

(iii) (b i ( i j j );b j ( i j j ))=(0;0)otherwise.

Note that (iii) encompasses the symmetric couple ( H

; H

) so that these

b eliefsimply that condition (A) iswrittenwith b ( H

; H

)=0.

It m ust b e shown now that the equilibrium strategy prescribe that: if the

state is (L,L) then  i

and  j

are (0;0) at eac h rm; If the state is (H,L) they

are  H = (p H ;a ) by rm H and  L = (p H =2;0) by rm L, with  H satisfying

condition3.2,3.3 andaconditionthatshallb ede nedb elo w; nally,if thestate

is(H,H) they are (c;0)by eac h rm.

First, if the state is (H,L) or (L,H) the strategies played m ust b e  H

, and

(p H

=2;0)resp ectively. The b est deviations from equilibrium under the assumed

system ofb eiefs are  H

for rm L and (p H

=2;0)for rmH. But thesedeviations

cannotb e pro tablesince H

satis esconditions(A)and (B)ab ov e. This couple

of actionsis therefore acouple of m utualb estreplies.

Second,ifthe stateis(L,L)then theplayofanullpriceandof noadvertising

byb oth rms m ust constitute a couple of b est replies: if a rm deviates to any

other strategy with a p ositiv e price it shall still b e p erceived as L, therefore it

cannot increase its pro t ab ov e zero. Therefore the couple ((0;0);(0;0)) is a

coupleof mutualb est replies.

Itremainstob ev eri edwhetherunderthestate(H,H)thecandidatethatthe

b eliefs prop ose, namely the couple((c;0);(c;0)) is a pair of m utualb est replies.

It is immediate to see that the only p ossible deviation left to either rm by the

b eliefs is the play of strategy  H

instead of (c;0). That deviation generates an

(5)

Pro ofof Prop osition 1

Pro of: Supp ose that (p;a ) satis es b oth (i) and (ii). Supp ose further that

(p;0) do es not satisfy either one or b oth (i) and (ii). Since (p;a ) satis es (ii)

(1+v0p)(p0c)0a0v+c0 then, a fortiori (1+v0p)(p0c)0v+c 0.

The lowest value of a that can b e associated to p is therefore giv en by (i) and

is (1+v 0p)p 0v  a +

. Then, assume  = (p;a +

) . By continuity of the

function (1+v 0p)p there exists a price p 0 > p such that (1+v 0p 0 )p 0 = v.

Hence assume the strategy  0

= (p 0

;0), whic h v eri es constraint (i), is used

instead of . Constraint (ii) for  0 writes as (1+v 0 p 0 )(p 0 0 c) 0 v +c 

0. Since  satis es (ii), for this inequality to b e v eri ed, it is sucien t that

(1+v0p 0 )(p 0 0c)0v+c(1+v0p)(p0c)0v+c0a +

. But this writesas

(1+v0p 0 )p 0 0c(1+v0p 0 )0v 0 (1+v0p)cor(1+v0p 0 )(1+v0p), whic h is true since p 0

> p and since b oth expressions in parentheses are non-negative.

The argumen tapplies also for allstrategies (p,a) with a>v +

and it is therefore

complete. 2

Pro ofof Lemma1:

Pro of: Supp ose that at an equilibrium rm L plays an action  0

di eren tfrom L

whenthestateis(H,L).Byde nitionofaSeparating

Equilibrium rm Lisp erceiv edasL anditsopp onent asH.However,

bydeviatingtostrategy L

,ifitisstillp erceiv edasLwithprobabilit y

one it will play according to its b est reply against  H

, and if it is

p erceiv edas L with probabilit yless than one, it will receiv e ahigher

demand thanif it iswith probabilit yone, and itspro t will increase,

therefore, in either case there is no strategy available to rm L that

dominates L

: 2

7. Appendix 2: Existence

The necessary conditions for the existence of a separating equilibrium are

suf- cien t to c haracterize the nature of the signal used, but are not sucien t to

guarantee existence ingeneral. Recall furthermorethat the analysis has b een so

farconcernedonlywith theactionsprescribedinstates(H,L)and(L,H) ,andthe

otheractions haveb een left unsp eci ed.

The presen tSection lls this gap and shows that existence is guaranteed, in

(6)

Journalof PoliticalEconom y , 86, pp. 485-503.

[16] Shaked,A.andJ.Sutton[1982]: "RelaxingPriceCompetitionthroughPro

d-uct Di eren tiation". Review of Economic Studies ,49, 3-13.

[17] ShapiroC.[1982]"Consumer Information,Pro duct Qualit y,and seller

Rep-utation", BellJournal of Economics, 13, pp. 20-35.

[18] ShapiroC. [1983]:"Premiumsfor HighQualit yPro ducts as Ren tsto

Repu-tation",QuarterlyJournal of Economics, 98, pp. 659-680.

[19] Sp ence,M. [1973]"JobMark etSignaling",QuarterlyJournalof Economics

87, 355-374.

[20] Tirole, J.[1989]: The Theory of Industrial Organization. Cam bridge, MA,

MIT Press.

[21] Wolinsky,A.[1983],"Prices asSignalsof Pro duct Qualit y",Review of

(7)

[1] Bagwell, K. and G. Ramey [1990], "Adv ertisingand Limit Pricing". Rand

Journalof Economics ,19, 59-71.

[2] Bagwell,K.andM.Riordan[1991],"HighandDecliningPricesSignalPro

d-uct Qualit y",AmericanEconomic Review 81, 224-239.

[3] Co op er, R.andT.Ross[1984], "Prices,Pro ductQualities andAsymmetric

Information", Review of Economic Studies 51, 197-208.

[4] Dorfman, R. and Steiner, P. [1954]: "Optimal Adv ertising and Optimal

Qualit y".AmericanEconomic Review , 44, 826-836.

[5] Fudenb erg, D. and J. Tirole [1986], "A Signal-Jamming Theory of

Preda-tion",Rand Journal of Economics 17, 366-376.

[6] Gabszewicz, J. and J-F. Thisse [1979]: " Price Competition, Qualit y, and

Income Disparities".Journal of Economic Theory,20, 340-359.

[7] Kihlstrom, R.and M.Riordan [1984],"Advertising asa Signal", Journal of

PoliticalEconom y94, 796-821.

[8] Klein B. and Leer K. [1981]:" The role of Mark et Forces in Assuring

Contractual Performance".Journal of PoliticalEconom y ,81, pp.615-641.

[9] Martin, S. [1993]: Advanced Industrial Economics. Cam bridge,MA, Blac

k-well.

[10] Matthews,S.A. and D. Fertig [1990],"Advertising Signals of Pro duct

Qual-it y", NorthwesternUniv ersity,CMSEMS D.P. 881.

[11] Milgrom,P.and J.Rob erts [1986],"PricesandAdv ertisingSignalsof Pro

d-uct Qualit y",Journal of PoliticalEconom y94, 796-821.

[12] Nelson, P. [1974],"Advertising as Information", Journal of Political

Econ-om y81, 729-754.

[13] Riordan, M. [1986]: "Monopolistic Competition with Exp erience Go o ds",

QuarterlyJournal of Economics ,90, pp.629-650.

[14] Sc hmalensee,R.:[1972]: TheEconomicsof Advertising.Amsterdam:

(8)

is quite obvious, while it is not so in the context of a monopoly. By fo cussing

on the single- rm problem some authors hav e stressed the importance of low

in tro ductory prices as a signaling p olicy (see Sc hmalensee[1978]), while others

hav eattributed more importanceto high-price p olicies as signaling devices (see

Bagwelland Riordan[1991]). The explanation we usehere do es not con ictwith

either those views of low (or of high) in tro ductory prices as signals, although

we stress the importance of advertising componentsof a signaling strategy that

allow a price p olicy to b e a signal 21

. In our example a price higher than the

fullinformation price|whenrev elationissp ontaneous the fullinformation price

itself|ma ysignalwithoutadvertising. Buttherearesituationswhereadvertising

is necessary for prices higher than, equal to, or lower than the full information

priceto separate.

The oldideathat theen tryof new rms mayb edeterredthrough advertising

nds itscounterpartin our model in the idea that signaling through advertising

allows a lower price than signaling through price alone and reduces the mark et

share for the rivals. The low qualit y rm by advertising can use a price low

enough to discourage en try of a rm with a still lower qualit y, while without

advertising the same price cannotdeter the en trants.

This typ e of advertising is most lik ely to continue over time and to vary in

a pro-cyclical fashion, as it is directed not so m uch to separate one incum b ent

from another incum b ent but to discourage, through time, the mimic king from

p otentialen trants.

21

ThesignalingroleofpricesisalsothefocusofWolinski[1983]andCo op erandRoss[1984]. Inthese works,ho wever,some consumersare informedab outpro duct qualitiesor canacquire informationatacost(i.e. thepro ductisasearchgoo d). Thequestiontherestudiedisthenthe extenttowhich equilibriumprices transmitinformation frominformed touninformed agents. Itishighprices thatha vereceivedthemostattentionin thoseworks.

(9)

cyclical b ehaviorof demand. Forinstanceassume that the mass of consumers is

M t

wherethesubscripttdenotesadate. Ifadvertisingisusedasanen trybarrier,

then it p ersists; howeverit m ust also varyaccording todemand conditions: the

no-jamming constraintprev enting a op to en terand imitate rm L b ecomes:

p L 2 (p H 0p L 0 v 2 )M t a L : (4.3) As M t

variesalong the cyclesovariesthe minim umamoun tof advertisingof

rm L.

As a conclusion weget that advertising varies pro-cyclically.

5. Concluding remarks onadvertising, signaling, andprice

policy

When advertising is used as an en try barrier, since in that case it is clear that

aprice-alone strategy can always b e mimic kedbyan en trantof lowerqualit y,it

arisesasanaturalequilibriumb ehavior. Whenitisjustusedbyahighqualit yto

separatefroma low qualit yitsrelativ e meritsov era price-alonestrategy should

b e in tuitively explained as they seem to apply for the duop oly and not for the

monopoly case.

Note rst that to obtain separation when one rm is of high qualit y and the

other is of low qualit y the former m ust discourage imitation by visibly wasting

some resources. This can b e done by an abnormal increase (for the case of a

decreaseinpriceseecommentsb elo w)inprice,orbyaddingthe xedcostsof an

advertisingcampaign. A highprice,howeverispainfulintermsoflost customers

that switc h to the rival supplier. This e ect is larger the more elastic is the

rm's demand to the price di erence. For high enough prices if qualit y do es

not command a high premium the high qualit yis out of the mark et and cannot

separate. The qualit ypremium for rm H inour model is higher (in p ercen tage

terms)the loweristhe parameterv. Ahighv inducesalowp ercen tagedi erence

inutility(u H 0u L )=u L =(m0p H +p L )=(v0p L

). Thisaccounts inpart for the

dicult ytoobtainseparation through a highprice and noadvertising when v is

high. By using advertising and an appropriate price together, the high qualit y

rm ma y succeed in discouraging imitation without raising its price to o m uch

ab ov ethat of the rival.

Second, the price of the low qualit y rm at a separating equilibrium is a

decreasing functionof the price of the highqualit y rm (in monopolythis is not

so). Thismeansthatthe lowerthe pricethatisused bythe latterforseparation,

the lower the margins that it can enjoy by imitating the low qualit y rm. This

mak es it easier for the high qualit y to b e discouraged from mimic king the low

qualit y. Also in this resp ect the use of advertising, since itallows areduction in

(10)

f L L His m 3 =p H 0p L

0(0: 5)v. The totaldemand addressed tothe two rms Land

0is thereforeequal tom 3

.

Assume then that at state (L,H,0) rm L m ust deter the op with a

(sepa-rating) strategy  L = (p L ;a L

). A necessary condition for this strategy to deter

the y er-by-the-nightisthat if this en tersand playsexactly  L

it mak esa loss.

The no-jamming constraint prev entinga op fromen tering and imitate rm

L is: p L 2 (p H 0p L 0 v 2 )0a L 0: (4.1)

The R.H.S. in the inequality ab ov eisjusti ed by the factthat the full

infor-mationpriceand pro t of a op iszero. The L.H.S.issimply the pro t of a op

that enters and imitates a rm of typ e L that is playing (p L ;a L ) against a rm H that isplaying (p H ;a H ). If a L

= 0 the y er-by-the-night always nds it pro table to en ter so that

advertising isnecessary for rm L tok eepthe op out 20

. Obviously,anecessary

conditionforthelowqualit y rm toseparatefromthe opthroughadvertisingis

thatbysodoingitmak esapro t higherthanifitaccommodatesthe en tryofthe

y er-by-the-night. The rm L pro t after accommodation is p L 2 (p H 0p L 0 v 2 ).

Andthe price p L

that maximizesthis pro t is equal top `

(1=2)(p H

0v=2). A

deviationfromthe en tryprev entingprice p L thatb elongs to L =(p L ;a L ),top `

en tailsapro tequalto(1=4)(p H

0 v=2) 2

. Thereforeaconditionthat L =(p L ;a L )

m ustsatisfy is that

p L (p H 0p L )0a L >(1=4)(p H 0v=2) 2 : (4.2)

Finally, the b ehavior of rm H m ust assure that rm L do es not want to

imitate rm H inaduop oly,i.e. that p L (p H 0p L )0a H >p L (p H 0p L )0a L (this

againis the no-jamming condition).

The lesson from this second example is that the theory here exposed can

ac-count for (i) low quality rms advertising together with high quality rms, (2)

persistence of both rms advertising over time.

Note that the existence region for high-price and no advertising for rm H

shrinks further if the low qualit y rm advertises. In fact the L.H.S. of all

con-ditions (A) and (B) is lowered by the xed amoun t a L

, so that the incentive

constrainta z

(p) isshifted upward,and therange ofv-valuesthat are largerthan

the ro ot z c shrinks accordingly asa L is increased. 20

Whenthestateis(L,L,0)thetwo rmsoftyp eLcan detertheentryofthethird rmby

usingaprice-advertisingcouple(p LL

;a LL

)satisfyingtheno-jammingconditionp LL 1 3 0 a LL 0 ;

which ensures that entry is not pro table. The two typ e L rms mak e nonnegativ epro ts

providedp LL 1 2 0a LL

 0. Clearly, there are man y such couples (p LL ;a LL ), one exampleis (p LL ;a LL )=(v ; v 3

(11)

p eriod. Similarly, the rep etition of promotion campaigns on television by some

brands ma y hav e the same function. Furthermore, as it shall b e shown b elo w

p ersistence can o ccur when renownbrandsma yb e inducedto advertiseinorder

todeter the en tryof other rms.

4. Advertising as an Entry Barrier

Assume there is a third rm in addition to the twopreviously considered. This

rm isa pro ducer oflowestqualit y, inthe sense that it can b e one of twotyp es:

L or0, but not of typ e H. Ifa buyer purc hasesthe go o d 0 atprice p 0 she gets a utilit y u 0 =0 p 0

. Tosimplify matters assume that a op can nev erb e mistak en

forahighqualit ypro duct, butitisindistinguishablefromalowqualit ypro duct,

anassumption that could b e relaxed ina more complete v ersionof this game.

To k eep with the approach of short term in teraction, we assume that the

en try, price, and advertising decisions are all sim ultaneously taken. If we had

assumedrep eat purc hase,and a m ulti-p eriod framew ork,itwouldb e p ossible to

modifythe examplehere inorder toconsiderthe b ehaviorof incum b entsagainst

p otentialen trants.

In this game it is common knowledge that there can b e such a pro ducer,

called a " yer-by-the-night", in the sense that ev erybo dy knows that if three

rms are on the mark et one could b e a op and m ust lea ve the mark et after

havingdeceiv edconsumers once. Weshall concen trateonthe asymmetric states

(H,L,0) and those obtained p erm utingthe order of the en tries. Since qualit y H

cannotbyhyp othesisb eimitated bythe op,inthesestatesthe opm ustimitate

the low qualit yif itwantsto en ter. Then if the state (H,L,0) realizes rm L has

toprev entthe op from en tering 19

.

Ifthe typ e0 rmen tersand replicates the strategyof rm Lthen consumers

cannot distinguish whic h rm is L and whic h is the op. Firm H, by contrast

ma yseparatefrom rm L using a strategy that can b e describedanalogously to

whathasb een doneab ov esoastoprev ent rmLfromimitating|forbrevitywe

donot re-state the conditions (A) and (B) for this game.

Note that if the op en ters and imitates rm L, then this rm and the op

are b oth b eliev ed to b e ops with probabilit y 0.5 and will share the demand

addressed to the corresp onding expected qualit y as it shall b e brie y shown.

The consumers know that there is atmost one op, hence their expected utilit y

from buying from one of the two rms that uses the strategy  L = (p L ;a L ) is 19

Undertheassumptionthata highqualit ycannotb emistak enwith a op, whenthestate is (H,H,0) the yer{by-the-night cannot enter. If the state is (L,L,0) , under revelation one wouldobserveadv ertisingfromb othlo wquality rms,asitisbrie yexplainedinthefollo wing footnote. Finally, ifthe state is (L,L,L)all rmspla y the Bertrandequilibrium with a price equaltomarginalcost(here zero)andnoadv ertising.

(12)

leftpart ofFigure 4,whic hhas b een drawnfora value ofv >2,and of c=0: 25:

thecurv ea (p)isrecognized asstarting fromtheorigin, andA (p)starts fromthe

p oint c, where a y

(p H

) = 0. The two constraints there cross at the p oint where

p H

=2c, and at the p oint wherep H

=1. It can b e shown that for c<1=2 these

crossingso ccurinthe same wayasdepicted inthatgraph 17

,i.e. one has thatfor

p H

intherange[2c;1]theinequalit yA (p)>aholds,whileoutsidethatrangethe

opp ositeinequalit yholds(the righthandgraphin Figure4 hasb een drawnfor a

value of c=0,the valid regionshrinks insize as cis increased formzero to0.5).

It follo ws that in general,for c<1=2, the region to whic ha separating strategy

 H

m ustb elong isformed by(i)anon-empt yregionwhereadvertising isstrictly

p ositiv e; (ii) a (p ossibly empt y) region consisting of all p oints on the abscissa

lyinginthe in terval[ v;z(c)]. This prov esthe rst partof the Prop osition.

(Toprov ethe part of the prop osition regarding the case c>1=2 consider Figure

5, whic h has b een drawn under the hyp othesis that c = 0: 7. It is apparent

from Figure 5 that if v > 2, then a separating equilibrium if it exists is giv en

bythe p oint of tangency of A (p H

) and a (p H

). The same situation arises for all

values of c  0: 5, b ecause for these values one can easily c heck that A (p H

) has

a unique maximum at the p oint where p H = 2c, and that at p H = 2c one has a (p H )=A (p H

), sothat the equilibrium  H

isunique in these cases 18

). 2

We will nally note that a price couple equal to the full information prices

forthe high and lowqualit ymayb e used ata separating equilibrium if the high

qualit y rm accompanies the full information price byan adequate expenditure

inadvertising. Forinstance, forc=0: 25one has that p 3 H

=5=6 and a=a s

(5=6)

ma yconstitute aseparating strategy  H

for H.

The importantfeatureof the resultssofarobtained isthat they,in thesame

lineasforthe monopolycase,establishacharacterizationofequilibria,preceding

acomplete sp eci cation ofb eliefs.

Existence isdiscussed inApp endix 2.

3.4. Persistence

Asitisclear, the plausibilit ythat advertisingm ustb e usedina one-shotset-up,

impliesthatitcanb eusedalsoinsituationswhererep eatpurc haseisnotpresen t.

For instance in the case of the restaurants quoted in the In tro duction. The

luxurious setting of one of two restaurants can b e in terpreted as an ostentation

17

Inparticular,p H

=2 c isthep oin twherea s ;a y ;a z ;allcross; p H

=1 isalw aysone ofthe ro otsof theequationa

s =a

z

,and itisthelargest ro otforc<1 = 2. 18

Whenc = 1 then 2 c = 2 and the p oin t of tangency of the twoconstraints implies that p

H

=2. For c >1 thetwoconstraintsdo nottouch an ymore sinceA(p H

)<0 for c >1,so thattheseparatingequilibriawithadv ertisingdonotexistforthesevaluesofc.

(13)

of Figure4 b elo w)and it is de nedby A (p H )=(p H 0c)  10 p H 2  =a y (p H ); (3.10) for p H <2c, andby A (p H )=(p H 0c)  10 p H 2  0 1 2  p H 2 0c  =a z (p H ); (3.11) for p H >2c. Notethat a z =a y at p H =2c sothat A (p H )is continuous 15 . Furthermore2c is also a maximum of A (p H ) if p H

> 0: 5. While for c< 0: 5 the maximumis at

the rightof p H

=2c and it coincides with the maximumof a z

(p H

).

The in terplay of the curv es a (p H

) and A (p H

) determines the regions in the

space of (p H

;a ) couples to whic h the separating strategy  H

; in state (H,L) or

(L,H),mustnecessarily b elong if a separating equilibrium of the game exists.

Note, to start,that if v <2 strategies in volving azero amoun tof advertising

and that satisfy the constraint derived by condition (A) always exist. If and

onlyifA (p H

) in tersectsthehorizontalaxistothe rightof v 16

, however,thereare

price-alone strategies that also satisfy constraint(B).

For the purp oses of the presen t Section|namely to demonstrate that the

p ossibilit yofadvertisingenlargesthesetofparameterrangesforwhic hseparation

ma y o ccur ov er and ab ov e the range for whic h separation ma y obtain through

prices only|it is sucient to show that, when v > minfz(c);2g, there exist

ranges for the cost parameter c for whic h a strategy  H

with strictly p ositiv e

advertising satis esnecessary conditions (A)and (B).

Prop osition 3. For c < 1 and v > minfz(c);2g the only non-trivial separating

equilibriathatma yexistin volveastrictlyp ositiv elev elofadvertisinginthestates

(L,H) and (H,L). Furthermore, for c <1/2 the high qualit y rm has a strictly

p ositiv epro t, while for c  1/2 its pro tsare null but itsmark etshare p ositiv e.

14

Thepro tto rmHiftheequilibrium strategiesarepla yedis

(p0c)d H (p H ;p H = 2)0a=(p0c)(10p H = 2)0a:

ThenconditionBwritesas

(p H 0c)  10 p H 2  0amax  1 2  p H 2 0c  ;0  : (3.9)

Therigh thand termiszeroifp H

<2 c, that iswhen rmH wouldmak enegativ epro ts if itdeviatedtothepriceofitsrival,butjustneedstomak enon-negativ epro ts.

15

The functionA coincides with the Min[ a z

;a y

] o ver the relevant range and so it can b e moreeasilyretained.

16

Theconditionisthat thelargestro otofa z (p H ),denotedz(c)= 1 4 (3+2 c+ p 904 c+c 2 ) islargerthanv .(Notethatforallc>0,z(c)>c andthatthero otsofa

y

(14)

separation.

In fact if v >2 the only prices that can signal rm H in the state (H,L) are

higher than 2, but then this rm receiv es zero demand as stated in Subsection

3.1 ab ov e, and the equilibrium. The fact that a high value of v destroys the

equilibriawhere price-alone strategies are used can b e in terpreted economically.

Welea vethis discussion howeverfor the concluding Section.

It is now p ossible to ask whether with v >2 advertising can separate where

price-alonestrategiescannot. Letusturnnowforthispurp osetotheimplications

of condition (B).

Note that when rm L playsaccording toits b est reply, ^ p L =p H =2; then the

demand to rm H underfull information is giv enby

d H (p H ; ^ p L )=10p H =2: (3.8) sothat  H ( H j L ;1)= (10p H =2)p H :

Notetostart thatthe demandto rmH (andto rmL)instate (H,L)if Hplays

the same strategy as rm L, should ingeneral b e determinedby the b eliefs held

by consumers when the couple ( L

; L

) is observed. However, in the case here

analyzedthe matterissimpli ed byLemma1: sinceL playsaccordingtoitsfull

information b est resp onse to p H

, it is nev erthe case that p L

> v. If b oth rms

playp L

, therefore, allconsumers buy ev enif they think that the state is (L,L),

sothat the mark etis en tirelycovered foranyb elief b ( L

j L

). Therefore,again,

b eliefs are irrelevant here, and the tie is broken as it is customary by assuming

that demand is split inequal parts 13

.

Condition(B)describesafunction,A (p H

),thatde nesthemaximalamoun ts

of advertising consistent with separation. This function lo oks generically as a

quasi-concav e function with a kink at the p oint where p H

= 2c, (see discussion

13

InamodelwhereD L

(p)hasnoverticalsections,butitisdo wnwardslopingwhereveritis p ositiv evalued,thentheassumptionthatb (

L ;

L

)=0wouldpla ytherolethattheassumption b (

H ;

H

)=0pla ysforcondition(A).Thecalculationswouldonlyb ecomelessstraightforward whilethemain conclusionswouldnotb ealtered.

(15)

termisthenequaltoD b

1 H

2

0a,wherebrepresen tstheb eliefsheldbyconsumers.

Thenfor p H

>v the condition writes as

(1+ v0p H b ) p H 2 0a  p H 2  2 (3.7)

Conditions (3.6) and (3.7) when com bined de ne a continuous function a (p H

),

whic hisconcaveshap edandithasakinkincorresp ondenceofp H

=v,asdepicted

inFigure 3. The graphof the functionto the right of v is shifted, withits piv ot

inthe p oint with co ordinates (v;a (v)),in the rightdirection asb increases from

0to 1. Forb= 0it coincides with the horizontal axis 12 , in fact,p b =v for b= 0 and p b >v for b >0. The function a (p H

) delimits the lower b ound for a in a separating strategy

 H

foranygiv en levelof p H

.

Third, and nally, note that the range of prices for whic h a high price and

no advertising strategy can separate is the in terval of prices at the right of the

in tersectionofthe curv ea (p H

) with the priceaxis. Since this in tersectionmo ves

rightwardasb ( H

j H

)isincreased,therangeofpricesthatcanseparatewithout

advertisingislargertheloweristhevalueofb ( H

j H

). Togetthedesiredresult,

thereforeweshallsetb ( H

j H

)=0soasto"considerthelessfav orablecasefor

the necessit yof advertising".

Assumption 2: b ( H

j H

)=0:

Notethat under this assumption the function a (p H

) dep ends onthe value of

b ( H

j  H

) only if v < 2, in whic h case it lo ok as in the right graph depicted

inFigure 3, in the opp osite case in fact there is nokink and a (p H )= (p H =2)0 (p H =2) 2

,as depicted in Figure4 b elo w.

A rstconditionforaprice-alonestrategytoworkasasignalisthattheprice

is higherthan v (and in particular higher than p b

if violating Assumption 2 one

had b > 0). Notice inciden tally that such a price may b e higher than the full

information price, in particular it must b e higher than that price if v > p 3 H

=

2(1+c)=3.

Simply lo oking at Figure 3 it is clear that if v is larger than 2 then a

price-alone strategy implies p H

> 2 and therefore it yields zero demand to rm H.

Thenwe can state

12 Formally,leta s (p H )= p H 2 0  p H 2  2 ,forp H <v ,andleta b (p H )=p H 0 b +v 2 b 1 +p 2 H 0 2+b 4 b 1 forp H 2[ v ;p b ],wherep b =(2b+2 v )= (2+b )isthehighestro otofa b (p H

). Then,recallingthat d H =0 forp H 2, a(p H )=a s (p H ) if p H <v ; a(p H )=a b (p H ); if p H 2[ v ;p b ] ; anda(p H )=0ifp>p b .

(16)

alone strategies it can b e obtained with a p ositiv elevel of advertising

expendi-tures.

To b egin with, letde nea strategy fora rmas afunction that describesits

c hoiceof action for eac h giv enstate of nature| rms mo vesim ultaneously after

thestate is c hosenbyNature. Anaction isacouple (p;a ) constituted byaprice

anda lev elof xedcost advertising,a , with a0. First,itis useful tostate the

follo wingresult.

Lemma 1: Independen tly of the consumers' system of b eliefs, at a separating

equilibrium of the game rm L inthe state (H,L)plays L =( ^ p L ;0).

Pro of: seeApp endix 1.

Obviouslythis resultgreatly simpli es the analysis. Thestrategy playedby rm

L instate (L,H) is, inthe duop oly example here considered, ^ p L =p H =2 and zero

advertising,or, inour notation,  L =  p H 2 ;0  .

Let usanalyze the implications of condition (A)for our duop oly example.

If rm L imitatesthe price and advertising usedby rmH then consumers

willformb eliefsb ( H

j H

). It isthen crucialtodeterminethe roleof b ( H

j H

)

for the c haracterization of the separating strategies and to c ho ose the value for

itthat do es not in terferewith the results.

First, giv en b ( H

j  H

), then the consumer's utility from buying when b oth

rms use the same strategy is giv enby b ( H j H )u H +(10 b ( H j H ))u L ; that

is,shortening the notation, b (v+m0p)+(10b )(v0p). This implies that the

consumer indi eren tb et weenbuying and not buying is m b = (p0v)=b. So that total demand is D b =1+v=b0p=b if p>v and it isD b =1if p<v 11 .

Second, going back to condition A (no-jamming), for a couple of price and

advertising (p H

;a ), used by rm H , if p H

< v , then all consumers buy one

unit, and demand to eac h rm is 1/2, so that if it plays  H rm L has pro t  L ( H ; H ;b ( H j H ))= p H 2 0a . Ifitplays ^ p L (p H )= p H 2

(andzeroadvertising)

itmak es pro t  L ( L j H ;0)=  p H 2  2 . In the case p H

<v,then the valueof p H and of a that satisfy condition A m ustsatisfy

p H 2 0a   p H 2  2 ; (3.6)

Note that for p H

> v condition A do es not write as in (3.6) ab ov e, b ecause

demandisnot inelastictopriceasnot allconsumersnecessarilybuy. The L.H.S.

11

NotethatthedemandD b

corresponds todemandD H

ofSection2,whenb=1,andtoD L whenb=0. Whileitselasticpart rotatesdo wnwardasb isdecreasedform1 tozero.

(17)

H H

the signal sent by rm H. But it do es not necessarily imply that L is p erceiv ed

as H by consumers: whether on or out the equilibrium path, in fact, observing

 H

at b oth rms ma ynot lead to the b elief that with probabilit yone the state

is(H ;H).

These twoprop erties lead, taken in turn, totwo necessary conditions on the

equilibrium strategies. For k;j 2 fH ;Lg , let  j ( j ; k ;b ( j j  k

)) denote the pay o to rm of typ e

j when (i) playing strategy  j

against strategy  k

playedby its rival of typ e k,

and (ii) giv enthe inducedb elief on itstyp e b ( j

j  k

).

Necessary condition A, whic hma yb e called "no-jamming condition" 9 writes as: (A)  L ( H ; H ;b ( H j H )) L ( L j H ;0):

Necessary condition (B) writes as

(B)  H ( L ; L ;b ( L j L ) H ( H j L ;1):

Recall again that the analysis is, for the time b eing, concerned with the

prop-erties of the strategies played when the state is (L,H) or (H,L) in a separating

equilibrium. These twoconditions are necessary, forany giv enb elief system

im-plying particular values of b ( H j  H ) and b ( L j  L ). If it is p ossible to show

thatthe c hoiceof these twovaluesis irrelevantforthe follo wingresults then the

last in uence of b eliefs on the analysis of necessary conditions (A) and (B) is

remo ved 10

.

3.3. Necessary conditions and signaling

The plan of the argumen tin this subsection is the follo wing:

1) We show that condition (A) is most favorable to price-alone strategies when

b ( H j H )=0. 2)Weassumeb ( H j H

)=0andshowthatwithprice-alonestrategiesseparation

cannotalwaysobtain.

9

The term signal jamming is used by Fudenb erg and Tirole 1986, in a context where an informedpla yerma ypreventanotheruninformedpla yerfrom observingasignal.

10

Notethat since  H

and  L

are byde nition part of a separating equilibrium theyreveal thestate(H,L) or (L,H) andthere is nochoicebut 0 or 1,as it applies,for theR.H.Sin the inequalitiesin (A)and (B).

(18)

of the two represen tations is self eviden t. For our conv enience we adopt the

second.

A t rstsightthetaskofde ningb eliefsasafunctionfromcouplesofstrategies

( i

; j

)(four-dimensionalv ectors)tocouplesofprobabilities(b i ( i ; j );b j ( i ; j ))

isformidable. Luckilyitisp ossible toobtainresults thatare b elief-independent,

in the sense that they apply to any system of b eliefs that sustain a separating

equilibrium. Thisexemptsusfromthetaskofjustifyingthec hoiceofaparticular

structure of b eliefs forthe resultsthat shall b eobtained inthe presen tSection.

Let us transform the notation sligh tly so that b i

( i

j  j

) denotes the

proba-bilit yassigned to the ev ent that rm i is H when this rm plays  i

against the

playof  j

by rm j. The notation mak esitclear that rmsare symmetricinthe

sense that for any , one has b i

( j ) = b j

( j ), i.e. that rms are assigned

the same b elief when they play iden ticalstrategies 7

. Even if this is an obvious

prop ert ywe state it formally for convenience:

Prop ert y1: Ifthe rmsadoptidenticalstrategiesthentheconsumersbeliefsassign

to either rm an identical probability of being of high quality.

A(fully)separatingequilibriumisonewhere: (i)b eliefsassignauniquev ector

(b i ( i j j );b j ( i j j

)) toeac h strategy pro le;(ii) the rms playpure strategies

thatleadtoadi eren tactionpro leineac hstate; (iii)b eliefsareconsistentwith

the rms' strategies 8

. Let us pro ceed to the analysis of the strategies ( H

; L

),

that m ustb e playedat a separating equilibrium when the realized state is(H,L)

or (L,H)| the strategies that are played at the other states of Nature do not

a ect the c haracteristicsof ( H

; L

) that are discussedhere.

(A) First, it must b e a prop ert y of any separating equilibrium of the game

that when the state is (L,H) or (H,L) the rm of typ e L must not nd it

convenient to deviate, from the action  L

that the equilibrium prescribes,

tothe use ofthe action  H

that is prescribed forthe typ e-H rm.

(B) Second, itm ust b e aprop ert y of anyseparating equilibrium that when the

stateis (L,H) or(H,L)the rm of typ eH has no incen tiveto deviate from

the equilibrium prescription  H

and use the strategy  L

that is prescribed

to rm L.

7

A more stringentsymmetry requirementis that b i ( 0 j 00 )= b j ( 0 j  00 ). Although this restrictionisquitenaturalitisnotnecessaryfortheanalysisherep erformed,aditistherefore notused.

8

Notethatp oin t(i)in thede nitionexcludes partiallyseparating equilibria,i.e. equilibria wheresomebut notallstatesarep o oled. Wearenotin terestedin partialseparationhere.

(19)

and it is p

L

=v otherwise. A Nash equilibrium in prices under full information,

with b oth rms enjoying a p ositiv emark etshare 5

, and p ositiv epro ts o ccurs if

the tworeaction functions cross at the p oint (p 3 L ;p 3 H )giv enby (p 3 L ;p 3 H )= 2(1+c) 3 ; (1+c) 3 ! : (3.5)

Whenthecrossingofthetworeactionfunctionso ccursatthisp ointb oth rms

enjoyap ositiv emark etshare(seeFigure2foranexample). Thisisp ossibleifand

onlyif cand v are inthesetS de nedbyS =f(c;v)jc2[0;2];v (1+c)=3g 6

.

It ishenceforth assumed:

Assumption 1 : (c;v)2S.

3.2. Necessary Conditions

Toanalyzetheduop oly caseweshall useamethodologythatparallelsascloseas

p ossible the one traditionally used for the monopoly. First the necessary

condi-tions for separation shall b e determined, then a c haracterization of the

separat-ingequilibriais derived,and nally (in anApp endix) theexistence ofseparating

equilibriais demonstrated.

Unlik e in the monopoly case the consumers observe two couples of

price-advertising strategies, one for eac h rm. There are four states of Nature: (L,L)

designs the state where b oth rm are of typ e L, (L,H) and (H,L) design the

states where one rm is H and the other is L, (H,H) designs the state where

b oth are of typ e H. Let ! denote the set of states that are admissible, here let

therefore b e ! = f(L;L);(L;H);(H ;L);(H ;H)g . A tthe rst stage of the game

Nature c ho osesa state and onlythe two rms observe Nature'smo ve,theythen

sim ultaneouslyc ho ose, at the second stage, their strategies ( i

; j

) : Finally, at

thethird stage, consumers formb eliefs and purc hasefromone or theother rm.

Thereareatleasttwoequivalentwaysofrepresen tingb eliefs. The rstassigns

toanystrategypro le( i

; j

)playedbythetwo rms,iandj,afourdimensional

v ector of probabilities, one for eac h state. The second assigns to eac h strategy

pro le an ordered couple (b i ( i ; j );b j ( j ; i

)) that represen ts the probabilities

5

Weare notin terested to equilibria where only one rmsurvives, for this reason also we ha venotin tro duced xedcostsinthecostfunctions. Notethatintheexampleheretreatedthe lo wqualit y rm at equilibrium alw aysenjoys a p ositiv emark etshare, whilethe high qualit y rmcouldb eoutofthemark etifitscostisto ohigh. Itisthissituationthat weruleout.

6

Thepricep 3 L

isan equilibriumpriceifandonly ifp 3 L  (1+c ) 3 v . Thepriceof rmHis higherthanc ifc< 2(1+ c ) 3

, i.e. ifc<2. Furthermore, sincethemarketdemand to rmH at prices (p 3 L ;p 3 H

)is equal to (10((1+c)= 3)), theconditionc <2 also guaran teesthat this b e p ositiv e. Whencethede nition oftheset S inthetext.

(20)

3.1. The structure of competition under full information

Beforethe studyof the signaling game describedin theIn tro duction ab ov e, itis

necessarytobrie ypresen tthefullinformationresultsofcompetitioninduop oly.

Assume that the cost conditions are the same as for the monopoly case, again

withoutlossofgeneralit y,assumethatc L

=0. Letthebuyer'sutilityb edescribed

bytheexpressions(2.1)and(2.2)ab ov e. Theconsumerthatisindi eren tb et ween

buyingqualit yHorLhastasteparameterm (p H ;p L )=p H 0 p L ;if0<p H 0 p L <1.

Thenitiseasytoshowthatthedemandtothehighqualit y rmisthecontinuous

functiond H (p H ;p L ) describedby d H (p H ;p L )=minf10p H +min[ p L ;v] ;1g; if p H <1+min[ p L ;v] (3.1) d H (p H ;p L )=0 otherwise.

Notehere thatfor anyvalue of p L , d H (p H ;p L )=0 for p H 2.

Similarly,demand to rm of typ eL is giv enby

d L (p H ;p L )=minfp H 0p L ;1g; if p L <p H ; (3.2) and d L (p H ;p L )=0 otherwise.

The twodemand functionsare depicted inFigure 1.

The reaction function for rm H is describedby the function

^ p H =(1=2)(1+p L +c); if max[c01;0]p L v; (3.3)

if bycontrast p>v , then rm L has zero demand and rm H reaction function

consistsof its monopoly price, that is

^ p H =(1=2)(1+v +c); if p L >v:

Finally,if rm Lquotesapricelowerthan c01then rm Hcannotquoteaprice

equalto (1=2)(1+p L

+c) asthisis lowerthanits marginalcost, thereforewe set ^ p H =c for c01p L .

Similarly,the reaction functionfor rm L is

^ p L = p H 2 ; if p H 2 <v; (3.4)

(21)

monopolist has no incentive to deviate from the play of  to the play of (v;0)

if the qualit y is high. Note that, as Milgrom and Rob erts[1986] show, although

man y other deviations are p ossible from the equilibrium prescription, the two

deviationsherestatedare generallyregardedascrucial. Infactitisgenerallythe

casethatone can ndasystem ofb eliefsthat supp ortsthe playof instate(H)

and of (v;0)instate (L)if  satis esthe two conditions (i)and (ii).

The c haracterization of  is usually the fo cus of the analysis, since it c

har-acterizes the typ e of signal used. In general  is not uniquely iden ti ed by the

constraints, while these rather iden tify a set of regions in the space of couples

(p;a ) to whic h m ustb elong. In this resp ect the follo wingresult indicates that

the use of advertising is not essentialfor separation.

Proposition 1: Undermonopolyifthereexistsastrategy =(p;a ) witha>0,

thatsatis es constraints(i) and(ii), then there also existsa strategy 0

=(p 0

;0)

satisfying the same constraints.

Pro of: See the App endix.

Note that Prop osition-1 can b e in terpreted as follo ws: the existence of a

separatingequilibrium,undermonopolyandinaone-shotgame,do esnotimpinge

onthe p ossibilit yof advertising;in other terms,the p ossibilit y toadvertisedo es

not enlarge the sp ectrum of circumstances under whic h separation ma y o ccur.

This however do es not mean that advertising can b e excluded on some game

theoretic ground b ecause to do that one needs use of an equilibrium selection

criterion. This isdone in Milgromand Rob erts[1986] who showthat advertising

resists the application of standard re nements criteria onlyif rep eat purc haseis

assumed, and we donot restate the argumen t here.

There isaneconomic reason, however,whyadvertising isnot necessary: itis

more costly to signal through advertising than through price. The pro of of this

statemen t is simple: de ne a (p) = (1+v 0p)p 0v. For any giv en price this

function giv es the minimal amoun t of advertising that is necessary to separate.

Then, maximize the pro t of the high qualit ymonopolist with resp ect to price

and advertising under the constraint that a  max [0;a (p)]. This is equivalent

to maximizewith resp ect to p the function 5( p)=(1+v 0p)p0a (p) overthe

in terval p in [0;p a

] where p a

is the highest ro ot of 0 = (1+v 0p)p0v. Since

a (p) is monotone decreasing in p, and since 5 0

(p) > 0 the pro t so written is

increasing inprice, sothat a maximumobtains atp a

with a (p a

)=0.

Sayingitdi eren tly,consideragainthe separationconstraint(i)ab ov e. Since

advertising adds to the costs of a high qualit y monopoly as m uch as it adds to

thoseof alowqualit yitisrelativ elyeasier toimitatethan ap olicyof highprice:

the latter is more costly in terms of lost pro t (unit margin times demand) for

amonopoly of lowqualit y,who pro duces ata lowmarginal cost, than for one of

(22)

H

taste parameter, ranging in the unit in terval [0;1], uniformly distributed over

this range so that the consumers p opulation has unit mass. If a consumer do es

not buy her utility is zero. The consumer that is indi eren tb et ween buying or

not ahighqualit y pro ducthas taste parameterm =p H

0v. The demandto the

monopolist of high qualit yunder p erfectinformation is then

D H (p H )=min[1+v 0p H ;1] if 1+v p H , and D H =0 if 1+v p H .

If the go o d is of low quality, then utilit y is assumed to b e in variant with m

and tob e giv enby u L =v0p L : (2.2)

Thereforedemand tothe monopolist of low qualit y under p erfectinformation is

D L =1 if v p L , and D L =0if v <p H .

The p erfect information prices are p m H = maxf(1+v+c)=2;vg and p m L = v

for the high and the low qualit yresp ectively. The case where (1+v+c)=2 v

is of no in terest since then the monopolist has no incen tive to separate when

her qualit y is H. Eliminating this case, then, the full information pro ts are

resp ectivelygiv enby  m H =[(1+v0c)=2] 2 and  m L =v.

Under imperfect information, assume that consumers know that the

monop-olistma yb e ofone ofthe twoqualities,sothatthere areonlytwop ossiblestates

ofNature, denoted by (H)and (L). Let denote acouple (p;a ) of price and

ad-v ertisingused bythemonopolist. Foranyobservationof the consumers forma

b eliefab out the pro duct qualit y. More preciselya b eliefis de ned asa function

b ()indicatingthe probabilit ythatthe pro duct isofhigh qualit y. Then, giv ena

particular priceand advertising com binationc hosenbythe rm, the consumer's

expected utilit yis u ()=b ()(v+m0p)+(10b ())(v0p),with u ()=u H

if

b ()=1and u ()=u L

if b ()=0.

A separating equilibrium, lo osely sp eaking, is constituted by strategies and

b eliefs with the prop ert y that , say, is playedif the state is H, and  L

, say,is

playedifthestateisL,andtheb eliefsaresuchthatnodeviationfromthese

strat-egyprescriptionispro table. Itiswellknownthat applicationofthe elimination

ofdominatedstrategiesimpliesthatataseparatingequilibrium L

=(p m L

;0),and

weshallnotrep eat theargumen there(see Milgromand Rob erts1986). This

im-pliesthat thereare twonecessaryconditionsforseparation,theyare resp ectiv ely:

(i ) (1+v0p)p0av; (2.3)

and

(ii ) (1+v0p)(p0c)0a v 0c: (2.4)

Condition (i)states that astrategy can b e part ofaseparating equilibrium

(23)

advertising. In terestingly,this typ eofadvertisingdo es notserv etoseparatefrom

existing rms,but fromp otentialen trants,and thereforewhere observed itmay

b eerroneously ascrib edtootherreasons notdep ending onimperfectinformation

ab outqualit y. Thistyp eofadvertisingp ersistsovertimeaslongasthe threatby

lowerqualit y rms p ersists, anditvariesinthe samedirection asdemand varies,

namelyit variespro-cyclically.

Ourexplanationoftheen trybarrierattributeofadvertisingisnotbasedup on

the e ects on consumers' tastes, as advertising is here purely dissipativ e. It is

in teresting to note that in an article on limit pricing and advertising Bagwell

and Ramey [1988] used price-adv ertisingcouples as informativ estrategies by an

incum b enttosignalits costofpro duction toanen trant. Theyfound thatpurely

dissipativ eadvertising couldnot b e usedas an en trybarrier.

Othermodels (ofrep eat-purchase)thatare relatedtothe studyofexperience

go o ds are the reputationmodels lik eKleinand Leer [1981]andShapiro [1983],

based on the importance of qualit y premia for the inducemen t to a monopolist

to in tro duce a high qualit y instead of a low one (see also Riordan [1986]). In

this resp ect our analysis also has some consequences, inspite of the assumption

we mak ethat the typ es of the rms are c hosenby Nature. Where rms cannot

rely on rep eat purc hase, in fact, a high qualitycan b e in tro duced only if it can

separatefromthe lowqualities.

2. Monop oly in the Absence of Repeat-Purchase

Themainimplicationsoftheanalysiscanb ederiv edviaasimpleexample,already

usedforthe monopolycasebyBagwelland Riordan[1991],and hererearranged,

whereneeded, toin tro ducea v ertically di eren tiatedduop oly 4

.

Consider amark etwhere amonopolistsells ago o d that can b e of qualit yH

or L. Unit pro duction costs are c for the high qualit y and 0 for the low qualit y.

There are no xedcosts of pro duction, although the monopolist can v oluntarily

add a discretionary amoun t of xed cost, a; in the form of wasteful advertising

campaigns. Eachconsumerbuyseitheroneunitornone. Ifthego o d isofqualit y

H the consumer's utilit yafter purc hase is

u H =v+m0p H (2.1) 4

Although several examples exist (see Gabszewicz and Thisse[1979] seminal pap er, and Shak edand Sutton [1981]) we do not ha ve yet a generaltheory of verticaldi erentiation in oligoplyundercompleteinformation. Theauthorsha veworkedoutasetofnecessaryconditions foraduop olytob ede ned asverticallydi erentiated,andha vecheckedthatthemain charac-teroftheanalysiscanb epreservedinthatmoregeneralset-up. Theexp osition,ho wever,then b ecomes mingled to conceptualissues p ertainingto thefull information generalizationofthe verticaldi erentiationmodelsthatdonotaddan yusefulinsigh tswhilethedi erenceb etween ourexplanationofadv ertisingandthetraditionalone b ecomesunclear.

(24)

uninformed party, here the consumers, holds ab out the other rm's typ e aswell

asof itsown typ e.

Thisfeatureof b eliefsdo es notdep endonthe numb erofinformedparties|it

is eviden tthat in the job mark etgame of Sp ence [1973]there are man yworkers

but the b eliefs of the emplo yer on the typ e of worker Smith do not take in to

account what worker Jones has sp en t on education 3

. It dep ends rather on the

prop ert y that the pay o (here the pro ts) of an informed play er dep ends up on

what the other playerdo es (e.g. itspricing b ehavior), and on the fact that this

reciprocal dep endence is common knowledge to allplayers.

The pap erpro ceeds asfollo ws. First, inSection2,the problem withone rm

only is in tro duced and the notation established. The main reason for Section

2 rests not up on its novelty but on the need to brie y clarify why in the

ab-senceof rep eat purc hasethe monopolycase is insucien ttoexplain anyformof

advertising,and to pavethe wayfor the remaining Sections.

Section 3, then, constitutes the core of the pap er. Two rms compete, eac h

b eing of one amongtwop ossibletyp es, eac h rm knows itstyp e and that of the

competitor whileconsumers donot observethe realizedstateof Nature butonly

the strategies played by the rms. On the basis of this observation consumers

formtheirb eliefs. Thepurp oseoftheSectionisthatofderivingac haracterization

of separating equilibria that do es not dep end on the b elief system adopted by

consumers. As a method, the approach consists of writing down two necessary

conditionsthatthe strategiesplayedbythetwo rmsm ustsatisfyinthestatesof

Naturewhentheirqualitiesdi er: thesesimply saythatneither rmmust ndit

pro tabletomimicexactlytheb ehaviorof therivalwhentheirtyp esdi er. Only

usingthese twoconditions it isshown that there are cases in whic haseparating

equilibrium, if it exists, must in volve the use of advertising bythe high qualit y

rm if the qualit ytyp es di er. The rst task ofthe presen tresearc his therefore

accomplished inSection 3, since an explanation is providedfor advertising that

do es not impinge onrep eat purc hase.

The issueof existenceofseparating equilibriaissolvedinApp endix2,whic h,

however, do es not dev elop a complete typ ology of p ossible b elief systems since

thisisirrelevantforourresultsanditwouldleadusastrayfromthenon-technical

issues that motivatethe analysis.

In Section 4, the idea that en try ma y b e deterred through advertising is

examined using the results of Section 3. The Section presen ts an extension of

the basic model obtained in tro ducing the p ossibilit y that the duop oly mark et

structure b e contestable by a rm whic h can also b e of two typ es: low and

"very low", and whic h can mimic the b ehavior of a low qualit y. A true

low-3

Similarly,iftheindustryisp erfectlycompetitiv ea rmcannota ectwithitsb ehaviorthe b eliefsthatconsumers holdab outthetyp esoftheother rms. KhilstromandRiordan[1984] modelthereforedo esnotsharethisfeature.

(25)

brands of long-established reputation also sp end considerable sums on

advertis-ing. This phenomenon cannot b e explained by Nelsonian theories, unless one

admits a constant in ow of new consumers at eac h p eriod who are unaware of

the qualit y of the brand b eing advertisedand who en ter a rep eat purc hase

pro-cess. Another way out would b e to assume that consumers do not have p erfect

recall|orthatthec haracteristicsofthepro ductsonthemark etc hangefastwhile

purc haseshapp eninfrequen tly 2

. Itseemsdiculttoaccepttheideahoweverthat

themagnitudeofthisin owb elargeenoughtojustifythe largestreamov ertime

of advertisingexpenditures that one observes. Similarly,this typ eof advertising

isnotencompassedbytheexplanationinKihlstromand Riordansince,although

they do not assume rep eat purc hase, rms need to advertise only in the in

tro-ductoryphase.

Third,theempiricalliterature(aswellassomenon-signaltheoriesof

advertis-ing lik eDorfman and Steiner[1954])had long since notedthe relativ e constancy

of the advertising-sales ratio of rms over time (see for instance Sc hmalensee

[1972]). This means inparticular that advertising varies pro-cyclically.

Fourth,invariousmark etswhere rmsofhigherqualit ycompeteagainstrivals

oflowerqualit y,advertising bylowqualit y rmsisconspicuous andsometimesit

compares tothat of the high qualit y rms.

The explanation provided in the presen t pap er encompasses all these four

typ es of advertising and it therefore enlarges the quota of advertising

expendi-tures that can b e attributed to a signaling e ort by rms. Furthermore, quite

surprisingly, the explanation we provide con rms an old idea, namely that high

advertising expenditures can b e used as a strategic en try barrier. It is worth

emphasizing that the en tryprev enting attribute of advertising is derivedin the

presen t pap er without recurring to ad hoc assumptions ab out the in uence of

advertising onconsumers' tastes.

Themodelb elo wabandonsthe repeatpurchase(whic hhowevercouldb e

rein-tro duced at no cost) and the one- rm assumptions. It explicitly considers the

problem of nding separating instrumen tsfor the case where two rms compete

onthe samemark et. Noneofthesetwo rmshas anestablishedreputationbutit

canb e ofhigh orof lowqualit y;thereforethe consumers are confrontedwith the

problemof in terpretingthe price-adv ertisingstrategies of b oth rmsatthe same

time. In terestingsituations ma yarise that donot app ear in the monopoly case.

For instance, a rm ma y try and exploit the consumers imperfect information

by copying the strategy adopted by its rival. This typ e of mimic king b ehavior

is rather di eren tfrom the one to whic hthe signaling literature in economics is

used: in our case, in fact, it is quite obvious that if two rms send the same

signal(whether along oroutside the equilibrium path) then theconsumers must

attribute them the same probabilit y of b eing of a giv en typ e. In other words

2

(26)

The 1

explanation of advertising by Nelson [1974] is well known. In a context

whereconsumers rep eatedlypurc haseanexperiencego o d|namely oneofa

qual-it ythatcanb elearnedonlyafterpurc hase|apparentlywastefuladvertising

cam-paignsinduce rational consumers torealize that the advertisedbrands sell a

su-p erior pro duct. Consumers would not re-purc hase the go o d after a rst trial if

it were not of go o d qualit y, and the in tro ductory advertising costs could not b e

reco veredin subsequentp eriods.

Nelson argumen ts|thatapply toanyformofwastefulexpensesby rms and

notjusttoadvertising|restonrepeatpurchase. Alongthoselinessev eralauthors

hav edev elop ed rigorous models to scrutinize the validity of Nelsonian theories

(see among others Milgrom and Rob erts [1986], Sc hmalensee [1982]; see also

Martin[1994]forarecen tsurvey). The limitationintheNelsoniantheoryisthat

it can account only for in tro ductory advertising whic h is terminated after the

pro ductb ecomesknown. Moreover,mostofthe literature considersthe strategic

in teraction among one rm and the p opulation of consumers. Kihlstrom and

Riordan[1984]represen tanexceptionsincetheyconsiderap erfectlycompetitive

industry with free en try, where rms are price-tak ers. They obtain the result

that advertising equilibria can arise also when the in teraction among rms and

consumers is not rep eated, provided the high qualit y rms hav e lower variable

costs than their lowqualit yrivals.

There seems to lac k a uni ed theory whic h can explain sev eral phenomena

relatedto advertising.

First, advertising or other wastefuland observable selling expenses are

com-mon in mark etswhere consumption has a transient nature. Think for instance

tothe ownersofrestaurantsthatcrowdsometouristicareas (thisexampleisalso

inTirole [1989]), who in vestsometimes considerable sums inluxurious settings,

silv er cutlery, numb er of waiters, and the lik e, although they do not expect to

receiv e the same customers rep eatedly. One ma y observe that some restaurants

will sp end money on these items and others will not. However, a theory based

onrep eat purc hase cannotassign an informativ erole tothese expenditures.

1

C-D.Fluet,D epartementdesSciencesEconomiques,Univ ersityofQueb ecatMon treal,CP 8888Succ.A,Mon treal-Queb ec-(Canada),H3C3P8. PaoloG.Garella,Dipartimen todiScienze Economic he,Strada Maggiore 45, 40123 Bologna (Italy), e-mailgarella@boph01.cineca.it.We wish to thank participan ts in workshops at Univ ersitat Autonoma de Barcelona, I.G.I.E.R. (Milano), and Johns Hopkins Univ ersity (Baltimore). F. Forges provided commen ts on an earlierversionbut do esnotsharean yresponsibilityon theremainingerrors. A suggestionby JacquesRobertwasparticularlyusefulat anearlierstage ofthepap er. The nancialsupp ort fromtheQueb ecFundFCARisgratefullyakno wledged.

(27)

Claude D. Fluet

University of Queb ec atMontreal

and Paolo G. Garel la Universitadi Bologna Septemb er1995 JELClassi cation : L13, L41 Abstract

The presen tarticle providesa uni edexplanation for sev eralphenomena related

to advertising by rms. (i) Adv ertising without rep eat purc hase of the pro

d-uct, (ii) advertising from established brands, or p ost-intro ductory, (iii) sim

ul-taneous advertising from low and high qualit y rms, (iv) its p ersistence and

pro-cyclicalit y. The explanation is originalb ecause it rests up on oligop olistic

in-teraction. The analysis hinges up on two fundamental results. The rst is that

advertising allows separation when a signal via prices only do es not. The

sec-ondisthatpurelydissipativ eadvertisingcanb eused tostrategically deteren try.

Hence, alink isestablished b et weenen trydeterrence and signaling.

Keywords: Adv ertising,signaling,en trydeterrence,imperfectinformation,

Riferimenti

Documenti correlati

• Marketing objective is to increase a site’s ranking on the Search Engine Page Results (SERPs) for keywords that are important to the business. • preferably making it to the

the liquid contained in the gallbladder or the cerebrospinal fluid (CSF)] (fig. Si tratta di una sequen- za SE in cui all’impulso a 180° sono fatti precedere e seguire due gradien-

Governments which perceive themselves as relatively stable and which expect that the enforcement of a clause against their own countries is unlikely, tend to support the adoption

On the other hand, the action of the uncertainty connected with the lure/repulsion effect of the advertising message is deeper (and it is not so surprising because the

The second test signal employed was the “Music” signal, which was based on the average 1/3 of octave spectrum.. of all the music pieces stored on the measured DAPs (more than

professionals employed by ad agencies in the Pan-Amazon region, and discuss the results from the point of view of sustainability. The study also sought to identify and

With the beginning of the twentieth century, the consumer goods were at the center of the symbolic represen- tations that have filled the anomic void of society, the emptiness

It is now clear that β-arrestin binds directly to β 2 -ARs following phosphorylation by a specific GPCR kinase (GRK2 or β-ARK) and has a central coordinating role in processes