Che materia stai cercando?

Anteprima

ESTRATTO DOCUMENTO

− −

monopolist’s profit function is given as The

π(Q, A) = P (Q, A)Q C(Q) κA.

first-order conditions for profit maximization are then:

0

− (4.13)

π = P C + QP = 0

Q Q

− (4.14)

π = QP κ = 0

A A

denote the solution to (4.13), and let represent the solution

Let (A) A (Q)

Q

M M

to (4.14). The monopoly solution, is then defined by

(Q , A ), Q = Q (A )

M M M M M

and = A (Q ).

A

M M M

Social welfare may be represented as

Z

Q − − (4.15)

W (Q, A) = P (X, A)dX C(Q) κA.

0

Suppose now that (i). the monopolist begins at its monopoly solution, , A ),

(Q

M M

and (ii). when advertising is changed to some nearby level the monopolist

A,

responds with its profit-maximizing quantity, Then, using (4.15),

Q (A).

M Z

Q

M

dW (Q (A), A)

M 0 0

| −k.

= [P (Q , A )−C (Q )]Q (A )+ P (X, A )dX

A=A M M M M A M

M

M

dA 0 (4.16)

This expression can be signed under two conditions. Assume first that the

value of advertising is greater for infra-marginal than marginal consumers: <

P

AQ

Assume second that advertising beyond the monopoly level does not decrease

0. 0 ≥

the profit-maximizing level of output: Then, by (4.16),

Q (A ) 0.

M

M

dW (Q (A), A)

M 0 0

| − −

> [P (Q , A ) C (Q )]Q (A ) + Q P (Q , A ) k

A=A M M M M M A M M

M

M

dA 0

−Q

= P (Q , A )Q (A )

M Q M M M

M

≥ 0, the equality uses (4.13) and (4.14), and

where the first inequality uses < 0,

P

AQ

0 ≥

the second inequality uses With this, a second conclusion is now

Q (A ) 0.

M

M 66

0 ≥

and then advertising is supplied by a

established: if < 0 Q (A ) 0,

P

AQ M

M 69

monopoly to an extent that is socially inadequate.

The key intuition can be easily summarized. As (4.14) reveals, when the

monopolist chooses advertising while holding quantity fixed, it balances the cost

of additional advertising against the benefit of selling the fixed quantity at a

higher price. The extent to which price rises is in turn measured by the benefit

that additional advertising brings to the marginal consumer. On the other hand,

as (4.15) suggests, a social planner balances the cost of additional advertising

against the benefit it brings to all of the monopolist’s consumers. If P < 0,

AQ

then the marginal consumer gets the least benefit from additional advertising;

thus, for a given quantity, the monopolist provides too little advertising. As is

standard, for any given level of advertising, the monopolist also provides too little

quantity. It follows that welfare would rise if the monopolist were to increase

advertising without decreasing quantity.

This analysis dovetails nicely with the first example. Under constant marginal

is established

costs, this example satisfies both of the two conditions: < 0

P

AQ

70

0 ≥

above, and is easily confirmed. Thus, the first example offers a

Q (A ) 0

M

M

concrete illustration of the sufficient conditions just derived. At the same time, the

analysis clarifies the general features that are embodied in the example and that

underlie the finding of inadequate advertising. Of course, the second conclusion

holds as stated whether advertising is complementary due to the information it

conveys or the social prestige that it facilitates. The former case is captured by

the second example, but in this example when individual demands are

P > 0

AQ

not too convex. The second conclusion appears most relevant for advertising that

facilitates social prestige.

Nichols (1985) presents a related sufficient condition that applies when con-

sumer welfare is the maximized value of where and are chosen

U(g(A)X, Y ), X Y

subject to a budget constraint: Nichols derives that consumer

P X + P Y = I.

x y 0 0

welfare rises with greater advertising if and only if [g A/g P (A)A/P (A)] > 0,

M

M

where is the monopoly price of good when is given. Thus, social

P (A) X A

M

welfare rises when the monopolist increases advertising above the monopoly level

69 Becker and Murphy (1993) also discuss the case of price-increasing advertising, and they de-

< 0

rive an expression corresponding to (4.16). They do not derive the two conditions (P and

AQ

0

Q (A ) ≥ 0) reported here that suffice for inadequate price-increasing monopoly advertising.

M

M

As I discuss below, Nichols (1985) derives a related sufficient condition for the characteristic

approach that he adopts. 0

dD(P (A),A) Ncg (A )

70 M m

= > 0.

|

In particular, A=A 2

dA 2[g(A )]

M M

67

if and only if additional advertising increases “prestige productivity” (i.e., in

g)

a greater percentage than it increases the price of the advertised product. This

requirement is automatically satisfied if advertising is price-maintaining or price-

decreasing. A limitation of this approach is that is not easily observed.

g

Finally, it is important to remark on the possibility raised in Section 2.4 that

advertising is a bad that is sold jointly with some other good (e.g., TV programs

may compensate viewers for watching TV ads). This possibility is not included

in the analysis above, where I follow the conventional modeling approach and as-

sume that advertising is a good whose quantity is determined by the monopolist.

But Becker and Murphy argue that the first conclusion above continues to hold

when advertising is a bad that is jointly sold. Intuitively, if the consumer volun-

tarily accepts additional advertising and the price of the good does not rise, then

additional monopoly advertising again induces a consumer surplus gain that the

monopolist is unable to appropriate.

4.3. Summary

In this section, I summarize research on the positive and normative theory of

monopoly advertising. The Dorfman-Steiner model offers a positive theory of a

monopolist’s price and advertising selections, and the first-order conditions pro-

vide a formal interpretation for some of the endogeneity concerns raised in the

previous section. Two examples are also examined. These examples confirm

Chamberlin’s insight that advertising’s effect on price is related to the elasticity

and scale effects of advertising, where the elasticity effect is determined by the

purpose of advertising. The second (informative) example also captures Ozga’s

rationale for diminishing returns to advertising.

Dixit and Norman provide a foundation for the normative theory of persuasive

advertising. They argue that, if the consumer welfare that advertising renders is

measured relative to a standard that remains fixed as advertising changes, then a

monopolist provides price-increasing advertising to an extent that is socially ex-

cessive. Proponents of the informative and complementary views, however, argue

that the fixed-standard approach ignores consumer-welfare gains from advertising

that are associated with information and social prestige. Under the alternative

approach that their work suggests, a monopolist provides price-maintaining and

price-decreasing advertising to an extent that is socially inadequate. Furthermore,

under conditions that are plausible when advertising facilitates social prestige, a

monopolist provides even price-increasing advertising to an extent that is socially

68

inadequate. On the whole, the research described above suggests that a profit-

maximizing monopolist may advertise to an extent that is socially inadequate.

At the same time, it is important to highlight two assumptions of the models

presented here. First, as mentioned in Section 2.4, the models do not include ads

that are utility reducing (bads) and unavoidable. For example, an objectionable

ad on a city bus, streetcar or taxi is difficult to avoid, and internet “pop-up” ads

are also intrusive. If a monopolist can profit from ads that are objectionable and

unavoidable, then the possibility of excessive monopoly advertising would gain

renewed credibility. Second, the models assume that the monopolist is unable to

segment the market by targeting its ads to certain groups and then practicing

price discrimination. If a monopolist can segment its consumers, then it may be

able to appropriate the increase in surplus that its advertising creates. It then

71

becomes more likely that the monopolist advertises at a socially optimal level.

5. Advertising and Price

Monopoly advertising may be inadequate, since the monopolist cannot appropri-

ate the consumer surplus that additional advertising creates. But in markets with

multiple firms advertising is also an important instrument of competition. The

advertising of one firm may steal the business and thus diminish the profit of

another. This business-stealing externality raises the possibility that advertising

may be excessive. In multi-firm markets, it is thus unclear, a priori, whether

advertising is inadequate, excessive or optimal.

This tension is recognized by Marshall (1919), who acknowledges both the ben-

eficial constructive and wasteful combative roles that informative advertising may

play. In the context of persuasive advertising, an early formalization is offered

by Dixit and Norman (1978), who consider not just monopoly but also multi-

firm markets. Due to the business-stealing externality, they find that advertising

then may be excessive even when it results in a lower price. With important

exceptions, however, the recent theoretical literature emphasizes informative ad-

72

vertising. I summarize here recent theoretical analyses of multi-firm markets in

71 For further discussion, see Adams and Yellen (1977) and Lewis and Sappington (1994).

72 For example, see Friedman (1983) and Schmalensee (1972, 1976b) for positive theories of

oligopolistic advertising competition, under the general assumption that a firm’s advertising in-

creases the demand for its product. For other models in which advertising plays a persuasive role,

see Banerjee and Bandyopadhyay (2003), Baye and Morgan (2004), Bloch and Manceau (1999),

Chioveanu (2005), Doraszelski and Markovitch (2004), Kotowitz and Mathewson (1979b), Von

69

which advertising provides price information.

5.1. Homogeneous Products

In a classic paper, Butters (1977) offers the first equilibrium analysis of infor-

mative advertising in a multi-firm model. In Butters’s model, firms produce a

homogeneous product at a constant unit cost There are consumers. As in

c. N

the second example above, a consumer can learn of a firm’s existence and price

only by receiving an ad from that firm, and ads are distributed randomly across

consumers at a cost of per ad. Finally, and in contrast to the second example,

κ −

consumers have symmetric unit-demand functions, so that is the surplus

R P

that a consumer enjoys when a unit is purchased at price . To ensure that

P

production has social value, assume that R > c + κ.

In this multi-firm setting, there are three kinds of consumers. Some consumers

uninformed:

are they receive no ads. Uninformed consumers never learn of any

captive:

firm, make no purchase and receive zero utility. Other consumers are they

receive ads from only one firm. A captive consumer knows of one firm and thus

buys from that firm, provided that the price does not exceed Finally, some

R.

selective:

consumers are they receive ads from more than one firm. A selective

consumer buys from the lowest-priced known firm, if that price does not exceed

If there is more than one such firm, a selective consumer picks one at random.

R. The number of uninformed consumers is determined by the total number of

ads, that firms send. Let denote the probability that a consumer receives at

A, Φ −

least one ad. The probability that the consumer is uninformed is then 1 Φ =

−A/N

A

− for large. If a proportion of consumers are to receive at

(1 1/N) N Φ

e

≈ · −

least one ad, then a total of ads must be sent. The social

A = N ln[1/(1 Φ)]

cost of advertising so that a proportion of consumers are not uninformed is thus

Φ

· − (5.1)

A(Φ) = κN ln[1/(1 Φ)].

Each firm chooses which price (or prices) to advertise and the number of ads

to send out at each such price. As Butters shows, when the number of firms is

finite, firms adopt mixed strategies in any Nash equilibrium. To see the forces at

hand, hypothesize an equilibrium in which all firms advertise the same price. If

this price were to exceed then a firm could do better by sending the same

c + κ,

der Fehr and Stevik (1998) and Tremblay and Martins-Filho (2001). Becker and Murphy (1993),

Hochman and Luski (1988), Nichols (1985) and Stigler and Becker (1977) consider complemen-

tary advertising in perfectly competitive markets.

70

number of ads but deviating to a slightly lower price. The firm then increases its

expected profit, since it wins its selective consumers with probability one. On the

other hand, if the candidate equilibrium price is or lower, then a firm earns

c + κ

negative profit, since some recipients are selective and choose a different firm. The

sunk cost of sending an ad, is then not covered, and a firm would do better by

κ,

sending no ads.

It is possible, however, to describe simply the limiting behavior that obtains

when the numbers of firms and consumers are sufficiently large. Each seller is

then negligible relative to the market. The behavior of any individual seller is

indeterminant, but equilibrium does constrain market behavior. Butters shows

that every must be advertised by some firm and that every such

P [c + κ, R] ∈

price must generate zero expected profit. For let denote

P [c + κ, R], x(P )

the equilibrium probability that an ad with price would be accepted by the

P

consumer that receives it. Then is the probability that a consumer does not

x(P )

receive an ad with a price below . It follows that is strictly decreasing. In

P x(P )

∈ − −

fact, since every earns zero profit, is defined by

P [c + κ, R] x(P ) (P c)x(P )

This implies that

κ = 0. x(c + κ) = 1 > κ/(R c) = x(R).

In effect, is a downward-sloping demand curve that confronts each firm

x(P )

in equilibrium. The firms compete with one another, but each firm also possesses

some monopoly power, due to the informational product differentiation that ad-

vertising creates. The demand curve is thus not perfectly elastic. But firms earn

zero profit, once the cost of advertising is included. Butters thus offers a first

equilibrium model of monopolistic competition with informative advertising.

What are the normative implications? Given that consumers possess identi-

cal unit demands, price plays no welfare role. A social inefficiency occurs only if

the advertising choice is excessive or inadequate, so that too few or many con-

sumers are uninformed. In the market equilibrium, the probability that a

x(R)

consumer purchases when the highest possible price is received must equal the

e

probability that the consumer does not receive any other ad. Hence,

1 Φ

e −

− Consider now the social planner’s choice. When

= x(R) = κ/(R c).

1 Φ

an additional consumer learns of the existence of some firm, the social benefit is

− But there is also a cost to reaching a previously uninformed consumer.

R c. · −

Using (5.1), the advertising cost per-consumer is A(Φ)/N = κ ln[1/(1 Φ)],

which is increasing in Balancing these considerations, a social planner chooses

Φ.

∗ to solve

Φ {Φ(R − − · −

max c) κ ln[1/(1 Φ)]}.

Φ ∗

− − − which implies that the market

The first-order condition is ) = 0,

R c κ/(1 Φ

71 ∗

e

equilibrium level of advertising is socially optimal: = Φ .

Φ

This is a striking finding. To see the intuition, consider the private benefit

73 −

to a firm of sending an ad at the price This benefit equals times

R. (R c)

the probability that the consumer receives no other ad. But this is also the

social benefit from sending an ad, since the ad increases social surplus (in amount

− only when the consumer receives no other ad. Put differently, the highest-

R c)

priced firm appropriates all consumer surplus and steals no business from rivals;

therefore, it advertises at the socially optimal rate. Now consider the private

benefit to a firm from sending an ad at a lower price, Such an ad

P < R.

generates consumer surplus that the firm does not appropriate, and it also may

steal business. Given that every earns zero profit, however, the

P [c + κ, R]

private benefit to a firm from sending an ad is the same whether or

P < R P = R.

Private and social benefits thus agree even for ads with P < R.

Butters’s model has been extended in many interesting directions. Stegeman

(1991) assumes that the numbers of consumers and firms are large, and then mod-

ifies Butters’s model with the assumption that consumer valuations are hetero-

geneous. He shows that informative advertising is then inadequate. Intuitively,

in equilibrium, the highest-priced firm sets its price strictly below the highest

consumer reservation value; therefore, a firm that advertises the highest price no

longer captures all of the surplus from the new sales that it creates. Since such

a firm does not steal business from any other, it advertises at a socially inad-

equate rate. Additional ads at lower prices would increase social surplus by at

least as much, and so an increase in advertising at any advertised price would

increase welfare. Likewise, Stahl (1994) reports that equilibrium advertising is

inadequate, when the Butters model with a finite number of firms is extended

to allow for downward-sloping individual demand curves (as in the second ex-

ample above) and general advertising technologies. Stahl shows that the unique

mixed-strategy Nash equilibrium is symmetric, and he finds that sellers choose a

common advertising level while mixing over prices.

The Butters model also may be extended to allow a more active role for con-

sumers. Suppose that consumers are aware of the existence of firms and seek only

price information. This is plausible in an established industry. Ads are one source

of price information, but a consumer might also obtain price information through

costly search. In comparison to the work described above, an important new

feature is that uninformed consumers may search for firms and make purchases.

73 Butters is unable to offer an intuition for his welfare finding. My discussion here draws on

Tirole (1988, Section 7.3.2) and Stegeman (1991).

72

Robert and Stahl (1993) provide an analysis of price advertising in an optimal

74 Assuming that firms make simultaneous advertising and pricing

search model.

choices, Robert and Stahl characterize a unique and symmetric price-dispersion

equilibrium, in which a firm either charges a high price that is not advertised or

selects from an interval of lower prices that are advertised. The high price may be

interpreted as a “list price.” Firms that charge this price sell only to uninformed

consumers, and the list price is set at a level that dissuades such consumers from

further search. In the interval of advertised “sales,” an interesting prediction is

that advertising intensity is greater at lower prices. Intuitively, the marginal ben-

efit of advertising is greater at lower prices, since such prices are more likely to

75

attract the recipient (who may be selective).

Interesting findings also arise when the model is extended to allow for sequen-

tial choices by firms. McAfee (1994) posits that firms first choose their advertising

rates and then choose their prices. A firm’s advertising rate determines the prob-

ability that a consumer obtains its price offer. An asymmetric equilibrium then

exists, wherein one firm advertises more than do other firms, who all advertise

equally. Prices are mixed, and now the firm with the higher advertising rate

higher

charges prices (in the sense of first-order stochastic dominance). The rea-

son is that such a firm has a greater stock of captive consumers. Roy (2000) also

considers a sequential-choice game, and he allows further that firms may “target”

76

the individual consumers to whom their respective ads are delivered. Working

74 Butters considers an extended model with search, but he does not analyze optimal search.

Baye and Morgan (2001) also examine a model of price advertising in which the information-

gathering activities of consumers are endogenized. In their model, a “gatekeeper,” such as a

magazine or an internet site, charges fees to firms that advertise prices and to consumers who

choose to access the list of advertised prices. Baye and Morgan (2004) extend the model to allow

that firms may also engage in brand advertising, where brand advertising by a firm increases

the number of consumers that are loyal to the firm’s product.

75 Bester (1994) conducts a related analysis in a monopoly model. Consumers must sink a

search cost in order to visit the monopolist’s store, and the monopolist thus seeks a device

through which to commit to low prices. Price advertising is such a device. A mixed-strategy

equilibrium is constructed, in which the monopolist advertises only low prices. Modifying the

model to allow that the monopolist is privately informed as to its costs of production, Caminal

(1996) constructs a pure-strategy equilibrium, in which the monopolist advertises low prices

when its costs are low.

76 For other studies of targeted advertising, see Adams and Yellen (1977), Esteban et al

(2001a), Esteban et al (2001b), Galeotti and Moraga-Gonzales (2004), Hernandez-Garcia (1997)

and Manduchi (2004). The related possibility of coupon targeting is considered by Bester and

Petrakis (1996), Moraga-Gonzalez and Petrakis (1999) and Shaffer and Zhang (1995), for ex-

ample. 73

with a duopoly model, he finds that the two firms divide the entire market into

mutually exclusive captive segments within which each firm operates as a local

monopolist. Under the assumption that consumers have identical unit demands,

the resulting equilibria are socially efficient: firms appropriate all consumer sur-

plus by pricing at and the social cost of informing consumers is minimized

R,

(every consumer receives exactly one ad).

5.2. Differentiated Products

Grossman and Shapiro (1984) extend the Butters model to include horizontal

product differentiation. In their model, firms are located around a circle, and the

number of firms may be endogenous. Following Tirole (1988, Section 7.3.2), I

focus here on a duopoly model in which firms are located on a line. Even when

simplified in this way, the Grossman-Shapiro model offers novel insights, and at

the same time provides a unified framework within which to interpret a broad

range of issues that arise in earlier writings and empirical efforts.

Consider then the following model. A unit mass of consumers are uniformly

distributed along a line of unit length. Each consumer has reservation value R

for a single unit of an ideal product, and suffers a transportation cost per unit

t

of distance from the ideal. There are two firms, located at opposite endpoints.

Advertising operates as in Butters’s model: a consumer can learn of a firm’s exis-

tence and price only by receiving an ad from that firm, and each ad is distributed

randomly over consumers. The cost of reaching a fraction of consumers is

Φ

i

Grossman and Shapiro allow for general advertising technologies

denoted ).

A(Φ

i

(of which the Butters’s technology given in (5.1) is a special case), and I specify

2

here a quadratic relationship: where

A(Φ ) = a(Φ ) /2, a > t/2.

i i

There are again three kinds of consumers. If firms and advertise at levels so

1 2

that fractions and of consumers are reached, respectively, then a fraction

Φ Φ

1 2

− of consumers receive no ad and are uninformed. A fraction

][1 Φ ]

[1 Φ

1 2

− receive only firm ads and are thus captive to firm likewise, a

Φ [1 Φ ] 1’s 1;

1 2 −

fraction are captive to firm Finally, a fraction consumers

Φ [1 Φ ] 2. Φ Φ

2 1 1 2

receive ads from both firms and are thus selective. Suppose that is sufficiently

R

large that a consumer purchases if any ad is received. Suppose also that the

number of selective consumers is of sufficient size that the firms compete for this

common demand. This is the case if the cost of advertising is not too great.

What demand function does firm confront? If the firms choose prices

1 P

1

and , respectively, then the marginal selective consumer is located at

P x =

2 74

− When a firm chooses its advertising expenditure, it equivalently

(P + t)/2t.

P

2 1

chooses its reach. Firm demand function thus may be written as follows:

1’s − −

D (P , P , Φ , Φ ) = Φ [(1 Φ ) + Φ (P + t)/2t].

P

1 1 2 1 2 1 2 2 2 1

The informative view holds that a firm faces a more price-elastic demand in mar-

kets with greater advertising. This elasticity effect is confirmed here. Firm 1’s

and is easily

elasticity of demand when evaluated at = P = P Φ = Φ = Φ

P

1 2 1 2

shown to be which is increasing in and thus in the market level

[ΦP/(2 Φ)t], Φ

of advertising.

Consider now a game in which the two firms simultaneously choose their prices

and advertising levels. If the marginal costs of production are constant, firm 1

thus chooses It is now

and to maximize

P (P , P , Φ , Φ ) A(Φ ).

Φ [P c]D

1 1 1 1 1 2 1 2 1

straightforward to derive price and advertising reaction curves and then solve for

e e e e e e

a symmetric equilibrium, The equilibrium is

and

P = P = P = Φ = Φ .

Φ

1 2 1 2

characterized as follows: √

e

2 Φ

e · (5.2)

= c + t = c + 2at,

P e

Φ 2

e p (5.3)

=

Φ ,

1 + 2a/t

2a

e p (5.4)

= ,

Π 2

(1 + 2a/t)

e

where is the equilibrium profit earned by a single firm.

Π

At a positive level, these equations yield a number of important implications.

Consider first the equilibrium price. As (5.2) reveals, it is higher than c + t,

which is the price that would emerge were consumers informed of all prices. The

reason is that demand is less elastic in the presence of informational product

differentiation. This does not mean that advertising increases prices; indeed,

the market would close in the absence of advertising. As Stigler (1961) and

Ozga (1960) suggest, it is consumer ignorance that leads to higher prices, while

advertising provides information and lowers prices. This may be confirmed by

noting that the equilibrium price falls when the cost of advertising falls (i.e., when

decreases). Second, using (5.3), the equilibrium advertising level is higher when

a

advertising is less costly and when products are more differentiated (i.e., when

is greater). The latter effect suggests that greater product differentation leads

t 75

to more advertising. This contrasts with the empirical interpretations offered by

Comanor and Wilson (1967, 1974), who posit that advertising induces product

differentiation. In the Grossman-Shapiro model, where advertising is endogenized,

product differentiation induces advertising.

As (5.4) shows, equilibrium profit is increasing in product differentiation and,

more surprisingly, the cost of advertising. When increases, the direct effect is

a

that each firm experiences a cost increase, but the resulting decrease in advertis-

ing also gives rise to a strategic effect: each firm faces a less elastic demand and

thus charges a higher price. The strategic effect dominates here, and firms benefit

overall when advertising is more costly (but not prohibitively so). This finding

provides a formal interpretation of work by Benham (1972) and others (see Sec-

tion 3.2.4) suggesting that some professions encourage legal restictions on adver-

77

tising. It also offers a formal interpretation of the profit-advertising relationship

described by Comanor and Wilson (1967, 1974) and others (see Section 3.2.2). In

the Grossman-Shapiro model, advertising does not cause profit, nor does profit

cause advertising. Instead, as (5.3) and (5.4) confirm, advertising and profit are

both endogenous variables that are jointly determined from exogenous variables

corresponding to the extent of product differentiation and the cost of advertising.

In a given sample of industries, as the extent of product differentiation varies,

advertising and profit move together; however, as the cost of advertising varies,

advertising and profit may move in opposite directions. From this perspective,

the sign of an observed correlation between advertising and profit simply reflects

which of the exogenous variables varies most in the sample at hand.

The key normative finding in this model is that advertising may be inadequate

78

or excessive. To understand the various effects, consider first additional adver-

tising by a firm that reaches a consumer who otherwise would be uninformed. The

social benefit of such advertising exceeds the private benefit, since the firm is un-

able to appropriate the resulting consumer surplus. This suggests that advertising

is inadequate. Consider next the effect of additional advertising by a firm that

reaches a consumer that also receives an ad from the other firm. Social surplus is

created if the consumer is located closer to the firm that undertakes the additional

advertising. The advertising firm does not internalize this matching benefit, and

77 But see also Peters (1984), who considers a model in which firms sell a homogeneous good,

face capacity constraints, and are privately informed as to their respective costs of production.

He argues that advertising restrictions may benefit high-cost (harm low-cost) producers, and it

can happen that prices are lower when advertising is restricted. See also LeBlanc (1998).

78 See Tirole (1988, p. 294) for a formal confirmation of this finding.

76

so the matching effect also suggests that market advertising is inadequate. But

the firm is motivated by the profit margin that it would enjoy on the “stolen”

consumer, while social welfare is not impacted by the re-distribution of margins

from one firm to another. This business-stealing externality parallels Marshall’s

(1919) notion of combative advertising and suggests that the market advertising

may be excessive.

Bester and Petrakis (1995) offer an interesting extension. In their model,

consumers live in one of two regions, where each region has a single firm. All

consumers are informed of the existence of both firms, and every consumer also

knows the price of the “local” firm. A consumer forms an expectation as to the

price charged by the distant firm, and a consumer learns the actual price if an ad

is received from the distant firm. In this setting, Bester and Petrakis character-

ize a symmetric mixed-strategy equilibrium. Their findings share features with

those of Robert and Stahl and also Grossman and Shapiro. In equilibrium, with

some probability a firm posts a “list price” and attracts only local consumers and

with the remaining probability a firm advertises a low “sale” price and tries to

79

attract distant consumers. Moreover, firms gain from an increase in the cost of

advertising, and market advertising may be inadequate or excessive.

Finally, Rogerson (1988) considers a model in which firms advertise prices and

also select product qualities. Each consumer observes the advertised prices, selects

a firm, observes the product quality offered by this firm, and then decides whether

to purchase or engage in sequential search. Consumers have heterogeneous search

costs and differ also in their willingness to pay for quality. Rogerson characterizes a

monopolistically competitive equilibrium, in which firms that offer higher-quality

products also enjoy larger markups. Consumers infer quality from the advertised

price, and those that are more willing to pay for quality select firms that advertise

higher prices. Intuitively, a firm will not “rip off” its consumers with a lower

quality, if the implied cost savings on those consumers that remain would be

small in comparison to the markup that would be lost on those consumers that

search again. At higher quality, the potential cost savings are greater, and a

higher markup is needed to dissuade the firm from cheating. With U-shaped

average costs and a zero-profit requirement, larger markups must be paired with

lower sales; thus, Rogerson finds that higher-quality firms are smaller. Rogerson

also examines a no-advertising benchmark, finding that social welfare is higher

when advertising is allowed.

79 See also Bester (1994) and Caminal (1996), as discussed in footnote 75.

77

5.3. Non-Price Advertising

Bagwell and Ramey (1994a) emphasize two features of the modern retail market.

First, in many retail categories, large-scale discount firms co-exist with small firms,

and the large firms share a common set of attributes: high sales volumes, heavy

advertising expenditures, low prices and large investments in advanced selling

technologies. Second, competition between retail firms often occurs through non-

price advertising. The typical TV or radio retail ad, for example, contains little or

no direct (“hard”) information. Motivated by these features, Bagwell and Ramey

develop a new model of the retail firm and offer an equilibrium interpretation of

80

non-price advertising by retailers.

The retail-firm model is easily described with reference to a monopolist that

expects consumers, where each consumer possesses the positive and

N > 0

downward-sloping individual demand function, The monopolist chooses

d(P ).

≥ ≥

a price and a level of investment The cost of investment

P 0 K 0.

is per unit, and the benefit of greater investment is that the marginal

r > 0 0

cost of selling is thereby reduced: The firm’s net revenue is thus

c (K) < 0.

≡ − − For given let and be

R(P, K, N) [P c(K)]Nd(P ) rK. N, P (N) K (N)

M M

the price and investment levels that jointly maximize Assuming that second-

R.

order conditions are satisfied, these monopoly values satisfy Let

R = R = 0.

P K

∗ ≡ denote the maximized value of net revenue.

Π (N) R(P (N), K (N), N)

M M

Bagwell and Ramey establish a “coordination economy” that is enjoyed by a

firm and its consumers when the firm gets larger. First, using a standard envelope

argument, it follows that a firm does better when it expects more consumers:

∗0 Second, a consumer also does better when a firm expects more

Π (N) > 0.

consumers. To see this, observe that at the monopoly values (since

R = 0 R =

P N P

0

0 0 0

−c −c

and It follows that and

0), R = d > 0 R = Nd < 0. K (N) > 0

KN KP M

0 a monopolist invests more and prices lower when more consumers are

P (N) < 0 :

M

expected. Intuitively, an investment that reduces marginal cost is more attractive

when a higher sales volume is anticipated, and the reduction in marginal cost in

turn makes a lower price more attractive.

At a general level, it is now possible to anticipate a role for non-price adver-

80 Advanced selling technologies include advanced information systems (electronic-scanner

checkout systems, privately owned satellites) and superior delivery systems (privately owned

warehouses and trucks). Bagwell and Ramey observe, too, that large firms offer greater product

variety, and their retail-firm model accounts for this attribute as well. I present a simplified

single-product model below. It is, however, useful to keep the multi-product version in mind,

since the focus on non-price advertising is most compelling for a retailer with many products.

78

tising. Imagine that consumers do not observe the firm’s price until after a search

cost is incurred. Consumers do, however, observe the firm’s non-price advertising.

0 Then, a higher

Suppose now that advertising attracts consumers: (A) > 0.

N

advertising level leads to greater expected sales, which in turn induces greater

investment and thereby a lower price. Therefore, if it is supposed that consumers

respond to advertising, then a firm that advertises heavily also adopts a low price,

and so the supposed responsiveness of consumers to advertising becomes justified.

From this perspective, it is entirely rational for consumers to respond to non-price

advertising. This conclusion holds as well in multi-firm markets, if a firm expects

greater market share when it advertises more heavily.

To go further, an equilibrium model of retail advertising is required. Consider a

three-stage game. In the first stage, firms decide whether to enter. Entry entails

a sunk cost, In the second stage, the firms simultaneously make price,

σ > 0.

investment and advertising selections. Finally, in the third stage, each consumer

picks a firm from which to buy, based on the information that the consumer

possesses. There are two kinds of consumers. Informed consumers observe the

81

firm that makes the greatest advertising expenditure. Uninformed consumers do

not observe advertising efforts. There is a unit mass of consumers in total. Let I

and denote the exogenous proportions of informed and uninformed consumers,

U ∈

respectively, where and No consumer observes a firm’s

I + U = 1 U (0, 1).

price and investment selections prior to picking a firm. The firms thus select their

monopoly price and investment levels, given the number of consumers that they

respectively expect. random equilibrium

As a benchmark, consider the that obtains when con-

sumers are not responsive to advertising and thus pick firms at random. Entering

firms then choose zero advertising and divide the market. Thus, if firms enter,

n

each firm expects consumers. Ignoring integer constraints, the equilib-

N = 1/n ∗

rium number of firms is the value Each firm adopts

that satisfies

n (1/n ) = σ.

Π

r r

the price P (1/n ).

M r advertising equilibrium,

Now consider an in which informed consumers adopt

the rule of thumb of buying from the firm that advertises the most. As Bag-

well and Ramey show, for any a mixed-strategy equilibrium is induced.

n 2,

This equilibrium is characterized by a distribution function that ensures

F (A)

for each firm that the higher cost of additional advertising is balanced against

81 Bagwell and Ramey thus posit a different advertising technology than do Butters (1977)

and Grossman and Shapiro (1984). In the latter work, an individual consumer that receives an

ad (or ads) has no further information as to the respective advertising expenditures of firms.

79

the benefit of a higher expected sales volume. In an advertising equilibrium,

when a firm advertises at level it expects to win the informed consumers with

A,

n−1

probability . Therefore, when a firm advertises at level it expects

F (A) A,

n−1 consumers, which is indeed an increasing function. The

I + U/n

N(A) = F (A) ∗ n−1 ∗

distribution function is formally defined by (F (A) I +U/n)−κA = Π (U/n),

Π

∗ is the profit that is enjoyed by a firm that chooses zero advertis-

where (U/n)

Π

ing. Observe that the simple rule used by informed consumers is rational. When

n−1

a firm advertises at level it sets the price and so higher-

A, P (F (A) I + U/n),

M

advertising firms indeed offer lower prices. This is consistent with the general

discussion above. Finally, ignoring integer constraints, entry occurs until ex-

pected profit is zero. Since a firm is indifferent over all advertising selections in

the support, the equilibrium number of firms is the value that generates zero

n

a

profit when a firm selects zero advertising: (U/n ) = σ.

Π a

The random equilibrium would obtain, for example, if advertising were pro-

hibited, while the advertising equilibrium might be predicted when advertising is

legal. It is thus interesting to compare these two equilibria. Observe first that

the market with advertising has fewer firms that are on average larger: .

n > n

r a

∗ ∗ ∗0

This observation follows, since and imply

Π (1/n ) = σ = Π (U/n ) Π > 0

r a

that Observe second that, with probability one, in a market with

1/n = U/n .

r a

advertising every firm offers a lower price than in a market without advertising.

To see this, observe that the highest possible price in the advertising equilibrium

occurs when a firm does not advertise, and the price then charged equals the price

that is always offered in the random equilibrium: There-

P (U/n ) = P (1/n ).

M a M r

fore, expected consumer welfare is higher when advertising is allowed. Since firms

make zero expected profit either way, social welfare is higher when advertising is

allowed than when it is not.

Bagwell and Ramey capture and build upon a number of themes from ear-

lier work. In line with Chamberlin’s (1933) work, they construct a monopolisti-

cally competitive equilibrium, in which profits are dissipated through advertising

expenditures and entry, and advertising operates through a scale effect to facil-

itate lower prices. But the scale effect that they utilize is a “long-run” effect,

under which greater expected sales volume leads to additional cost-reducing in-

82

vestments. In addition, their advertising equilibrium exhibits endogenous firm

heterogeneity: some firms advertise heavily, enjoy high expected sales, choose low

82 Bagwell and Ramey show that a short-run scale effect, corresponding to marginal costs that

decline with output, would reinforce their findings. Declining marginal costs may be relevant

when large retailers receive quantity discounts.

80

prices and make large investments, while other firms advertise less but expect low

sales and set high prices while making small investments. Bagwell and Ramey

also provide a formalization of Nelson’s (1974b) signaling-efficiency effect. In par-

ticular, they find that a choice of heavy advertising is paired with a selection of

large investment, and so rational consumers indeed can use ostensibly uninforma-

tive advertising expenditures as an indication of low costs and thus low prices.

Finally, they offer an equilibrium interpretation for the empirical finding of Ben-

ham (1972) and others (see Section 3.2.4) that the introduction of even non-price

advertising leads to the entry of large-scale firms and lower prices.

Bagwell and Ramey (1994b) offer an alternative formulation. In a first model,

where one firm is known to be more efficient than a second, they show that the

possibility of advertising ensures that all consumers coordinate on the efficient

firm. If another equilibrium were posited, then the efficient firm could break this

equilibrium by advertising heavily. Sophisticated consumers would understand

that the efficient firm could then possibly profit, only if it were to receive a large

number of consumers and price at the associated low monopoly price. Advertising

is not required on the equilibrium path, though, once consumers are coordinated

on the efficient firm. In a second model, a firm is privately informed as to whether

it is more efficient than its rival. In the (refined) separating equilibrium, when

this firm is more efficient, it advertises a positive amount on the equilibrium

path, in order to signal its low costs and the associated low monopoly price. This

prediction confirms Nelson’s (1974b) signaling-efficiency effect, when a firm’s level

of efficiency is exogenous and privately known.

In the Bagwell-Ramey (1994a,b) models, a consumer desires to visit a firm

that expects a large number of other consumers. In this sense, an “indirect” net-

work externality exists between consumers. Chwe (2001), Clark and Horstmann

(2001) and Pastine and Pastine (2002) consider the related but distinct case of

a “direct” network externality among consumers, whereby a consumer enjoys the

social prestige that is associated with purchasing from a firm that actually sells

83

to a large number of other consumers. Under both approaches, advertising may

promote improved coordination and welfare gains.

83 For further discsussion of research on network externalities, see the contribution to this

volume by Farrell and Klemperer. 81

5.4. Loss Leaders

The discussion above emphasizes extreme cases, in which a firm can advertise all

or none of the prices of its products. In many categories, retailers carry thousands

of items, and it is clearly not realistic to assume that all prices can be meaningfully

advertised. As mentioned in Section 3.2.4, however, a firm may then advertise

the price of particular “loss-leader” products.

I describe here the “commitment” and “signaling” theories of loss-leader pric-

84

ing. The commitment theory is advanced by Lal and Matutes (1994). In their

duopoly model, one firm is located at each endpoint of the Hotelling line, each

firm offers two products and each firm can advertise the price of just one product.

For any firm, consumers observe one advertised price at zero cost and must pay

a search cost to observe the other, unadvertised price. Except for their locations,

consumers are identical and have independent unit demands for both goods. For

simplicity, suppose that consumers have a common reservation price for each

R

good. A firm then faces a commitment problem: it is unable to credibly promise

that it will charge a price below on an unadvertised good. In the absence of any

R

advertising, therefore, consumers would foresee that all products are priced at R

and choose not to visit any firm, thus saving the search cost. In the presence of

advertising, however, a firm can use an advertised loss-leader price to guarantee

sufficient consumer surplus to justify costly search, even though the unadvertised

good is priced at Notice that consumers rationally expect that the price of the

R. 85

unadvertised product is independent of the advertised loss-leader price.

Lal and Narasimhan (1996) extend the commitment theory to include a pre-

ceding stage in which the manufacturer of the loss-leader good selects a wholesale

price and a level of advertising. Manufacturer advertising raises the demand for

the manufacturer’s good; specifically, it increases the reservation value that con-

sumers have for a second unit of the loss-leader good. Lal and Narasimhan argue

that manufacturer advertising may lower the retail price and raise the wholesale

price of the loss-leader good, so that the retail margin is reduced and the whole-

sale margin is increased. This theory provides a formal foundation for work by

84 See also Lal and Rao (1997) and Wernerfelt (1994). Gerstner and Hess (1990) and Hess and

Gerstner (1987) offer related analyses that feature loss-leader pricing, bait-and-switch tactics

and rain checks.

85 In their contribution to this volume, Farrell and Klemperer discuss a related “bargains-

then-ripoffs” theme that arises in dynamic models with switching costs. For interesting recent

contributions to the commitment theory, see Ellison (forthcoming), Konishi and Sandfort (2002)

and Rao and Syam (2001). 82

Steiner (1973, 1978, 1984, 1993) and others, as discussed in Section 3.2.4.

Building on Nelson’s (1974b) signaling-efficiency effect, the signaling theory

holds that a firm uses an advertised low price to signal low costs. Consumers then

rationally expect the price of the unadvertised product to be relatively low as

well. Bagwell (1987) offers a formalization of this general idea. He considers a two-

period model, in which demand is downward-sloping, consumers must pay a search

cost to observe the current price and a firm is privately informed as to whether its

costs are high or low. A low-cost firm may signal its costs with an “introductory

sale.” The firm then obtains greater repeat business, since consumers rationally

expect the firm to charge a low price in the future. As Bagwell (1987, p. 384)

notes, his two-period, single-good model may be reinterpreted as a single-period,

two-good model, in which an advertised loss-leader price signals low costs and thus

a low price on the unadvertised product. Simester (1995) develops this loss-leader

86

model in detail and records a number of interesting predictions.

5.5. Summary

The multi-firm models described above yield a striking set of predictions. Some

of these predictions confirm and extend ideas found in earlier writings. The for-

mal models also offer a number of new predictions. At a normative level, this

work offers support for the presumption that retail markets perform better when

advertising is possible. This is true for both price and non-price advertising. At

the same time, there is no presumption that the level of advertising is optimal.

As in the normative theory of monopoly advertising, inadequate advertising may

arise, since a firm does not internalize the consumer surplus that an additional ad

may generate. Furthermore, when there are multiple firms, excessive advertising

may occur, since a firm privately benefits from the sale that an additional ad may

generate, even when this sale is “stolen” from another firm and offers no or modest

social benefit. Finally, in many retail categories, large retailers sell thousands of

products. The amount of direct price information that advertising can convey is

then necessarily limited. Important future work might consider further the role

of price and non-price advertising activities by multi-product retailers.

86 See also Bagwell and Ramey (1994b). They provide an extended model in which loss-leader

pricing is used to ensure that consumers coordinate on the most efficient firm.

83

6. Advertising and Quality

Nelson (1974b) predicts a positive relationship between advertising and product

quality, especially for experience goods. In support of this prediction, he identifies

the signaling-efficiency, repeat-business and match-products-to-buyers effects. As

discussed in Section 3.2.5, however, the empirical literature offers mixed support

for this prediction. I consider now recent theoretical analyses of advertising and

quality. I organize this discussion around the three effects that Nelson (1974b)

identifies. The signaling-efficiency effect is formalized using a static model. A

related dynamic model is then presented, so that the repeat-business effect may

be examined. This is followed by a short discussion of the match-products-to-

buyers effect. Finally, I also discuss research that considers advertising in the

context of the quality-guarantee effect.

6.1. Signaling-Efficiency Effect

Nelson (1974b) argues that demand expansion is most attractive to efficient firms.

Such firms may enhance demand by advertising heavily, setting low prices and

providing high quality; consequently, consumers may draw inferences as to the

deal that a firm offers after observing its advertising. Above, I discuss the manner

in which observed advertising may signal efficiency and thereby price for a retailer

that offers search goods. I consider now how observed advertising and price may

signal efficiency and thereby the (exogenous) quality of an experience good. This

analysis may be most relevant for the manufacturer of a new product.

My approach is to draw on techniques developed by Bagwell and Ramey (1988)

for signaling games with multiple signals. They analyze price and advertising as

signals of cost in an entry-deterrence model (see Section 7.2 below). As Bagwell

(1992) and Overgaard (1991) observe, these techniques also can be used to analyze

87

how a high-quality monopolist best uses multiple signals to signal its quality. In

88

particular, Overgaard examines the static model that I now summarize.

87 Here and in Section 7.2 below, I illustrate these techniques using a simple Lagrangian

argument. A more general treatment is available in the original papers.

88 See also Zhao (2000), who places additional structure on the demand function and derives

a related set of findings. For models in which a high-quality monopolist signals its quality using

only price, see Bagwell and Riordan (1991) and Bagwell (1991, 1992). Under the assumption

that a higher-quality product entails a higher marginal cost, they show that the high-quality

monopolist adopts a high (supra-monopoly) price. This prediction is maintained below, when

the model is expanded to include advertising as a signal.

84

Formally, suppose a monopolist privately observes whether its product-quality

∈ {L, ≥

type is low or high, and then selects a price and an advertising

t H}, P 0

≥ ∈

level Consumers observe and form some belief

A 0. P A, b = b(P, A)

as to the likelihood of high quality, and then demand units,

[0, 1] D(P, A, b) > 0

≥ or

where and Advertising may be dissipative (D

< 0 < D = 0),

D D 0.

P b A A

it may contain information and/or induce social prestige and thus be demand-

enhancing Let denote the constant marginal cost of production

(D > 0). c(t)

A

when quality is type If then the high-quality monopolist is also

t. c(H) < c(L),

the efficient (low-cost) monopolist. This is the case to which Nelson’s (1974b)

signaling-efficiency effect refers. As Schmalensee (1978) emphasizes, however, it

may be more plausible to assume that a high-quality product has a higher marginal

89

cost: Both cases are considered here.

c(H) > c(L). ≡ − −

A monopolist of type makes profit

t Π(P, A, b, t) (P c(t))D(P, A, b)

For fixed and assume has unique maximizers, and

κA. t b, Π(P, A, b, t) P (t, b)

m

and is strictly concave in In the complete-information benchmark, the

A (t, b), P.

m ≡ ≡

monopoly selections are and

(P (H), A (H)) (P (H, 1), A (H, 1)) (P (L), A (L))

M M m m M M

The complete-information monopoly profits are

(P (L, 0), A (L, 0)). π (H) =

m m M

and Assume

Π(P (H), A (H), 1, H) π (L) = Π(P (L), A (L), 0, L). π (t) >

M M M M M M

∈ {L,

for so that both quality types are profitable.

0 t H},

Perfect Bayesian Equilibrium {P

A is a set of strategies, , and be-

(t), A(t)}

t=L,H

∈ {L,

liefs, such that: (i) for each maximizes

b(P, A), t H}, (P (t), A(t)) Π(P, A, b(P, A), t),

and (ii) is derived from the equilibrium strategies using Bayes’ rule when-

b(P, A) separating equilibria, 6

ever possible. I focus here on in which (P (H), A(H)) =

and thus In a separating

(P (L), A(L)) b(P (H), A(H)) = 1 > 0 = b(P (L), A(L)).

equilibrium, the low-quality monopolist is “found out.” It can do no better than to

make its complete-information selections, and

(P (L), A(L)) = (P (L), A (L)),

M M

90

earn the corresponding profit, Thus, if the high-quality monopolist is to

π (L).

M

separate, then it must choose some pair that the low-quality monopolist

(P, A)

89 While it is plausible that unit costs tend to be higher for higher-quality products, this

relationship may fail if higher-quality products achieve greater market share and thereby enjoy

scale economies. Phillips et al (1983) use PIMS data and report that businesses with higher

relative quality often have higher market shares and lower relative unit costs.

90 (L), A (L)). Then the low-quality monopolist could deviate

Suppose (P (L), A(L)) 6 = (P

M M

to (P (L), A (L)) and earn strictly higher profit, since

M M

Π(P (L), A(L), 0, L) < Π(P (L), A (L), 0, L) ≤ Π(P (L), A (L), b(·), L)

M M M M

b(·) ≡ b(P (L), A (L)). D > 0 π (L) > 0,

where The final inequality follows since and with

M M b M

P (L) > c(L).

the latter implying that M 85

would not mimic: ≤ (6.1)

(L).

Π(P, A, 1, L) π M

To make the problem interesting, assume that signaling is costly for the high-

quality monopolist: does not satisfy (6.1).

(P (H), A (H))

M M

least-cost separating equilibrium,

In the the high-quality monopolist separates

in the way that it finds most profitable. The least-cost separating equilibrium

is of particular interest, and it is also selected when standard refinements are

∗ ∗

employed. Formally, define as the price-advertising pair that solves

(P , A ) subject to (6.1). (6.2)

Max Π(P, A, 1, H)

P,A ∗ ∗ Following argu-

In a least-cost separating equilibrium, , A ).

(P (H), A(H)) = (P

ments by Bagwell (1992) and Overgaard (1991), the existence of the least-cost

separating equilibrium may be established. Here, I focus on the characterization

of this equilibrium.

To gain some intuition, consider any two price-advertising pairs, (P , A )

1 1

and that yield the same profit for a mimicking low-quality monopolist:

(P , A ),

2 2 Observe that

Π(P , A , 1, L) = Π(P , A , 1, L).

1 1 2 2

− (6.3)

Π(P , A , 1, H) Π(P , A , 1, H)

2 2 1 1

− − −

= [Π(P , A , 1, H) Π(P , A , 1, H)] [Π(P , A , 1, L) Π(P , A , 1, L)]

2 2 1 1 2 2 1 1

− , A , 1) D(P , A , 1)].

= [c(H) c(L)][D(P

1 1 2 2 to that

Suppose that and consider a change from , A ) (P , A )

c(H) > c(L), (P

1 1 2 2

leaves the low-quality monopolist indifferent. According to (6.3), if demand is

lower at the new price-advertising pair, then the high-quality monopolist gains

from the change. The key idea is that demand-reducing changes are more at-

tractive to the high-quality monopolist when marginal costs increase with qual-

ity, since the demand reduction then offers a greater cost savings. Similarly, if

then a change that leaves the mimicking low-quality monopolist

c(H) < c(L),

indifferent and enhances demand is preferred by the high-quality monopolist.

Further insight may be gained by analyzing the program given in (6.2). The

≡ −

Lagrangian is Using (6.3),

L(P, A, λ) Π(P, A, 1, H) + λ[π (L) Π(P, A, 1, L)].

M 91

it may be verified that at the optimum. The Lagrangian may be

λ (0, 1)

91 ∗

The costly-signaling assumption implies that λ > 0. Given A , the first-order condition

L = 0 P as the solution to

determines

P 86

rewritten as −

c(H) λc(L) −

− − (6.4)

(L).

)D(P, A, 1) κA} + λπ

L(P, A, λ) = (1 λ){(P M

1 λ

As the bracketed term in (6.4) reveals, in the least-cost separating equilibrium,

the high-quality monopolist makes the same price-advertising selection as it would

were it to produce at constant marginal cost and offer a

c [c(H)−λc(L)]/(1−λ)

o ∈ −

product of known high quality. Observe that implies

λ (0, 1) sign{c c(H)} =

o

sign{c(H) c(L)}.

Consider first the case in which a high-quality product entails a greater marginal

cost: In the least-cost separating equilibrium, the high-quality mo-

c(H) > c(L).

nopolist then undertakes a “cost-increasing distortion,” in that it sets the same

price-advertising pair as it would were its quality known but its marginal costs

higher It is natural to assume that under complete information a

(c > c(H)).

o

monopolist would choose a higher price and less demand-enhancing advertising

92

were its constant marginal costs increased. Under this assumption, in the least-

cost separating equilibrium, the high-quality monopolist distorts its price upward

∗ ∗

and its demand-enhancing advertising downward

> P (H)) (A < A (H)).

(P M M

A high-quality monopolist thus best signals its quality with a high price and a low

level of demand-enhancing advertising. In essence, the high-quality monopolist is

signaling that it has high costs and is willing to reduce demand.

Consider second the case in which a high-quality product entails a lower

marginal cost: Then, in the least-cost separating equilibrium, the

c(H) < c(L).

∗ ∗ ∗

Π (P, A , 1, H) − Π (P, A , 1, L) = (λ − 1)Π (P, A , 1, L)

P P P

c(H) > c(L).

Take the case in which The left-hand side is then positive. Consider the right-hand

e

e ∗ ∗

(t) maximize Π(P, A , 1, t). Suppose first that P < P (L). Then let (P , A ) =

side. Let P

m m 1 1

∗ ∗ ∗

e

∗ ∗ ∗ ∗ ∗ ∗

, A ) and (P , A ) = (P , A ), where P > P (L) satisfies Π(P , A , 1, L) = Π(P , A , 1, L).

(P 2 2 m

∗ ∗

∗ ∗ ∗ ∗

P > P , A , 1) > D(P , A , 1). Using c(H) > c(L) and (6.3), the high-quality

implies D(P ∗ ∗ ∗ ∗

, A ), which contradicts that (P , A ) solves (6.2). Suppose

monopolist strictly prefers (P e

e e

∗ P

= P (L). It follows from c(H) > c(L) that (L) < P (H). Starting at

second that P m m m

∗ ∗ ∗ ∗ ∗

(P , A + ε, A ), satisfies P + ε ≤

), consider a small price increase, so that the new pair, (P

e

P (H). Given the strict concavity of profit in price, the high-quality (low-quality) monopolist

m ∗ >

strictly prefers the new (old) pair, and again a contradiction is reached. It must be that P

e ∗ ∗

P (L). Π (P, A , 1, L) < 0 P ; hence, the right-hand side is positive if and

This implies that at

m P

λ < 1. c(H) < c(L).

only if A similar argument applies when

92 For instance, it may be verified that this assumption is satisfied when demand is described

by either of the two examples considered in Section 4.1.2.

87 The

high-quality monopolist undertakes a “cost-reducing distortion” (c < c(H)).

o

∗ and an upward distortion

result is a downward pricing distortion < P (H))

(P M

in the level of demand-enhancing advertising As Nelson (1974b)

(A > A (H)).

M

predicts, the high-quality monopolist best signals its quality with a low price and

high level of demand-enhancing advertising. Fundamentally, the high-quality mo-

nopolist is signaling that it has low costs and welcomes an expansion in demand.

What if advertising is dissipative? Whether marginal cost rises or falls with

product quality, dissipative advertising would not be used by a monopolist with

93

a known high-quality product; thus, such advertising is not used as a signal. In

this model, advertising is used as a signal of quality only if it is demand-enhancing.

While the model is static, the findings suggest a dynamic perspective. In

particular, once the monopolist’s product is sufficiently mature, consumers are

presumably informed about its quality, and so the high-quality monopolist then

sets its price and advertising at their complete-information levels (P and

(H)

M

Over the long run, the model thus predicts that the high-quality prod-

A (H)).

M

uct’s price declines and its demand-enhancing advertising increases, if marginal

cost rises with quality. The opposite prediction (rising price, declining advertis-

ing) applies when marginal cost falls with quality. Whether its product is new or

mature, the monopolist would never use dissipative advertising.

A further prediction is that the correlation between advertising and quality

fluctuates across market settings. Suppose that marginal cost rises with qual-

ity and consider a new product. Relative to the complete-information bench-

mark, a high-quality monopolist distorts its advertising downward (A(H) = A <

while a low-quality monopolist does not distort its advertising (A(L)

A (H)), =

M But it is not clear whether the level of complete-information advertising

A (L)).

M

is greater when quality is high or low. Intuitively, complete-information advertis-

ing is expected to be greater when quality is high, if marginal cost rises slowly

with quality and the marginal impact of advertising on demand rises quickly with

quality. Pulling these themes together, it is possible that the advertising-quality

correlation is positive under complete information and thus for a mature product,

94

and yet the correlation is negative for a new product (A < A (L) < A (H)).

M M

More generally, when the advertising-quality correlation is stronger

c(H) > c(L),

(more positive, less negative) for a mature product. These findings offer a possible

93 L (P, A, λ) = κ[λ − 1] < 0,

Formally, if advertising is dissipative, then where the inequality

A

λ < 1

follows since (as shown in footnote 91). Thus, when advertising is dissipative, it is

∗ = 0.

optimally set at a boundary: A

94 For further discussion, see Orzach et al (2002) and Zhao (2000).

88

interpretation for empirical efforts (see Section 3.2.5) that report a generally weak

advertising-quality correlation that is stronger for established products.

The model may be extended to consider a monopolist of intermediate age,

so that some but not all consumers are informed of quality. If represents the

ζ

fraction of uninformed consumers, then the profit for a monopolist now depends

and

upon its type through its marginal cost the demand of informed consumers.

For example, the profit for a high-quality monopolist becomes ζΠ(P, A, b, H) +

− Linnemer (2002) develops a static model of this kind. When

(1 ζ)Π(P, A, 1, H).

and an intermediate number of informed consumers exists, he shows

c(H) > c(L)

that dissipative advertising may be used along with a high (supra-monopoly) price

to signal high quality. Linnemer’s model shares important formal features with

the Milgrom-Roberts (1986) model, as I explain below.

It is also possible to extend the model to allow for multiple sellers. Under the

assumption that advertising is dissipative, Kihlstrom and Riordan (1984) explore

a model in which quality is high or low and firms are competitive price takers,

where the price that is “taken” may differ depending upon whether a firm is

perceived to offer a high- or low-quality product. In this context, advertising

can be understood as an “entry fee” that is necessary to enter the high-quality

market. They show that dissipative advertising can signal high quality even in

a static model, if marginal cost is sufficiently lower when quality is high. The

idea is that a high-quality firm then enjoys a larger mark-up from a sale in the

high-quality market, and so the advertising expenditure can fall in a range that

95

only a high-quality firm would be willing to incur.

As Fluet and Garella (2002) and Hertzendorf and Overgaard (2001) demon-

strate, dissipative advertising may also signal high quality in a static duopoly

model. In the Hertzendorfer-Overgaard model, exactly one seller offers a high-

quality product, but consumers do not know the identity of this seller. A key

feature of this model is that the sellers share private information as to the identity

of the high-quality firm. As a consequence, one seller’s price-advertising selection

provides potential information concerning the other seller’s quality. This enriches

95 Wiggins and Lane (1983) also consider the manner in which advertising may signal quality

when prices are fixed. In their model, consumers are risk averse, and advertised products are of

more uniform quality. Horstmann and Moorthy (2003) examine a model in which competitive

firms face uncertain demand. Advertising by a firm can improve its capacity utilization in

low-demand states, by attracting consumers who otherwise would be uninformed. Since lower-

quality firms may have greater excess capacity in low-demand states, this particular benefit from

advertising can be greater for low-quality firms.

89

96

the set of signaling possibilities. Under the assumption that marginal cost is

independent of quality, dissipative advertising is sometimes used as a signal of

quality, and the correlation between advertising and quality is highest when the

quality difference is intermediate in size. Allowing that marginal cost increases

with quality, Fluet and Garella conduct a related analysis and find that any sep-

arating equilibrium entails positive advertising by the high-quality firm, provided

that the quality difference is not too great.

6.2. Repeat-Business Effect

Nelson (1974b) argues that advertising rekindles memories of experiences with the

advertised product. As recollections are more likely to prompt repeat business

when the quality of product is high, a high-quality product may be advertised to

a greater extent, and even new consumers may thus infer high quality from heavy

advertising. I now summarize several recent efforts that use explicit dynamic

models in order to capture a repeat-business effect. These efforts differ somewhat

from Nelson’s conception, in that a memory-activation process is not modeled;

instead, the repeat-business effect emerges in the following sense: the return from

advertising and thereby achieving an initial sale may be greater for a high-quality

product, due to the greater repeat purchases that come from satisfied customers.

Schmalensee (1978) offers a first formal investigation. As noted above, he

argues that the marginal cost of production is greater when a high-quality good is

produced. Under the assumption that all sellers must charge the same price, the

value of an initial sale may be greater when a low-quality good is sold, as then the

mark-up is larger. This “reverse” signaling-efficiency effect favors low-quality firms

and can counter the repeat-business effect that favors high-quality firms. Indeed,

Schmalensee demonstrates that low-quality products are more heavily advertised,

if consumers are responsive to advertising and marginal cost is sufficiently greater

for a high-quality product.

As Schmalensee acknowledges, a weakness of his model is that consumer behav-

ior is irrational: consumers are responsive to advertising, even though advertising

is associated with low-quality products. This weakness is addressed by Kihlstrom

and Riordan. I discuss above their finding for a static model, but they also con-

sider a two-period formulation that allows for a repeat-business effect. Due to this

effect, the value of an initial (first-period) sale is greater for a high-quality firm,

96 For other multi-sender signaling models, see Bagwell and Ramey (1991), de Bijl (1997) and

Matthews and Fertig (1990). 90

and so dissipative advertising can signal high quality even if low-quality firms

enjoy a modest marginal-cost advantage. The precise extent of the critical ad-

vantage varies with the particular assumption that is made as to the information

held by second-period consumers.

Working with a monopoly model, Milgrom and Roberts (1986) allow that

consumers may draw product-quality inferences from advertising and price. In

effect, they extend the static model with dissipative advertising from Section 6.1

to include a second period. The product is non-durable, and consumers have unit

demands in each period and heterogeneous reservation values. When a product

is consumed, the consumer discovers whether he is satisfied with the product. A

satisfied consumer enjoys the gross surplus (measured by the reservation value)

that the product offers, while an unsatisfied consumer receives zero gross surplus.

Product quality is operationalized as the probability that a randomly selected con-

sumer finds the product satisfactory. If the product is satisfactory for a consumer

in the first period, then it will remain so for this consumer in the second period.

In the second period, the monopolist sells only to consumers that purchased in

the first period and had a satisfactory experience.

The main features of their analysis may be understood with reference to a two-

period profit function, where

V (P, A, b, t) = Π(P, A, b, t) + δe

π (P, b, t), δ (0, 1)

is the discount factor, is the profit function used above in the static

Π(P, A, b, t)

e

model and is a reduced-form profit function for the second period. As above, the

π

consumers’ belief derives from first-period price and advertising observations:

b e

I assume that is decreasing in and increasing in since a firm

b = b(P, A). π P b,

can earn greater second-period profit if it sold to a larger number of consumers in

e

the first period. More importantly, I assume that embodies a repeat-business

π

e

effect in the following sense: is higher when than when

t = L t = H.

π P

The intuition for the repeat-business effect is as follows. For any given belief,

when the monopolist raises its first-period price, some consumers elect not to buy.

Consider whether these “lost” consumers are of greater value to a low- or high-

quality monopolist in the second period. There are two considerations. First,

lost consumers might be more painful for the high-quality monopolist, since a

greater fraction then would have been satisfied and thus given repeat business.

Second, if marginal cost increases with quality, then lost consumers might be

less painful for the high-quality monopolist, since a smaller markup then would

be enjoyed on those lost consumers that did offer repeat business. The first

(second) consideration works in favor of (against) the assumption made above.

The assumption therefore holds if the high-quality monopolist has a weak cost

91

advantage (c(H) or if any cost disadvantage of high-quality production

c(L))

is sufficiently modest.

Consider now the implications of the assumption that advertising is dissipa-

tive. First, the first-period profit function, depends directly upon

Π(P, A, b, t),

−κ. A second implication,

advertising only through the cost of advertising: =

Π

A

e

already reflected in the notation, is that depends on only through the belief

π A

function Third, advertising occurs (if at all) only in the introductory pe-

b(P, A).

riod. The monopolist would not advertise in the second period, since advertising

does not directly alter demand and no opportunities for signaling remain (for all

second-period consumers the product is already known to be satisfactory). Finally,

in a separating equilibrium, if the monopolist offers a low-quality product, then

it selects zero advertising. In analogy with the discussion above, in a separating

equilibrium, the low-quality monopolist picks its complete-information monopoly

price-advertising pair. When advertising is dissipative, the complete-information

solution entails zero advertising.

In the least-cost separating equilibrium, is it possible that the high-quality

monopolist picks positive advertising? Let denote the discounted two-

v (L)

M

period profit that the low-quality monopolist earns in a separating equilibrium.

In the least-cost separating equilibrium, the price-advertising pair selected by the

high-quality monopolist solves the following program: ≤

subject to (6.5)

Max V (P, A, 1, H) V (P, A, 1, L) v (L).

M

P,A

Suppose that the solution to (6.5) entails positive advertising. Then the first-

Given that

order condition for advertising is (P, A, 1, H) = λV (P, A, 1, L).

V

A A

advertising is dissipative, this condition reduces to Consider next the

λ = 1.

first-order condition for price. Using this can be written as

λ = 1, e

− − (6.6)

Π (P, A, 1, H) Π (P, A, 1, L) = δ[e

π (P, 1, L) (P, 1, H)].

π

P P P P

Thus, if the high-quality monopolist chooses a positive amount of dissipative

advertising, then the high-quality price must satisfy (6.6).

Consider first the case in which Then the left-hand side of (6.6)

c(H) > c(L).

is positive. The right-hand side of (6.6) is also positive, due to the repeat-business

effect. In this case, therefore, it is possible that a high-quality monopolist signals

its quality with a positive level of dissipative advertising along with a distorted

price. Milgrom and Roberts discuss the specific circumstances under which such a

92 97

separating equilibrium occurs. Consider second the case in which c(H) c(L).

Then the left-hand side of (6.6) is non-positive. Given that the right-hand side

is positive under the repeat-business effect, (6.6) cannot be satisfied. A main

conclusion is now apparent: in the least-cost separating equilibrium, the high-

quality monopolist uses dissipative advertising to signal its quality only if the

marginal cost of production is greater for a high-quality product.

The underlying intuition is as follows. When the monopolist raises its first-

period price, sales for the first period are reduced. If a high-quality product

entails a higher marginal cost, this first-period effect is less painful for a high-

quality monopolist. The price hike also reduces sales in the second period, since

there are then fewer satisfied consumers that emerge from the first period. Under

the repeat-business effect, this second-period effect is more painful for a high-

quality monopolist, as a greater fraction of its first-period consumers would have

had a satsifactory experience. Due to these offsetting effects, the cost of a price

increase can be equalized across the low- and high-quality types of monopolists

(i.e., (6.6) can hold). As both types also experience the same cost from dissipative

advertising, the monopolist may have no better option than to use both a distorted

price and a positive advertising expenditure when signaling high quality. By

contrast, if marginal cost (weakly) falls with quality, then the cost of a price hike

is (weakly) greater for a high-quality monopolist. The high-quality product is

then best signaled with a low price and no advertising.

It is interesting to compare the predictions of the Milgrom-Roberts model with

those of the static model. In the static model, dissipative advertising is not used as

a signal. Furthermore, when advertising is demand-enhancing and a

c(H) > c(L),

high-quality monopolist distorts its advertising downward, with advertising rising

in the future (once consumers are informed) to its undistorted level. By contrast,

in the dynamic model, if then a high-quality monopolist may use

c(H) > c(L),

dissipative advertising as a signal, with advertising falling in the future to its

undistorted level of zero. The inclusion of the repeat-business effect thus generates

novel predictions, illustrating further the complex relationship between advertising

and product quality. Finally, recall Linnemer’s (2002) extension of the static

model. In his model, the profit earned on informed consumers plays a role similar

98

to that played by second-period profit in the Milgrom-Roberts model.

97 In the two-period model, future demand depends directly upon actual quality, and it is

possible that a separating equilibrium exists even when c(H) = c(L).

98 In Linnemer’s model, the formal analog of the repeat-business effect emerges as follows:

over the range of prices that a high-quality monopolist might choose, a price increase diminishes

93

Hertzendorf (1993) offers an interesting extension. He supposes that consumers

observe the monopolist’s advertising expenditure with error. By contrast, the

monopolist’s price is perfectly observed. If no advertising is observed, it may be

unclear whether the firm failed to advertise or the consumer failed to observe

the advertising. In this setting, if the monopolist’s price reveals quality, then

the monopolist will not use advertising as a signal. Intuitively, if the monopolist

were to use advertising, then it could deviate to a lower advertising level, without

being detected and without altering the consumers’ belief (since price already

reveals quality). Advertising may be used, however, when the monopolist’s price

99

is independent of product quality. In this case, if repeat-business effects are

sufficiently large and/or marginal cost does not rise too swiftly with quality, then

the high-quality monopolist advertises to a greater extent.

Horstmann and McDonald (1994) consider a different kind of noise. In their

model, consumers observe price and advertising perfectly, but the consumption

experience generates only an imperfect indication of quality. Specifically, they

consider a two-period model in which a monopolist privately observes whether

the quality of its product is high or low, where the marginal cost of production

is independent of quality and in each period a higher-quality product yields a

satisfactory experience with a higher probability. A consumer’s experience with

the product is then not fully informative: a product may offer a satisfactory ex-

perience in the first period and fail to do so in the second period. In the first

period, there is no basis for the monopolist to use price and advertising as signals

of quality. Imperfect signaling is possible in the second period, however, since

this period has a greater expected number of satisfied consumers when quality is

high. In a refined equilibrium, second-period play takes the following form: the

high-quality monopolist prices high and advertises, while the low-quality monop-

olist sometimes adopts this behavior and otherwise sets a low price and does not

advertise. The high price is such that a consumer purchases in period two only

if the product yielded a satisfactory experience in period one. Two predictions

follow. First, advertising does not signal the quality of newly introduced goods.

Second, advertising can signal the quality of an established good, but even then

the signal is imperfect. These predictions offer further interpretations for em-

pirical efforts (see Section 3.2.5) that report a generally weak advertising-quality

the profit earned on informed consumers by a greater amount for a high-quality monopolist.

99 Moraga-Gonzalez (2000) offers a related finding in a model in which advertising provides

direct information about product quality but not all consumers observe advertising efforts.

94 100

correlation that is stronger for established products.

6.3. Match-Products-to-Buyers Effect

I consider next Nelson’s (1974b) match-products-to-buyers effect, whereby even

seemingly uninformative advertising can provide indirect information that im-

proves the match between product and buyer, since a firm has greater incentive

to send its ads to those consumers that value its product the most. Aspects of

this effect appear in some of the preceding discussion. In particular, Grossman

and Shapiro (1984) provide conditions under which advertising that contains di-

rect information as to a product’s existence, attributes and price serves to increase

consumer surplus by generating improved matches and expanded sales. In the fol-

lowing, I consider work in which the matching effect operates in markets for which

consumers are already informed of the existence of products. This work empha-

sizes advertising that provides information as to the attributes of the advertised

product, where the information may be direct or indirect.

Meurer and Stahl (1994) offer a model in which advertising provides direct

information as to horizontal attributes. In their model, there are two firms, and

each consumer desires one unit of the product. For a given consumer, one product

is a good match and offers gross utility , while the other product is a bad match

V

and offers zero gross utility. In the first stage of the game, firms simultaneously

choose advertising levels. An ad provides direct and truthful information as to

the attributes of the advertised product. A recipient of an ad thus knows whether

the advertised product offers a good match. If it does not, then the other product

must. For a consumer that receives no ad, the two products are homogeneous

and each provide an expected gross utility of Advertising therefore induces

V /2.

product differentiation. This is consistent with some of the arguments advanced

by proponents of the persuasive view, although here advertising-induced product

differentiation derives not from a change in tastes but from the information that

advertising provides. In the second stage of the game, each firm sets its (publicly

observed) price. The marginal cost of production is c < V /2.

As Meurer and Stahl show, the effects of advertising on social surplus are

non-monotonic. The equilibrium characterization entails mixed strategies, but

the key ideas are easily related. One the one hand, as advertising increases, more

consumers are “informed” (i.e., receive an ad) and thus obtain a good match. On

100 For other interpretations, see Orzach et al (2002), Zhao (2000) and the discussion above in

Section 6.1. 95

the other hand, at higher levels of advertising, the extent of product differentiation

is greater and each firm has more market power. In particular, each firm is then

especially tempted to raise price to and profit on those informed consumers

V

for which its product offers a good match. But expected sales are then reduced,

since uninformed consumers are unwilling to purchase at this price. The better-

matching and reduced-sales effects of advertising are conflicting. The result is a

non-monotonic relationship between advertising and social surplus. Building on

these themes, Meurer and Stahl show further that the Nash advertising level may

be excessive or inadequate.

A tension between the better-matching and reduced-sales effects of advertising

also arises in the monopoly models analyzed by Lewis and Sappington (1994) and

Johnson and Myatt (2004). Lewis and Sappington consider a monopolist that

may use advertising to supply pre-purchase information to buyers. Advertising

provides direct but possibly noisy information about product attributes and thus

raises the expected value of the product for some consumers while lowering it

for others. As Johnson and Myatt emphasize, advertising then induces greater

dispersion in consumers’ expected valuations and thereby generates a clockwise

rotation of the demand curve. In these models, the monopolist can vary the

precision of the information, by varying the content of the ads. At one extreme, if

the monopolist provides no information, then each consumer regards himself as an

“average type,” and the monopolist selects the monopoly price for that type. At

the other extreme, if the monopolist provides perfect information, then consumers

learn their respective valuations, and the monopolist then sets a higher price that

is attractive only to consumers with above-average valuations. This latter strategy

facilitates better matching but also entails reduced sales. The main finding in this

work is that the monopolist’s expected profit often achieves its maximum at one of

the extremes; thus, a profit-maximizing monopolist either provides no or perfect

information about product attributes. Further, the latter option is more attractive

when consumer valuations are heterogeneous and costs are high.

Anderson and Renault (forthcoming) also analyze a model of monopoly ad-

vertising. In their model, however, search costs play an important role, and

and

advertising may provide direct information as to product attributes price.

The basic model has a single consumer, who seeks one unit of the monopolist’s

product. The consumer can learn the product’s price and his “match” (reserva-

tion) value for the product by incurring a search cost. The monopolist is also

uncertain of the match value. When the search cost is sufficiently low, the con-

sumer is willing to incur the cost, even though the monopoly price is anticipated,

96

since he will enjoy positive consumer surplus if a high match value is realized. If

the search cost is higher, however, the consumer is unwilling to incur the cost,

unless the monopolist provides some information that raises the expected benefit

from search. In line with the discussion in Section 5.4, if the monopolist could

use advertising to transmit price information only, then it would raise the benefit

of search by advertising its commitment to a sub-monopoly price. Anderson and

Renault go further, however, and allow that the monopolist may use advertis-

ing to transmit price and/or attribute information. The consumer’s match value

may be determined by several product attributes, and the monopolist may elect

to offer partial match information. Importantly, such information may raise the

expected benefit of search for the consumer, by reassuring the consumer that the

match value is not too low. Building from these points, Anderson and Renault

find that the monopolist uses advertising to transmit partial match information

for intermediate levels of the search cost, and uses advertising to transmit price

and partial match information when the search cost is higher.

Bagwell and Ramey (1993) present a multi-firm model in which advertising

101

offers indirect information as to vertical attributes. In their model, marginal

cost is increasing in quality, and consumers possess downward-sloping demands.

Some consumers prefer high-quality, high-priced goods, while others prefer low-

quality, low-priced goods. Advertising may then provide information that better

enables buyers to match with their respective preferred products. Formally, they

consider a three-stage game. In the first stage, firms choose whether to enter.

If a firm enters, then at the same time if chooses its price, quality level and

advertising activities (i.e., claims and expenditures). In the second stage, each

consumer observes advertising activities, but not price and quality choices, and

picks a single firm to visit. Finally, in the third stage, each consumer observes the

102

price and quality at the selected firm and chooses a purchase quantity.

Advertising claims need not be truthful. A firm that offers one quality of

product may mimic the advertised claims and expenditures of firms that offer the

101 For an early discussion in which advertising plays a matching role in a market with vertically

differentiated products, see Rosen (1978). He proceeds under the assumption that advertised

claims are truthful. For another model in which advertising provides indirect information,

see Anand and Shachar (2004a). They explore a duopoly model in which advertising content

provides direct but noisy information; furthermore, the fact that a firm chooses to target its

ad to particular media channels provides indirect information that the firm’s product may be a

good match for consumers that are exposed to those channels.

102 In this game, the product is a search good. Bagwell and Ramey also analyze the possibility

of an experience good, in which case quality is not observed at the time of purchase.

97

other quality of product. The benefit of misrepresentation is that a firm thereby

“tricks” consumers that prefer the alternative price-quality offering into visiting

its store. But there is also a cost: the misrepresenting firm loses those consumers

that prefer its (true) price-quality offering and to whom it otherwise would have

sold. The net gain from misrepresenation hinges upon the differences in market

share that accrue to firms offering the different qualities. An equilibrium in which

advertising provides information is thus possible only if prices, advertising activ-

ities, and market shares satisfy incentive-compatibility and free-entry conditions.

Fortunately, a sorting condition is available: a quality-sensitive consumer may

yield more profit to a high-quality firm, since the demand expansion that ensues

is sufficient to overwhelm the higher marginal cost. If the market shares are suf-

ficiently similar across qualities, then costless advertising claims (“cheap talk”)

are credible, as under the sorting condition a firm does not gain from trading

consumers that prefer its product for a similar number of consumers that do not.

But if market shares differ sufficiently across qualities, then firms that offer the

low-market-share quality are tempted to misrepresent, and so firms that provide

the high-market-share quality must use dissipative advertising expenditures to

discourage mimicry and signal quality. In a free-entry equilibrium, high market

shares are associated with high fixed costs. Thus, if fixed costs are roughly con-

stant across quality levels, then cheap talk credibly communicates quality. But if

fixed costs vary significantly with quality, then dissipative advertising is used by

firms offering the quality of product that has the higher fixed costs.

6.4. Quality-Guarentee Effect

Up to this point, I have emphasized the extent to which advertising signals prod-

uct quality, when quality is exogenous or determined by a once-and-for-all choice.

In many markets, however, firms offer experience goods and must be given incen-

tive to provide a high-quality good in each period. An intertemporal tradeoff is

suggested. On the one hand, a firm’s short-run incentive is to save costs and offer

unsuspecting consumers a low-quality product. Balanced against this short-term

benefit, however, is the long-run cost of a lost reputation for quality. A firm that

saves costs and provides a low-quality good today foresakes its reputation and

thus the profit that it could earn on repeat sales tomorrow.

Where does advertising fit in? The reputational argument just advanced pre-

sumes that the firm is not anonymous. Clearly, a firm must be identifiable if it is

to be rewarded with repeat business only when it provides high-quality products.

98

In turn, a firm may acquire a “name” by advertising its brand. From this per-

spective, advertising is associated with higher-quality products, since a “known”

firm is reluctant to lose its reputation by offering a shoddy product. Advertising

thus has a quality-guarantee effect that is reassuring even to first-time buyers.

The quality-guarantee effect is emphasized by early writers. Fogg-Meade

(1901), Marshall (1919) and Shaw (1912) all argue that the advent of large-

scale advertising gave manufacturers a significantly greater incentive to provide

high-quality products. As observed in Section 2.2, Braithwaite (1928) takes an

opposing view and argues that the quality-guarantee effect is modest, while Gal-

braith (1958, 1967) and Packard (1957, 1969) go further and suggest that brand

advertising has powerful and negative social consequences. The same debate con-

tinues in the modern era, perhaps with even greater intensity, as the effects of

The Economist

“globalization” are scrutinized. (2001), for example, argues that

a brand name removes the curtain of anonymity and makes a firm accountable for

the quality of its product and the working conditions of its laborers. But Klein

(2001) contends that persuasive (life-style) advertising is an important means by

which brand-name multinationals influence media, shape culture and generally

distort the economic and social aspirations of individuals.

This on-going debate is not resolved here. Accommodating aspects of both

views, I assume that a monopolist’s brand is known to consumers by name and

that advertising is demand-enhancing (perhaps due to its persuasive powers). In

this general context, my goal is to investigate the theoretical underpinnings of the

quality-guarantee effect. Two questions are asked. First, in what manner must

the monopolist distort its price and/or advertising selections, in order to provide a

quality-guarantee effect? Second, among those price-advertising selections that do

guarantee a high-quality product, which selection is preferred by the monopolist?

By answering these questions, I hope to determine whether advertising may play

a quality-guarantee role, even when consumers already know the brand name and

the monopoly can also provide quality assurances with its price.

Formally, I consider an infinitely repeated game. In each period, the monop-

olist chooses a price , an advertising level and a quality level where quality

P A t,

∈ {L,

is either low or high: Consumers observe and but not form a

t H}. P A t,

belief as to the probability that the monopolist has selected a high-quality prod-

b

uct, and then demand a quantity After any consumption experience

D(P, A, b). ≡ − −

is concluded, the monopolist earns profit Π(P, A, b, t) (P c(t))D(P, A, b) κA

and consumers observe the chosen quality Assume that a high-quality prod-

t.

uct involves a higher marginal cost of production (c(H) Departing

> c(L)).

99

from the structure developed above, assume further that a consumer would never

knowingly purchase a low-quality product (D(P, The stage game is

A, 0) = 0).

then repeated. I focus on stationary subgame perfect equilibria. In a stationary

equilibrium, along the equilibrium path, the monopolist makes the same price,

advertising and quality choices in every period. not

To fix ideas, suppose for the moment that the stage game is infinitely

repeated. In a static model, for any given price and advertising expenditure,

if consumers were to form a belief that results in a positive demand, then the

monopolist would surprise consumers with a low-quality product, as it would

thereby save costs without affecting demand. This logic also carries through in

any finite-horizon game. The firm would “cheat” and provide low-quality in the

last period. Using backward induction, it follows that no transaction ever occurs.

In the infinitely repeated game, however, the short-run cost savings that ac-

company a low-quality selection can be balanced against an associated long-run

reputational cost. Suppose that consumers believe that the monopolist will pro-

vide a high-quality product if and only if it has always done so before and the

price and advertising selections fall in a range that guarantees quality. Formally,

quality is guaranteed for a reputable firm when the price and advertising selections

fall in the range for which X

∞ t

− ≤

Π(P, A, 1, L) Π(P, A, 1, H) (6.7)

δ Π(P, A, 1, H),

t=1

where is the discount factor. The left-hand side represents the short-run

δ (0, 1)

cost savings that the monopolist enjoys when it cheats and surprises consumers

with a low-quality selection. The right-hand side captures the long-run reputa-

tional cost that the monopolist incurs, if it cheats in the current period and thus

103

sacrifices its reputation and the prospect of repeat purchases at all later dates.

Recalling the first question raised above, I now characterize the price and

advertising selections that quarantee quality. The incentive constraint captured

in (6.7) may be re-written as follows:

δ {(P − −

− ≤ (6.8)

c(H))D(P, A, 1) κA}.

(c(H) c(L))D(P, A, 1) −

1 δ

103 If the monopolist cheats, then in all future periods play reverts to the Nash equilibrium of

the static game, whereby the monopolist does not advertise and offers a low-quality product

while consumers do not purchase. The monopolist then earns zero profit.

100

Let the interest rate be defined by It is now straightforward to

r δ = 1/(1 + r).

re-write (6.8) as: − − − ≥ (6.9)

[P (c(H) + r(c(H) c(L))]D(P, A, 1) κA 0.

As (6.9) reveals, the monopolist has the incentive to provide a high-quality product

if and only if its price and advertising selections would generate non-negative

profit, under a hypothetical situation in which the firm’s marginal cost is c(H) +

− and consumers believe that product quality is high.

r(c(H) c(L))

An important implication is that the monopolist provides a high-quality prod-

uct only if the price strictly exceeds the true marginal cost, Intuitively, the

c(H).

monopolist will forego the current-period opportunity to cheat consumers only if

profitable repeat business then would be lost in the future. Notice that advertis-

ing is not essential for the quality-guarantee effect (once the name of the product

is known). To see this, put and observe that (6.9) holds if and only if

A = 0

≥ − (6.10)

P c(H) + r(c(H) c(L)).

When price exceeds this critical level, the quality-guarantee effect is achieved

through price alone.

Consider now the second question raised above. Among those price-advertising

selections that guarantee quality, which one maximizes the monopolist’s profit?

The profit-maximizing selection solves the following program:

subject to (6.9). (6.11)

max Π(P, A, 1, H)

P,A

The associated Lagrangian can be expressed as

λ

− − − (6.12)

L(P, A, λ) = (1 + λ){[P (c(H) + r(c(H) c(L)))]D(P, A, 1) κA},

1+ λ

where is the Lagrange multiplier. As the bracketed term in (6.12) reveals, in

λ

the most-profitable stationary equilibrium the monopolist offers a high-quality

product and makes the same price-advertising selection as it would were it to

λ −

produce at constant marginal cost and offer a

c r(c(H) c(L))

c(H) +

1 1+λ

product of known high quality.

The reputation model exhibits a surprising similarity to the static signaling

model of Section 6.1. As in the static model, a cost-increasing distortion is implied:

the high-quality monopolist does best when it sets the same price-advertising pair

101

as it would were its quality of product known but its marginal costs higher (c >

1

In comparison to the complete-information and high-quality monopoly

c(H)).

price and advertising selections, an upward distortion in price and a downward

distortion in demand-enhancing advertising is again predicted. Intuitively, the

upward distortion in price and downward distortion in advertising contribute to a

downward distortion in demand, thereby reducing the short-run cost savings that

104

would be gained if the monopolist were to cheat.

Klein and Leffler (1981) offer an early formalization of some of these themes.

They establish that a competitive firm has incentive to offer a high-quality prod-

uct only if price exceeds marginal cost for the high-quality product (so that repeat

business has value). Their expression for the “quality-assuring price” is analogous

105

to (6.10). They also introduce advertising as an investment in brand-name cap-

ital that is forfeited if a firm degrades its reputation. In light of such advertising

expenses, they argue that the zero-profit requirement of competitive markets may

be reconciled with a positive markup. An implication is that an observed correla-

tion between advertising and profit (see Section 3.2.2) may reflect the rents that

are necessary for high-quality performance rather than the presence of market

power that is brought forth by an advertising-induced barrier to entry.

The reputation model presented above may be modified to illustrate the invest-

ment interpretation of advertising that Klein and Leffler advance. In particular,

consider an equilibrium in which the monopolist does not advertise through time

(i.e., but does advertise at the time of entry. Suppose further that the

A = 0),

initial advertising creates actual brand-name capital, in that it causes a (rep-

A

0

utable) monopolist’s demand to grow through time, where the (constant) rate

of growth increases with the initial advertising outlay. Two implications follow.

First, if the initial advertising outlay is increased, then the monopolist faces a

greater long-term loss from cheating (since a faster-growing consumer demand is

forfeited), and so a lower quality-guaranteeing price can be achieved. Importantly,

advertising that creates brand-name capital may thus represent a means through

which a firm can offer a more competitive price while maintaining its incentive

to offer a high-quality product. Second, if consumers require a sufficient up-front

investment in advertising, it remains possible to reconcile a positive markup with

104 Likewise, if advertising were dissipative, then the high-quality monopolist would guarantee

quality most profitably by not advertising (i.e., setting A = 0).

105 The formal expression in (6.10) is first derived by Shapiro (1983). See also Telser (1980) for

related themes and Stiglitz (1989) for further discussion.

102

106

a zero-profit condition. Interesting future work might expand this framework

to allow for multiple firms that can choose to invest in advertising at any date.

Klein and Leffler also discuss the possibility that consumers may be uninformed

as to firms’ costs. They introduce the provocative idea that a firm’s dissipative

advertising expenditure may signal its cost type and thereby influence consumers’

quality perceptions. Intuitively, a firm benefits if consumers believe it to have low

costs (where a firm has “low costs” if for that firm is low),

c(H) + r(c(H) c(L))

since it can then offer a lower quality-guranteeing price. While the idea is simple

and intuitive, the appropriate formalization is non-trivial, as it involves dynamic

signaling in a rivalrous environment. Rogerson (1986) offers a formal investigation

of this kind. This area, too, represents a promising direction for further work.

Finally, consider the implications of the reputation theory for multiproduct

firms. Suppose that the framework above is extended to allow that the monopo-

list carries two products. If the products do not have the same brand name, then

consumers may be unaware that the products are linked. The quality-guaranteeing

price for each product might then be determined by the product-by-product ap-

plication of the incentive constraint captured in (6.9). Now suppose that the

products have a common brand name. They are then linked in the consumers’

minds, and the monopolist may lose repeat business on both products if it cheats

on either. In the relevant incentive constraint, the product-by-product incentive

constraints are pooled (i.e., added together): the gains from cheating on both

products must be no greater than the loss in profits on both products that cheat-

ing would imply. It is possible that the monopolist can be induced to supply

high-quality products, even when the price-advertising selection for one product

would fail the incentive constraint for that product alone. Branding may thus

benefit a firm by expanding the set of quality-guaranteeing price-advertising se-

107

lections. This discussion reinforces the argument that advertising can motivate

106 t

) d(P, 1), where g(A ) − 1 is the

Formally, suppose that demand at time t is given as g(A

0 0

demand growth rate, P is the stationary price selection and b = 1 is the belief. Suppose that

0

g(0) = 1, g > 0, and δg(A ) is bound below unity. Then δ in (6.8) is replaced by the effective

0

(growth-included) discount factor, δ ). The quality-guaranteeing price is again given

≡ δg(A

e 0

by (6.10), when r is replaced by the effective interest rate, r )]/g(A ). Observe

≡ [1 + r − g(A

e 0 0

that sign{r − r ) − 1} > 0 for A > 0. Finally, the monopolist earns discounted

} ≡ sign{g(A

e 0 0

profit in amount Π(P, 0, 1, H)/(1 − δ ) − κA , which is driven to zero at a finite and positive

e 0

level for A .

0

107 For further formal analyses of branding, see Bagwell (1992), Cabral (2000), Choi (1998),

Montgomery and Wernerfelt (1992) and Wernerfelt (1988). The general idea that incentive

constraints are relaxed when pooled is also exploited in the collusion literature. See Telser

103

high-quality choices by reducing anonymity and making brand names known.

6.5. Summary

A huge theoretical literature analyzes the relationship between advertising and

product quality. The relationship is subtle, and it varies across circumstances.

One set of work analyzes the manner in which advertising may signal quality. In

this context, the advertising-quality relationship can be understood with reference

to the three effects that Nelson (1974b) identifies. These effects provide a basis for

a positive relationship between advertising and product quality. But the signaling-

efficiency effect may be reversed in a number of environments, since higher-quality

goods may use more expensive materials and thus have higher marginal costs. In

such environments, greater advertising may be associated with higher quality

if Nelson’s other effects are prominent. The main empirical implication is that

no systematic correlation between advertising and quality is expected, since the

relationship reflects market circumstances and the simultaneous use of price and

advertising as signals of quality. This implication is consistent with the empirical

work summarized in Section 3.2.5. It also motivates new empirical work (as

discussed in Section 8) that considers price and advertising as joint signals.

A second set of work investigates the extent to which advertising may pro-

vide an incentive for the continued selection of high-quality products. As early

writers argue, advertising can play an important role by making brands known

and identifiable, so that brand reputations can be forged and maintained. Once

brands are known, however, if advertising enhances current demand, then the

quality-guarantee effect is most profitably generated when price is distorted up

and advertising is distorted down. If advertising also creates brand-name capital

by enhancing future demand, then it appears possible that a firm may distort its

advertising upward, in order to be able to offer a lower quality-guaranteeing price.

7. Advertising and Entry Deterrence

With a few exceptions, the theory summarized above does not address the rela-

tionship between advertising and entry. This is an important ommision, since the

persuasive view hypothesizes that advertising exerts an entry-deterrence effect. As

discussed in Section 3, the empirical support for this hypothesis is mixed. In the

(1980) for a first formalization. Further analysis is offered by Bernheim and Whinston (1990).

104

absence of an empirical resolution concerning the relationship between advertising

and entry, theoretical analyses may be of special value.

In the tradition of Bain’s (1949) limit-pricing model, many of the first mod-

els of advertising as an entry barrier employ the assumption that the incumbent

can credibly commit to maintain its pre-entry advertising expenditures if entry

108

occurs. A high pre-entry advertising expenditure may then imply a hostile

environment for a potential entrant, in which case such advertising may deter en-

try. But the credibility of this commitment is questionable. As Needham (1976)

argues, in the absence of such a commitment, an incumbent’s pre-entry adver-

tising influences the entry decision only if there is some link between pre-entry

109

advertising and the entrant’s post-entry expected profit.

I consider here two possible links. First, advertising may have a goodwill

effect, so that some consumers favor the incumbent in the post-entry period when

the incumbent advertises heavily in the pre-entry period. While the empirical

studies reviewed in Section 3.1.1 suggest that the goodwill effect of advertising

is often modest, the effect may be pronounced in certain industries. Second, the

incumbent’s pre-entry advertising behavior may signal the incumbent’s private

information and thereby affect the entrant’s expected profit from entry.

7.1. Advertising and Goodwill

If advertising generates goodwill for the incumbent, then it is natural to expect

that an incumbent could deter entry by engaging in heavy pre-entry advertis-

ing. But is this expectation confirmed in an equilibrium model? To answer this

question, the source of the goodwill effect must be specified. An “informational

goodwill effect” is present, if an incumbent’s pre-entry advertising provides con-

sumers with hard information of durable value concerning the incumbent’s exis-

tence and prices. The incumbent might include its location and phone number

on the ads, for instance. Alternatively, as persuasive-view advocates emphasize

(see Section 2.2), the incumbent’s pre-entry advertising generates a “reputational

goodwill effect,” if in some general sense it reinforces consumers’ past experiences

so as to differentially reward an established firm. For example, the incumbent’s

advertising efforts may reinforce any reputation that it has for providing reliable

and high-quality products.

In Schmalensee’s (1983) model, an informational goodwill effect is posited. He

108 See, for example, Salop (1979), Spence (1980) and Williamson (1963).

109 See also Cubbin (1981). 105

considers a homogenous-products market served by an incumbent and potentially

an entrant. Consumers learn of a firm’s existence and price through an adver-

tising technology of the kind proposed by Butters (1977). The three-stage game

proceeds as follows. In the first (pre-entry) stage, the incumbent sends out ads to

consumers. A consumer who receives such an ad is informed of the incumbent’s ex-

istence and can learn the incumbent’s (eventual) price at zero cost. In the second

stage, after observing the incumbent’s advertising behavior, the entrant considers

whether to incur a sunk cost and enter. If entry occurs, then the entrant sends out

its own ads. In this event, each firm then has a set of captive consumers, and there

is also a set of selective consumers. Finally, in the third (post-entry) stage, active

firms play some simultaneous-move oligopoly game. As Schmalensee observes, if

entry occurs and firms choose prices, then a pure-strategy Nash equilibrium does

not exist. He thus supposes that the firms compete in quantities.

In this model, advertising is a durable investment and entry entails a sunk cost.

It is thus tempting to reason by analogy with Dixit’s (1980) entry-deterrence

model and conclude that the incumbent strategically overinvests in advertising

in order to deter entry. But this analogy is false. As Schmalensee shows, the

reduced

incumbent can deter entry, but it does so with a advertising expenditure.

Intuitively, if the incumbent were to advertise heavily, then it would have many

captive consumers. The incumbent would then be tempted to set a low output,

so as to sell only to these consumers at a high price. A rational entrant would

thus perceive that the incumbent would be a “soft” competitor. Consequently, if

the incumbent seeks to deter entry, it should underinvest in advertising, thereby

ensuring that it has few captive consumers and would respond to entry with

vigorous competition for selective consumers.

Ishigaki (2000) modifies this three-stage game to allow that post-entry compe-

tition occurs in prices. After characterizing the mixed-strategy pricing equilibria

that entry induces, Ishigaki finds that the entry is either blockaded (the incumbent

deters entry when it behaves as it would were there no entrant) or accommodated

(the incumbent optimally allows entry and sets its Stackelberg advertising level).

There is no parameter region for which the incumbent strategically distorts its ad-

vertising choice in order to deter entry. Together, the models of Schmalensee and

Ishigaki suggest the following striking conclusion: in homogeneous-products mar-

kets, when the goodwill effect of advertising is informational, a profit-maximizing

incumbent does not deter entry by investing more in advertising than it would

were there no entry threat. These models therefore provide no formal support for

the entry-deterrence effect. 106

As Fudenberg and Tirole (1984) establish, a similar conclusion may obtain

when the incumbent and entrant sell differentiated products. They consider a

simple two-period model that captures some of the central themes raised above.

In the first (pre-entry) period, the incumbent (firm 1) chooses a fraction of

Φ

1

consumers to inform of its existence and price. As in the Grossman-Shapiro (1984)

model, the cost to the incumbent of informing a fraction where

is

Φ ), A(Φ )

A(Φ

1 1 1

Assume further that initially it is

is positive, increasing and convex for > 0.

Φ

1 ∞.

prohibitively costly to reach all consumers: The informational goodwill

A(1) =

effect is captured with a strong assumption: consumers who receive an ad in the

first period do not bother to read any ads that they may receive in the second

(post-entry) period, and they thus remain captive consumers for the incumbent

throughout the game. The incumbent selects its monopoly price in the pre-entry

period and achieves a net revenue of per consumer. The incumbent’s

R > 0

M

pre-entry profit is thus Φ R ).

A(Φ

1 1

M

In the second period, the incumbent and the entrant (firm 2) make advertising

and pricing selections. Under the goodwill assumption, consumers remain

1 Φ

1

in the second period that are not captive to the incumbent. Fudenberg and Tirole

assume that the firms advertise so as to cover the remaining market. Let denote

A

the second-period advertising expenditure incurred by each firm in the course of

creating selective consumers. The second-period prices of the incumbent

1 Φ

1

and entrant, respectively, are denoted as and . In the second period, the

P P

1 2 from a selective con-

incumbent enjoys per-customer net revenues of (P , P )

R

1 1 2

∞)

sumer and from a captive consumer. The entrant sells only to selective

R (P ,

1 1

consumers and enjoys a per-customer net revenue of Assume that the

R (P , P ).

2 1 2

net revenue functions are differentiable, concave in own prices, increasing in rival

2

∂ R

prices and characterized by positive cross-partials (i.e., for i = 1, 2, > 0).

i

∂P P

1 2

The final assumption indicates that prices are strategic complements.

For this two-period game, payoff functions are defined as follows:

∞)+(1−Φ

−A(Φ (7.1)

Π (Φ , P , P ) = [Φ R )]+δ[Φ R (P , )R (P , P )−A]

1 1 1 2 1 1 1 1 1 1 1 1 2

M − − (7.2)

Π (Φ , P , P ) = δ[(1 Φ )R (P , P ) A],

2 1 1 2 1 2 1 2

where is the common discount factor. Assume a Nash equilibrium

δ (0, 1)

∗ ∗ for the second-stage subgame exists and satisfies the first-order condi-

(P , P )

1 2

tions: ∞)

∂Π (Φ , P , P ) ∂R (P , (P , P )

∂R

1 1 1 2 1 1 1 1 2

− (7.3)

= δ[Φ + (1 Φ ) ]=0

1 1

∂P ∂P ∂P

1 1 1

107 ∂R

∂Π (Φ , P , P ) (P , P )

2 1 1 2 2 1 2

− (7.4)

= δ(1 Φ ) = 0.

1

∂P ∂P

2 2

Throughout, the dependence of on is suppressed.

P Φ

1

i

Given that prices are strategic complements, (7.3) implies that

∗ ∗ ∗

∞)

∂R ∂R

(P , (P , P )

1 1

1 1 2 (7.5)

> 0 > .

∂P ∂P

1 1

In the second period, the incumbent thus would like to raise its price on captive

consumers and lower the price that it offers to selective consumers. The incumbent

thus picks a second-period price that optimally balances these considerations.

The following relationships are now direct from (7.3), (7.4) and (7.5):

2 ∗ ∗

∂ Π (Φ , P , P )

1 1 1 2 (7.6)

> 0,

∂P ∂Φ

1 1

2 ∗ ∗

∂ Π (Φ , P , P )

2 1 1 2 (7.7)

= 0,

∂P ∂Φ

2 1

2 ∗ ∗

Π (Φ , P , P )

∂ 1 1 1 2 (7.8)

> 0,

∂P ∂P

1 2

2 ∗ ∗

Π (Φ , P , P )

∂ 2 1 1 2 (7.9)

> 0.

∂P ∂P

2 1

According to (7.6), and as in Schmalensee’s model, when the incumbent’s pre-

entry advertising is greater, it becomes more attracted to higher post-entry prices.

In the formulation considered here, as (7.7) confirms, the incumbent’s pre-entry

advertising does not directly alter the entrant’s preferred price. But, as (7.9)

indicates, if greater pre-entry advertising leads the incumbent to price higher,

then the entrant becomes attracted to higher prices for this reason.

Under a standard stability condition, it is now easy to confirm that (7.6)-(7.9)

yield the anticipated conclusion:

∗ ∗

∂P

∂P

1 2 110

and (7.10)

> 0 > 0.

∂Φ ∂Φ

1 1

110 With P on the y-axis, the stability condition indicates that the second-period pricing

1

reaction function of the incumbent is flatter than that of the entrant.

108

Thus, as the incumbent advertises more heavily in the pre-entry period, a greater

number of captive consumers are created, and so the incumbent prices higher in

the post-entry period. Given that prices are strategic complements, the entrant

prices higher as well.

Consider now the advertising level at which the incumbent accommodates the

entrant in the most profitable manner. Assuming that the second-order condition

is satisfied, the incumbent maximizes its payoff when it chooses in the pre-entry

∗ ∗

(Φ )

dΠ ,P ,P

∗ 1 1

period the value Using

that satisfies the first-order condition

Φ = 0.

1 2

1 dΦ

1

(7.3), this condition may be re-stated as ∗ ∗ ∗

(P , P ) ∂P

∂R

1 1 2 2

∗ ∗ ∗ 0

∞) − − (7.11)

R + δ[R (P , R (P , P )] + δ(1 Φ ) = A (Φ ).

1 1 1 1

M 1 1 2 ∂P ∂Φ

2 1

On the left-hand side of (7.11), the first two terms are positive and capture the

direct effect of greater pre-entry advertising on first- and second-period net rev-

enue. The third term is also positive. This term represents the strategic effect

of greater pre-entry advertising. As established in (7.10), when the incumbent

advertises more heavily, the entrant prices higher. The incumbent thereby earns

greater profit in the post-entry period. Finally, the term on the right-hand side

captures the cost of additional advertising.

It is interesting to compare with the value that would occur if the entrant’s

Φ

1

post-entry price were unresponsive to the incumbent’s pre-entry advertising. In

the absence of the strategic effect, the left-hand side would be smaller. Given the

convexity of the function it follows that the optimal value for would

A(Φ ), Φ

1 1

∗ It thus may be concluded that the incumbent overinvests in

then fall below .

Φ

1

pre-entry advertising, in order to create a larger captive group of consumers and

thereby commit itself to a higher post-entry price, so that the entrant will respond

with a higher price of its own. As Fudenberg and Tirole put it, the incumbent

111

best accommodates the entrant by overinvesting so as to become a “fat cat.”

111 Boyer and Moreaux (1999) consider a different demand specification, under which the en-

trant also has captive consumers. In their formulation, the incumbent’s advertising level exerts

entrant’s

a strategic effect through its impact on the price reaction curve. For example, when

the incumbent and entrant sell substitute products and prices are strategic complements, if the

incumbent advertises more heavily, then the entrant has a smaller set of captive consumers,

and so the entrant’s price reaction curve shifts downward. In contrast to Fudenberg and Tirole,

Boyer and Moreaux argue that the incumbent best accommodates entry by underinvesting in

advertising. Furthermore, this finding holds for a variety of sequential-move games and whether

the products are substitutes or complements. See Fershtman and Muller (1993) for an earlier

discussion of the underinvestment finding when products are substitutes.

109

Suppose now that the entrant must incur a sunk cost if it chooses to enter.

Rather than accommodate the entrant, the incumbent might then choose to deter

entry. But how is this achieved? Intuitively, if the incumbent seeks to deter entry,

then it may achieve an indirect benefit by underinvesting in advertising, so as to

create a small group of captive consumers and thereby commit itself to a low price

in the event of entry. Fudenberg and Tirole refer to this as the “lean-and-hungry

look.” But an argument also can be made that the incumbent should overinvest

in advertising, since it thereby achieves the direct benefit of reducing the entrant’s

possible market. To see these competing effects more clearly, use (7.4) and note

that the overall effect of pre-entry incumbent advertising on post-entry profit to

the entrant is given as follows:

∗ ∗ ∗ ∗ ∗

dΠ (Φ , P , P ) (P , P ) ∂P

∂R

2 1 2

1 2 1 2 1 ∗ ∗

− − (7.12)

= δ[(1 Φ ) (P , P )].

R

1 2 1 2

dΦ ∂P ∂Φ

1 1 1

The first term is positive under (7.10) and captures the indirect benefit to the

incumbent of reduced pre-entry advertising, but the second term is negative and

reflects the direct benefit to the incumbent of increased pre-entry advertising. As

Fudenberg and Tirole observe, in an important set of environments, the indirect

benefit of reduced pre-entry advertising dominates, and entry deterrence again

112

requires underinvestment in advertising.

The models developed above, however, all posit an informational goodwill

effect. What if instead advertising induces a reputational goodwill effect? To

113

begin, it is useful to distinguish between two issues. A first issue is whether

an incumbent with an existing reputation for reliable and high-quality products

has an advantage relative to an entrant with no existing reputation. Undeniably,

this is often the case. Consumers are naturally willing to pay a premium for a

product from a reputable firm relative to that which they would pay for a product

from an unknown firm. This suggests that informational product differentiation

may be a barrier to entry. As Bagwell (1990), de Bijl (1997), Farrell (1986) and

Schmalensee (1982) demonstrate, this suggestion is readily confirmed in formal

114

models. A second issue concerns the extent to which advertising is the source of

this entry barrier. With respect to this issue, it is noteworthy that the incumbent

112 In recent work, Doraszelski and Markovich (2004) use numerical methods to compute the

Markov-perfect equilibria for a dynamic game with an informational goodwill effect. In their

overinvesting

model, an incumbent may deter entry by in advertising, while the optimal accom-

modation strategy can vary with market characteristics.

113 This distinction is explored in Section 2.2. It is also emphasized by Demsetz (1982).

114 Schmalensee (1982) offers a first formalization of the brand loyalty that consumers exhibit

110

is not allowed to advertise in the formal entry-deterrence models just mentioned.

In these models at least, there is clearly no formal sense in which advertising is

necessary for informational product differentiation to act as an entry barrier.

But might advertising somehow reinforce consumers’ past experiences with the

established product and thereby exacerbate the informational barrier to entry? It

is, of course, possible to assume that advertising is more effective when consumers

115

have greater experience with the advertised product. But a more compelling

model would yield the reinforcement effect as an implication of optimizing behav-

ior. I am not aware of a model of this kind.

Drawing on the earlier writings, let me highlight one approach that may war-

rant formalization. As Braithwaite (1928, p. 32), Marshall (1919, p. 307) and

Comanor and Wilson (1974, Chapter 4) explain, a firm’s ad must bid for the

consumer’s attention, and it may be more costly for a new firm to get the con-

sumer’s attention when the consumer is already overloaded with related ads from

established firms. As Comanor and Wilson (1974, p. 47) put it:

“To the extent that the advertising of others creates ‘noise’ in the mar-

ket, one must ‘shout’ louder to be heard, so that the effectiveness of

each advertising message declines as the aggregate volume of industry

advertising increases. In this case, it will be necessary for new entrants

to spend more today to gain an established market position than ex-

isting firms spent yesterday, when aggregate industry advertising was

probably far less. From these circumstances also, new entrants may

have differentially higher advertising costs than did established firms

at their entry into the market.”

This “noise effect” suggests that the incumbent may strategically overinvest in

toward pioneering brands of known quality. He shows that a high-quality incumbent can earn

positive profit without inducing the entry of an equally efficient, high-quality entrant. Bagwell

(1990) extends this model to allow that consumers rationally infer quality from price and that

the entrant may offer a superior product. The key finding is that a low-quality incumbent

may deter entry, even when the entrant actually offers a high-quality (and socially efficient)

product. de Bijl (1997) establishes a similar finding for search goods when search costs are

high, and shows as well that the entry barrier may be diminished if the incumbent is informed

of the entrant’s quality. Farrell (1986) extends the analysis to consider the manner in which

the incumbent’s behavior affects the entrant’s incentive to choose a high-quality product. See

also Schmalensee (1979) for a detailed discussion of the product-differentiation advantages that

accrued to pioneering brands in the market for lemon juice.

115 See Comanor and Wilson (1974, Chapters 3 and 4) for a model in which it is assumed that

advertising’s effectiveness varies with consumers’ experience with the advertised product.

111

advertising, in order to jam the message space and force the entrant to be more

stentorian with its advertising efforts. Incumbent advertising would then raise

the entrant’s advertising costs and exacerbate the entry barrier. This cost-raising

strategy is informally discussed by Hilke and Nelson (1984), who provide evi-

dence in the U.S. coffee market that Maxwell House used such a strategy when

facing entry by Folgers. Future work might revisit this noise effect, in a model

that endogenizes the manner in which consumers with finite information-storage

capabilities manage (as possible) their exposure to advertising.

7.2. Advertising and Signaling

I consider now the possibility that the incumbent’s pre-entry behavior may signal

its private information. This information may be relevant for the entrant’s calcula-

tion of the expected profit from entry. In this case, an informational link connects

the incumbent’s pre-entry behavior and the entrant’s post-entry expected profit.

Milgrom and Roberts (1982) establish that a low-cost incumbent may distort

its pre-entry price downward in order to signal its costs and thereby deter entry.

Bagwell and Ramey (1988) extend the Milgrom-Roberts analysis to allow that the

low-cost incumbent may signal its costs by distorting its pre-entry price and/or

advertising. As I explain below, they find that the low-cost incumbent deters entry

most profitably, when its pre-entry price is distorted downward and its demand-

116

enhancing advertising is distorted upward. They thus provide a theory in which

117

an incumbent overinvests in advertising in order to deter entry.

Bagwell and Ramey consider a signaling game with two periods. In the pre-

∈ {L, ≥

entry period, an incumbent of cost type selects its pre-entry price

t H} P

≥ ≡

and advertising level The incumbent earns pre-entry profit

0 A 0. Π(P, A, t)

−c(t))D(P, ≥

where and Advertis-

(P A)−κA, D > 0, D < 0, D 0 c(H) > c(L).

P A ∈ {L,

or dissipative (D For

ing may be demand-enhancing (D > 0) = 0). t H},

A A

assume further that is strictly concave in , with a unique maximizing

Π(P, A, t) P

116 Methodologically, the analysis presented below is closely related to that presented above in

Section 6.1. Notice, though, that the incumbent now uses price and advertising to signal its

cost to an entrant, whereas in Section 6.1 the monopolist uses price and advertising to signal

quality to consumers.

117 Milgrom and Roberts offer an information-theoretic foundation for Bain’s (1949) prediction

that an incumbent can deter entry by limit pricing. Likewise, Bagwell and Ramey provide a the-

oretical counterpart to an interesting extension of Bain’s approach that is offered by Williamson

(1963). In Williamson’s model, the incumbent deters entry, by making a pre-entry commitment

to a low price and a high level of advertising. 112

This pair denotes the monopoly price-advertising selection.

pair, (t), A (t)).

(P

M M ≡

The corresponding monopoly profit is (t) Π(P (t), A (t), t).

π M M M

At the start of the post-entry period, a single entrant observes the pre-entry

price and advertising level, but not the incumbent’s type, and forms some belief

∈ as to the probability that the incumbent has high costs. The

b = b(P, A) [0, 1]

entrant then enters or not, where (E denotes (no) entry. If entry

E = 1 = 0)

does not occur, then the incumbent earns monopoly profit in the post-entry

π (t)

M

period. If entry does occur, then the entrant learns the incumbent’s type, and the

incumbent and entrant play some post-entry duopoly game, earning and

π (t)

D

eD eD

respectively. The sunk cost of entry is included in

π (t), π (t).

For a given price , advertising level entry decision and incumbent type

P A, E

the incumbent and entrant payoffs are

t, − (7.13)

(t) + (1 E)π (t)]

V (P, A, E, t) = Π(P, A, t) + δ[Eπ D M

eD (7.14)

u(E, t) = δEπ (t),

respectively, where is the common discount factor. At the time of the entry

δ

decision, the entrant’s expected profit from entry is

≡ − (7.15)

U (P, A, E, b) [bu(E, H) + (1 b)u(E, L)]/δ.

Using (7.14) and (7.15), it follows that and

U(P, A, 0, b) = 0 U(P, A, 1, b) =

eD eD

bπ (H) + (1 b)π (L).

Further structure is provided by three key assumptions. First, whatever its

type, the incumbent prefers that entry not occur: Second, the

π (t) > π (t).

M D

entrant earns positive profit from entry if and only if the incumbent has high

eD eD

costs: Third, the incumbent gains at least as much from

π (H) > 0 > π (L). ≥

entry deterrence when its costs are low as when its costs are high: π (L)−π (L)

M D

− The first assumption is unobjectionable, the second assumption

π (H) π (H).

M D

holds in standard duopoly models if the sunk cost of entry falls in an intermediate

range, and the third assumption reflects the differential benefit of greater sales to

a lower-cost firm and holds in many popular duopoly models.

{P

A Perfect Bayesian Equilibrium is a set of strategies (t), A(t), E(P, A)}

t=L,H

∈ {L,

and beliefs such that: (i). for each maximizes

b(P, A) t H}, (P (t), A(t))

(ii). for all maximizes

V (P, A, E(P, A), t), (P, A) 0, E(P, A) U(P, A, E, b(P, A))

and (iii). is derived from the equilibrium strategies whenever possible.

b(P, A) 6

I again focus on separating equilibria. For such equilibria, (P (H), A(H)) =

113

and thus The entrant

(P (L), A(L)) b(P (H), A(H)) = 1 > 0 = b(P (L), A(L)).

infers the incumbent’s type and enters if and only if the incumbent has high costs.

As the high-cost incumbent is “found out,” it can do no better than to make its

118

monopoly selection, and then face entry.

(P (H), A(H)) = (P (H), A (H)),

M M ≡

The high-cost incumbent thus receives the payoff (H), A (H), 1, H)

V (P

M M

Separation then requires that the low-cost incumbent choose some pair

V (H).

M that the high-cost incumbent would not mimic:

(P, A) ≤ (7.16)

V (P, A, 0, H) V (H).

M

To ensure that separation is costly, I assume that the low-cost incumbent’s monopoly

does not satisfy (7.16).

selection (L), A (L))

(P

M M

In the least-cost separating equilibrium, the low-cost incumbent makes the

∗ ∗ ∗ ∗

selection where is the price-advertising selection

(P (L), A(L)) = (P , A ), (P , A )

that solves the following program: subject to (7.16). (7.17)

Max V (P, A, 0, L)

P,A

Bagwell and Ramey (1988) establish that the least-cost separating equilibrium

exists. I focus here on the characterization of such an equilibrium.

To gain intuition, consider any two price-advertising pairs, and

(P , A ) (P , A ),

1 1 2 2

that leave the mimicking high-cost incumbent indifferent. Since each pair deters

Thus

entry, indifference means that , A , H) = Π(P , A , H).

Π(P

1 1 2 2

− (7.18)

Π(P , A , L) Π(P , A , L)

2 2 1 1

− − −

= [Π(P , A , L) Π(P , A , L)] [Π(P , A , H) Π(P , A , H)]

2 2 1 1 2 2 1 1

− , A ) D(P , A )].

= [c(H) c(L)][D(P

2 2 1 1

As (7.18) reveals, given that the low-cost incumbent prefers the

c(H) > c(L),

pair at which demand is highest. Intuitively, a demand-increasing change is more

attractive to a low-cost incumbent, since the demand increase then translates into

a smaller cost increase.

118 (H), A (H)). Then the high-cost incumbent could achieve a

Suppose (P (H), A(H)) 6 = (P

M M

strict gain with a deviation to (P (H), A (H)), since

M M

V (P (H), A(H), 1, H) < V (P (H), A (H), 1, H) ≤ V (P (H), A (H), E(·), H),

M M M M

E(·) ≡ E(P (H), A (H)). π (H) > π (H).

where The final inequality follows, since

M M M D

114

For further insight, consider the program given in (7.17). The Lagrangian is

≡ It may be verified that

(H) V (P, A, 0, H)].

L(P, A, λ) V (P, A, 0, L) + λ[V

M

119

∈ at the optimum. Using (7.13), the Lagrangian may be re-written as

λ (0, 1) −

c(L) λc(H) −

− − (7.19)

)D(P, A) κA} + K(λ, δ),

L(P, A, λ) = (1 λ){(P −

1 λ

where is independent of and As the bracketed term in (7.19) reveals,

K(λ, δ) P A.

in the least-cost separating equilibrium, the low-cost incumbent makes the same

price-advertising selection as it would were its constant marginal cost known to

≡ − − ∈

be Observe that implies

c < c(L).

[c(L) λc(H)]/(1 λ). λ (0, 1) c

2 2

In the least-cost separating equilibrium, the low-cost incumbent thus under-

takes a “cost-reducing distortion,” in that it selects the same price-advertising pair

as it would were its costs known and lower than they truly are. Put differently, the

low-cost incumbent behaves as it would were it a monopolist operating in a single-

period setting with constant marginal cost It is natural to assume that

c < c(L).

2

a single-period monopolist would lower its price and raise its demand-enhancing

120

advertising were its constant marginal cost of production reduced. Under this

assumption, the low-cost incumbent distorts downward its price (P < P (L))

M

and upward its demand-enhancing advertising (A Intuitively, the

> A (L)).

M

low-cost incumbent undertakes these distortions in order to demonstrate its will-

ingness to increase demand. Finally, dissipative advertising is not used as a signal,

since it would never be used by an incumbent with known costs.

As in the Milgrom-Roberts model, profitable entry is not deterred in a sepa-

rating equilibrium. The entrant infers the incumbent’s cost type and resists entry

exactly when entry would be unprofitable (i.e., when the incumbent has low costs).

The incumbent’s pre-entry behavior credibly reveals its cost type, however, only

when the low-cost incumbent distorts its pre-entry selection. In the least-cost sep-

arating equilibrium, the low-cost incumbent deters (unprofitable) entry by limit

pricing and overinvesting in demand-enhancing advertising.

With these predictions at hand, it is interesting to revisit the relationships

between advertising, profitability and entry (see Sections 3.2.2 and 3.2.3). The

Bagwell-Ramey model predicts that greater incumbent advertising is associated

with higher profitability and lower rates of entry. These predictions match closely

those suggested by Comanor and Wilson (1967, 1974) and other persuasive-view

119 The proof is analogous to that given in footnote 91 for the product-quality signaling model.

120 As observed in footnote 92, this assumption holds in each of the two examples discussed in

Section 4.1.2. 115

advocates. The predictions, however, are not attributable to advertising-induced

brand loyalty; instead, they arise because an efficient incumbent advertises more,

earns more and faces less entry than would an inefficient incumbent. The Bagwell-

Ramey model thus offers some support for the “superior efficiency” interpretation

advanced by Demsetz (1973, 1974) and Nelson (1974b, 1975).

The basic model can be extended in several directions. Bagwell and Ramey

(1990) suppose that the incumbent’s private information concerns the level of in-

dustry demand. The incumbent now deters entry by signaling that demand is

low. They establish that a “demand-reducing distortion” occurs: the low-demand

incumbent behaves as if it were a single-period monopolist but demand is lower

than it truly is. Under natural assumptions, in the least-cost separating equilib-

rium, the low-demand incumbent’s price and demand-enhancing advertising are

both distorted downward. Thus, entry deterrence entails limit pricing, whether

the incumbent is privately informed of its costs or the level of industry demand;

however, entry deterrence results in an underinvestment in demand-enhancing

advertising when the incumbent is privately informed as to the level of industry

demand. Bagwell and Ramey also consider the possibility that the incumbent

may wish to signal that demand is high, so as to influence the entrant’s beliefs

and accommodate entry in the most profitable manner possible. In this case, a

“demand-increasing distortion” occurs, with the implication that the high-demand

121

incumbent distorts upward both price and demand-enhancing advertising.

The work described here also can be extended to analyze manufacturer-retailer

relations. Suppose that a manufacturer has private information concerning the

eventual demand for its new product. The retailer may wish to carry the manu-

facturer’s product only if the retailer believes that there is a high demand for this

product. The manufacturer thus may wish to use its advertising expenditure and

(wholesale) price to signal to the retailer that demand is high. Following the logic

just described, the manufacturer signals that it offers a high-demand product by

engaging in a demand-increasing distortion, whereby it distorts upward both price

122

and demand-enhancing advertising. This discussion provides a formal counter-

121 Bagwell and Ramey also provide propositions that characterize necessary features of re-

fined pooling equilibria. But Albaek and Overgaard (1992a) show that, in fact, refined pooling

equilibria fail to exist in this model.

122 See Chu (1992) for a formalization of this extension, wherein the retailer learns demand if

it decides to carry the incumbent’s product. Albaek and Overgaard (1992b) suppose that the

retailer carries the product but does not learn demand prior to setting the retail price. The

retailer’s beliefs then impact its price choice. Now, a manufacturer may undertake a demand-

reducing distortion in order to signal that demand is low, as it thereby enourages the retailer to

116

part to a common argument expressed in earlier writings that a manufacturer uses

heavy advertising to communicate its confidence in the new product to retailers.

See, for example, Berreman (1943) and Chamberlin (1933, p. 121).

Finally, Linnemer (1998) offers an interesting extension, in which an incumbent

and

firm has private information with respect to its product quality production

costs. He considers a two-period model, in which the incumbent’s first-period

price and dissipative advertising outlays are used by consumers to infer quality

and by a potential entrant to infer costs. Specifically, the incumbent either has

a low-quality product and low costs, a high-quality product and medium costs,

or a high-quality product and high costs. Consumers know product quality in

the second period, but they must infer it in the first period. The entrant knows

the incumbent’s product quality; however, the entrant does not know whether a

high-quality incumbent has medium or high costs. The entrant wants to enter,

unless the incumbent has a high-quality product that it produces at medium cost.

The interesting point is that the high-quality incumbent with medium costs has a

conflict. As in the static signaling model presented in Section 6.1, it is tempted to

distort price upward in order to signal quality to consumers. But, as in the limit-

pricing literature discussed just above, it is also tempted to distort price downward

in order to signal that its costs are not high and thus that entry would not be

profitable. In rough analogy with the Milgrom-Roberts (1986) model, given these

conflicting considerations, the high-quality medium-cost incumbent may have no

better option than to use a distorted price and a positive dissipative advertising

expenditure when signaling its type. As in the Bagwell-Ramey (1988) model, the

consequent overinvestment in advertising deters entry that is unprofitable.

7.3. Summary

In summary, when the goodwill effect of advertising is informational, the the-

oretical literature emphasizes that an incumbent firm that seeks to deter entry

may underinvest in pre-entry advertising. It is also possible that advertising gen-

erates a reputational goodwill effect, by reinforcing consumers’ experiences with

the established product and exacerbating informational product differentiation.

I am not aware, however, of an equilibrium model of this kind. On the whole,

the entry-deterrence effect of advertising is not strongly supported by the existing

theoretical models that emphasize advertising’s possible goodwill effects. Future

work might endogenize the “noise effect” that is emphasized in earlier writings.

set a low price and hence mitigates the double-marginalization problem.

117

I also consider the possibility that the incumbent’s pre-entry pricing and adver-

tising behavior may signal its private information and thereby affect the entrant’s

expected profit from entry. When the incumbent has private information about its

costs, a low-cost incumbent may limit price and overinvest in advertising, in order

to signal its costs and thereby deter entry. On the other hand, if the incumbent

has private information as to the level of industry demand, a low-demand incum-

bent may limit price and underinvest in advertising, in order to signal demand

and thereby deter entry. The overinvestment finding provides some support for

the entry-deterrence effect of advertising; however, it must be noted that entry is

deterred only when it is intrinsically unprofitable. In other words, the low-cost

incumbent’s heavy advertising does not make entry unprofitable; rather, it reveals

that entry would be unprofitable.

8. Empirical Analyses

While inter-industry studies offer useful descriptions of economy-wide empirical

regularities, they often suffer from important endogeneity and measurement con-

cerns (as detailed in Section 3) and ultimately fail to identify the underlying

structural parameters that describe how individual markets work. As mentioned

in the Introduction, the modern (second-group) empirical analyses of advertising

increasingly use new data sets, which are often constructed at remarkably dis-

aggregated levels, and emphasize consumer and firm conduct. Strategic theories

of advertising (as reviewed in Sections 4-7) influence the specification of demand

functions and supply relationships in these analyses. In this section, I offer a brief

and non-technical review of this empirical literature.

8.1. Advertising and the Household

I begin with a group of empirical studies that examine the impact of advertising

on brand purchase decisions. The studies utilize household brand purchase panel

data and often household advertising exposure data. With such disaggregated

data, it is possible to gain insight into the respective roles of advertising and

experience in explaining household brand purchase behavior. Likewise, it is possi-

ble to better distinguish between the informative, persuasive and complementary

effects of advertising.

How are such data obtained and analyzed? One approach is to use a controlled

field experiment. In this way, Krishnamurthi and Raj (1985) examine the effect of

118

an increase in advertising on the elasticity of demand for an unnamed frequently

purchased brand. Household brand purchase data are obtained through panel

diaries maintained by households in a test city over a 52-week pre-test period and

a 24-week test period. Advertising exposure is controlled through a split-cable

TV experiment: a test panel of families is connected to one cable while a control

panel of families in connected to another cable, and then the level of (non-price)

advertising for the brand is increased in the test period for the test panel. At

the family-panel level, Krishnamurthi and Raj specify a log-linear demand for

the brand, where the log of the (relative) price of the brand is interacted with

a time (pre-test, test) dummy variable. They report that demand for the brand

becomes significantly more inelastic in the test panel of families once advertising

is increased.

Guadagni and Little (1983) advance an alternative approach. They obtain

household brand purchase data through supermarket scanner data. These data

include individual item sales and prices by store by week, promotional activities

within the store, and histories of purchases for samples of households. The multi-

nomial logit choice model of brand choice is well suited for the analysis of such

data. Guadagni and Little illustrate the power of this approach, by using scanner

data on 32 weeks of purchases of coffee by 100 households and estimating the

parameters that govern consumers’ optimal brand-size choices. ik

In the multinomial logit model, consumer enjoys utility from alternative

i u

(i.e., a brand-size choice), where utility consists of deterministic and random

k ik i

components: Under an appropriate (extreme value) distibutional

u = v + ² .

k

k

assumption for , consumer optimally chooses alternative with probability

² i k

k X i

i v

ik v

= e / e ,

p j

k i

j∈S 123

i

where is the set of alternatives under consideration by consumer Next,

S i.

the deterministic term is decomposed into a linear combination of attributes that

are associated with alternative k: X

i ijk,

= b x

v j

k j∈T

ijk

where denotes the value of attribute for consumer under alternative The

x j i k.

set of attributes includes price, promotion and also brand and size experience

T

123 See McFadden (1974). 119

124

measures. The econometrican observes consumer choices and attribute values

parameters using maximum likelihood methods.

and then estimates the b

j

How is brand experience measured? At the time of the coffee purchase of

nth

consumer the experience that this consumer has with the brand associated with

i,

brand-size alternative is an attribute of this alternative that is measured as a

k

weighted average of past purchases of the brand, where past purchases are treated

125

as variables. The experience variables are then initialized using household

0 1

purchase observations for previous weeks. Using this approach, Guadagni and

Little report that brand and size experience are the most important attributes in

explaining consumer brand-size choice. Guadagni and Little do not have house-

hold advertising exposure data, however, and so their analysis leaves open an

important question: What are the respective roles of advertising and experience

in explaining household brand purchase behavior?

This question is the focus of subsequent work. For 251 households in a test city,

Tellis (1988) obtains scanner data for purchases of 10 brands of toilet tissues over

a 52-week period, and he also obtains TV meter records of household exposure to

brand advertising. Tellis seeks to explain both brand choice and volume, where

explanatory variables include brand experience, volume experience, advertising

exposure and price. Like Guadagni and Little, Tellis uses purchase behavior in a

pre-test period to develop experience measures. Using advertising exposure data,

he is also able to assess the impact of advertising on brand choice and volume,

both directly and interacted with experience. In line with Lambin’s (1976) work,

Tellis reports that experience is the strongest determinant of purchase behavior,

and that other marketing variables like price are more important than advertis-

126

ing. Advertising appears to have only a small effect on brand choice. According

to this evidence, pioneering firms may enjoy important experience-based advan-

tages; however, advertising itself is not one of the more important determinants

of purchase behavior.

These findings are evaluated in further work that uses scanner and advertising

124 Measurements of past purchase behavior are also sometimes referred to as indicating brand

(or size) loyalty.

125 ibk (n) denote the brand experience that consumer i has at the time of the nth purchase

Let x ibk ibk

occasion for the brand associated with the brand-size alternative k. Then x (n) = α x (n −

b

ik ik

1) + (1 − α )d (n − 1), where d (n − 1) is a 0 − 1 dummy variable that takes value 1 if consumer

b

i bought the brand associated with brand-size k at purchase occasion (n − 1). The smoothing

constant α is selected by trial and then refined. Size experience is measured similarly.

b

126 Pedrick and Zufryden (1991) consider the yogurt product category. In comparison to Tellis’s

(1988) study, they report a stronger direct effect of advertising exposure on brand choice.

120

exposure data and multinomial logit models to explain household brand-choice

behavior. Kanetkar et al (1992), for example, consider the product categories

of aluminum foil and dry dog food. Their explanatory variables include brand

experience, advertising exposure and price. They also find that the direct impact

of advertising appears small in comparison to other marketing variables like price.

Examining the interaction of price with advertising exposure, they further report

that increased advertising exposure is associated with greater brand choice price

sensitivity. One interpretation is that advertising increases the “identifiability” of

127

different brands and thereby promotes price comparisons. Deighton et al (1994)

consider ketchup and detergent. They find a large inertia (loyalty) effect, in that a

buyer is likely to purchase the same brand as was bought on the previous shopping

trip. Allowing for interactions between previous purchase and advertising, they

find that advertising does little to change the repeat-purchase probabilities of

consumers that have just purchased the brand. Advertising can be effective,

however, in attracting consumers who have not recently purchased the brand.

The studies above are published in marketing journals, but economists are

now also conducting related analyses. Ackerberg (2001) constructs a binary logit

model to explain the household choice of whether to purchase a newly introduced

yogurt product, Yoplait 150. Explanatory variables include previous purchase

measures, advertising exposure, price and time. Advertising is also interacted

with an experience variable, where a consumer is experienced (inexperienced) if

he has (never) purchased Yoplait 150 in the past. Advertising’s effect on inex-

perienced consumers is positive and significant, whereas advertising has only a

small and insignificant effect on experienced consumers. Ackerberg also considers

a specification in which experience is measured in terms of the number of previous

purchases. He finds that the effectiveness of advertising declines as the consumer

128

becomes more experienced (i.e., as the number of previous purchases increases).

The models described above endogenize consumers’ current brand choices but

are nevertheless “reduced form.” Consumers’ past purchases are regarded as ex-

ogenous data that generate a brand experience attribute with which to better

explain current brand choices. On the other hand, in a “structural” empricial

model, the consumers’ dynamic choice problem is fully specified, and the pa-

127 This is in the spirit of Steiner’s (1973, 1978, 1984, 1993) work. See Section 3.2.4.

128 Shum (2004) reports similar findings in his study of household brand choice in the breakfast

cereal category. His investigation uses scanner data for 50 brands of breakfast cereal combined

with an aggregate measure of advertising exposure (namely, quarterly brand-level national ad-

vertising expenditures). 121

rameters of the consumers’ utility function and/or constraints are estimated. A

structural model thus may offer greater insight into the process through which

advertising affects consumer purchase behavior.

In Erdem and Keane’s (1996) structural model, the utility that a consumer

derives from the purchase of a brand is a function of the brand’s attributes and

a random component; however, for each brand there is now an attribute (“qual-

ity”) whose value is uncertain and experienced with noise. The utility function is

parameterized to allow that the consumer may be risk averse with respect to the

experienced value of this attribute. The consumer seeks to learn the mean value

of a brand’s attribute, and the precision of the consumer’s information may be

improved by direct experience with the brand and observation of brand advertis-

ing messages. A forward-looking consumer thus may experiment and purchase a

brand today, in order to acquire information. The method of simulated maximum

likelihood is used to estimate parameters that describe the utility function and

the precision of experience and advertising exposure signals, so as to best explain

brand choices. Using scanner data and household advertising exposure data for

different brands of laundry detergent, Erdem and Keane report that consumers are

risk averse and that experience is much more informative than advertising. The

how

model thus provides insight into brand loyalty is formed: due to risk aversion,

consumers are loyal to brands that have delivered positive use experiences.

Ackerberg (2003) offers a related structural model of brand choice. In his util-

ity specification, however, the consumer may be interested in observed advertising

for two reasons. First, if a consumer’s prior belief is that a brand’s advertising

intensity is positively associated with the value of its attribute, then observed

brand advertising provides indirect (signaling) information as to the brand’s at-

tribute value. Second, if a consumer directly values the prestige effect that higher

brand advertising intensity is perceived to imply, then greater observed brand

advertising is indicative of a higher direct utility from brand purchase. Using

scanner data and advertising exposure data for Yoplait 150, Ackerberg conducts

a structural estimation. Identification of the informative and prestige effects is

possible, since the informative effect suggests that advertising affects the purchase

probabilities of inexperienced consumers only whereas the prestige effect implies

that advertising also affects the purchase probabilities of experienced consumers.

Ackerberg reports that advertising has a large and significant informative effect

129

and an insignificant prestige effect.

129 The studies above emphasize the effect of advertising on household purchase behavior for

various consumer goods. By contrast, Shachar and Anand (1998) consider the effect of “tune-

122

Finally, as Ippolito and Mathios (1990) illustrate, the effect of advertising on

household purchase behavior also may be examined using event studies. They fo-

cus on the ready-to-eat cereal market. In response to growing evidence of fiber’s

potential cancer-preventing benefit, a regulatory ban in the U.S. on health-claim

advertising by cereal producers was lifted in 1985. Using brand-level cereal con-

sumption data, Ippolito and Mathios find that fiber cereal consumption increased

significantly, once the ban on health-claim advertising was removed. They also

use brand-level cereal consumption data for samples of individuals in 1985 (prior

to most health-claim advertising) and 1986 (more than a year after health-claim

advertising began). The household data suggest that advertising lowered the cost

of acquiring health-related information for individuals who were not well reached

by other health information sources.

In broad terms, the studies described above point toward a number of striking

conclusions. For a set of frequently purchased consumer goods: (1) experience is a

very important determinant of household purchase behavior; (2) advertising also

influences household purchase behavior, but experience is the more powerful input;

(3) advertising and experience are substitutes, in that advertising is less effective

in influencing purchase behavior for households that have recent experience with

the brand; and (4) much of advertising’s effect derives from the information that it

contains or implies. On the whole, the studies provide support for the informative

view of advertising.

The studies are of particular interest in light of the long-standing debate as

to whether advertising deters entry. As discussed in Section 7.1, existing the-

ory demonstrates that informational product differentiation may be a barrier to

entry; however, the theoretical literature to this point does not clearly identify

a sense in which advertising “reinforces” consumers’ experience and exacerbates

in” ads (i.e., TV ads in which a network advertises one of its own shows) on the TV viewing

decisions of individuals. Viewers are assumed to possess greater prior information about the

existence and attributes of regular shows than specials; thus, a differential effect of tune-in ads on

viewing decisions across the two show categories may suggest that advertising has informational

content. Using a Neilsen data set that records individual characteristics and viewing behavior,

Shachar and Anand specify a nested multinomial logit model and report estimates indicating

that a differential effect is indeed present. Anand and Shachar (2004b) provide further support

for the informative content of tune-in ads. Consistent with the models reviewed in Section 6.3,

they provide evidence that advertising enables buyers to better match their respective tastes

with the product attributes offered by different shows. Finally, Byzalov and Shachar (2004) also

study TV viewing decisions and report that advertising has a negligible direct effect on utility;

instead, advertising provides information and thereby reduces the uncertainty that risk-averse

consumers face when contemplating purchase of the advertised product.

123

the informational barrier to entry. Likewise, the studies described above support

the idea that consumer experience is an important asset for pioneering brands;

however, they suggest that advertising itself does little to reinforce experience.

At the same time, it must be emphasized that the studies have important

limitations. First, they focus on a narrow set of consumer goods. An important

task of future work is to determine the extent to which the conclusions of these

studies extend to other goods. Second, the studies treat price and advertising ex-

posure as exogenous variables. This is a concern, since brand choice may depend

upon attributes that are observable to market participants but unobservable to

the econometrician. In this case, price and advertising exposure may be corre-

130

lated with the error term. The possibility of endogeneity bias thus motivates

a structural approach that jointly estimates demand function parameters along

with parameters that determine firm behavior. Work of this kind is considered in

the next subsection.

8.2. Advertising and Firm Conduct

I consider next empirical studies that reflect the strong influence of the intervening

theoretical work and emphasize firm conduct. Some studies adopt a reduced-

form approach and evaluate the predictions of strategic theories of advertising,

while others adopt a more structural approach and specify demand functions, cost

functions and supply relationships.

Consider first the reduced-form studies that assess the predictions of strategic

advertising theories. While some recent papers discussed in Section 3 report

evidence that is relevant for the descriptive validity of intervening theoretical

work, I illustrate this style of analysis here using papers by Thomas et al (1998),

Horstmann and MacDonald (2003) and Ellison and Ellison (2000). I do this for

two reasons. First, these papers identify and assess predictions that are tightly

linked with the intervening theoretical work. Second, it is useful to collect as

many papers as possible in Section 3, so that the topic treatments found there

may be more self contained.

Using auto industry data, Thomas et al assess the advertising-quality relation-

ship. They provide evidence that models priced higher than the full-information

price tend to have higher advertising levels. Referring to the Milgrom-Roberts

(1986) model, the authors emphasize that this behavior is consistent with the

hypothesis that manufacturers of high-quality models signal unobservable quality

130 For further discussion, see Berry (1994), Villas-Boas and Winer (1999) and Nevo (2001).

124

attributes by setting prices above full-information levels and advertising expen-

131 The

ditures beyond those incurred by manufacturers of low-quality models.

signaling interpretation is further supported by the finding that these relation-

ships are weaker for older models. Finally, as the repeat-business effect suggests,

they find that automobiles that experience higher sales five years after introduc-

tion are characterized by greater advertising in the introductory period. These

findings are broadly consistent with Nelson’s (1974b) reasoning, but he does not

address the possibility that price and advertising serve as joint signals of qual-

ity. The predicted relationships between price and advertising are thus strongly

influenced by the intervening theoretical work.

Horstmann and MacDonald (2003) provide a related analysis that focuses on

the compact disc player market. Using panel data on advertising and pricing

during 1983-92 and controlling for product features, firm heterogeneity and ag-

gregate effects, they provide evidence that advertising increases after a player is

introduced and price falls from the outset. As Horstmann and MacDonald observe,

this pattern is not easily reconciled with signaling models in which advertising is

dissipative. A possible interpretation of this pattern is provided by the static sig-

naling model of Section 6.1, however, when advertising is demand-enhancing and

a higher-quality product has a higher marginal cost. The high-quality monopolist

then best signals its quality by distorting its demand-enhancing advertising down-

ward and its price upward. In the dynamic perspective suggested by that model,

the high-quality product’s demand-enhancing advertising increases over time and

its price falls over time.

Ellison and Ellison (2000) consider the behavior of pharmaceutical incumbents

in the period of time that precedes the loss of patent protection. The incentive

to deter entry is greatest in intermediate-sized markets, since entry deterrence

is unnecessary (impossible) in markets that are sufficiently small (large). For

prescription drugs, incumbent advertising has a public-good aspect, in that some

of the benefits may accure to generic entrants; thus, an incumbent operating in

an intermediate-sized market has a potential incentive to reduce advertising and

thereby reduce the profitability of entry. This rationale for diminished advertising

is weakened in larger markets, as the incumbent’s focus switches from deterrence

131 Specifically, the authors first regress model i’s price at time t on the model’s observable qual-

ity attributes (horsepower, etc.) and other variables. The residual is interpreted as capturing

deviations from the full-information price that are due to unobservable quality attributes. Sec-

ond, they regress the advertising level for model i at time t on the corresponding price residual

(and other variables). A positive coefficient is consistent with the described hypothesis.

125

to accommodation. Arguing in this way, Ellison and Ellison build on intervening

theoretical work (e.g., Fudenberg and Tirole (1984)) and offer a novel prediction:

advertising may be reduced most rapidly in years prior to patent expiration in

markets of intermediate size. Using data on 63 drugs that faced patent expirations

over 1986-1992, they also report evidence that supports this prediction.

Consider next empirical studies that follow the methodology of the “new em-

pirical industrial organization” (NEIO) and adopt a more structural approach. It

is instructive to contrast the NEIO approach with the earlier structure-conduct-

performance paragidm (SCPP) that underlies the inter-industy studies of Bain

(1956), Comanor and Wilson (1967, 1974) and followers. In broad terms, the

SCPP makes two assumptions: (1) across large groups of industries, a stable and

causal relationship runs from exogenous structural characteristics through con-

duct to performance; and (2) market-power measurements of performance may

be calculated from available (e.g., accounting profit) data. As Breshahan (1989)

explains, the NEIO is distinguished from the SCPP in several respects. Among

these are: (1) the assumption of symmetry across industries is abandoned, and

instead an econometric model of a single industry (or a closely related set of mar-

kets) is developed; (2) market power is not treated as observable, and instead

the analyst infers marginal cost from firm behavior; and (3) firm and industry

conduct are not treated as simple implications of market-structure variables, and

instead the analyst specifies behavioral equations that are based on theoretical

models and uses estimates to test between models. 132

The standard NEIO analysis has three basic ingredients. First, demand

functions are specified. For example, a firm’s output may be a linear function of

own and rival prices as well as exogenous demand variables like income. Second,

marginal cost functions are specified. A firm’s marginal cost might be a linear

function of its output and exogenous cost variables like input prices, for instance.

Third, supply relationships are specified. A firm’s supply relationship corresponds

to a first-order condition for optimizing behavior. Once the marginal cost func-

tions are substituted into the firms’ respective supply relationships, the demand

functions and supply relationships constitute an econometric system of equations,

in which outputs and prices are endogenous variables, and the demand, marginal

cost and any conduct parameters may be estimated.

How are the supply relationships specified? Under one approach, the sup-

ply relationships include a conjectural variations or conduct parameter that is

estimated as a continuous variable. Under appropriate conditions, the conduct

132 See Bresnahan (1989), Church and Ware (1999) and Kadiyali et al (2001).

126 133

parameter can be identified and a performance inference thereby obtained. The

conjectural variations approach includes as special cases a number of hypotheses

as to firm behavior. The analyst may then test among these hypotheses using

nested methods. But the approach also has limitations: the estimated conduct

parameter may not correspond to any particular model of firm behavior, and some

interesting types of behavior (such as asymmetric collusion) may not be included

as special cases. An alternative approach is to consider a menu of models. For

example, the Bertrand, Stackelberg and Collusive models imply distinct supply

relationships that may be individually considered. Under the menu approach, the

analyst may test among models using non-nested methods and then emphasize

parameter estimates for the preferred model.

Some recent NEIO studies include advertising as an endogenous variable. The

models are then more complex. Each firm may have multiple choice variables;

furthermore, if a goodwill effect is allowed, then the demand functions and supply

relationships must be dynamic. If the conjectural variations approach is adopted,

then dynamic conduct parameters may be specified and estimated, where such

parameters indicate a firm’s perception as to how a change in its current be-

havior would alter rival behavior in the future. The identification of structural

parameters then requires that some restrictions be placed on the dynamic conduct

parameters. Finally, it is desirable that the specification of demand functions be

sufficiently flexible to include the primary (market-size) and selective/combative

(market-share) effects of advertising.

Roberts and Samuelson (1988) offer an early study of this general nature. They

develop an analysis of dynamic non-price rivalry among U.S. cigarette manufactur-

134

ers in high- and low-tar cigarette markets over the 1971-82 period. The demand

functions are specified in a multiplicatively separable fashion that facilitates the

identification of the market-size and market-share effects of advertising. Mak-

ing use of factor demand data, Roberts and Samuelson estimate marginal costs

135

directly. Finally, the supply relationships are captured as dynamic first-order

133 The conjectural variations approach is pioneered by Iwata (1974). See Nevo (1998) and

Corts (1999) for discussion of identification problems under the conjectural variations approach.

134 Cigarette advertising was banned from TV and radio over this period, but substantial

advertising expenditures were made in magazines, newspapers and outdoor media.

135 The approach here is to specify a total cost function, use Shephard’s lemma to derive

a system of factor demands and then estimate the parameters of this system. With the cost

parameters thus estimated, the estimated value of marginal cost can be determined as a function

of input prices and output volumes. See Bresnahan (1989, pp. 1039-40) for discussion of

the strengths and weaknesses of this approach relative to the alternative approach mentioned

127

conditions for firms’ goodwill choices, where a firm’s dynamic conduct parameter

is restricted to describe the extent to which an increase in the firm’s goodwill

stock at date would induce rivals to increase their goodwill stocks in period

t

Their estimates suggest that advertising is not combative; in fact, adver-

t + 1.

tising in new-product categories (i.e., in the low-tar market) appears to expand

market sizes and constitute a public good among firms. They further report that

the estimated dynamic conduct parameters are negative. Evidently, firms are not

naive: each recognizes that an increase in its own advertising would encourage

less (market-size-expanding) advertising from rivals in the future.

Using data on the Coca-Cola and Pepsi-Cola markets over the 1968-86 period,

Gasmi et al (1992) illustrate the menu approach. They specify a demand func-

tion for each product, where sales depend on own and rival price and advertising

selections. The demand specification presumes that advertising has no goodwill

effect. Marginal cost is constant at a value that is specified to be linear in input

prices. Using the demand and cost specifications, they then turn to the supply

relationships and derive first order conditions for each firm in price and advertis-

ing, where the parameters of these conduct equations take different restrictions

as different oligopoly games (Nash in prices and advertising, Nash in prices and

collusion in advertising, etc.) are considered. For any given game, the two de-

mand and four conduct equations can be simultaneously estimated, where the

six endogenous variables are the prices, advertising levels and quantities of the

two firms. After determining the best-fitting game, the authors then emphasize

the associated parameter estimates. Their analysis suggests that Coca-Cola was

a Stackelberg leader in price and advertising until a mid-sample period (1976).

After this period, duopoly conduct is characterized by collusion in advertising and

possibly price. In this context, their estimates suggest that advertising in the cola

market is largely combative.

This approach is also used by Kadiyali (1996), who analyzes the U.S. pho-

tographic film industry. In the 1970s, Kodak had a virtual monopoly of this

industry; however, Kodak accommodated entry by Fuji in the 1980s. Kadiyali

refers to 1970-1980 (1980-1990) as the pre-entry (post-entry) period. She specifies

a demand function and a constant marginal cost for each firm, and then considers

the two periods separately. In the pre-entry period, only Kodak is active, and

the supply relationship is described by Kodak’s pricing and advertising first-order

conditions. In the post-entry period, the supply relationship is described by pric-

ing and advertising first-order conditions for both firms, where the parameters

previously, whereby marginal cost is inferred from the supply behavior of firms.

128

of these conditions assume different restrictions as different post-entry games are

considered. Kadiyali’s parameter estimates for the pre-entry period indicate that

Kodak maintained its monopoly position by using limit pricing and high adver-

tising. As in the Bagwell-Ramey (1988) model, a possible interpretation is that

Kodak reduced price and raised advertising in order to signal low costs. Kadiyali’s

estimates for the post-entry period suggest several conclusions, including: (1) Ko-

dak was compelled to accommodate Fuji by 1980, since Fuji enjoyed demand

and cost advantages; (2) Kodak and Fuji then colluded in price and advertising,

putting a large weight on Fuji’s profit; and (3) advertising expanded market size

and constituted a public good across firms.

Finally, Slade (1995) develops a dynamic “state-space” approach with which

to study price and advertising brand rivalry. In this formulation, firms adopt

Markov strategies that determine price and advertising behavior, given the cur-

rent state of play. The empirical model is described by demand and strategy

equations. The endogenous variables of the strategy equations are the size and

probability of price and advertising changes, while the exogenous variables in-

clude factor prices (costs) and past endogenous choices (goodwill). Using weekly

price, sales and advertising data for four brands of saltine crackers sold in grocery

stores in a small town, Slade obtains estimates suggesting that a brand’s sales

are decreasing (increasing) in own price (advertising) and increasing (decreasing)

in rival-brand price (advertising). Advertising is thus combative, but it has an

overall positive effect on market size. Given the specification of linear demand

and costs, the demand coefficient estimates may be used to draw inferences about

the strategic environment. Slade reports cross-brand evidence that advertising

efforts are strategic substitutes, prices are strategic complements, and low prices

and high advertising make a brand “tough” (i.e., reduce rival-brand profits). In

the dynamic game, firms thus compete aggressively in advertising and accom-

modate when setting prices, but the resulting high prices do not reflect collusive

136

behavior. A further implication is that entry-deterring behavior would involve

limit pricing and high advertising.

Placing firm conduct at centerstage, the empirical studies reviewed here are

strongly influenced by the intervening theoretical work. While the NEIO analysis

of advertising is just getting underway, it is already clear that one conclusion of

the earlier empirical work is retained: the effects of advertising vary importantly

across markets. The recent work also generates some interesting specific findings.

136 Nevo (2001) draws a similar conclusion in his analysis of the ready-to-eat cereal industry.

See also Vilcassim et al (1999). 129

First, in some markets, there is evidence that firms choose advertising in a collu-

sive manner. This contrasts with a common view that firms compete aggressively

in non-price variables, although support for the common view is found in other

markets. Second, while advertising is often combative, there is also some support

for the market-size effect (e.g., in new-product categories). Finally, some studies

offer new evidence that limit pricing and high advertising may deter entry. While

these are interesting findings, the primary contribution of the existing NEIO ad-

137

vertising studies is methodological. The studies reviewed here pave the way for

what should be an active and valuable research area in the coming years.

8.3. Summary

The research described here constitutes an important advance in the empirical

analysis of advertising. While the earlier inter-industry analyses searched for evi-

dence of general causal relationships from structure to performance, the studies re-

viewed above emphasize the limitations of the inter-industry approach and instead

use new disaggregated data sources to explore household and firm conduct. One

set of studies examine purchase decisions, using household-level brand-purchase

and advertising-exposure panel data. These data offer a remarkable opportunity

to study a long-standing and fundamental question in the economic analysis of

advertising: Does advertising reinforce consumer experience and insulate pioneer-

ing firms from entry? A second set of studies integrate game-theoretic models of

advertising into the empirical investigation. Some studies examine the descriptive

validity of the models, while others implant a model of the supply relationship

into the system of equations that is to be estimated. These studies offer a window

into the strategic conduct of firms.

9. Sunk Costs and Market Structure

As Sections 5 through 7 reveal, an important lesson of game-theoretic models in

industrial organization is that details may matter. Empirical efforts that follow

the SCPP and seek inter-industry confirmation of sweeping causal hypotheses are

thus too ambitious. But what are the alternatives? As illustrated by the NEIO

137 The work described above highlights two advantages of the structural methodology: it may

be used to estimate unobserved economic parameters and to compare the predictive power of

alternative theories. A further advantage is that an estimated structural model may be used to

make policy recommendations. Dube et al (forthcoming) perform an analysis of this kind.

130

studies reviewed in the previous section, one alternative empirical strategy is to

focus on a particular industry, where more details are observed and the theory

imposes tighter restrictions. As Sutton (1991) emphasizes, a second strategy is to

cull from the game-theoretic models a few robust implications and then examine

those implications at inter-industry and industry levels. In this section, I provide

138

a brief review of work by Sutton and others that follows this second strategy.

9.1. Main Ideas

Sutton develops robust predictions that concern the manner in which the en-

dogeneity of sunk costs and the “toughness of price competition” influence the

139 To this end, he models

relationship between market size and concentration.

industry equilibrium in terms of a multi-stage game, in which firms enter, sink

costs, and then compete (e.g., in prices) in the product market. Considerable lat-

titude is allowed as to whether firms sell horizontally or vertically differentiated

products, move sequentially or simultaneously within given stages of the game,

sell single or multiple products, or choose prices or outputs. exogenous

Sutton distinguishes between two categories of industries. In an

sunk cost industry, the only sunk costs are exogenous setup costs. These are the

costs of acquiring a single plant of minimum efficient scale and perhaps advertising

at some threshold level. An exogenous sunk cost industry may be an industry with

endogenous

homogeneous or horizontally differentiated goods, for example. In an

sunk cost industry, by contrast, a firm incurs advertising (or R&D) outlays which

result in an enhanced demand for that firm’s product in the subsequent stage of

140

product-market competition. As in the first example described in Section 4,

an industry is characterized by endogenous sunk costs if products are vertically

differentiated as a consequence of brand-image advertising, for instance. Sutton is

not concerned with the reason that advertising works in such an industry; rather,

he assumes that it does and then examines the implications.

Consider first the case of an exogenous sunk cost industry. To illustrate the key

predictions, imagine that firms sell products that are differentiated in a symmetric

sense, so that the equilibrium price when firms enter may be represented as

N

138 For further discussion, see Bresnahan (1992), Schmalensee (1992), Sutton (1997a) and Sut-

ton’s contribution to this volume.

139 Sutton’s analysis builds on that in Shaked and Sutton (1983, 1987, 1990).

140 The analsyis is extended to include endogenous R&D sunk costs in Sutton (1997b, 1998).

For an early analysis of this kind, see Dasgupta and Stiglitz (1980).

131 141

| where denotes the toughness of price competition. For example,

p(N θ), θ

may correspond to transportation costs or competition policy. A firm’s cost

θ

function is where denotes the exogenous setup costs that

C(q) = cq + σ, σ > 0

are associated with entry. An increase in the market size is accomplished

S > 0

through successive replications of the consumer population. This ensures that the

distribution of tastes is not altered, so that the equilibrium price does not depend

directly upon market size. Ignoring the setup cost it is then possible to denote

σ, |

a firm’s equilibrium gross profit function as In most such models,

SΠ(N θ). 142

| |

is nonincreasing in and for all Assume then that

p(N θ) N p(N θ) > c N.

| | → → ∞.

is positive and decreasing in with as The

Π(N θ) N, Π(N θ) 0 N

|

equilibrium level of entry is determined by SΠ(N θ) = σ.

Using this example, two predictions can be described. First, as market size S

increases indefinitely relative to the setup cost the equilibrium concentration,

σ,

measured as converges monotonically to zero. Intuitively, an increase in

1/N,

market size always raises profit and invites further entry, where the additional

entry restores the zero-profit requirement by reducing each firm’s market share

without increasing its markup. Economies of scale thus become unimportant as

a barrier to entry in markets that are sufficiently large. The second prediction

concerns the effect of an increase in the toughness of price competition. An

| |

increase in is associated with a reduction in Assume then that

θ p(N θ). Π(N θ)

is decreasing in Under this assumption, a second prediction follows: an increase

θ.

in the toughness of price competition shifts the equilibrium concentration upward.

This simply reflects the familiar intuition that fewer firms can enter in a zero-profit

equilibrium, when price competition is more vigorous.

As Sutton discusses, the main features of this example generalize across a

wide range of models. In some of these models, multiple equilibria may arise. For

example, if products are horizontally differentiated, then there may exist many

single-product firms or a smaller number of multi-product firms. The functional

relationship just described between concentration and market size is thus replaced

by a lower bound relation. More generally, as Sutton (1991, p. 308) states, the

two robust predictions for exogenous sunk cost industries are: (i) the function

that gives the lower bound to equilibrium concentration converges monotonically

to zero as market size increases; and (ii) this lower bound shifts upward, in re-

141 For example, Dixit and Stiglitz (1977), Shubik and Levitan (1980) and Sutton (1997a, 1998)

provide models of this kind. My discussion here follows that in Sutton (1997a).

142 An exception is the case of Bertrand competition with homogeneous goods. In this case,

only one firm enters, regardless of market size.

132

sponse to an increase in the toughness of price competition. These predictions

are illustrated in Figure 3a.

Consider second the case of an endogenous sunk cost industry. As Sutton

(1991, p. 47) puts it, the main point is then as follows:

“If it is possible to enhance consumers’ willingness-to-pay for a given

product to some minimal degree by way of a proportionate increase

fixed

in cost (with either no increase or only a small increase in unit

variable costs), then the industry will not converge to a fragmented

structure, however large the market becomes.”

As this quotation suggests, in endogenous sunk cost industries, the negative rela-

tionship between concentration and market size breaks down.

Formally, suppose that a firm’s product is described in terms of a single vertical

attribute, where the willingness-to-pay of all consumers is increasing in In

u, u.

an endogenous sunk cost industry, a firm’s advertising expenditures may affect its

brand image and thus therefore, let the advertising response function,

u; A(u),

indicate the sunk expenditure that a firm must incur in order to achieve where

u,

is non-negative and increasing. A firm’s total fixed cost expenditure is then

A(u) ≥

. Let denote the firm’s unit cost of production. Now,

F (u) = A(u) + σ c(u) 0

assume that there exist constants and such that by incurring

α > 0 K > 1 K

times more fixed costs than any of its rivals, a firm will achieve a final stage

(i.e, gross) profit that is no less than where corresponds to total consumer

αS, S

expenditure in the market. This assumption can be understood as embodying

two features: (i) a sufficiently high attribute implies a certain minimal level of

profit in the final stage (i.e., does not increase too quickly with and (ii) a

c(u) u),

certain large increase in fixed advertising expenditures translates into a sufficiently

high attribute (i.e., is increasing and continuous, with an elasticity that is

F (u)

bounded above). Under this assumption, as Sutton (1991, pp.73-4) establishes, a

non-convergence property is implied: there exists some such that some firm

B > 0

must enjoy at least a fraction of total industry sales at any subgame perfect

B 143

equilibrium, independent of the size of the market.

143 The proof below is developed for a three-stage game between single-product firms, in which

firms enter, sink costs and then compete, with simultaneous moves in each stage. In some

settings, equilibria may fail to exist, and so only the necessary features of equilibria are char-

acterized. As Sutton explains, the proof may be applied to a variety of related games. For

example, if firms sink costs sequentially within the second stage of the game, then the proof

applies once the deviant firm is identified as the firm that comes last in the sequence. See Sutton

(1997b, 1998) for further discussion of such games.

133 _

u

The proof is instructive and simple. For a given equilibrium, let denote

_

m denote the highest share

the highest value of offered by any firm, and let

u

of industry sales enjoyed by any firm. The profit in the final stage to any firm

_ _

m u

clearly cannot exceed Hence, if the firm that offers is to earn non-negative

S.

profit in the game, then it is necessary that

_ _

m u).

≥ (9.1)

S F (

Now suppose a firm were to deviate and advertise to such an extent as to incur

_

u). Using the assumption stated above, the deviant firm enjoys

fixed costs KF (

net profit that is at least _

u).

− (9.2)

αS KF ( _

m

−K

Using (9.1) and (9.2), it follows that the deviant firm earns at least αS S =

_

m]S.

− Of course, in equilibrium, a firm cannot earn net profit in excess of

[α K

_ _ _

m m m]S,

≥ −

Therefore, an equilibrium exists only if or equivalently,

S. S [α K

α

_

m≥ ≡ (9.3)

B.

1+ K

Thus, as (9.3) confirms, regardless of the size of the market, in any equilibrium

the maximal market share must exceed the constant B.

Intuitively, under the assumption that a given proportionate increase in a

firm’s advertising outlay relative to that of rivals can induce some some fixed

fraction of consumers to purchase that firm’s product at a price that exceeds the

firm’s unit variable cost, a fragmented market structure cannot stand: some firm

would deviate with a large advertising outlay and earn greater profit. In equilib-

rium, as market size increases, the tendency toward fragmentation is offset by a

competitive escalation in advertising outlays. This suggests that the relationship

between market size and concentration may be non-monotonic. Sutton provides

some examples in support of this suggestion. In summary, two robust predictions

for endogenous sunk cost industries are that the lower bound to equilibrium con-

centration (i) is bounded away from zero as market size increases, and (ii) is not

144

necessarily monotonic in market size. Figure 3b illustrates.

144 Sutton (1991, p. 308) mentions further the robust prediction that an increase in setup cost

(σ) results in an increase in the lower bound for concentration.

134

9.2. Econometric Tests and Industry Histories

Sutton (1991) next confronts the predictions of the theoretical analysis with a

careful analysis of twenty narrowly defined food and drink manufacturing indus-

145

tries across six countries. He divides the industries into two groups. In the first

group, firms sell homogeneous products, and advertising outlays are very low. The

homogeneous-goods industries are examined in light of the theoretical predictions

for exogenous sunk cost industries. In the second industry group, advertising out-

lays are moderate to high. The advertising-intensive industries are thus analyzed

with reference to the theoretical predictions for endogenous sunk cost industries.

Cross-country comparisons afford the necessary variation in market size.

The empirical effort begins with a cross-sectional econometric analysis of ob-

served concentration levels. The empirical regularities that Sutton uncovers are

consistent with the predictions of the theory. In the group of homogenous-goods

industries, he reports a strong negative correlation between (four-firm) concentra-

tion and the ratio of market size to setup cost, He observes further that the

S/σ.

lowest levels of concentration found at large values of are very small (below

S/σ

5%). By contrast, in the advertising-intensive group, the lowest level of concen-

tration is 25%. Several of these industries also have large values of Bounds

S/σ.

regression analysis offers further support for the predictions of the theory.

With the inter-industry statistics in place, Sutton next presents a remark-

able set of industry studies. Interesting on their own, these studies offer fur-

ther opportunities for assessing the theory. For example, in his discussion of

homogeneous-goods industries, Sutton considers the salt and sugar industries and

identifies international differences in competition policy that suggest variation in

the toughness of price competition. This variation facilitates an examination of

the prediction that, for exogenous sunk cost industries, tougher price competition

is associated with more concentrated markets. Broadly, the industry experiences

are consistent with this prediction. Likewise, within the group of advertising-

intensive industries, Sutton offers convincing qualitative support for the main

theoretical ideas. For example, consistent with the hypothesis of a lower bound,

in the frozen food industry, a wave of new entry resulted in a fragmentation of

the market that sparked a competitive escalation in advertising outlays by lead-

ing firms, leading back to a more consolidated structure. Finally, Sutton’s case

studies also confirm that details matter. Above the lower bound, a rich array of

strategic interaction is observed.

145 The six countries are France, Germany, Italy, Japan, the U.K. and the U.S..

135

9.3. Related Work

Sutton’s research is related to several strands of work that are discussed above.

I describe here four such relationships. I then mention some recent work that

extends Sutton’s theoretical and empirical analyses.

Consider first the inter-industry studies of advertising and concentration. As

discussed in Section 3.2.1, while the advertising-concentration relationship is cen-

tral to Kaldor’s arguments and the focus of a number of inter-industry studies,

the relationship has defied a simple characterization. Sutton (1991, pp. 125-8)

explains that his work encompasses a possible interpretation: if the relationship

between market size and concentration varies in kind between homogeneous goods

and advertising-intensive industries, then the earlier studies, which use pooled

data and ignore this switch of regime problem, are misspecified. This interpreta-

tion explains further why such studies occasionally report a positive and significant

relationship between advertising intensity and concentration. Consistent with Sut-

ton’s theory, suppose that (i) a negative (null) relationship exists between concen-

tration and market size for the homogeneous goods (advertising-intensive) group,

and (ii) the mean level of concentration is higher in the advertising-intensive group.

Under this supposition, if pooled data are used and concentration is regressed on

the market size/setup cost ratio and advertising intensity, then a positive coeffi-

cient is expected on the advertising intensity variable.

Second, while Sutton studies manufacturing industries, similar relationships

may also emerge in retail industries. As discussed in Section 3.2.4, some evidence

suggests an association between advertising and the growth of large-scale retail

firms. Consider the retail eyeglass industry. In the 1960s, considerable variation

existed across states in the U.S. with respect to the legal restrictions imposed

on advertising in the retail eyeglass industry. Depending on the scope of other

sunk cost outlays, it may be appropriate to regard a retail eyeglass market as

characterized by exogenous (endogenous) sunk costs when advertising is (not)

restricted in the corresponding state. Interestingly, work by Benham (1971) and

others (see footnote 55) suggests that large-scale retail firms operated in states

that permitted advertising. Likewise, as Steiner (1978) and Pashigian and Bowen

(1994) argue, the growth in manufacturer brand advertising, instigated by the

emergence of TV and the growth in (relative) earnings by females, may have

substituted for retail service and facilitated the emergence of a more concentrated

retail market structure.

In this context, it is also interesting to compare Sutton’s theoretical approach

with that of Bagwell and Ramey (1994a). As discussed in Section 5.3, Bagwell

136

and Ramey explore a multi-stage model of retail competition, in which firms

first enter and then make their advertising, pricing and (cost-reduction) invest-

ment decisions. It is straightforward to extend their model to include a market

size variable, corresponding to the total mass of consumers. If advertising is

S,

banned, a zero-profit “random” equilibrium obtains, in which each of the enter-

n

ing firms sells to consumers. As in Sutton’s exogenous sunk cost industries,

S/n

the market fragments as gets large. On the other hand, when advertising is en-

S

dogenous, a zero-profit “advertising” equilibrium obtains, in which entering firms

make heterogeneous decisions. As in Sutton’s endogenous sunk cost industries,

146

the market does not fragment as gets large.

S

Third, Sutton’s multi-stage approach, in which advertising outlays are sunk

prior to price competition, may be questioned in light of the empirical stud-

ies discussed in Section 3.1.1, which find that the effects of advertising on sales

are often brief. While this concern has some merit, it should be noted that

the no-fragmentation prediction for advertising-intensive industries may also hold

when advertising outlays do not precede price choices. Using a variant of the

Schmalensee (1976b) model of advertising competition, Schmalensee (1992, pp.

130) suggests that this prediction may be maintained whenever market share is

fixed

“sufficiently sensitive to variations in costs, so that rivalry is both tough and

147

focused on fixed outlays, not on per-unit price-cost margins.”

Fourth, it is interesting to compare Sutton’s theoretical findings with the

persuasive-view (see Section 2) and game-theoretic (see Section 7.2) examina-

tions of advertising’s entry-deterrence effect. Sutton offers some support for the

entry-deterrence effect, in that the scope for profitable entry is limited when ad-

vertising expenditures escalate. At the same time, it must be noted that Sutton

does not offer a theory in which an incumbent firm strategically advertises at a

high level in order to deter subsequent entry. In fact, advertising follows the entry

choice in Sutton’s basic model, so that it is the expectation of future advertising

rivalry that restrains entry.

Sutton’s theoretical and empirical analyses has been extended in several re-

cent efforts. Symeonidis (2000a) considers the theoretical effect of tougher price

competition on concentration in endogenous sunk cost industries. When price

146 In particular, the highest-advertising firm achieves a share of industry sales that is bounded

from below by the fraction of informed (i.e., advertising-responsive) consumers, I. See also

Bagwell et al (1997), who offer a related dynamic model of price competition that describes the

evolution of a concentrated retail market structure.

147 Likewise, in the Bagwell-Ramey (1994a) model, a no-fragmentation prediction occurs, even

though advertising and pricing decisions are simultaneously made.

137

competition is tougher, final-stage profits are reduced, giving firms less incentive

to sink advertising expenditures in the penultimate stage. As both gross profit

and sunk costs are then lower, the overall effect on net profit may be ambiguous.

As a general matter, then, in industries with endogenous sunk costs, an increase

in the toughness of price competition has a theoretically indeterminant effect on

concentration. In recent empirical work, Symeonidis (2000b) examines the evo-

lution of concentration in U.K. manufacturing industries following a change in

competition law that prohibited price-fixing agreements. The resulting increase

in the toughness of price competition is associated with greater concentration in

exogenous and even endogenous sunk cost industries. The relationship between

concentration and market size is negative in exogenous sunk cost industries, and

the relationship breaks down in industries with high advertising. These findings

are consistent with Sutton’s predictions. Other supportive empirical studies of

manufacturers are offered by Bronnenberg et al (2005), who study the geographic

distribution of brand market shares across U.S. metropolitan markets for several

consumer package goods industries, Matraves (1999), who examines the global

pharmaceutical industry, and Robinson and Chiang (1996), who use PIMS data.

Looking across U.S. metropolitan areas, Berry and Waldfogel (2004) study the

newspaper industry and offer evidence consistent with Sutton’s predictions for

endogenous sunk cost industries. Finally, Ellickson (2001a,b) considers the retail

supermarket industry. He reports that endogenous sunk costs associated with

investments in store size and information and distribution networks are an impor-

tant source of concentration in this retail market.

9.4. Summary

Sutton’s innovative effort contributes importantly at both methodological and

substantive levels. Methodologically, Sutton demonstrates an eclectic approach

that evaluates game theoretic models by employing traditional inter-industry

(SCPP) and recent industry-study (NEIO) empirical methods. This approach

invites theorists to explicitly distinguish between the robust and particular impli-

cations of their models. Robust implications, such as those associated with the

lower bound, may be examined using traditional inter-industry regressions. But

there is also a rich set of observed behaviors that occur above the lower bound.

The specific experiences of a given industry can be further interpreted using par-

ticular strategic models, historical analyses and recent industry-study methods.

At the substantive level, Sutton convincingly makes the fundamental point that

138

endogenous sunk costs in the form of advertising outlays often play a critical role

in the evolution of market structure. The role of (brand and retail) advertising

in the evolution of concentrated retail structures represents a promising direction

for future research.

10. New Directions and Other Topics

In this section, I briefly discuss two new directions for advertising research. The

first direction concerns the use of advertising in media markets. This is a long-

standing research topic that has enjoyed renewed attention in the past few years.

The second direction is at an earlier stage and concerns the potential implications

of findings in the fields of behavioral economics and neuroeconomics for advertising

research. Finally, despite the length of this survey, many topics remain untreated.

At the end of the section, I mention a few such topics and identify some research

for further reading.

10.1. Advertising and Media Markets

In the research reviewed above, sellers choose advertising levels and incur a cost

when delivering advertising messages to consumers. The models, however, gener-

148

ally treat the cost of advertising as exogenous. How is the price of an advertising

message determined? As emphasized by Kaldor (1950), advertising and enter-

tainment are often jointly supplied to the consumer: much advertising reaches

consumers through media markets. A viewer of a commercial TV broadcast, for

example, encounters frequent advertisements, and advertisements are also promi-

nent in magazines, newspapers and radio broadcasts. Advertising revenue is a

major source of income for media companies, and such companies naturally have

some control over the price of an advertising message. But an advertiser is will-

ing to pay only so much for a message, and the advertiser’s willingness-to-pay is

driven by the number of potential consumers that the message might reach.

149

two-sided market.

It is useful to think of the media market as a In a two-

sided market, two groups interact through an intermediary or platform, and inter-

148 Baye and Morgan (2001) offer a notable exception. As discussed in footnote 74, they focus

on a single information gatekeeper (i.e., a “monopoly platform”) that sells advertising to firms

and information to consumers. My focus below is on research that characterizes the price of

advertising when multiple media companies exist and compete for firms’ advertising messages.

149 For more on two-sided markets, see Armstrong (2004), Caillaud and Jullien (2003) and

Rochet and Tirole (2003). 139

group network externalities are present in that members of one group are directly

affected by the number from the other group that use the same platform. In the

commercial TV market, for example, the two groups that interact are consumers

and advertisers, and the platform is the broadcast company. Advertisers benefit

when the broadcast has more viewers, since those viewers represent potential

consumers for the advertised products. Thus, for a given broadcast, a positive

network externality flows from viewers to advertisers. At the same time, viewers

may regard ads as a nuisance; and if the nuisance cost outweighs any other benefit

that is associated with the ad, then a negative network externality flows from

advertisers to viewers. The broadcast company must then ensure that consumers

150

stay on board, by bundling the ads with entertainment.

Anderson and Coate (forthcoming) provide a theory of commercial broad-

151

casting and advertising that captures many of these features. In particular,

their model permits a welfare analysis concerning how well the commercial broad-

cast market fulfills its two-sided role of delivering programming to viewers and

enabling advertisers to contact potential consumers. Advertising has a social ben-

efit in their model, since it is the means through which firms inform consumers of

the existence of their respective products. But advertising also has a social cost;

namely, a viewer incurs a nuisance cost when an ad is viewed. In this general set-

ting, when a broadcaster chooses a level of advertising, it determines the number

of viewers and thereby induces a price for advertising at which firms are willing

to demand the chosen level of advertising.

The basic model has two channels, where each channel carries one program

and a program can be of two possible types (e.g., news or sports). A given viewer

can watch only one program, and viewers have different preferences over program

types. Viewers are distributed along a Hotelling line, with a viewer’s location

defining that viewer’s ideal program type; and the two possible program types

152

are located at the respective endpoints of that line. A viewer’s benefit from

viewing one of the possible program types decreases with the distance of this type

150 As noted in Section 2.4, some similar themes appear in work by Barnett (1966), Becker and

Murphy (1991) and Telser (1978).

151 See Armstrong (2004), Dukes (2004), Kind et al (2004) and Nilssen and Sorgard (2001) for

related efforts. Berry and Waldfogel (1999) provide a related empirical analysis of the radio

broadcasting market. I do not provide an extensive survey of work on advertising and the

media here. Instead, I refer the reader to Anderson and Gabszewicz (2004), who provide a

comprehensive and recent review of such work.

152 Anderson and Coate also discuss the endogenous determination of programs. I focus here

on their analysis of advertising when programming decisions are given.

140

from the viewer’s ideal type. All viewers also suffer a (common) nuisance cost

from watching ads. Ads are placed by firms with new goods and inform viewers

of the existence and nature of these goods. Firms are differentiated with regard

to their desire to advertise: one firm may offer a product that is more likely to

be satisfactory to consumers than is the product of another firm. Each firm is a

monopoly in its product market, facing consumers who each desire at most one

unit and have a common reservation value for a satisfactory product. Once a firm

advertises, it therefore prices at the reservation value, sells to those consumers

who regard the product as satisfactory, and collects all social surplus associated

with the introduction of the new good. Under the assumption that a viewer can

watch only one program, each broadcaster has monopoly control over the access

153

by firms to its viewers. When the broadcaster chooses a level of advertising for

its program, a price for advertising is induced, and those firms with products that

are more often satisfactory elect to advertise.

Anderson and Coate show that the equilibrium level of advertising is below

(above) the socially optimal level if the nuisance cost of advertising is low (high).

Intuitively, broadcasters determine the level of advertising with the objective of

maximizing advertising revenue; thus, the nuisance cost of advertising affects the

level of advertising provided by the market only insofar as a broadcaster perceives

that additional advertising would induce a marginal viewer to switch off or over

to the other program. It is particularly interesting that the market may provide

programs that have too few ads. This finding reflects two considerations. First,

broadcasters compete for viewers, and they can do so only by lowering advertising

levels. Second, for any given set of viewers, each broadcaster has a monopoly

in delivering those viewers to advertisers. A broadcaster may thus hold down

advertising levels, in order to drive up the price of advertising. A further and

related finding is that the level of advertising would be higher if the two programs

153 In terms of the literature on two-sided markets, each viewer can use only one platform (i.e.,

single-homes.

watch only one program) and thus By contrast, a firm can use both platforms

multi-home.

(i.e., advertise on both programs) and may thus As Armstrong (2004) shows, in

such a situation a “competitive bottleneck” arises: platform competition is more intense over

the party that single-homes. See also Caillaud and Jullien (2003) and Rochet and Tirole (2003).

In an extension of their advertising model, Anderson and Coate allow that viewers may be

charged subscription fees. Consistent with the literature on two-sided markets, they find that

competition often drives such fees to zero (when subsidies for viewing are infeasible; see also

footnote 19). Focusing on newspapers, Gabszewicz et al (2001) establish a similar finding. The

single-homing assumption is perhaps more natural with respect to newspapers than with TV

channels, since with the latter consumers may switch platforms more frequently.

141

were operated by a monopoly. The key intuition is that a monopoly broadcaster

does not reduce advertising levels in order to compete for viewers; instead, the

monopolist is concerned only that greater advertising might cause some viewers

to watch no program.

The basic model can be extended in a variety of directions. For example, the

assumption that advertising generates a nuisance cost is more plausible in some

media markets than in others. Rysman (2004) offers an empirical analysis of the

market for Yellow Pages directories. His estimates indicate that consumers value

advertising; thus, the nuisance cost of advertising is negative. This is consistent

with the idea that consumers visit the Yellow Pages platform to obtain information

that is embodied in ads. Similarly, Gabszewicz et al (2001) focus on newspapers,

where ads are easily avoided. They make the plausible assumption that ads do

not generate a nuisance cost in this context.

At this point, it is useful to remark on some recent trends in the advertising and

media industries. Several commentators argue that, over the past several years,

154

firms have increasingly opted for ads that target specific consumer groups. At

a broad level, the greater emphasis on targeted advertising seems to reflect two

related considerations. First, the returns from mass-audience advertising may

be lower due to an underlying fragmentation of media platforms. The commer-

cial TV platform is now a less dominant means of reaching potential consumers,

since such consumers increasingly enjoy a range of alternative media platforms,

including internet sites, cable-TV programs and specialty magazines. Second, the

relative returns from targeted advertising may be higher due to advances in digital

technology. For example, personal video recorder devices, such as TIVO, enable

consumers to rapidly skip through TV ads and thus reduce the effectiveness of

some mass-audience advertising. At the same time, internet ads that are affili-

ated with keywords on search engines better enable firms to target their ads to

interested consumers and then measure the impact of these ads.

If the reported trends are accurate, what might they suggest for future research

on advertising? First, research on targeted advertising and price discrimination

is of special importance. Several recent studies of this kind are mentioned briefly

in Section 5 (footnote 76). Second, empirical studies of the substitutability across

different media of the demand for advertising may be of particular value. For

recent work of this kind, see Fare et al (2004), Seldon et al (2000), and Silk et

al (2002). Third, the described patterns suggest a greater role for ads that offer

information. Consumers are more likely to view such ads, and relevant informa-

154 See, e.g., Bianco (2004), Delaney (2005), Lewis (2000) and The Economist (2005).

142

tion may be more easily transmitted using digital media platforms. Theoretical

work that further analyzes Nelson’s (1974b) match-products-to-buyers effect may

be especially relevant. Some recent work on this topic is described in Section

6.3. Finally, as the nature of advertising evolves, so, too, will the industry that

“produces” advertising content. Silk and Berndt (1993, 1995, 2004) study the

production of advertising and estimate the cost structure of advertising agencies.

Interesting future work might further study the on-going evolution of this industry.

10.2. Advertising, Behavioral Economics and Neuroeconomics

As discussed in Section 2, some of the early proponents of the persuasive view,

such as Braithwaite (1928) and Robinson (1933), emphasize that advertising al-

ters consumers’ tastes and creates brand loyalty. As detailed in Section 3, the

empirical implications of this view have been extensively assessed; however, much

less attention has been given to the process by which advertising distorts tastes.

According to the complementary view, for example, advertising does not change

tastes and instead enters as an argument in a stable utility function. As discussed

in Section 4, Dixit and Norman (1978) offer a sophisticated normative treatment

of persuasive advertising, but they remain somewhat agnostic as to the underlying

mechanism through which advertising shifts tastes.

Given this state of affairs, it is natural that economists would seek insights

from other disciplines. Two related approaches stand out. First, over the past two

decades, behavioral decision research in psychology has contributed to the field of

behavioral economics. Work in this field is motivated by the desire to increase the

psychological realism of economic models by imposing assumptions that are rooted

in psychological regularity. Thus, preference functions or associated behavioral

rules that have experimental support are embedded in theoretical models, in order

to achieve new theoretical insights and better predictions. Second, in recent years,

neuroscience research has used imaging of brain activity and other methods to

gather insight into the way that the brain works. This works informs the new and

emerging field of neuroeconomics, which seeks to understand economic decision

155

making at a more foundational level.

Recent work by Gabaix and Laibson (2004) illustrates the behavioral ap-

156

proach. They endow some consumers with a behavioral bias by assuming

155 For overviews of behavioral economics and neuroeconomics, respectively, see Camerer and

Loewenstein (2004) and Camerer et al (2005).

156 See also Brekke and Rege (2004) and Krahmer (2004). The former paper considers how

143

that these consumers are naïve and fail to foresee ”shrouded attributes,” such

as maintenance costs, expensive add-ons and hidden fees. For example, when a

guest checks into a hotel, the guest pays a room charge but may not fully an-

ticipate the additional expenses attributable to large markups on extra services

(parking, meals, minibar, phone, etc.). In a standard model of price competition

between firms, if price advertising were costless, firms would reveal all expenses

and compete over the total price. Information revelation may break down in the

presence of naïve consumers, however. Firms will not compete by publicly un-

dercutting their competitors’ add-on prices, even when advertising is costless, if

add-ons have close substitutes that are only exploited by sophisticated consumers

and many naïve consumers would drop out of the market altogether once the

add-on expenses were made more salient.

The model suggests some novel predictions for advertising theory. First, the

competitive pressure that is normally associated with price advertising may be

suppressed when pricing is complex and some consumers are thus naïve. Second,

in markets with naïve consumers, advertising content is more likely to shroud

negative product information. Finally, in comparison to the loss-leader literature

reviewed in Section 5.4, a new prediction is that loss-leader behavior (e.g., a low

room rate with large markups on extra services) is used by profit-maximizing

firms, even when it is costless for firms to make commitments as to add-on prices.

Recent work in neuroscience suggests that human behavior is the outcome of

an interaction between distinct neural systems. McClure et al (2004a) use func-

tional magnetic resonance imaging and report evidence that parts of the limbic

system are activated by decisions involving immediately available rewards while

regions of the prefrontal cortex are engaged by intertemporal choices. This work

provides neurological support for models in which decision makers use a hyperbolic

discounting function. More generally, as McClure et al (2004a, p. 506) explain,

recent imaging studies “suggest that human behavior is often governed by a com-

petition between lower level, automatic processes that may reflect evolutionary

adaptations to particular environments, and the more recently evolved, uniquely

human capacity for abstract, domain-general reasoning and future planning.”

The imaging studies motivate new two-system models of decision making.

Loewenstein and O’Donoghue (2004) develop a model in which decisions reflect

an interaction between a deliberative system that assesses options using a goal-

advertising may impact consumers’ assessments as to the popularity of a brand, while the latter

paper offers a formalization of Nelson’s (1974b) memory-activation role for advertising (see

footnote 13 and Section 6.2). 144

based perspective and an affective system that encompasses emotions and moti-

vational drives. Environmental stimuli might activate one or both systems. With

the exertion of willpower (cognitive effort), which is a scarce neural resource,

the deliberative system may partially override the affective system. Formally,

Loewenstein and O’Donoghue represent the decision-making process as a kind of

principal-agent model. The deliberative system (the principal) chooses behavior

to maximize its objective function subject to the constraint that it must incur

the cost of exerting the willpower that is required to get the affective system (the

157

agent) to carry out the chosen behavior. Focusing on addiction, Bernheim and

Rangel (2004) develop a related model in which the brain can operate in a “cold

mode” or a “hot mode.” At a broad level, the cold (hot) mode is analogous to the

deliberative (affective) system. In the Bernheim-Rangel model, however, at any

given point in time, either the cold mode or the hot mode is in total control. The

model is also dynamic: when an individual makes a decision in the cold mode, he

takes into account the associated probability that cues will be encountered that

trigger hot modes in the future.

In such two-system models, what is the appropriate measure of decision-maker

welfare? Loewenstein and O’Donoghue suggest that the deliberative system objec-

tive function guide welfare calculations, but they offer arguments for and against

including the cost of exerting willpower. Bernheim and Rangel, on the other

hand, unambiguously recommend that welfare be measured using cold-state pref-

erences. In their view, hot-mode decisions are cue-triggered errors that correspond

to imperfections in the process by which the brain delivers choices.

What has this to do with advertising? As Braithwaite, Robinson and other

persuasive-view advocates argued long ago, advertising is often designed to elicit

emotions and motivational drives. In other words, advertising content may be

designed to serve as an environmental cue that activates the affective/hot-mode

system. If advertising indeed plays this role, then it may be possible to use models

similar to those just described and reconsider the welfare effects of persuasive

advertising. For example, in the Dixit-Norman (1978) model reviewed in Section

4, the pre-advertising (post-advertising) demand curve may be broadly associated

with the deliberative system or the cold mode (the affective system or the hot

mode). These models may also give rise to new rationales for bans on advertising

of addictive products. More generally, as further advances are achieved in the

analysis of two-system models, important new tools may be created for positive

and normative analyses of advertising.

157 For further discussion of self-control and willpower, see Benabou and Tirole (2004).

145

Neurological studies may also provide insight into the elusive concept of brand

loyalty. McClure et al (2004b) offer a first study of this kind. In a blind taste

test, they find that subjects split equally in their preferences for Coke and Pepsi.

When one cup was labeled “Coke,” however, individuals showed a significant

bias for the labeled cup (even though the unlabeled cup also contained Coke);

further, when the subjects were informed that they were drinking Coke, brain

regions associated with memory were activated. By contrast, brand knowledge

of Pepsi did not have similar effects on choice or brain activity. This study gives

striking neurophysiological evidence that is consistent with the hypothesis that

some consumers exhibit brand loyalty toward Coke. The full implications of this

study are not yet clear; however, it does at least raise the possibility that future

neurological studies may provide important and novel insight as to when and how

advertising may instill brand loyalty.

10.3. Other Topics

Advertising is a huge research area, with key contributions from various disciplines

including economics, marketing, psychology, neuroscience and political science.

Clearly, it is not possible to summarize all of this work in one survey. Here, I

simply mention a few omitted topics and offer suggestions for further reading.

First, I largely ignore the literature that considers the economic consequences

of laws against deceptive advertising. Pitofsky (1978) describes the rationale be-

hind the government regulation of truth-in-advertising. Sauer and Leffler (1990)

provide an empirical assessment of the implications of such regulation for adver-

tising content. In a recent effort, Barigozzi et al (2002) show that laws concerning

the veracity of comparative advertisements can enhance the signaling potential of

advertising. Second, I ignore many aspects of advertising that are emphasized in

other social sciences. Advertising plays an important role in political contests, for

example. For recent work of this kind, see Coate (2004) and Prat (2002). Finally,

the success of a given ad depends in part on the associations that it triggers in

consumers’ minds and thus hinges on specific psychological considerations that

are not considered here. For research of this kind, see Kardes (2002).

10.4. Summary

I discuss in this section two new directions for advertising research. First, recent

work returns to a long-standing research topic and analyzes the role of advertising

in media markets. This work highlights the two-sided nature of the media mar-

146


PAGINE

180

PESO

1.22 MB

AUTORE

Atreyu

PUBBLICATO

+1 anno fa


DESCRIZIONE DISPENSA

Materiale didattico per il corso di Impresa, management e mercati della prof.ssa Ornella Tarola. Trattasi del testo di Kyle Bagwell dal titolo "The Economic Analysis of Advertising", avente ad oggetto i seguenti argomenti: le componenti della pubblicità, gli effetti sulle vendite e sulla fidelizzazione del cliente, pubblicità come economia di scala, la "monopoly advertising", il rapporto pubblicità - prezzo e pubblicità - qualità, nuove forme di pubblicità e marketing.


DETTAGLI
Corso di laurea: Corso di laurea magistrale in scienze delle pubbliche amministrazioni
SSD:
A.A.: 2011-2012

I contenuti di questa pagina costituiscono rielaborazioni personali del Publisher Atreyu di informazioni apprese con la frequenza delle lezioni di Impresa management e mercati e studio autonomo di eventuali libri di riferimento in preparazione dell'esame finale o della tesi. Non devono intendersi come materiale ufficiale dell'università La Sapienza - Uniroma1 o del prof Tarola Ornella.

Acquista con carta o conto PayPal

Scarica il file tutte le volte che vuoi

Paga con un conto PayPal per usufruire della garanzia Soddisfatto o rimborsato

Recensioni
Ti è piaciuto questo appunto? Valutalo!

Altri appunti di Impresa management e mercati

Static Games and Cournot Competition
Dispensa
Qualità e varietà del prodotto in regime di monopolio
Dispensa
Price Competition
Dispensa
Bilancio di esercizio
Dispensa