Você está na página 1de 20

JOURNAL OF OPTIMIZATION THEORY AND APPLICATIONS: Vol. 119, No. 2, pp.

261–280, November 2003 ( 2003)

Dynamic Advertising under Vertical


Product Differentiation1
L. COLOMBO2 AND L. LAMBERTINI3

Communicated by G. Leitmann

Abstract. We investigate a dynamic advertising model where product


quality is endogenous. In the differential game between single-product
firms, there exists a parameter range where the low-quality firm uses a
more efficient advertising technology and earns higher profits than the
rival. Moreover, we show that equilibrium qualities are the same under
duopoly, multiproduct monopoly, and social planning, the only distor-
tion being concerned with the output levels.

Key Words. Advertising, product quality, differential games, optimal


control.

1. Introduction

There exists a wide literature dealing with dynamic advertising either


in monopoly model or in oligopoly model; see Refs. 1–3 and Ref. 4 (Chap-
ter 11). To the best of our knowledge, the endogenous interplay between
advertising and product quality has not received a large amount of attention
so far. The few existing contributions in this field consider usually the opti-
mal control of advertising efforts and product quality in relation with build-
ing up the stock of goodwill, under incomplete consumer information.4 One
interesting exception dealing with a full information model of dynamic
advertising and product quality is in Ref. 8, extending the well-known
Lanchester model to account for the interplay between market shares and
quality.5 Others have investigated, adopting either static or dynamic
1
The authors thank Gustav Feichtinger for an insightful discussion that inspired this paper,
Roberto Cellini, and the seminar audience in Bologna for useful comments.
2
Research Fellow, Department of Economics, University of Bologna, Bologna, Italy.
3
Professor of Economics, Department of Economics, University of Bologna, Bologna, Italy.
4
See Refs. 5–7, Ref. 3, and the references therein.
5
For the formulation of the Lanchester model, see Ref. 2, Ref. 4 (Chapter 11), and Refs. 9–10.
261
0022-3239兾03兾1100-0261兾0  2003 Plenum Publishing Corporation
262 JOTA: VOL. 119, NO. 2, NOVEMBER 2003

approaches, the strategic use of product qualities as the firms instruments


to build up market shares (Refs. 11–13).
We study persuasive advertising in a dynamic market where fully
informed consumers may choose between two goods characterized by differ-
ent quality levels which are determined endogenously. We consider (i) the
differential game between two single-product firms, (ii) the optimal control
problem faced by a monopolist supplying both varieties so as to maximize
profits, and (iii) the optimal control problem solved by a benevolent planner
supplying both varieties so as to maximize social welfare.
Our main results can be summarized as follows. In the duopoly game,
there exist parameter ranges wherein the low-quality firm earns higher pro-
fits than the high-quality firm, due to the fact that the number of consumers
choosing the low-quality good is considerably larger than the number of
consumers buying the high-quality good. This is in contrast with the
acquired wisdom coming from the existing static games describing quality
competition in oligopoly (Refs. 14–18, inter alia), where the possible interac-
tion between quality and advertising is disregarded completely and the equi-
librium market share of the high-quality good is always larger than the that
of the low-quality good. While in the static literature on this issue it is
always true that supplying a superior quality enables the firm (i) to extract
a higher profit margin from each unit and (ii) to obtain a larger market
share than that of the rivals, here we show that being able to attract the
richest set of costumers is insufficient to ensure higher profits because
inferior qualities may serve larger market shares. This appears to be
realistic, e.g., if one considers that selling a luxury motorbike like the Ducati
999 may well entail a larger unit price-cost margin than selling a Honda
Hornet, but the number of customers that can afford a Hornet is surely
much larger than the number of customers who may afford a Ducati 999.
Moreover, in our model, the quality levels turn out to be independent
of the market regime; i.e., they are the same at the duopoly, monopoly, and
planning equilibria. Again, this finds no correspondence in the static litera-
ture on vertical product differentiation, where it emerges usually that the
degree of vertical differentiation at the duopoly equilibrium is larger than
the monopoly optimum and the social optimum, due to the incentive for
firms to increase product differentiation so as to soften price competition
(Refs. 16 and 19). In addition, our result also tells that there is no quality
distortion at the monopoly equilibrium, as compared to the first best. While
in static models we observe the incentive for the monopolist to undersupply
or oversupply product quality in order to induce self-discrimination across
consumers (Refs. 19–21), here we observe only a downward distortion in
output levels due to monopoly power.
JOTA: VOL. 119, NO. 2, NOVEMBER 2003 263

The remainder of the paper is structured as follows. The basic setup is


laid out in Section 2. The duopoly setting is investigated in Section 3, while
Section 4 contains the analysis of the monopoly and first best equilibria.
Concluding remarks are given in Section 5.

2. Model

Traditionally, models dealing with vertically differentiated duopoly


postulate the existence of a consumers gross surplus function which is
assumed to be nonseparable in its arguments. Instead, we assume that all
the arguments of the utility function are separable additively. Consumers
are indexed by a marginal willingness to pay θ ∈[0, θ 1 ], with density equal
to one, so that the total mass of consumers is θ 1. The distribution of con-
sumers and the support are invariant w.r.t. time. The market exists for
t∈[0,S), with the time being considered as continuous. At each t, the mar-
ket is supplied by two single-product firms offering goods of quality qi (t),
with iG{H, L}, qH (t)„qL (t)„0, which are perfectly observable, together
with the price vector, by consumers before purchase. Production entails an
instantaneous variable cost of quality improvement, which is assumed to be
convex in the current quality level,
Ci (t)Gci xi (t)[qi (t)]2, (1)
with constant parameters 0FcH FcL indicating that the high quality firm is
more efficient in the production of quality.
From the consumption of the high quality good, a consumer of type θ
draws the following net surplus:
UH GθCqH (t)ApH (t), (2)
where pH (t) is the market price of variety H at time t. Similarly, from the
consumption of the low quality good, a consumer indexed by θ draws the
following net surplus:
UL GsθCqL (t)ApL (t), (3)
where pL (t) is the market price of variety L at time t and s∈(0, 1) is a
positive and time-invariant parameter capturing the idea that gross satisfac-
tion of such consumer from buying the low quality good is lower. If a
consumer does not buy either variety, the resulting utility is nil. In order to
obtain the expressions of market demands, we compute the threshold of θ
which characterizes the consumer who is indifferent between buying from
the high quality firm and buying from the low quality firm,
θ̂ (t)G[ pH (t)ApL (t)AqH (t)CqL (t)]兾(1As), (4)
264 JOTA: VOL. 119, NO. 2, NOVEMBER 2003

and the analogous threshold of θ characterizing the consumer who is indif-


ferent between buying from the low quality firm and not buying at all,
θ̃ (t)G[ pL (t)AqL (t)]兾s. (5)
The direct demand system follows the relations
xH (t)Gθ 1Aθ̂ (t), (6)
xL (t)Gθ̂ (t)Aθ̃ (t), (7)
provided that θ̃ (t)H0, ensuring that we are in the nondegenerate case where
partial market coverage prevails. If so, we can write the inverse demand
system,
pH (t)Gθ 1CqH (t)AxH (t)AsxL (t), (8)
pL (t)GqL (t)Cs[θ 1AxH (t)AxL (t)]. (9)
Observe that the relations (4)–(9) define correctly the demand system if and
only if θ̂ (t)„ θ̃ (t), which amounts to requiring that
s[ pH (t)AqH (t)]„pL (t)AqL (t).
In the remainder of the paper, we will check that this condition indeed holds
at the duopoly steady-state equilibrium. The economic interpretation of this
requirement is that one should exclude the case of quality leapfrogging in
either direction, i.e., any situation such that the firm endowed with a techno-
logical efficiency parameter cH [respectively, cL] decides to produce a lower
[resp., higher] quality than the firm using the less [respectively, more]
efficient technology.
The instantaneous profits are written as follows:
π i (t)G[ pi (t)Aci [qi (t)]2]xi (t)Abi [ai (t)]2, (10)
2
where bi [ai (t)] is the instantaneous cost of investing in advertising, ai (t)
being the advertising effort of firm i at time t. We assume that the sales
evolve over time in response to the advertising investments, according to
the following kinematic equation (as in Ref. 22):
∂xi (t)兾∂t ≡ xi Gai (t)Aδ xi (t), iGH, L, (11)
where δ denotes the depreciation (disaffection) rate, constant over time and
common to both firms. It is worth stressing that this advertising technology
has asymmetric effects on the market demand of the two firms. The reason
is that the low-quality firm may increase demand by attracting consumers
in the lower part of the preference spectrum [i.e., pushing θ̃ (t) down] as well
as in the intermediate range of preferences [i.e., pushing θ̂ (t) up], while the
high-quality firm can fight only against the rival for θ̂ (t), since the upper
JOTA: VOL. 119, NO. 2, NOVEMBER 2003 265

bound of the willingness to pay θ 1 is given. This preludes to the possibility


for the low-quality firm to perform better than the high-quality firm in
equilibrium.
Firm i aims at maximizing the discounted profit flow,
S

Πi (t)G 冮0
π i (t) e −pt dt, (12)

w.r.t. the controls qi (t) and ai (t), under the constraint given by the state
dynamics (11).
For future reference, we define also the consumer surplus,
CS(t)GCSL (t)CCSH (t), (13)
where
θ̂ (t)

CSL (t)G 冮θ̃ (t)


(szCqL (t)ApL (t)) dz

Gs[xL (t)]2兾2, (14)


θ1

CSH (t)G 冮θ̂ (t)


(zCqH (t)ApH (t)) dz

G[xH (t)兾2] [xH (t)C2sxL (t)]. (15)


If we disregard the issue of surplus distribution between consumers and
producers, and we confine our attention to Pareto efficiency, we can define
welfare as follows:
W(t)GCSH (t)CCSL (t)Cπ H (t)Cπ L (t). (16)

3. Duopoly Equilibrium

The current value Hamiltonian function of firm i is

H i (t)Ge −ρt[π i (t)Cλ ii (t)ẋiCλ ij (t)ẋ j ]. (17)


The first-order conditions (FOCs) on the controls are (henceforth, we omit
the indication of time for brevity)6

∂H i 兾∂qi G0 ⇒ xi (1A2ci qi)G0 ⇒ qi G1兾2ci , (18)

∂H i 兾∂ai G0 ⇒ A2ai biCλ ii G0 ⇒ λ ii G2ai bi ⇒ λ̇ ii G2ȧi bi . (19)

6
Second-order conditions are always met throughout the paper. They are omitted for brevity.
266 JOTA: VOL. 119, NO. 2, NOVEMBER 2003

The above FOCs entail that the present game is a linear state game, produc-
ing subgame perfect or Markov perfect open-loop Nash equilibria.7 Notice
also that conditions (18)–(19) do not contain λ ij because the present game
features separated dynamics.8 Therefore, the problem admits the solution
λ ij G0 for all t∈[0,S) and j ≠ i. Accordingly, we specify only one costate
equation per firm,

∂H H 兾∂xH G−δλ HHCθ 1CqHAcH q2HA2xHAsxL


Gρλ HHAλ̇ HH , (20)

∂H L 兾∂xL G−δλ LLCqLAcL q2LCs(θ 1AxHA2xL)


Gρλ LLAλ̇ LL , (21)

along with the transversality condition

lim µi (t)xi (t)G0 (22)


t→S

and the initial conditions xi (0)H0.


Now, using (19) and the costate equations, we write

ȧH Gλ̇ HH 兾2bH G−1兾4cHAθ 1C2aH bH (δCρ)C2xHCsxL , (23)


ȧL Gλ̇ LL 兾2bL G−1兾4cLC2aL bL (δCρ)Cs(−θ 1CxHC2xL). (24)

The steady-state equilibrium requires ȧH G0, ȧL G0, yielding

aH G[1C4cH (θ 1A2xHAsxL)]兾8bH cH (δCρ), (25)


aL G[1C4cL s (θ 1AxHA2xL)]兾8bL cL (δCρ), (26)

which can be plugged into the system of state equations (11), simplifying as
follows:

ẋH G[1C4cH (θ 1A2xHAsxL)]兾8bH cH (δCρ)Aδ xH , (27)


ẋL G[1C4cL s(θ 1AxHA2xL)]兾8bL cL (δCρ)Aδ xL . (28)

7
See e.g. Refs. 23–24. For an exhaustive exposition of linear state games, see Ref. 4 (Chapter 7).
8
This also entails that rewriting the model over a finite-time horizon with an appropriate scrap
value, this problem could be solved also with the alternative coordinate transformation method
of Ref. 25.
JOTA: VOL. 119, NO. 2, NOVEMBER 2003 267

Fig. 1. Parameter space.

We are interested in investigating the dynamics of the system in the positive


quadrant of the space {xi , ai }, which is described in Fig. 1. The system
{ẋi G0, ȧi G0} yields the following steady-state sales:
2bL cL δ (1C4cH θ 1)(δCρ)As{−2cLCcH [1C4cL θ 1 (sA2)]}
xSS
H G , (29)
4cH cL (4bL δ (δCρ){1CbH δ (δCρ)C[4C4bH δ (δCρ)As]s}
−cL sC2cH [1C2cL θ 1 sCbH δ (δCρ)(1C4cL θ 1 s)]
xSS
L G . (30)
4cH cL (4bL δ (δCρ){1CbH δ (δCρ)C[4C4bH δ (δCρ)As]s}
These expressions can be plugged into (25)–(26) to yield the optimal invest-
ments in advertising, aSS
i . The following proposition holds.

Proposition 3.1. The steady state defined by {aSS SS


i , xi } is a saddle
point.

Proof. See the Appendix (Section 6).

The above output levels are acceptable if and only if they are nonnega-
tive. Of course, this depends upon the relative size of the demand and cost
parameters {s, θ 1 , bi , ci }. First, note that the denominator of xSS
i is positive
for all admissible values of parameters. Since we are treating the issue of
enlarging market shares through advertising, it is interesting to examine the
nonnegativity of the outputs in terms of the parameters bi and their interac-
tion with production cost parameters ci. We can prove the following lemma.

Lemma 3.1. The equilibrium output of the high-quality firm xSS


H is
always positive.

Proof. Examine the numerator of (29). The first term,


2bL cL δ (1C4cH θ 1)(δCρ),
268 JOTA: VOL. 119, NO. 2, NOVEMBER 2003

is clearly positive. Then, it can be quickly established that


−s{−2cLCcH [1C4cL θ 1 (sA2)]}H0
as well. To see this, rewrite it as
s[2cLAcHC4cH cL θ 1 (2As)],
which is positive for all s∈(0, 1) and cH FcL . 䊐

Now, we examine the output of the low-quality firm.

Lemma 3.2. If
cH ∈(cL s兾2(1C2cL θ 1 s), cL),
then xSS
L H0 for all bH H0. If instead,

cH ∈(0, cL s兾2(1C2cL θ 1 s)],


SS
then x H0 for all
L

scLA2cH (1C2cL θ 1 s)
bH Hb̄H ≡ „0.
2δ cH ( ρCδ )(1C4θ 1 cL s)
L G0 for all bH ∈[0, b̄H ].
Otherwise, xSS

Proof. In general, xSS


L H0 for all

scLA2cH (1C2cL θ 1 s)
bH Hb̄H ≡ .
2δ cH ( ρCδ )(1C4θ 1 cL s)
This is surely true for all admissible bH if
cH ∈(cL s兾2(1C2cL θ 1 s), cL),
since this implies b̄H F0. In the case where
cH ∈(0, cL s兾2(1C2cL θ 1 s)],
then b̄H „0. Therefore, in this range of cH, we have that xSS
L „0 iff
bH „b̄H . This concludes the proof. 䊐

Now, on the basis of Lemma 3.2, one can also check that, in equilib-
L ≡ θ̂ (t)Aθ̃ (t) ensures also that θ̂ (t)„ θ̃ (t).
rium, the nonnegativity of xSS
Lemmas 3.1 and 3.2 imply directly a relevant corollary.

Corollary 3.1. If cH ∈(0, cL s兾2(1C2cL θ 1 s)], then the high-quality firm


is monopolist for all bH ∈[0, b̄H ].
JOTA: VOL. 119, NO. 2, NOVEMBER 2003 269

When the efficiency of the advertising technology of firm H is very high


(or conversely, the cost of advertising is sufficiently low), there is no room
in the market for the inferior variety.
For (29)–(30) to be acceptable, it must be also that
θ 1 HxSS SS
H CxL ;

i.e., indeed, firms cover the market only partially.

Lemma 3.3. θ 1 HxSS SS


H CxL for all bHHmax{b
¡ H , 0}, where
cL [sC2bL δ (δCρ)]CcH [2A8bL cL δθ 1 (δCρ)As(1C4cL θ 1)]
b¡ H ≡ ,
2cH δ (δCρ){4cL θ 1 [2bL δ (δCρ)Cs]A1}
with ∂b¡ H 兾∂bL F0 always.

Proof. The derivation of b¡ H is straightforward. Then, observe that


∂b¡ H cL [1C4cH θ 1 (1As)]
G− F0. 䊐
∂bL cH {4cL θ 1 [2bL δ (δCρ)Cs]A1}2

The above result entails that, intuitively, partial market coverage is


easier to obtain in equilibrium, the higher the level of both the advertising
cost parameters. Now define
2bL cL δ (δ + ρ)+3cL sAcH {2+s+4cL θ 1 [−2bL δ (δ + ρ)+(−1+s)s]}
b̂H ≡ (31)
2cH δ (δ + ρ)(1+4cL θ 1 s)
as the level of bH such that xSS SS SS SS
H GxL . Assessing xH AxL , one can prove
9
easily the following lemma.

Lemma 3.4. If b̂H H0, then xSS H HxL for all bH ∈(0, b̂H); xH FxL for
SS SS SS

SS SS
all bH Hb̂H . If instead b̂H ‚0, then xL HxH .

The foregoing discussion (in particular, Corollary 3.1 and Lemma 3.4)
opens the possibility that there exist admissible parameter constellations
such that both firms enjoy positive market shares and profits in equilibrium,
with the low-quality firm serving more consumers and earning higher profits
than the high-quality firm.

9
As we already know, for all s∈(0, 1), the denominator of xiSS is always positive. Therefore,
SS
the value of bH such that xH AxLSS G0 is unique and it is given by b̂H .
270 JOTA: VOL. 119, NO. 2, NOVEMBER 2003

We now turn to the comparative statics on the steady-state level of


i 兾∂c j H0, i ≠ j, can be inter-
sales w.r.t. all parameters. First, the property ∂xSS
preted easily. All else being equal, as firm j becomes less efficient in supply-
ing quality (i.e., qj decreases as cj increases), the product of firm j becomes
less appealing and some customers switch to firm i. With regard to an
increase in the own cost of quality improvement, we have the expected
property
∂xSS
i 兾∂ci F0.

Not surprisingly,
i 兾∂θ 1 H0.
∂xSS
That is, as the market becomes more affluent, demand for all varieties
increases.
Now consider the effects of a change in the efficiency of the advertising
investments on market shares. First of all, notice that
H 兾∂bH F0,
∂xSS ∂xSS
L 兾∂bH H0, (32)
in the whole admissible parameter range. This is what one would expect
from the outset: as the advertising campaign of firm H becomes less expen-
sive (or more effective), the demand for the high-quality good is enhanced
while that for the inferior variety shrinks.
The effect of a change in bL on xSSi is more involved,
H 兾∂bL H0 and ∂xL 兾∂bL F0,
∂xSS SS
for all cH Hc̃H , (33)
scL
c̃H G (34)
2[1C2cL θ 1 sCbH δ (1C4cL θ 1 s)(δCρ)]
and conversely for all cH ∈(0, c̃H ]. This means that, when vertical product
differentiation is relatively low (which holds when cH Hc̃H ), any increase in
bL drives some consumers to switch from the inferior variety to the high-
quality good. This does not happen if the degree of differentiation is large
enough.
We are now in a position to assess the firms relative performance in
terms of steady-state profits,
[1CbH δ (δC2ρ)]Ω2
π SS
H G , (35)
16c2H c2L [4bL δ (δCρ)[1CbH δ (δCρ)]C[4C4bH δ (δCρ)As]s]2
Ω ≡ cH [1C4cL θ 1 (−2Cs)]sA2bL cL δ (1C4cH θ 1)(δCρ)A2cL , (36)
[bL δ (δC2ρ)Cs]Ψ2
π SS
L G , (37)
16c2H c2L [4bL δ (δCρ)[1CbH δ (δCρ)]C[4C4bH δ (δCρ)As]s]2
Ψ ≡ cL sA2cH [1C2cL θ 1 sCbH δ (δCρ)(1C4cL θ 1 s)]. (38)
JOTA: VOL. 119, NO. 2, NOVEMBER 2003 271

By evaluating the sign of the difference between π SS


H and π L , we obtain a
SS

critical value b̃H at which π H Gπ L entailing the following lemma.10


SS SS

Lemma 3.5. If b̃H H0, then π SS H Hπ L for all bH ∈(0, b̃H); π H Fπ L for
SS SS SS

all bH Hb̃H . If instead b̃H ‚0, then π L Hπ H .


SS SS

Moreover, comparing b̂H against b̄H , we obtain the following lemma.

Lemma 3.6. b̂H Hb̄H always.

Proof. It suffices to observe that


b̂HAb̄H T2bL cL δ (1C4cH θ 1)(δCρ)As{−2cLCcH [1C4cL θ 1 (sA2)]}
with the expression of the r.h.s. coinciding with the numerator of (29), that
is always positive as we already know from Lemma 3.1.11 䊐

Lemmas 3.4 to 3.6 yield a partition of the parameter space into three
regions, according to the value assumed by the advertising efficiency param-
eter of high-quality firm, bH. This partition is illustrated in Fig. 1, assuming
that the parameter set is such that
b̃H Hb̂H H0Hmax{b¡ H , b̄H }.
This means that we exclude the trivial case where the market is a monopoly
for firm H.

Proposition 3.2. Take values of {s, θ 1 , ci } such that b̃H Hb̂H H0. If
H HxL , aH HaL , and π H Hπ L . If bH ∈(b̂H , b̃H), then
bH ∈(0, b̂H), then xSS SS SS SS SS SS

xH FxL , aH FaL , and π H Hπ L . If bH Hb̃H , then xSS


SS SS SS SS SS SS SS SS
H FxL , aH FaL
SS

and π H Fπ L .
SS SS

The above proposition illustrates a situation where the parameter


values are such that there exists an admissible range wherein both firms are
active and earn positive profits in steady state. Moreover, in contrast with
the conclusions commonly drawn from the static approach to vertical differ-
entiation in oligopoly, in one subset of the parameters the low-quality firm
performs better than the high-quality firm in terms of both market share
and equilibrium profits. This fact can be interpreted as follows. First of all,

10
The equation π H
SS
Gπ LSS yields two roots. While the smaller is always negative, the larger may
or may not be positive. The expressions are omitted for brevity.
11
One such numerical example is in the Appendix (Section 6).
272 JOTA: VOL. 119, NO. 2, NOVEMBER 2003

consider that the critical levels b̃H and b̂H are a function of bL. Then, exam-
ine the case bH ∈(0, b̂H). Here, the advertising technology of firm H is suffic-
iently efficient, as compared to that of firm L, to imply that firm H acquires
a dominant position in the market. If instead bH ∈(b̂H , b̃H), the relative
decrease in the efficiency of its advertising campaign induces firm H to invest
less in advertising, which in turn entails that its equilibrium demand
becomes lower than that of firm L. However, this is not yet sufficient to
reverse the inequality on profits because the reduction in the market share
is still more than offset by the quality differential. In the third range, where
bH Hb̃H , the distribution of consumers between firms induces a reversal of
the profit ranking. All this amounts to saying that having a larger market
share in equilibrium is not sufficient to earn higher profits than the rival.
Observe that, in line of principle, b̂H Hb̃H is admissible. However, this
case would not allow for a sensible interpretation in terms of the underlying
economics. In particular, if that were indeed the case, one would observe a
nonmonotone behavior of the relative profits of firms. In particular, as the
advertising technology of firm H becomes less efficient, we would obtain
that initially π SSH Hπ L , then π H Fπ L
SS SS SS
for bH ∈(b̃H , b̂H), then again
π H Hπ L for bH Hb̂H . For this reason, we can dismiss such a case.
SS SS

4. Monopoly and Social Planning

Now, we examine the regimes where, both plants are controlled,


respectively, by (i) a profit-seeking monopolist and by (ii) a benevolent plan-
ner maximizing social welfare. In the monopoly regime (M), the current
value Hamiltonian function turns out to be

H M Ge −ρt (π HCπ LCλ H ẋHCλ L ẋL), (39)

while under social planning (SP), we have

H SP Ge −ρt (WCλ H ẋHCλ L ẋL), (40)

where the instantaneous social welfare function W is defined as in (16).

The first relevant result is stated in the following proposition.

Proposition 4.1. Under both monopoly and social planning, the opti-
mal qualities are the same as under the profit-seeking duopoly (D):
JOTA: VOL. 119, NO. 2, NOVEMBER 2003 273

qki G1兾2ci , iGH, L and kGD, M, SP.

Proof. To show this, it suffices to observe that, irrespective of the


property structure of the two plants (i.e., of the regime k), the first-order
condition on the Hamiltonian w.r.t. quality qi is

∂H k兾∂qi G∂H D
i 兾∂qi G∂π i 兾∂qi G0,
since the consumer surplus is independent of quality levels. 䊐

Another obvious result needs no proof. This regards the profit perform-
ance of the industry in the three regimes at stake. Clearly, the highest attain-
able profits in equilibrium accrue to the monopolist selling both varieties,
that, by definition, must perform better than the sum of two independent
single-good duopolists. Then, it is also straightforward that the profit-seek-
ing duopolists earn strictly larger profits than a social planner taking care
of consumer surplus into his objective function.
Concerning the performance of the monopolist selling both varieties,
we can prove the following proposition.

Proposition 4.2. Under monopoly, the optimal advertising efforts for


generic output levels are

1C4cH (θ 1A2xHA2sxL)
aM
HG ,
8bH cH (δCρ)
1C4cL s(θ 1A2xHA2xL)
aM
L G ,
8bL cL (δCρ)

while the steady-state output levels are

bL cL δ (1C4cH θ 1)(δCρ)Cs[cLCcH (−1C4cL θ 1A4cL sθ 1)]


xM
HG ,
8cH cL (bL δ (δCρ){1CbH δ (δCρ)C[1CbH δ (δCρ)As]s}
−cL sCcH (1CbH δ (δCρ)(1C4cL θ 1 s))
xM
L G .
8cH cL (bL δ (δCρ){1CbH δ (δCρ)C[1CbH δ (δCρ)As]s}

The monopoly equilibrium is a saddle point.

Proof. See the Appendix (Section 6).

If a planner controls both firms, we obtain the following proposition.


274 JOTA: VOL. 119, NO. 2, NOVEMBER 2003

Proposition 4.3. Under social planning, the optimal advertising efforts


for generic output levels are
1C4cH (θ 1AxHAsxL)
aSP
H G ,
8bH cH (δCρ)
1C4cL s(θ 1AxHAxL)
aSP
L G ,
8bL cL (δCρ)
while the steady-state output levels are
2bL cL δ (1C4cH θ 1)(δCρ)Cs[cLCcH (−1C4cL θ 1A4cL sθ 1)]
xSP
H G ,
4cH cL (2bL δ (δCρ){1C2bH δ (δCρ)C[1C2bH δ (δCρ)As]s}
−cL sCcH [1C2bH δ (δCρ)(1C4cL θ 1 s)]
xSP
L G .
4cH cL (2bL δ (δCρ){1C2bH δ (δCρ)C[1C2bH δ (δCρ)As]s}
The social optimum is a saddle point.

Proof. See the Appendix (Section 6).

One last remark is in order, concerning the different incentives to invest


in advertising in the three regimes. The state dynamics (11) (the same across
regimes) and the dynamics of advertising efforts ai (changing across regimes)
are drawn in Fig. 2. Without further proof, the properties of this graph,

Fig. 2. Comparisons.
JOTA: VOL. 119, NO. 2, NOVEMBER 2003 275

together with the appropriate expressions of optimal advertising efforts in


the three regimes, produce the following proposition.

Proposition 4.4. For a given xj it follows that aiSP HaiD HaM


i implying
that xiSP HxD M
i Hxi .

As the planner aims at maximizing the sum of profits and consumer


surplus, and as qualities are the same across regimes, then clearly the social
incentive to invest in advertising is driven by the fact that increasing the
extent of market coverage amounts to increasing welfare. This explains also
that, in both profit-seeking regimes, which by definition do not take into
account the consumer surplus, the advertising expenditure must be lower
than the socially desirable investment. Moreover, if each variety is supplied
by an independent firm, it surely invests more to advertise its product than
a monopolist would do for the same variety. The reason is that a single-
product duopolist wants to steal costumers from the demand basin of the
rival, which of course cannot be the case for a multiproduct monopolist.

5. Concluding Remarks

We have proposed a differential duopoly game where each firm may


invest both in product quality and advertising campaigns. We have assumed
that sales evolve over time in response of advertising investments, while
product quality improvements do not require any capital accumulation to
take place.
Contrary to the results we are familiar with from static analyses, we
have shown that there exists a range of parameters in which the low-quality
firm gains a higher market share as well as higher profits than the rival. The
reason behind this result lies in the relative efficiency of the advertising
technologies used by the two firms.
Moreover, we have shown that the level of product quality provided
by a duopoly corresponds to the one which is socially desirable. This is true
also under a monopolistic regime. The unique distortions which arise at
the privately optimal equilibria are on the output side. These downward
distortions due to market power are induced obviously by pricing above
marginal cost and therefore are qualitatively the same as in the static
literature.

6. Appendix: Proofs

Proof of Proposition 3.1. We consider first the system composed by


(11) in combination with the appropriate kinematics of the control variable
276 JOTA: VOL. 119, NO. 2, NOVEMBER 2003

ai , that is, (23) or (24), alternatively. In the two cases, that can be treated
in isolation because of the separated dynamics assumed in the model, the
system can be written in matrix form as follows:

−δ
冤ȧ 冥 G 冤 2 冥冤 冥 冤 冥
ẋH 1 xH 0
C ,
H 2bH (δCρ) aH sxLA1兾(4cH)Aθ 1
−δ
冤ȧ 冥 G 冤 2s 冥 冤a 冥C 冤sx A1兾(4c )Asθ 冥 .
ẋL 1 xL 0
L 2bL (δCρ) L H L 1

Since the determinants of the above 2B2 matrices are both negative, the
equilibria that we have obtained are two saddles. From the phase diagram
(Fig. 3), it is clear that these can be approached only along the north-west
arm of the saddle path. 䊐

Fig. 3. Phase diagram.

Numerical Duopoly Example. Take the following numerical values of


the parameters:
θ 1 G1, bL G1兾10, cH G7兾10, (41a)
cL G3兾4, sG9兾10, δ Gρ G1兾20. (41b)
If so, then
π SS
L Hπ H ,
SS
for all bH Hb̃H ⯝22.105.
JOTA: VOL. 119, NO. 2, NOVEMBER 2003 277

Setting bH G22.3, the numerical values of the equilibrium variables are


π SS
H ⯝0.2037, π SS
L ⯝0.2041, H ⯝0.4178,
xSS xSS
L G0.4760. (42)

Proof of Proposition 4.2. The first-order conditions relative to the


Hamiltonian (39) are
∂H 兾∂qi G0 ⇒ xiA2ci qi xi G0 ⇒ qi G1兾2ci ,
M
(43)

∂H 兾∂ai G0 ⇒ A2ai biCλ i G0 ⇒ λ i G2ai bi ⇒ λ̇ i G2 ȧi bi ,


M
(44)

∂H 兾∂xH G−δλ HCθ 1CqHAcH q2HA2xHA2sxL Gρλ HAλ̇ H ,


M
(45)

∂H 兾∂xL G−δλ LCqLAcL q2LCs (θ 1A2xHA2xL)Gρλ LAλ̇ L ,


M
(46)
along with the same initial and transversality conditions as in the duopoly.
From (45) and (46), we have
λ̇ H G(δCρ)λ HAθ 1AqHCcH q2HC2xHC2sxL , (47)
λ̇ L G(δCρ)λ LAqLCcL q As (θ 1A2xHA2xL),
2
L (48)
while from (44),
ȧH Gλ̇ H 兾2bH G−1兾4cHAθ 1C2aH bH (δCρ)C2xHC2sxL , (49)
ȧL Gλ̇ L 兾2bL G−1兾4cLC2aL bL (δCρ)Cs (−θ 1C2xHC2xL). (50)
The steady-state equilibrium requires {ȧH G0, ȧL G0}
1C4cH (θ 1A2xHA2sxL)
aM
HG , (51a)
8bH cH (δCρ)
1C4cL s(θ 1A2xHA2xL)
aM
L G . (51b)
8bL cL (δCρ)
Then, simplifying the state kinematics yields
1C4cH (θ 1A2xHA2sxL)
ẋH G , −dxH , (52)
8bH cH (δCρ)Aδ xH
1C4cL s(θ 1A2xHA2xL)
ẋL G , −dxL . (53)
8bL cL (δCρ)Aδ xL
At the resulting unique steady state, the sales are
bL cL δ (1C4cH θ 1)(δCρ)Cs[cLCcH (−1C4cL θ 1A4cL sθ 1)]
xSS
H G , (54)
8cH cL (bL δ (δCρ){1CbH δ (δCρ)C[1CbH δ (δCρ)As]s}
−cL sCcH [1CbH δ (δCρ)(1C4cL θ 1 s)]
xSS
L G . (55)
8cH cL (bL δ (δCρ){1CbH δ (δCρ)C[1CbH δ (δCρ)As]s}
278 JOTA: VOL. 119, NO. 2, NOVEMBER 2003

The dynamic system can be written in the matrix form


−δ

冤冥冤 冥冤 冥 冤 冥
ẋH 1 0 0 xH 0
ȧH 2 2bH (δCρ) 2s 0 aH −1兾4cHAθ 1
G C .
ẋL 0 0 −δ 1 xL 0
ȧL 2s 0 2s 2bL (δCρ) aL −1兾4cLAsθ 1
By computing the four eigenvalues, it is easy to assess that
λ 1 Gλ 3 H0 and λ 2 Gλ 4 F0.
Hence, the equilibrium is a saddle point. 䊐

Proof of Proposition 4.3. The first-order conditions relative to the


social planner Hamiltonian (40) are

∂H 兾∂qi G0 ⇒ xiA2ci qi xi G0 ⇒ qi G1兾2ci ,


SP
(56)

∂H 兾∂ai G0 ⇒ A2ai biCλ ii G0 ⇒ λ ii G2ai bi ⇒ λ̇ ii G2ȧi bi ,


SP
(57)

∂H 兾∂xH G−δλ HCθ 1CqHAcH q2HAxHAsxL


SP

Gρλ HHAλ̇ HH , (58)

∂H SP
兾∂xL G−δλ LLCqLAcL q2LCs (θ 1AxHAxL) (59)
Gρλ LLAλ̇ LL ,
with the same initial and transversality conditions as in the previous cases.
From (58) and (59), we have
λ̇ H G(δCρ)λ HAθ 1AqHCcH q2HCxHCsxL , (60)
λ̇ L G(δCρ)λ LAqLCcL q As(θ 1AxHAxL),
2
L (61)
while from (57),
ȧH Gλ̇ H 兾2bH G−1兾4cHAθ 1C2aH bH (δCρ)CxHCs xL , (62)
ȧL Gλ̇ L 兾2bL G−1兾4cLC2aL bL (δCρ)Cs (−θ 1CxHCxL). (63)
The steady-state equilibrium requirement {ȧH G0, ȧL G0} yields
1C4cH (θ 1AxHAsxL)
aSP
H G , (64a)
8bH cH (δCρ)
1C4cL s (θ 1AxHAxL)
aSP
L G , (64b)
8bL cL (δCρ)
JOTA: VOL. 119, NO. 2, NOVEMBER 2003 279

that can be used to rewrite the state equations as


1C4cH (θ 1AxHAsxL)
ẋH G , −dxH , (65)
8bH cH (δCρ)Aδ xH
1C4cL s(θ 1AxHAxL)
ẋL G , −dxL . (66)
8bL cL (δCρ)Aδ xL
Solving the system {ẋH G0, ẋL G0}, one obtains the steady-state sales,
2bL cL δ (1C4cH θ 1)(δCρ)Cs[cLCcH (− 1C4cL θ 1A4cL sθ 1)]
xSS
H G , (67)
4cH cL (2bL δ (δCρ){1C2bH δ (δCρ)C[1C2bH δ (δCρ)As]s}
−cL sCcH [1C2bH δ (δCρ)(1C4cL θ 1 s)]
xSS
L G . (68)
4cH cL (2bL δ (δCρ){1C2bH δ (δCρ)C[1C2bH δ (δCρ)As]s}
Finally, the dynamic system can be written in the matrix form
−δ

冤冥冤 冥冤 冥 冤 冥
ẋH 1 0 0 xH 0
ȧH 1 2bH (δCρ) s 0 aH −1兾4cHAθ 1
G C .
ẋL 0 0 −δ 1 xL 0
ȧL s 0 s 2bL (δCρ) aL −1兾4cLAsθ 1
Again, computing the four eigenvalues, one can check that
λ 1 Gλ 3 H0 and λ 2 Gλ 4 F0.
Therefore, the equilibrium is a saddle point. 䊐

References

1. JØRGENSEN, S., A Surûey of Some Differential Games in Adûertising, Journal of


Economics Dynamics and Control, Vol. 4, pp. 341–369, 1982.
2. ERICKSON, G. M., Dynamic Models of Adûertising Competition, Kluwer, Dord-
recht, Holland, 1991.
3. FEICHTINGER, G., HARTL, R. F., and SETHI, P. S., Dynamic Optimal Control
Models in Adûertising: Recent Deûelopments, Management Science, Vol. 40,
195–226, 1994.
4. DOCKNER, E. J., JØRGENSEN, S., LONG, N. V. and SORGER, G., Differential
Games in Economics and Management Science, Cambridge University Press,
Cambridge, UK, 2000.
5. SCHMALENSEE, R., A Model of Adûertising and Product Quality, Journal of Pol-
itical Economy, Vol. 86, pp. 1213–1225, 1978.
6. KOTOWITZ, Y., and MATHEWSON F., Adûertising, Consumer Information, and
Product Quality, Bell Journal of Economics, Vol. 10, pp. 566–588, 1979.
280 JOTA: VOL. 119, NO. 2, NOVEMBER 2003

7. CONRAD, K., Quality, Adûertising, and the Formation of Goodwill under Dynamic
Conditions, Optimal Control Theory and Economic Analysis, Edited by G.
Feichtinger, North-Holland, Amsterdam, Holland, Vol. 2, pp. 215–234, 1985.
8. OUARDIGHI, F. E., and PASIN, F., Adûertising and Quality Decisions under
Dynamic Conditions, 10th International Symposium on Dynamic Games and
Applications, State University of St. Petersburg, St. Petersburg, Russia, July 8–
11, 2002.
9. CASE, J., Economics and the Competitiûe Process, New York University Press,
New York, NY, 1979.
10. SORGER, G., Competitiûe Dynamic Adûertising: A Modification of the Case
Game, Journal of Economic Dynamics and Control, Vol. 13, pp. 55–80, 1989.
11. MOORTHY, K. S., Product and Price Competition in a Duopoly Model, Market-
ing Science, Vol. 7, pp. 141–168, 1988.
12. MOTTA, M., Endogenous Quality Choice: Price ûs Quantity Competition, Journal
of Industrial Economics, Vol. 41, pp. 113–32, 1993.
13. DUTTA, P. K., LACH, S., and RUSTICHINI, A., Better Late than Early: Vertical
Differentiation in the Adoption of a New Technology, Journal of Economics and
Management Strategy, Vol. 4, pp. 563–589, 1995.
14. GABSZEWICZ, J. J., and THISSE J. F., Price Competition, Quality, and Income
Disparities, Journal of Economic Theory, Vol. 20, pp. 340–59, 1979.
15. GABSZEWICZ, J. J., and THISSE J. F., Entry (and Exit) in a Differentiated Indus-
try, Journal of Economic Theory, Vol. 22, pp. 327–338, 1980.
16. SHAKED, A., and SUTTON J., Relaxing Price Competition through Product Dif-
ferentiation, Review of Economic Studies, Vol. 69, pp. 3–13, 1982.
17. SHAKED, A., and SUTTON J., Natural Oligopolies, Econometrica, Vol. 51, pp.
1469–1483, 1983.
18. LEHMANN-GRUBE, U., Strategic Choice of Quality When Quality Is Costly: The
Persistence of the High-Quality Adûantage, RAND Journal of Economics, Vol.
28, pp. 372–384, 1997.
19. CHAMPSAUR, P., and ROCHET, J. C., Multiproduct Duopolists, Econometrica,
Vol. 57, pp. 533–557, 1989.
20. SPENCE, A. M., Monopoly, Quality, and Regulation, Bell Journal of Economics,
Vol. 6, pp. 417–429, 1975.
21. MUSSA, M., and ROSEN S., Monopoly and Product Quality, Journal of Econ-
omic Theory, Vol. 18, pp. 301–317, 1978.
22. VIDALE, M. L., and WOLFE, H. B., An Operations Research Study of Sales
Response to Adûertising, Operations Research, Vol. 5, pp. 370–381, 1957.
23. LEITMANN, G., and SCHMITENDORF W. E., Profit Maximization through Adûer-
tising: A Nonzero-Sum Differential Game Approach, IEEE Transactions on
Automatic Control, Vol. 23, pp. 646–650, 1978.
24. FEICHTINGER, G., The Nash Solution of an Adûertising Differential Game: Gen-
eralization of a Model by Leitmann and Schmitendorf, IEEE Transactions on
Automatic Control, Vol. 28, pp. 1044–1048, 1983.
25. DOCKNER, E. J., and LEITMANN, G., Coordinate Transformations and Deri-
ûation of Open-Loop Nash Equilibria, Journal of Optimization Theory and
Applications, Vol. 110, pp. 1–15, 2001.