You are on page 1of 13

Optimal Advertising Policies for Diffusion Models of New Product Innovation in Monopolistic Situations Author(s): Engelbert Dockner and

Steffen Jrgensen Source: Management Science, Vol. 34, No. 1 (Jan., 1988), pp. 119-130 Published by: INFORMS Stable URL: http://www.jstor.org/stable/2632189 Accessed: 16/04/2010 06:22
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=informs. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

INFORMS is collaborating with JSTOR to digitize, preserve and extend access to Management Science.

http://www.jstor.org

MANAGEMENT SCIENCE Vol. 34, No. 1, January 1988 Printed in U.S.A.

OPTIMAL ADVERTISING POLICIES FOR DIFFUSION MODELS OF NEW PRODUCT INNOVATION IN MONOPOLISTIC SITUATIONS*
ENGELBERTDOCKNER AND STEFFENJ0RGENSEN Institute of Economic Theoryand Policy, Universityof Economics, Augasse 2-6, A-1090 Vienna,Austria Departmentof Economics, Universityof Saskatchewan, Saskatoon,Saskatchewan,S7N OWO, Canada Institute of TheoreticalStatistics, CopenhagenSchool of Jconomics and Business Administration, Julius ThomsensPlads 10, DK-1925 Frederiksberg C, Denmark
This paper deals with the determination of optimal advertising strategies for new product diffusion models. We consider the introduction of a new consumer durable in a monopolistic market and the evolution of sales is modelled by a flexible diffusion model. Repeat sales and possible entry of rivals are disregarded but we allow for discounting of future revenue streams and cost learning curve. Using standard methods of optimal control theory we characterize qualitatively the structure of an optimal advertising strategy for different versions of the diffusion model. (MARKETING; MARKETING-ADVERTISING; MARKETING-NEW PRODUCTS)

1. Introduction

In the last 15-20 years there has been a growinginterestin applyingmethods of of salesresponse dynamicoptimizationto marketing problems,cast in the framework modelsor diffusionmodels of new productinnovation.In the areasof optimalpricing thereis now a considerable or advertising the readeris referred to the body of literature; surveyarticlesby Sethi (1977), Little (1979) who deal with the advertisingarea, and Rao (1984) who is concernedwith pricing.Surveypapersthat particularly stressthe and Chatterjee competitiveaspectsareEliashberg (1985), Dolan et al. (1986),J0rgensen (1986). Diffusion models of new product innovation are treated by Mahajan and Muller(1979), Mahajanand Wind (1986). Innovationdiffusionmodels are the main focus of this paperand we wish to design optimal advertisingstrategiesfor such models. Briefly stated, a diffusion model attempts to describethe spreadof a new productinnovation among a set of potential adopters.Typically,the model considersthe evolution of the numberof adoptersover time, and the diffusioncomes aboutthroughword of mouth and is due to advertising. observedphenomThe potentialvalue of such modelslies in explanationof frequently ena associatedwith new productinnovation (e.g. the S-shapeddiffusioncurve). FurA third application,which we thermore,diffusionmodels can be used in forecasting. shall considerin this paper,is for normativepurposes,that is, to resolveproblemsof optimalmarketing strategies. In ?2 we give a short review of the literatureon diffusionmodels that incorporate is anadecision variables.In ?3, a flexiblediffusionmodel with advertising marketing lyzed by optimal control methods.A main result of this section is Theorem 1 where qualitativestatementsabout an optimal advertisingpolicy are made under minimal but disregard We includecost experience hypotheses. discounting.It can be shownthat in a diffusionmodel of new productinnovationit pays to increase(decrease) advertis* Accepted by John R. Hauser; received November 20, 1985. This paper has been with the authors 9 months for 3 revisions.

119 0025-1909/88/3401/01 19$01.25


Copyright ?) 1988, The Institute of Management Sciences

120

ENGELBERT DOCKNER AND STEFFEN J0RGENSEN

ing over time if sales increase(decrease)as penetrationincreases.In ?4 which, in our opinion, contains the main resultsof the paper, we look at more specificcases and results(Corollary1 and Theorem2). A model of this sectiongeneralizes obtainstronger to a seriesof otheradvertisthe Horskyand Simon (1983) model and has relationships and suggestsfutureavenuesof ing models. ?5 containssome policy recommendations research.
2. Literature Review

Any innovationdiffusionmodel restson a numberof simplifyingassumptionsthat discussedand, wheneverpossible,empirically tested.Mostly,these mustbe recognized, are assumptionsmade to facilitateanalyticalinsights into the problem and we enbalancebetweencomplexityand tractacounterthe problemof obtaininga reasonable bility. The majorassumptionsunderlyinga diffusionmodel, and variousattemptsto relax these assumptions,are well summarizedin Mahajanand Wind (1986, Tables
1-1, 1-2).

are often criticizedfor their failureto Diffusionmodels havingconstantparameters take into account the effects of the marketingvariables.To relax this assumption, variousproposalshave been made to let the parameters dependon relevantmarketing variables.But no unifiedtheoretical frameworkexists to provideguidelinesfor what can be foundin Robinto include,and where.Differentsuggestions variables marketing son and Lakhani (1975), Dodson and Muller (1978), Mahajanand Muller (1979), Monahan (1984), Horsky and Simon (1983), Kalish (1983), Lilien, Rao and Kalish (1981), Thompsonand Teng (1984), Kalish (1985), Kalishand Sen (1986). A unifiedtheory,supported by conclusiveevidence,dealingwith the questionhow to decisionvariables into diffusionmodels,is stillnot apparent. the marketing incorporate Theoretical models have been built with the purposeof remedyingsome shortcomings of a certainmodel, or makingsome kind of generalization. Only in very few casesthe modelling effortshave been empiricallyvalidated.Neither empiricalnor theoretical workhave providedfirmguidelinesfor the choice of a particular functionalformto be used for a specificdiffusionscenario,and the modellingof the impactsof e.g. advertisratherad hoc. ing in innovationdiffusionprocesseshas progressed In this paperwe suggestthat a feasibleavenue of researchmay be to assume more flexible,i.e. moregeneraldiffusionmodels.' Then we avoid, at leastto a certainextent, the problemthat optimalmarketing strategies may be highlysensitiveto the particular functionalformchosenfor analysis.But this benefitis obtainedat a cost. Ourapproach involves a risk that the conclusions drawn from such 'general'models may be less specific.However,we cannot hardlyexpectmorethan qualitativerecommendations to emerge from general models; this is, of course, a drawbackcomparedto the more clearcutprescriptions obtainedfrom highlyspecializedmodels. 3. Model Formulation We begin the analysisby statinga model with as few assumptionsas possible(given This model will yieldgeneral,but also less specificconclusions. the overallframework). In ?4 we considera more specializedmodel whichwill yield more preciseconclusions. Considerfirstpurchasesof a new consumerdurablein a monopolisticmarket.We its adverrestrictthe analysisto the case of a monopolistfirmwhich only manipulates over a fixedplanningperiod. tisingexpenditure Let t denote time such that 0 < t < T. The lengthof the planningperiod, T, is fixed.
I

This approach resembles the one used by Sasieni (1971) and Kalish (1983).

OPTIMAL ADVERTISING FOR NEW PRODUCT DIFFUSION

121

Define x = x(t): cumulativesalesvolume (demand)by time t, A = A(t):the firm'srate of advertising at time t. expenditure A flexiblediffusionmodel is given by
x= (t) =

g(x(t), A(t)),

x(O) = xO? 0

and fixed,

(1)

where x representsthe currentsales rate. By (1), the currentsales rate is relatedto accumulatedpast sales and the currentrate of advertising expenditure.Functiong is and twice differentiable (2) with respect to that where a subscripton a variabledenotes partial differentiation in (2) implythat salesarenonnegative variable. The inequalities and concaveincreasing as a functionof advertising.23 We introducea cost learningcurve by assumingthat marginalcosts, denoted by c, depend on cumulativesales (production)such that marginalcosts decreasewith increasingcumulativeoutput (experience):
c = c(x); dc(x)/dx = c'(x) ? 0. (3) gA >O; < 0; gAA

for all (x, A)

Note that marginalcosts could be constant(c' = 0). The learningcurve phenomenon has been applied in a number of innovation diffusion models: see Robinson and Lakhani(1975), Bass(1980), Kalishand Lilien (1983), Kalish(1983, 1985). Assume that the firm chargesa constant price,p, over the planningperiod.4This made for mathematical convenience.Ideally,the optimization assumptionis primarily problemshouldincludepriceas well as advertising See (andothermarketing variables). Thompsonand Teng (1984), Kalish(1985) whereboth priceand advertising are decision variablesof the firm. It should be noted, however,that the interactionbetween optimal price and optimal advertisingmay be sensitive to the particularfunctional formof the diffusionmodel. In the model (1), pricereductionscouldpossiblysubstitute in orderto increaseinstantaneoussales. It is, however,our conincreasedadvertising jecturethat in a model includingprice as well as advertising, the qualitativestructure of an optimal advertising policy will be the same as the ones that will be establishedbelow.5
The instantaneous profit stream to the firm is given by P(x, A) = (p
-

c(x))g(x, A)

- A. A convex advertising term in P(x, A) was suggested expenditure by Gould (1970) but was questionedby Schmalensee(1972). Instead of the term -A, Gould (1970) introduced a term:-h(A) wherefunctionh is convex,but assumedg,s = 0, thatis, sales arelinearlyincreasing with advertising. (Noticethat Gould'sproblem(convexadvertising cost and linear sales) will yield a solution which is structurally equivalentto the solutionof our problemwheresalesare concaveand advertising cost is linear.) The intertemporal optimizationproblemfor a profitmaximizingfirmcan be stated
2In diffusion models with advertising, a concave function has been employed by e.g. Horsky and Simon (1983), Kalish (1985). 3The concavity assumption guarantees that it is never optimal to increase advertising without bound. Hence we avoid nonsensical conclusions. The assumption in (2) means that the firm is operating on a concave portion of the advertising-sales response function. See Little (1979), Hauser and Shugan (1983), Lilien and Kotler ( 1983) for a further discussion of possible shapes of the response function. 4 In Theorem 2 below (where marginal costs are constant) we also need to assume that the (constant) term p - c (i.e. the unit margin) is strictly positive. This is not an unusual assumption because otherwise the firm would possibly not produce the product at all. 5 This conjecture has support in calculations done for a general price-advertising model in a preliminary paper by Dockner et al. (1985).

122

ENGELBERT DOCKNER AND STEFFEN J0RGENSEN

as follows.Determinean advertising policy over the fixed time intervalfrom t = 0 to


t = T such that the total discounted profit
T

J(x, A)

exp(-rt)[(p

c(x))g(x, A)

A]dt

(4)

is maximized,subjectto the dynamicconstraint(1). In (4), r is a constantnonnegative discountrate.As admissiblecontrolsA = A(t)we takethose thataretwicedifferentiable in t and satisfyA(t) 2 0 for all relevantt.6 we apply To solve the optimalcontrolproblem(4) subjectto (1) andA(t)admissible, Hamiltonian the maximumprinciple(Arrowand Kurz 1970).Definethe current-value
H=H(x,A, X) =g(x,A)(p-c(x)+ X)-A (5)

costatevariable(shadowprice of x) which satisfies whereX = X(t)is the current-value the differential equation X = rX- gx[p - c + X] + c'g (6)
with the transversality condition at t = T X(T) =O. (7)

In (6), and in the sequel,we omit the arguments(x, A) and x of functionsg and c (and theirderivatives) whereno confusioncan arise. For an optimal nonnegativeA we have the firstordernecessarycondition that HA < 0. Confiningour interestto interiorsolutionswe have
H4 = 0 == gA[p-c

+ X] = 1

(8)

which admits a unique solution because of the assumptionsin (2). A second order conditionfor H-maximizationis (9) is strictlypositive.Note thatgenerally wheneverthe termin brackets whichis satisfied it is not sufficientfor optimalitythat (9) is satisfied.A sufficientcondition is the following:the maximizedHamiltonianis concavein x (Arrowand Kurz 1970).Whenapplying this condition to our control problem it turns out that sufficiencyis not always at least we cannot provethat the sufficientconditionis met in the general guaranteed; model (1) with cost learningand a positivediscountrate.7 We state some immediateconsequencesof the optimalityconditions(6)-(8). For an optimalA > 0 we get by substitutionfrom (8) into (6)
HAA<?0=g4A[p-c+X]<0

X = rX -

- + gc'
gA

(10)

in (10) (using(7)) yields and integration X = exp(rt) J [


g9A

rTx

gc' exp(-rs)ds.

(11)

To explainthe content of (11) firstnotice that the costatevariableXhas the interpreta6 The assumption of differentiability may seem unnecessarily strong. However, if the Hamiltonian is strictly concave in A then the necessary optimality conditions uniquely define an optimal A, say, A*. When strict concavity holds it follows from the necessary conditions that the state and the costate are continuous and A* is differentiable whenever it is an interior solution. See also Kalish (1983, p. 158). 7 However, for the diffusion models of ?4, the sufficient conditions are satisfied under mild restrictions on the parameters. A detailed proof is omitted because of space limitations; it can be obtained from the authors upon request.

OPTIMAL ADVERTISING FOR NEW PRODUCT DIFFUSION

123

tion X = uJ*/ax, where J* is the optimal value of the profit functional. Hence the increasein cumulathe dollarvalue(at time t) of a marginal costatevariable represents tive sales. (Equation(10) shows how this dollarvalue changesover time.) Next notice that gx > 0 (gx < 0) means that additionalsales now will increase(decrease)future demand. Put in another way, gx > 0 (gx < 0) reflectspositive (negative)'diffusion' effectson the demand side. Finally notice that the costate variabledepends on three importantdynamic factors:diffusion effect on the demand side, cost learning and discounting.From (1 1) it is easily seen that if (a) gx is uniformlypositive,then from (10) the costateis positive for all t. FurtherThis seems to be more, if r = 0, then the costatevariableis monotonicallydecreasing. consistentwith intuition, since in this case the firm has no preference for earlyprofits and gx positivemeansthat additionalsalesnow stimulatesaleslateron. In this case we to would expect an optimal advertising policy to be one of initiallyheavy advertising reapthe benefitsfrom rapidlyincreasingcumulativesales which will stimulatefuture demand. Then advertisingcan be graduallydecreasedas the diffusionprocessgains momentum. (b) gx is uniformlynegativeand cost learningis absent(i.e. c' = 0), then the costate variableis negativebut increasingfor all t. Here, increasedpenetrationhas an adverse effecton futuredemandand we wouldexpectthat optimaladvertising shouldbe low in the beginningof the planningperiod.Also, concerningcosts, there is no incentiveto increasecumulativesalesin orderto benefitfrom lowercosts in the future. We are ready to state a result for the case where sales are given by the general specification (1), i.e. x = g(x, A). Thereare cost experienceeffectsbut a zero discount rate.(The case wherethe discountrateis positiveis apparently intractable.)
THEOREM 1. Withdemandgiven by (1), a cost learningcurvebut no discounting, thenan optimaladvertising policy is characterized by: A is decreasingovertime if gx > 0 and gAx < 0 for all t, A is increasing overtime if gx < 0 and gAx> 0 for all t.8

PROOF. See the appendix. The resultsin Theorem 1 can be given the followinginterpretations: 1. The decrease(increase)of A is global,i.e. A decreases for all t, and the (increases) advertising path is monotonic. 2. An optimaladvertising policy is independentof the cost learningcurve since the assumptionr = 0 removesthe effectof the costatevariable. (See also equation(0) in the appendix.)Hence the shape of an optimal advertising path is only influencedby the diffusioneffects,embodiedin the signs of the partialsgx and gAx, respectively. 3. The conditiongx > 0 and gAx < 0 means that demand (sales)will increasewith penetrationor, put in another way, additional sales now increase future demand. i.e. we Moreover,the increasein currentsalesis weakerfor higherlevelsof advertising, have a kind of 'diminishingreturns'phenomenon.The policy implicationis to advertise heavily in the beginningof the planning period in order to stimulate adoption which in turn stimulatesfuturesales.The conditiongx < 0 and gAx> 0 is relatedto a situation where demand decreaseswith penetrationand the decreaseis weaker for higher levels of advertising.The policy recommendationis to increase advertising

8 Noticethatthe cases(gx > 0, gAx > 0) and (g, < 0, gAx < 0), respectively, in Theorem1. arenot considered There are two reasons for this omission. First, the cases treated in the theorem are those we found analytically tractable (cf. the appendix, Equations (0), (i)). Second, a situation where, for instance, g, > 0 and gAx > 0 would imply that sales are increasing with penetration, and the increase in (current) sales being higher for higher levels of (current) advertising, i.e. a 'increasing returns' phenomenon. We conjecture that the opposite case is the more realistic one, but this has to be decided in each particular application.

124

ENGELBERT DOCKNER AND STEFFEN J0RGENSEN

the decreasein currentsales graduallyover the horizon in order to counterbalance caused by increasingpenetration.(See also the remarksconcerning(11).) A typical situationfor a durablegood is as follows:at introduction,a positive word of mouth stimulatesdemand (i.e. g, is positive),while lateron, saturationeffectsdominateand with additionalsales now (i.e. g, is negative).In this case an futuredemanddecreases optimaladvertising policy may not be monotonicas the ones in Theorem 1, but rather U-shaped. the case 4. Notice thatalthougha zerodiscountratemay seem somewhatunrealistic, to situationswherethe discount rate is relativelylow. can serveas an approximation Also notice that a zero discount rate means that the firm is indifferentof having one dollartoday or one dollarat some instantin the future.
4. Specific Functional Forms

In this section, the demand function g(x, A) is specifiedas a Bass (1969) diffusion rate.Consider the followingspecimodelwith parameters dependingon the advertising fication9
x = g(x, A) = [a + Of (A) + yx + af (A)x](M-x) (12)

are nonnegativeand constant. For , = 6 = 0, a and 'y can be where all parameters interpreted as in the Bass (1969) model; a is the innovation coefficientand 'y is the imitation coefficient.The parameter,Bis related to the effectivenessof advertising of advertising toward the innovators,whereas6 is relatedto the effectiveness vis-'a-vis the imitators. and The advertising efficiencyfunction,f (A) is twice differentiable (13) implying decreasingreturns from advertising.(See also assumption (2).) Teng and Thompson(1983), Thompsonand Teng (1984) considereda model of the same structure as (12) but with f(A) linear. (HencefA = 0, i.e. there are constant returnsto advertisingand (13) is not satisfied.)Due to the linearity,they obtainedadvertising arepiecewiseconstant;sometimes policiesof the bang-bang type suchthatexpenditures is zero, and sometimesit is at the maximallevel. advertising Let us considersome specialcases of (12). Putting6 = 0 yields
x = g(x, A) = [a + Of (A) + yx](M-x). (14) f '(A) > 0; f "(A)< 0

In (14) it is assumedthat advertising informsthe innovators; imitatorsareinformedby word of mouth. The model (14) could describea situationwhere the productis an is targetedat the innovators innovationat an early stage of the life cycle. Advertising and emphasizesnewness,uniquenessand personalvaluefor the potentialadopter.The diffusionmodel in (14) has been analyzedby Horskyand Simon (1983) who employed a specificadvertising f (A) = ln A. The model (14) is also relatedto efficiencyfunction: the modelsof Stigler(1961),Fourtand Woodlock(1960), namelyfor ,B= y = 0. Taking and introducinga decay the innovationcoefficientas a linear function of advertising termon the right-hand side of (14), say, -dx, turnsthe modelinto the Vidaleand Wolfe (1957) model. Optimal advertisingstrategiesfor this model have been designedby Gould (1970), Sethi (1973). Putting ,B= 0 in (12) yields
x = g(x, A) = [a + yx + af (A)x](M-x). (15)

9 See also Simon and Sebastian ( 1987) where an empirical study of some variants of ( 12) is carried out, using data from an advertising campaign of the German telephone company. The purpose of the analysis was to test for goodness-of-fit various subcases of (12); hence, no optimization was involved.

OPTIMAL ADVERTISING FOR NEW PRODUCT DIFFUSION

125

In (15) it is supposed that advertising informs the imitators. The model could describe a situation where an innovation is at an intermediate stage of the life cycle. The product is now an object of social communication which can be stressed by advertising that emphasizes buyer evaluation, demonstration effects and social pressure. The diffusion model in (1 5) is related to the models of Ozga (1960), Mansfield (1961), namely for a = 6 = 0. Taking the imitation coefficient as a linear function of advertising and appending a decay term yields the optimal advertising model treated by Gould (1970), Sethi (1979). In the model (12) we distinguish four components of the diffusion process: a(M - x): adoption without interaction with previous adopters and without influence from advertising, 3f (A)(M - x): adoption without interaction with previous adopters but stimulated by advertising, 'yx(M - x): adoption by interaction with previous adopters but without influence from advertising, af (A)x(M - x): adoption by interaction with previous adopters and stimulated by advertising. In the two subsections to follow we analyze the optimal advertising problem (4), subject to the evolution equation (12). First we assume a zero discount rate; next we derive some results for the more realistic case of positive discounting. In the latter we specify the advertising efficiency function as logarithmic (cf. Sethi 1975, Horsky and Simon 1983).

4.1. The Undiscounted Case


Throughout this subsection, the discount rate, r, is equal to zero. The following is an implication of Theorem 1.
COROLLARY 1. If the demandfunctionis given by (12), then optimal advertising are increasingif ab - 0,y > 0, and decreasing expenditures for ab - fry < 0.

PROOF. See the appendix. The result of Corollary 1 can be seen as a nonlinear counterpart to the result in Theorem 2 of Teng and Thompson (1983). However, these authors could prove their result for a positive discount rate. The inequalities that must hold for parameters to generate the two different types of policies correspond to those employed by Teng and Thompson (1983). Notice that advertising should, for example, be increased if ab - fiY > 0 O a/y > A/6, i.e. if the independent innovative effect is stronger than the imitative effect. For the two special cases of (12), that is (14) and (15), respectively, we can obtain more definite answers. (a) If the demand function is given by (14), i.e. advertising informs the innovators, then optimal advertising expenditures are decreasing. (This is easily seen from the corollary by putting 6 = 0.) As noted, equation (14) is a generalization of the Horsky and Simon (1983) model to the case of a general, concave advertising efficiency function. Hence we have demonstrated that the result of Horsky and Simon (1983) holds in a more general set-up. Moreover, we have proved that a policy of decreasing advertising expenditures is also optimal in a model incorporating cost learning. Horsky and Simon (1983) proved their results for a model with a logarithmic advertising efficiency function and a constant marginal cost. (b) If the demand function is given by (15), i.e. advertising informs the imitators, then optimal advertising expenditures are increasing. (This is easily seen from the corollary by putting A = 0.) The result is the opposite of the above but when advertising informs the imitators it is beneficial to postpone expenditures. (In contrast hereto,

126

ENGELBERT DOCKNER AND STEFFEN J0RGENSEN

informsthe innovatorssince in that whenadvertising shouldbe decreased expenditures to informthe innovaamount of advertising case it paysto use initiallya considerable tors and get the diffusionprocessstarted.) 4.2. The DiscountedCase Throughoutthis subsectionthe discount rate is strictlypositive. FollowingHorsky and Simon (1983) we specifythe advertisingefficiencyfunction as logarithmic: f(A) = ln A. Furthermore we assume that there is no cost learning.This implies constant marginalcosts. We state a resultfor the demandfunction(12), recallingthat (14) and can be obtainedas specialcases of ( 12). (15), respectively,
THEOREM 2. For a constantmarginalcost, c, such thatp - c > 0, and a demand functiongiven by (12) withf (A) = ln A, thenthefollowingholds: expendi(i) If ab - 0ry> 0 and x(T) ? (M/2) - (f/26), thenthe optimaladvertising turesare increasing(for all t).

TABLE 1 Summary of Results > Demand Specification ~~Discounting/ lll No Discounting Cost Learning Discounting No Cost Learning

(a + ff(A) + yx +6xf(A))(M-

x)

g(x, A) g(x,~A).

=g(a A) ,Bf(A)+ yx + 6xf(A))(M- x) 1! ab y <0

|tt**

g(x, A) = (a + Off(A)+ 'yx)(M- x)

g(x, A)

= (a

+ yx + 6xf(A))(M - x)

*)f(A) = log A, **) x(T) not too large, ***) x(O) not too small

OPTIMAL ADVERTISING FOR NEW PRODUCT DIFFUSION

127

< 0 and x(O)2 (M/2) - (f/26), thenthe optimaladvertising (ii) If ab - 0,By expendituresare decreasing (for all t). See the appendix. Theorem2 tends to confirmthe resultsthat werederivedfor the undiscountedcase. The theoremstatesthat (i) optimaladvertising is increasing underthe sameconditionas in the undiscounted case, providedthat penetrationat the horizon date is not too high. This additional conditioncan be satisfiedif, for instance,the marketpotentialis sufficiently large. (ii) optimal advertisingis decreasingunder the same conditions as in the undiscounted case, providedthat the initial penetrationexceeds a certain(minimal)level. Thisextraconditioncan be satisfied is relatively if, for example,initialpenetration high comparedto the maximalpenetrationlevel. As in the undiscountedcase we can obtain strongerresultsfor the two models (14) and (15). as in Theorem2, exceptthatdemandis givenby (14), (a) Underthe samehypotheses then the optimal advertisingexpendituresare decreasing.(This is easily seen from Theorem2 with 6 = 0.) This resultconfirmsa conjectureof Horskyand Simon (1983) that the structureof an optimal policy will be the same in the discountedas in the undiscounted case.As alreadymentioned,Horskyand Simon (1983) only provedtheir resultsfor the case of a zero discountrate. (b) Under the same hypothesesas in Theorem 2, except that demand is given by (15), then the optimal advertising expendituresare increasingwheneverit holds that x(T) ? M/2. (Thisis easilyseen fromTheorem2 with ,B= 0.) We obtainthe policythat shouldbe increasedas long as saturation effectsare not domiadvertising expenditures nating, in the sense that the market is not 'too much saturated'at the end of the planningperiod. In Table 1 our resultshave been summarizedin the form of diagramsdepictingthe The focus is of the time path of optimaladvertising qualitativestructure expenditures. on the resultsfrom the threediffusionmodels in ?4.
PROOF.

5. Concluding Remarks This paperhasdealtwith the problemof determining optimaladvertising policiesfor monopolistic new product diffusion models. In particular,we have tried to design models of some flexibility(cf. (1) and (12)). Results of Horskyand Simon (1983) on decreasingadvertisingstrategieshave been extendedto cases with a generalconcave advertising efficiencyfunction, cost learningor a positive discount rate. A nonlinear to the model of Teng and Thompson(1983) was analyzedbut only with a counterpart zero discountrate. The followingmanagerial implicationsare emphasized. * Our approachto modellingdiffusionof a new productexplicitlyrecognizesthree relevant.First,there are dynamiceffectson the dynamicfactorsthat are managerially demandside, reflectedin the diffusiondifferential equation.Such effectsarisebecause of phenomenasuchas wordof mouthand saturation. Second,therearedynamiceffects on the cost side,due to the presenceof a cost learningcurve.Third,dynamicconsiderations are involved via the discount rate which revealsthe manager'stime preference for money. * In all caseswe obtainedmonotonic advertising strategies over the entireplanning should eitherbe initiallylow, and then increasing, period.Hence, optimaladvertising or initially heavy and then graduallydecreased.It should be noticed, however,that could dependon the fact that the profitfunctional(4) did not these types of strategies includea salvagevalueterm.If the firmputs a valueon the finalamountof cumulative

128

ENGELBERT DOCKNER AND STEFFEN J0RGENSEN

sales, then nonmonotonic strategies may be optimal. Moreover, such strategies could emerge in cases where the demand diffusion effects are not uniform over the interval [0, T]. Finally, it should be stressed that in a number of cases monotonicity was dependent on the length of the planning period. * Strategies of decreasing advertising typically occur when there are positive diffusion effects throughout the planning period, or advertising is targeted at the innovators. This type of policy has been reported in the majority of cases treated in the literature; apparently it is quite robust against the choice of model assumptions. * Strategies of increasing advertising occur when negative diffusion effects are present and advertising is aimed at the imitators. This case may, however, be the less interesting case since the product suffers from negative word of mouth. Moreover, advertising toward imitators has only indirect effects. The diffusion models under study in this paper are restricted to the case of a monopolist firm which only manipulates its advertising. Evident extensions are multiplayer models, incorporating several marketing decision variables. For such models, see J0rgensen (1982), Eliashberg and Chatteree (1985), Dolan et al. (1986), J0rgensen (1986), Kalish and Sen (1986). Recently, also studies have begun of marketing policies for diffusion in stochastic environments. See Eliashberg et al. (1987), Raman (1987).10
'? The research of the first author was supported by a grant from the Osterreichischen Bundeswirtschaftskammer (Exportakademie).

Appendix
PROOF OF THEOREM 1. Differentiation with respect to time in (8), and using (10) yields -A

AA X
=

(g)

(r SgA+

g'

g-"g
)2

(0)

For r = 0 we obtain in (0):


gAA
(g -)2

gx
gA

gxA
(g)

(i)

and the results follow immediately.


PROOF OF COROLLARY 1.

Q.E.D.

For r = 0 we obtain(i) from(0), and evaluation of


-gxgA + gAxg =

F(x, A)

(ii)

for demand specification (12) shows that F(x, A) = f'(A)(M Using (i)-(ii) completes the proof. Q.E.D.
PROOF OF THEOREM 2.
- x)2 [ab -

ly].

Withdemandgivenby (12) andf(A)


A = (O + bx)(M -x)

= -

ln A, we obtainfrom(O)
(iii)

03y)1(ii Xr + A(ca (o + 6X)2 ]J* _ + c(iv)

From (8) follows A and substitution of (iv) into (iii) yields

A = A[r

a-yM

_]-(d J)

+ x)(M-x)(p-c)r.

(v)

Differentiation in (v) with respect to time and evaluation for A = 0 yields A|


lAO X -A(ab-y)

+ A(a
x

y)(M - x) + r(p-c)[w(M-x)(fl+ 6X)2

- x]-

(vi)

OPTIMAL ADVERTISING FOR NEW PRODUCT DIFFUSION Whenever A is constant, it must hold that (cf. (iii) and (iv))

129

fi+ 5x
Substitution of (vii) into (vi) yields AIA=o =
XI-d+A

Ar+

r(p

c)6(M

x) = _ A(a

(i+ 6X)2(v)

y)(M

-x)

(vii)

[a

y - r] - r(p - c)[26(M- x) -

-x]}

(viii)

First consider the case where


ab - de > 0. (ix)

From (iii) and (ix) it is clear that A(T) > 0. If, in (vi), it holds that x c (M/2) - (fl/26) then A < 0 along A = 0. Hence A > 0 for all t. If, in (viii), it holds that x c (2M/3) - (fl/36), and the stronger condition (than (ix)) ab - ,ly - br 2 0 is satisfied, then A < 0 along A = 0. Hence A > 0 for all t. This shows the first part of the theorem. Next consider the case ab - fy < 0. (x) From (iii) and (x) we haveA(T) < 0. If, in (vi), it holds that x0 2 (M/2) - (j3/26)then A > 0 forA = 0 and hence A < 0 for all t. This completes the proof. Q.E.D.

References
ARROW, K. J. AND M. KuRz, Public Investment, The Rate of Return, and Optimal Fiscal Policy, Johns

Hopkins Press, Baltimore, 1970. BASS,F. M., "A New Product Growth Model for Consumer Durables," Management Sci., 15 (January 1969), 215-227. , "The Relationship Between Diffusion Rates, Experience Curves and Demand Elasticities for Consumer Durable Technical Innovations," J. Business, 53 (1980), S51-S67. DOCKNER, E., G. FEICHTINGER AND G. SORGER, "Interaction of Price and Advertising under Dynamic Conditions," Working Paper, University of Economics/Technical University, Vienna, 1985. DODSON, J. A. AND E. MULLER, "Models of New Product Diffusion Through Advertising and Word-ofMouth," Management Sci., 24 (November 1978), 1568-1578. DOLAN, R. J., A. P. JEULAND AND E. MULLER, "Models of New Product Diffusion: Extension to Competition against Existing and Potential Firms over Time," in Innovation Diffusion Models of New Product Acceptance, V. Mahajan and Y. Wind (Eds.), Ballinger Publ. Co., 1986, 117-150. ELIASHBERG,J. AND R. CHATTERJEE,"Analytical Models of Competition with Implications for Marketing: Issues, Findings, and Outlook," J. Marketing Res., 22 (August 1985), 237-261. , C. TAPIERO AND Y. WIND, "Innovation Diffusion Models with Stochastic Parameters," The Wharton School, University of Pennsylvania, Working Paper 1987. FOURT, L. A. AND J. W. WOODLOCK, "Early Prediction of Market Success for New Grocery Products," J. Marketing, 26 (October 1960), 31-38. GOULD, J. P., "Diffusion Processes and Optimal Advertising Policy," in Microeconomic Foundations of Employment and Inflation Theory, E. S. Phelps et al. (Eds.), W. W. Norton, New York, 1970, 338-367. HAUSER, J. R. AND S. M. SHUGAN, "Defensive Marketing Strategies,"Marketing Sci., 2 (Fall 1983), 319-360. HORSKY, D. AND L. S. SIMON, "Advertising and the Diffusion of New Products," Marketing Sci., 2 (Winter 1983), 1-18. JORGENSEN, S., "A Survey of Some Differential Games in Advertising," J. Economic Dynamics and Control, 4 (November 1982), 341-369. , "Optimal Dynamic Pricing in an Oligopolistic Market: A Survey," in Dynamic Games and Applications in Economics, T. Basar (Ed.), Springer-Verlag,New York, 1986, 179-237. KALISH, S., "Monopolist Pricing with Dynamic Demand and Production Cost," Marketing Sci., 2 (Spring 1983), 135-160. , "A New Product Adoption Model with Price, Advertising and Uncertainty," Management Sci., 31 (December 1985), 1569-1585. AND G. L. LILIEN, "Optimal Price Subsidy Policy for Accelerating the Diffusion of Innovation," Marketing Sci., 2 (Fall 1983), 407-420. AND S. K. SEN, "Diffusion Models and the Marketing Mix for Single Products," in Innovation Difusion Models of New Product Acceptance, V. Mahajan and Y. Wind (Eds.), Ballinger Publ. Co., 1986, 87-116. LILIEN, G. L., A. G. RAO AND S. KALISH, "Bayesian Estimation and Control of Detailing Effort in a Repeat-Purchase Diffusion Environment," Management Sci., 27 (May 1981), 493-506. AND P. KOTLER, Marketing Decision Making, Harper and Row, New York, 1983.

130

ENGELBERT DOCKNER AND STEFFEN J0RGENSEN

LITTLE, J. D. C., "Aggregate Advertising Models: The State of the Art," Oper. Res., 27 (July/August 1979), 629-667. "Innovation Diffusion and New Product Growth Models in Marketing," J. MAHAJAN, V. ANDE. MULLER, Marketing, 43 (Fall 1979), 55-68. ANDY. WIND (EDS.), Innovation Diffusion Models of New Product Acceptance, Ballinger Publ. Co., 1986. MANSFIELD, E., "Technical Change and the Rate of Imitation," Econometrica, 29 (October 1961), 741-766. G. E., "A Pure Birth Model of Optimal Advertising with Word-of-Mouth," Marketing Sci., 3 MONAHAN, (Spring 1984), 169-178. OZGA,S. A., "Imperfect Markets Through Lack of Knowledge," Quart. J. Economics, 74 (February 1960), 29-52. RAMAN,K., "Dynamic Optimization of the Marketing Mix in a Stochastic Environment," Auburn University, Department of Marketing, Working Paper, 1987. RAo, V. R., "Pricing Research in Marketing: The State of the Art," J. Business, 57, 2 (1984), 39-60. Sci., 21 B. AND C. LAKHANI, "Dynamic Price Models for New Product Planning," Management ROBINSON, (June 1975), 1113-1122. M. W., "Optimal Advertising Expenditure," Management Sci., 18 (December 1971), 64-72. SASIENI, R., The Economics of Advertising, North-Holland, Amsterdam, 1972. SCHMALENSEE, SETHI, S. P., "Optimal Control of the Vidale-Wolfe Model," Oper. Res., 21 (July/August 1973), 998-1013. , "Optimal Control of a Logarithmic Advertising Model," Oper. Res. Quart., 26 (1975), 317-319. , "Dynamic Optimal Control Models in Advertising: A Survey," SIAM Rev., 19 (October 1977), 685-725. , "Optimal Advertising Policy with the Contagion Model," J. Optim. Theory Appl., 29 (December 1979), 615-627. "Diffusion and Advertising: The German Telephone Campaign," ManageH. ANDK. H. SEBASTIAN, SIMON, ment Sci., 33 (April 1987), 451-466. STIGLER, G., "The Economics of Information," J. Political Economy, 69 (June 1961), 213-225. "Oligopoly Models for Optimal Advertising When Production Costs Obey TENG, J. T. AND G. L. THOMPSON, a Learning Curve," Management Sci., 29 (September 1983), 1087-1101. THOMPSON, G. L. AND J. T. TENG, "Optimal Pricing and Advertising Policies for New Product Oligopoly Models," Marketing Sci., 3 (Spring 1984), 148-168. "An Operations Research Study of Sales Response to Advertising," Oper. M. L. ANDH. B. WOLFE, VIDALE, Res., 5 (June 1957), 370-38 1.

You might also like