Managing Advertising and Promotion for Long-Run Profitability

Kamel Jedidi • Carl F. Mela • Sunil Gupta
Graduate School of Business, Columbia University, New York, New York 10027, College of Business Administration, University of Notre Dame, Notre Dame, Indiana 46556, of Graduate School Business, Columbia University, New York, New York 10027,

In recent years, manufacturers have become increasingly disposed toward the use of sales promotions, often at the cost of advertising. Yet the long-term implications of these changes for brand profitability remain unclear. In this paper, we seek to offer insights into this important issue. We consider the questions of i) whether it is more desirable to advertise or promote, ii) whether it is better to use frequent, shallow promotions or infrequent, deep promotions, and iii) how changes in regular prices affect sales relative to increases in price promotions. Additional insights regarding brand equity, the relative magnitude of short- and long-term effects, and the decomposition of advertising and promotion elasticities across choice and quantity decisions are obtained. To address these points, we develop a heteroscedastic, varying-parameter joint probit choice and regression quantity model. Our approach allows consumers’ responses to short-term marketing activities to change in response to changes in marketing actions over the long term. We also accommodate the possibility of competitive reactions to policy changes of a brand. The model is estimated for a consumer packaged good category by using over eight years of panel data. The resulting parameters enable us to assess the

effects of changes in advertising and promotion policies on sales and profits. Our results show that, in the long term, advertising has a positive effect on “brand equity” while promotions have a negative effect. Furthermore, we find price promotion elasticities to be larger than regular price elasticities in the short term, but smaller than regular price elasticities when longterm effects are considered. Consistent with previous research, we also find that most of the effect of a price cut is manifested in consumers’ brand choice decisions in the short term, but when long-term effects are again considered, this result no longer holds. Last, we estimate that the long-term effects of promotions on sales are negative overall, and about two-fifths the magnitude of the positive short-term effects. Finally, making reasonable cost and margin assumptions, we conduct simulations to assess the relative profit impact of long-term changes in pricing, advertising, or promotion policies. Our results show regular price decreases to have a generally negative effect on the long-term profits of brands, advertising to be profitable for two of the brands, and increases in price promotions to be uniformly unprofitable. (Long-Term Effects ; Promotions ; Advertising ; Price Sensitivity Scanner Data ; Choice ; Purchase Quantity ; Switching Regression

; )

Cop yr ig ht 1 9 9 9 , Ins t it u t e f o r Op e ra t i on s R e se a rc h a nd t he Ma n a ge me n t Sc i e nc e s

Marketing Science/Vol. 18, No. 1, 1999 pp. 1–22


1. Introduction
In the last few years, Procter and Gamble has been to lead the consumer packaged goods trying industry by reducing trade promotions and coupons ( Street January 15 and January 22, 1997), Journal, advertising and brand building, and emphasizing follow-EDLP pricing strategy. Two driving factors ing an ap- to be at the core of P&G’s strategy—cost pear reduction better supply chain management through (Kurt Associates 1993), and a belief that, in the Salmon long advertising is good and promotions are bad term, for brands. However, a recent survey of manufacturers and retailers shows that many companies in the industry have not followed P&G’s lead (Cannondale Associates 1998). In fact, this survey shows that the proportion of marketing budget allocated to trade promotions has grown from 44% in 1997 to 47% in 1998, a trend that continues the increase of the past decade. Companies continued reliance on promotionsfrom the fact that, while it is easier to may stem assess the short-term effects of promotions (a topic that many academic studies have also focused on), it is much to determine the long-term effects of harder promotions and advertising. The task of assessing longrun effects is exacerbated by the fact that competitors often respond to changes in marketing policy. Unless com- can measure, quantify and compare the panies short- long-term effects of promotions and and advertising on brand sales and profits, it is difficult to imagine how they may arrive at the appropriate budget allocationthese two marketing elements. Although between important, this issue is relatively underresearched et al. 1995, Bucklin and Gupta 1998, (Blattberg Gupta 1993). The purpose of our study is to fill this gap by determining and comparing shor - and lon -term the g consumers’ effects of promotion and advertisin t on s g behavior (including both choice and purchase and quantity) consequently on the long-run profitability of a brand. Accordingly, we develop a model and use eight years of disaggregate data to address this goal. In the process,

we also broach such tactical questions as i) “is it better frequent shallow discounts or infrequent to offer deep discounts,” and ii) “is it better to charge high The Wall regularand offer deep discounts or vice price versa?” 1.1. Related Research and Contributions of the Pape r A few recent studies have also examined some of the questions that we are addressing in this paper. It is, therefore, appropriate to highlight how our effort dif- from and builds upon previous research. The fers main objective of our paper is to provide substantive insights and an approach to managing advertising and promotion for the long-run profitability of brandstaking into account changes in consumers’ while and competitors’ behavior over time. Specifically, our pa- addresses the following per issues. Advertising-Promotion 1. We Tradeoff: the strategic issue of how best to allocate resources between advertising and promotion. Using model results simulation that accounts for both shortand a and long-term effects, changes in consumers’ behavior, and competitors’ reactions, we conclude that it is perhaps for all brands to unilaterally increase unwise advertising and cut promotions or vice versa. In other words, the ad-promotion tradeoff is brand specific. It depends on brand-specific advertising and promotion effects as as the current level of resources allocated to well these two decisions. In contrast, many recent studies have examined the long-term effects of either advertising or promotion, Therefore, these studies are unable to but not both. of- any insight on the issue of advertisingfer promotion tradeoff. For example, Dekimpe and Hanssens (1995a, include only advertising, while Papatla 1995b) and Krishnamurthi (1996) and Mela et al. (1998)


include only promotions. Studies that incorporated both these decisions also have significant limitations. For example, Boulding et al. (1994), and Mela, Gupta, and Lehmann (1997—hereafter MGL) provided only direc- results (i.e., they do not consider whether tional short- effects outweigh long-term effects or the term relative different strategies) and may have costs of inadver- the impression that somehow advertising tently left is “good” and promotions are “bad.” Sethuraman and Tellis (1991) examined the trade-off between advertis- price discounting. However, their study ing and has
Marketing Science/Vol. 18, No. 1, 1999

Consistent with Fader et al. (1992) we define short-term effects of

promotion and advertising as current effects (e.g., the effects of this week’s promotions on sales), and long-term effects as those occurring over several quarters or years.


g. Longand Short-Run Our paper allows Tradeoffs: us to explicitly capture the that while choice accounts for a large sults proportion of the total promotion elasticity in the short(consistent with Bucklin et al. “. This in turn in uences the profit impact of these marketing variables. MGL) not provide could Impact on Choice and tradeoff. lowers sion of the elasticity of price. we may not see any changes in brand shares. brand equity. 1. this issue has not been addressed in Once again. while. we also decompose cantly the advertising effects on choice and quantity. it is important to go beyond share In other and 2 this In addition Our analysis captures the long term effect of promotions and ad- vertising on brand intercepts which have been interpreted as a measure of brand equity (Kamakura and Russell 1993). .an approach to address this issue. Inclusion of competitive reactions is also a relatively new aspect of the paper which is absent from studies addressing the long-run issue (e. These competitive moves may offset each other leaving the market shares unchanged. If promotions make con..e. 1998) focus on category-specific. Our re. 18. for three of the four brands set analyzed.studies. Although of researchers have decomposed the number a effects of 3 brand while promotion hurts it. vious 6.more price sensitive. promotions.when both short..” (page 172). our previous discussion (regarding current expenditure levels) suggests why it may not be appro-for a brand to continue increasing priate its advertising . long-run effect of advertising on price the sensitivity).of marketing policies on purchase quantity. Finally.. and Krishnamurthi 1996). (1998) and Papatla and Krishnamurthi (1996). we find that these effects vary signifi-by brands.” effect of advertising on price elasticity . In this scenario. We capture the main effects of Equity: advertising and promotions on consumers’ purchase behavior.. then competitors may sumers intensify promotions. . No. not brand-specific. AND GUPTA Managing Advertising and Promotion for Long-Run Profitability two major limitations as the authors themselves indi. We find that the depth elasticities are larger than frequency elasticities in the run but become smaller when long-run short effects are considered. . Second.First. 1998. our study provides richer insights for both researchers and managers . words. 7. in general. 2 Results indicate that advertising has a longrun positive effect on brand choice while the opposite promotions. Regular Price Versus Is it better for a brand Promotion: regular price and offer price to raise its promo. to raise prices and lower promotions it is better (i. . MELA. Further. pre. For our vides data we find that.. these brands have been hurting their profits on both price and promotion dimensions). 3. issues. and Krishnamurthi 1996). MGL. . Quantity Decisions: to brand choice. infrequent. previous papers assessing the long-term effect of promotions and advertising on choice (e.of promotions (depth and frequency) are fects about two-fifths of the short-term effects.. “we did not analyze . Mela et builds al. Papatla. long-term ef.JEDIDI.. Promotion Depth Versus Is it better to offer Frequency: deep discounts or frequent. for the other it is better to lower price and lower brand. Mela et al.and long-run effects are ticity considered.g. others accounted for interactions and not the main effects (e. DeKimpe and Hanssens 1995a.e. 5.these reaction functions. the (page . our paper explicitly captures the ef. . As expected.g. our measures are only short-term cated. 4. elas. et al. inclu. Competitive We explicitly model and account Effects: for competitive reactions in assessing the longrun impact of marketing decisions. Note three equity things.g. many studies (e. although consumers have become more price sensitive over time due to increased promotions. even if advertising has a positive effect on Third. on average. elasticities. Advertising may have longer term effects172). fects This on previous studies such as MGL. Due to modeling and technical limitations. For example. First. . MGL. In a loose sense this holds for confirms managerial intuition that advertising enhances understand changes in consumer and competitor the brand better off offering lower tions or regular price with limited price promotions? Our paper pro. By explicitly studying these changes.and long-run tradeoffs. shallow discounts? To the best of our knowledge no other study has addressed this tactical issue empirically in context of long-term effects. promotion and advertising. we find that. 1999 . while some studies (e. Papatla. Marketing Science/Vol. . Mela. Brand 2. Kamakura and Russell 1993) captured the main effects and not the interactions (i. 1998 and run Gupta quantity accounts for the majority of this 1988).g. Second.

Conversely. the parameters and bbqj t jt should vary over time and be allowed to change with changes in long-term marketing strategy of a brand. 407–419 for a discussion). Previous studies (e. Our modeling approach may summarized be as follows: • Consumer’s choice of brand c jt 3 Strictly speaking. price and promotion of various and consumer-specific factors (e. Krishnamurthi and Lee Raj show that ignoring this correlation can lead 1988) to biased parameter estimates. Selectivity bias refers to the bias in parameter estimates that results from ignoring the dependence between the choice and quantity models. long-term variables do not capture all the changes in these parameters. Section 3 discusses estimation issues.g. extensive advertising overyears may make consumers less sensitive to the short.price discounts. the choice and quantity models to be correlated due to omitted variables which affect both these decisions (Dubin and McFadden 1984). However. consumption rate. as we will discuss shortly. are qthe short-term variables affecting Xj t pur. frequent term discounting may make consumers more price by a brand sensitive. However. 1988).quantity (some of these could be the chase same variables that affect choice. this study provides significant insights about managing advertising and promotion for longrun profitability of brands—insights that were not available from previous studies. We would like our model to accommodate three key c characteristics.97. To capture these key features. MELA. j at time tf () c Xt bj tcj • Consumer’s quantity decision given choice of bran j at time t g ()bqqj q j t X jt t d 4 Marketing Science/Vol. predicted expected quantities from each model were within 1% while estimated elasticities differed less than 0. However. Specifically.and draw managerial implications about lations the long-term value of promotions and advertising.g. regression models assume quantity to be contin- uous. The models are estimateda using maximum likelihood approach. Raj the 3 Zc t j 2. 8. and and are long-term marketing variables (e. In sum.. ours is the first to conduct a similar Gupta analysis for advertising. Sec. we use these results to conduct simu. For example. quantity for many packaged goods (such as the one we analyze) is actually discrete. pp. brand loyalty). brands may alter consumers’ sensitivity to short-term marketing actions. AND GUPTA Managing Advertising and Promotion for Long-Run Profitability promotions across behaviors (cf. in our application. In addition. cccc • Choice bcZj t j t hc () jtj parameters Quantity qqqq • bcZj t j t hq () jtj parameters where are the short-term marketing variables Xc t j affecting brand choice (such as weekly price and e e promotion).03%.. Model A consumer’s decision of which brand to buy and how much quantity of that brand to buy depends on brandspecific factors (e. In § 6. Finally. First. such as price). Section 4 describes the data and § 5 presents the results.g. 1. when in fact.the model should allow the error terms in ond.. of the paper is organized as follows.and Trost 1978. Conceptually thissimilar to the varying-parameter regression is models developed in econometrics (see Johnston 1984.and ing econometric literature. Bucklin et al. This would e ically make the choice and quantity models heteroscedastic. 18.g. The error terms and are j t ecq jt specif-included to capture this aspect. This suggests that consumers’ sensitivities to short-marketing activities can vary over time as a term function of long-term marketing actions.. Methodology Finally. In § The we2 describe our model.7 presents conclusions and offers directions tion for future research. brands) consumer’s brand loyalty. and his/her sensitivity to price and promo-Further. long-term marketing activities of tion). product inventory. This model is new to both the market.g. the error terms in choice model should be correlated across brands the to avoid the IIA assumption.g. the specification of a continuous quantity model has little practical effect on our results and greatly simplifies model estimation.JEDIDI. 1999 .. No. Lee 1982. Lee and Trost 1978). we found the correlation between parameters estimated from regression and parameters estimated from an ordered logit to be 0. Zqt j ing) advertisand consumer-specific factors (e. the paper provides a : odological contributionmethby developing a varying parameter multinomial probit and regression model with selectivity bias. Conceptually this is a straightforward extension of the approachLee and Trost (1978) and Krishnamurthi used by and (1988).. we use a heteroscedastic probit model for brand choice and a heteroscedastic regression model which controls for selectivity bias for the quantity model (e. Sec.

Next we allow the short-term thesensitivities(including be affected by a intercepts and cq tijto bi j0 ) 0t brand’s marketing actions. It is now straightforward to develop a two-stage estimation procedure. LL U ij t i j kt i j kt i j t bcc c X k where 1.Xis brand j’s 0quantity intercept. the proceeds to brand a certain quantity of that brand. 1. Next we ˆ and Vˆ j t i imize log Lj . As of because the high nonlinearity of the likelihood function 1983. No. p. We first develop a varyingparameter maximum likelihood probit method to esˆ timate the choice parameter H c by using all Hc ijt (5) si j t by (2) qq where 1. Marketing Science/Vol.the quantity of cally. L )isthe lth short-term marketing vari. As brand terms andc qcould bethe error correlated due to ijtijt omitted variables which may affect a consumer’s decision which brand to buy and how much of both quantity brand to of that buy. the conditional likelihood function for brand j’s quantity data is: 1 Qi j t i j t s Lj u tQ 0 i ij jtt i j t u where (•) is the standard normal density function. the discrete choice modeling stream. Ideally. this is very difficult H q ).. b t i j0tij and (l q Xi j l t 1. a simple regression be used to capture this behavior. Although this is possible. j.. we develop a consistent two-stage maximum likelihood procedure for estimating our model. We then use toH c ˆ of the W i j t(•) calculate (•) for all i. (Maddala as (H c . the choice of a Given j. buy Following Krishnamurthi and Raj (1988).. choice j of 1 are derived j. Specifically. (3) . (4) Wi j t we bqq lqqj q c ij t l c 0 m j l m i j mt i j l t Ze where ccq j l m are moderating jand km parameters. Following we assume that consumers choose a brand to maximize choice utility. ijt ijt ijt Ze . max. t (see Appendix B). cc c brand 0j’s choice X i j0tij is b tijkt intercept. iance Note L depends on both the choice and that j parametersquantity . price of a bc associated parameter sensitivitiesshort-term to the i j kt re ecting consumers’ consumer marketing activity. MELA. Lee and 1983.. the utility their brand of bran j for i at purchase d by consumer occasion (1) X 3. j 1 . Equations (B-1) and (B-4)).(see Appendix shown in Appendix rived A. j and 1 . Trost (1978)... K )isthe kth short-term marketing variable affecting i’s brand choice behavior consumer brand in a certain week). model may Specifi.. is the conditional mean and is the u2 conditional i at var. Lee 1982).decision of how able affecting i’s consumer quantity of j to buy.. (A-10) to get asymptotically e efficient estimates and correct standard errors for both the choice and the quantity model H parameters mentioned above.g.. and is the (e. the likelihood function can be de. j bought brand t is given by consumer time Q i jt i j l t i j l t i jt bqq q X l Given the normality assumption. To formulate this procedure. muchindicated earlier. This effect is captured b long-term as follows: cc bccktc c ij j k c 0 m j km i j mt ij kt vector observations. obtain the conditional expectation and we need to the conditional variance Qi j t given of These conditional moments for brand brand in Appendix B. However. the standard errors of the estimates may notexact since the choice parameters are computed be independently of the quantity parameters and because the second stage estimation ignores the fact that V .. Alternatively. function in Equation (A-10). Lee and Trost 1978). and Krishnamurthi and Raj (1988). 5 . (k 1.we can maximize the joint likelihood eters. 18. such t is given a procedure can be very cumbersome (see Maddala 224)..given J the estimated W (•) and Vˆ (•) to obtainvalues estimates for the ˆ bias parameters selectivity quantity and the r qc j model parameters . The parameter estimates produced by the twostage procedure are consistent (Heckman 1979.. Estimation To obtain estimates for the quantity and choice param. the errors in Equations (1)–(4) to be c Assuming normally distributed.. J are estimated. Following Lee (1982). AND GUPTA Managing Advertising and Promotion for Long-Run Profitability details of the model derivation and estimation are quite different from the previous approaches. 1999 should maximiz LL in Eq.JEDIDI.

The model is estimated at the brand rather than the brand-size level for several reasons. management believes most mar. no brand entries or exits occurred over the data period. The data were collected in a medium sized Midwestern market. AND GUPTA Managing Advertising and Promotion for Long-Run Profitability suggested by Lee and Trost (1978.JEDIDI. LL in only market share in this category (our selection procedure to Krishnamurthi and Raj (1988) who is similar chose brands comprising 80% share). Second. 18. The demographics of the static panel do not deviate substantially from the national averages. this rather approach is consistent with many previous studies (e. Basic descriptives for the are given in Table data 1. Second. and did observations in which households did not purchase one of the four were eliminated from the analysis. 4 Details of variable operationalization are given in the next section. Households’ me. 7 for details). price (PRICE) and price promotion (PROM) sensitivity parameters are further reparameterized as functions of long-term advertising long-term promotion (LTPROM). because of the normality assumption and the er. Households three who not buy any of these four brands. Specifically. randomly sampled about half of We therefore the households.1. specifying a utility function that outlines how consumers respond changes in the short-term. two sizes and switching between all sizes is common. p. store.2. productlife and product introduction factors are not likely cycle to impact our analysis. demographic. In addition. As a result. the the initial 2SML estimates have the same asymptotic properties as thosethe single step of MLE.g. First.estimates all of the variance components of neously the quantity model. The panel is a static panel (no households enter or exit) and consists households observed over an eight and of 1. Fourth. Unlike OLS. 4. mature We used IRI product category. 1. ML regression explicitly accounts for the heteroscedasticity of the error terms and simulta. Papatla and Krishnamurthi 1996). (7) Note we are suggesting that advertising and promotion policies may affect choice (and subsequently ways: via a direct effect on brand profits) in two choice probabilities (changing intercepts) and via an indirecton a household’s response to price and effect promotions (changing sensitivities).in parameters. There are four brand-sizes in our data. The large quantity of data (both observations and brands) makes model estimation virtually intractable. and then specify to price a model of how consumers adapt their responses by 4. Data 4.590 one quarter year period running from 1984 to 1992.and mean interpurchase times are 6 and 12 dian weeks respectively. 1999 . utility of j for brand 4 consumer i at time t is defined as U i j t i j bcc0tijc bc b 2t j tPROM (6) 1 tjtijPRICE ijt where the intercept. Because parameter values are consistent. the large number of brand alternatives makes model estimation size considerably more difficult. Additionally. for the same reason. and (LTADV). The data are comprised of panel. The medium account for over 75% of all purchases. First. and trip data. consistently estimated by the inverse of the hessian (see Krishnamurthi and Raj 1988. we account for size in our quantity analysis. we use a varying-parameter ror multi. we use a varying-parameter ML regression instead of OLS to estimate the quantity and the selectivity bias parameters. Third.actions are intended to promote the keting brand than a specific brand-size. MELA. No. Marketing Science/Vol. Several features distinguish our estimation approach from that of Krishnamurthi and Raj (1988). we adopt a twostep maximum likelihood (2SML) estimation which utilizesNewton-Raphson method to the maximize one iteration using the consistent estimates as starting The asymptotic covariance matrix is values. p. Descriptives scanner data for a nonfood. This left brands us with 691 households who made 13.664 purchases.probit method instead of a logit to estimate nomial the choice parameters. we confined our analysis to four largest brands which comprise 71% of the the 6 to changes in advertising and price promotion policies over time. brand loyalty (LOY) as follows: bccckt j k c LTAD j t ij 0 k1 c V cccc LTPROM j t k k2 c 3 LOY i j t i j kt e . Variables Choice Model We begin Specification.. 368) and Krishnamurthi and Raj (1988).

8 4 29.10 0.4 29.35 0.8 Promotion Depth (% off) 1 11.used in MGL. price is operationalized as the lowest price per ounce acrossbrand’s different sizes. Operationalization of Choice Model Variables. The interaction of loyalty with price and promotion sensitivities captures this effect.050 0. represents approximately 48 weeks of purchases and is the du.055 3 0.6 3 10.61 29.2 Advertising 1 Mean for 1988–1991 1984–1987 ($) 1 66.13 3 0.of (i. 1999 Table 1 Descriptive Statistics of the Data Mean for Variable Brand Market Share 1 0. 18.16 0.for heterogeneity in consumer preferences for trols different brands (Guadagni and Little 1983).25 1 Advertising represents average in ation-adjusted advertising dollars in thousands spent in a quarter. The price (PRICE) of a selected brand is the regular (nonpromoted) price per ounce. are likely to be less price sensitive than nonloyal consumers.3 17. self perception theory choice. advertising strengthen brand image and build is likely to equity (Aaker 1991).to avoid in uencing the loyalty measure chases with promotional purchase events (Lattin 1990).054 2 0.9 2 12.056 4 0. the PROM variable contains information about the pres. As with price.e.price across brand-sizes re ects the lowest mum price available to a given household for the nonselected minimum price operationalization is brand. As consumers that commonly switch across brand sizes in this category.78 2 25.60 29. Loyalty (LOY) is defined as a household’s share of purchases of a brand over the last four nonpromoted The four period purchase cycle purchases. frequency) and the magnitude of ence (i.58 3 28..70 4 28.98 12.2 25. Moreover. 1978). regular price elasticities.052 0. the mini. We used only nonpromoted ration formulation (analogous mum minimum price) for nonselected brand to sizes.. i. In contrast. is also significant theoretical support for There the indirect effect of advertising and promotions on consumers’ choice.26 26. For suggests that consumers who buy on promotions are likely to attribute their behavior to the presence of promotions their personal preference of the and not to brand (Dodson et al.55 3 45.4 2 8.11 0. 1994).4 33.27 4 28. In other loyal consumers (almost by definition) words.72 28. price promotions thereby enabling us to depth) disentangle the effects of frequency and depth on brand and profitability. Frequent use of promotion is therefore likely to reduce consumers’ intrinsic prefer. excellent of marketing studies which also summary conclude similar interaction effect of advertising on price sensitivity.that promotional price elasticities may dence exceed Marketing Science/Vol.3 3 13.e. on the other is likely to reduce product differentiation hand. and therefore increase consumers’ price sensitivity et al. The PROM variable enables us to accommodate this possibility. For each brand. it may hurt “brand ence for equity” (Kamakura and Russell 1993).20 28. we use a sales maxi.12 Purchase Quantity 1 27.13 4 0.051 0.JEDIDI. Blattberg et al.1 17.04 30.the brand. Increased use of price promotions.7 32.36 2 0. AND GUPTA Managing Advertising and Promotion for Long-Run Profitability Several theories support the existence of a direct effect of advertising and promotions on consumers’ example.053 Promotion Frequency (% of occasions) 1 15.e.8 20. The betterweighted averages at capturing brand-level than price variance in categories where size switching is common. No.52 17. (1995) outline evi. Economic theory suggests that advertising leads to product differentiation which reduces consumers’ price sensitivity (Comanor and Wilson Kaul and Wittink (1995) provide an 1974).048 0.42 Regular Price per Ounce ($) 1 0. Price promotion (PROM) re ects the discount offered by the brand. MELA. 1. assess the impact of long-term promotional To or 7 .41 per Occasion (oz) 2 26.3 4 6.. The main effect of loyalty simply con.8 16. These effects are expected to (Boulding be moderated by consumers’ loyalty to brands.

some not 5 5 which is 6 t is advertising. AND GUPTA Managing Advertising and Promotion for Long-Run Profitability advertising activity of a brand on consumers’ price and promotion sensitivities. the intercept and sensitivity parameters are specified to vary as a func.9. However. a week.t 1 is the quantity bought by i on e store visit t household 1 is the interval of time 1. since our focus is on weeks in assessing the long-term (not weekly) impact of advertis.t 1 CR i* I . t 2 k PROM •••. Inventory was computed and then mean-centered as in Bucklin and Gupta (1992). Quantity Model Specification and Variable Operationalization. Specifically.7. Specifically. t store visit t 1 and t between i is and CR the average weekly consumption rate for i.44) in MGL.. 18. we also estimated our choice model using 0.t 2 k AD . we defined these longterm variables as a geometric series of past promotional and advertising activities. 1999 . 6 cq q l3 7 LTAD j t l c 2 LTPROM V LOY i j t i j l t e . Mean householdpurchased by a household was included quantity (MQTY) to control for heterogeneity in households’ buying pat. Ii .t j 7 Using the decay affect our results significantly. (in ounces) of the product bought by a household over the entire duration of the data divided by the total number purchases over the same of its period. t 3 as PRICE b 2 tjt PROM . our approximation does not appear to be a serious limitation. the mean weekly price promotion parameter and activity of a brand across stores. Accordingly. Since our unit of firm anal. ( a y basis. Consistent with the choice model.t 1 (11 . LTADV jt j ADV k23 AD V and LTPROM PROM jtj .of long-term tion promotion and loyalty as follows: q bqqqcj l ijlt c 0 parameter Since price and promotion vary on advertisingweekl to be 0. Households This formulation is consistent with the Koyck model specification which has been extensively used in the literature to study the carryover effects of advertising (Clarke 1976.97 for each week. MGL also find that the lag values for advertising and promotion were not significantly different. Given that previous studies have shown negligible short-term effects of advertising (e. Once the weekly advertising decay parameters are defined.t 1 k2 3 PROM j.84) and the mean quarterly lag (0.97. long-term To test the stability of our results acrossdiffering values of the decay k values of parameters. Using (1976) estimated the decay quarterl y k) data.6. we chose a decay factor of 0. .6 after one quarter. the variable and long.JEDIDI.g. Model fit was best with a lag of 0.t 3 q bq ctit jIN b 4t i iMQTY j3 jt V (10) Price (PRICE) and (PROM) have the same operationalization as in the choice model. The quantity of j that brandt is specified consume i buys at time r Qi j t i j q bqb qij 0t 1 tjtij j.8. All of the coefficients and most of the standard deviations were very stable for the various lags. 1. INV i t i Q k PROM .deviations from this assumption should ing. 8 Marketing Science/Vol. 0. (12) l1 jt We acknowledge that it would be better to use actual weekly ad- This decay or lag value is between those reported by Papatla and vertising data. Quarterly. in ation-adjusted advertising expenditures were furnished by the advertising agency of the that supplied us the data. Leone 1995).t 1 V k ADV •••. data limitations preclude us from using it (note single source data did not exist in 1984). we similarly obtained the long-term promotion variable (LTPROM) from Equation (9).advertising variable (LTADV) for any term week obtained using Equation (8).t 1 i INV ) wher Q i . equivalent to a decay close to 0. price Inventory a household was included to capture the (INV) of impact of stockpiling in previous purchase occasions on future purchase quantity decisions. our unit of analysis is a week. (8) j. (9) j. No. Clarke of observing a price for their brand choice also face the same when deciding how much to buy. Krishnamurthi (1996) for an average quarterly purchase cycle (0. MELA.This variable was defined as the total amount terns. 0. This procedure implicitly advertising to be constant over the assumes thirteen a quarter. we created an approximation of ysis weekly advertising spending (ADV) by dividing the quarterly advertising by 13 weeks. Tellis 1988).

001 4.251 8. the long-term effects of promotions and advertising.079 0.882 0. brand 1.319 0. Dev.137 0. Table 2 Results of Model Estimation Choice Model Quantity Model Variables Parameter Estimate Standard Error Parameter Estimate Standard Error Intercept Brand 1 0 Brand 2 Brand 3 Brand 4 Price Brand 1 0.963 0.1.965 0.509 Brand 4 Std.090 0. 5.462 0. However.173 0.700 8. 1999 9 .106 0.365 0. Dev. 8.e.152 0.this vein. the brand intercepts capture “the additional utility not explained by measured attributes” and have been used as measures of brand equity by several researchers (Kamakura and Russell 1993).618 0.255 0.143 0. 0.340 0.876 the intercept term is but one component tured of brand equity (Swait et al. Therefore. 1.001 2.098 Long-Term Promotion Loyalty 0.510 Brand 2 Std. Results 5. 0.088 There is some controversy about whether the intercept in these models really captures brand equity. Choice Model Short-Term Table 2 shows that three of Effects.840 8. Gupta 1988).100 0.130 Brand 2 Brand 3 Brand 4 Price Promotion Brand 1 Brand 2 Brand 3 Brand 4 Main Effect Long-Term Advertising Long-Term Promotion Loyalty Brand 1 Std. be.099 Long-Term Promotion Loyalty 0. Dev.305 Moderators of Price Promotion Sensitivity (Interaction Effect) Long-Term Advertising 0.020 Brand 4 Std.154 0.619 1. The market leader.405 5.898 Brand 3 Std.338 0. Dev.555 6. Dev.001 1.288 0.100 0. Table that both promotions and advertising have 2 shows a significant long-term impact on brand intercepts.244 2.088 0. 0. one would expect advertising to In increase brand equity due to the positive brand messages 8 01 3 7.158 0. has the rectly low-regular price sensitivity. Dev.933 0. Guadagni and 1983.166 0. Dev. AND GUPTA Managing Advertising and Promotion for Long-Run Profitability Finally.JEDIDI. Dev.307 0.018 0.075 0.179 1.727 2.123 0..073 0.148 0. 2 1 5.094 0.549 1.400 0.428 0.090 0.835 0.142 1.a large number of studies which show tent with significant short-term effects of price and promotions on consumers’ brand choice behavior (e.signed.079 0. we will use the term brand equity somewhat loosely to indicate an increase in base choice probability. MELA. 0. Both views imply in the brand intercepts contribute to the changes over-equity of the 8 all brand.019 1.063 0.921 Brand 2 Std.175 0.076 0.172 0. Finally. 0. Dev.263 Brand 4 Std.122 Brand 2 Std. Dev.276 0.162 0. 1.760 Brand 3 Std.131 2.153 0.066 0. Long-Term Of greater interest are Effects. i.594 1.advertising and price) were then created by tween calculating the product of these standardized variables..073 0.503 0.181 0. Dev.000 1 0.175 0.103 0. No.057 0.272 0. Marketing Science/Vol.g.109 Brand 3 Std.818 1. This was done to facilitate zero the comparison of effect sizes.278 0.100 0..752 0.849 5. Interaction effects (e.075 0.181 7. Recall that the intercept in the brand model represents the base probability of choice purchasing a brand controlling for price and promotional activity.111 Brand 1 Std.455 0. 18.739 Moderators of Price Sensitivity (Interaction Effect) Long-Term Advertising 0. four regular price terms are the significant and all are cor.106 0.111 Brand 1 Std. These results are consis. 0. 0.367 1.060 0. they have a mai effect on consumers’ significant n brand choice utility.960 0. All four promotion est terms are positive and significant.g. the covariance Little matrix errors show the non-IIA structure for the across brands.170 0. Dev.075 0. all independent variables (in both the choice and the quantity models) were standardized to meanand unit variance. 1993).057 0. researchers have suggested that the equity Other cap. 6.476 0.

Table 2 shows that long-term promotions make consumers more sensitive to changes in regular price but sensitive to promotional discounts. the quantity model are significant.095 0. It is sensitivity terms in interesting to note that brands 1 and 4 have the two smallest price parameters in choice.JEDIDI. 1’s price promotion parameter is positive Brand and significant. 1990).076 1 Fixed for identification purposes. consumers will need an time.184 1. even discount to react positively to a higher brand. the main effect of longterm Marketing Science/Vol.. 3 Significant parameters are highlighted in bold. advertising is insignificant. This will not reduce consumers’ sensitivity to promotional discounts over In other words. after controlling for the observed variables.99. the effect of brand 4’s price promotion on quantity bought is negative and significant. Consistent with this expectation we find that the advertisingpositive and significant.791 Brand 4 1.g.251 1.49 may be acceptable to consumersthe same regular price may seem too high in year 1. In sum. The interaction of long-term advertising with promotion was also insignificant. This suggests that regular price cuts of brands 2 and 3 lead while consumers to switch to these brands. 5. the choice and quantity decisions are largely independent in this category.131 1. our results show a negative and significant main effect of long-term promotion on consumers’ brand choice utility. over to reduce brand equity (Blattberg and Neslin 1989. Many theories effect is suggest the long-term. but they have the two largest price parameters in quantity. it may be considered a gain in year 4.487 1.368 Log-likelihood 8118. An by brands aggregate sales elasticity (e.907 Avg. suggesting that its deals increase the quantity bought.024 Brand 3 1. mean quantity purchased has a positive and significant effect while household inventory has a negative and significant effect on households’ purchase quantity decisions. Contrary to expectation.g. 1999 Selectivity Bias Brand 1 1. Inspection of the data reveals that brand 4’s price deals are nearly exclusively for smaller sizes to this unexpected result. often embodied in national brand advertising. Frequent less promotions may increase consumers’ sensitivity to regular by lowering price (Kalyanaraman 10 their reference price . such price changes 1 and 4 make consumers buy more. This would suggest that while from a regular price of $1. For example. Quantity Model Short-Term Only two of the price Effects. However. AND GUPTA Managing Advertising and Promotion for Long-Run Profitability Table 2 (Continued) Results of Model Estimation Choice Model Quantity Model Variables Parameter Estimate Standard Error Parameter Estimate Standard Error and Winer 1995). frequent promotions of Coke may lower consumers’ reference price of Coke $1.022 Brand 4 0.269 0. Consistent with this view. 1995b) or Papatla and Krishnamurthi 1996) would not be able to uncover this interesting phenomenon. 18.46 0.964 2.541 0. As expected. Purchase Quantity Inventory Covariance Matrix for Errors Brand 1 Brand 2 Brand 3 Brand 2 0. 1.49 to $0. price promotions are likely that. Long-Term While the main effect of Effects. In § 6 we will expand on the managerial implications of these results. MELA. Economic theory as well as previous studies suggest that advertising reduces consumers’ price sensitivity.258 Brand 3 1. discount may be considered a significant “gain” in year 1. it seems that increased price promotions and reduced advertising have a negative main effect on brands’ value and choice . The selectivity leading bias terms are insignificant for all four brands suggesting that.960 Brand 2 0.080 6.00 49053.266 0. Standard deviation of the error in a brand’s intercept or response param- 2 eter..2. for our data set. No. in 4. this effect was not significant. Frequent promotions (say 50 cents off) are year also to increase the reference discount level likely over This would imply that while a 50 cents time. DeKimpe and Hanssens a focus on only brand choice (e.

1999 jj h1 . Competitive Reaction Functions Before assessing the impact of changes in a brands’ marketing activity.prefers Coke. Post’s • Given a fixed budget. consistent with this them lie wait heuristic is the finding that the long-term in effect of promotions on promotion sensitivity is also positive. is it better to offer frequent. In this formulation. this consumer is likely to stock up on his/ her favorite brand. In this section. they conceptually 9 while represent Inclusion of lagged competitive activity did not significantly im- prove fit. s/he does not have to switch sumer to Pepsi since this consumer expects Coke to be on promotion soon (perhaps next week). First differences for e changes promotion and advertising are defined similarly. it still leaves our key research questions unresolved. that long-term promotions make Note consumers less promotion sensitive in choice but more promotionin quantity. these changes in policy.or shallow infrequent.JEDIDI. DPRICE jt 6. This suggests that in response to repeated expo. consumers learn to lie in wait sure to for especially good deals and then stockpile when they see (Mela et al. Managerial Implications Our previous discussion highlights the directional ef. the efficacy of these discounts may be diminished significantly. we would like our results to help address the following important us questions: the negative long-term effects of • Do promotions offset their positive short-term effects? are the relative effects of advertising • What and promotions on choice and quantity? are the managerial implications of these • What results for resource allocation across advertising and promotions ? • Is it better to lower regular price or to promote more often? The problem is illustrated by a major pol-shift by Post cereals in 1996. we estimated the following competitive reaction functions on first differences using 9 OLS. it is important to consider the potential competitive responses they may induce. When Coke offers a promotion.of advertising and promotions in the longfects term. Specifically. advertising effects are measured at the quarterly level. If competitors respond to those discounts (a very likely scenario). 11 . No.j j t DPRICE h2 j j t DPROM jj e1j t . respectively. if Pepsi is on promotion this week and a con. This suggests that frequent sensitive promotions of brands makes it unnecessary for consumersbrands (as it becomes increasingly likely to switch that on the favored brand will be forthcoming) a deal but makes them more likely to stockpile when their favorite brand is on promotion (because they fulfill a greater portion of their demand in promoted periods). we did not specify advertising reactions in the pricing/promotion equations and pricing/promotion reactions in the advertising equations for two key reasons. 1. Further.i) an understanding of how competitors will quire re. This sparked other manufacturers to follow move. In addition to an understanding of consumer ior behav(modeled earlier). (14) e3j t . we address each of these three steps. Second.affect consumers.t 1 representsth jt DPRICE in price over time. First. pricing/promotion effects are weekly. 1996). For example. For example. AND GUPTA Managing Advertising and Promotion for Long-Run Profitability promotions (captured through the intercept) is positive.1. MELA. answers to these questions re. and iii) a comparison of tions the incremental response to the incremental cost of these policy changes. deep discounts? Marketing Science/Vol. simulating the effect of an increase in discounts in the absence of competitive reaction could lead to an optimistic assessment of the effects of discounts. However. Post reduced their icy discounting level and cut their regular prices by twenty percent.promotions. 6. Following Lee ang and Wittink (1992. ii) a simulation of act how policy changes and ensuing competitive these reac. (15) jj DAD V jt jj where PRICE j t PRICE j . 1998). (13) DPROM jt jj h3j j t DPRICE h4j j t DPROM h5j j t DAD V e2 j t .

1.694 — 0.027) (0.020) (0.021) (0. P refers to PRICE.007 — (0.017) (0. 1999 . we conducted a market Table 3 Competitive Reactions—Parameter Estimates and (Standard Errors) A.011) (0. Market Response Simulations: Procedure To evaluate the effects of changes in marketing policy on brand sales and profits.004 0.009) — (0.014) (0.165 0.021) (0. Finally.003 — 0.019) — (0.186 — (0.016) (0.247) D AD 4 0. it seems less likely that a price war will spark an much adver.024) (0.022 — 0. The results of this analysis are presented in Table 3. 6.025) — (0.024) 4 0. For example.072 0. Advertising Reactions Effect of Effect on D D AD 1 AD 1 D AD 2 D AD 3 D AD 4 — 0.022) (0.006 0.003 — 0.012 0.016) — (0.016) (0. AND GUPTA Managing Advertising and Promotion for Long-Run Profitability distinct marketing practices. 12 Marketing Science/Vol.225 0.016 0.024) 0.153) (0.035 D P2 DP 3 0. Significant parameters are highlighted in 3.034) 2 0.333 0.148 0.009 0.000 D PR1 D PR2 D PR3 D PR4 — (0.142 (0.007 0.394) (0.027) (0.189 — (0.017 (0.037 0.019) — (0.019 (0.021) — (0.034 0.023) — (0.023 0.010) (0.404) — (0.035) (0.174) (0.018) (0.011 — 0.013 — 0.036) (0. Price and Promotion Reactions Effect of Effect on D DP DP DP DP P 1 — 0. However. No.032) (0.083) (0. Brands 2 and 4 do not react to changes in price and promotions.031 — 0.031) — — 0.021) — (0.021) (0. in response to an increasepromotional activity of brand 3.140) — 1.023 0.085 D P4 0.039 0. The results show a high level of competitive reactivitybrands 1 and 3. Brand 1 has a similar effect on brand 3.019 3 (0.020 — (0.013 0.013 — 0.006 — 0.013) — (0.020 0. these brands show competitive reactions in their advertising.018) 0.586 0.507 0.234) 0. Further.014) — (0.011) (0.010) (0.018) 0.017 — 0.013) — (0.016) (0.024) (0.018) (0.009 0.044 — 0.150) (0.009 0.006 0.140 0. MELA.response than a price tising response. 18. brand 1 in the reduces its price and increases its promotional activity.012 — 0.016) (0.287) (0.015) (0.269 D AD 2 D AD 3 (0. 2.026) (0.043 0. bold. a reduction in the advertising of brand 3 leads to a reduction in the advertising of brand 1.014) — (0.212 0.024 0.205 D PR3 D PR4 0.015) (0.2.013) (0.004 0.068 0.103) 0.022 0.015 0.093) — (0.001 — D PR1 D PR2 0.746 0.019) — B.023 1 — (0. Intercepts are not reported to conserve space. A decrease in the price between of brand 3 results in lower regular prices and higher discounts for brand 1.010 — 0.015) (0. and PR refers to PROM.JEDIDI.

we need to separate the impact of price promotion on choice and quantity. we estimated history of the probabilit and the base (conditional y choice) for each brand. Next. on average. in price promotional frequency is preferable to an increase in the discount level. Three key emerge the price elasticity for the four brands ranges total from to 1. In the same scenario. 18. 1999 promotion variable. impact of a regular price change by a brand. Choice probabilities for the target in Table brand then recomputed using the manipulated were prices. findings Table from this table. Price Promotion While Elasticities. Simulation Results Price Simulation results are given in Elasticity. 1. and advertising only ap. Using the updated longable term Marketing Science/Vol. MELA. 4. a 1% increase in the frequency price of price promotions would imply a promotion frequency (101 discounts over 1000 weeks). level For example. thus yielding the combined price elasticity. shortterm effect gave us an estimate of the long-term effect of promotions.reacted as indicated in Table petitors 3.1% (one) discount would be randomly mental distributed across the remaining 90% of the nonpromotedin this example) and its discount weeks (900 level would be 20%. we obtained the total (shortplus long-term) effect of promotions on choice and quantity and quantity effects were separated (choice following the procedure used for price and advertising). This meant (a) updating the short-term promotion variable by increasing either the frequency or depth of promotions. if a brand offered an average 20% discount the time (over 1000 weeks this would imply 10% of 100 promotions). appears only as a short-term variable. In each simulation. but the long-term promotion variables at their kept original a long-term variable in our model.2%).quantity household and brand purchase occa. Comparing these results with the baseline estimates. The of 10.. Subtracting the (choice) elasticity from the second (total) first provided the quantity elasticity. This calculation yielded base sales estimates against which the effects of changes in policy can be judged. The simulations proceeded by first calcu.The expected quantity for each brand was sion. Calculating Base Level Choice Probabilities and Quantity.long-term. we assumed assess com. A similar simulation was performed where price was reduced in both the choice and quantity models. 6. We and the increased price promotions by 1%. then computed by multiplying the choice probability by quantity conditioned on choice and summing across occasions all for all households. Price and Advertising To Elasticities. These the were used to estimate the choice probabilities and then conditional quantities. and adjusted the regular prices and promotions of the other brands by using the competitive response functions reported 3.3.43 with an average of 0. price pears promotion appears both as a short.0.1% incre. The simulation was repeated to assess the short-term impact of promotion on choice and quan-The difference between the total and the tity. 0. No. an increase in the of price promotions would be simulated by depth increasing each discount by 1% (in this per equation (9). we increased the short-term promotions by 1%. Therefore. AND GUPTA Managing Advertising and Promotion for Long-Run Profitability simulation.79. To assess the effect of advertising on choice and quantity we followed the same procedure . Note that the forgoing procedure relies on an increase in either the frequency or depth of promotions. Second. Using the parameter estimates from the choice and quantity models and the actual purchase each household.and long-term variable. the frequency of promotions by 1%. First.e.expected shares and purchase quantities in lating the absence of any changes in marketing policy. we the reduced the price of the target brand by 1%. To increase we randomly selected nonpromoted weeks and inserted the incremental promotions (using the brand’s mean of discount) into those nonpromoted weeks. we computed the intercepts and response parameters as per our model.JEDIDI. By simulating both frequency price and we can ascertain whether or not an increase depth. both in the short. to an average of 20. choice probabilities multiplied with The new the brand’s base quantity (i. This was done as follows. choice for about 75% while quantity accounts accounts for 13 base choice on .37 decomposition of the total price elasticity into choice and quantity components shows that. calculated without a price cut) yielded an estimate of the choice elasticities. and (b) updating the long-term promotion vari.

036 0. 18.0504 0.002 0.0434 Long-term 0.0046 0.0239/0.6%) across brands.0252 0.001 0.0120 Brand 2: Short-term 0.0334 Total 0. (1995a) and 0. that advertising elasticities vary with an average of 0.072 Promotion Frequency Elasticity Brand 1: Short-term 0.624 Brand 4 0. found elasticities to be significantly larger (almost price ten times) than the advertising elasticities.quantity behavior is quite similar to the chase findings of Bucklin et al.0290 Total 0.0230 that almost all of the advertising effect sults is on consumers’ choice decision. No. The corresponding depth elasticity ranges from nearly 0 to 0. Advertising and Promotion Elasticities—Simulation Results about 25% of the total elasticity.0526 0.558 0. as well as shortelasticity and long-term) of promotion ranges from 0.0206 Brand 2: Short-term 0.0222 Brand 3: Short-term 0.01*frequency)* a 0.0165.0018 0.0726 0.0400 0.040 0.0306 Long-term 0. our re.0038 Total 0.0150 Long-term 0.0034 0.071 0.0244 0.0182 0.042 Brand 4 0.a 1% increase in frequency (or depth) is plies tanta.JEDIDI.0292 0.0224 0. Third.0292 with an aver-of 0. Thus.switching behavior relative to their ers’ brand pur.718 Advertising Elasticity Brand 1 0.0062 0.0134 Total 0.079 Brand 2 0. The market leader in this category has a lower choice elasticity than quantity elasticity suggesting its price changes have less impact on consum.0186 0. First.0156 0.079 0.0278 is similar to the advertising elasticity for mature products of 0.127 Brand 3 0.0152 Total 0. Promotion Table 4 yields a number Elasticities.0060 0.0239.0314 Total 0.0184 0.0426 0.366 Brand 2 1.0030 0.0194 0.0032 0.0380 0.434 Brand 3 0.0214 0. 1999 14 .0078 0.000 0.0272 with an average of 0.0256 Brand 4: Short-term 0.08.67. While these elasticities may appear age small to regular price elasticities. MELA. Advertising Elasticity.0028 Brand 3: Short-term 0. The decrease in a brand’s expected price arising from a 1% increase in promotional frequency is given by (0.13 Brand Choice Purchase Quantity Total Price Elasticity Brand 1 0.0152 0.05 esti.0306 Lodish et al.0578 0.0222 0. Presumably.029 0.338 0. this is due to the brand building nature of the national advertising .0056 0.439 0. it is important relative to a 1% change in the frequency (or depth) of note promotions has a much smaller effect on the average price charged than a 1% change in the regular price.switching (because a large number of ers’ brand consumers buy this brand) but more impact on their quantity decisions.0272 Long-term 0. of findings about the effects of increasing the frequency of promotions.8%) im. slightly Marketing Science/Vol.0218 Long-term 0.123 0. This average The results in Table 4 show from 0.0146 0. 1.0332 0. the total and depth frequency(choice and quantity.0568 Long-term 0.0040 0.0206 to 0.0448 0.005 1. Using the average depth and frequency (21. AND GUPTA Managing Advertising and Promotion for Long-Run Profitability Table 4 Price.380 0.0048 0. mount an equivalent 1% discount in price (in the form of increased frequency) yields a total (total longterm short-term) elasticity of 0.0292 Brand 4: Short-term 0.0020 0.0216 0.0268 0.0056 Total 0.0586 Long-term 0. this (16.04 to 0.0272 Promotion Depth Elasticity Brand 1: Short-term 0. Our result that price changes have about three times the impact on consum.0024 0. our results extend this finding further by suggesting significant differences across brands.0212 0.0158 0.066 0.0304 0.0326 0. (1998).0242 0.004 0.0340 0.15 by Assmus mated et (1984).036% cut in the regular price. Consistent with prior literature we al.337 0.0356 Long-term 0.0018 0. Further.0132 0.0142 Total 0.0024 0.

79 Discount Frequency 0. the price and short-term Fourth. This significantly than higher short-term response to promotion compared to adver-may explain the increasing budget allocated tising to promotions in recent years. promotion elasticities in Table 5 are over 9–14 times higheradvertising elasticities..46 Marketing Science/Vol. by of virtue vividness. reduce suggest that price. and subtract advertising and promotion expenses. and current expenditure levels. that deeper discounts indicate lower quality. multiply the sales by the calculate increases in brands’ sales. This highlights a they key benefit of our modeling approach. ap. are considered. However. The profitability of these various strategies clearly depends upon costs. the finding is reversed (the total elasticity for quantity). market response. This is also consistent with the finding of Gupta (1988). and the long-term elasticities term of promotional frequency are negative and about 27% of positive short-term effects. when long-term effects are considered.but has the lowest overall total effect.of 0. The Long-Term Impact of Price Decreases. As indicated earlier. or promote more frequently. This is because i) the is greater deleterious long-term effects of promotions on brand equity the effects on choice and ii) the training of reduce consumers to stockpile when they observe an especially increases the effects on good deal quantity. we calculate base profits by multi.brands’ market level sales by gross margins plying and subtracting advertising and promotional then expenses. MELA. these results do not necessarily firms should advertise less. Using model and reasonable assumptions made in results consultation with the management of the sponsoring company. 6. going beyond directional results to assessing the relative effect sizes separating the short. while the long-term effects of promotion depth are consistently negative for all brands. on average. we arrived at rough estimates of the longterm profit impact of competing marketing activities. to assess the effect of a 5% price cut on brands’ profitability we use the estimated price elas. and Table results. these are mixed for promotion frequency.46. Fifth. depth has the greatest overall short-term ef. Next.6 outlines our procedure. promotions have a substantial impact In sum.being about two-fifths of the positive shortfects term Moreover.greater for increased depth of promotions pear than do for increased frequency. short-term promotions Table 5 Price and Promotion Elasticities Compared have a substantially larger impact on making consumers switch brands rather than making them buy more quantity. On effects average brands). in the long-run. their How-this effect is reversed when long-term effects ever. This table highlights that deep promotions. Put differently.67 Discount Depth 1. 1. In this table. The percent change from the base profit is then calculated.e. It is also possible that consumers begin to believe. 18. Mela and Urbany (1996) find evidence of such attributions in consumer protocols. assumptions. a 1% increase in the depth of promotions yields a total elas. on consumers’ purchases with the negative long-term ef. ticity Second. compared to price and promotional Third. No. their the negative long-term effects are nearly two-fifths the positive short-term effects. in the longalso run consumers come to expect discounts and show lower sensitivity to these promotions. AND GUPTA Managing Advertising and Promotion for Long-Run Profitability lower than the regular price elasticity. Advertising and Promotions on Brand Profits Our previous discussion suggests that advertising has a small effect on brand sales compared to price or promotions. the long-term elasticities of (across promotional depth are negative and about 58% of their short-positive effects. Table 4 shows that almost 90% of the positive short-term effects of promotions is accounted for by brand choice.JEDIDI.10 0. frequency. Similarly.and long-term and effects.the average price elasticity across brands pares withaverage “equivalized” (on a 1% price change the basis) and short-term frequency and depth total elasticities. In other words.79 0. We ticities then recalculate the lower gross margins. the negative long-term effects effects.91 0. We proceed similarly for advertising and promotions by using elasticities to calculate Short-term Total (Long-Term Short-Term) Regular Price 0.4. Table 5 fect. However. 1999 15 . i. generate a very high response. com.

007.333 6.601. 1999 16 .287.548 6.773 12.536 159.JEDIDI.957 17.032.874 174.546 177.977 231.000 7. MELA.725 158.161 160.199.31* Assume 20% markup (Dhar and Hoch 1996).669 5.02 0.939 14.05 level.045 0.050 Manufacturer Price/oz Manufacturer Profit/oz Ounces Sold 3 1 2 ($) 0.13 0.03 0.043 ($) 0.324 195.542.434 159.042.02 0.052 0. The Effect of Regular Price Changes on its.032 0.454 158.32* 0. 3% of regional advertising budget is allocated to the market area under consideration.457 161.344 Base Profit ($) 470. Promotion and Advertising on Profits Brand 1 Brand 2 Brand 3 Brand 4 Base Profits Retail Price/oz ($) 0.03 Ounces Sold with 5% Depth Increase 17. and subtracting the increased advertising and promotion expenses from revenues in or.541 19.08 0.31* 0.294.935 5. 2 3 4 5 changes in demand.054 0.887 128.Prof.381 % Change in Profit Profit Impact of Promotion Depth Changes Depth Elasticity 0. 18. Assume 30% variable costs. No.667 5.491 198.052 0.03 0. Marketing Science/Vol.682 Gross Margin ($) 560.386 % Change in Profit *Significant at 0.477.33* 0.083 194.795 5 Market Price Promotion Profit Impact of Price Changes ($) 42.159.333 Gross Margin ($) 559.69* 0.498 161.672 7.02 0.684.306.38* 0.400 689.369 % Change in Profit Profit Impact of Promotion Frequency Changes Frequency Elasticity 0.194 231.151.37 1.00 0.838 7. Market population is 3% of the region population.33* 0.07 Ounces Sold with 5% Advertising Increase 17.030 17. Therefore.338 128.148 Gross Margin ($) 541. AND GUPTA Managing Advertising and Promotion for Long-Run Profitability Table 6 Long-term Impact of Changes in Price.221 % Change in Profit Profit Impact of Advertising Changes Advertising Elasticity 0. Inferred from average frequency and depth and 20% markup.748 194.22 2.138 7.99 0.13 0.indicated by Table 6.280 162.966 158.294.43 obtain brands’ profits. Approximate der standard profits were calculated using the errors for delta method.694.956 6.436 5.166.678 34.186 6.950 231.045 0.272 20.646 128.342 178.88* 2.360 659. while brand 2 is better off.280 1.811 178.779 Price Elasticity 0.086 5.509 226.367 Profit with 5% Advertising Increase ($) 470.567 178.04 0.49* 0. Store data projected to all outlets.03 Ounces Sold with 5% Frequency Increase 17. 1.047 0.754.116 7.944 6.33* 0.138 Profit with 5% Frequency Increase ($) 468.63 1.62 0.919 Regional Advertising ($) 1. and 4 are at a disadvantage with an Brands 1.033 0.041. additional 5% price decrease.361 Profit with 5% Retail Price Cut ($) 452.524 Gross Margin ($) 560. 1 3.31* 0.01 0.160.840 Market Advertising 4 ($) 46. decreases in As regular price are not generally profitable in this category. 3.143 Profit with 5% Frequency Increase ($) 468.032 0.702.181 126.296 128.72 Ounces Sold with 5% Retail Price Cut 18.

1.036% cut in price. Marketing Science/Vol. brand 3 should reduce its advertising the most. in our product category. and iii) whether it is better to change regular prices or price promotion levels. entirely on manufacturers’ perspective. While comparing the small. that develops marketing budget recommenda.5. The Effect of Advertising Changes on 5%Profits. The retailers’ motiva. the underlying the approach herein enables managers to answer developed important marketing questions such as i) whether or not to increase advertising and decrease promotions. No. brands three and four should reduce their advertising spending levels. and profit impact of policy Previous research has typically stopped changes.upon short. For change example. Finally. Brand 1’s mean advertising level is nearly optimal as the advertising profit elasticity is near zero. our cute results are based on a single category in a single market. decreasing price would generally not be profitable (the exception is brand 2). For example. 18. However.31% to 0. 7. Our than results are consistent with this finding. Brand 2 also stands to benefit from increased advertising.Furthermore.increase sales and revenues. although price decreases can signifi. 1999 . In the process. then promotions may be even less profitable than they ap. advertising has a very small effect for mature categories. (1995a. Nonetheless. and changes in advertising expenditure have a much smaller effectchanges in advertising copy and quality. we examined the 17 small effects of marketing instruments on profits suggest that although the market may not be in perfect equilibrium. We attempted to make these assumptions as have reason. In sum. we find that. to our knowledge.suggest that firms should increase regular sults price and reduce the level of price promotions.recall that a 1% increase in frequency or tant to depth of promotion is equivalent to an average of Table 6. However. b) suggest that. they also cantly have a significant negative impact on margins leaving the overall profits lower. we believe our model is a simple and powerful approach that enables researchers to repli-and generalize the results across products cate and markets. However. Second. our analysis focuses almost port. increasing advertising would have mixed effects on profitability and increased promotions would have a deleterious impact on profits.Prof.possible with the help of the sponsoring able as com.JEDIDI. The Effect of Price Promotion Changes on its. 10 ii) whether to increase or decrease the frequency or depth of promotions. our of certain re. this paper seeks to provide a means to answer several questions regarding the long-term im. it may not be feasible to increase price without also offering price discounts to obtain retailer sup. the realities of the pear marketplace some constraints on the actual also impose execution marketing strategies. In contrast. making it difficult to tion may exe.g. Conclusions Substantively. margins.and long-term effects tions predicated as as competitive reactions. 6. it is not far from optimality.of promotions and advertising on brand pact choice and purchase quantity. the most recent in the data indicate brand 1’s spending quarters has fallen substantially below that mean indicating that brand 1 should consider increasing its advertising again. profit due to price or promotion changes. Limitations and Contributions of the Simulation We recognize that our simulations have quite different. due to lower pass-through by the retailer). in general. Like regular price cuts. It has nearly double the amount of brands with spent similarsuggesting that this prescription to cut sales advertising may well be reasonable. it is impact impor. in advertising has a very small impact increase on profits (relative to price) for most brands. of the intended strategies. Lodish et al. MELA. the well paper develops an integrated methodology consisting of three phases—model. profit estimates are based on First. AND GUPTA Managing Advertising and Promotion for Long-Run Profitability In other words. if the cost of promotions is higher than our assumption (e. the results pany’s A could for different sets of assumptions. after first phase. Indeed. It is the first paper. In particular. and retailer pass-through. albeit impact on profits. increasing the frequency or depth of promotions has a negative. simulation of consumer and competitive response. in practice.. our assumptions 1 0 The about costs.5% increase in promotions (frequency A depth) or significantly affects brand profits from 0. However. this is one of the main contributions of this manuscript.

To address these issues. as well as the effects on consumers’ brand choice versus purchase decisions. and a comparison of the tactics of decreasing regular price with the alternative of increasing discounts. MELA. 1. short-term price promotion elasticities (on an equiva. purchase incidence and the effects of marketing activ. and two anonymous reviewers for their suggestions and insights as well as IRI and an anonymous consumer packaged goods company for financial support and their comments on this project. Conversely. and promotions. remain several important areas for future There research. to assess the relative profit impact of longterm changes in price. in the longers’ brand term.profitability of advertising and promotions. (1988). advertising. Area Editor. on quantity average. we To conducted several simulations to assess the relative effects of various marketing activities and also to separate these effects into short. the relative across ef.versus long-term.price basis) are about 1.79. the effects of competitive reactivity. we performed additional simulations by making reasonable assumptions about costs and margins. we developed a new model and an estimation approach that allows for (a) varying parameters to capture changes in consumers’ response to short-term marketing activities due to changes in long-term marketing actions of brands. and (c) decisions correlation among brands to avoid the IIA assumption. The results also show that promotional frequency increases are generally less deleterious than promotional depth increases (although the result varies by brand). 1 1 The authors would like to thank the Editor. long. Marketing Science/Vol. the effect of promotions on quantity may be greater that than of choice.56) are about 30% than regular price elasticities. whether promotion’s short-term positive effects outweigh its possible long-term deleterious effects. These results suggest that. in the long-term.JEDIDI. Both price lower and 18 price promotion elasticities are larger than advertising elasticities (0.choice decision. price and promotions on fects profits. (b) correlation between errors in brand choice and purchase quantity to avoid selectivity bias. This leads to a heteroscedastic.of advertising. promotions make it more difficult to increase regularand increasingly greater discounts need to prices be offered to have the same effect on consumers’ choice. Area Editor. and two anonymous reviewers for their suggestions and insights as well as IRI and an anonymous consumer packaged goods company for financial support and their comments on this project. zation Although we model brand choice and purchase quantity.07). AND GUPTA Managing Advertising and Promotion for Long-Run Profitability impact of promotions and advertising on “brand equity”. Our results show that. Our hasadvertising and significant effect on “brand a positive equity” while promotions have a negative effect. how the long. could facilitate their implementation in This managerial settings.this aspect in the model as well.responses to advertising and promotion term differ choice and quantity decisions. we did find that in the long. results show that. 11 The authors would like to thank the Editor. 1999 . A formal dynamic these optimi-could further yield important insights. Furthermore.00 compared to lent 1% regular price elasticities of 0. then estimated our model using over eight We years of scanner panel data for 691 households for a consumer packaged good. varying-parameter probit and regression model which also controls for selectivity bias. No. 18. However. We would like to reiterate the need to make results generalizable.purchase timing could further affect the ity on long. Finally.promotions make consumers more price term. term It would be desirable to assess these effects in future Retailer behavior may also be affected by studies. advertising and short-term promotion was on consum. Consistent with Gupta we found that most of the effect of price. the relative efficacy of the frequency and depth of promotions. in the longterm. Finally. porate it would be desirable to develop approaches to simplify the application of the techniques outlined in this paper. Although we not find a significant effect of advertising on did consumers’ price activity and it will be helpful to term incor. sensitive and less discount sensitive in their brand choice decision. Results show increases in advertising and decreases in price have mixed effects on brand profitability would across the brands while further increasing promotions would have a uniformly negative impact on long-term profits.move beyond the directional results. the long-term effects of promotions are negative and are about two-fifths the positive short-term effects. total (longplus short-term) promotion elasticities (0.

18.JEDIDI.c f i jt ij tjj ) ) r r .inthe qc l c qqqq ij t jl lm ZX . (2). . (A-4) where i1 2t ij dt i 13t i1 t ccc l l ll ll cc c i2 ti 1t i 3 ti 1t i 4 ti 1t (A-9) d d 1 2t i d 1 3ti 1 4t • f (f q1ti . ] iis rq c W 1 U Q where ij t i j t i jt lcc f lqq f . . The log-likelihood for the data can therefore be written as: N 4 1 j 1 t T i f cccc ij t ij kt i jk t i j t k eX . Marketing Science/Vol. Then L L r r J M cccc iJ t J Without loss of generality assume that brand J2 JJ rr 0 0 1 for J p. . MO . . 0 j lm i jm t ij lt vi jd t . (A-1) 0 qqqc ij t jj j cqcc ij t j j j . (A-6) i 12 1 rq c rq c 1 ij t i j t i jt is the covariance between choice and quantity errors. both the choice and quantity models have heteroscedastic errors.. Second.qcf i jt ij t j 1 2 Conceptually j0 it is straightforward to extend the likelihood expres- 2 Note that the intercept error variances and are not estimable hd2 j0 c and are absorbed in and respectively. d2 ). . (A-8) j is brand 1. 1. ij Var( f c t)j k ij kt j j i jt h2 ( X c )2 ij k number of purchase occasions for household r cc w ..Rc ) N 0 rr 1 1 2 1 1 M . This is evident if we rewrite the reduced form of Equations (1). choice model. N (0 . T i is the E( f q t ) 0. Cov( f c q .Rq c ) j N rr . 1999 19 . 1t Several important features of the error structure should be noted. h2 ). (A-2) rr L ec i jk t j k N (0. and ccc i 1ti 2t c ccc c c c i1 t 13 12 2 3 c ccccc ccc c c i1 t 14 12 2 4 ww 2r 1 2 Wwrrr i22 || | ww i1 t i1 ti 3t 1 4 13 34 2r 13 | | ww i 1 ti 4t ccccc l c cj k ij t km 0 jk m ij ZX jk t mt i . J and . 3. wrrr | wrrr 2r 14 are the covariances of the utility difference variables. (A-7) 1. Var( f q t ) i jl t j j i jt d2 ( Xq )2 ij jl l r qq w . MELA. The likelihood expressions for brands 2. ) which is multivariate normal with null vector as the mean and the following covariance matrix: WW i1 1 WW i1 2 W1t i where Wi1 1 i1 2 i2 2 [w q1t the variance of the quantity error. and E( f c ) 0. Third. No. (3) and (4) as follows: . 63) 4 brands. . f qqqq ij t ij lt i jl t ij t l eX . dj . and 4 are derived in the same fashion. 12 the likelihood of this purchase is (Maddala 1983. We restricted the number of brands to four for ease of exposition and because our application involves four major brands. the conditional expectation of Q i1t Q i1 t Cov( f c c . ) 2ti i 1 c f cff q i 1 4t i jt id t i 1 3ti 1 4ti eq i jl t jl and f(•) is the joint density function of ( . the choice errors are normally distributed and are correlated across brands. AND GUPTA Managing Advertising and Promotion for Long-Run Profitability Appendix A Likelihood Function We assume the errors in Equations (1)–(4) to be normally distributed as follows: c ccc i1t U ij t id t i jt id t i d t or f cfclclc U i jt d and Q ij t i j t i jt l q q f N (0 . the errors in the choice and quantity models are correlated as suggested by Lee and Trost (1978) and Krishnamurthi and Raj (1988). (A-3) N (0. First. ij t LL i 1 ln L ij t (A-10) where N is the number of households in the sample and i. (A-5) . rr q jjjj sion for more than four brands. but are assumed to be zero across brands.. This gives us a multinomial probit model of brand choice. Note that the covariances between choice and quantity errors are nonzero within a brand. Appendix B: Conditional Moments of Conditional Expectation of given choice of brand 1 is Q i1 t From Equations (A-7) and (A-8). When consumer i chooses brand j.

) 1 i t |.2 ) ( . 4.). i t 14 )( c . u2 i 1 ti Var( f qcc | 1t i 1 2ti ll 2t i 1ti . (B-5) q2 4 1 4 2 4 2 4U 2 3 .4 a A 3 where 1 ll ccc i t ist ( t is t i s r s i lt 1 E .) Tallis’ (1961) results.). . 225). 2. (B-1) l c3t c c lc i i Var( W W i1 2 . 3 a A A 1 q q ls 24 3 4 1 4 1 4 2 4 2 4 3 2 3 .4 a a q A )U( )U( ) 2 . ) trivariate standard normal vector with covariance elels 1 23 1 3 1 3 2 3 2 3 4 24 .) using The elements of ( i t E by Tallis (1961. the conditional variance of of brand 1 may be written as 13 t i1t l E( f 1 ti | 1 2ti ll 2t i 1 ti . 1 . i 1 3t . q . A q . 3. f q i1t i1 2t .. These are given by ccc i ti |. . )[( A q24 2 q 2qq3 4 2 3 3 4 4. Note that ( c ( |. 1 1 where m i t 12 . 3. )[ 2 13 14 34 23 24.) standard bivariate normal distribution function. and i1 4t follow a multivariate normal disq 1 i t i 1 ti | l lcc 2 ti 1 ti 1 3t tribution.2 qq a ( )U ( A A a . ) q q ) 2 A 2 3 a 3 1 3 2 a 2 3 4 24 . 1 3t Conditional Variance of Q i1 t Q given choice Using Equations (A-7) and (A-8) and the properties of the multivar1t l lc c c c i3 t i qq cc i 1ti cccc i3 t i 1 ti . (1 1 22 a q sr. . 3 1 3 E 12 ww 1 E z z 2 r1 2 a z r1 2 a z a 2 ( )U ( 12 12 12 1 3 1 3 1 4 1 4 23 2 3 (1 2 1 2 1 3 2 3 24 23 a a ( A A A )U( 3 ww 1 2 {( 1 2 2 2 3 24 3 4 . MELA. 14ti ll 4 ti 1t ). A .. E |. Because .JEDIDI. Transforming z 1 i t 12 2 t r to a standard multinormal •{ 1 1 2 1 2 2 2 3 2 4 3 4 .) 1 ( |. 24 (1 24 3 4 2 13 1 4 34 3 ) ( 2 (. 13 i t ) is also mul( l c3t c c lc i i qc 2 1 i 1 .2 ( )U ( A . 1 ti 1 i2 2 1 4t i l l4 t i 1 2t i 1t ) . 1 22 c 1 |. a a . q . q ) q ( . ) (1 24 ( ) ( . AND GUPTA Managing Advertising and Promotion for Long-Run Profitability s i 1t i E( Q 1 ti | 1 2t i ll cc 2 ti 1 ti .) 1 i t 1 ( |. 1 3t ll 1 ti . . 2334 42 q qq )]}.) 12 1 22 (| 1 12 cc 2 1 13 ( 1 qc 2 1 c 1 l qc 1 qc c lc ctic c i 1 22 ti ti ll ti t 1 WW S i i t 22 1 i t S ( 1 |. p. . . We obtain the ele- S i t ( ). 1 ) i ti () E f 1 |. 14 l l4 W1 . the conditional density of ( | .). 20 . 1 2 2 12 )( 1 3 2 13 ) i ti t correlation coefficient between i t * z z q z 1 23 23 12 12 2 2 3 24 3 4 . E 2.l and i rt 1 . ( . . 1 choice probability of brand 1. 14ti ll 4 ti ) iate normal distribution. (B-2) ) .4.

Stud. 1984. John D. Hardie. Thomas A. Lee ang. Dick R. Joe A. Impact of deals and deal retraction on brand switching. Dekimpe. 21. P. Marketing Res. Lehmann. Generalizing what is known about temporal aggregation and advertising Carryover. Marketing Res. Measuring preference from scanner panel J. How pro——. Marketdata: Filtering out the effects of price and promotion. John W. what. A summary of fifty-five in-market experimental estimates of the long-term effect of TV advertising. Mary Ellen Stevens. Stuart Kalmenson. 14 (Part 2 of 2) G109–G121. Hanssens. ometrics 8 357–382. S. The effect of advertising on competition.. Robert C.V. S. International J. Determining segmentation in sales response across consumer purchase behaviors. J. Wagner. Clarke. 1–20.JEDIDI. 1988. Wittink. 14 (3) (Fall) G151–G160.. Bucklin. Res. McGraw Hill. 1s 1r When. 1998. 2 (3) (Summer) J. Econometric Methods. Lattin. Johnston. Sunil.). The persistence Mela. New York. Marketing Sci. Fader. 1992. UK. 1993. How T. Diagnosing competitive reactions using (aggregated) scanner data. Gary Russell. Institute. 14 (3) (Summer) G122–G132. Little. ——. Farley. David. in Marketing 9 39–57. Beth Lubetkin. Bruce G. Kaul. purchase incidence. and segmentation: An integrated approach. Mary Ellen Stevens. Sunil Gupta. 1997.V. Econometrica 47 (January) 153–161. and and b z sr .. 1992. ——. Gurumurthy. Marketing Res. Marketing Sci. C. Marketing Res. Working paper. Some approaches to the correction of selectivity bias. 1995. J. 1995. 18 453–476. Anil. NJ. Donald R. Limited-Dependent and Qualitative Variables in Cambridge University Press. Empirical generalizations about the impact of advertising on price sensitivity and price. Lodish. 1 (December) 81–97. 1995a. Sunil Gupta. C. advertising experiments. ——. ——. Fox. J. A logit model of brand choice calibrated on scanner data. Report No. 14 (3) (Fall) G133–G140. Jeanne Livelsberger. Marketing Sci. 1995b. Marnik. of a Brand Name. Washington. There are similar expressions for the remaining i 1j t i 1 k t elements Heckman. Rev. Peter S. London. Marketing Res. 1978. 98-103. Marketing Sci. Boulding. 1990. Marketing Sci. Marketing Sci. 1992.l a rsl. Managing Brand Equity: Capitalizing on the Value The Free Press. 1995a. Jeanne Livelsberger. 1991. Englewood Cliffs. 1995b.. Blattberg. z on z J. 1. Marketing Science/Vol. 1998. Guadagni. Little. DC. Eunkyu Lee. 1983. Brand choice. G. 14 (3) International J. EconEfficient Consumer Response: EnFood Marketing hancing Consumer Value in the Grocery Industry. The long-term impact of promotions and advertising on consumer brand choice. 1982. CA. and How Much to Buy’. keting Sci. 15 (February) 72–81. 1988. Peter S. Kurt Salmon Associates. ——. ——. ——. J. k 2. 3. r . Sales promotion: The long and short of it. ber) 522–524. . Commercial adoption of advances in the analysis of scanner data. Stanford. ——. 1999 21 . Lehmann. Beth Lubetken. Bruce Richardson. A model of brand choice and purchase quantity sensitivities. Donald R. Dodson. Raj. in Marketing 10 (1) 9–22. 1976. 18. Lee. Maddala. James M. and how much to buy. 1983. 13 (4) (November) ——.. Inc. J.. Comanor. Marketing Res. J. Methods.r are the partial regression coefficients of |. Marketing Res. Mar- Assmus. AND GUPTA Managing Advertising and Promotion for Long-Run Profitability Ar a1 ssr (1 b . J. 1996. Darral G. Edward J. j. Lakshman. Gert. Trade promotion spending and merchandising: Promotion paradox. Makato Abe. Competitive reaction versus consumer response: Do managers overreact? 119. Econometric measurement of the duration of advertising effects on sales. McFadden. S. 29 (2) (May) 201–215.. Impact of sales promotions on when. advertising works: A meta-analysis of 389 real world split cable T. sales force. Marketing Res. 49 355–372. Econometrica 52 (March) 345–362. Trost. William. Marketing Res. Empirical generalizations about market evolution and stationarity. Econometric Res. in Marketing 13 103– International J. MA. Richard Briesch. MSI Working Paper Series. MELA. 1l respectively. Siddarth. No. 7 (1) (Winter) Krishnamurthi. and Strategies. 1984. 35 (2) (May) 189–197. Kalyanaram. 4. . Winer. 1984. 14 (3) Reference s Aaker. Magid Abraham.. Marketing Sci. 1990. Scott A. of marketing effects on sales. Alice M. ——. Marketing Letters ——. John D. S. 345–357. Bruce Richardson. (Summer) G141–G150. 1994. Leone. Jeffrey A. Wilson. 1998. How advertising affects sales: A meta-analysis of econometric results. Robert P. Empirical generalizations from reference price research. Cambridge. Kamakura. Daniel L. Leonard M. Res. 1995. Econometrics. H. ——. CT. 1989. Re ections on ‘Impact of Sales Promotions on What. William S. Brian Sternthal.l and E( b sl . 14 (1) (Winter) 1– J. 1993. 32 (2) (May) 125–139. Prentice-Hall.. ing Res.r 1 b q a 1l ) ls q2 2)(1 sr sl. Cannondale Associates. 1995. 1979. Sample selection bias as a specification error. motions work. Dick R. 34 (2) (May) 248–261. An econometric analysis of residential electric appliance holdings and consumption. 1978. 30 (4) (Novem——. Richard Staelin. 31 (2) (May) 159–172. Mastering the mix: Do advertising. ——. Sales Promotion: Concept. Peter M. Wittink. Gupta. G151–G160. Stanford University. New York. Russell S. Randolph E. Lung-Fei. 1993. Robert P. James J. Estimation of some limited dependent variable models with application to housing demand. Working Paper. Measuring brand value with scanner data.. Econom. Tybout. 25 (4) (November) 342–356. J. Neslin. myth orreality? Wilton. and promotion activities lead to differentiation? J. Carl F. Marketing Sci. Dubin. Market response with purchase feedback. 203–208. Dominique M. London Business School. 21 (1) (February) 65–74. Res. 1974.

MELA. 22 Marketing Science/Vol. UT. Advertising exposure. Internat. J. Joel Urbany. Provo. Merrie Brucks. The long-term impact of promotions on consumer stockpiling. Purushottam Lakshman Krishnamurthi. in Marketing J. 1988. Chris Dubelaar. Promotion over time: Exploring expectations and explanations. Ratchford. Raj Gerard J. 1997. Swait. Joffre. Debbie MacInnis (Eds. (May) 160–174. Soc. An analysis of tradeoff between advertising and price discount. 1991. Measuring the dynamic effects of promotions on brand choice. Papatla. 1. Tulin Erdem. M. (Series B) 23 223– Tallis. 33 (February) 20–35. 35 (2) Sethuraman. ——. Royal Statist. search. Tellis. 1999 . Douglas Bowman. 10 (1) 23–45. Marketing Association for Consumer ReJ. The moment generating function of the truncated multi-normal distribution.JEDIDI. 1998. 25 (May) 134–144. Gerard J. J. Res. and has been with the authors 14 months for 2 revisions. 1993. Res. 18. G. Advances in Consumer Research. Kamel Jedidi. AND GUPTA Managing Advertising and Promotion for Long-Run Profitability ——. Marketing Res. Jordan Louviere. 28 (2) This paper was received November 25. The equalization price: A measure of consumer-perceived brand equity. J. Marketing Res. J. 229. (May) 250–262.). loyalty and brand purchase. 1998. processed by Brian T. No. Tellis. Marketing Res. 1996. 529–535. 1961.