Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Intern. J. of Research in Marketing 17 Ž2000. 281–305 www.elsevier.comrlocaterijresmar Market share response and competitive interaction: The impact of temporary, evolving and structural changes in prices Shuba Srinivasan a,) , Peter T.L. Popkowski Leszczyc b, Frank M. Bass c a The A. Gary Anderson School of Management, UniÕersity of California, RiÕerside, CA 92521, USA b Marketing Department, UniÕersity of Alberta, Edmonton, AB, Canada T6G 2R6 c UniÕersity of Texas at Dallas, Richardson, TX 75083-0688, USA Received 1 February 2000; accepted 1 September 2000 Abstract Managing pricing is a challenging task due to the significant impact on shares and the likelihood of strong consumer and competitor reaction. The major contributions of this paper are to assess comprehensive share response to temporary, evolving and structural changes in prices and to determine the level of market share as a function of levels of prices. For the empirical analysis, we examine two consumer product categories and find that it is valuable to distinguish among temporary, evolving and structural changes in prices, as their impact on market shares tends to differ. Further, we find that subsequent competitive reaction will influence predictions of price response. Accordingly, it is important for managers to use conjectures regarding competitive price reactions in assessing the impact of policy changes. We conclude with the strategic implications of the findings and discuss a number of opportunities for future research. q 2000 Elsevier Science B.V. All rights reserved. Keywords: Cointegration; Structural breaks; Competition; Marketing mix; Vector-Error Correction Models 1. Introduction Pricing decisions are important to managers due to the significant and immediate impact of price changes on shares and profits and the potential for strong reactions from consumers and competitors. Pricing ranked third in overall importance among 15 marketing issues in a survey by Davidson and Stacey ) Corresponding author. Tel.: q1-909-787-6447; fax: q1-909787-3970. E-mail address: [email protected] ŽS. Srinivasan.. Ž1997. and was cited as Aextremely importantB by 78% of the respondents. Companies continually change their price strategies and tactics with a view to increasing sales or profitability. Some price changes are temporary movements around a fairly stable level consistent with I Ž0. behavior. For example, a brand that offers a 2-week price discount off a fairly stable level engages in such a temporary effort. Other price changes are permanent if there is no return to the previous level. If a company engages in a regular practice of discount policies, this would lead to evolving prices or I Ž1. behavior, an example of a permanent change. Another type of permanent 0167-8116r00r$ - see front matter q 2000 Elsevier Science B.V. All rights reserved. PII: S 0 1 6 7 - 8 1 1 6 Ž 0 0 . 0 0 0 2 3 - 9 282 S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 change is a structural change in price in favor of a new level different from the previous level. If a company announces a one-time 20% price cut, leading to a new price level, this is an example of a structural change. As an illustration, Fig. 1 shows the market share and price series for the Miller brand of beer over 365 weeks while Fig. 2 shows the market share and price series for the Blue Bonnet brand of margarine over 104 weeks. The price series for Miller indicates there is evolution in price or a permanent price change, accompanied by an evolution in market share. Fig. 2 shows that Blue Bonnet brand has a structural change in price along with a structural change in market share in week 62. Finally, it is evident from the graphs that every 4 to 5 weeks both brands have temporary price promotions. There are other real-world illustrations for the three types of pricing behavior. See the following, for example. Ž1. Companies adopt short-term promotional tactics ranging from 25 cents off a six-pack of soda to thousands of dollars off an automobile in order to gain sales in terms of brand switching, repeat purchase, stockpiling and consumption. These tactics correspond to temporary price changes. Ž2. A company offers a price discount that generates immediate positive response resulting in subsequent regular practice of offering discounts. The airline pricing tactics in 1992 resulting in price wars is an example of permanent, evolving price behavior. While this type of price behavior may result in short-run gains, it would result in sustained losses due to competitive action and reaction ŽDekimpe and Hanssens, 1999.. On the other hand, sustained increases in prices can result in substantial long-run profits. Ž3. A company reduces prices to a new level by offering a significant one-time price cut. For exam- Fig. 1. Ža. Market share series Žin logs. for Miller. Žb. Market share series Žin logs. for Budweiser. Žc. Price series Žin logs. for Miller. Žd. Price series Žin logs. for Budweiser. S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 283 Fig. 2. Ža. Market share series Žin logs. for Blue Bonnet. Žb. Market share series Žin logs. for Parkay. Žc. Price series Žin logs. for Blue Bonnet. Žd. Price series Žin logs. for Parkay. ple, in the dry cereal category, Kellogg’s announced a 20% across-the-board price cut as a result of declining shares. These three scenarios raise important questions from a managerial standpoint. First, managers are interested in assessing the comprehensive share response to temporary, evolving and structural changes in prices. Managers need to know the share elasticities in response to these types of price changes so that they can make strategic decisions about levels of prices Žlist prices. vs. tactical decisions on the extent of price discounts. Second, managers are also concerned with whether or not, and to what extent, the reaction from competitors will influence predictions of price response. Omission of competitive reaction may lead to biased estimates of market response to price and other marketing variables Žsee, e.g., Leeflang and Wittink, 1992, 1996.. Indeed, Bucklin and Gupta Ž1999. argue that these two managerial issues are unresolved and point to an immediate need for academic research in this area. Our approach is based on two modern empirical frameworks, Vector Auto-Regressive Models ŽVAR, hereafter. and Vector Error-Correction Models ŽVEC, hereafter.. Our model follows the VAR model proposed by Dekimpe and Hanssens Ž1995b, 1999. and estimates market share, price and competitive prices as a simultaneous system of equations. We derive the static equilibrium conditions for the VAR model. Structural changes in prices and shares are determined using the structural break tests. Once a structural break is identified, the stability of the model is crucial to the task of evaluating the impact of structural changes within the system. Parameter stability is tested using diagnostic tests such as the plot of recursive residuals and the CUSUM test. If the parameters of the data generating process are invariant to structural shifts, one can then assess the impact of 284 S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 structural changes in prices on market shares using the static equilibrium conditions of the VAR model. To assess the impact of temporary changes in price and evolving prices on market shares, we use the time-series technique of cointegration analysis and impulse–response analysis. These techniques provide us with the long-run price elasticities—that is, the response to evolution in prices—and the short-run price elasticities—the response to temporary price reductions. We address the following research questions in this paper: Ž1. What is the impact of temporary changes in prices, evolving prices and structural changes in prices on market shares? Ž2. Do competitors and consumers respond to permanent changes Žboth evolving and structural. in prices in the same way that they respond to temporary changes in these variables? Our analysis offers new contributions to the literature in this area. First, we introduce the typology of temporary, evolving and structural changes in prices and discuss the managerial implications of these types of changes. Second, we estimate the comprehensive share Žcomprehensive in the sense of a multivariate system. and competitive response to these types of price variations. Finally, we formally gauge the influence of unilateral price-setting behavior in a dynamic-system context. The paper is organized as follows: in Section 2, we highlight how our effort differs from and builds on extant literature in the area. Section 3 discusses our modeling approach, Section 4 describes the scanner data we use and Section 5 provides the empirical results. Section 6 presents conclusions and offers directions for future research. 2. Background and overview 2.1. Long-run effectiÕeness of marketing-mix strategies A number of recent articles are closely related to our work. Table 1 summarizes the key findings of these studies; we emphasize the main points in the following discussion. Research has found that for frequently purchased consumer goods, market shares are predominantly stationary ŽBass and Pilon, 1980; Ehrenberg, 1994; Dekimpe and Hanssens, 1995a; Lal and Padmanabhan, 1995.. While Dekimpe and Hanssens Ž1995a. observed that a majority of sales series is evolving, Srinivasan and Bass Ž2000. demonstrate, theoretically and empirically, that this finding is consistent with market shares being stationary, provided brand sales and category sales are cointegrated. These studies’ findings are consistent with the view that marketing-mix variables have only a temporary effect on shares. Some studies, however, have documented long-run effects of marketing activity on shares. Focusing on markets with evolving shares, Bronnenberg et al. Ž2000. observed that distribution has a positive impact on market share through the early growth stage, but this effect diminishes over time. These results indicate that distribution has a long-run impact in the early stages of the product life cycle but not in the later stages. Franses et al. Ž1999. are primarily concerned with aberrant observations that hamper the quality of data and find different results across robust and non-robust methods for the long-run marketing effects. They find that distribution, advertising and promotions have long-run impact on sales. Franses et al. Ž2000. propose unit root tests that take into account the logical consistency constraint on market shares. Dekimpe and Hanssens Ž1995b, 1999. have shown that when marketing-mix variables influence an evolving series, it is possible for temporary changes in a marketing-mix variable to have permanent effects. On the other hand, in stationary markets, Dekimpe et al. Ž1999. have shown that when both market shares and marketing-mix variables are stationary, permanent offsetting competitive effects due to temporary changes in marketing effort are unlikely. Studying the phenomenon of hysteresis, Hanssens and Ouyang Ž2000., have developed optimal spending rules in terms of both trigger and maintenance marketing. The over-time impact of price promotions on category sales was analyzed for over 500 categories in Nijs et al. Ž2000., who Ža. found limited evidence of persistent effects, and Žb. tested the moderating impact of marketing intensity, competitive conduct and competitive structure on the short- and long-run primary-demand elasticity of price promotions. Pauwels et al. Ž2000. investigated the long-term impact of price promotions on three components of brand sales, i.e., category incidence, Table 1 Studies looking at the long-run salesrshare effects of marketing mix variables Model Findings 54 yearly observations Weekly grocery data 1991–1996 Ž257 weeks. of ready-to-drink tea VECM VAR Dekimpe and Hanssens Ž1995a. Data base of 400 prior analyses Unit root tests Dekimpe and Hanssens Ž1995b. Dekimpe and Hanssens Ž1999. Monthly advertising and sales data. Ž1. Monthly pharmaceutical data for 5 years. Ž2. BRANDAID data VAR Persistence analysis VECM Dekimpe et al. Ž1999. 113 Weeks scanner data four categories VAR Impulse response analysis Franses et al. Ž1999. Weekly scanner data for grocery product Cointegration analysis Franses et al. Ž2000. Weekly scanner data for 2 years for three categories Unit root tests Hanssens and Ouyang Ž2000. Printer sales for 4 years Theoretical and econometric analysis Jedidi et al. Ž1999. Scanner panel data, 691 households from 1984 to 1992 Individual-level choice model and quantity model. Lal and Padmanabhan Ž1995. Yearly data from IRI Factbook Regression analysis Mela et al. Ž1998. Scanner panel data, 691 households from 1984 to 1992 Dynamic market structure and cluster-wise logit model Mela et al. Ž1997. Scanner panel data, 691 households from 1984 to 1992 Two-stage approach Nijs et al. Ž2000. Data for 4 years for 560 categories VAR Persistence analysis Pauwels et al. Ž2000. Scanner data for 2 years for two categories VAR Persistence analysis Srinivasan and Bass Ž2000. Weekly scanner data and store-movement data for grocery products Cointegration analysisrVECM Advertising and sales are cointegrated. Distribution influences long-run share, through early growth and diminishes during later stages of product life cycle. Sales series are mostly evolving while shares are stationary. Advertising has a long-term effect on sales. Ž1. Sales calls, advertising and price differential have long-run impact on sales. Ž2. Advertising, promotion and prices have no long-run impact on sales. Promotions only have temporary effect in stationary market. In non-stationary markets, small lasting effect, positive for private label, and negative for two national brands. Distribution, advertising and promotions have long-run impact on sales. Propose unit-root and cointegration tests that take the logical-consistency properties of market-share series into account. Quantifies effect of hysteresis on sales and derives optimal spending rules. Advertising has long-run positive effect on sales; promotions have a negative long-run effect. Relative promotional expenditures have no impact on long-run shares. Increases in price promotions and reductions in advertising have led to decreased brand differentiation. Increases in price promotions and reductions in advertising have led to increased consumer price and promotion sensitivity. Limited persistent effect of promotions on category sales. Quantifies the long-term effect on category incidence, brand choice and purchase quantity. A majority of sales series being evolving is consistent with market shares being stationary, provided brand sales and category sales are cointegrated. 285 Data Baghestani Ž1991. Bronnenberg et al. Ž2000. S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 Study 286 S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 brand choice, and purchase quantity, and found the relative importance of these components to differ from the short-term categorization. Mela et al. Ž1998., studying dynamic market structure, showed that increases in price promotions and reductions in advertising led to decreases in brand differentiation. Jedidi et al. Ž1999. found that promotions have a negative effect while advertising has a positive effect on brand equity in the long term. Mela et al. Ž1997. observed that increases in price promotions and reductions in advertising have led to increases in consumer price and promotion sensitivity. In summary, while some of the above-mentioned studies report the absence of long-run effects on shares, others find positive long-run effects on shares. Thus, no conclusive findings are currently available on the long-run effects of marketing-mix variables. Given the potential for strong reactions to pricing decisions from competitors, we explore several important studies on competitive reaction. 2.2. CompetitiÕe reactions to marketing-mix strategies Empirical work on the nature of competitive reactions has been broadly based on two approaches: a non-cooperative game theory approach and a timeseries causal approach. Non-cooperative game theory based approaches have taken a variety of forms. For example, the AmenuB based approach requires specifying a priori the alternative forms of competitive interaction and using non-nested hypothesis tests to ascertain which type of interaction fits the data Že.g., Gasmi et al., 1992; Kadiyali et al., 1996.. Alternatively, conjectural variation approaches estimate competitive interaction directly without the need to specify the form of the interaction a priori Žsee e.g., Bresnahan, 1989; Putsis and Dhar, 1998, 1999.. One potentially valuable use of a game theory based approach is it provides a broad-based understanding of the nature of competitive interaction across categories, marketing instruments and strategic groups. The interested reader is referred to papers by Bresnahan, 1989, and Cotterill et al., 1999. Time series approaches employ time series data to infer the reactions of firms across marketing instru- ments. The most comprehensive studies on competitive reactions are those by Leeflang and Wittink Ž1992, 1996.. Following earlier studies by Lambin et al. Ž1975., Leeflang and Reuyl Ž1985., and Leeflang and Wittink Ž1992. ŽLW, hereafter. have studied competitive reaction functions of changes in price and promotion variables in response to changes in these variables by competitors. In general, when reactions occur within four periods, they are thought to be retailer-dominated, while longer-duration responses are thought to be manufacturer-dominated. In a subsequent study, LW Ž1996. examined competitive interactions among marketing variables and their impact on market shares and derived equilibrium conditions. The empirical reaction function was used to assess the nature and extent of competitive reaction in the marketplace. LW Ž1996. investigated 664 cases over different brands and different marketing variables, reporting that competitors generally overreact to price changes. Brodie et al. Ž1996. have replicated the study by LW Ž1996., examining three product categories with data from New Zealand. Jedidi et al. Ž1999. used the same response functions as LW Ž1996. to estimate competitive reactions. Our study builds on these important studies assessing the long-run effectiveness of and competitive reactions to marketing-mix strategies. A critical point of departure between our study and these important studies is that our approach Ži. makes a distinction among share responses to temporary changes in prices, evolving prices and structural changes in price and Žii. assesses how competitors and consumers respond to these three types of price variations. 3. Methodology Our analysis proceeds sequentially. First, we test for non-stationarity of the different time series. We use the Perron and Vogelsang Ž1992. procedure to control for structural breaks in the series. If the series are stationary, we estimate a VAR model in levels. We test for parameter constancy of the VAR model with shift dummy variables using several diagnostic tests ŽMaddala and Kim, 1998.. If the parameters are constant, we can then assess the impact of structural S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 changes in prices on market shares using the static equilibrium conditions of the VAR model. If the series are evolving, we assess whether there is a long-run cointegrating relationship among market share and prices. On the one hand, if there is a cointegrating relationship, we estimate a VEC model. On the other hand, if there are unit-roots but there is no cointegrating relationship, we estimate a VAR in differences since regressions on the levels of evolving variables may produce spurious results ŽGranger and Newbold, 1974.. Finally, to assess the impact of temporary changes in prices and evolving prices on share and on each other, we use impulse–response analysis. We describe these steps below. 3.1. Testing uniÕariate equilibrium share models When a series may be appropriately modeled as depending on a constant plus a coefficient times a lag of the series plus a random term, testing whether a series is stationary or evolving may be accomplished by means of the well-known Dickey–Fuller ŽDF. test for the presence of a unit root. When more than one lag is involved, the appropriate test is the Augmented Dickey–Fuller test ŽADF. ŽBanerjee et al., 1993.. We use these tests in examining the stationarity of market shares and marketing-mix variables. We apply the testing procedure in Perron and Vogelsang Ž1992. to test for structural breaks in the evolving series. Dekimpe et al. Ž1997, 1999. have used structural break unit root tests Že.g., Perron and Vogelsang, 1992. to control for effect of new-product introductions. 3.2. The multiÕariate Õector autoregression model Our model follows the Vector Auto-Regressive Model proposed by Dekimpe and Hanssens Ž1995b, 1999.. A five variable VAR is constructed, consisting of a brand’s market share, its price and the prices of the major competitive brands. As exogenous variables, we included own and competitive features, displays and price specials. Since these variables are zero-one type, we treat them as exogenous variables. We apply logarithmic transformations to market share and prices, providing a constant elasticity model ŽFranses et al., 1999.. As an example for the mar- 287 garine market, the following VAR model was estimated for the Blue Bonnet brand: MS BB ,t PBB ,t PPK ,t P FL ,t P PL ,t i i i i i a11 a12 a13 a14 a15 a10 i i i i i a 21 a 22 a23 a24 a 25 a 20 N i i i i i a 32 a 33 a34 a 35 s a 30 q Ý a 31 is1 i i i i i a 40 a 41 a 42 a 43 a44 a 45 a50 i i i i i a51 a52 a53 a54 a55 D D D D a11 a12 a13 a14 D D D D a 21 a22 a23 a24 D D D D a32 a33 a34 q a 31 D D D D a 41 a42 a43 a44 D D D D a51 a52 a53 a54 F F F F a11 a12 a13 a14 F F F F a 21 a22 a23 a24 F F F F a32 a33 a34 q a 31 F F F F a 41 a42 a43 a44 F F F F a51 a52 a53 a54 S S S S a11 a12 a13 a14 S S S S a 21 a22 a23 a24 S S S S a32 a33 a34 q a 31 S S S S a 41 a 42 a 43 a 44 S S S S a51 a52 a53 a54 MS BB ,tyi PBB ,tyi P PK ,tyi PFL ,tyi PPL ,tyi D BB ,t D PK ,t D FL ,t D PL ,t FBB ,t FPK ,t FFL ,t FPL ,t e BBS ,t S BB ,t e BBP ,t S PK ,t q e PKP ,t , S FL ,t e FLP ,t S PL ,t e PLP ,t Ž 1. with endogenous variables being market share and price of Blue Bonnet ŽBB. and prices of Parkay ŽPK., Fleischman ŽFL. and Private Label ŽPL. brands; FBB , D BB and SBB indicate the feature, display and price specials variables of Blue Bonnet Žsimilar defi- 288 S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 nitions apply for the other feature, display and price specials variables., e jt Ž j s BBS, BBP, PKP, FLP and PLP. denotes the error terms, and N is the order of the model determined using the log-likelihood criterion. To avoid potential degrees-of-freedom problems when estimating extended VAR models Že.g., when the market shares of several competitors and marketing-mix variables are included in a simultaneous system., we estimated separate models for each brand. We can write the VAR in matrix notation as: yt s A 0 q A1 yty1 q A 2 yty2 q PPP qA N ytyN q A F x1 t q A D x 2 t q AS x 3 t q et , Ž 2. where yt is a 5 by 1 vector containing market share and prices at time t; x 1 t , x 2 t and x 3 t are 4 by 1 vectors containing the feature, display and special variables; and A F , A D and A S are the 5 by 4 matrices of the respective coefficients. Some features of this model should be noted. First, it is a comprehensive share and price response model incorporating feedback effects ŽBass, 1969., purchase reinforcement effects ŽGivon and Horsky, 1990., competitive reactions ŽLW, 1992., lagged response effects ŽBass and Clarke, 1972. and decision rules. Since lagged market shares may influence prices, we treat price and competitive prices as endogenous. The rationale behind this is that managers may adjust their prices based on market performance in past periods ŽVillas-Boas and Winer, 1999; Winer, 1986..1 Second, we use market share rather than sales; the use of market share as a dependent variable enables us to make relative comparisons across brands with regard to the nature of competitive activity ŽLW, 1996.. Finally, when the variables are evolving, omitting the long-run relationship will underestimate the effects of marketing mix variables on market share ŽVanden Abeele, 1994.. Thus, when variables are evolving and cointegrated, it is appropriate to estimate a VECM model rather than a VAR model. Incorporating the long-run relationship among variables allows us to estimate long-run elasticities as well as short-run elasticities ŽEnders, 1995; Johnston and Dinardo, 1996.. 3.3. Impact of a structural change in prices on market share using structural break analysis To gauge the influence of unilateral price-setting behavior in a dynamic-system context, we first establish the conditions for the existence of static equilibrium for the one-lag VAR and then extend it to the more general case. Recall from Eq. Ž2. that the reduced form of the first order VAR with no exogenous variables is: yt s A 0 q A1 yty1 q e t . Eq. Ž3. becomes: yt s A 0 q A1 Ž A 0 q A1 yty2 q e ty1 . q e t s Ž I q A1 . A 0 q A21 yty2 q A1 e ty1 q e t . With successive lagging and using the identity Ž I y A1 .Ž I q A1 q A21 q PPP qA1ny1 . s Ž I y A1n ., the preceding equation can be written as: yt s Ž I y A 1 . While we account for this in the model, our empirical results in Table 3 indicate that these effects are not significant for the two product categories we studied. y1 Ž I y A1n . A 0 q A1n ytyn q e t q A1 e ty1 q A21 e ty2 q PPP qA1ny 1 e tyŽ ny1. . Thus, yt will go to a constant matrix plus a matrix consisting of a weighted sum of past random shocks provided that A1n goes to zero as n becomes large. Further, we require Ž I y A1 . to be nonsingular. The preceding equation may be written as: yt s Ž I y A 1 . 1 Ž 3. y1 A 0 q Ž I y A1 L . y1 et , Ž 4. assuming A1n goes to zero when n is large. This will be true if each of the eigenvalues Žcharacteristic roots. of A1 has an absolute value less than 1. This S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 is the condition for the stability of the VAR model.2 The static equilibrium is: yeq s Ž I y A1 . y1 A 0 s Py1 1 A0 . Ž 5. Using a similar approach, we can derive the static equilibrium for the VAR model of order N with exogenous variables x 1 t , x 2 t and x 3 t . Starting with Eq. Ž2., we take the expected value of both sides of the equation and set y’s equal to the equilibrium value: yeq s A 0 q A1 yeq q A 2 yeq q PPP qA N yeq q A F x 1 t q A D x 2 t q A S x 3t . Ž 6. We solve for yeq to find: yeq s Ž I y Ý A i . y1 289 Once a structural break is identified, the stability of an estimated model is crucial to the task of evaluating the impact of structural changes within the system. If a model is not invariant to structural interventions, no meaningful discussion AWhat would happen if . . . B could be undertaken since the Lucas Ž1976. critique would apply with full force and simulation analyses would become problematic. A number of tests are available to examine whether the parameters of the model are stable, such as the Chow test, and tests based on recursive estimates such as the plot of recursive residuals, the CUSUM test, the CUSUM-of-squares test and the recursive coefficient estimates ŽCharemza and Deadman, 1997; Wolter et al., 1998.. If the parameters of the data-generating process are invariant to structural shifts, one can determine the impact of structural changes in prices on market shares. = Ž A 0 q A F x 1eq q A D x 2 eq q A S x 3eq . s Py1 Ž A 0 q A F x 1eq q A D x 2 eq q A S x 3eq . . Ž 7. The eigenvalues of P are the complements of the eigenvalues of Ý A; therefore, the static equilibrium will exist when each of these eigenvalues has an absolute value less than 1. We use Eq. Ž7. to derive the market share response to unilateral price setting behavior in a dynamic-system context. It is important to note that the share level depends on a simultaneous equation system in which the level of market share depends upon the levels of prices and vice versa. Prices are determined by managerial policies that reflect a general level Žthe intercept. and by responsiveness to the other variables. If the managerial policy is changed, the level will also change, and this change will influence the level of market share. Therefore, we can use the static equilibrium relationships to examine the effects on market share of structural changes in prices. 2 Each sequence in the stationary VAR has a finite and time-invariant mean and variance after controlling for the structural shift, and the stability condition holds Žsee Enders, 1995.. We confirm that this is the case by a check of eigenvalues of the matrix, A t , in the empirical analysis. 3.4. Impact of eÕolÕing prices and temporary changes in prices on market shares If the market share series and price series’ are evolving, we investigate if there is a long-run equilibrium or cointegrating relationship among the variables ŽBanerjee et al., 1993.. Since own and competitive features, displays and price specials are zero–one type variables, we treat them as exogenous and abstain from a long-run analysis for these variables. As an example, if there is a long-run cointegrating relationship in the beer market among the market share of Miller ŽML. and prices of Miller ŽML., Budweiser ŽBD., Busch ŽBH., Old Milwaukee ŽOM. and Milwaukee’s Best ŽMB., then they are related by an equilibrium relationship such as: MS ML ,t s b 0 q b 1 P ML ,t q b 2 PBD ,t q b 3 PBH ,t q b4 POM ,t q b5 PMB ,t q e MLS ,t . Ž 8. The existence of a long-term relationship implies that equilibrium error e ML S,t represents a stationary process. One rationale for this kind of long-run cointegrating relationship is that different price levels correspond to different long-run demand or sales levels, and these, in turn, are associated with different levels of shares. S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 290 Following Johansen Ž1995., we can rewrite Eq. Ž2., where yt is a k-vector of non-stationary I Ž1. variables as: Ny1 D yt s A 0 q P yty1 q Ý Gi Dytyi is1 q A F x1 t q A D x 2 t q AS x 3 t q et Ž 9. N where P s Ý is1 A i y I and Gi s yÝ Njsiq1 A j . Engle and Granger Ž1987. assert that if the coefficient matrix P has reduced rank r - k, then there exists k = r matrices a and b with rank r such that P s ab X and b X yt is stationary; r is the cointegrating rank, and each column of b is the cointegrating vector. The elements of a are known as the adjustment parameters. Johansen’s Ž1995. method is to estimate the P matrix in an unrestricted form, then test whether we can reject the restrictions implied by the reduced rank of P . We will apply the FIML approach of Johansen to test for cointegration and to construct the lagged error terms Žsee Johansen, 1988; Johansen and Juselius, 1990.. A major problem when testing for cointegration among N I Ž1. series is that one can have up to N y 1 cointegrating vectors. Thus, a direct interpretation of the cointegrating coefficients is difficult ŽLutkepohl and Reimers, 1992.. Since impulse–response functions do not have this ambiguity, we use them to estimate the long-run elasticities. The long-term dynamics can be analyzed by tracing the impact of an innovation on a single endogenous variable on the evolution of the entire system. A shock to the k th variable directly affects the k th variable and is also transmitted to all of the endogenous variables through the dynamic structure of the VAR ŽVEC. model. The results of such an analysis are called impulse–response analysis. ŽIn the interests of space, we do not discuss this in detail and refer the interested reader to Lutkepohl and Reimers, 1992; Dekimpe and Hanssens, 1999; Bronnenberg et al., 2000, for examples.. Using impulse–response analysis, one can incorporate both the direct effect of price on share and the indirect effect of price on share through competitive prices. Hence, we use impulse–response analysis to estimate comprehensive share and competitive response to temporary changes in prices and evolving prices. 4. Data description To estimate our models, we use data supplied by Information Resources and AC Nielsen. The former consists of aggregate weekly store level scanner data for beer in a test market in the US from 1989 to 1996. This results in 365 weekly observations. Information is available on weekly unit sales, price, price specials, local print advertising and display activity. We used AC Nielsen household scanner data on purchases in the margarine category to construct time series of shares, prices and promotional variables. This data set has been used extensively in the recent marketing literature Že.g., Chintagunta, 1993.. The brand-level variables are created from skulevel variables. This aggregation from sku- to brandlevel may be subject to aggregation bias due to possible heterogeneity among promotional variables ŽChristen et al., 1997; Pesaran and Smith, 1995.. To control for this bias, we performed pooling tests to determine whether we can pool the different varieties for a brand; over 95% of sales could be pooled.3 However, our estimates of elasticity are consistent with prior estimates Že.g., Tellis, 1988., thus alleviating, to a certain extent, the likelihood of substantial aggregation bias. Further, as pointed out by Kopalle et al. Ž1999. and Bucklin and Gupta Ž1999., there are also some tradeoffs in using a sku-level specification, such as collinearity and large state space. For the beer category, we considered the top five brands: Budweiser, Miller, Busch, Old Milwaukee and Milwaukee’s Best. For the margarine category, we considered four major brands: Blue Bonnet, Parkay, Fleischman and a major private label brand. Market share of a brand is calculated as the share of category volume, while price is the inflation-adjusted average weekly price per ounce. Price specials, displays and advertising are indicator variables. However, since not all stores promote a brand uniformly, we specified the promotional variables based on the 3 Christen et al. Ž1997. point out that the aggregation bias is likely to be quite small in data characterized by three conditions: frequent promotions, frequent price cuts, and small own priceelasticities. We note that all three conditions are met for the margarine data while the first two conditions are met for the beer data. S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 percentage of units sold on deal, advertising and display in a given week Žsee, e.g., Wittink et al., 1988.. 5. Empirical results 5.1. UniÕariate stationarity tests and testing for structural breaks Fig. 1 shows the market share and price series for Miller and Budweiser brands Žto preserve space, we only include graphs of the major brands, but we summarize the findings below.. Table 2 displays the results of the unit-root tests. To assess the stable or evolving character of market shares and marketingmix variables, we applied several versions of the Augmented Dickey–Fuller ŽADF. test. Table 2 lists the test statistics under the column labeled ADF.4 Results for the beer category indicated that the market shares and prices of the four major brands of beer—Budweiser, Miller, Busch and Milwaukee’s Best—were evolving while Old Milwaukee’s share and price were stationary. All advertising, display and price special variables were stationary for the five major brands of beer. The Perron and Vogelsang Ž1992. procedure was used to control for structural breaks in the data with evolving series Žsee column labeled Min ta ., but our substantive conclusions were not affected for the beer category. Fig. 2 shows the market share and price series for Blue Bonnet and Parkay brands of margarine. ŽAgain, to preserve space, we only include graphs of the two major brands, but we summarize the findings below.. For the margarine category, results shown in Table 2 indicated that the market shares of Blue Bonnet and Fleischman were evolving while the market shares of Parkay and the private label brand were stationary.5 4 We started by including 12 lags, determined the greatest lag with a significant t-value and retested using this number of lags. A constant was included, and we used both the Schwarz criterion and the log-likelihood ratio test. All results were consistent. 5 The remaining brands consisted of smaller brands and other private labels and were not included in the analysis; given that shares sum to one, it is conceptually impossible to have one share series evolving while the other four shares are evolving. 291 However, the results showed that the prices of all the brands were evolving. All advertising, display and price special variables were stationary except for the price specials of Parkay and Blue Bonnet. Our substantive conclusions were affected as described below when the Perron and Vogelsang Ž1992. procedure was used to control for structural breaks Žsee column labeled Min ta .. The market share and price of Blue Bonnet were stationary with a break point occurring in week 62 as seen in Fig. 1c and d. The market share and price of Fleischman were found to be stationary with a level shift occurring in week 74. The price of Parkay was stationary with the break point occurring in week 75. The Perron and Vogelsang Ž1992. procedure indicated the presence of structural breaks in these series; nevertheless, we additionally conducted F-tests to test for equality of means before and after the structural break for these series. These test results shown in Table 2 were consistent with the results obtained from the structural break tests in that we rejected the null hypothesis of equality of means of these series before and after the break point. In week 62, Blue Bonnet reduced its price, resulting in a significant increase in market share Žsee Fig. 2a and b.. Parkay responded both with price specials in the short run and with a structural price change in week 75. Parkay’s share had a small decline after the price decrease for Blue Bonnet and a slight increase after a reduction in Parkay’s price; however, its share remained stationary as evidenced by the rejection of the unit-root test. The market shares of Fleischman and the private label brand were not affected by the change in the pricing strategy of Blue Bonnet; most of the increase came at the expense of smaller brands and the remaining private labels. Fleischman’s break in week 74 was accompanied by a share increase; this too was at the expense of smaller brands and the remaining private labels. Price reductions of national brands resulted in an asymmetric loss of share of the small brands and private label brands rather than other national brands, consistent with the findings of Blattberg and Wisniewski Ž1989.. Thus, in the short run, there is an immediate response by major competitors in the form of temporary price specials; therefore, their shares remain stable. However, it takes 12 and 13 weeks for Fleischman and Parkay, respectively, to respond with a 292 S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 Table 2 Non-stationarity tests for beer and margarine data Variable ADF Min ta ADF outcome Structural break test outcome Žbreak point. Beer Share Budweiser Share Busch Share Miller Share Old Milwaukee Share Milwaukee Best Price Budweiser Price Busch Price Miller Price Old Milwaukee Price Milwaukee Best Advertising Budweiser Advertising Busch Advertising Miller Advertising Old Milwaukee Advertising Milwaukee Best Display Budweiser Display Busch Display Miller Display Old Milwaukee Display Milwaukee Best Special Budweiser Special Busch Special Miller Special Old Milwaukee Special Milwaukee Best y2.78 y1.86 y2.46 y3.33 y1.61 y2.58 y2.54 y2.07 y3.98 y1.99 y19.27 y19.14 y18.55 y17.15 y8.61 y11.38 y11.42 y12.02 y6.11 y5.30 y12.02 y5.15 y12.21 y3.46 y3.79 y3.01 y2.09 y2.68 Evolving Evolving Evolving Stationary Evolving Evolving Evolving Evolving Stationary Evolving Stationary Stationary Stationary Stationary Stationary Stationary Stationary Stationary Stationary Stationary Stationary Stationary Stationary Stationary Stationary Evolving Evolving Evolving Stationary Evolving Evolving Evolving Evolving Stationary Evolving Stationary Stationary Stationary Stationary Stationary Stationary Stationary Stationary Stationary Stationary Stationary Stationary Stationary Stationary Stationary Margarine Share Blue Bonnet Share Parkay Share Fleischman Share Private Label Price Blue Bonnet Price Parkay Price Fleischman Price Private Label Advertising Blue Bonnet Advertising Parkay Advertising Fleischman Advertising Private Label Display Blue Bonnet Display Parkay Display Fleischman Display Private Label Special Blue Bonnet Special Parkay Special Fleischman Special Private Label y2.33 y7.52 y2.54 y6.26 y1.65 y2.34 y2.80 y1.65 y12.76 y10.81 y10.42 y3.17 y6.87 y6.21 y3.74 y7.25 y2.49 y1.57 y10.38 y10.61 y8.74 Evolving Stationary Evolving Stationary Evolving Evolving Evolving Evolving Stationary Stationary Stationary Stationary Stationary Stationary Stationary Stationary Evolving Evolving Stationary Stationary Stationary Žweek 62. Stationary Stationary Žweek 74. Stationary Stationary Žweek 62. Stationary Žweek 75. Stationary Žweek 74. Evolving Stationary Stationary Stationary Stationary Stationary Stationary Stationary Stationary Evolving Stationary Stationary Stationary y1.83 y2.84 y2.91 y2.27 y2.24 y9.60 y3.97 y9.55 y9.93 y1.47 y2.78 y3.35 F-test outcome for equality of meansa Reject equality of means Reject equality of means Reject equality of means Reject equality of means Reject equality of means Reject equality of means a F-test for equality of means before and after the break point for series where unit-root tests indicated a structural break; t-statistic for the parameter estimate associated with the lagged level in the ADF equation, to be compared with the 5% critical value of y2.93; critical values for the structural-break tests are from Perron and Vogelsang Ž1992.. S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 structural change in prices. In general, when reactions occur within four periods, they are thought to be retailer-dominated while longer duration responses are manufacturer-dominated Žsee LW, 1992; Kopalle et al., 1999.. Even though the shares of Fleischman and Parkay are not significantly affected Ždue to their immediate response in the form of price specials., these brands did respond to the structural price change since their shares relative to Blue Bonnet had decreased. 5.2. Impact of a structural change in prices on market share using structural break analysis We included three shift dummy variables I1, I 2 and I 3 in the VAR model for margarine corresponding to the structural breaks. The results of the parameter estimates of the reduced form of the VAR model for margarine are provided in Table 3. Results in Table 3 show that only the own structural price break for a series is significant.6 To guide model specification with respect to the number of lags, model testing was conducted using the log-likelihood criterion and the Schwarz criterion.7 The results were consistent; a second-order VAR was selected for the margarine data. Following Dekimpe and Hanssens Ž1999., we only included the parameter estimates that had a t-value greater than one.8 The 6 As an example, for Blue Bonnet, we include the dummy variable for break 1 Žindicating Blue Bonnet’s price break.. To determine the impact of competitive price breaks, we first estimate a model that includes a dummy for break 1 and break 2 followed by a model that includes break 1 and break 3 dummies. This is done since the dummies for break 2 and break 3 are highly collinear. 7 Though the likelihood-ratio test and AICrSIC information criterion approach are equivalent, the real difference is in the significance levels, which are usually fixed at 5% in hypothesis testing, and are selected automatically Žtypically starting at more than 10% in small samples but converging to 0 for larger ones. in the information criterion approach. Loose significance levels are not necessarily the best choice and the discussion on this issue cannot be regarded as closed with a clear guideline. Model choice by residual correlation statistics Žsuch as Ljung-Box Q. has often been reported to yield the poorest results. 8 Since lagged market share is not significant in any of the price equations estimated, the price equations are the same for the four VARs estimated Žsee Eq. Ž1... 293 results of these analyses are used to estimate the impact on shares of a change in prices. However, we first need to check for parameter constancy. For Blue Bonnet’s market share equation, the plot of recursive residuals does not indicate instability, as seen in Fig. 3a. We also performed a series of Chow forecast tests, the CUSUM test based on recursive residuals and the CUSUM-of-squares test and examined the recursive estimates of the parameters of the VAR; the results were consistent. These tests conducted for all the remaining brands also did not indicate instability and therefore are not included. Thus, the structural shift in price does not lead to a change in the response coefficients. To graphically show the impact of the change in pricing strategy for Blue Bonnet, we forecast market share from the break in week 62 onward using the VAR parameter estimates, assuming no price break in the price series of Blue Bonnet. This gives us a new market share series that can be compared with the actual market share series to get a measure of the effect of the structural shift in prices. This is illustrated for Blue Bonnet in Fig. 3b and c, before and after the structural break, respectively. There is a substantial one-time increase in share in response to a structural price change after which share stabilizes. We predict market shares from week 74 using the VAR parameter estimates under two scenarios. Under scenario one, we assume that prices of both Fleischman and Parkay do not have a structural change or are at the price level before the structural break. Under scenario two, Parkay’s price is set to the new level corresponding to the price level after the structural break while Fleischman’s price is set at the price level before the structural break. We compare these scenarios with the actual market share corresponding to the case where both the prices have a structural break. We find that Fleischman’s share is lower under scenario two as compared to scenario one; subsequent competitive reaction does affect the prediction of price response and needs to be accounted for by managers. Therefore, it is important for managers to use conjectures regarding price reaction in assessing the impact of policy changes ŽJeuland and Shugan, 1988.. Using the expression for the equilibrium of the VAR model in Eq. Ž7. and estimates from Table 3, we assess the effect of a changed level of price on 294 Table 3 Results of VAR model for margarine Share Blue Bonnet Share Parkay Share Fleishman Share Private Label Price Blue Bonnet Price Parkay Price Fleishman Constant Share Žy1. Break 1 Break 2 Break 3 Price Fleishman Žy1. Price Fleishman Žy2. Price Blue Bonnet Žy1. Price Blue Bonnet Žy2. Price Parkay Žy1. Price Parkay Žy2. Price Private Label Žy1. Price Private Label Žy2. Advertising Fleishman Advertising Blue Bonnet Advertising Parkay Advertising Private Label Display Fleishman Display Blue Bonnet Display Parkay Display Private Label Special Private Label y0.593 ) y1.689 ))) 0.277 ))) y0.162 ) y1.451))) 0.205 ))) y1.347 ) y1.252 ))) y1.662 ))) y2.139 ))) a 0.277 ))) 0.080 ))) y0.083 ))) 0.260 ))) y0.091) y0.085 ))) 0.407 ) 0.638 y0.184 ) 0.171) 0.380 ))) 0.516 ))) y0.233 ) 0.179 ) y0.619 ))) 0.709 ))) y0.051) y0.210 ))) 0.729 ))) y0.249 ))) 0.332 ) y0.158 ))) y0.354 ))) y0.186 0.206 0.192 ))) 0.336 ))) y0.267 )) y0.099 y0.210 ))) 0.295 ))) First-differenced. Indicates p- 0.10 Žvia two-tailed tests.. )) Indicates p- 0.05. ))) Indicates significant at p- 0.01. ) Price Private Label a y0.388 ))) 0.057 y0.225 ))) y0.242 ) 0.745 ))) y0.202 ))) y0.246 ) y0.494 ))) y0.248 ))) y0.413 y0.420 ))) 0.158 ) y0.572 ))) y0.064 0.180 ))) S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 Variable S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 295 Fig. 3. Ža. Test of parameter stability: plot of recursive residuals for Blue Bonnet. Žb. Market share for Blue Bonnet brand prior to the structural break. Žc. Market share for Blue Bonnet brand after the structural break. Žd. Market share for Fleischman brand. the level of market share. The values of prices for the brands of margarine are set equal to the new level in the system of simultaneous equations indicated in the reduced form equations; the system solved for the new levels of the remaining variables. The results are shown in Table 4. The predicted levels of market share for Blue Bonnet are 16.9% before the structural break in week 62 and 20.3% after; actual shares are 17.6% and 20.4%, respectively. For Fleischman, the predicted levels of share are 8.4% before the structural break in week 74 and 10.8% after; actual shares are 8.2% and 10.3%, respectively. The predicted share levels are quite good, as evident in Fig. 3. The own-price elasticity estimates corresponding to a structural change in price for Blue Bonnet, Parkay and Fleischman brands Table 4 Impact of structural changes in prices for margarine Brand Blue Bonnet Fleishman Parkay Private Label a Market share before price break a Market share after price break Actual Ž%. Predicted Ž%. Actual Ž%. Predicted Ž%. 17.6 8.2 28.9 7.4 16.9 8.4 29.8 7.6 20.4 10.3 28.9 7.4 20.3 10.8 29.8 7.6 Share response elasticity y1.37 y1.55 0 – The price breaks for Blue Bonnet, Fleischman and Parkay occur at week 62, 74 and 75, respectively; market shares do not add up to 1 since smaller brands and other private labels account for the rest. 296 S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 Table 5 ŽA. Impact of temporary changes in prices for margarine Žstandard error in parentheses. Effect of independent variable Response effects of price on share Blue Bonnet Parkay Fleischman Private Label Lagged response effects of price on share Feedback effects of share on price Purchase reinforcement effects of share Žon share. CompetitiÕe price reaction effects Blue Bonnet Parkay Response elasticity Žeffect on market share or price of. Blue Bonnet Parkay Fleischman Private Label y0.578 )) Ž0.114. 0.541 Ž0.369. n.s. 0.187 ) Ž0.245. y0.439 )) Ž0.201. n.s. n.s. n.s. 0.683 Ž0.429. n.s. 0.194 Ž0.125. n.s. y0.119 Ž0.073. y0.392 ))) Ž0.151. n.s. 0.213 )) Ž0.109. y0.439 ) Ž0.255. 0.133 )) Ž0.065. 0.305 Ž0.259. n.s. n.s. 0.231))) Ž0.081. 0.119 Ž0.115. – 0.123 Ž0.123. – 0.218 Ž0.186. 0.290 ) Ž0.153. – n.s. Private Label 0.076 Ž0.073. 0.056 Ž0.055. n.s. Decision rules effects of price Žon price. 0.447 ))) Ž0.115. Fleischman 0.159 )) Ž0.077. n.s. 0.316 ))) Ž0.089. y0.495 Ž0.322. n.s. y0.223 ) Ž0.132. n.s. 0.073 ) Ž0.04. 0.337 ))) Ž0.079. n.s. n.s. y0.035 Ž0.029. – y0.340 )) Ž0.087. ŽB. Impact of temporary changes in prices for beer Žstandard error in parentheses. Effect of independent variable Response elasticity Žeffect on share or price of. Budweiser Busch Miller Old Milwaukee Milwaukee’s Best y1.810 ))) Ž0.203. 0.363 ))) Ž0.111. 0.522 ))) Ž0.116. n.s. 0.430 ))) Ž0.162. y2.420 ))) Ž0.124. n.s. 0.658 ))) Ž0.171. n.s. 0.457 ))) Ž0.157. 0.434 ))) Ž0.107. n.s. 0.371)) Ž0.166. n.s. 0.365 ))) Ž0.118. n.s. 0.521))) Ž0.120. y1.233 ))) Ž0.077. CompetitiÕe price reaction effects Budweiser – n.s. Busch n.s. – Response effects of price on share Budweiser Busch Miller Old Milwaukee Milwaukee’s Best Lagged response effects of price on share n.s. y1.832 ))) Ž0.144. n.s. y0.622 ))) Ž0.088. y1.857 ))) Ž0.160. 1.030 ))) Ž0.133. 0.212 )) Ž0.097. 0.466 ))) Ž0.064. 0.268 ))) Ž0.037. 0.174 ) ) Ž0.078. y0.141))) Ž0.047. n.s. 0.436 ))) Ž0.103. 1.025 ))) Ž0.131. y1.640 ))) 0.164. n.s. n.s. n.s. S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 297 Table 5 Ž continued . Effect of independent variable Response elasticity Žeffect on share or price of. Budweiser Busch Miller Old Milwaukee Milwaukee’s Best 0.274 ))) Ž0.045. n.s. – n.s. n.s. – Milwaukee’s Best 0.212 ))) Ž0.062. 0.109 ) Ž0.060. n.s. n.s. n.s. Decision rules effects of price Žon price. 0.222 ))) Ž0.035. 0.527 ))) Ž0.026. 0.545 ))) Ž0.032. 0.193 ))) Ž0.074. 0.320 ))) Ž0.030. 0.119 ) Ž0.064. 0.200 ))) Ž0.07. – CompetitiÕe price reaction effects Miller Old Milwaukee 0.232 ))) Ž0.029. n.s. indicates estimates with t-value less than one. ) Indicates p - 0.10 Žvia two-tailed tests.. )) Indicates p - 0.05. ))) Indicates significant at p - 0.01. are y1.37, 0 and y1.55, respectively. Thus, for two out of the three brands with structural breaks in price, shares are quite responsive to structural changes in prices. In Section 5.3, we compare the relative effect of temporary and structural changes in price for margarine. 5.3. Impact of temporary changes in prices on market share To assess the impact of temporary changes in prices on shares for the margarine and beer markets, impulse–response analyses are performed. For the beer data, where shares and prices are evolving, cointegration analysis is first performed on market shares, price and competitive prices Žsee Eq. Ž8... Advertising, display and price specials are included as exogenous variables. Separate models are estimated for each brand in each product class. Once the number of lags is determined, the VECM is estimated using the Johansen and Juselius Ž1990. method. The PcFiml software is used for estimation purposes ŽDoornik and Hendry, 1996.. This method determines the cointegration rank, the number of significant cointegrating vectors and the associated cointegrating vectors with the long-run parameters. Due to interpretation problems with cointegration vectors, we use impulse–response functions to assess both the short-run and the long-run price elasticities. In the interests of space, we do not discuss in detail the coefficients of the five separate VEC models estimated for each of the five brands but rather interpret the coefficients of the impulse–response analysis incorporating both direct and indirect effects. Since the impulse–response functions were estimated for the transformed data expressed in natural logarithms, these were used to assess the short-run elasticities or the percentage change in the dependent variable due to a percentage change in the shocked variable.9 A potential limitation of the impulse–response analysis is that results may not be invariant to the ordering of the endogenous variables when the error terms are correlated ŽLutkepohl and Reimers, 1992.. However, when the residual correlations are small, results are robust to the causal ordering. Enders Ž1995. suggests a cut-off value of 0.20; if the correlation coefficient between, for example, e BBP,t and e PKP,t in Eq. Ž1., N r N is - 0.2, the correlation is deemed not to be significant. In all instances, we found correlations below 0.20. Furthermore, we also 9 To see this, let the impulse–response be denoted by n as a result of a one standard deviation Ž s . shock to price. Since we are dealing with natural logarithms, we have lnŽMS tq 1 .ylnŽMS t . s n or lnŽMS tq 1 .rlnŽMS t . s n . With algebra, one can show that percentage change in market share is DMS t rMS t sexpŽ n .y1. In a similar manner, the percentage change in price is D Pt r Pt s expŽ s .y1. Therefore, the price elasticity of market shares ŽexpŽ n .y1.rŽexpŽ s .y1.. The standard errors are obtained using the Delta method Žsee Kmenta, 1997.. 298 S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 performed impulse–response analysis using different ordering of the variables, and results were very similar.10 The results of the impulse–response analysis for margarine and beer data are reported in Table 5a and b, respectively. The parameter estimates for the short-run own and cross-price elasticities are consistent with our expectations and with the results reported in the prior literature ŽTellis, 1988; Wittink et al., 1988.. From Table 5a, the short-run price elasticities of share response for the margarine brands are smaller and range from y0.22 to y0.57; however, these results are consistent with Blattberg and Wisniewski Ž1989. for the margarine category. In terms of the cross-elasticities, temporary changes in price of the private label brand seems to be more effective in drawing share from Fleischman and Blue Bonnet than vice versa. This is consistent with Bronnenberg and Wathieu Ž1996. whose results suggest that private label brands could promote more effectively than national brands where the former have advantageous positions in terms of quality relative to price. With regard to temporary price changes, on the one hand, consumers may believe that the change is temporary and hence buy more on deal or accelerate their purchase. On the other hand, if a consumer sees a brand’s price that is outside the acceptable range of prices, it is contrasted with the acceptable range and becomes noticeable, consistent with AssimilationContrast Theory ŽKalyanaram and Little, 1994.. Thus, structural price changes could lead to larger sales increases. Consistent with this, our results suggest that for two of the three brands with a structural change in prices, Blue Bonnet and Fleischman, shares are more responsive to structural price changes than to temporary price changes. There is a significant lagged response effect of price on share for Parkay ŽBass and Clarke, 1972.. The influence of lagged prices on current prices is positive and significant for Parkay and Blue Bonnet, consistent with inertia in decision-making. Current period market shares are positively related to past market shares for Parkay and the private label brand, 10 Dekimpe and Hanssens Ž1999. utilize the information in the residual correlations matrix to derive a vector of expected shock values and to simultaneously shock all error terms in the system. consistent with purchase reinforcement ŽGivon and Horsky, 1990.. Only for the private label brand do we find significant feedback effects; price depends on market performance in the prior period ŽVillasBoas and Winer, 1999.. The results of competitive price reactions show that the price reactions are positive between Parkay and Fleischman, indicative of manufacturer pricing activity ŽLW, 1992.. Table 5b shows that the short-run price elasticity of share response for the beer brands ranges from y1.64 to y2.42 while the cross price elasticity ranges from 0.36 to 1.03. The brands with the highest levels of share ŽBudweiser and Miller. tend to have lower sensitivity to price compared to lower-selling brands ŽBusch and Old Milwaukee., suggesting that temporary price changes Žprice promotions. would be more effective in the short run for the latter brands than for the former. There are significant lagged response effects for three brands; the effect for Old Milwaukee is consistent with alternating promotions. However, we do not observe significant feedback and purchase reinforcement effects for any of the brands. It appears that in the nonstationary beer market, market share is mostly influenced by own and competitive marketing mix variables. With regard to competitive reaction, there is only one negative price reaction, indicative of the alternating price promotions of retailers ŽLW, 1992.. For the rest of the brands, price reactions are positive; this reflects the effect of competitive manufacturer pricing activity, as in LW Ž1992.. The estimated price reactions are close to those obtained by Kopalle et al. Ž1999.. 5.4. Impact of eÕolÕing prices on market share The impact of evolution in prices on market shares is shown in Table 6. Old Milwaukee’s price and share are stationary; hence, the results for Old Milwaukee are not shown in Table 6. The calculation of the long-run price elasticities is similar to Dekimpe and Hanssens Ž1999, footnote 11.. For example, a shock to Miller’s price results in a sustained change in Miller’s price of 0.58%. This change results in sustained share losses of y0.68%. Therefore, the long-run price elasticity of Miller is y1.174. All other elasticities are estimated in a similar manner, and standard errors are obtained using the Delta method ŽKmenta, 1997.. S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 299 Table 6 Impact of evolving prices for beer Žstandard error in parentheses. Effect of independent variable Response elasticity Žeffect on price or share of. Budweiser Busch Miller Milwaukee’s Best n.s. n.s. n.s. y2.319 ))) Ž0.024. n.s. n.s. n.s. Milwaukee’s Best y0.745 ))) Ž0.084. y0.384 )) Ž0.168. 0.288 ))) Ž0.085. n.s y1.174 ))) Ž0.022. n.s. 0.270 ))) Ž0.047. y1.029 ))) Ž0.367. CompetitiÕe price reaction effects Budweiser – 0.555 Ž0.514. – 1.570 ))) Ž0.371. 0.382 ))) Ž0.058. – 1.030 Ž0.715. 0.101 Ž0.085. 0.388 ) Ž0.204. – Response effects of price on share Budweiser Busch Miller Busch Miller Milwaukee’s Best y0.087 Ž0.080. 0.323 )) Ž0.108. 0.203 ))) Ž0.076. n.s. 0.378 )) Ž0.160. 0.178 ))) Ž0.069. 0.465 ))) Ž0.132. n.s. indicates estimates with t-value less than one. ) Indicates p - 0.10 Žvia two-tailed tests.. )) Indicates p - 0.05. ))) Indicates significant at p - 0.01. With regard to evolving prices, on the one hand, consumers may become price sensitive over time due to lowered price levels ŽMela et al., 1997.. On the other hand, a succession of price discounts may reduce reference prices of consumers, leading to a lower level of utility for a given level of discount ŽKalyanaram and Winer, 1995.. Thus, price reductions occurring after discounts would lead to smaller sales increases Žprice sensitivity could be lower.. Our results show that the short-run price elasticities, averaging close to y2, are greater than the long-run price elasticities, averaging approximately y1.3. With regard to price promotions, consumers may be cued that the change is temporary and therefore may stockpile andror switch to other brands to take advantage of the deal. With respect to the long run, the cross-price elasticities are consistent with our expectations except for Busch’s price; it has a negative impact on Budweiser’s share. Since both Busch and Budweiser are sister brands of the same parent company, Anheuser–Busch, there may be a certain degree of complementarity between them. On the one hand, we find that Miller’s share is more sensi- tive to temporary price changes in Budweiser than vice versa. On the other hand, Budweiser is relatively more responsive to Miller’s long-run price levels. The long-run own and cross-price elasticities of Budweiser and Miller are consistent with the share patterns over time, which indicate that Miller has gained share relative to Budweiser, while prices of both brands have decreased over time. Fig. 4a–b show the short- and long-run share response of market share of Miller to its prices and those of Budweiser. Thus, evolving prices have permanent effects on market shares in the US beer market. This is consistent with the results reported by Johnson et al. Ž1992. in their analysis of the Canadian beer market province-by-province. With respect to competitive reaction elasticities, there are positive price reactions suggesting that there is Aprice-matchingB behavior in the long run. This provides evidence of the Adouble jeopardyB effect of price promotions. A negative price shock results in a strong competitive reaction. This, in turn, leads to a persistent reduction both in price and competitive prices. This is in contrast to the response 300 S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 Fig. 4. Ža. Long-run impact of a negative unit % shock to Miller’s price on Miller’s share. Žb. Long-run impact of a negative unit % shock to Budweiser’s price on Miller’s share. Žc. Long-run impact of a negative unit % shock to Budweiser’s price on Miller’s price. Žd. Long-run impact of a negative unit % shock to Miller’s price on Budweiser’s price. to structural changes in prices where competitors respond with a lag of 12 to 13 weeks. Structural price changes elicit delayed competitive response since strategic decisions concerning price levels tend to be manufacturer-dominated while temporary changes in prices and evolving prices elicit immediate response because tactical decisions concerning temporary promotions tend to be retailer-dominated ŽLW, 1992.. Moreover, competitive price response is likely to be delayed when prices have been relatively stable and thus the price change is unanticipated— structural change—as opposed to situations where price movements have been continual-evolving prices. Fig. 4c–d show the impulse–response function of price due to shocks to competitive price for Budweiser and Miller. LW Ž1996., for instance, provide empirical evidence for the fact that managers tend to react to changes in a competitor’s marketing instrument, even when the brand’s performance is not affected by that competitive change. Taken together, the results suggest that due to competitive interactions of prices, short-run promotions may lead to permanent reductions in price levels which impact long-run shares due to different own- and cross-elasticities. Finally, from a strategic marketing perspective, we now address the issue of whether a structural decrease in price is preferred to prices that are evolving downwards.11 First, our empirical results show that evolution in prices causes evolution in market shares, whereas a structural change in price results in an immediate shift in level of market share and hence allows the firm to get to 11 We are thankful to Mike Hanssens and an anonymous referee for motivating the following discussion. S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 the desired market share level sooner. Second, competitive reaction is likely to be delayed because of the unexpected nature of a structural change and due to the fact that structural price changes tend to be manufacturer-dominated. Finally, consumers tend to notice and contrast prices that lie outside their range of acceptance with the acceptable range, and hence, share response to structural price changes could be higher. Thus, a structural decrease in price could be more profitable. As a result, a company that is profitable in the short run may be able to replicate its marketing tactics, a potentially beneficial outcome in the long run. v 6. Conclusions We offer novel findings and new contributions on the three types of price changes—temporary, evolving and structural—to the literature in this area. Table 7 summarizes the key findings of our study; we emphasize the main managerial implications, from both a tactical and a strategic perspective, in the following discussion. v 6.1. Managerial implications based on our empirical analysis v Distinguish among temporary changes in prices, eÕolÕing prices and structural changes in prices, as their impact on market shares tends to differ. While temporary price changes only cause a temporary change in shares, both structural changes in price and evolving prices have permanent effects on market shares. Structural shifts to levels of price do shift levels of market share for the margarine data. On the one hand, in the margarine category, for two out of the three brands, share response to structural price changes is greater than response to temporary price changes. On the other hand, we find that shortrun price elasticities—response to temporary changes—are greater than the long-run price elasticities—response to evolving prices—in the beer category. These differences may be partly due to category-specific effects; temporary price v 301 discounts may be more likely lead to increased consumption for beer than for margarine. When should firms use one type of price change Õs. another? From a tactical perspective, temporary price variations may be more effective at increasing shares for lower-selling brands rather than for higher-selling brands in the short run, as evidenced in the beer category. From a strategic perspective, managers may ask this question: If I am in a price decrease mode, should I offer a series of small but continuous price decreases or a single large price decrease? Structural price cuts allow the firm to get to the desired market share level sooner and elicit delayed competitive reaction. Moreover, they generate greater share response because when consumers see a brand’s price that is outside the acceptable range of prices, it is contrasted with the acceptable range and becomes more noticeable. Hence, our results suggest that a structural price cut is preferable to a series of continuous price decreases. Competitors do not respond the same way to the three types of price changes. Structural price changes elicit delayed competitive response since strategic decisions concerning price levels tend to be manufacturer-dominated while temporary changes in prices and evolving prices elicit immediate response since tactical decisions concerning temporary promotions tend to be retailer-dominated. Moreover, competitors are likely to respond with a delay when prices have been relatively stable and therefore the price change is unanticipated as opposed to situations where price movements have been continual. Our results provide evidence of the Adouble jeopardyB effect of price promotions as a result of temporary price changes. From a manager’s point of view, it is clearly not prudent to promote a brand frequently, thus inviting immediate competitive reaction. This, in turn, leads to a persistent reduction of prices in the category in the long run. Anticipate competitiÕe response to price changes. Subsequent competitive reaction affects predictions of price response. Firms should try to anticipate their competitors’ behavior, i.e., form conjectures regarding competitive price re- 302 Table 7 Pricing behavior: comparison of temporary price changes, evolving prices and structural price changes Illustration Empirical treatment Exemplar Substantive implications based on our empirical analysis ŽI. Temporary price changes Short-term promotions Že.g., 50 cents off a six-pack of beer.. ADF unit-root test All major brands in margarine and beer categories Ži. Only a temporary effect on shares. Žii. Short-run response elasticities are greater than long-run elasticities Ževolving prices.. Žiii. Temporary price changes could be more effective for lower-selling brands than for higher selling brands. Živ. Strong competitive reaction consistent with retailer-dominated reactions. ŽII. Evolving prices Regular practice of offering discounts Že.g. airline pricing tactics in 1992.. ADF unit-root test Four major brands in the beer category Ži. Significant permanent effects on share could lead to evolution in share. Žii. Response elasticities lower than those for temporary price changes. Žiii. Immediate competitive reactions consistent with retailer-dominated reactions. Živ. Long-run price matching behavior Žprice wars.. Žv. Could be less profitable than structural changes since the latter allow the firm to get to desired market share level sooner. ŽIII. Structural changes in price A company offers a significant one-time price cut or increase to a new level Že.g. Kellogg’s price-cut of 20% for cereal.. Structural break test ŽPerron and Vogelsang, 1992. Three brands in the margarine category Ži. Permanent effects on share could lead to shift in share level. Žii. No change in consumer response coefficients. Žiii. Subsequent competitive reaction does affect the prediction of price response and needs to accounted for by managers. Živ. Share response could be greater than response to temporary changes. Žv. A structural decrease in price could be more profitable since firm gets to desired share level sooner. Žvi. Major competitors respond with a 12 to 13 week lag to an unexpected price break consistent with retailer-dominated reactions. S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 Pricing behavior S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 actions in assessing the impact of policy changes. Firms should use competitive intelligence to understand competitor’s motives, capabilities and potential reactions. 6.2. Limitations and future research Our study has some limitations. First, we estimated separate models for each brand due to data restrictions and degrees-of-freedom problems when estimating extended VAR models in a simultaneous system. In terms of future research, it would be fruitful to develop a model in which all competitors are considered simultaneously. Second, our analysis pertains to two categories. Future research concerning the nature of the long-run response to permanent changes in levels of prices may benefit from systematic collection of data across different conditions and circumstances so as to provide empirical generalizations. Third, while these findings are very important to management, more research is needed to determine the profitability of these results. However, this issue has been under-researched due to the unavailability of suitable data. Jedidi et al. Ž1999. and Dekimpe and Hanssens Ž1999. have made a first attempt to address the profitability of different long-run marketing strategies. Finally, some pricing decisions need attention at the brand manager level while others need attention at the retail-account level. This, in turn, leads to a demand for analysis at the store level rather than the brand level. Academic research has only now begun to address this topic ŽBucklin and Gupta, 1999.. These are key issues for future research. Acknowledgements We are indebted to Information Resources and Peter Rossi, University of Chicago, for providing the data. We thank the Editor, Jan-Benedict Steenkamp and Bart Bronnenberg, Marnik Dekimpe, Philip Hans Franses, Dominique M. Hanssens and Dick R.Wittink. We also thank the seminar participants at the University of California, Riverside, the University of California, Irvine, the 1999 Marketing Science Con- 303 ference and the Conference on Competition and Marketing, Mainz, Germany, for their invaluable comments and suggestions. Funding for this research was provided by the Academic Senate Research Fund from the University of California, Riverside and the Social Sciences and Humanities Research Council of Canada. References Baghestani, H., 1991. Cointegration analysis of the advertising– sales relationship. Journal of Industrial Economics 34, 671– 681. Banerjee, A., Dolado, J., Galbraith, J.W., Hendry, D.F., 1993. Cointegration, Error-Correction, and the Econometric Analysis of Non-Stationary Data. Oxford Univ. Press, New York. Bass, F.M., 1969. A simultaneous equation regression study of advertising and sales of cigarettes. Journal of Marketing Research 6, 291–300. Bass, F.M., Clarke, D.G., 1972. Testing distributed lag models of advertising effect. Journal of Marketing Research 9, 298–308. Bass, F.M., Pilon, T.L., 1980. A stochastic brand choice framework for econometric modeling of time series market share behavior. Journal of Marketing Research 17, 486–497. Blattberg, R.C., Wisniewski, K.J., 1989. Price-induced patterns of competition. Marketing Science 8 Ž4., 291–309. Bresnahan, T.F., 1989. Empirical studies of industries with market power. In: Schmalensee, R., Willig, R. ŽEds.., Handbook of Industrial Organization vol. 2. North-Holland, New York. Brodie, R.J., Bonfrer, A., Cutler, J., 1996. Do managers overreact to each others’ promotional activity? Further empirical evidence. International Journal of Research in Marketing 13, 379–387. Bronnenberg, B.J., Wathieu, L., 1996. Asymmetric promotion effects and brand positioning. Marketing Science 15 Ž4., 379– 394. Bronnenberg, B.J., Mahajan, V., Vanhonacker, W., 2000. The emergence of market structure in new repeat-purchase categories: a dynamic approach and an empirical application. Journal of Marketing Research 37, 16–31. Bucklin, R.E., Gupta, S., 1999. Commercial use of UPC scanner data. Marketing Science 17 Ž3., 247–273. Charemza, W., Deadman, D.J., 1997. New Directions in Econometric Practice: General to Specific Modelling, Cointegration and Vector Autoregression. Edward Elgar, Cheltenham, UK. Chintagunta, P.K., 1993. Investigating purchase incidence, brand choice and purchase quantity decisions of households. Marketing Science 12 Ž2., 184–208. Christen, M., Gupta, S., Porter, J.C., Staelin, R., Wittink, D.R., 1997. Using market-level data to understand promotion effects in a nonlinear model. Journal of Marketing Research 34, 322–334. Cotterill, R., Putsis Jr., W.P., Dhar, R., 1999. Assessing the 304 S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 competitive interaction between private labels and national brands. Journal of Business, forthcoming. Davidson, T.A., Stacey, C.E., 1997. 1996 marketing analysis and modeling survey. Summary Report, Prevision, Wellesley, MA. Dekimpe, M.G., Hanssens, D.M., 1995a. Empirical generalizations about market evolution and stationarity. Marketing Science 14 Ž2., G109–121. Dekimpe, M.G., Hanssens, D.M., 1995b. The persistence of marketing effects on sales. Marketing Science 14 Ž1., 1–21. Dekimpe, M.G., Hanssens, D.M., 1999. Sustained spending and persistent response: a new look at long-term marketing profitability. Journal of Marketing Research 36 Ž4., 397–412. Dekimpe, M.G., Steenkamp, J.B., Mellens, M., Vanden Abeele, P., 1997. Decline and variability in brand loyalty. International Journal of Research in Marketing 14, 405–420. Dekimpe, M.G., Hanssens, D.M., Silva-Risso, J.M., 1999. Longrun effects of price promotions in scanner markets. Journal of Econometrics 89, 269–291. Doornik, J.A, Hendry, D.F., 1996. Modelling Dynamic Systems: Using PcFiml 9.0 for Windows. International Thompson Business Press, Boston. Ehrenberg, A.S.C., 1994. Theory or well-based results: which comes first? In: Laurent, G., Lillien, G.L., Pras, B. ŽEds.., Research Traditions in Marketing. Kluwer Academic Publishing, Boston, MA, pp. 79–105. Enders, W., 1995. Applied Econometric Time Series. Wiley, New York. Engle, R.F., Granger, C.W.J., 1987. Cointegration and error correction: Representation, estimation and testing. Econometrica 55 Ž2., 251–276. Franses, P.H., Kloek, T., Lucas, A., 1999. Outlier robust analysis of long-run marketing effects for weekly scanning data. Journal of Econometrics 89, 293–315. Franses, P.H., Srinivasan, S., Boswijk, P., 2000. Testing for unit roots in market shares. Working Paper, The A. Gary Anderson Graduate School of Management, University of California, Riverside, CA. Gasmi, F., Laffont, J.J., Vuong, Q., 1992. Econometric analysis of collusive behavior in a soft-drink market. Journal of Economics and Management Strategy 1 Ž2., 277–312. Givon, M., Horsky, D., 1990. Untangling the effects of purchase reinforcement and advertising carryover. Marketing Science 9, 171–187. Granger, C., Newbold, P., 1974. Spurious regressions in econometrics. Journal of Econometrics 2, 111–120. Hanssens, D.M., Ouyang, M., 2000. Modeling marketing hysteresis. Working Paper, The Anderson Graduate School of Management, University of California, Los Angeles, CA. Jedidi, K., Mela, C.F., Gupta, S., 1999. Managing advertising and promotion for long-run profitability. Marketing Science 18 Ž1., 1–22. Jeuland, A., Shugan, S., 1988. Channel of distribution profits when channel members form conjectures. Marketing Science 7 Ž2., 202–210. Johansen, S., 1995. Likelihood-Based Inference in Cointegrated Vector Autoregressive Models. Oxford Univ. Press, New York. Johansen, S., 1988. Statistical analysis of cointegration vectors. Journal of Economic Dynamics and Control 12, 231–254. Johansen, S., Juselius, K., 1990. Maximum likelihood estimation and inference on cointegration with applications to the demand for money. Oxford Bulletin of Economics and Statistics 52 Ž2., 169–210. Johnston, J., Dinardo, J., 1996. Econometric Methods. McGrawHill, New York. Johnson, J.A., Oksanen, E.H., Veal, M.R., Frets, D., 1992. Shortrun and long-run elasticities for Canadian consumption of alcoholic beverages: An error-correction mechanismrcointegration approach. The Review of Economics and Statistics 74, 64–74. Kadiyali, V., Vilcassim, N.J., Chintagunta, P., 1996. Empirical analysis of competitive product line pricing decisions: Lead, follow or move together. Journal of Business 69 Ž4., 459–487. Kalyanaram, G., Little, J.D.C., 1994. An empirical analysis of latitude of price acceptance in consumer packaged goods. Journal of Consumer Research 21, 408–418. Kalyanaram, G., Winer, R.S., 1995. Empirical generalizations from reference price research. Marketing Science 14 Ž2., G161–G169, Part 2 of 2. Kmenta, J., 1997. Elements of Econometric. 2nd edn. Univ. Michigan Press, Ann Arbor, MI. Kopalle, P., Mela, C.F., Marsh, L., 1999. The dynamic effect of discounting on sales: Empirical analysis and normative pricing implications. Marketing Science 18 Ž3., 317–332. Lal, R., Padmanabhan, V., 1995. Competitive response and equilibria. Marketing Science 14 Ž2., G101–G108. Lambin, J.J., Naert, P.A., Bultez, A.V., 1975. Optimal marketing behavior in oligopoly. European Economic Review 6, 105– 128. Leeflang, P.S.H., Reuyl, J.C., 1985. Competitive analysis using market response functions. 1985 AMA Educators Proceedings, Washington, DC. pp. 388–395. Leeflang, P.S.H., Wittink, D.R., 1992. Diagnosing competitive reaction using Žaggregated. scanner data. International Journal of Research in Marketing 9, 39–57. Leeflang, P.S.H., Wittink, D.R., 1996. Competitive reaction versus consumer response: Do managers overreact? International Journal of Research in Marketing 13, 103–119. Lucas, R.E., 1976. Economic policy evaluation: A critique. In: Brunner, K., Meltzer, A.M. ŽEds.., The Phillips Curve and Labor Markets. Carnegie–Rochester Conference Series on Public Policy vol. 1. North-Holland, Amsterdam. Lutkepohl, H., Reimers, H.E., 1992. Impulse–response analysis of cointegrated systems. Journal of Economic Dynamics and Control 16 Ž1., 53–78. Maddala, G.S., Kim, I.M., 1998. Unit Roots, Cointegration and Structural Change. Cambridge Univ. Press, Cambridge, UK. Mela, C.F., Gupta, S., Lehmann, D.R., 1997. The long-term impact of promotions and advertising on consumer brand choice. Journal of Marketing Research 34, 248–261. Mela, C.F., Gupta, S., Jedidi, K., 1998. Assessing long-term promotional influences on market structure. International Journal of Research in Marketing May 15 Ž2., 89–107. S. SriniÕasan et al.r Intern. J. of Research in Marketing 17 (2000) 281–305 Nijs, V.R., Dekimpe, M.G., Steenkamp, J.-B.E.M., Hanssens, D.M., 2000. The category demand effects of price promotions. Working Paper, Catholic University of Leuven, Leuven, Belgium. Pauwels, K., Hanssens, D.M., Siddarth, S., 2000. The long-term effect of price promotions on category incidence, brand choice and purchase quantity. Working Paper, Anderson Graduate School of Management, University of California, Los Angeles, CA. Perron, P., Vogelsang, T.J., 1992. Non-stationarity and level shifts with an application to purchasing power parity. Journal of Business and Economics Statistics 10 Ž4., 301–320. Pesaran, M.H., Smith, R., 1995. Estimating long-run relationships from dynamic heterogeneous panels. Journal of Econometrics 68, 79–113. Putsis, W.P., Dhar, R., 1998. The many faces of competition. Marketing Letters 9 Ž3., 269–284. Putsis, W.P., Dhar, R., 1999. Category expenditure, promotion and competitive market interaction: Can promotions really expand the pie? Working Paper, London Business School. Srinivasan, S., Bass, F.M., 2000. Cointegration analysis of brand sales and category sales: Stationarity and long-run equilibrium 305 in market shares. Applied Stochastic Models in Business and Industry 16 Ž3., 159–177. Tellis, G.J., 1988. The price elasticity of selective demand: A meta-analysis of econometric models of sales. Journal of Marketing Research 25, 331–341. Vanden Abeele, P., 1994. Commentary to: Diagnosing competition: Development and findings. In: Laurent, G., Lillien, G.L., Pras, B. ŽEds.., Research Traditions in Marketing. Kluwer Academic Publishing, Boston, pp. 79–105. Villas-Boas, M.J., Winer, R.S., 1999. Endogeneity in brand choice models. Management Science 45 Ž10., 1324–1338. Winer, R.S., 1986. A reference price model of brand choice for frequently purchased products. Journal of Consumer Research 13, 250–256. Wittink, D.R., Addona, M.J., Hawkes, W.J., Porter, J., 1988. SCAN ) PRO: The estimation, validation and use of promotional effects based on scanner data. Working Paper, Cornell University, Ithaca, NY. Wolter, J., Terasvirta, T., Lutkepohl, H., 1998. Modeling the demand for M3 in the unified Germany. Review of Economics and Statistics, 399–409.