|How retailer and competitor decisions drive the long-term effectiveness of manufacturer promotions for fast moving consumer goods
March 17, 2006
While both retailer and competitor decisions contribute to long-term promotional effectiveness, their separate impact has yet to be evaluated. For 75 brands in 25 categories, the author finds that the long-term retailer pass-through of promotions is 65%, yielding a long-run wholesale promotional elasticity of 1.78 before competitive response. However, competitors partially match the wholesale price reduction by 15%, which decreases promotional elasticity by 10%. The range of retailer and competitor response across the analyzed cases is very wide, and is affected by category and brand characteristics. As to the former, large categories yield stronger retailer response, while concentrated categories yield stronger competitor response. As to the latter, smaller brands face a fourfold disadvantage compared to leading brands: they obtain lower retail pass-through, lower retail support, and lower benefits from competing brand’s promotions, while their promotions generate higher benefits to competitors. Interestingly, the mid-90s move from off-invoice allowances towards scan-back deals only partially improves their promotional effectiveness compared to that of leading brands.
Keywords: sales promotions, long-term retailer pass-through and competitor response, jeopardy, impulse response restrictions
The long-term success of manufacturer marketing actions often depends not only on consumer response, but also on retailer and competitor reaction. For instance, Proctor & Gamble’s “value pricing” initiative was met with competitor price cuts and retailer skepticism and did not meet its objectives (Ailawadi, Lehmann and Neslin 2001). As a result, marketing managers are urged to consider all the long-run effects of their actions, including the dynamic response of consumers, retailers, and competitors (Besanko, Dubé and Gupta 2005; Chen 1996; Dekimpe and Hanssens 1999). Unfortunately, little is currently known on the impact of retailer and competitor actions on the long-run effectiveness of marketing actions.
A particularly important marketing activity for fast moving consumer goods are sales promotions, which represent the majority of manufacturers’ marketing budgets, amounting to 16% of their revenues (Canondale Associates 2001). However, manufacturers question the effectiveness of this practice as (1) the retailer may fail to pass-through and support price cuts over time (Armstrong 1991; Chevalier and Curhan 1976), (2) competitors may retaliate with their own promotions (e.g. Leeflang and Wittink 1996), and (3) consumers may ‘lie-in-wait’ for promotions (Mela et al. 1998). Moreover, the move towards category management (Progressive Grocer 2001) implies that retailers may respond to wholesale price promotions with changes in prices (and feature and display activity1) for competing brands (Zenor 1994; Hall, Kopalle and Krishna 2002); a practice referred to as ‘cross-brand pass-through’ (Besanko et al. 2005; Moorthy 2005). While recent research has estimated the immediate own and cross-brand pass-through of manufacturer promotions (Besanko et al. 2005), it did not account for dynamic effects and feedback loops among the retailer and the competing manufacturers, nor calculate their impact on the long-term promotional effectiveness.
In this paper, we conduct restricted impulse response functions, based on VAR models, for 75 brands in 25 product categories to answer three related research questions. First, to what extent are manufacturer promotions passed through by the retailer, and induce reaction by competing manufacturers? Second, how do these retailer and competitor reactions drive the long-term sales response to the initiating wholesale price promotion? Third, do these reactions and sales effects depend on category characteristics and the brand’s competitive position? In particular, with respect to long-term promotional sales effects, are smaller brands in jeopardy compared to leading brands in the category?
2. Long-term promotional effects and the role of retailer and competitor response
Consider a manufacturer reducing the wholesale price for its brand (hereafter focal brand) to the retailer2; a manufacturer promotion that can last several weeks (Armstrong, 1991). In response, the retailer may (1) adjust the consumer price3 of the focal brand (own brand pass-through), and (2) adjust the consumer price for competing brands in the category (cross-brand pass-through). Moreover, competing brand manufacturers may react by offering their own promotions to the retailer. We discuss these reactions in turn.
2.1 Long-Term Retailer Pass-through for the promoting brand
Acknowledging that the manufacturer promotion increases the retailer’s margin on the focal brand, the retailer may change the focal brand’s price in at least four documented ways:
1) the retailer may cut the price to the consumer by the same amount; i.e. pass-through 100% of the manufacturer promotion to the consumer.
2) the retailer may cut the price to the consumer by a higher amount; i.e. pass-through more than 100% of the manufacturer promotion
3) the retailer may cut the price to the consumer by a lower amount; i.e. pass-through less than 100% of the manufacturer promotion
4) the retailer may not cut the price to the consumer at all, i.e. ‘pocket’ the promotion.
Based on previous literature, the choice between these actions likely depends on (1) the retailer’s estimate of consumer price sensitivity, (2) the main retailer goal (profit versus traffic/market share), and (3) the retailer’s consideration for the manufacturer’s requests and goals. First, the retailer has little incentive to pass-through when consumer sales are not very sensitive to price reductions. Moreover, profit-maximizing pass-through rates depend on the specification of the demand function: concave demand functions (including linear and homogeneous logit) yield pass-through rates of less than 100%, whereas multiplicative (constant elasticity) demand functions yield pass-through rates of over 100% (Tyagi, 1999).
However, retailer goals may differ from mere short-run profit maximization. Indeed, increasing sales volume is often cited as a retailer objective, whether in the form of increasing store traffic or increasing market share vis-à-vis other retailers (Ailawadi 2001). This volume objective is more likely to trump the profit objective in large categories and for leading national brands (Bronnenberg and Mahajan 2001). A focus on traffic or market share may lead retailers to pass-through more than 100% of the deal, even though this is not optimizing short-run profits.
Finally, pass-through may be affected by the ability of brand manufacturers to ensure retailer compliance with the conditions for accepting the manufacturer promotion. This ability has traditionally been poor, but may be improving: the 1990s witnessed the move from off-invoice allowances to scan-back deals, which limit the retailer’s freedom to ‘pocket’ the deals. With an off-invoice allowance, the retailer gets rewarded with a price reduction for units she purchases in a given deal period, while with a scan-back deal, the retailer gets rewarded only for as much product as she can prove was sold to consumers in the given deal period (Drèze and Bell 2003)
Current empirical evidence demonstrates that (1) short-run pass-through rates are typically lower than 100% (Besanko et al. 2005), and (2) short-run pass-through rates are significantly higher for high-share brands and large categories, both of which are believed to draw more traffic and thus increase retail store revenues (Chevalier and Curhan 1976).
2.2 Long-Term Retailer cross-brand pass-through
The move towards category management (Progressive Grocer 2001) implies that retailers may respond to wholesale price promotions with price changes for other brands in the category (Zenor 1994). Analytical models provide the rationale for such cross-brand pass-through.
Negative cross-brand pass-through is motivated by retailer category profit maximization and Hotelling-like demand models (Moorthy 2001) or a combination of logit demand and manufacturer Stackelberg interaction (Sudhir 2001). Intuitively, the promoted brand attracts brand switchers, which leaves competing brands with only hardcore loyal consumers. Therefore, the retailer creates volume with the promoted brand, while increasing margin on the non-promoted brands to “mix-back” to the desired category profit levels (Grier 2001). In the general formulation by Moorthy (2005), the retailer will increase prices on other brands when demand-substitution effects dominate. Such action would accelerate desired substitution towards the promoted brand. What would prevent retailers from engaging in such negative cross-brand pass-through? First, raising prices on (category) traffic drawing brands may adversely affect overall retailer performance. Second, discontent by manufacturers (or consumers) of large brands may translate into a credible threat to the retailer. In contrast, little prevents the retailer from raising prices on the smaller brands in the category.
Positive cross-brand pass-through is motivated by strategic complementarity among brands in a category (Moorthy 2005). In this case, the marginal profit from each brand to the retailer is an increasing function of the other brands’ prices. Retail competition adds “external” strategic complementarity: the marginal brand profit to one retailer increases if another retailer increases prices. An alternative explanation for positive cross-brand pass-through is simple retailer brand profit maximization4 (Sudhir, 2001). Intuitively, a wholesale price promotion reduces overall retailers’ costs, which they may then use to reduce prices on competing brands for a sales lift (Hall et al. 2002). The study by Besanko et al. (2005) reports a frequent occurrence of both positive and negative cross-brand pass through in the short run. Which brands should particularly benefit from such positive cross-brand pass-through? As argued before, retailers are more likely to promote larger brands, which are believed to generate substantial category expansion.
2.3 Competing brand manufacturer reactions
In principle, the price response of competing brand manufacturers may be aggressive (reducing wholesale price), accommodating (increasing wholesale price) or passive (Chen 1996). Their choice among these options likely depends on (1) how their own sales were affected by the retailer’s pass-through and (2) whether they perceive that their reaction will be beneficial.
First, the jury is still out as to the extent of brand switching versus category expansion effects of price promotions. On the one hand, a high degree of brand switching implies that competing brands suffer substantially (e.g. Gupta 1988). On the other hand, a high degree of category expansion implies that competing brand sales may hardly be affected, or even increase (Pauwels et al. 2002; Sun et al. 2003; van Heerde et al. 2003). Logically, competing brand manufacturers would contemplate an aggressive response if their sales substantially decreased, which is more likely in concentrated categories (Chen 1996).
However, such aggressive response may not be beneficial if (1) the retailer is unlikely to pass-through (most of) the promotion or (2) the passed-through promotion will lead competitors to retaliate in turn, and the resulting ‘price skirmish’ is undesirable. Both conditions appear more likely for a small brand manufacturer, who is contemplating how to respond to a larger brand manufacturer’s promotion (Emerson, 1972). An additional reason against aggressive response is low consumer price sensitivity, which often applies to manufacturers of high-end niche brands. In such case, accommodating response is often optimal, especially if the brand is driven out of price-sensitive segments (Hauser and Shugan 1983; Pauwels and Srinivasan 2004).
Finally, despite extensive study of competitor response, its impact on the initiating brand sales has received little empirical analysis. On the one hand, several authors envision substantial damage, and argue that the net effectiveness of a marketing action largely depends on competitive response (Bass and Pilon 1980; Chen 1996). On the other hand, competing brands may perceive minimal damage from each other’s marketing actions (Chen and MacMillan 1992; Steenkamp et al. 2005), depending on the relative importance of brand switching versus category expansion from price promotions (Neslin 2002).
2.4 Are small-share brands in jeopardy regarding long-term promotional sales effectiveness?
Many of the above arguments imply that small-share brands end up with a smaller sales impact of their own promotions, and experience more harm from competing brand promotions.
First, both theoretical predictions (e.g. Lal et al. 1996) and empirical evidence (Chevalier and Curhan 1976; Walters 1989) support that promotions by smaller brands are less likely to be passed-through and supported. Retailers appear more willing to pass-through and support promotions of leading brands, as these are believed to generate substantial category expansion (Bronnenberg and Mahajan 2001) and may draw business away from competing retailers that do not offer consumers such promotion and/or guard against loosing business to those who do (Moorthy 2005).
Second, wholesale promotions by smaller brands are more likely to yield positive cross-brand pass-through for larger brands, while the reverse is not the case (Moorthy 2005). Interestingly, both phenomena would constitute a form of retailer-driven jeopardy for smaller brands, in addition to the consumer-driven jeopardy observed by Fader and Schmittlein (1993).
3. Research methodology
To investigate our research questions, we apply an atheoretical (reduced-form) econometric model that captures the dynamic reactions of consumers (sales), retailers and manufacturers (competitors). We opt not to consider a theoretical model for two reasons. First, marketing theory is often unclear as to the exact timing and direction of dynamic effects5, even when it is very informative about the direction and magnitude of immediate effects (Dekimpe and Hanssens 1999). Second, theoretical models typically require assumptions on the form of consumer demand or the managers’ pricing behavior (Besanko et al. 2005). These assumptions may then drive estimated own-brand and cross-brand pass-through rates (Moorthy 2005); including a predisposition to find negative cross-brand pass-through (e.g. the nested logit model) or imply that if one brand generates positive cross-brand effects, the other generates negative effects (e.g. the linear demand model). Instead, we prefer to discover reaction patterns with a reduced form approach, and use past theoretical literature to interpret the empirical findings.
Evidently, this choice comes at a cost: as a reduced-form model merely identifies and summarizes historic data patterns (Franses 2005), it can not disentangle demand versus supply drivers of managerial decisions (e.g. Besanko et al. 2005), and its predictions may not hold up when such drivers substantially change. Therefore, we relate estimated response to supply and demand factors in a second stage; and perform a split-sample analysis to investigate the move from off-invoice allowances to scan-back deals.
The particular reduced-form model we estimate is a Vector AutoRegression (VAR) model, which regresses the vector of all endogenous variables on the lagged vectors of these variables (hence the name Vector Autoregression) and the exogenous control variables. Because of this formulation, the VAR-model captures the dynamic interactions among the endogenous variables of interest. VAR-models have been used to analyze a wide variety of long-term marketing effects (Dekimpe and Hanssens 1999; Pauwels et al. 2002, 2004; Srinivasan et al. 2004).
3.1 Model specification
Specifically, our VAR-model for each category includes as endogenous variables: log of sales, wholesale prices, retail prices, feature and display for the top three brands (hereafter brands A, B and C) in the category. The lag order of the VAR-model is 1, which guards against curve fitting and is also selected in all cases by the Schwarz Bayesian Information Criterion (Lütkepohl 1993). As exogenous variables, we control for (i) a deterministic-trend variable (t) to capture the impact of omitted, gradually changing factors, (ii) seasonal dummy (0/1) variables that capture the shopping periods around major holidays (Pauwels and Srinivasan 2004), and (iii) new product introductions in the category. Importantly, wholesale promotions for any brand may affect retail actions for any considered brand in the category; which is necessary to detect cross-brand pass-through (Moorthy 2005). In addition, the model allows each endogenous variable (sales, retailer and manufacturer actions) to be influenced by the past of all endogenous variables. Therefore, we account for a rich interplay of dynamic effects, including:
past marketing actions may affect current sales because of consumer stockpiling and reference prices (a typical negative impact of past price on current sales);
performance feedback and decision rules may imply that marketing actions get repeated, or alternate over time (e.g. if the retailer puts a brand on display, she typically puts it again/does not put the same brand on display the next week)
competitive reaction induces current changes as a result of past marketing actions.
The standard VAR-model does not specify the contemporaneous effects, i.e. which variables impact others in the same week, which are instead estimated through the residual covariance matrix using the generalized impulse approach (Pesaran and Shin 1998). Model fit is verified by the Schwarz’s Information Criterion (balancing log likelihood with model parsimony), and by diagnostic tests for residual correlation (Durbin Watson test and LM tests), residual normality (Jarque-Bera test) and heteroskedasticity (White test).
3.2 Restricted impulse response functions
As it is infeasible to interpret the estimated VAR-coefficients directly (Sims 1980), researchers use the estimated coefficients to calculate the unrestricted impulse response function. This “conceptual experiment” simulates the over-time impact of a change (over its baseline) to one variable on the full dynamic system and thus represents the net result of all modeled actions and reactions (e.g. Pesaran and Shin 1998). Recently, Pauwels (2004) developed conceptual experiments that only allow some variables to react, restricting the other variables to remain at their baseline level, as predicted by the VAR-model. We adapt this methodology to our setting; starting from a brand’s wholesale price promotion.
First, we calculate the long-term response of wholesale price itself to its own one-unit (i.e. $1) reduction. This quantity represents the ‘effective manufacturer promotion’, as it indicates how long the typical manufacturer promotion for this brand lasts. As (immediate) pass-through is defined as the extent to which a change in wholesale price is passed through by the retailer in shelf price (Besanko et al. 2005), our long-term equivalent requires we divide the estimated long-term retailer (and competitor) response by this ‘effective wholesale promotion’ to calculate long-term retailer pass-through (and competitive reaction). In order to obtain the long-term retailer and competitor response, we estimate separate impulse response functions by restricting different variables to remain unaffected6 by the manufacturer promotion, as detailed next.
The first conceptual experiment (E1) allows long-term changes to the wholesale price and retail price of the focal brand and to sales of all brands. This represents the base case scenario, isolating long-term retailer pass-through and its long-term sales response. The second conceptual experiment (E2) adds retailer promotion support by also allowing long-term changes to feature and display of the initiating brand. The third conceptual experiment (E3) adds long-term changes to the retail prices, feature and display of competing brands, which represents the retail category management decisions. Finally, the fourth conceptual experiment (E4) also allows long-term changes to competitive wholesale prices. This scenario corresponds to the conventional unrestricted impulse response function, as all variables in the dynamic system are allowed to respond. Calculation of the standard errors for each conceptual experiment allows a formal comparison of the impulse response functions, as they are all based on the same estimated coefficients from the same VAR-model.
3.3 Second-stage weighted least squares analysis
Our second-stage analysis relates the estimated long-term responses to brand and category characteristics. This second stage employs weighted least-squares regression, using as weights the inverse of the standard errors of the first-stage response estimates, which serve as the dependent variables. The independent variables are brand market share, category size, category concentration, brand ownership (national brand versus store brand), brand expensiveness, brand wholesale price volatility, category expensiveness, category wholesale price volatility, product storability and impulse buy (Narashimhan et al. 1996; Srinivasan et al. 2004).
4. Data description
Our time series are based on scanner data from a large mid-western supermarket chain, Dominick's Finer Foods. With 96 stores in and around Chicago, this is one of the two largest in the area. These data are publicly available at the University of Chicago website7. In order to allow comparison with recent research (Srinivasan et al. 2004), we study the same 25 fast moving consumer products: analgesics, bathroom tissue, beer, canned soup, canned tuna, cereal (cold, hot), cheese, cookies, crackers, dish detergent, fabric softeners, front-end candies, frozen dinners, fabric softener, juice (bottled, frozen, refrigerated), laundry detergent, paper towels, shampoos, snack crackers, soaps, soft drinks, toothbrushes and toothpastes.
The relevant variables include unit sales at the SKU level, retail prices, and feature (‘price special’) and display (‘bonus buy’) activity8. Additionally, retail margin data allow us to calculate the average acquisition cost of each SKU to the retailer, which is a useful measure of wholesale price (Chintagunta 2002; Srinivasan et al. 2004). Sales are aggregated from the SKU to the brand level using the standard practice (e.g. Pauwels et al. 2002) in adopting constant weights, rather than varying (current-period) weights to compute the weighted prices9. All price data are appropriately deflated using the Consumer Price Index. The data period runs from September 1989 till May 1997. This extended time period also enables us to compare the early-90s period in which manufacturer promotions mostly took the form of off-invoice allowances, and the late-90s period during which scan-back deals became more prominent (Ailawadi 2001; Drèze and Bell 2003). The former period reflects Armstrong’s (1991) situation in which manufacturers offer rather general performance requirements with little enforcement mechanisms. Based on previous research (Srinivasan et al. 2004), we use September 1994 as the cut-off point for the split-sample analysis. Given our interest in retailer chain-level response to changes in wholesale prices, we aggregate the data across stores10.
We focus on the three top-selling brands in each category, capturing on average 87% of the total category volume. For ease of exposition, we display results across 25 categories by brand market share: leading brands (hereafter brands A), have the highest market share in their category, on average 45%. Smaller brands B and C have considerably less market share, on average respectively 24% and 15%. Table 1 provides more details on these brands, which do not differ much on price characteristics: both price level and wholesale price volatility are similar for the three groups (e.g. the higher price for some C-brands denotes their niche status). As for ownership, 3 of the B-brands and 6 of the C-brands are store brands (private labels). The additional measures for our second-stage analysis are shown in appendix.