|
Article Excerpt
Long-term marketing effectiveness is a high-priority research topic for managers, and emerges from the complex interplay among dynamic reactions of several market players. This paper introduces restricted policy simulations to distinguish four dynamic forces: consumer response, competitor response, company inertia, and company support. A rich marketing dataset allows the analysis of price, display, feature, advertising, and product-line extensions.
The first finding is that consumer response differs significantly from the net effectiveness of product-line extensions, price, feature, and advertising. In particular, net sales effects are up to five times stronger and longer-lasting than consumer response. Second, this difference is not due to competitor response, but to company action. For tactical actions (price and feature), it takes the form of inertia, as promotions last for several weeks. For strategic actions (advertising and product-line extensions), support by other marketing instruments greatly enhances dynamic consumer response. This company action negates the postpromotion dip in consumer response, and enhances the long-term sales benefits of product-line extensions, feature, and advertising. Therefore, managers are urged to evaluate company decision rules for inertia and support when assessing long-term marketing effectiveness.
Key words: long-term marketing effectiveness; dynamic consumer and competitor response; company inertia and support; vector autoregressive (VAR) models; impulse-response functions; policy simulation restrictions; postpromotion dip
History: This paper was received June 5, 2003, and was with the author 3 months for 3 revisions; processed by Pradeep Chintagunta.
1. Introduction
Besides consumer demand, current marketing decisions often influence future company and competitor marketing activity. The 0% financing deals initiated by General Motors as an emergency measure after September 11, 2001 were quickly copied by competitors and continued a year later, even on the new 2003 models (Wall Street Journal, 2002). Likewise, escalation of advertising expenditures has been demonstrated in many industries (Metwally 1978). As a result, marketing managers are urged to consider the net long-term impact of their decisions, which includes dynamic consumer and competitor response, as well as associated future company actions (Chen 1996, Dekimpe and Hanssens 1999, Jedidi et al. 1999, Krishna et al. 2000).
Recent econometric studies compute this net long-run marketing impact by means of impulse-response functions derived from vector autoregressive (VAR) models (Bronnenberg et al. 2000; Dekimpe et al. 1999; Nijs et al. 2001; Pauwels et al. 2002; Srinivasan et al. 2000, 2004). In essence, an impulse-response function is the outcome of a "conceptual experiment" that tracks the full chain of events set in motion by a change to the marketing variable (Pesaran and Shin 1998). These events include consumer reactions such as promotion-induced stockpiling (Neslin 2002), competitor reactions such as retaliation (Leeflang and Wittink 1996), and company decision rules that back the initial marketing action by (1) prolonging it over time ("inertia"), such as keeping prices low for several weeks after a price promotion (Krishna et al. 2000, Srinivasan et al. 2004), or (2) supporting it with other actions, such as backing a product-line extension with advertising (Keller 1998). The overtime result of this chain of events is estimated as the net effect of the marketing action on sales. Figure 1 presents two typical examples of such impulse-response functions, showing the net sales elasticity of a price promotion and a product-line extension. While only the product-line extension effect shows "wear-in," i.e. it takes a number of weeks before the peak sales impact is reached, both impulse-response functions show "wear-out," i.e., it takes several weeks after the peak impact before sales effects die out.
Even though the net performance effect of a marketing action has strong managerial relevance, it remains unclear which part of it is due to dynamic consumer response, to dynamic competitor response, to company inertia, or to company support. As a result, it is often hard to understand sign and magnitude of the net dynamic impact, and to reconcile it with marketing theory and managerial intuition. Two examples serve as an illustration for tactical as well as strategic marketing actions. First, the absence of a significant postpromotion sales dip in several empirical studies (Blattberg et al. 1995) may be due to the fact that prices do not return to their regular levels for several weeks (Srinivasan et al. 2004). A plausible reason for such prolonged company action, as confirmed in recent experiments, is the managerial tendency to weigh past prices when setting future prices (Krishna et al. 2000). When confronted with evidence of price inertia, managers often want to find out to what extent it contributes to performance impact.
Second, advertising may fail to affect sales due to its inability to generate consumer response for established brands (Abraham and Lodish 1990), or due to competitive retaliation campaigns that cancel any demand gain (Bass and Pilon 1980). The distinction between these explanations is crucial, especially if managers obtain information that competitors will not respond as they have in the past. Indeed, in both examples, managers typically have little reason to believe that dynamic consumer response will deviate from historically observed patterns, but may have good reason to expect changes to company actions and/or competitor response. Moreover, several authors have demonstrated the relevance of combining the output of quantitative models with managerial judgment (Blattberg and Hoch 1990, van Bruggen et al. 1998). Finally, while recent research appropriately models marketing-sales endogeneity (e.g., Dekimpe and Hanssens 1999, Swait and Andrews 2003), Lodish (1980) and Shugan (2004) point out that marketing decision makers may face exogenous constraints and prior commitments that are unrelated to the context of the model. Therefore, it appears useful to combine the flexibility of vector autoregressive models that allow estimation of these dynamic forces, with policy simulations that allow their restriction and separation in conceptual forecasting experiments.
In this study, we introduce such restricted policy simulations to answer three related research questions. First, what is the dynamic response of consumers (demand) to marketing actions? Second, to what extent do dynamic competitor response, company inertia, and company support add to consumer response in order to form net marketing effectiveness? Finally, do these dynamic forces play a different role for tactical actions, such as (price) promotions, versus more strategic actions, such as advertising and product-line extensions?
The following section introduces our methodological framework. Section 3 introduces a rich marketing dataset, while [section]4 reports on the empirical findings. Finally, [section]5 provides conclusions, implications, and areas for future research.
2. Framework
All of the above dynamic forces are captured in the time-series modeling framework presented in Table 1. First, we examine the time-series properties for all sales and marketing variables (as extensively discussed in Dekimpe and Hanssens 1995). Based on these properties, we formulate models of the dynamic interactions among sales, marketing, and competitive marketing. Next, we use the estimated coefficients to simulate the net impact of a marketing action on sales, known as the impulse-response function. Finally, we extend the methodology by restricting endogenous variables to their steady-state baseline, which allows separation of consumer response, competitor response, company inertia, and company support. Calculation of the standard errors of these restricted policy simulations allows their formal comparison.
2.1. Vector Autoregressive (VAR) Model
In the absence of cointegration, which is rarely found among sales and marketing actions in consumer packaged goods (e.g., Nijs et al. 2001, Srinivasan et al. 2004), vector autoregressive (VAR) models are estimated with the stationary variables in levels and the evolving variables in differences. Equation (1) displays the basic of our model:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
with [[u.sub.S,t], [u.sub.FM,t], [u.sub.OM,t], [u.sub.CM,t]]' ~ N(0, [[SIGMA].sub.u],) and lag number K also known as the order of the model.
First, the vector of endogenous variables log of sales (S), log of focal marketing action (FM), log of own other marketing actions (OM), and log of competitive marketing actions (CM) is related to its own past, allowing complex dynamic interactions among these variables. Second, the vector of exogenous variables typically includes (i) an intercept [alpha], (ii) a deterministic-trend variable t, to capture the impact of omitted, gradually changing variables, and (iii) seasonal dummy variables SD (Nijs et al. 2001). Finally, Equation (1) explicitly displays the modeling of immediate (same-week) interactions among the endogenous variables. First, sales may be affected immediately by all company and competitive marketing actions through coefficients [[beta].sup.0.sub.12]-[[beta].sup.0.sub.14] . Second, the model assigns causal priority to the focal marketing action, which can immediately affect sales ([[beta].sup.0.sub.12]), own other marketing actions ([[beta].sup.0.sub.32]) and competitive marketing actions ([[beta].sup.0.sub.42]), but not vice versa. In other words, this causal ordering assumes that companies cannot change their focal marketing activity, observe immediate reaction, and adapt their actions again within the same time period, which makes sense for weekly data of retail-distributed consumer packaged goods (Leeflang and Wittink 1992, Dekimpe et al. 1999). Our explicit treatment of restrictions on the immediate reactions reflects a structural VAR approach to identification (Bernanke 1986), which is more appropriate (1) for subsequent policy restrictions than weak (implicit) identifying assumptions in the form of either the Choleski decomposition (e.g., Sims 1980, Dekimpe et al. 1999) or of the generalized impulse-response analysis (e.g., Pesaran and Shin 1998, Nijs et al. 2001). Moreover, we choose to...
|
