Marketing Science

Kalyanam, Kirthi

doi: 10.1287/mksc.15.3.207pmid: N/A

The advent of optical scanning devices and decreases in the cost of computing power have made it possible to assemble databases with sales and marketing mix information in an accurate and timely manner. These databases enable the estimation of demand functions and pricing/promotion decisions in real time. Commercial suppliers of marketing research like A. C. Nielsen and IRI are embedding estimated demand functions in promotion planning and pricing tools for brand managers and retailers.This explosion in the estimation and use of demand functions makes it timely and appropriate to re-examine several fundamental issues. In particular, demand functions are latent theoretical constructs whose exact parametric form is unknown. Estimates of price elasticities, profit maximizing prices, inter-brand competition and other policy implications are conditional on the parametric form assumed in estimation. In practice, many forms may be found that are not only theoretically plausible but also consistent with the data. The different forms could suggest different profit maximizing prices leaving it unclear as to what is the appropriate pricing action. Specification tests may lack the power to resolve this uncertainty, particularly for non-nested comparisons. Also, the structure of these tests does not permit seamless integration of estimation, specification analysis and optimal pricing into a unified framework.As an alternative to the existing approaches, I propose a Bayesian mixture model (BMM) that draws on Bayesian estimation, inference, and decision theory, thereby providing a unified framework. The BMM approach consists of input, estimation, diagnostic and optimal pricing modules. In the input module, alternate parametric models of demand are specified along with priors. Utility structures representing the decision maker's attitude towards risk can be explicitly specified. In the estimation module, the inputs are combined with data to compute parameter estimates and posterior probabilities for the models. The diagnostic module involves testing the statistical assumptions underlying the models. In the optimal pricing module the estimates and posterior probabilities are combined with the utility structure to arrive at optimal pricing decisions.Formalizing demand uncertainty in this manner has many important payoffs. While the classical approaches emphasize choosing a demand specification, the BMM approach emphasizes constructing an objective function that represents a mixture of the specifications. Hence, pricing decisions can be arrived at even when there is no consensus among the different parametric specifications. The pricing decisions will reflect parametric demand uncertainty, and hence be more robust than those based on a single demand model.The BMM approach was empirically evaluated using store level scanner data. The decision context was the determination of equilibrium wholesale prices in a noncooperative game between several leading national brands. Retail demand was parametrized as semilog and doublelog with diffuse priors for the models and the parameters. Wholesale demand functions were derived by incorporating the retailers' pricing behavior in the retail demand function. Utility functions reflecting risk averse and risk neutral decision makers were specified. The diagnostic module confirms that face validity measures, residual analysis, classical tests or holdout predictions were unable to resolve the uncertainty about the parametric form and by implication the uncertainty with regard to pricing decisions. In contrast, the posterior probabilities were more conclusive and favored the specification that predicted better in a holdout analysis. However, across the brands, they lacked a systematic pattern of updating towards any one specification. Also, none of the priors updated to zero or one, and there was considerable residual uncertainty about the parametric specification.Despite the residual uncertainty, the BMM approach was able to determine the equilibrium wholesale prices. As expected, specifications influence the BMM pricing solutions in accordance with their posterior probabilities which act as weights. In addition, differing attitudes towards risk lead to considerable divergence in the pricing actions of the risk averse and the risk neutral decision maker. Finally, results from a Monte Carlo experiment suggest that the BMM approach performs well in terms of recovering potential improvements in profits.

Bowman, Douglas; Gatignon, Hubert

doi: 10.1287/mksc.15.3.222pmid: N/A

Order of entry has been demonstrated to have a significant effect on market share. A number of explanations for this effect have been suggested in the marketing and strategy literatures. To date, the market share advantage gained by pioneers has typically been treated as a main effectan automatic regularity. Treating order-of-entry as a main effect implies that there is no penalty on the effectiveness of a brand's marketing instruments for late entry and that a late entrant can compensate for being late by dedicating sufficient marketing resources to their product.In this study, we investigate the influence of order-of-entry into a market on the effectiveness of a firm's marketing mix decisions by asking the question, Can followers compensate for not being first by their marketing mix decisions? Also, even if they can compensate for being late, does this effort become increasingly more difficult with later entry? That is, are there asymmetries in the effectiveness of a brand's marketing mix variables that relate to its order of entry into the market, or as has been typically assumed to date, is order of entry strictly a main effect? An asymmetry exists, for example, if the market response to advertising is different for the first entrant versus the second or third entrant. An asymmetry also exists if the effects of, say, a price change by the first entrant on the second entrant are different than the effects on the third entrant. We develop a market share attraction model where the parameters vary as a function of order-of-entry. Our main contribution is in modeling the sources of order-of-entry advantage as asymmetries in the effectiveness of a brand's marketing instruments. Hence, distinct from previous research we explain why there are inherent order-of-entry effects. This paper is potentially of interest to researchers developing market share models and studying the effectiveness of marketing-mix variables. The substantive implication of our results concern directly academics interested in marketing strategy as well as the practicing marketing strategists.We model asymmetries in the market response of early entrants versus late entrants using data from two durables and three nondurables categories. With one exception, all data sets are established from the inception of the category and hence do not suffer from the possible bias of excluding pioneers who have failed. Results show that asymmetries in the effectiveness of a brand's marketing mix variables are an essential source of order-of-entry effects; we find that the main effects of order of entry are minimal. Order-of-entry effects do not necessarily lead to lower shares, but overcoming these effects is not without substantial cost to the late entrant.Our results support previous research that has demonstrated advantages to early entry. In addition, we provide guidelines for how late entrants should compete. Later entry tends to reduce a competitor's price sensitivity, suggesting that they not instigate in a price war with earlier entrants in order to gain share. Order-of-entry tends to decrease response to quality and to promotion. To achieve the same impact on market share, later entrants need a bigger change in quality and need to spend more on promotion. Our data did not support an asymmetric effect on advertising.

Papatla, Purushottam

doi: 10.1287/mksc.15.3.243pmid: N/A

The issue of consumer heterogeneity in discrete choice analysis has been attracting much attention recently. Research has suggested that heterogeneity can result in biased parameter estimates which, in turn, can lead to incorrect conclusions. Among the many methods proposed in the literature to handle heterogeneity, fixed effects models seem to be the most attractive from a substantive point of view. However, in order to provide consistent estimates, these models typically require long purchase histories. This difficult constraint has prevented their widespread use.In this paper, we propose a new model which offers the benefits of fixed effects models without requiring long purchase histories. Our approach differs from the classic formulation in two ways. First, we calibrate the common and fixed-effects sequentially rather than simultaneously. This two-stage estimation permits us to obtain unbiased estimates of the common parameters even when the sample has households with very few observations. Second, we incorporate the fixed effects in a multiplicative rather than in an additive form.Our assumptions regarding the error term result in a probit model with heteroskedastic and temporally correlated random utilities. We develop a method of moments based estimator to calibrate parameters of models with this type of error structure. This procedure exploits the property that the method of moments yields estimates which are asymptotically as efficient as the maximum likelihood estimates given an appropriate starting point and a matrix of instruments. This is a very general procedure in that its mechanics are not tied to the proposed model. It can be readily applied to other problems where probit models are used to analyze data with serial correlation. This is a methodological contribution of the paper.The model and estimation procedure are illustrated on the A. C. Nielsen scanner data base for the liquid detergent category. Our results indicate that the proposed model can provide a better fit than random effects and loyalty based models and better predictive performance than random effects models. From a substantive perspective, our analysis provides the following findings.(a) the distribution of price sensitivities has a low variancethus, most households have very similar price sensitivity in this category;(b) the promotion response distribution has a high variancesuggesting that there are some households that are highly responsive to promotions while there are quite a few households that are less influenced by promotions;(c) preferences and price sensitivity have a negative covariancewhich is an indication that a reduction in preference for a brand is associated with an increase in price sensitivity and vice versa;(d) preferences and promotion response have a positive covariancethus, as the preference for a brand increases, response to promotions by the brand also increases; and(e) price sensitivity and promotion response have a negative covarianceimplying that, as households become more price sensitive, purchases may be driven more by price than by promotions; in other words, rather than responding to promotions such as displays and features as signals of a reduced price, price sensitive households actively evaluate prices in arriving at a choice.The proposed approach can be used by theoretical as well as applied researchers of brand choice behavior. Specifically, estimates of fixed-effects provided by the model can be used to cluster households into groups with similar parameters. Profiling the resulting groups in terms of demographics would provide interesting insightsfrom a theoretical perspectiveregarding the household characteristics that are associated with different types of response behavior. It would also help managers to better target their brands.Using the proposed approach for cross-category analyses of fixed-effects would be an interesting area of future research. For instance, given a sample of households and their purchase histories in multiple categories, the proposed model can be used to explore the relationship between household behavior and category characteristics. In particular, issues such as the following can be investigated by comparing household-level estimates across categories:(a) are households that are highly price sensitive in one category also price sensitive in other categories?(b) what household characteristics are associated with high (low) price sensitivity or high (low) promotion response across categories? and(c) in what types of categories are households likely to exhibit high (low) price sensitivity or high (low) promotion response?

Gnl, Fsun; Srinivasan, Kannan

doi: 10.1287/mksc.15.3.262pmid: N/A

We examine the basic premise that consumers may anticipate future promotions and adjust their purchase behavior accordingly. We develop a structural model of households who make purchase decisions to minimize their expenditure over a finite period. The model allows for future expectations of promotions to enter the purchase decision. Households with adequate inventory of the product may face a trade-off of buying in the current period with a coupon or defer the purchase until next period, given their expectations of future promotions. Thus, we provide a framework for examining the impact of consumer expectations on choice behavior.The target audiences for our paper are (a) empirical researchers who intend to make structural models part of their applied research agenda; and (b) managers who value and seek to understand the impact of consumers' coupon expectations on current purchase behavior. Our research objective is to provide an empirical framework to examine whether and to what extent consumers anticipate future coupon promotions and adjust purchase behavior. The central premise of our approach is that a rational consumer minimizes the present discounted value of the cost of a purchase where cost in a single period consists of purchase price, inventory holding cost, gains from coupons, and potential stockout cost. We aim to test whether our hypotheses regarding the various elements of the cost structure are supported and that whether consumers take into account future discounted cost when making current purchase decisions.The research methodology we adopt is relatively new in econometrics and known as the estimable stochastic structural dynamic programming method. The methodology amounts to incorporating a maximum likelihood routine embedded in a dynamic programming problem. The dynamic programming problem is solved several times within a maximum likelihood iteration for each value of the state space elements and for each value of the parameters in the parameter set. The state space in our model consists of purchase and nonpurchase alternatives in each time period, coupon availability and no coupon availability in each time period, level of inventory in each time period for each household, and consumption rate of each household.We use scanner panel data on purchases in the disposable diaper product category and promotions. We estimate the inventory holding and stockout costs, brand-specific value of coupons, and consumers' expectations of future coupons. The key insights and lessons learned can be summarized as follows: (1) Our results are consistent with the notion that consumers hold beliefs about future coupons, and that such beliefs affect the purchase decision. We find that the dynamic optimization model performs significantly better than a single-period optimization model and a naive benchmark model. (2) We find a high and significant stockout cost, consistent with the essential nature of the product category. Our estimate of the holding cost yields a reasonable annualized percentage value when converted to the cost of capital. We find that consumer valuation of coupons differ markedly across brands. (3) Our empirical evidence supports the notion that consumers hold beliefs about future coupon availability. We also find that the expectations about future coupons, estimated endogenously, differ depending upon whether or not a coupon was available in the current period. Thus, the proposed model structure yields rich managerial insights and facilitates several what if scenarios.A possible limitation of our model, and estimable structural models in general, is the computational cost. While it is possible to conceptually extend the state space to accommodate variations across households and add a richer parameter structure, each addition multiplies the size of the state space and the computation time. For this reason, we have kept the state space as tight as possible and refrained from additions that would otherwise enable us to incorporate heterogeneity in consumer decisions. For example, we assumed that consumers are similar other than reflected by their purchase behavior. We built a category purchase incidence model rather than a brand choice model. We refrained from including unobserved heterogeneity in the parameters. We chose to opt out of modeling autocorrelation and other time-dependent error term patterns in the likelihood function. Thus, we have made an effort to build a structural model that reasonably reflects consumer purchase behavior without requiring expensive computation. Currently, there are developments in econometrics to approximate the computation of the valuation functions without sacrificing much accuracy. When these methods are well developed we expect that structural models will become more commonplace in marketing.

Roy, Rishin; Chintagunta, Pradeep K.; Haldar, Sudeep

doi: 10.1287/mksc.15.3.280pmid: N/A

In this paper we develop a general class of dynamic brand choice models, called Lightning Bolt (LB) models, which are consistent with the theory of random utility maximization of consumer choice behavior. The underlying random utility process is Markov, and the inter-temporal evolution of the (utility-maximizing) brand choice process is also Markov. The models permit parsimonious parameterizations of the random utility process in brand choices with the resulting switching probabilities being functionally related to explanatory variables. The model allows for structural state dependence (feedback), habit persistence (inertia), and unobserved heterogeneity. The theoretical development shows that several well known stochastic brand choice models can be deduced from random utility maximization theory.From a managerial perspective, the usefulness of the proposed model stems from its ability to separate out the effects of habits, state dependence and heterogeneity. Strong state dependence effects imply incentives for inducing trial of a brand (e.g., product sampling). In contrast with state dependence, a strong habit persistence effect may be indicative of buyer behavior where inducement of trial of a different brand may not be sufficient to maintain a (sustained) defection of the consumer from the habitually purchased brand to the trial brand. Failure to distinguish between these two effects has important implications. For example, a model that only accounts for state dependence effects would, in the presence of only habit persistence, incorrectly attribute it to state dependence. Based on this the manager could decide to embark on an expensive sampling program that might prove ineffective due to the absence of state dependence. It is also important to distinguish between the effects of unobserved heterogeneity and state dependence. In the absence of true state dependence, failing to account for unobservable variations across households (such as differences in price sensitivities), results in the temporally persistent unobservable elements showing up as state dependence in the model. Hence, a manager may incorrectly opt for sampling as the appropriate marketing action, whereas, a couponing or price promotion strategy should have been preferred.We estimate the model parameters using the AC Nielsen household scanner panel data set on catsup purchases. Further, we investigate empirical techniques for overcoming the initial conditions problem that affects many dynamic models of brand choice. Through simulation analysis using the estimated parameters, we show that the calculated profitability of a promotion must take into account the multi-period impact due to state dependence. Further, we demonstrate how ignoring heterogeneity can result in spurious state dependence, thereby making marketing tactics such as product sampling appear far more attractive than they actually are. Specifically, the model that accounts only for habit persistence and state dependence effects predicts a share increase of 2.4 points through sampling for one of the brands in the empirical analysis. Once the effects of unobserved heterogeneity are accounted for, however, this number drops precipitously to 0.3. While even this low number might fulfill the objectives of the brand manager, it ensures that expectations are not eight times that number.It is important to note that this paper is a first attempt at analyzing the three fundamental dimensions of dynamic brand choice behavior. Our formulation represents a reduced form approach as opposed to a structural approach. Other limitations of the model include our focus on time-invariant choice sets. Extensions of the model to situations where choice sets vary over time are possible, but are difficult. Another limitation is the dependence on multivariate extreme value distributions. An alternative would be to use multivariate normal distributions, i.e., a probit-like structure which has certain useful properties. A comparison of the two approaches would be useful. Empirical extensions of the model to allow for higher-order processes and nested logit type model structures could also be fruitful. Explicitly incorporating variety-seeking behavior in the model would further enrich the theoretical framework proposed in this paper.