|
Article Excerpt
The recent marketing literature reflects a growing interest in structural models, stemming from (1) the desire to test a variety of behavioral theories with market data, and (2) recent developments that facilitate estimation of and inference for these models. Whether one should always go through the effort of developing such tightly parameterized models with the associated computational burden of estimating them and whether it pays off to make strict behavioral assumptions in terms of better decisions remain open questions. To shed some light on these issues, we provide examples of structural approaches to consumer choice and demand as well as examples where the goal is to study the nature of competition in the marketplace. From that review comes our discussion of issues in the development and application of structural models, including their estimation, testing, and validation, their applicability in the practice of marketing, and their usefulness for normative as well as descriptive purposes.
Key words: structural models; heterogeneity; competition; endogeneity; dynamic demand models.
History: This paper was received September 14, 2004, and was with the authors 3 months for 2 revisions; processed by Kannan Srinivasan.
1. Introduction
The recent marketing literature reflects a growing interest in structural models. By structural models we mean those that rely on economic and/or marketing theories of consumer or firm behavior to derive the econometric specification that can be taken to data. In particular, structural models are typically derived based on optimizing behavior of agents (e.g., utility maximizing by consumers, profit maximizing by firms, etc.). Furthermore, because one can assess the role of the behavioral assumptions in driving empirical findings, the appropriateness of these assumptions can be investigated. Consequently, the structural approach allows us to test the theories from which the models are derived, and obtain behavioral predictions that are invariant to the effects of policy changes. Hence, the increased interest in the development and application of structural models in marketing stems from (1) the desire to test a variety of behavioral theories with market data, and (2) recent developments that facilitate estimation of and inference for these models.
Although it is clear that there are situations where a structural approach may be preferred by marketing researchers, whether structural modeling will become the dominant paradigm in marketing, whether one should always go through the effort of developing such tightly parameterized models with the associated computational burden of estimating them, and whether it pays off to make strict behavioral assumptions in terms of better decisions remain open questions. Our objective in this paper is to shed some light on these issues. For this, we first provide examples of structural approaches to consumer choice and demand--both static and dynamic--as well as to the nature of competition in the marketplace. From that review comes our discussion of several important issues in the development and application of structural models, including their estimation, testing, and validation, their applicability in the practice of marketing, and their usefulness for normative as well as descriptive purposes. Interested readers are also referred to related reviews by Kadiyali et al. (2000), Dube et al. (2002), Chintagunta et al. (2004).
We aim to provide a critical assessment of our views of the strengths and weaknesses of structural modeling and its future in marketing. There are two factors that can facilitate more widespread application of structural models in marketing in particular. The first is the availability of stated and revealed preference data at unprecedented levels of disaggregation (right up to the individual consumer or household). Second, residing at the intersection of applied economics and psychology, marketing is in a unique position to develop new structural models which incorporate behavioral theory from both economics and psychology. Given the richness of data in marketing, more realistic and hopefully more powerful theories of consumer and firm behavior can be developed and tested.
2. Structural and Reduced-Form Models
The goals of econometric modeling are to explain past behavior of agents and to predict their future behavior However, the relative weights placed on data fitting (which is easier, ceteris paribus, the less structure is imposed on data) versus relying on theory in building an econometric model has been the topic of much debate. Whereas proponents of "reduced-form" (1) models prioritize fit to the empirical data and propose flexible functional forms to reflect variations in the data and let the data "speak," those of structural models prioritize theory as a guiding source for the empirical specification. Haavelmo (1944) and Lucas (1976) provide arguments for the use of structural econometric models in economics that have been derived from, or at least are consistent with, an underlying economic theory. They argue that structural models can be used to predict the effects of a shift to a policy regime that is different from that observed in the historical data.
Reduced-form models in general aim at representing consumers' or firms' historical decision rules as derived from available marketing data. The resulting estimates are used to predict the behavior of these agents in the future. Predictive validation of (reduced-form) models in hold-out data has received much attention in the marketing literature. By hold-out data we mean time-series observations from the agents that were used in calibrating the model. Assessing predictive validity on such hold-out data has become the standard for both comparing the predictive power of new versus old models as well as for assessing their validity and robustness. It is important to realize that a reduced-form model might have excellent predictive validity (particularly if assessed on holdout samples of consumers) and still provide misleading predictions of the effects of (marketing) strategy changes, if the impacts of such changes are too far from the historical data. For a model to be useful for predicting strategy changes, ideally there must be a similar change in the relevant strategy variables in the data.
Thus, reduced-form models run the risk of producing misleading forecasts of the effects of strategy changes that change the stochastic context in which decisions are made. This arises when a strategy change affects the agents' decision rules themselves, i.e., the parameters of the decision model or even its functional form. Historical marketing data may not display sufficient variability in strategy regimes to reliably estimate the effect of strategy changes on the parameters representing the agents' decision process. Structural models seek to parameterize the way in which the stochastic decision context created by policies affects the agents' preferences and decision rules (Rust 1994). Whereas in economics, policy changes mostly constitute changes in the economic environment (for example, privatization, deregulation) or a change in the nature of the strategic interaction of agents (for example, price setting or taking, or competitive versus collusive behavior), in marketing strategy changes of interest, next to the last mentioned strategic interactions, often constitute changes in the behavior of a particular agent (a change in promotional strategy of the firm), which have a more limited impact on the economic system. In many cases structural parameterizations are derived from the economic assumption of "optimality" of behavior of agents and the market equilibrium paradigm. This often involves assuming that given the information available to them, consumers and firms are rational expected-utility-maximizing entities, which allows optimal behavioral specifications to be derived that are applicable under a wider variety of decision contexts and economic environments. Recently, methods have been developed that facilitate implementation of such models. For example, dynamic programming approaches (e.g., Rust 1994) have enabled the representation of consumers' forward-looking decision making, and the empirical industrial organizational literature (e.g., Berry 1994, Reiss and Wolak 2002) has produced joint models of supply and demand in which estimable aggregate relations are derived from assumptions of the microbehavior of agents.
At the same time, the strong identifying--mostly parametric--assumptions prove to be the weakness of structural models as well. "We know for the onset in an enterprise like this that what will emerge--at best--is a workable approximation that is useful in answering a limited set of questions" (Lucas 1987). Reduced-form models are often simpler, require less and much weaker assumptions, and allow for more flexible semiparametric and nonparametric estimation methods. In contrast, many structural models are only parametrically identified (Rust 1994) and cannot be estimated using fully nonparametric methods. Economic theory often has little to say about the functional form, which is then chosen for reasons of analytical tractability. Therefore, reduced-form methods are particularly useful in exploratory analyses and specification testing, i.e., in formulating theory based on empirical data, and in empirically investigating the validity of structural model assumptions and predictions, and fit well in the empirical research tradition in marketing.
Another potential drawback of structural models is that if "too much" structure is imposed, then the models might be less useful for certain decision-making purposes. Consider, for example, a situation where we estimate demand parameters under the assumption that market data are the pricing outcomes of firms' profit-maximizing decisions with firms' interactions following a specific equilibrium concept (e.g., Nash-Bertrand pricing). In this case, we cannot use the demand parameters to come up with "improved" pricing decisions for a firm under that same equilibrium assumption (i.e., Nash-Bertrand), because those parameters have themselves been obtained by imposing "optimal" behavior of firms (see for a related discussion Shugan 2004). Nevertheless, a structural model whose parameters...
|
