Final Report Summary - SEARCHSTRUC (The Effect of On-Line Product Search on the Market Structure of Consumer Durables)
The project focuses on the role of on-line consumer search and its ultimate influence on consumer product choices. It covers formal topics related to the econometric identification of preferences and search behavior in empirical settings. It also covers the types of data that would be amenable to estimating preferences and search behavior in practice. Finally, it covers practical empirical problems such as consideration sets and their formation, search costs, the role of promotions and other marketing influences to facilitate search and choice, and the implications of search behavior for firms’ marketing strategies.
The project contributes to the marketing and economics literature on consumer demand estimation that has addressed the fact that individuals may not consider or search every available product alternative when making purchase decisions. This fact complicates the application of the revealed preference approach to inferring demand from choice data. Several papers have estimated models that separate the formation of the considered set of products and the actual choice, conditional on the considered or searched set. A major weakness of this literature is that the identification of preferences and search set formation is often attempted with choice data alone. Another weakness is that the considered set is often postulated without being specific about its underlying search-theoretic primitives.
In the project, we use both descriptive as well as structural approaches to resolved these problems. We also collect extensive data that track product choices as well as aspects of the underlying search process leading up to the purchase itself. Such data enables the estimation of a structural discrete choice model of demand that separates consideration and choice.
The substantive interest in this project is twofold.
1. The effects of search on competitiveness and consumer outcomes.
We investigate the effects of consumer search, and of seller-provided on-line search-tools, on the structure and competitiveness of markets and on the welfare of consumers. We show that consumer search is driven by three groups of variables: (1) preferences, (2) uncertainty, and (3) search cost. We further present an empirically estimable model that incorporates how the consumer combines these three variables in a fashion that leads to optimal sequential search, i.e. the search strategy that maximizes expected utility minus search costs. Empirically, the model is applied to the online market for camcorders using data from Amazon.com and is used to answer questions about market structure and competition. We also address policy-maker issues about the effect of selectively lowered search costs on consumer surplus outcomes. We find that consumer search for camcorders at Amazon.com is typically limited to 10–15 choice options and that this affects estimates of own and cross elasticities. Typically, the estimated levels of price sensitivity are found to be considerably higher than those from a typical model of choice that assumes consumers search all products and observe all prices. This finding alone has important implications for pricing analysis conducted by marketers, industrial organization economists, and antitrust policy workers. In a policy simulation, we also find that the vast majority of the households benefit from Amazon.com’s product recommendations via lower search costs.
In a follow up project, we present a joint model of costly consumer search and choice decisions. In this model, consumers construct a limited choice set guided by incremental benefit and cost of an extra search and choose the utility maximizing alternative from this set. Our model explains why some products are more popular during the search stage than their expected utility merits, by explicitly modeling pre-search consumer uncertainty about choice alternatives. Applying our model to aggregate search and choice data from the camcorder category at Amazon.com we find that consumers search more for products with new attributes, but do not necessarily value these attributes highly on average. Ignoring the search-stage suggests revealed preference for these attributes, i.e. confounds willingness to pay and willingness to search. We demonstrate how manufacturers can use our model to make decisions regarding the adoption of a new feature in its product line management, accounting for consumer uncertainty about the new attribute.
The paper entitled “Online Demand under Limited Consumer Search,” BY Jun B. Kim, Paulo Albuquerque, and Bart J. Bronnenberg, which appeared in Marketing Science in 2011 was the recipient of the 2011 Frank M. Bass Award given by the Institute for Management Science for the best paper published in Marketing Science and Management Science based on a doctoral dissertation.
2. The use of search data to study demand
Second, we investigate the extent to which direct observation of consumer search outcomes informs us about demand. This is especially relevant in the case of studying markets for durable goods. In particular, a policy maker generally has difficulty studying demand in such markets from choices alone because the consumer typically makes a single choice. On the other hand, observing a consumer’s search behavior typically yields multiple observations on the same consumer, thus enabling to distinguish between the preferences of two consumers who make the same choice but for different reasons. This part of the study therefore focuses on estimating consumer heterogeneity in markets with durable goods, essential for assessing the competitiveness of markets, and price sensitivity.
We develop a new method to visualize browsing behavior in so-called product search maps. Manufacturers can use these maps to understand how consumers search for competing products before choice, including how information acquisition and product search are organized along brands, product attributes, and price-related search strategies. The product search maps also inform manufacturers about the competitive structure in the industry and the contents of consumer consideration sets. The advantages of the approach are twofold. First, the authors simultaneously visualize the positions of products and the direction of consumer search over products in a perceptual map of search proximity. Second, they explain the formation of the map using observed product attributes.
In another project, still ongoing, we parse computer records from the Comscore online consumer panel to create a dataset of individual search activity in a durable-product category, digital cameras, across time, products and online retailers. We observe alternatives searched, in which order, for how long, and which product was bought. We aditionally observe an extensive list of product characteristics, and retailer characteristics. We find that search outcomes (disregarding choices) are highly predictive of consumer choices. We find evidence that suggests that in the sequence of search, later stages of search are more predictive of choice than earlier stages. Importantly, we relate this to whether the consumer searches with constant preferences, or that consumer preferences are updated as the consumer search is in process. In other words, we seek to determine whether the consumer, in addition to learning about products, also learns about his or her preferences based on the search environment. The latter is important as it suggests that there are welfare implications to the design of search engines.