UCL School of Management is delighted to welcome Jae Hyen Chung, Chicago to host a seminar discussing “Inferring Automobile Market Structure from Consumer Search”.
The objective of this paper is to build a structural model to analyze market structure in a product category by leveraging online consumer search data and aggregate market share data corresponding to products in that category. Recent approaches to recovering the competitive structure underlying a market have typically required temporal variation in product attributes and market share (aggregate data) or individual level panel data on brand choices. Such methods would however, pose a challenge when studying product categories with long inter-purchase times (e.g., automobiles) for two reasons. First, with aggregate data, if information is available only over a short duration (e.g., a few months), then aggregate data may not required temporal variation in car attributes. Second, with individual panel data, the long inter-purchase times could result in changes in consumer preferences, incomes, etc. which could affect the validity of market structures derived from the panel information. By combining aggregate data (that fully reflect brand performance in the market) with individual search data collected close to the time of purchase (that capture the “consideration sets” of consumers), we demonstrate how researchers and marketers can better understand the competitive interactions among brands with readily available data from such categories. To characterize the online search behavior of consumers, we use the sequential search algorithm proposed by Weitzman (1979) to rationalize consumers’ observed search sequences. Further, predicted model choices are aggregated to provide the analog to observed market shares in the data. In estimating the model parameters, this paper makes two important methodological contributions. The first is a new simulated maximum likelihood method to estimate the parameters of the sequential search model with individual search sequence data and the market share data. The second, driven by the specific manner in which information is revealed during search in our empirical application, is that we allow consumers to search for a subset of product characteristics and idiosyncratic match values (rather than either a specific product characteristic such as price or the match value). Via simulation studies, we demonstrate that (i) the proposed method can recover the true parameter values and (ii) the market structure generating the data cannot be recovered if the search process is ignored. Next, we propose to estimate the structural model using search and market share data from the SUV market a submarket of the US automobile industry. Taken together, the paper makes a substantive contribution in terms of understanding market structure while providing useful methodological tools for researchers interested in estimating the parameters of sequential search models.