Structural Demand Estimation with Varying Product Availability

Demand estimation based on discrete choice theory has become an important tool in Marketing and empirical Industrial Organization. In particular, estimation techniques have been developed that allow the use of individual-based discrete choice methods in situations where only aggregate data are available (e.g. Berry et al. 1995). In essence, these methods allow making inferences on the distribution of individual preferences over products or attributes by integrating purchase probabilities over heterogeneous individuals and relating them to the aggregate market shares observed in the data. However, if not all products were available in every store and/or on every purchase occasion, the observed market share will be a convolution of two different –albeit related– factors: consumer choice and the probability of finding the products available in the store. Failing to account for the varying degree of availability would produce incorrect demand parameter estimates and would lead to wrong inferences about consumer preferences, competitive reactions, etc. This paper develops a model that extends the current methodology to account for varying levels of observed product availability, although the actual store/trip product assortments faced by the consumer are not observed by the researcher. The model parameters are estimated by simulating potential product assortment vectors by drawing multivariate Bernoulli vectors consistent with the observed aggregate level of availability. We show that our model is a generalization of the traditional random coefficients multinomial logit and similar estimation methodologies can be used. Furthermore, results from a Monte Carlo experiment show that utility parameters can be recovered correctly using the proposed estimation method, notwithstanding what causes product availability to vary. The model is applied to the UK chocolate confectionery market, focusing on the convenience store channel. We compare the parameter estimates to those obtained from not accounting for varying availability and analyze the pricing and competitive implications.