Combining Micro and Macro Data to Study Retailer Pricing in the Presence of State Dependence

Consumers of repeat-purchase goods have a higher probability of choosing products that they have purchased in the past. This form of persistence or state dependence has emerged in scanner panel data for many product categories. Considering the existence of state dependence by firms is important for the better understanding of consumer purchase behavior and pricing. If state dependence is a function of loyalty, then firms may want to engage in strategic pricing to control the evolution of preferences. However, manufacturers and retailers have limited access to scanner panel data. In addition, scanner panel data are often not suitable for use in pricing decisions because they provide price information only for those items that a consumer has purchased in a particular store and on a certain day. In this paper, we will show how firms can use readily available store-level scanner data in combination with tracking data, which firms routinely collect, to estimate the impact of state dependence on consumer purchase behavior and determine the resulting effect on the pricing decisions of firms. We model demand using a flexible, random coefficient logit model for aggregated data that takes into account the heterogeneity of brand perceptions and customer responses to pricing and promotions. This model also accounts for the possibility that competing brands exhibit flexible substitution patterns. The results indicate that consumers may change their purchase behavior if they have recently purchased a particular brand. We then use the demand side estimates on the supply side to show how retailer pricing and profitability are affected if a retailer does or does not anticipate state dependence in predicting consumer purchase behavior.