Contact:
Email: m.hanzroh@queensu.ca
Department of Economics
Queen's University
94 University Avenue
Kingston, ON, K7L 3N6
Job Market Paper
Abstract: This paper develops and estimates a consumer search model for experience goods - products that are challenging to evaluate without direct use - available across multiple retailers. Beyond price uncertainty, consumers also face uncertainty about product suitability (match values) and form expectations about the quality of the match-value information available at different retailers, available for instance through consumer reviews. Consumers direct search across retailers based on their expectations about prices, retailer preferences, and expectations about match information quality. Consumers gain more precise match signals at retailers with higher quality match information, which makes finding a well-matched product more efficient. Analyzing clickstream data on camera searches, I document search behaviors that cannot be explained by models lacking expectations about information quality. Structural estimation indicates that larger retailers, such as Amazon and Walmart, provide higher-quality information, which I quantify enhances consumer welfare by 8.35%. Additionally, I show that retailers with superior information quality have an increased capacity to steer consumers and extract rents.
Presented at CEA 2024
Abstract: Many demand models rely on the characteristics-space approach to represent products and estimate consumer preferences. A practical limitation with the approach in some markets is that if demand-relevant characteristics are not observed, the substitution patterns the model predicts are unreliable. To address this limitation, this paper proposes a method of learning substitution patterns directly from search data. The approach is to treat the sets of products that consumers search for as their revealed consideration set, and measure product substitution between a pair of product by their frequency of co-searches across all consumers’ search sets. This substitution measure can then be mapped to vectors of latent characteristics representing each product. I validate the latent characteristics by using them as an input to a simple predictive demand model applied to data on online shopping at a large UK eCommerce platform. The aim is to predict which product a consumer will purchase given the set of previously searched products, as in a recommender system. I find that representing products with latent characteristics leads to improvement in prediction performance. These findings are supported by replicating the empirical analysis within a Monte Carlo simulation of consumer search.
Presented at CEA 2023