STRATEGIC MERCHANT COMPETITION AND LARGE ASSORTMENT MANAGEMENT IN ONLINE MARKETPLACES

Abstract

In this dissertation research, I study strategic merchant competitions on a retail deal platform and propose a new modeling approach for discovering consumer preference structures in online shopping environments with large and frequently changing assortments. In the first essay, I examine the impact of alternative platform policies and changes in market primitives on payoffs of players on a retail deal platform where merchants make frequent strategic decisions based on competitive considerations in a two-sided market. To this end, I construct a dynamic game model and conduct counterfactual simulations. My analyses reveal novel insights regarding the interplay between platform policies and merchants’ profits and relationships among competing merchants, and identify several mutually beneficial growth opportunities for all parties involved via policy changes. They also provide a comprehensive assessment of the net contribution of each merchant’s action to the platform eco-system, which sheds new light on how to identify and foster “star brands” in such marketplaces. In the second essay, I develop a scalable stock-keeping-unit-level modeling framework of discovering consumers’ preference structures in large and frequently changing assortments at the store/marketplace level. My proposed model identifies the underlying “topics of interest” that drive online browsing and purchase activities concerning the entire store assortment. It overcomes several limitations of standard Latent Dirichlet Allocation models for handling large assortments and enables demand forecast for products not in the existing assortments. I apply the proposed framework to investigate “topics” driving browsing and purchase activities in an online marketplace of fashion products, which reveals distinct topics driving these two types of shopping activities and their time-varying patterns of relevance. Finally, I propose a personalized product recommendation system that is based on individuals’ preference structures inferred from the model and powered by the Bayesian optimization algorithm. Hold-out comparisons show that our approach substantially outperforms benchmark recommendation algorithms commonly used in practice (such as user-, item-, and content-based collaborative filters), and it is particularly powerful in prescribing recommendations for products that do not exist in the previous assortment. This dissertation research offers a range of policy recommendations and valuable managerial insights to marketing researchers and E-commerce practitioners.