Towards A Better Design of Online Marketplaces

Abstract

Online markets are staggering in volume and variety. These online marketplaces are transforming lifestyles, expanding the boundaries of conventional businesses, and reshaping labor force structures. To fully realize their potential, online marketplaces must be designed carefully. However, this is a significant challenge. This dissertation studies individual behavior and interactions in online marketplaces, and examines how to enhance efficiency and outcomes of these online marketplaces by providing actionable operational policy recommendations.

An important question in the context of open-ended innovative service marketplaces is how to manage information when specifying design problems to achieve better outcomes. Chapter 1 investigates this problem in the context of online crowdsourcing contests where innovation seekers source innovative products (designs) from a crowd of competing solvers (designers). We propose and empirically test a theoretical model featuring different types of information in the problem specification (conceptual objectives, execution guidelines), and the corresponding impact on design processes and submission qualities. We find that, to maximize the best solution quality in crowdsourced design problems, seekers should always provide more execution guidelines, and only a moderate number of conceptual objectives.

Building on the same research setting, Chapter 2 looks into another important yet challenging problem---how the innovation seeker should provide interim performance feedback to the solvers in online service marketplaces where seekers and solvers can interact dynamically. In particular, we study whether and when the seeker should provide such interim performance feedback. We empirically examine these research questions using a dataset from a crowdsourcing platform. We develop and estimate a dynamic structural model to understand contestants’ behavior, compare alternative feedback policies using counter-factual simulations, and find providing feedback throughout the contest may not be optimal. The late feedback policy, i.e., providing feedback only in the second half of the contest, leads to a better overall contest outcome.

Moving to a wider application, Chapter 3 leverages consumer clickstream information in e-commerce marketplaces to help market organizers improve demand estimation and pricing decisions. These decisions can be challenging, as e-commerce marketplaces offer an astonishing variety of product choices and face extremely diversified consumer decision journeys. We provide a novel solution to these challenges by combining econometric and machine learning (Graphical Lasso) approaches, leveraging customer clickstream information to learn the product correlation network, and creating high-dimensional choice models that easily scale and allow for flexible substitution patterns. Our model offers better in- and out-of-sample demand forecasts and enhanced pricing recommendations in various synthetic datasets and in a real-world empirical setting.