Date of Completion


Embargo Period



Mechanism Design, Matching Theory, Team Formation, Peer-to-peer lending, Game Theory, Bi-level optimization, Stability, and Fairness

Major Advisor

Robert Day

Associate Advisor

David Bergman

Associate Advisor

Mike Shor

Field of Study

Business Administration


Doctor of Philosophy

Open Access

Open Access


In this dissertation we analyze matching markets using two broad methodologies. In the first two of three essays, we use integer programming to design new practical markets, solving difficult organizational problems in which items like a seat in a classroom course or membership on a project team are allocated to agents who express their preferences to a centralized planner. In the third essay, we employ empirical modeling to study the signals sent among three types of agents in a peer-to-peer (P2P) lending market: borrowers, lenders, and a (rating-service) platform. The first two essays use extensive simulations, while the third uses statistical analysis on a large empirical dataset: four years of loan application and payment history from a prominent online P2P lending platform.

Our findings in the first essay show that large but manageable course allocation problems can be solved with various multi-stage optimization algorithms, providing much better outcomes than existing benchmarks from the literature on metrics of efficiency, fairness, and incentive-compatibility. We demonstrate robustness of our techniques by simulating over a variety of market parameters, including varying degrees of manipulation over a range of common to private value utility functions. These results show promising new practical designs that can satisfy more organizational objectives than previous methods.

In the second essay, we find that the much harder (interpersonal) quadratic-interaction optimization in an agent-based team formation game requires advanced computational technique. In the pursuit of a balance between the competing objectives efficiency and group stability, we explore the cutting-edge of computational operations research. In contrast to existing draft-based systems that favor distributed, intuitive heuristics, we solve the extremely difficult centralized optimization problems directly, with positive results using a new technique for bi-level optimization in this context, based on a customized column generation framework. Again, the results produce a promising new optimization framework for a practical problem, and our results compare favorably with existing methods.

Finally, the results of the third essay show just how complex behavior in real-world matching markets can be through the empirical analysis of a real market. We verify a few intuitive results, but also find some counterintuitive interactions, in which a monotonically increasing signal from the market platform results in a non-monotonic return on investment (ROI), for example. Overall, we find risk-seeking behavior among peer investors that does not tend to pay off, and some strange disconnects in which, for example, investors favor “Debt Consolidation” loans despite inferior ROI, and have a prejudice against “Business” loans despite no significant evidence of poor performance.