Dan Fragiadakis defends his Ph.D. dissertation.
His three papers in his thesis are:
Improving Welfare in Assignment Problems: an Experimental Investigation (with Peter Troyan).
Many institutions face the task of allocating objects (such as university dormitories) to individuals (students) without the use of monetary transfers. A common solution to this problem is the Random Serial Dictatorship (RSD): agents are ordered randomly, and one at a time, each is assigned her favorite good according to her submitted preferences. While RSD provides each agent with a dominant strategy of ranking objects truthfully, it may produce socially undesirable outcomes whereby it is possible to make some agents substantially better off at only a small cost to others. In this paper, we study the prospect of raising welfare in assignment problems by incentivizing agents to report goods they value similarly as indifferent. Specifically, we modify RSD by ordering agents earlier who report more indifference, a method similar to that used by the Stanford Graduate School of Business to assign MBA students to educational trips abroad. While theory predicts weak welfare gains in equilibrium, this requires agents to calculate nontrivial best response strategies that deviate from simple truth-telling. In practice, it is unknown whether agents will be able to find these equilibria and, if they cannot, what the welfare implications of using such mechanisms will be. Motivated by these observations, we run a lab experiment where we find that many agents follow natural heuristics that entail reporting indifferences between objects that are similar in value. Average earnings increase significantly compared to RSD, but the way in which indifference is rewarded can alter the variance in earnings. This suggests that institutions that use RSD can benefit by rewarding indifference, but should choose how to do so carefully.
Identifying Predictable Players: Relating Behavioral Types and Predictable Subjects (with Muriel Niederle and Daniel Knoepfle).
Behavioral game theory models are useful in organizing data of strategic decision making. However, are subjects classified as behavioral types more predictable than unclassified subjects? Alternatively, how many predictable subjects await new behavioral models to describe them? In our experiments, subjects play two-person guessing games against random opponents and are subsequently asked to replicate or best respond to their past choices. We find that existing behavioral types capture two thirds of strategic subjects, i.e. individuals who can best respond. However, there is additional room for non-strategic rule-of-thumb strategies to describe subjects that can merely replicate their actions.
Market Design under Distributional Constraints: Diversity in School Choice and Other Applications (with Peter Troyan).
Distributional constraints are important in many market design settings. Prominent examples include the minimum manning requirements at each branch in military cadet matching and diversity in school choice, whereby school districts impose constraints on the demographic distribution of students at each school. Standard assignment mechanisms implemented in practice are unable to accommodate all of these constraints. This leads policymakers to resort to ad-hoc solutions that eliminate blocks of seats ex-ante (before agents submit their preferences) to ensure that all constraints are satisfied ex-post.
We show that these solutions ignore important information contained in the submitted preferences, resulting in avoidable inefficiency. We introduce a new class of dynamic quotas mechanisms that allow the institutional quotas to dynamically adjust to the submitted preferences of the agents. We show how a wide class of mechanisms commonly used in the field can be adapted to our dynamic quotas framework. Focusing in particular on a new dynamic quotas deferred acceptance (DQDA) mechanism, we show that DQDA Pareto dominates current solutions. While it may seem that allowing the quotas to depend on the submitted preferences would compromise the strategyproofness of deferred acceptance, we show that this is not the case: as long as the order in which the quotas are adjusted is determined exogenously to the preferences, DQDA remains strategyproof. Thus, policymakers can be confident that efficiency will be improved without introducing perverse incentives. Simulations with school choice data are used to quantify the potential efficiency gains.
I blogged about him before
He will be joining the Economics department at Texas A&M next year.
Congratulations, Dan, and welcome to the club!