Learning and Market Design

Technological progress and the growing need for a fair or efficient allocation of limited resources in a globalized world has generated numerous instances where institutions have been designed to find and implement efficient, fair, or stable allocations. For instance, kidney transplantation became a more and more successful substitute for the traditional dialysis treatment of kidney insufficiencies asking for a timely and transparent allocation procedure of donor kidneys. Capacity constraints of schools and universities together with an ever-increasing mobility of families and students ask for a fair and easy-to-handle process of school choice and student-to- university matching. Or globalized labor markets with thousands of mobile and highly qualified potential employees ask for a coordinated effort of employers possibly including a centralized assignment of standardized jobs (see, e.g., the US-based entry-level labor market for young scientists or medical interns). In response to these increasing needs for efficient, fair and stable allocation procedures, institutions have been developed where participants (e.g., kidney donors and patients, students, employers, or employees) submit their preferences and are (based on these preferences) assigned a resource or match (e.g., a kidney, a school, a study program, or a job). When designing these institutions, scientists have to adopt some assumptions regarding the behavior of participants. Typically, it is assumed that participants recognize so-called dominant strategies (i.e., a behavior that cannot be improved upon regardless of the behavior of other participants) or that participants coordinate on some kind of equilibrium behavior, or act according to simple heuristics. For instance, a student can never benefit from stating something else than her true ranking of schools when the so-called deferred acceptance algorithm is used. But recent laboratory studies of such institutions suggest that individuals are not always able to learn such optimal behavior. For instance, subjects in laboratory experiments frequently restrict applications to (laboratory) schools that are known to be save bets in the sense that these schools surely have the capacity etc. to admit the corresponding student. The project will develop a learning model to analyze which type of behavior will be exhibited depending on the situation at hand (e.g., the information given to participants, the number of participants, the preference intensity, etc.). We will test the predictions of the learning process with data generated by accompanying laboratory studies with students who participate in the different allocation mechanisms. With the theoretical toolkit of learning models developed in the project and the empirical evidence gathered in the laboratory studies, it will become easier to describe the behavior of participants as a basis for real-life institutional design.

Based on models and laboratory experiments this project investigates the manipulability and fairness perceptions of mechanisms and algorithms for the allocation of scarce ressources (e.g., donor organs, school choice, college admissiojn, or asylum seeking). While market-like mechanisms are manipulable and manipulated, these mechanisms are perceived as fairer than sequential mechanisms that are not manipulable but allocate based on an exogenous or randomly determined sequence. As markets grow large, manipulability vanishes and allocations of different mechanisms converge - while fairness perceptions remain unaltered.