Transparency in Centralised Allocation: Theory and Experiment

Abstract

Many algorithmic allocation mechanisms suffer from a verifiability problem: participants cannot check if their assignments are correct. This problem is compounded if there are suspicions that the designer has deviated from the true allocation. We formalise these concerns and propose solutions in an information-based framework. A participant's assignment is `verifiable' by her if any other assignment contradicts her information. A stronger requirement is `transparency', where the designer cannot deviate from the true allocation without being detected. We show how the communication of `terminal-cutoffs' and the use of `predictable' multi-stage mechanisms each provide information to participants that verifies their assignments. Even though the information from predictable mechanisms and terminal-cutoffs can each be manipulated by a dishonest designer without detection, in our main result we show that they nevertheless achieve transparency if used together. We suggest transparent environments for use in school admissions, single-object auctions and house allocation. We support the effectiveness of our solutions via a school admissions laboratory experiment.

Keywords: Mechanism design, information, designer incentive-compatibility, dynamic mechanisms, cutoffs, school admissions experiment

JEL Classification: C78, C73, D78, D82