The Instability of Matching with Overconfident Agents

Many centralized college admissions markets allocate seats to students based on their performance on a single standardized exam. The exam's measurement error can cause the exam-derived priorities to deviate from colleges' aptitude-based preferences. Previous literature proposes to combine pre-exam preference submission with a Boston algorithm (a PreExam-BOS mechanism). This paper examines the proposed mechanism in an experiment where students are not fully informed of their relative aptitudes. The results show pre-exam preference submission is distorted by overconfidence and PreExam-BOS fails to achieve stable matching with respect to aptitudes. Compared to a post-score Serial Dictatorship mechanism, which is robust to overconfidence but more prone to the exam's measurement error, PreExam-BOS creates more mismatches and a greater variance in the extent of mismatches: some students receive a large advantage while others are hurt considerably. Moreover, PreExam-BOS rewards overconfidence and punishes underconfidence. The observed overconfidence cannot be mitigated with an improved information condition.