On the Complexities of Understanding Matching Mechanisms

We study various novel complexity measures for two-sided matching mechanisms, applied to the popular real-world school choice mechanisms of Deferred Acceptance (DA) and Top Trading Cycles (TTC). In contrast to typical bounds in computer science, our metrics are not aimed to capture how hard the mechanisms are to compute. Rather, they aim to capture certain aspects of the difficulty of understanding or explaining the mechanisms and their properties. First, we study a set of questions regarding the complexity of how one agent's report can affect other facets of the mechanism. We show that in both DA and TTC, one agent's report can have a structurally complex effect on the final matching. Considering how one agent's report can affect another agent's set of obtainable options, we show that this effect has high complexity for TTC, but low complexity for DA, showing that one agent can only affect another in DA in a quantitatively controlled way. Second, we study a set of questions about the complexity of communicating various facets of the outcome matching, after calculating it. We find that when there are many more students than schools, it is provably harder to concurrently describe to each student her match in TTC than in DA. In contrast, we show that the outcomes of TTC and DA are equally hard to jointly verify, and that all agents' sets of obtainable options are equally hard to describe, showcasing ways in which the two mechanisms are comparably complex. Our results uncover new lenses into how TTC may be more complex than DA. This stands in contrast with recent results under different models, emphasizing the richness of the landscape of complexities of matching mechanisms. Our proofs uncover novel structural properties of TTC and DA, which may be of independent interest.