Incentive-Compatible Selection Mechanisms for Forests

TL;DR

This paper investigates incentive-compatible selection mechanisms in forest-graphs, establishing bounds on their quality, introducing novel mechanisms, and proving impossibility results for certain desirable properties.

Contribution

It provides new bounds on the quality of IC mechanisms, introduces two novel mechanisms with different fairness and exactness properties, and proves an impossibility result for combined fairness and exactness.

Findings

Upper bound of 4/5 on mechanism quality.

Lower bound of approximately 0.36 achieved by two mechanisms.

Existence of mechanisms that are fair but not exact, and vice versa.

Abstract

Given a directed forest-graph, a probabilistic \emph{selection mechanism} is a probability distribution over the vertex set. A selection mechanism is \emph{incentive-compatible} (IC), if the probability assigned to a vertex does not change when we alter its outgoing edge (or even remove it). The quality of a selection mechanism is the worst-case ratio between the expected progeny under the mechanism's distribution and the maximal progeny in the forest. In this paper we prove an upper bound of 4/5 and a lower bound of $ 1/\ln16\approx0.36 $ for the quality of any IC selection mechanism. The lower bound is achieved by two novel mechanisms and is a significant improvement to the results of Babichenko et al. (WWW '18). The first, simpler mechanism, has the nice feature of generating distributions which are fair (i.e., monotone and proportional). The downside of this mechanism is that it is not exact (i.e., the probabilities might sum-up to less than 1). Our second, more involved mechanism, is exact but not fair. We also prove an impossibility for an IC mechanism that is both exact and fair and has a positive quality.