Low-Rank Payoffs and Limit Uniqueness in Global Games

Abstract

When does the global game information structure select a unique equilibrium? Limit uniqueness in two-player supermodular games fails exactly when a risk-dominant better response cycle exists (Veiel, 2025). We show that rank-one factor structure on payoffs eliminates such cycles entirely, so every rank-one supermodular game admits a generalized ordinal potential and limit uniqueness follows for any number of actions. The boundary is sharp: an explicit three-action rank-two game carries a length-six cycle, no supermodular game carries a cycle of length four, and every game within a quantified sup-norm margin of a nondegenerate rank-one game is cycle-free. Rank-one structure can also be manufactured: when players compete across many independent markets with common latent payoffs, the stacked observation matrix is rank one plus sparse, and a Robust PCA estimator leaves residual noise that vanishes with the signal scale yet stays positive at any finite sample, even under partial observation.