This paper extends global games theory to Generalized Pareto Distribution (GPD) noise. Our framework encompasses all empirically relevant tail behaviors through the shape parameter ξ, including Pareto-type distributions (0 < ξ < 0.5), exponential tails (ξ = 0), and light-tailed cases (ξ < 0). The central challenge is that GPD-Gaussian mixing yields non-Gaussian posteriors even with Gaussian priors, invalidating the Morris-Shin analytical approach based on closed-form Gaussian calculations. We establish various results of theoretical and empirical relevance. This work establishes that the unique switching equilibrium can survive heavy-tailed information ubiquitous in financial crises, providing theoretically justified and empirically implementable tools for testing coordination thresholds in crisis data without imposing untenable distributional assumptions.