Collective deterrence as a classification problem: Voting rules, deterrence credibility, and escalation risk

TL;DR

This paper models deterrence coalitions as a classification problem, analyzing how institutional design impacts deterrence credibility and escalation risk through a signalling model and ROC curve analysis.

Contribution

It introduces a formal framework linking deterrence decision-making to classification, providing empirical analysis of different institutional choice functions.

Findings

ROC curves reveal tradeoffs between deterrence credibility and escalation risk.

Certain institutional designs outperform others across various environments.

Abstract

Deterrence coalitions that collectively own their deterrence technology, need an institutional design to decide when to retaliate against an attack or incident. This choice of institutional design, formalized through a social choice function, introduces a tradeoff between credible deterrence and escalation risk. We study this tradeoff via a simple signalling model, and use it to construct an associated binary classification problem to determine institutional designs that perform well in a variety of environments. For a small coalition of four members, we compute and study the statistics of the empirical ROC curves associated to a variety of choice functions and probability distributions for retaliation and false positives.