Predicting Proliferation: High Reliability Forecasting Models of Nuclear Proliferation as a Policy and Analytical Aid

Abstract

Researchers have now produced a number of quantitative studies of the determinants of nuclear proliferation, using data on all known nuclear weapons programs. But while scholars have laid important groundwork in understanding the causes of nuclear pursuit, these studies are primarily focused on explaining rather than predicting proliferation. Drawing from existing quantitative work, this project uses statistical learning methods to construct a predictive model of proliferation, focusing on the ability of different nuclear proliferation theories to make accurate out-of-sample predictions. This study makes two contributions to the literature on nuclear proliferation and the larger policy debate. First, it identifies for the first time an empirically grounded set of nuclear “triggers”— conditions under which countries are most likely to shift from latent nuclear capacity to a full-fledged nuclear weapons effort. Understanding these triggers has become increasingly important, as more states have begun to pursue a nuclear hedging strategy in which they seek dual-use nuclear capabilities without committing to a weapons program. Second, this study helps to reconcile conflicting empirical findings in the literature. Predictive analytics provide a new and useful way of understanding the substantive significance of existing empirical findings, and of comparing the relative importance of different theoretical approaches.