Arms Race Modeling

Abstract

The Richardson model has been employed frequently in the study of arm races, but seldom with empirically satisfactory results. For each of twelve pairs of arms expenditure series which are commonly proposed as arms races, we consider two potential sources for the failure of Richardson modeling efforts. First, they may not constitute arms races. Second, for those which do, the Richardson model specification may not be adequate. The methodology which enables us to explore these issues is "statistical causality analysis," as recently promulgated in time series research. Using this methodology, we find that arms expenditures are independent in a majority of the twelve cases examined. For those cases where interdependence is found, those which exhibit arms competition, the traditional Richardson model specification is not empirically supported. Finally, we provide a specific methodology for improved model specification derived from a general distributed lag model for any proposed arms race.