International war and conflict threaten the lives and well-being of millions of people. On the basis of theoretical work in international relations, the co-investigators develop multivariate, Bayesian vector autoregression time series models (BVARs) for short and medium term conflict forecasting and for the analysis of counterfactuals of various kinds, more specifically, Markov-switching models of this type (MS-BVARs) for the Balkans, Israeli-Palestinian, and India-Pakistan conflicts. The models will incorporate and test theoretical work on conflict phase shifts as well as the results of research on the impact of elections and democratic transitions on conflict. Scaled events data from two sources (KEDS/TABARI and IDEA) will be used to estimate the model. The models will yield short and medium term, case specific, quantitative predictions of conflict and cooperation; they eventually will incorporate assessments of the welfare consequences of conflict. Finally, by developing priors for the model coefficients and constructing posterior inferences for the models' predictions, for one of the first times in political science, explicit assessments of the impact of model uncertainty on (political) forecasts accuracy will be produced. Intellectual Merit. The proposed has theoretical and practical value. To begin, it will produce statistically sound characterizations of conflict phase sequences. The investigators test formally for the number of phases in the Balkans, Israeli-Palestinian, and Indian-Pakistani conflicts, and also provide numerical estimates of the transition probabilities between these phases (along with measures of the precision of these estimates). They then will produce quantitative, weekly and monthly predictions of the future course of the three conflicts conditional on the realization of specific conflict phases and on the steady state probabilities of the conflict phases for each case. In addition, the fitted MS-BVARs will illuminate similarities and differences in the three conflicts' dynamics. For instance, the fitted models will show if there are common degrees of persistence in them, the extent to which the three conflicts display the same patterns of reciprocity and triangularity, whether the conflicts tend toward the same long-term (fixed) mean levels of conflict, and whether provision for electoral forces and transitions to (from) democracy enhance the predictive power of the MS-BVARs. Impulse response analysis will yield insights into the possible impact of hypothetical, surprise peace initiatives by third parties (conditional on the conflict phase). The methods of conditional forecasting (with the BVARs and MS-BVARs) will be used to analyze counterfactual histories of the conflicts. For example, by inserting counterfactual variables for elections in Pakistan in the late 1990s we will examine of the counterfactual consequences of that country not experiencing a democratic reversal on its conflict with India. Broader impact. The results will be disseminated in several ways. First, a web-site will be constructed. The website will contain the investigators' computer code, data, and examples. It also will contain a tutorial on how to construct and apply BVARs and MS-BVARs for selected international conflicts. Second, the investigators will offer short-courses on how to build and apply MS-BVARs. These courses will be offered at such gatherings as the International Studies Association (ISA) , Peace Science Society (PSS), and American Political Science Association (APSA) meetings. Every effort will be to include "unrepresentative groups" in these short course and training sessions. Scientifically, the project will demonstrate the usefulness of events data in political forecasting, and advance our understanding of Bayesian time series methods in the social sciences. American and global society will benefit from being better able to anticipate international conflict weeks and months ahead and also being able to evaluate counterfactuals of various kinds.
