Collaborative Research: Development of a Technology for Real Time Ex Ante Forecasting of Intra and International Conflict and Cooperation

合作研究：开发实时事前预测内部和国际冲突与合作的技术

• 批准号: 0921018
• 负责人: John Freeman
• 金额: $ 16.84万
• 依托单位: University of Minnesota-Twin Cities
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2009
• 资助国家: 美国
• 起止时间: 2009-10-01 至 2013-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0921018&HistoricalAwards=false
• 关键词: Collaborative; Research; Development; Technology; Real

This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5).Conflicts continue to rage in the Levant, South Asia, and elsewhere. Governmental, nongovernmental, and intergovernmental organizations are seeking a technology that will help them anticipate the outbreak of violence. Some progress has been made predicting single events and political instability two years ahead. But we lack scientifically sound tools for gauging the statistical uncertainty of these and other forecasts, making predictions in real-time, analyzing the behavior of groups of belligerents, and making contingent forecasts (forecasts conditional on possible interventions by third parties). We develop a scientifically advanced technology that meets these desiderata: a tool that produces on an open source website in real-time, ex ante forecasts of intra- and international conflict and cooperation for three parts of the world. The time series data is produced by state-of-the art TABARI automated coding software using the CAMEO event and actor coding system; the source are news feed aggregators such as Google News and the European Monitor accessed via real simple syndication feeds (RSS). Bayesian multivariate time series models with Markov-switching are fit to these data. These models then produce forecast probability densities for systems of directed dyadic behaviors. Model performance is based on density evaluation; model refinements are made via developed recalibration methods. The website will report forecasts for selected systems of belligerents in the Levant, South Asia, and East Asia; it will disseminate the coded event data, real-time forecasts of these cases with measures of forecast uncertainty, contingent forecasts based on possible interventions, information on model performance. The technology will meet most, if not all, of the desiderata for forecasting in international relations. Three specific contributions are be made. First, the feasibility and value of automated real-time events coding is demonstrated. Until now, this coding has lagged real-time; and the value of the data has not been realized. The application of current statistical work on forecasting is a second contribution. For the first time in political science we produce and evaluate density (not point) forecasts. And we refine our forecasting models with newly developed recalibration techniques. Third, the well known tendency for conflict and cooperation phase shifts are explicitly analyzed and incorporated in our forecasts. The Markov-switching element of our model capture these nonlinearities. Using the Teragrid we produce probability weighted forecasts of behavioral systems occupying certain states, report transition probabilities, and include estimates of the uncertainty of both. Thus the final forecasting models will be original in international relations. The forecasts we produce will be available for use by NGOs, IGOs, and any party interested in anticipating conflict in the Levant, South Asia, and East Asia. To help these parties use our forecasts we will make presentations at meetings held by the A.P.S.A, I.S.A, and I.P.S.A. We also will present our work and publicize our website in countries like Taiwan via institutions like the Academia Sinica. Eventually, our website will report contingent forecasts based on inputs from scholars and policy makers' forecasts of what will transpire in the short-term if certain kinds of intervention occur. Estimates of the probability of phase shifts and of forecast densities will accompany these contingent forecasts.

该奖项是根据2009年美国复苏和再投资法案（公法111-5）资助的。政府、非政府组织和政府间组织正在寻求一种技术，帮助他们预测暴力事件的爆发。在预测未来两年的单一事件和政治不稳定方面取得了一些进展。但是，我们缺乏科学可靠的工具来衡量这些预测和其他预测的统计不确定性，进行实时预测，分析交战方群体的行为，并进行应急预测（以第三方可能的干预为条件的预测）。我们开发了一种满足这些需求的科学先进技术：一种在开源网站上实时生成的工具，预先预测世界三个地区的内部和国际冲突与合作。时间序列数据是由最先进的TABARI自动编码软件使用CAMEO事件和演员编码系统制作的;数据来源是通过真实的简单的联合供稿（RSS）访问的谷歌新闻和欧洲监测等新闻供稿聚合器。贝叶斯多变量时间序列模型与马尔可夫转换拟合这些数据。这些模型，然后产生预测的概率密度系统的有向二元行为。模型性能是基于密度评估，模型的改进是通过开发的重新校准方法。该网站将报告对黎凡特、南亚和东亚交战国的选定系统的预测;它将传播编码事件数据、对这些情况的实时预测以及预测不确定性措施、基于可能干预的应急预测、模型性能信息。这项技术将满足国际关系中预测的大多数（如果不是全部的话）需求。有三个具体的贡献。首先，论证了自动化实时事件编码的可行性和价值。到目前为止，这种编码方法存在实时性差，数据的价值没有得到充分的发挥。第二个贡献是将目前的统计工作应用于预测。在政治科学中，我们第一次制作和评估密度（而不是点）预测。我们用新开发的重新校准技术改进我们的预测模型。第三，众所周知的冲突和合作的相变趋势进行了明确的分析，并纳入我们的预测。我们的模型的马尔可夫切换元素捕捉这些非线性。使用Teragrid，我们产生概率加权预测的行为系统占据某些状态，报告转移概率，并包括两者的不确定性的估计。因此，最终的预测模型在国际关系中将是原创的。我们的预测将提供给非政府组织、政府间组织和任何有兴趣预测黎凡特、南亚和东亚冲突的各方使用。为了帮助这些机构使用我们的预测，我们将在美国邮政协会、国际邮政协会和国际邮政协会举行的会议上发表演讲。我们也会透过中央研究院等机构，在台湾等国家介绍我们的工作，并宣传我们的网站。最终，我们的网站将根据学者的投入和政策制定者的预测报告应急预测，如果发生某种干预，短期内会发生什么。对相移概率和预测密度的估计将伴随这些应急预测。