This paper proposes a Global Vector Autoregression (GVAR) mortality model to simultaneously model and forecast multi-population mortality dynamics. The proposed GVAR model decomposes the global regression model into population-wise local systems. Each local system consists of an intra-population autoregressive component and a small set of global factors, which contain systematic mortality information of all populations. Such a decomposition substantially reduces the extra estimation cost of including new populations compared to unconstrained VAR models, and makes the GVAR model an efficient tool for analyzing the joint mortality dynamics of a large group of populations. Further, under fairly general assumptions, the proposed GVAR model could generate coherent mortality projections between any two ages in any two populations. Using single-age mortality data of 15 low-mortality countries, we find that the global factors have substantial explanatory and forecasting power of mortality changes of individual populations, and the proposed GVAR model could produce satisfying mortality forecasts under various settings.

本文提出了一种全球向量自回归（GVAR）死亡率模型，以同时建模和预测多种群死亡率动态。提出的GVAR模型将全局回归模型分解为按人口分类的局部系统。每个本地系统都由一个人口内部自回归组件和一小部分全局因素组成，其中包含所有人群的系统死亡率信息。与无约束的VAR模型相比，这种分解大大减少了包括新种群在内的额外估计成本，并使GVAR模型成为分析大量人群联合死亡率动态的有效工具。此外，在相当普遍的假设下，建议的GVAR模型可以在任何两个人口的任何两个年龄之间生成一致的死亡率预测。使用15个低死亡率国家的单年龄死亡率数据，我们发现全球因素对单个人群的死亡率变化具有重要的解释和预测能力，并且所提出的GVAR模型可以在各种环境下产生令人满意的死亡率预测。