VAR Model with Sparse Group LASSO for Multi-population Mortality Forecasting
提出一种时空加权向量自回归模型，结合稀疏组LASSO方法，用于多人口死亡率建模与预测，通过整合时空权重提升估计精度和预测表现。
We introduce a spatial–temporally weighted vector autoregressive (SWVAR) model for modeling and forecasting mortality rates across multiple populations. First, we stack the mortality rates of the populations and build a vector autoregressive (VAR) model. Next, we apply the sparse group least absolute shrinkage and selection operator (sparse group LASSO) for fitting to avoid overparameterization. Furthermore, we integrate spatial–temporal weights, derived from age differences and geographic centroid distances, into the grouped penalty term. These approaches allow the resulting model to effectively combine information from multiple populations and reduce confounding factors associated with combined modeling. We demonstrate through a series of empirical experiments that the spatial–temporally weighted VAR model enhances estimation accuracy and exhibits superior in-sample fitting and out-of-sample forecasting performance.