基于美国抽烟数据的时空自回归模型的统计推断

Inference of a Spatio-Temporal Autoregressive Model Based on US Smoking Data

针对目前文献中大多数时空模型的建立大都需要提前指定相关的空间权重矩阵,而当空间权重矩阵形式设定错误时,模型的解释性变得极为不可靠,预测能力也大大降低的问题,提出了更为一般的时空自回归模型来拟合美国抽烟需求数据并进行单步与多步预测,模型将空间权重矩阵转化为带估的时空系数矩阵,采用基于Yule-Walker方程的广义矩估计法和基于Yule-Walker方程的最小二乘两种方法来估计系数矩阵,最终的两种预测方法结果均表明建模效果较好。另外模型分析结果也表明:人均吸烟包数与人均可支配收入呈正相关关系,而人均吸烟包数与每包香烟的零售价格呈负相关关系,均与现实意义相符。

In view of the establishment of most of the spatiotemporal models in the current literature,most of them need to specify the relevant spatial weight matrix in advance,and when the spatial weight matrix form is set incorrectly,the explanatory nature of the model becomes extremely unreliable,and the prediction ability is greatly reduced,a more general spatiotemporal autoregressive model is proposed to fit the US smoking demand data and make single-step and multi-step prediction,and the model converts the spatial weight matrix into a spacetime coefficient matrix with estimation.The generalized moment estimation method based on Yule-Walker equation and the least squares method based on Yule-Walker equation were used to estimate the coefficient matrix,and the final results of both prediction methods showed that the modeling effect was better.In addition,the model analysis results also show that the per capita number of smoking packs is positively correlated with per capita disposable income,while the per capita number of smoking packs is negatively correlated with the retail price of each pack of cigarettes,which is consistent with practical significance.