Forecasting State-to-State Migration Rates
利用面板数据模型和诊断图形方法，分析美国州际迁移率的方差结构，发现百分比变化是平稳的一阶自回归过程，可用于短期预测。
Models of internal migration can be found in the demographic, geographic, and economic literatures. Unlike most of these models, which are based on cross-sectional analysis of migration rates, the approach here is to incorporate longitudinal as well as cross-sectional aspects by using models from the panel-data literature. Empirical modeling using this approach is possible using a data base recently revised and updated by the U.S. Bureau of the Census. The focus of this article is on model-selection aspects, relying heavily on diagnostic and graphical methods. Understanding the variance structure of the data is particularly important for forecasting purposes. An interesting aspect of the variance structure is the presence of rampant cross-sectional heteroscedasticity, moderate contemporaneous correlation, and mild longitudinal autocorrelation of the transformed data. An important conclusion of the model selection and diagnostic verification is that, based on the available data, percentage changes of state-to-state rates are stationary and can be modeled by an autoregressive process of order 1. A corollary of this conclusion is that the recent migration rate takes on an important role in short-term forecasting of internal migration.