On the Relationship between the Estimates of Level Models and Difference Models
证明，当线性回归模型包含常数项且扰动项方差协方差矩阵正定时，其斜率系数的广义最小二乘估计与对应差分模型的估计相同；并说明若模型设定不当（如遗漏常数项或误用普通最小二乘），估计效率会大幅损失。
Abstract This paper proves that if a linear regression model has a constant term and its disturbances have a positive‐definite variance‐covariance matrix, its slope coefficients have the same generalized least squares estimates as the slope coefficients of the corresponding difference model. It is also illustrated that if the original model is homogenous (i.e., with no constant term) and/or a constant term is inadvertently included in the difference model and/or the ordinary least squares estimator is adopted for the difference model, one can incur a great loss in estimation efficiency.