Definition
根据 Linear Projection Model#求解参数 可知
The moment estimator of is
define as the minimizer of
and call as the least squares estimator of
注意, 是总体参数,是一个数; 是一个样本估计量,是一个随机变量,其随机性来源于抽样。
Solution
定义 Sum of Squared Risiduals 为 ,易知
写作矩阵形式
先展开再求导:
F.O.C.(称之为正规方程组)
therefore
S.O.C.
这要求 是正定矩阵。
也可以先求导再展开:
F.O.C. (前导不变,后导转置)
其中
Residual Properties
根据前面的(※)可知
若自变量包含常数项,则有
若自变量包含虚拟变量,则残差的分组均值为零。
附录:no-matrix calculus
单变量回归的情形
F.O.C
therefore
多变量回归的情形
注意: 是一个数因此转置后不变,即 ,这样变换将有助于后续应用矩阵求导公式。
F.O.C.
注意: 是一个对称矩阵所以转置后不变。
therefore
S.O.C
which is a positive semi-definite matrix
二阶条件对应总体模型的假设