On Constrained Quasi-Likelihood Estimation
本文证明拉格朗日乘子法可推广到约束拟似然估计，即使没有目标函数也能进行最优估计，适用于半参数模型等无法使用最大似然估计的问题。
For maximum likelihood or least squares parameter estimation subject to a constrained parameter, the standard approach is to use the method of Lagrange multipliers. In this paper it is shown that the same formal procedure applies very generally for constrained quasi-likelihood estimation even though there is ordinarily no objective function to maximize or minimize. This allows for optimal estimation in a wide range of problems where maximum likelihood cannot be used, such as for semiparametric models. The method is very similar to projecting the free estimator or projecting the equations which it solves.