基于模拟退火算法的工具变量选取方法

工具变量估计是解决模型内生性问题的基本方法，但其有限样本表现对工具变量的选取十分敏感. 近年来，对于“多工具变量”模型，文献中提出了基于近似最小均方误差的工具变量选取方法来解决这一问题，但这些方法或依赖于工具变量的排序，或受限于工具变量的数目. 本文采用基于模拟退火算法的工具变量选取方法很好地克服了这些缺陷. Monte Carlo 模拟结果表明该算法有效可行.

Selecting instrumental variables based on simulated annealing algorithm

Instrumental variables estimation provides a general solution to the problem of an endogenous explanatory variable, but the finite sample properties of instrumental variable estimators are sensitive to the choice of instruments. In recent years, several approaches based on minimizing the approximate mean square error have been proposed in the literature for the models with many instruments. However, these methods either depend on the order of the instruments or are limited by the number of instruments. In this paper, the selection method of instruments based on simulated annealing algorithm is proposed to solve these problems. Monte Carlo simulations have demonstrated the effectiveness of the proposed algorithm.