广义矩估计模型平均

GMM-Based Model Averaging Method

模型平均方法以其稳健性好，遗失有用信息少等诸多优点而成为目前统计学和计量经济学界研究的热门问题，在经济，金融，生物，医学等领域有着广泛的应用前景．在模型平均的理论研究过程中如何选取权重是最重要的问题．现存的模型平均方法大都是在最小二乘估计的基础上研究的，并且大多数通过光滑AIC，光滑BIC和最小化Mallow准则得到组合的权重．但是在广义矩估计的基础上模型平均的研究还很不完善，文章在广义矩估计条件下提出了通过最小化目标参数平均估计量的渐近方差来获取权重．这种方法使得所得到的平均估计更加的稳健．蒙特卡洛模拟实验显示文章获得的模型平均估计量的风险与光滑AIC，光滑BIC和MMA相比相对较低．

Model averaging has become a hot topic in the field of statistics and econometrics, with its robustness and loss of less useful information. It has wide ap- plications in many fields such as economics, finance, biology and medicine. However, how to select weights is the most important question in the theoretical research of the model average. As we all know, existing model average methods are based on the least square estimate mainly, and most of them are weighted by smoothed AIC, smoothed BIC and the minimization of Mallow criteria. The different work of this paper is that minimizing the estimation of asymptotic variance of estimator for interesting parameter to get the weights is proposed under the generalized method of moments （GMM）. This method could make the average estimate more stable. The simulation experiments show that the risk of GMM model average estimator is relatively low compared with those based on S-AIC, S-BIC and MMA.