固定效应部分线性可加动态面板模型的惩罚二次推断函数估计

Penalized quadratic inference estimation for partially linear additive dynamic panel model with fixed effects

工具变量法是估计动态面板模型的常用方法,但该方法并没有充分利用现有矩条件,导致所得估计有效性不足.为此,本文首先采用变量变换法消除模型的内生性,再用惩罚二次推断函数法推导出个体内具有一阶自相关结构的固定效应部分线性可加动态面板模型中未知参数和函数的估计;进一步,证明了所得估计量的一致性和渐近正态性,同时还用Monte Carlo模拟实验比较了该方法和半参数GMM法在有限样本下的表现;最后将所述方法应用于实际数据分析中.

Instrument variable method is a common method for estimating dynamic panel model.However,the method leads to reduce validity of estimation because it doesn't make full use of the existing moment conditions.To overcome this drawback,endogeneity of the model is first eliminated by variable transformation method and then penalized quadratic inference function method is used to derive the estimation of unknown parameters and functions of partially linear additive dynamic panel model with fixed effects which have first order autocorrelation structure within subjects.Furthermore,consistency and asymptotic normality for the obtained estimators are proved and the finite sample performances of the proposed method and semi-parametric GMM method are compared by Monte Carlo simulation studies.Lastly,the proposed method is illustrated in the analysis of a real data set.