时变固定效应动态短面板模型的变量选择估计及应用

Variable selection,estimation and application of dynamic short panel data models with time-varying individual fixed effects

虽然常见的微观面板数据时间维度较短,但仍具有一定的时间跨度,个体固定效应依旧可能存在时变性,导致基于个体固定效应的面板数据模型推断不可信.因此,考虑时变个体固定效应能够降低模型设定偏误.采用修改的Mundlak-Chamberlain投影方法,不仅能控制不可观测的时变个体固定效应,还能识别时不变的个体解释变量的影响效应,如工资的性别歧视.另一方面,解释变量的影响效应同样会因政策等外部宏观环境因素的变动存在时变性.投影和时变效应的引入都会极大增加模型中待估参数的个数,降低模型估计的精度.为此,重要变量的筛选及其时变性的识别尤为关键.通过引入投影方程参数的约束条件,采用惩罚条件拟似然和ECM算法,能同时实现变量的选择和估计.模拟结果显示新方法表现良好.最后,基于该模型和方法识别了不同的高管激励方式(分薪酬激励和股权激励)如何影响企业配置金融资产的动机.

Although the time dimension of common micro panel data is short,it often has a long time span,which may lead to incorrect inference based on the usual fixed effects panel data model.However,panel data model with time-varying individual fixed effect can alleviate model specification error.We adopt the modified Mundlak-Chamberlain projection which enables us to identify the influence of time invariant individual explanatory variables,such as gender discrimination in wages.Moreover,the effects of observed explanatory variables may also be time-varying due to external environmental factors such as policies.This,together with the above projection will greatly increase the number of parameters,and reduce the accuracy of estimation.Therefore,we study variable selection and introduce the constraint conditions of the projection parameters.The conditional penalized quasi-likelihood method and ECM algorithm are used to obtain selection and estimation simultaneously.Simulation studies show that our methods perform well.Finally,we identify the influence of different executive incentives(including salary incentive and equity incentive) on the motivation of corporate financial asset allocation by our procedures.