时变交互效应面板模型：估计、检验与应用

As China’s economy arrives at the state of new normal, most areas of the economy has experienced deep transition, and significant time-varying features prevail in economic relationship. Due to the imposed restriction of non-time-varying relationship between variables, existing panel data models with interactive fixed effects are limited in real applications with the transition economy. Given this, the project introduces time-varying coefficients of regressors and factor loadings, proposes a time-varying panel data model with fixed interactive effects, and thus develops the corresponding existing models essentially. ..First, this project obtains estimators from the model based on nonparametric kernel methods, and derives properties such as consistency and asymptotic distributions of the estimators. Second, considering asymptotic bias possessed by the estimators, the project obtains the asymptotic unbiased estimators by bias correction----preliminary Monte Carlo simulations approves the method. Third, the project tests the specification of time-varying models, derives the asymptotic distribution of the test statistic, and improves the test’s finite sample performance via bootstrap. Last, based on the time-varying panel data models with interactive fixed effects, the project studies important practical issues such as operation efficiency of state-owned firms, as well as fairness of allocating environment returns. It will look into the time-varying features with transition economy, so as to reflect dynamic effects of policy implementation objectively, and also to provide suggestions with regard to the timing and intensity of government regulation.

随着中国经济发展进入新常态，经济体中大多数领域都在经历着较深刻的转型，经济关系普遍呈现出明显的时变特征。现有交互效应面板模型由于对变量间关系施加了非时变限制，而在转型经济的应用中突显出局限性。基于此，本项目在现有分析框架中引入时变的回归系数和因子载荷，提出时变交互效应面板数据模型，是对相应现有模型的实质性拓展。.首先，本项目基于非参数核函数方法得到模型估计量，并推导估计量的一致性和渐近分布等性质；其次，考虑到估计量存在的渐近偏误，通过偏误纠正方法得到渐近无偏的估计量——初步蒙特卡洛模拟显示方法切实可行；再次，检验模型设定是否存在时变性，推导检验统计量的渐近分布，并借助Bootstrap改进检验的有限样本表现；最后，利用时变交互效应面板模型研究国有企业经营效率和环境收益分配公平性等现实问题，着重考察转型时期经济机制的时变特征，客观地反映政策实施的动态效果，为政府调控的时点和力度选择提供建议。

该课题旨在具有因子结构和时变特征的面板模型中构造新的计量方法，具体体现在：首先，现有文献对于面板Pobit模型的估计虽然考虑了个体固定效应，却忽略了个体效应的潜在时变特征，该课题通过因子结构（交互效应）的引入，可以更为一般性地允许个体时变效应的存在，并且在即使时间维度固定的情形下，仍能够对平均边际效应等重要因果变量估计和准确推断，从而发展了微观计量中重要的应用工具；其次，在宏观和金融实证预测研究中，基于扩散指数的模型因其降维的有效性成为最有影响力的预测工具之一。然而在现有扩散指数模型预测框架下，无论因子载荷还是预测模型的系数均被限定为非时变参数，预测结果无法很好地解释现实经济中的结构变化。该课题打破了传统框架下对于参数的时变限制，允许因子载荷与回归系数均具有时变特征，并通过对于二者分别的非参数估计得到最终稳健的预测结果，实证对比发现该方法对主要宏观变量的预测结果具有显著的优势。进一步地完善基于时变因子増广模型回归和预测的理论框架，提出分布理论，从而有助于现实检验推断和预测区间的构造；最后，对未知因子个数的准确估计是所有基于因子结构模型使用的必要条件，然而现有的主流估计方法在样本容量相对较小时，精确度明显不足，该课题提出了基于双重交叉验证这一数据驱动方法估计因子个数的新思路。我们证明了该方法决定因子个数的一致性，并且在有限样本和实证应用中发现该方法的明显优势，这为因子模型的设定提供了重要的参考。