半参数面板数据模型的估计、检验及模型选择

发布时间：2018-05-29 04:12

本文选题：半参数面板数据模型 + 随机效应　；参考：《北京化工大学》2016年硕士论文

【摘要】：面板数据(Panel Data)是时点、个体两个维度的数据呈现形式,相对于横截面数据和时间序列优点突出,在应用实践中发挥了很大作用。半参数面板数据模型,综合了参数和非参数模型的优点,同时又避免了参数模型的“限制性强”与非参数模型的“维数灾难”的问题,近年来得到广泛关注。本文介绍了半参数面板数据模型中带有个体效应的部分线性模型和变系数模型的估计问题,采用截面最小二乘方法和修正的局部常数最小二乘方法对模型中的未知量进行了求解。我们知道个体固定效应和个体随机效应下,模型的估计方法和所得结果都有所不同。为了确定个体效应,提高估计、预报的精确度,本文针对部分线性面板数据模型,提出了参数Hausman检验和非参数Hausman检验方法。通过证明可知,在原假设(个体随机效应)成立时,参数Hausman检验统计量渐近服从卡方(χ2)分布,非参数Hausman检验统计量渐近服从正态分布。Monte Carlo模拟结果显示,参数Hausman检验在模拟中表现良好,且相比于非参数Hausman检验稳健性和可靠性更高。进一步,我们在地区生产总值影响因素的实证分析中,运用本文提出的估计方法和参数Hausman检验方法完成了数据的统计推断。通过数值模拟结果我们发现,在小样本下Hausman检验统计量的分布存在不确定性,为了避免可能造成的错误判断,本文引入Bootstrap抽样方法,构造了参数Bootstrap-Hausman检验统计量以及非参数Bootstrap-Hausman检验统计量,先求得统计量的分位点后再构造假设检验的拒绝域。模拟结果显示,小样本下,两种检验方法均能识别出个体效应,但参数Bootstrap-Hausman方法的稳健性和可靠性更高。最后,将参数Bootstrap-Hausman检验应用到经济增长与居民消费关系的实证研究中。
[Abstract]:Panel data (Panel data) is a time point, the data presentation form of two dimensions of individuals, compared with cross-section data and time series, has outstanding advantages, and has played a great role in application practice. The semi-parametric panel data model, which combines the advantages of parametric and non-parametric models and avoids the problem of "restrictive" and "dimensionality disaster" of parametric models, has been paid more and more attention in recent years. In this paper, the problem of estimating partial linear model and variable coefficient model with individual effect in semi-parametric panel data model is introduced. The unknowns in the model are solved by using the cross-section least square method and the modified local constant least squares method. We know that the estimation method and the results of the model are different under individual fixed effect and individual random effect. In order to determine individual effect and improve the accuracy of estimation and prediction, this paper presents parametric Hausman test and nonparametric Hausman test for partial linear panel data model. It is proved that when the original hypothesis (individual random effect) is established, the parameter Hausman test statistic is asymptotically distributed from chi-square (蠂 ~ 2), and the non-parametric Hausman test statistic is asymptotically obeyed from the normal distribution. Monte Carlo simulation results show that, The parametric Hausman test performs well in the simulation and is more robust and reliable than the non-parametric Hausman test. Furthermore, in the empirical analysis of the influencing factors of regional GDP, we use the estimation method and the parameter Hausman test method proposed in this paper to complete the statistical inference of the data. By numerical simulation, we find that there is uncertainty in the distribution of Hausman test statistics in small samples. In order to avoid possible misjudgment, Bootstrap sampling method is introduced in this paper. The parameter Bootstrap-Hausman test statistics and the nonparametric Bootstrap-Hausman test statistics are constructed. The sub-sites of the statistics are obtained first and then the rejection domain of the hypothesis test is constructed. The simulation results show that both methods can identify individual effects in small samples, but the parametric Bootstrap-Hausman method is more robust and reliable. Finally, the parametric Bootstrap-Hausman test is applied to the empirical study of the relationship between economic growth and consumption.
【学位授予单位】：北京化工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：O212.1

【参考文献】