计量经济模型中岭回归参数的估计研究

发布时间：2018-11-19 12:21

【摘要】：在计量经济学中,建立多元线性回归模型时,经常会出现模型所选取的变量之间存在多重共线性的情况,这可能导致建立的模型得到的参数不稳定,模型不能充分地解释所要研究的问题。由于模型存在共线性的变量,线性回归模型不满足计量经济学经典的假设条件,此时如何有效的处理模型变量之间的共线性成为模型面临的重要问题之一。而常用的解决模型变量之间多重共线性的方法都存在一定的缺陷,即不能很好的利用模型变量信息。如何能更好的充分利用模型变量的有用信息,不损失变量之间的内在联系去解决变量之间的共线性关系,岭回归就是在这样的背景下产生的。岭回归可以更好地保留变量的内在信息,不破坏它们之间的内涵关系,并且能够更好的抓住主要矛盾,去分析解决问题,进而是解决模型多重共线性的一个重要途径。因此,文章对岭回归问题进行研究具有理论意义与应用价值。在建模过程中使用岭回归方法时,计算岭参数是重要的关键问题。因此文章在研究岭参数计算上做了深入研究。首先,回顾了与岭回归相关的理论,从减小模型变量方差扩大因子角度出发,结合生物工程领域的遗传算法,计算最优的岭参数,然后根据得到的岭参数建立岭回归模型。其次,文章通过蒙特卡洛模拟方法对其进行了模拟实验。实验结果表明,基于遗传算法确定岭参数的岭估计可以有较好的估计特性。最后,文章采用所提出的方法,进行了关于我国卫生总费用问题的实证分析,得到了一些有价值的结论并给出一些相关的政策建议。文章通过遗传算法与方差扩大因子法相结合计算了最优岭参数,得到的模型参数更加稳定。这样计算岭参数一方面剔除了岭参数在确定过程中掺杂的人为主观性,另一方面把变量由于多重共线性导致其不合理的方差扩大因子控制在合理的区间内,稳定了模型的估计参数,并且提高了参数的估计精度。文章的研究工作为岭回归选取岭参数提供了一种新的选取方法,可以使得岭回归模型更好的应用到实际经济问题当中。其次,利用研究的成果,研究了影响我国卫生总费用变量之间的多重共线性问题,得出了结论并给出了有价值的政策意见。
[Abstract]:In econometrics, in the establishment of multivariate linear regression models, multiple collinearity often occurs between the variables selected by the model, which may lead to the instability of the parameters obtained by the established models. The model can not fully explain the problem to be studied. Because there are co-linear variables in the model, the linear regression model does not meet the classical econometric assumptions. How to deal with the collinearity between the model variables effectively becomes one of the important problems facing the model. However, the common methods to solve the multiple collinearity between model variables have some defects, that is, they can not make good use of model variables information. How to make full use of the useful information of model variables and not lose the internal relations between variables to solve the collinear relationship between variables, Ridge regression is produced in such a background. Ridge regression can better retain the internal information of variables, do not destroy the connotative relationship between them, and can better grasp the main contradiction to analyze and solve the problem, and then it is an important way to solve the multiple collinearity of the model. Therefore, the study of Ridge regression is of theoretical significance and practical value. When using ridge regression method in modeling, calculating ridge parameters is an important key problem. Therefore, the paper makes a deep study on the calculation of ridge parameters. Firstly, the theory related to ridge regression is reviewed, and the optimal ridge parameter is calculated from the angle of reducing the extended factor of variance of model variables and combining with genetic algorithm in the field of bioengineering, and then the ridge regression model is established according to the obtained ridge parameters. Secondly, the paper carries on the simulation experiment through the Monte Carlo simulation method. The experimental results show that the ridge estimation based on genetic algorithm can have better estimation characteristics. Finally, by using the proposed method, the paper makes an empirical analysis on the total health expenditure in China, obtains some valuable conclusions and gives some relevant policy suggestions. In this paper, the optimal ridge parameters are calculated by combining genetic algorithm with variance expansion factor method, and the model parameters are more stable. On the one hand, the artificial subjectivity of the ridge parameter is eliminated in the determination of the ridge parameter, on the other hand, the unreasonable variance expansion factor of the variable is controlled in a reasonable range due to multiple collinearity. The estimation parameters of the model are stabilized and the estimation accuracy of the parameters is improved. The research work in this paper provides a new method for selecting ridge parameters by ridge regression, which can make the ridge regression model better applied to practical economic problems. Secondly, by using the results of the research, this paper studies the multiple collinearity problem affecting the variables of the total health expenditure in China, and draws a conclusion and gives some valuable policy suggestions.
【学位授予单位】：天津财经大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：F224

【相似文献】