路径相关性研究：经验证据和模拟检验

发布时间：2018-07-09 10:03

本文选题：相关性研究 + 非参数模型　；参考：《浙江工商大学》2017年硕士论文

【摘要】：相关性分析是研究数据之间关系的一种方法,是变量随机分析的一个重要课题,而相关性分析的结果能够为发掘数据背后的信息提供有力地支持。从应用角度来看,现在金融保险等领域的投资风控、信贷评估,网络及APP领域的信息推送等等均和相关性分析有着紧密联系。早先学者们对相关性进行了研究,提出了许多度量变量相关性的方法,但这些相关性研究主要关注变量之间相关程度的分析,而对于变量之间相关模式的识别与研究并不是很重视。现有的一些相关系数,如Pearson相关系数,能对变量间的相关关系进行度量但不能对变量的相关模式进行识别,另一些相关系数,如Kendall相关系数、Spearman相关系数等,虽然可以一定程度上反映变量之间的相关关系,但仅仅片面地刻画了变量之间的相关结构。数据时代的到来给变量之间的相关性研究带来了挑战。从理论上来看,多个变量之间的相关性关系非常复杂,对于高维的数据更是如此。随着研究的深入,有学者发现原有的一些研究假设并不成立,这些不恰当的假设可能会导致严重的后果。本文受许冰(2010)路径设计的启发,并借鉴近来的一些研究成果,通过构建路径模型体系,综合考察变量间的相关模式及相关性的度量,进而对变量进行路径相关性分析,为变量间的相关性分析提供一种新的方法。本文使用Li and Racine(2004)的非参数变量筛选方法,对有关变量进行了分类;基于变量筛选结果构建非参数路径模型体系,分析变量间的整体效应、直接效应和间接效应。发现:(1)不管是在基准模型还是路径模型中,非线性分量的占比大于线性分量的占比,线性分量的波动大于非线性分量的波动,且非线性变量在模型中占主导地位;(2)单路径变量中用电量的整体效应最大,双路径变量中用电量和已用授信额度的整体效应最大;(3)用基准模型的外推精度取代变量间的因果分析,对具体数据进行了模拟分析。
[Abstract]:Correlation analysis is a method to study the relationship between data. It is an important subject of variable random analysis, and the result of correlation analysis can provide strong support for the information behind the data. From the perspective of application, the investment wind control, credit assessment, network and APP field of investment in financial insurance and other fields are pushed forward. There is a close relationship between equality and correlation analysis. Earlier scholars have studied the correlation and put forward a number of methods to measure the correlation of variables, but these correlation studies mainly focus on the analysis of the correlation between variables, but not much attention is paid to the identification and research of correlation patterns between variables. A number, such as the correlation coefficient of Pearson, can measure the correlation between variables but can not identify the correlation patterns of variables. Other correlation coefficients, such as Kendall correlation coefficient and Spearman correlation coefficient, can reflect the correlation between variables to a certain extent, but only one-sided depicts the correlation structure between variables. The arrival of the data age challenges the correlation between variables. In theory, the correlation between multiple variables is very complex, and it is more so for high dimensional data. As the research goes deep, some scholars have found that some of the original hypotheses are not established, and these inappropriate hypotheses may lead to serious problems. This article is inspired by the design of Xu ice (2010) path, and draws on some recent research results. Through the construction of the path model system, this paper comprehensively investigates the correlation patterns and correlation between variables, and then carries out path correlation analysis on variables, and provides a new method for correlation analysis among variables. This paper uses Li and Racine (2004) the non parametric variable selection method is used to classify the related variables; based on the variable screening results, the non parametric path model system is constructed, and the overall effect, the direct effect and the indirect effect between the variables are analyzed. (1) the proportion of nonlinear components is greater than the proportion of the linear component in the reference model or the path model. The fluctuation of component is greater than the fluctuation of nonlinear component, and the nonlinear variable is dominant in the model. (2) the overall effect of electricity consumption in single path variable is the largest, and the total effect of electricity consumption and credit line is the largest in the double path variable. (3) the effect analysis is replaced by the extrapolation precision of the reference model, and the specific data are carried out. The simulation analysis.
【学位授予单位】：浙江工商大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：C81

【参考文献】