基于时变Copula的房地产业与银行业尾部动态相关性研究  

Analysis of Dynamic Tail Dependence between Real Estate and Banking Based on Time-varying Copula

作　　者：江红莉[1] 何建敏[1] 庄亚明[1] 

机构地区：[1]东南大学经管学院,江苏南京211189

出　　处：《管理工程学报》2013年第3期53-59,共7页Journal of Industrial Engineering and Engineering Management

基　　金：国家自然科学基金资助项目(71071034)

摘　　要：文献多数从线性相关和静态相关的视角研究房地产业与银行业(金融业)的相关性,本文基于大智慧"板块指数"中的"房地产"(代表房地产业)和"银行类"(代表银行业)数据,采用GARCH、EVT和时变Copula模型研究房地产业和银行业的动态尾部相关性。实证结果表明,时变Copula模型刻画尾部相关性的效果优于静态模型,时变SJC-Copula的建模效果相对最好;样本考察期内,上、下尾相关系数均为正,分别落在区间[0.2287,0.6146]和[0.4666,0.7143]内,且下尾相关系数均大于上尾相关系数,说明市场低迷时,房地产业和银行业易产生共生风险,并且低迷时的相关性强于活跃时的相依性;上、下尾相关系数均具有自相关性、ARCH效应。分析月均下尾相关系数发现,房地产业和银行业的下尾相关性具有"政策效应":下尾相关系数相对较低的月份往往伴有严厉的房地产调控政策出台或之前的月份出台了调控政策,意味着房地产调控政策将弱化房地产业和银行业的下尾相关性。未来应继续执行或出台更严厉的调控政策,降低银行业与房地产业的共生风险,促进房地产业的健康发展。Real estate and banks play essential roles for economic development. The relationship between real estate and banks is becoming closer along with the development of the real estate industry in China. The current literature analyzes the relationship between real estate and banks from the perspectives of linear correlation and state correlation. These studies cannot reflect the non-linear and dynamic correlation between real estate and banks. The Time-varying Copula model can be used to calctilate dynamic correlations and dynamic tail dependence. We applied the GARCH model and Extreme Value Theory （EVT） to estimate the parameters of marginal distribution, and apply time-varying Copula functions to measure dynamic tail dependences between real estate and banking. The real estate and banking indices, dated from August 30, 2006 to May 12, 2011, and collected from the ＂plate index＂ in the Great Wind database, were used as the sample data. A total of 1140 data sets were used for analysis. In the first part, we estimate the marginal distribution of real estate and banks by using the GARCH model and Extreme Value Theory （ EVT）. The Kurtosis statistics and the JB test show that the revenue series of real estate and banks aren＇t normally distributed. The ARCH test result shows that these two series are heteroscedastic. It is suitable to use the GARCH to model the marginal distribution of real estate and banks. We adopt the maximum likelihood estimation method （MLE） to estimate parameters of GARCH （1, 1） -N, GARCH （1, 1） -T and GARCH （1, 1） -GED models. Our finding shows that GARCH （1, 1） -T is the most suitable for comparing the log-likelihood values. EVT only models the distribution of tail, which is helpful to depict fat tail characteristics. We employ GPD （Generalized Pareto distribution） to model the tail distribution of the standard residual sequence which is filtered by GARCH. The tail of underlying distribution indicates that GPD appears to fit the distribution fairly well. In

关 键 词：时变COPULA GARCH 极值理论 尾部相关性 房地产业 银行业 

分 类 号：F840.32[经济管理—保险]