项目名称: 含有变点的分位数回归模型:贝叶斯分析及应用
项目编号: No.71301099
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 管理科学
项目作者: 周影辉
作者单位: 上海大学
项目金额: 20.5万元
中文摘要: 变点模型因具有模型上的灵活性和应用上的广泛性,自提出以来即引起了众多计量经济学家的高度关注,目前仍是国内外计量经济学研究中的一个重要方向。关于变点问题的现有研究大都集中于线性回归模型中因变量的条件均值或条件方差中的变点,但考察其条件分位数中的变点在很多情况下更有意义。在分位数回归模型中考虑变点的现有研究基本上都是依赖于理论证明比较复杂、计算效率比较低的传统方法,而用较为简单有效的贝叶斯方法进行研究的尚不多见。为此,本项目将运用马尔科夫链蒙特卡罗(MCMC)抽样技术对以下三个问题进行研究:(一)含有变点的分位数回归模型的贝叶斯统计分析;(二)可看作变点模型的门限分位数自回归模型的贝叶斯统计推断;(三)含有变点的分位数回归模型和门限分位数自回归模型在我国宏观经济数据(如国内生产总值的季度数据)和金融数据(如上证指数的对数收益率数据)上的应用。
中文关键词: 分位数回归;变点;厚尾数据;贝叶斯方法;
英文摘要: As change-point models have high flexibility in modeling and universality in application, they have received a great deal attention by many econometrist since proposed. Now, they are still the important research topics in econometric. Most existing studies about change-point problems focus solely on structural changes in the conditional mean or conditional variance of response in the linear regression model, howerer, it may be of key importance to investigate structural changes in its conditional quantiles under many circumstances. The existing literatures on quantile regression with change points rely basiclly on the classical method with complicated theories and unefficiently computational algorithms. The related studies by using the relatively simple and efficient Bayesian method are few. Hence,this project will study the following three issues on the bases of Markov chain Monte Carlo (MCMC) sampling techniques: (1) Bayesian statistical analysis of quantile regression models with change points; (2) Bayesian statistical inference of threshold quantile autoregressive models, which can be regarded as change-point models; (3) Application of quantile regression models with change points and threshold quantile autoregressive models to macroeconomic data (for example, the quarterly data of the gross demestic product
英文关键词: Quantile regression;Change points;Heavy-tailed data;Bayesian method;