基于格兰杰方法和收敛交叉映射方法的因果分析研究.pdfVIP

摘 要 因果关系是一种超越相关关系的事物间的联系，更像是事物之间的内在规 律. 其已经被广泛应用于神经科学，经济学，气候学，物理和生物学等领域，目 前主流的因果分析方法主要是格兰杰因果分析，它是一种基于自回归模型的分 析方法，多用于检测线性系统的因果关系. 近些年，一种新的分析因果关系的方 法——收敛交叉映射(CCM) 方法也得到了广泛应用，它是一种基于几何学中关 于流形重构的分析方法，相比于格兰杰因果分析方法而言，其理论并不完善，但 是在检测非线性系统的因果关系时却具有显著的优势，本文主要探究这两种因 果关析推理的可靠性，并针对10 个行业指数进行因果网络分析，同时对汇率和 上证指数进行实证分析，来分析汇率市场和股票市场之间的联系. 本文针对格兰杰因果分析和收敛交叉映射因果分析主要做了如下工作： • 提出利用动态时间扭曲(DTW) 相似量来替代交叉收敛映射算法中涉及到 的欧氏距离以及皮尔森相关. • 构造了具有单向因果关系的耦合系统验证了格兰杰因果分析方法和收敛 交叉映射因果方法的合理性. 此外，分别探究了采样间隔对格兰杰因果和收敛交 叉映射因果的影响. • 最后，利用格兰杰因果进行 10 个行业之间的因果网络分析，同时基于格 兰杰因果分析和收敛交叉映射因果分析方法探究汇率市场和股票市场的联系. 关键词: 格兰杰因果分析，收敛交叉映射因果分析，动态时间扭曲，采样间隔， 行业分析 i ABSTRACT Causal analysis is a connection between things that goes beyond correlation anal- ysis, more like the internal laws between things. It has been widely used in the fields of neuroscience, economics, climatology, physics, and biology, etc. At present, the mainstream causal analysis method is Granger causality analysis, which is an analysis method based on autoregressive models, which is mostly used for detection Causality of linear systems. In recent years, a new method of causality analysis, the Convergent Cross Mapping (CCM) method, has also been widely used. It is an analysis method based on manifold reconstruction in geometry. In constrast to Granger method, the theoretical structure is not perfect, but it performs better in detecting the causality of non-linear systems. This dissertation mainly explores the reliability of these two causal analysis reasoning, and targets 10 industries. The index conducts causal network anal- ysis, and conducts relevant empirical analysis on the exchange rate and Shanghai Stock Index to analysis the relationship between the exchange rate market and the stock mar- ket.