基于传统金融理论视角构建的ARMA和ARCH族时间序列模型在金融实证分析中发挥着重要的作用，但应用它们去实证分析行为金融理论问题时出现了推断的偏差，导致的主要原因在于心理、行为等潜变量的测量误差被忽略了。因此本项目将潜变量分析与时间序列分析的思想相结合，拟通过构建含潜变量的ARMA模型和ARCH-M族模型（含参数与非参数形式）去解决行为金融实证建模的问题。项目主要采用非参数贝叶斯技术去完成模型的估计和检验，并编写相应的算法实现数值模拟，以检验统计推断的可行性，同时考察模型应用于真实市场时的实用性。研究结果可以为行为金融实证研究提供新的定量工具与对应的统计理论基础。

ARMA and a family of ARCH models constructed through the perspective of traditional financial theory play an important part in empirical analysis. Unfortunately, some inference biases always appear when these models are applied to analyzing behavioral finance problems. That is because the potential measurement error of psychological variable is always ignored in the modeling process. Combining the idea of latent variable analysis and time series analysis, this project is designed to solve the problems of empirical modeling in behavioral finance by constructing the ARMA and a family of ARCH-M models with a latent variable (including the parametric and non-parametric modeling). Nonparametric Bayesian method is employed for estimation and test, relevant algorism is complied and numerical simulation is realized for proving the feasibility of statistical inference. The project also examines whether models practically apply to real market. The research results will provide a new quantitative tools and corresponding theory base for empirical research in behavioral finance.