统计研究 ›› 2017, Vol. 34 ›› Issue (11): 98-108.doi: 10.19343/j.cnki.11-1302/c.2017.11.009

• • 上一篇    下一篇

贝叶斯视角下符号约束与时变随机波动SVAR模型的实现与应用

苏治等   

  • 出版日期:2017-11-15 发布日期:2017-11-25

The Realization and Application of Sign-restriction and Time-varying Stochastic Volatility SVAR Model under the Bayesian Perspective

Su Zhi Liu Chengcheng Song Zhigang   

  • Online:2017-11-15 Published:2017-11-25

摘要: SVAR模型的传统识别方法更多地涉及等式约束,即通过对模型中的结构参数或脉冲响应函数施加以严格的等式约束,从而实现模型参数的识别与估计,符号约束则是通过将已有经济理论背景转化为先验信息,借助不等式来约束各变量间关系,能有效地避免模型缺乏理论基础的问题;伴随着经济社会改革的不断深化,模型中变量间系数及扰动项的方差—协方差矩阵均随时间而变,表现出明显的时变特征,固定参数已不能有效地刻画经济运行中的动态变化关系,时变随机波动SVAR模型的优势日益凸显;贝叶斯估计的一个重要特征即是将先验信息与样本信息相结合,应用贝叶斯定理得到参数的后验分布以确定最终估计值,参数估计的精度及可信度大幅度提升。基于此,本文系统地阐述了TVP-SV-SVAR模型的建立及贝叶斯估计过程,将其应用于中国货币政策有效性的量化研究,并就Sign-SVAR、TVP-SVAR及TVP-SV-SVAR三种模型的效果进行了对比分析。实证结果进一步发现,结合贝叶斯估计方法的符号约束与时变随机波动SVAR模型在宏观经济政策问题的考察与分析中更具优势。

关键词: 符号约束, 时变参数, 随机波动, 贝叶斯估计

Abstract: The traditional identification methods of SVAR model are more concerned with equality constraints, that is, by means of strict equations to constrain the structural parameters or impulse response functions in the model, so as to implement the identification and the estimation of model parameters, while sign-restriction can avoid the lacking of theoretical basis effectively by transforming the existing economic theory background into a priori information and restricting the relationship between the variables by means of inequality. With the deepening of economic and social reform, the coefficients of variables and the variance―covariance matrix of error terms both change with time, showing obvious temporal characteristics, and the fixed parameters can’t effectively portray the dynamic changing relationships in economy, the advantages of TVP-SV-SVAR models have become increasingly prominent; an important feature of a Bayesian estimation is to combine the priori with the sample information, and to apply the Bayesian theorem to obtain the posterior distribution of the parameters to determine the final estimation. Thus, the accuracy and the reliability of estimation are greatly improved. Based on this, this paper systematically expounds the setup and the Bayesian estimation process of TVP-SV-SVAR model, and applies it to the study of the effectiveness of China’s monetary policy. In addition, it also compares the effectiveness of Sign-SVAR, TVP-SVAR as well as TVP-SV-SVAR model. The empirical results further show that TVP-SV-SVAR models are more advantageous in the analysis of Macroeconomic policy issues.

Key words: Sign-restriction, Time-varying Parameters, Stochastic Volatility, Bayesian Estimation