中国通货膨胀动态模型预测的实证研究  

Empirical Research on Chinese Inflation Forecast Using Dynamic Model

作　　者：郭永济[1] 丁慧[1] 范从来[1] 

机构地区：[1]南京大学商学院

出　　处：《中国经济问题》2015年第5期3-15,共13页China Economic Studies

基　　金：教育部长江学者和创新团队发展计划资助项目"经济转型背景下稳定物价的货币政策"(IRT13020);教育部人文社会科学重点研究基地重大项目(14JJD790006);江苏省普通高校博士研究生科研创新计划资助项目(KYLX_0004)的资助

摘　　要：通货膨胀形成机制的复杂性以及影响因素的多元性,要求通货膨胀预测过程必须关注模型与参数的不确定性以及信息的综合有效利用。本文构建了基于动态模型平均的时变向量自回归模型预测我国的通货膨胀,模型根据预测表现动态选择解释变量、系数时变程度和模型维度,在有效控制模型和参数不确定性的同时最大限度地综合利用宏观经济信息。基于我国20个宏观经济变量的实证结果表明,同时考虑解释变量动态选择、系数时变程度动态选择和模型维度动态选择的通胀预测模型,其预测表现优于单一维度的随机游走模型、时变向量自回归模型、贝叶斯模型平均模型,特别在经济波动较大时综合考虑这些因素的通胀预测模型表现更加出色,增加不同维度的子模型也提高了经济波动增大时的通胀预测能力。Inflation in complexity of the formation mechanism and diversity of influencing factors requires the forecast process of inflation must focus on the model and parameter uncertainty,and the comprehensive and effective use of information.This paper constructs the time- varying parameters vector auto- regressive model based on dynamic model average（ TVP-VAR- DMA） to forecast China＇s inflation. This model can dynamically select variables,parameters and model dimensions in line with the overall performance of inflation forecast,as well as maximizing the utilization of macroeconomic information in effective control the model and parameter uncertainty. Based on 20 macroeconomic variables of empirical results indicate that,the inflation forecast model which at the same time to consider explained variable dynamic select,coefficient time- varying and model dimensions dynamic select is superior to single dimension of Random walk model（ RW）,Time- varying vector auto-regressive model（ TVP- VAR）,and Bayesian model average model（ BMA） on forecast performance. Integrated consider these factors make forecast performance more excellent,especially in the case of economic fluctuation is bigger,and add different dimensions of sub- model increases the ability of forecast under the conditions of economic volatility increased.

关 键 词：通货膨胀预测 动态模型平均 遗忘因子 TVP-VAR-DMA 

分 类 号：F822.5[经济管理—财政学]