中文摘要：

模型和参数的不确定性以及信息的综合有效利用是影响宏观变量预测精度的主要因素。本文运用贝叶斯模型平均（BMA）方法建模并对样本外通胀进行预测，综合备选模型及变量的信息，以控制模型不确定性，并有效利用丰富的宏观数据信息。本文选取28个解释变量构建了含有2”个单一线性模型的集合，实证上采用了马尔科夫链蒙特卡洛模型综合算法（MC3）对备选模型进行抽签，抽签次数为1000万次。采用中国宏观数据的实证结果表明，通胀一阶滞后项与工业企业增加值增速作为预测因子几乎被选择在所有预测模型中；对于通胀的样本内拟合，贝叶斯模型平均（BMA）方法优于单一模型；对于样本外预测，在RMSE标准下，贝叶斯模型平均方法的预测能力优于较为流行的AR模型、主成分分析模型、菲利普斯曲线模型、利率期限结构模型、单一最优模型和五变量模型。

英文摘要：

This paper focuses on the modeling and forecasting of inflation out - of sample in China using Bayes- ian Model Averaging method to reduce the problem of model uncertainty and improving information efficiency. The general framework is a simple linear regression model using a large set of potential indicators, comprising 28 monthly time series covering a wide spectrum of Chinese economic indicators. We use the Markov chain Monte Carlo Model Combination （ MC3 ） method to choose the model which can best forecast inflation. The re- sult show that the BMA method do better than a wide range of popular models such as the AR model, the model using principal components, Philips Curve model, term structure model, the single best linear model and the five factor model.