基于动态模型平均的中国通货膨胀实时预测

Real-time Forecasting of Inflation in China Based on Dynamical Model Averaging

本文研究了动态模型平均方法 (DMA)及其参数估计。DMA方法允许方程所含变量、变量系数及模型所含方程同时变动,适用于对宏观经济指标进行实时预测。本文利用DMA对中国通货膨胀进行实时预测表明,DMA方法下的中国通货膨胀预测解释变量处于0～3;以CPI指数和GDP平减指数作为通货膨胀衡量指标的情况下,不同预测期的解释变量被包含概率是时变的;遗忘因子为0.95时,利用DMA方法对我国通货膨胀的预测效果最佳,优于贝叶斯模型平均和时变向量自回归模型。

In this paper, we describe the method of dynamic model averaging and how to estimate its parameters. It not only allows variables to change over nine but also allows coefficients of equation and the number of equations in the model to change at the same time. Using this method, we forecast China＇s inflation in real time. The results show that the dependent variables almost betwean 0 and 3. In different forecast periods, the probability of including dependent variables varies over time. When the forget factor is 0.95, the forecast performance under the method of DMA is better than the Bayesian Model Averaging and the time vary vector models.