中国经济增长与通胀的混频预测--基于Sims-Zha先验分布的BVAR模型

Forecasting of China's Economic Growth and Inflation with Mixed-Frequency Data--Based on BVAR Models with Sims-Zha Priori Distribution

Sims-Zha先验分布使用两组虚拟变量提取了时间序列中单位根和协整的先验信息,将GLP等主流模型一般化为其特例。本文使用GDP与CPI的环比和累计同比混频数据,参考混频数据和超参数选择的研究成果,根据中国经济特点进行推导和测试后建立了Sims-Zha先验分布的贝叶斯向量自回归模型,并通过对预测结果的RMSE数值与国际上关于此问题的其他流行模型预测结果的RMSE值进行比较。研究发现:当GDP的原始数据不平稳需要进行一阶差分处理时,设定没有截距项的Sims-Zha先验分布的贝叶斯向量自回归模型,预测效果比原模型更优,当GDP的原始数据平稳时,则Sims-Zha先验分布的贝叶斯向量自回归模型比原模型预测效果更优,有无外生变量对模型预测效果的影响不明显,无外生变量累计同比增长率预测模型的预测效果最优;相比常用模型Sims-Zha先验分布下的贝叶斯向量自回归模型在GDP上的短期预测效果更精准,在CPI的短期预测上预测效果差于GLP模型,但是优于其他模型.

Sims-Zha prior consists of two sets of dummy observations from the data to capture prior beliefs about unit roots(not just random walk)and cointegration in the time series,and it makes most popular models such as GLP model as the special type of Sims-Zha model.We use Bayesian vector autoregressive model with Sims-zha priori distribution to model China's growth and inflation with mixed-frequency data of monthly CPI time series data and quarterly GDP time series data,and use the RMSE value of the prediction results to compare the predicting models.According to empirical evidence,when the raw data of GDP is not stable and requires first-order differential processing,it is assumed that the Bayesian vector autoregressive model with Sims-zha priori distribution has no intercept entry,and the.prediction ffect is better than the original model.When the raw data of GDP is stable,The original model of Bayesian vector autoregressive model with Sims-zha priori distribution has bet ter prediction fect.And the influence of exogenous variables on the prediction ffect of the model is not obvious.The prediction ffect of the forecast model of cumulative yearon-year growth rate wit hout exogenous variables is optimal.Compared with other models the Bayesian vector autoregressive model under S ims-zha priori distribution has a more accurate short-term prediction effect on GDP.And the prediction effect is worse than the GLP model on the short-term forecast of CPL,but better than other models.