Abstract: The traditional VAR model limits the variables to the same frequency, which destroys integrity of the original data, thus affecting performance of the model. In this study, MF-BVRA model, which well integrates quarterly and monthly variables, is introduced. The recursive samples of GDP, PPI, CPI and PMI are divided into three of different quarterly information groups. The performance of MF-BVAR is compared with that of MIDAS and co-frequency model. The transmission mechanism among variables is analyzed by mixing Granger Causal Relation Test, impulse response analysis and variance decomposition analysis. The empirical results show that: 1) compared with QF-BVAR, MF-BVAR’s real-time prediction errors in PPI, CPI, PMI and GDP are about 70%, 80%, 75% and 20% lower respectively; 2) real-time and short-term prediction performance of MIDAS and MF-BVAR, quantitative and statistical tests tell, are significantly better than that of QF-BVAR, and the latter does better in real-time forecasting, while MIDAS is good for short-term forecasting; the more monthly information is fully used, the smaller the prediction error will be, the recursive samples tell; (3) there is a bidirectional causal relationship between GDP and CPI, PPI and PMI, CPI and PMI, and the conduction time is different; while the forward conduction duration of PMI and GDP, PPI and GDP, PPI and CPI are all three periods, and the reverse conduction path is blocked without an asymmetry.

Key words: MF-BVAR Model, mixing data of China’s Macro economy, real-time prediction, PMI, PPI, CPI