Real-time And Short-term Forecasting Of Chinese CPI

The CPI prediction is an important task during the period of China’s economic transition. The CPI growth rate prediction not only provides us with judging the future economic situation, but also helps government for macro-control. It is very necessary to analyze the driving factors of inflation, especially about commodity price in imported inflation. Along with the booming of China’s trade dependency, the price is more and more close to international commodity price. Analysis of the international commodity factor is conducive to improve the inflation forecasting accuracy, which not only can make the inflation forecasting more scientific but also control inflation in the perspective of commodities. Therefore, we need to establish a high precision inflation forecasting model and use effective variables to forecast the CPI. The traditional CPI forecasting model based on same frequency data can’t make real-time forecasting within month but only make prediction on outside sample, which will affect the timeliness of prediction ultimately.This article selects the stock index, interest rate, exchange rate and a variety of commodity prices as alternative predicting variables based on high frequency financial data. Firstly, in order to choose the highly correlated variables related to CPI and reduce the complexity of MIDAS model, this paper uses the SSVS method to filter 27 high frequency financial variables to select seven major variables as forecasting variable about the mixed frequency model. Secondly, we predict CPI establishing the uni-variate and multivariate mixed frequency model. And then by comparing the accuracy with OLS and AR model, we get the following conclusion:Firstly, compared with the OLS and AR benchmark model, the mixed frequency model has a comparative advantage(except for MIDAS(m,K,h)). Compared with the uni-variate MIDAS model, the absolute and relative accuracy of multivariate MIDAS model is higher.Secondly, the role of different variables for CPI prediction is different. For in-sample, iron ore price, stock price and oil price effect strongly, but exchange rate effect weakly.Thirdly, the real-time forecasting and short-term forecasting results of multivariate mixing model show that China’s inflation will ease over the next three months, and the price will fall back with deflation risk. In the short baseline-forecasting period, China’s inflation rate is between 1.2%-1.6%.Finally, the international commodity price fluctuation has a significant role in the CPI prediction. The absolute accuracy of all-variables model increased by about 35% compared with model excluding commodity price.