关键词: 混频数据, MIDAS模型, 居民消费价格指数, 短期预测

Abstract: In existing literatures CPI forecasting is usually based on monthly data. Daily observations from financial markets should be aggregated into monthly data, which will probably loss the intra-month information implied in them. To incorporate these information into CPI dynamics, in this paper we use a mixed-frequency data sampling approach to investigate the predictive information in the returns (first-order moments) and volatilities (second-order moments) of financial variables. Our results show that stock market, short-term interest rates and interest spreads changes only affect CPI at return levels, while both returns and volatilities of long-term interest rates, grain and energy commodities are related with short-term forecasting of CPI. And the influence of returns is more persistent than the influence of volatilities. Compared with monthly-data based CPI forecasting method, our mixed-data sampling model has better in-sample explanatory and out-of-sample predictive performance. Besides, the second-order volatility information is also useful to reduce forecasting errors.

Key words: Mixed-frequency Data, MIDAS Model, CPI, Short-Term Forecasting