基于混频大数据的宏观经济总量实时预测研究

A Study on Real-time Forecast of Macroeconomic Aggregate Based on Mixed-frequency Big Data

号脉中国经济高质量发展的主旋律亟需宏观经济的实时预测。本文基于2012年1月1日至2019年6月30日的混频大数据,利用动态因子模型构造高频舆情指数,拓展现有低频(至多到月度)预测模型,提出中国宏观经济总量实时预测的监测系统。研究发现:①高频舆情指数能够显著地反映宏观经济的基本面,对GDP增长率具有稳定和有效的解释力;②实时预测模型相比于基准预测模型既能提升预测精度,又能保证预测的稳健性,同时在样本外预测中,当基准预测期和滞后阶数取3的整数倍时预测效果最好;③实时预测模型的多步滚动向前动态预测具有现时性,在同等预测精度下,预测时间至少比AR、OLS等基准模型分别提前18天和30天。

The main theme of the high-quality development of the Chinese economy urgently requires real-time macroeconomic forecast. Based on the mixed-frequency big data from January 1, 2012 to June 30, 2019, this paper uses a dynamic factor model to construct a high-frequency public opinion index, and extends the existing low-frequency(up to monthly) forecast model, and propose a monitoring system for real-time forecast of China’s macroeconomic aggregates. The study finds that: ①the high-frequency public opinion index can significantly reflect the macroeconomic fundamentals, and has a stable and effective explanatory power for GDP growth rate;②the real-time forecast model can improve forecast accuracy and ensure forecast compared with the benchmark forecast model at the same time, in the out-of-sample forecast, the forecast effect is the best when the base forecast period and the lag order are taken as integer multiples of 3;③The multi-step rolling forward dynamic forecast of the real-time forecast model is current and has the same forecast accuracy. Next, the forecast time is at least 18 days and 30 days earlier than the benchmark models such as AR and OLS.