摘要
文章通过构建月度景气指标与季度实际GDP增长率之间的混频动态向量自回归模型,并采用期望最大值算法和卡尔曼滤波来实现混频数据和缺失数据的估计和迭代预测。大量月度景气指标的MFVAR模型的伪实时数据的多步滚动迭代样本外预测结果表明:虽然不同类别的月度景气变量在不同预测期的预测结果存在一定的差异,但实时预报、短期预测,以及组合预测结果均表明混频动态向量自回归预测模型对我国季度实际GDP增长率的实时预报和短期预测具有精确性、有效性与适用性。
By use of monthly climate indicators and real GDP growth rate to build a dynamic mixed-frequency vector autore- gressive (MFVAR) model, and by utilizing expectation maximum algorithm and Kalman filter, this paper realizes the estimation and iterative prediction on mixed data and missing data. The paper also employs MFVAR model with a large number of monthly climate indicators to do an iterative multi-step out-of-sample forecasting of pseudo real-time data. The result shows that the pre- diction result of different kinds of monthly climate variable is slightly different in different forecasting period, but the result of re- al-time forecast, short-term predication and combined forecasts all indicate that the dynamic MFVAR prediction model has accu-racy, validity and applicability in real-time forecasting and short-term prediction on Chinese quarterly real GDP growth rate.
作者
刘汉
刘营
王永莲
Liu Han Liu Ying Wang Yonglian(Business School, Jilin University, Changchun 130012, China School of Statistics, Jilin University of Finance and Economics, Changchun 130117, China)
出处
《统计与决策》
CSSCI
北大核心
2017年第21期29-33,共5页
Statistics & Decision
基金
全国统计科学研究项目(2017LD01)
教育部人文社会科学研究青年基金项目(15YJC790055)
教育部人文社会科学重点研究基地重大项目(16JJD790014)
关键词
混频数据
MFVAR模型
GDP增长率
实时预报
短期预测
mixed-frequency data
MFVAR model
GDP growth rate
real-time prediction
short-term forecasting