摘要
本文在实时数据基础上选取金融变量作为预测因子并通过混频数据抽样(MIDAS)模型对GDP增长率进行短期预测。结果表明:短期预测时MIDAS模型预测效果甚佳而且嵌入自回归项的MIDAS模型明显降低预测误差;数据修正对MIDAS模型的预测精度有负面影响;货币供应量等预测因子在包含自回归项MIDAS模型中预测精度较高,投资和出口依旧是拉动我国经济增长的重要因素;SPA检验及组合MIDAS模型的较好预测精度说明组合MIDAS模型预测能力占优。
The paper employs mixed frequency data sampling (MIDAS) model to nowcast and forecast GDP with financial variables in real time data. We find that : ( 1 ) The prediction of MIDAS model is very well in short- term and the model with autoregressive term significantly reduce the forecast error; (2)Data revision have a negative impact on accuracy of prediction with the MIDAS model; (3)Money supply and other predictors have the higher accuracy of prediction in MIDAS - AR model, which means investment and exports are still two important factors driving China's economic growth ; and (4) SPA test proves the combination of MIDAS model can better improve the accuracy of prediction that means the prediction power of combination of MIDAS model is dominant.
出处
《金融研究》
CSSCI
北大核心
2013年第9期16-29,共14页
Journal of Financial Research
基金
国家自然科学基金项目"货币政策规则非线性的理论模型与计量研究"(71001087)
"状态空间混频模型及其在宏观经济中的应用"(71371160)
福建省教育厅项目"我国经济波动与经济政策的DSGE计量建模及应用研究"(2009100074)
国家留学基金委公派访问学者(含博士后研究)项目(201208350111)资助
关键词
金融变量
GDP增长率
MIDAS模型
混频预测
数据修正
Financial variable, GDP growth rate, MIDAS model, Mixed -frequency forecasting, Data revision