基于混频回归类模型对中国季度GDP的预报方法研究

Short-term Prediction of Quarterly GDP in China Based on MIDAS Regression Models

根据混频数据计量经济模型的建模理论和分析技术,本文构建了中国季度GDP 5种不同权重函数的混频数据回归预测模型(MIDAS)和非限制MIDAS模型。结合传统分布滞后模型推导出MIDAS模型的最小二乘识别方法,并在此基础上对中国季度GDP进行短期预报,分析了高频解释变量滞后阶数变化效应及其对低频变量GDP的影响效应。根据6种模型拟合及预测结果,进一步构建混频回归联合预测模型,并考察了混频回归联合预测模型的预测精度及预测效果。研究结果表明:非限制混频数据回归预测模型的预测精度及拟合效果高于5种不同权重MIDAS模型,以BIC为权重构建的混频联合预测模型在对我国季度GDP短期预报时表现最优。

From the perspective of mixed frequency data＇s econometric models which are the theory and analytical techniques, this paper builds five different weight＇s functions of MIDAS and unrestricted MIDAS. Combined with the traditional estimation of autoregressive distributed lag model, the ordinary least squares estimation method of MIDAS is given out. Base on the estimation, we forecast the quarterly GDP in China, and analys the effects of high frequency explanatory lag order changes and its influence of low frequency variable in lag length on GDP forecasting. According to the results of the six MIDAS models the paper furthely builds the combined MIDAS model, and investigats the prediction accuracy and pre- diction effect of combined MIDAS model. Research conclusions show that the unrestricted MIDAS model＇s prediction accuracy and fitting effect is higher than others different weights MIDAS models, and the combined MIDAS model which uses BIC is the optimal performance in forecasting of China＇s quarterly GDP.