基于部分函数线性回归改进方法的CPI混频预测

CPI Mixed Frequency Forecasting Based on Improved Method of Partial Functional Linear Regression

针对混频数据,提出一种新的分析方法。首先把具有函数特征的高频数据看成是某个随机过程产生的函数,然后利用部分函数线性回归模型对混频数据进行分析,并根据金融高频数据的相依特征,提出基于残差协方差函数的模型估计改进方法。最后通过数值模拟和CPI预测实例与现有的部分函数线性回归模型及MIDAS模型进行对比分析,结果表明本文提出的改进方法的样本外预测精度最高。

A new analysis method is proposed for mixed frequency data.First,the high frequency data with functional characteristics are treated as functions generated by a random process,and then the mixed data is analyzed using a partial functional linear regression model.And based on the dependent characteristics of the financial high-frequency data,a improved model estimation method is proposed based on residual covariance function.Finally,the numerical simulation and CPI prediction example are used to compare with the existing partial functional linear regression model and MIDAS model.The results show that the improved method proposed in this paper has the highest out of sample prediction accuracy.