一种新的集成预测方法——GPVECM

GPVECM-A new integrated forecasting method

采用基于遗传规划的因子提取方法来解决从大量非平稳数据中提取非平稳因子的问题,并提出一个基于遗传规划与向量误差修正模型相结合的集成预测方法(GPVECM).此方法以预测误差最小化为目标优化了因子提取过程,将因子提取与预测模型有机地集成在一起.本文应用GPVECM方法对中国进出口贸易预测进行了实证检验,取得了良好的预测结果,表明此方法能够有效地提炼数据中的信息,从而提高预测精度,进行有效预测.

Considering the information demand of forecasting, we propose a method called GPVECM by combining genetic programming with vector error correction model, to summarize useful information from many economic variables and forecast. This method could construct new features directly from non-stationary time series, aiming to minimize forecasting error. We adopt a one-step process to assure the factors constructed satisfying the assumptions of the econometric model. Empirical results of Chinese gross trade forecasting show that the method could improve accuracy remarkably.