摘要
文章以1990年1月至2017年1月间我国CPI指数序列为研究对象,采用ARMA模型对序列进行拟合和预测,得到短期预测误差为3.599,长期预测误差为12.528。针对ARMA模型没有良好捕捉到CPI序列中非线性关系的缺陷,本文采用BP网络、RBF网络以及核方法对其作了改进。有非线性特征的三种模型长期预测精度与ARMA模型相当,而短期预测精度有较大提高,最大提高比例为51.85%。
This paper takes Chinese CPI index sequence from January 1990 to January 2017 as the research object, adopts ARMA model to fit and predict the sequence, and obtains the short-term prediction error 3.599 and the long-term prediction error12.528. Aiming at the defect that the ARMA model can not catch the nonlinear relation in CPI sequence, the paper uses BP neural network, RBF network and kernel method to improve it. The long-term prediction accuracy of the three models with nonlinear characteristics is comparable to that of the ARMA model, but the short-term prediction accuracy is greatly improved, with a maximum improvement of 51.85%.
作者
孙冠华
Sun Guanhua(School of Economics,Nanjing University,Nanjing 210093,China)
出处
《统计与决策》
CSSCI
北大核心
2018年第16期18-21,共4页
Statistics & Decision
