基于半参数混频误差修正模型的中国CPI预测研究  

Research on China CPI Forecast Based on Semi-parametric ECM-MIDAS Model

作　　者：鲁万波[1] 杨冬 Lu Wanbo;Yang Dong

机构地区：[1]西南财经大学统计学院 [2]西南财经大学统计学院数量经济研究所

出　　处：《统计研究》2018年第10期28-43,共16页Statistical Research

基　　金：国家自然科学基金"高维协高阶矩的估计及其在投资组合中的应用"(71771187);教育部"新世纪优秀人才支持计划"(NCET-13-0961);中央高校基本科研业务费专项资金"混频多因子模型高阶矩建模及其投资组合研究"(JBK1807047);西南财经大学数据科学与商业智能联合实验室的资助

摘　　要：鉴于宏观经济变量具有明显的非线性特征,本文将非线性误差修正项引入存在协整关系的非平稳混频数据抽样(MIDAS)模型中,构建了半参数混频数据抽样误差修正(SEMI-ECM-MIDAS)模型;使用广义似然比(GLR)检验,拓展了混频数据下模型函数形式的一致性检验问题;最后使用该模型研究了中国股票市场周数据、广义货币发行量月度数据和国际原油市场月度数据对中国CPI的短期预测效果;基于AIC准则,对包含半参数模型在内的4种混频数据抽样模型和2种同频模型的连续预测效果进行了全面比较。研究结果发现:GLR检验表明误差修正项具有明显的非线性特征且在回归中具有显著的反向修正机制,无论采用递归样本、滚动样本还是固定样本,本文提出的SEMI-ECM-MIDAS模型在进行连续预测时均具有最优的预测精度,且预测结果不受混频动态协整关系选择的影响。Considering that the macroeconomic variables have obvious nonlinear characteristics,a semi-parametric ECM-MIDAS model is constructed for the data which exist cointegration relation in the MIDAS model. By using the GLR test,the problem of the consistency test of the functional form of the parametric regression model is solved. The simulation result shows that SEMI-ECM-MIDAS has better forecast performance when the error term is nonlinear. In application part,China＇s stock market weekly data,China＇s monetary market and crude oil market international data are used to make short-term forecast of China ＇s CPI. A comprehensive comparison was made among the models. The results show that the error correction term has obvious nonlinear characteristics. The semi-parametric ECM-MIDAS model proposed in this paper always has the best prediction accuracy. In addition,the prediction results are not affected by the choice of dynamic mixed frequency cointegration relationship.

关 键 词：半参数混频误差修正模型 广义似然比检验 预测 月度CPI 

分 类 号：C812[社会学—统计学]