中国债券收益率的可预测性检验

基于工具变量(instrumental variable, IVX)统计检验方法,对我国主要宏观经济变量能否单独或联合预测债券收益率进行可预测性研究.IVX检验方法无需考虑预测变量的持续性先验信息,对任何属于单位根过程、近单位根过程、近平稳过程或平稳过程的预测变量都稳健.进一步使用AIC和BIC方法对多元变量回归模型进行变量筛选,仿真结果表明,两种准则均能准确筛选出模型重要变量并排除冗余变量.通过对国债和AAA、AA-级企业债三种不同信用等级债券收益率的长短期可预测性差异分析,研究发现:1) 1年期储蓄存款利率、人民币实际有效汇率、沪深300指数、工业增加值同比以及大部分宏观经济变量组合在10%显著性水平上能显著预测债券收益率;2)宏观经济变量长期预测能力强于短期;3)基于IVX方法的AIC和BIC准则能够有效地筛选模型重要变量.

Testing for Chinese bond return predictability

In this study, we investigate whether bond return can be predicted by macroeconomic variables based on the instrumental variable (IVX) based Wald statistics. The IVX-based testing procedure does not have any prior on the degree of persistence of the predicting variables and robustifies inference regardless of the predicting variable being stationary, or have moderate deviation from a unit root, or instead they are near unit root or unit root process. Meanwhile, AIC and BIC are further introduced into the IVX method to select important variables. Monte Caro simulations suggest that they can select the important variables and exclude redundant variable correctly. By analyzing the differences in the long-term and short-term predictability of the bond yields of three different credit ratings for government bonds, AAA and AA-grade corporate bonds, we find 1) The 1-year savings deposit rate, the real effective RMB exchange rate, the Shanghai and Shenzhen 300 index return, and most of the macroeconomic variables combinations can be used to predict the bond return; 2) Long-term predictability of macroeconomic variables is stronger than the short term; 3) Both AIC and BIC can effectively select important variables.