Predicting Chinese bond risk premium with machine learning
研究用机器学习方法预测中国国债一年期持有超额收益，发现收益率曲线与风险溢价存在非线性关系，货币政策与税收等宏观因素影响显著，且对长期债券更有效。
This paper investigates whether bond yield curve and macroeconomic factors have nonlinear relationships with bond risk premia in the Chinese bond market. We apply machine learning approaches to forecast Chinese treasury bond one-year holding period excess returns. Our results show that the bond yield curve has significant nonlinear predictive relationships with bond risk premia. We find evidence that ‘monetary policy’ and ‘tax’ macroeconomic groups have stronger nonlinear relationships with risk premia while ‘invest’ macroeconomic factors matter more for bonds with longer maturities. This paper provides statistical evidence for a significant relationship between expected bond risk premia and several economic drivers including range of forecast of GDP and bond volatility variables. We further document the economic values of our forecasting results by showing they can generate statistically higher certain equivalent values than those from the benchmark forecast.