债券信用溢价的共同因子与结构变化:基于机器学习方法

Common Factors and Structural Changes of Bond Credit Premium:Based on Machine Learning Methods

共同因子是刻画风险溢价的重要基础,将共同因子模型应用于公司债券市场有助于合理估计信用风险溢价。本文利用机器学习算法探究债券信用溢价因子的存在性以及结构变化后发现:规模、下行风险、价值、波动率以及流动性等五个公司债券共同因子对单个债券信用溢价有较好的解释能力,动量因子对信用溢价的解释能力较差,流动性因子具有较强的逆周期防御性功能。债券市场以及公司债券信用溢价因子在2015年前后存在明显的结构变化,利用稀疏学习和集成学习可以有效分析因子结构变化,建立风险预警。此外,在公司债券市场的市场制度和环境改变过程中,可以利用机器学习算法识别其对市场的影响,防范化解潜在的系统性风险。

The common factor is an important basis for characterizing the risk premium. Applying the common factor model to the corporate bond market helps to reasonably estimate the credit risk premium. Using machine learning algorithms to explore the existence and structural changes of bond credit premium factors,the study found that the five common factors of corporate bonds,including scale,downside risk,value,volatility,and liquidity,have a good ability to explain the credit premium of a single bond. Momentum factor is poor in explaining credit premium,and liquidity factor has a strong counter-cyclical defensive function. The bond market and corporate bond credit premium factors have obvious structural changes around 2015. Using sparse learning and integrated learning can effectively analyze the changes in factor structure and establish risk warnings. In addition,in the process of changes in the market system and environment of the corporate bond market,machine learning algorithms can be used to identify its impact on the market and prevent and resolve potential systemic risks.