统计研究 ›› 2021, Vol. 38 ›› Issue (4): 145-160.doi: 10.19343/j.cnki.11-1302/c.2021.04.011

• • 上一篇    

收益率曲面预测及其在信用债投资组合管理中的应用

王雷 闫红蕾 张自力   

  • 出版日期:2021-04-25 发布日期:2021-04-25

Prediction of the Yield Surface and Credit Bond Portfolio Management Strategy

Wang Lei Yan Honglei Zhang Zili   

  • Online:2021-04-25 Published:2021-04-25

摘要: 收益率曲线是信用债投资者的重要参考。在市场套利活动的作用下,跨期限和信用等级的债券收益率具有很强的内生联系,其变动规律具有整体性和连续性。以往研究将信用债收益率曲线拆分成无风险利率曲线和信用利差两个部分,前者关注期限的影响,后者关注信用等级的影响,但很少同时考虑两者的共同影响。本文在收益率曲线的基础上增加信用等级维度,将AAA+级到AA级收益率曲线视为一个相互关联的整体,定义为信用债的收益率曲面(Yield Surface)。相比收益率曲线,收益率曲面包含了跨等级的系统性预测信息,通过预测收益率曲面能够构建具有较高收益的投资管理策略。本文以中期票据市场为例,采用卷积神经网络模型预测1周后的收益率曲面,在此基础上计算债券的预测理论价格,发现该价格对未来交易价格的预测有显著作用。基于预测价差,本文提出了信用债投资管 理策略,应用该策略的投资组合能够获得显著的正收益。业绩归因分析发现,该策略取得的收益同时来 自投资组合在信用风险和久期风险上的暴露,预测价差可以刻画债券市场的“风险前沿”。本文采用了中债估值价格进行稳健性检验,主要结论均保持一致,具有较强的稳健性。

关键词: 信用债, 收益率曲面, 卷积神经网络, 信用债投资组合管理, 神经网络的可解释性

Abstract: The yield curve is an important reference for credit bond investors. Under the influence of market arbitrage activities, bond yields are connected endogenously across maturities and credit ratings, and the mechanism is holistic and continuous. Previous studies divide the credit bond yield curve into two parts--the risk-free interest rate curve and the credit spread. The former focuses on the impact of maturity and the latter focuses on the impact of credit ratings, without considering the combined effects of the two. This paper views the yield curve of AAA+ to AA bond as an interconnected whole based on credit ratings and maturity, which is defined as Yield Surface. Compared with the yield curve, the yield surface contains systematic forecast information across credit ratings, and the forecast helps make a portfolio management strategy for high yields. Taking the medium-term note market as an example, this paper uses a convolutional neural network model to predict the yield surface one week later. Based on this, the predicted theoretical price of the bond is calculated and it is found to have a significant predictive ability for future transaction prices. Besides, this paper proposes an investment strategy based on the spreads between predicted theoretical price and trading price, and the investment portfolio under the strategy can achieve significant positive returns. The performance attribution analysis indicates that the returns come from the exposure to credit risk and duration risk simultaneously. Thus, the predicted spreads can depict the “risk frontier” of the bond market. This paper performs the robustness test with the valuation of ChinaBond, and the main conclusions are consistent and quite robust.

Key words: Credit Bond, Yield Surface, Convolutional Neural Network, Credite Bond Portfolio Management, Interpretability for Neural Network