中国城投债风险溢价的及时性度量与预测 : 基于高维适应性算法的构建分析 (The Real-Time Measuring and Forecasting of the Chinese Chengtou Bond Risk Premia —Based on the Method of High-Dimensional Adaptive Algorithm )

Chinese abstract:债券风险溢价是衡量债券风险的晴雨表。使用参数模型进行债券风险溢价的预测不仅会受外部经济政策的冲击导致模型参数具有时变特征，同时还因经济变化的实时更迭带来解释变量的及时性选择问题。本文构建的高维适应性算法能及时探测参数与解释变量的最大化同质区间，实现在及时探测参数结构性变化的同时，也能对最佳解释变量进行实时选择。结果显示，该方法为市场提供了一种更为合理的参数建模方式以预判债券风险溢价的未来走势。我们发现：①高维适应性算法相比其他预测模型在城投债风险溢价的样本外预测与方向走势预测中具有显著的预测优势。②从同质区间识别角度来看，相比于2016与2017年，2018年与2019年参数模型的同质性区间变得更短，说明市场更加关注城投债的短期风险，在一定程度上印证和解释了城投债短期风险溢价高于长期风险溢价的“倒挂”现象。③从变量选择结果来看，不同省份在不同时间决定城投债风险溢价的作用不同。决定城投债风险溢价走势的因素集中，风险溢价较高、风险溢价波动较大以及债券发行数量较多的省份具有明显的网络溢出效应，对未来债券的风险溢价具有较大作用；而那些风险溢价较低、风险溢价波动较低以及债券发行数量较少的省份，其网络外溢效应不明显。债券平均规模与债券剩余期限的网络节点效应对债券未来风险溢价也有较大影响。从实证结果的政策含义上来看：一方面，准确、全面地披露城投债信息、管控城投债系统性风险，有助于维系债券市场价格信号的有效性，稳定债券市场；另一方面，在控制城投债风险溢价的过程中，政府需要重点关注风险溢价波动较高、债券发行数量较多的省份，同时也需要时刻关注债券的平均规模与剩余期限，在各省实行差异化的适应性金融政策，对深化金融供给侧结构性改革，维护脱贫攻坚成果具有重要意义。English abstract: As a barometer to measure the risk, the risk premia of bonds imply the risk of the bond market. When using the parametric model to predict the risk premia, the external policies and the real-time changes of the economy not only will lead to the time-varying parameters, but also bring about the changing of the explanatory variables in the model. The high-dimensional adaptive algorithm constructed in this paper can detect the maximum homogeneous intervals of the parameter, detect the structural changes of the variables and select the best explanatory variables in time. The results show that this method provides a more reasonable parameter modeling method to predict the future trend of bond risk premia. We find that, ①in the Chinese Chengtou Bond market, compared with other prediction models, the high-dimensional adaptive algorithm has significant prediction advantages in out of sample prediction and also has the outstanding performance in the trend prediction.②From the perspective of homogeneous interval identification, compared with 2016 and 2017, the homogeneous interval of the parametric model in 2018 and 2019 becomes shorter, which indicates that the market pays more attention to the short-term risk of Chengtou bonds in 2018 and 2019. This finding can help to explain why the short-term risk premia is higher than the long-term risk premia in the Chengtou market. ③From the results of variable selection, different provinces have different roles in determining the risk premia of Chengtou bonds at different times. The factors which determine the trend of risk premia of Chengtou bonds are the provinces with higher risk premia, or the provinces with greater volatility of risk premia, or the provinces that have more issued Chengtou bonds. These factors have obvious network spillover effects. The network spillover effects are not obvious in the provinces with lower risk premia, lower volatility of risk premia and less issued Chengtou bonds. The average bond size and the residual maturity also has a great influence on the future risk premia of Chengtou bonds. From the policy implications of the empirical results: on the one hand, disclosure of the information and managing of the systematic risk will help to stabilize the Chengtou bond market and maintain the effectiveness of the price. On the other hand, in the process of managing the Chengtou bonds, the government needs to focus on the provinces with high volatility of risk premia and with large issued Chengtou bonds. At the same time, we also need to pay attention to the average scale and the residual maturity of the Chengtou bonds. The implementation of the adaptive financial policies in different provinces is of great importance in maintaining the achievements of the “Anti-Poverty-War”