|
|
|
| Risk Co-movement and Forecasting of the Stock Market and Bond Market Based on the Forefront Perspective of Machine Learning |
| YANG Zihui, ZHANG Pingmiao, LIN Shihan
|
| Business School, Southern University of Science and Technology; Advanced Institute of Finance, Sun Yat-sen University |
|
|
|
Abstract At the Central Financial Work Conference in October 2023, President Xi Jinping emphasized the need to uphold risk prevention and control as the perennial focus of the financial profession and ensure that risks are identified, revealed, and resolved as early as possible. It remains a key future task for the financial profession to improve the risk monitoring and early-warning system and the ability to reduce and control financial risk. Considering the great attention aroused by tail-risk events, such as the simultaneous crashes of both the stock and bond markets, a major priority is to identify the co-movements of risk across different markets and enhance the risk early-warning system for both the stock and bond markets. In the context of stock-bond risk co-movement, it is of both academic and practical significance to forecast stock market and bond market risk accurately and effectively using state-of-the-art machine learning techniques. This will help to overcome the limitations of traditional risk prediction methods and improve the risk monitoring and early-warning system for both the stock market and bond market. It will also contribute to identifying risk co-movements across different markets, thus preventing extreme risk events like simultaneous crashes of the stock and bond markets.
This paper conducts an analysis based on a sample of Chinese listed companies in the period from January 2015 to September 2022. First, we employ the MVMQ-CAViaR model and quantile regression method to investigate the tail-risk co-movements between the stock market and bond market. Additionally, we distinguish between the heterogeneous risk spillover relationships of companies with and without default records. The results show that stock-bond risk co-movement is more significant for companies that have defaulted.
Next, this paper utilizes state-of-the-art machine learning techniques, namely the quantile regression forest, quantile gradient boosting model, and quantile regression neural network, to construct prediction models for tail risk based on the emerging perspective of stock-bond risk co-movement. On this basis, we further evaluate the bi-directional predictive power between stock market risk and bond market risk using the quantile loss function, quantile goodness of fit, and Diebold-Mariano test. For most prediction models for the “bond market risk → stock market risk” direction, the results indicate that considering the bond market risk increases the robustness and accuracy of the forecasting of stock market risk by improving the goodness of fit and reducing fitting errors. The application of a machine learning framework including the quantile gradient boosting model, quantile regression forest, and quantile regression neural network significantly strengthens the prediction of stock market risk. In contrast, in the prediction model for “stock market risk → bond market risk”, the quantile gradient boosting model outperforms the other models in predicting bond market risk. This is attributable to the gradient boosting method, which corrects fitting errors through iterative learning, thereby better capturing the stock market risk information.
Furthermore, we divide the sample according to industry attributes to assess the heterogeneous predictive power of different models. We find evidence of asymmetric predictive power between stock market risk and bond market risk, in which a unidirectional predictive power for “bond market risk → stock market risk” is documented in most industries while a bi-directional predictive power for “stock market risk → bond market risk” is only evident in the materials, daily consumption, finance, and real estate industries. Meanwhile, under the “stock market risk → bond market risk” prediction framework, the finance industry has the most forecastable bond market risk, as it can be identified by all three machine learning models.
Finally, this paper yields policy implications for strengthening the risk prediction system for key fields in China. First, policy makers should improve the tail-risk early-warning framework for China's capital market using the new perspective of stock-bond risk co-movement. Second, state-of-the-art machine learning techniques should be promoted in the financial regulation field to enhance the financial stability guarantee system. Third, regulators should improve the industry-level financial risk forecasting system and tail-risk early warning for key fields including the finance industry.
|
|
Received: 12 December 2022
Published: 17 July 2024
|
|
|
|
|
|
|
|
