Abstract:

With the continuous development of China’s futures market, the types of listed commodity futures are increasing. A large number of financial capital invested in commodity futures has become a new feature of commodity market operation. The importance of studying the covariance matrix of multiple commodity futures has become increasingly prominent. In this paper, the Bayesian shrinkage and sparse methods in machine learning are integrated into the VAR model with time-varying parameters, and a new SS-TVP-VAR model is constructed. The model is used to dynamically model and forecast the realized covariance matrices of multiple commodity futures. The empirical results show that the SS-TVP-VAR model can effectively predict the realized covariance matrix of China’s commodity futures market, and is superior to other fixed parameter models such as VAR-Lasso in terms of statistical accuracy and economic benefits. The variance and covariance of realized covariance matrix in China’s commodity futures markets have different driving structures, and variance is mainly driven by its own lags, while covariance is mainly driven by other lags. This paper helps investors and market managers perform asset allocation and risk management in terms of both covariance matrix driven structure and predicted future covariance matrix, which is of great practical significance to promote the high-quality development of China’s commodity futures market.

Key words:

realized covariance matrix, commodity futures, Bayesian shrinkage, sparsity

摘要：

随着中国期货市场的规模持续扩大，上市的商品期货种类不断增加，大量金融资本投资商品期货成为商品市场运作的新特点，研究多个商品期货协方差矩阵的重要性日渐凸显。通过将机器学习中的贝叶斯收缩和稀疏方法融入参数VAR模型，构建了一个崭新的SS-TVP-VAR模型，并运用该模型对多个商品期货的已实现协方差矩阵进行动态建模和预测。实证结果表明：SS-TVP-VAR模型能够有效预测中国商品期货市场的已实现协方差矩阵，在统计精度和经济效益方面均优于VAR-Lasso等其他固定参数模型；商品期货市场已实现协方差矩阵的方差项与协方差项具有截然不同的驱动结构，方差项主要受其自身滞后驱动，而协方差项则主要由其他滞后项驱动。本研究有助于投资者及市场管理者从协方差矩阵驱动结构和预测未来协方差矩阵两个方面进行资产配置和风险管理，对推动中国商品期货市场高质量发展具有重要的现实意义。

关键词:

已实现协方差矩阵, 商品期货, 贝叶斯收缩, 稀疏