系统管理学报

系统管理学报 ›› 2023, Vol. 32 ›› Issue (2): 332-342.DOI: 10.3969/j.issn.1005-2542.2023.02.010

• 数字经济与金融工程 • 上一篇    下一篇

基于双重XGBoost模型的农产品期货波动率预测——以玉米期货为例

胡越1，王桑原1，覃浩恒2，徐亮1，张一苇3   

  1. 1.西南财经大学工商管理学院，成都 611130；2.亚利桑那州立大学商学院，美国 亚利桑那州 850003；3.华期创一成都投资有限公司，成都 610213
  • 出版日期:2023-03-28 发布日期:2023-03-25
  • 作者简介:胡越（1998-），女，博士生。研究方向为大数据与优化。
  • 基金资助:

    国家自然科学基金资助项目（71971177，U1811462）

Volatility Prediction of Agricultural Products Based on Dual XGBoost Model: A Case Study of Corn Futures

HU Yue1, WANG Sangyuan1, QING Haoheng2, XU Liang1, ZHANG Yiwei3   

  1. 1. School of Business Administration, Southwestern University of Finance and Economics, Chengdu 611130, China; 2. W.P. Carey School of Business, Arizona State University, AZ 850003, USA; 3. Huaqi Chuangyi Chengdu Investment Co., Ltd., Chengdu 610213, China
  • Online:2023-03-28 Published:2023-03-25

PDF

摘要：

农产品期货的波动率在农产品衍生品定价、风险分散和农产品风险对冲等领域都起着关键性作用。对波动率进行预测，投资者可以依据波动率预测结果，对预期可能面临的风险采取相应的应对策略，更加精准地进行农产品风险管理。但波动率预测领域存在如下挑战：① 波动率的预测期限较短，仅为1天或3天，难以反映资产在未来较长时间的价格波动率情况；② 以往研究多关注于价格等信息，在波动率预测中对于基本面信息考虑较少；③ 神经网络、深度学习等预测模型的可解释性较差，网络构建和超参数的选择多依赖于经验选择。本文提出了一个基于XGBoost模型的波动率预测框架，考虑价格和基本面数据，对于波动率的长期趋势和短期变化进行了分析。实证结果表明，加入了更多信息维度的模型有助于提升波动率预测的精度，相比于传统的GARCH模型，均方误差MSE缩小了35%以上。

关键词:

农产品期货, 机器学习, 波动率预测, XGBoost模型

Abstract:

The volatility of agricultural futures plays a key role in the pricing of agricultural derivatives, risk diversification, and hedging of agricultural price risks. Accurate volatility forecasting allows investors to deal with adverse price changes and manage agricultural risks more effectively. However, there are several challenges in the field of volatility forecasting. The forecast period of volatility is short, only 1 day or 3 days, and it is difficult to reflect the price volatility of assets in the future for a long time. Previous studies have mostly focused on information such as price, and less consideration has been given to fundamental information in volatility forecasting. In addition, the interpretability of prediction models such as neural networks and deep learning is poor, and the selection of network construction and hyper parameters mostly depends on empirical selection. This paper proposes a volatility prediction framework based on the XGBoost model, considering the price and fundamental information, and analyzes the long-term trend and short-term changes of volatility. The empirical results show that the model with more information dimensions is helpful in improving the accuracy of volatility prediction. Compared with the traditional GARCH (1, 1), the MSE is reduced by more than 35%.

Key words:

agricultural futures, machine learning, volatility forecast, XGBoost model 

中图分类号: