主管:中国科学院
主办:中国优选法统筹法与经济数学研究会
   中国科学院科技战略咨询研究院

中国管理科学 ›› 2025, Vol. 33 ›› Issue (2): 61-70.doi: 10.16381/j.cnki.issn1003-207x.2022.1541cstr: 32146.14.j.cnki.issn1003-207x.2022.1541

• • 上一篇    下一篇

基于时变极值方法的VaR预测模型以及应用

宋诗佳, 田飞, 李汉东()   

  1. 北京师范大学系统科学学院,北京 100875
  • 收稿日期:2022-07-14 修回日期:2022-09-27 出版日期:2025-02-25 发布日期:2025-03-06
  • 通讯作者: 李汉东 E-mail:lhd@bnu.edu.cn

VaR Prediction Model Based on Time-varying Extremum Method and Its Application

Shijia Song, Fei Tian, Handong Li()   

  1. School of Systems Science,Beijing Normal University,Beijing 100875,China
  • Received:2022-07-14 Revised:2022-09-27 Online:2025-02-25 Published:2025-03-06
  • Contact: Handong Li E-mail:lhd@bnu.edu.cn

摘要:

基于高频数据的VaR预测是风险管理领域关注的热点。本文提出了一种基于高频数据的时变极值VaR预测模型,即ARFIMA-RGARCH-DPOT-VaR模型。该模型在基本的均值—方差模型的基础上,假设收益波动受到服从广义帕累托(GP)分布的极值影响,且GP的动态参数由已实现波动测度驱动。基于两步法进行建模和参数估计,首先,用ARFIMA-RGARCH模型对收益序列进行建模,得到标准化新息;其次,以RGARCH模型预测的已实现波动测度为协变量,对标准化新息超过阈值的超值(POT)的分布参数进行动态估计。在此基础上,实现对VaR的预测。在该方法的实证分析中,对中国股票市场和美国股票市场的五种主要指数即上证指数、沪深300指数、深证成指、标普500指数和纳斯达克指数的VaR预测结果表明,本文提出的方法比基准RGARCH模型和HEAVY模型更能精准预测尾部的风险。

关键词: VaR预测, RGARCH, 极值理论, 超值, 已实现波动

Abstract:

VaR is an important tool for risk management. This paper proposes a time-varying extreme VaR prediction model based on high frequency data, namely ARFIMA-RGARCH-DPOT-VAR model. Based on the basic mean-variance model, the model assumes that return fluctuations are affected by extreme values that obey generalized Pareto (GP) distributions, and GP dynamic parameters are driven by realized volatility measures. The model is modeled and estimated based on the two-step method. First, the ARFIMA-RGARCH model is used to model the return sequence and the standardized innovation is obtained. Secondly, the realized fluctuation measure predicted by RGARCH model is taken as the covariable, and the distribution parameters of POT with standardized innovation exceeding the threshold are estimated dynamically. On this basis, the prediction of VaR is realized. In the empirical analysis, the VaR prediction results of three major indexes in China's stock market and the USA’s stock market, namely Shanghai Composite Index, CSI 300 Index, Shenzhen Composite Index, S&P500 Index and Nasdaq Index show that the RGARCH-DPOT method proposed by us is more accurate than the standard RGARCH model and HEAVY model in predicting tail risks.

Key words: VaR predicting, RGARCH, extreme theory, POT, realized volatility

中图分类号: