Econ Papers Daily / 期刊论文
记忆结构在WTI原油期货尾部风险预测中的作用
The role of memory structure in tail risk forecasting for WTI crude oil futures
摘要
The West Texas Intermediate (WTI) crude oil futures market plays an important role in oil-price discovery and risk management, making accurate forecasts of downside tail risk valuable to market participants. Using daily WTI futures data from January 2000 to December 2025, this paper examines how the use of historical information affects downside Value at Risk (VaR) forecasting. We develop a unified Conditional Autoregressive Value at Risk (CAViaR) framework that jointly considers four aspects of the memory structure, which are window design, bandwidth length, kernel weighting, and the CAViaR updating specification. Together, these choices determine how historical oil-market information is retained, weighted, discounted, and updated. Our paper differs from related studies by exploring which memory structures are more informative for crude-oil tail-risk forecasting under different market conditions. The full-sample evidence identifies a set of superior forecasting structures, characterized by expanding windows, non-truncated kernel weighting, medium-to-long bandwidths, and asymmetric updating dynamics. Relative to conventional rolling-window, uniformly weighted CAViaR benchmarks, these structures deliver relatively more stable gains at the 95% VaR level, whereas gains at the 99% level are more concentrated in particular oil-market states. The findings provide useful implications for VaR-based downside risk management in the WTI crude oil futures market.
中文摘要
西德克萨斯中质原油（WTI）期货市场在石油价格发现与风险管理中发挥着重要作用，因此对下行尾部风险的准确预测对市场参与者具有重要价值。本文使用2000年1月至2025年12月的WTI期货日度数据，考察历史信息的使用方式如何影响下行风险价值（VaR）预测。我们构建了一个统一的条件自回归风险价值（CAViaR）框架，该框架同时考虑记忆结构的四个方面，即窗口设计、带宽长度、核加权以及CAViaR更新设定。这些选择共同决定了历史石油市场信息如何被保留、加权、折现和更新。本文与相关研究的不同之处在于，探讨在不同市场状态下哪些记忆结构对原油尾部风险预测更具信息含量。全样本证据识别出一组占优的预测结构，其特征为扩展窗口、非截断核加权、中长带宽以及非对称更新动态。相对于传统的滚动窗口、等权重的CAViaR基准，这些结构在95% VaR水平上带来相对更稳定的改进，而在99%水平上的改进则更集中于特定的石油市场状态。研究结论为WTI原油期货市场中基于VaR的下行风险管理提供了有益启示。