Scholarlink

<p>Small and medium-sized traders can be highly affected by losses caused by exchange rate changes. Therefore, the Value-at-Risk (VaR) methods, which provide estimates for a long-time horizon are of interest to them. In our previous backtesting research we realized that no classical VaR method is applicable for long-time horizon. Therefore, we study here several new approaches to long-time horizon VaR models, which improves the behaviour of classical short and medium-time VaR methods. A nonparametric approach with the decay of data importance for older data reveals the best performance with respect to the numerous precision tests. The backtesting is done on 6 currencies resulted as medoids from functional clustering. The results show that the standard methods based on random walk forecast model have at least by 97% bigger quantile regression loss function than our proposed approach for 1 year ahead forecast in average for all 6 currencies. Since 2023, its usage and practical application are publicly available at <a href="https://var.ef.jcu.cz/">https://var.ef.jcu.cz/</a> for 32 currencies listed by the European Central Bank.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

New value-at-risk method adjusted for a long time horizon with application to exchange rates

  • 期刊论文

  • 发表日期：

    2025年12月6日

查看全文
  • Tomáš Mrkvička,
  • Martina Krásnická,
  • Gabriela Hlásková

摘要

Small and medium-sized traders can be highly affected by losses caused by exchange rate changes. Therefore, the Value-at-Risk (VaR) methods, which provide estimates for a long-time horizon are of interest to them. In our previous backtesting research we realized that no classical VaR method is applicable for long-time horizon. Therefore, we study here several new approaches to long-time horizon VaR models, which improves the behaviour of classical short and medium-time VaR methods. A nonparametric approach with the decay of data importance for older data reveals the best performance with respect to the numerous precision tests. The backtesting is done on 6 currencies resulted as medoids from functional clustering. The results show that the standard methods based on random walk forecast model have at least by 97% bigger quantile regression loss function than our proposed approach for 1 year ahead forecast in average for all 6 currencies. Since 2023, its usage and practical application are publicly available at https://var.ef.jcu.cz/ for 32 currencies listed by the European Central Bank.