Scholarlink

<p>Management of financial risks, pricing of bonds, and economic forecasting all rely heavily on interest rate modeling. In particular, standard numerical methods frequently encounter difficulties in accurately depicting complex stochastic trends in interest rate dynamics. To address these challenges, this study introduces a Hybrid Physics-Informed Neural Networks (H-PINNs) for interest-rate modeling that integrates the Hull–White SDE into the training objective and adds a hybrid embedding layer (Fourier+polynomial) with automated selection to improve feature extraction and sample efficiency. The model is trained on US Treasury yield data for the 13-week and 10-year maturities spanning 2010–2023, and its out-of-sample performance is evaluated over the period from January to October 2024, covering diverse economic conditions and market fluctuations. We assess the proposed framework via numerical experiments that benchmark H-PINNs (Fourier+Polynomial) against the calibrated Hull–White (HW) model and PINN baselines, including V-PINNs, H-PINNs (Fourier), and H-PINNs (Polynomial), for stochastic interest-rate modeling. Model fidelity is quantified through comprehensive error metrics such as MSE, MAE, RMSE, MedAE, Max, <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(L_2\)</EquationSource> </InlineEquation> Error and <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(R^{2} \)</EquationSource> </InlineEquation> and further supported by rigorous statistical analyses, including ANOVA, the Wilcoxon test, and 95% confidence intervals for RMSE, to examine the model’s predictive accuracy, stability, and effectiveness across varying market scenarios. Furthermore, this paper explores potential future research directions, establishing a foundation for additional investigation into the PINNs framework.</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

A hybrid PINNs approach to capture interest rate dynamics in short-rate model

  • 期刊论文

  • 发表日期：

    2025年11月27日

查看全文
  • Indu Rani,
  • Chandan Kumar Verma

摘要

Management of financial risks, pricing of bonds, and economic forecasting all rely heavily on interest rate modeling. In particular, standard numerical methods frequently encounter difficulties in accurately depicting complex stochastic trends in interest rate dynamics. To address these challenges, this study introduces a Hybrid Physics-Informed Neural Networks (H-PINNs) for interest-rate modeling that integrates the Hull–White SDE into the training objective and adds a hybrid embedding layer (Fourier+polynomial) with automated selection to improve feature extraction and sample efficiency. The model is trained on US Treasury yield data for the 13-week and 10-year maturities spanning 2010–2023, and its out-of-sample performance is evaluated over the period from January to October 2024, covering diverse economic conditions and market fluctuations. We assess the proposed framework via numerical experiments that benchmark H-PINNs (Fourier+Polynomial) against the calibrated Hull–White (HW) model and PINN baselines, including V-PINNs, H-PINNs (Fourier), and H-PINNs (Polynomial), for stochastic interest-rate modeling. Model fidelity is quantified through comprehensive error metrics such as MSE, MAE, RMSE, MedAE, Max, \(L_2\) Error and \(R^{2} \) and further supported by rigorous statistical analyses, including ANOVA, the Wilcoxon test, and 95% confidence intervals for RMSE, to examine the model’s predictive accuracy, stability, and effectiveness across varying market scenarios. Furthermore, this paper explores potential future research directions, establishing a foundation for additional investigation into the PINNs framework.