Scholarlink

The modeling of asset price dynamics is a cornerstone of quantitative finance. However, traditional assumptions of constant volatility and the absence of jumps limit the ability to reflect real market behaviors. To address these limitations, stochastic volatility models with jumps (SVJ) have been developed, accounting for the stochastic nature of volatility, which is particularly relevant during periods of stress or uncertainty, as in the case of episodes of asset price bubbles [4]. Studies have validated the effectiveness of these models in capturing market anomalies, such as volatility smiles and skews [5, 6]. Furthermore, Jraifi [7] proposed an innovative approach for calibrating SVJ models using recurrent neural networks (RNNs) to model temporal dependencies in volatility data, demonstrating a significant improvement in prediction accuracy (RMSE reduced by 15% compared to traditional methods) and increased robustness to changing market conditions.

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

Deep Learning Enhanced Drift Calibration in Stochastic Volatility Jump-Diffusion Models: A Tikhonov Regularization Approach with Neural Networks

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Abdelilah Jraifi,
  • Ibtissam Medarhri,
  • Ikram El Bouhali

摘要

The modeling of asset price dynamics is a cornerstone of quantitative finance. However, traditional assumptions of constant volatility and the absence of jumps limit the ability to reflect real market behaviors. To address these limitations, stochastic volatility models with jumps (SVJ) have been developed, accounting for the stochastic nature of volatility, which is particularly relevant during periods of stress or uncertainty, as in the case of episodes of asset price bubbles [4]. Studies have validated the effectiveness of these models in capturing market anomalies, such as volatility smiles and skews [5, 6]. Furthermore, Jraifi [7] proposed an innovative approach for calibrating SVJ models using recurrent neural networks (RNNs) to model temporal dependencies in volatility data, demonstrating a significant improvement in prediction accuracy (RMSE reduced by 15% compared to traditional methods) and increased robustness to changing market conditions.