Scholarlink

<p>Financial time series data are characterized by significant noise and volatility, complicating accurate analysis and forecasting. Conventional denoising techniques often struggle with these datasets’ non-stationary, high-frequency noise. To address this, we propose&#xa0;SWiFTS-D, the first method to combine TCN-driven dynamic thresholding with Elastic Net regularization in a wavelet framework for financial denoising. This novel approach integrates sparsity-enhanced wavelet transforms with dynamic thresholding, utilizing a Temporal Convolutional Network (TCN) for adaptive threshold prediction and Elastic Net regularization for feature selection. This approach dynamically adjusts to changing market conditions. SWiFTS-D offers three key contributions: Adaptive volatility-aware denoising through a TCN-based dynamic thresholding algorithm that adjusts in real-time to market conditions; Sparsity-enhanced feature selection using Elastic Net regularization on thresholded coefficients to retain economically significant patterns; Continuous learning that recalibrates wavelet bases and thresholds via statistical optimization. Experiments across diverse financial datasets, including Bitcoin, Ethereum, EUR/USD, MSFT, and Crude Oil, demonstrate that SWiFTS-D outperforms traditional methods, achieving superior signal-to-noise ratio (SNR improvements of 14.5–54.7%), peak signal-to-noise ratio (PSNR gains of 13.3–55.5%), and correlation metrics. SWiFTS-D provides a robust tool for financial analytics with applications in algorithmic trading, risk management, and economic trend analysis.</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

Sparsity-enhanced wavelet transform with dynamic thresholding for financial time series denoising

  • 期刊论文

  • 发表日期：

    2026年4月3日

查看全文
  • Peter Tettey Yamak,
  • Yujian Li,
  • Ting Zhang

摘要

Financial time series data are characterized by significant noise and volatility, complicating accurate analysis and forecasting. Conventional denoising techniques often struggle with these datasets’ non-stationary, high-frequency noise. To address this, we propose SWiFTS-D, the first method to combine TCN-driven dynamic thresholding with Elastic Net regularization in a wavelet framework for financial denoising. This novel approach integrates sparsity-enhanced wavelet transforms with dynamic thresholding, utilizing a Temporal Convolutional Network (TCN) for adaptive threshold prediction and Elastic Net regularization for feature selection. This approach dynamically adjusts to changing market conditions. SWiFTS-D offers three key contributions: Adaptive volatility-aware denoising through a TCN-based dynamic thresholding algorithm that adjusts in real-time to market conditions; Sparsity-enhanced feature selection using Elastic Net regularization on thresholded coefficients to retain economically significant patterns; Continuous learning that recalibrates wavelet bases and thresholds via statistical optimization. Experiments across diverse financial datasets, including Bitcoin, Ethereum, EUR/USD, MSFT, and Crude Oil, demonstrate that SWiFTS-D outperforms traditional methods, achieving superior signal-to-noise ratio (SNR improvements of 14.5–54.7%), peak signal-to-noise ratio (PSNR gains of 13.3–55.5%), and correlation metrics. SWiFTS-D provides a robust tool for financial analytics with applications in algorithmic trading, risk management, and economic trend analysis.