Scholarlink

<p>Generating high-fidelity synthetic stock market data is essential for financial modeling and risk assessment. This study proposes the Deep Quantization Variational Autoencoder (DQVAE), a novel generative framework integrating Adaptive Attention Modulation (AAM), Adaptive Quantization (AQ), and the Adaptive Modulated Sigmoid (AMSig) activation function. These components enable the model to capture intricate stock price patterns while improving feature prioritization and data reconstruction quality. Experimental results demonstrate that DQVAE achieves high accuracy in replicating price features, maintaining strong statistical similarity with real stock data. However, challenges remain in modeling the variability of trading volume, highlighting areas for further refinement. This work contributes to advancing synthetic time-series data generation and holds potential for applications in financial forecasting, algorithmic trading, and secure data sharing.</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

DQVAE: leveraging adaptive modulated sigmoid, attention, and quantization for high-fidelity synthetic stock market data

  • 期刊论文

  • 发表日期：

    2025年6月14日

查看全文
  • Mohammad Diqi,
  • Ema Utami,
  • Kusrini,
  • Ferry Wahyu Wibowo

摘要

Generating high-fidelity synthetic stock market data is essential for financial modeling and risk assessment. This study proposes the Deep Quantization Variational Autoencoder (DQVAE), a novel generative framework integrating Adaptive Attention Modulation (AAM), Adaptive Quantization (AQ), and the Adaptive Modulated Sigmoid (AMSig) activation function. These components enable the model to capture intricate stock price patterns while improving feature prioritization and data reconstruction quality. Experimental results demonstrate that DQVAE achieves high accuracy in replicating price features, maintaining strong statistical similarity with real stock data. However, challenges remain in modeling the variability of trading volume, highlighting areas for further refinement. This work contributes to advancing synthetic time-series data generation and holds potential for applications in financial forecasting, algorithmic trading, and secure data sharing.