Scholarlink

<p>Constructing an optimal investment portfolio, which encompasses asset selection, return prediction, and capital allocation in line with investor objectives,—requires analytical rigor and market intuition. Recent advances in machine learning have introduced powerful tools to support these decisions. This paper proposes a dynamic, data-driven portfolio management framework using Nasdaq stock data from January 2017 to December 2024. The framework integrates three key components: a novel fuzzy clustering approach that accounts for outliers and interdependencies to enhance asset selection; a Long Short-Term Memory (LSTM) model to improve return forecasting accuracy; and a portfolio optimization module that dynamically allocates capital using multiple objective functions. Empirical results demonstrate that the proposed framework significantly outperforms benchmark approaches across portfolio performance metrics. Decomposition analysis further confirms that each module (clustering, forecasting, and optimization) contributes independently to the overall performance gains, highlighting the model’s robustness and practical value for portfolio managers.</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

Integrated machine learning framework for dynamic portfolio management: clustering, forecasting, and optimization

  • 期刊论文

  • 发表日期：

    2026年2月6日

查看全文
  • Hossein Dastkhan,
  • Alireza Mosaffa,
  • Frank J. Fabozzi

摘要

Constructing an optimal investment portfolio, which encompasses asset selection, return prediction, and capital allocation in line with investor objectives,—requires analytical rigor and market intuition. Recent advances in machine learning have introduced powerful tools to support these decisions. This paper proposes a dynamic, data-driven portfolio management framework using Nasdaq stock data from January 2017 to December 2024. The framework integrates three key components: a novel fuzzy clustering approach that accounts for outliers and interdependencies to enhance asset selection; a Long Short-Term Memory (LSTM) model to improve return forecasting accuracy; and a portfolio optimization module that dynamically allocates capital using multiple objective functions. Empirical results demonstrate that the proposed framework significantly outperforms benchmark approaches across portfolio performance metrics. Decomposition analysis further confirms that each module (clustering, forecasting, and optimization) contributes independently to the overall performance gains, highlighting the model’s robustness and practical value for portfolio managers.