Scholarlink

<p>This paper implements the use of deep learning (DL) techniques to enhance sparse index tracking portfolios for the Nasdaq 100 (NDX) index. Specifically, we use autoencoders (AEs) and variational autoencoders (VAEs) to construct sparse tracking portfolios under cardinality constraints, thereby reducing the number of required stocks while maintaining close performance to the NDX index. By using AEs and VAEs, we extract the complex non-linear relationships between the index and a subset of its constituents. Using historical data from 2019 to 2023, our empirical results show that the proposed DL-based models achieve robust index replication in both in-sample and out-of-sample periods. Furthermore, these models exhibit reduced tracking errors and cumulative returns comparable to the benchmark. The results highlight the potential of deep learning techniques as a powerful tool for efficient index tracking.</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

Enhancing Sparse Index-Tracking Portfolios Using Deep Learning Models

  • 期刊论文

  • 发表日期：

    2025年2月20日

查看全文
  • Carlos Andres Zapata Quimbayo,
  • Daniel Aragón Urrego,
  • John Freddy Moreno Trujillo,
  • Oscar Eduardo Reyes Nieto

摘要

This paper implements the use of deep learning (DL) techniques to enhance sparse index tracking portfolios for the Nasdaq 100 (NDX) index. Specifically, we use autoencoders (AEs) and variational autoencoders (VAEs) to construct sparse tracking portfolios under cardinality constraints, thereby reducing the number of required stocks while maintaining close performance to the NDX index. By using AEs and VAEs, we extract the complex non-linear relationships between the index and a subset of its constituents. Using historical data from 2019 to 2023, our empirical results show that the proposed DL-based models achieve robust index replication in both in-sample and out-of-sample periods. Furthermore, these models exhibit reduced tracking errors and cumulative returns comparable to the benchmark. The results highlight the potential of deep learning techniques as a powerful tool for efficient index tracking.