Scholarlink

<p>Forecasting volatility in financial time series with a single model is challenging due to its non-linear, volatile, and noisy nature. This article introduces semi-parametric decomposition-based hybrid models RGARCH-M-J and REGARCH-M-J, where “J” represents the decomposition techniques EEMD, CEEMDAN, or ICEEMDAN. These models enhance volatility forecasting by integrating data decomposition techniques within the Realized GARCH framework. Initially, modes are extracted using the decomposition method and categorized into high-frequency, low-frequency, and trend clusters (HLT) using sigmoid functions and k-means clustering. This approach aggregates only the trend cluster modes before applying them to the RGARCH and REGARCH models. The proposed models are evaluated on four financial stock index datasets, demonstrating significant improvements in prediction accuracy and computational complexity over standard time series and machine-learning models across various performance metrics. The model confidence set (MCS) test is used to validate performance.</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

An Integrated Framework for Volatility Prediction: Leveraging Decomposition Techniques with Realized GARCH Models

  • 期刊论文

  • 发表日期：

    2025年5月19日

查看全文
  • Koushik Bera,
  • Prakash Raj,
  • N. Selvaraju

摘要

Forecasting volatility in financial time series with a single model is challenging due to its non-linear, volatile, and noisy nature. This article introduces semi-parametric decomposition-based hybrid models RGARCH-M-J and REGARCH-M-J, where “J” represents the decomposition techniques EEMD, CEEMDAN, or ICEEMDAN. These models enhance volatility forecasting by integrating data decomposition techniques within the Realized GARCH framework. Initially, modes are extracted using the decomposition method and categorized into high-frequency, low-frequency, and trend clusters (HLT) using sigmoid functions and k-means clustering. This approach aggregates only the trend cluster modes before applying them to the RGARCH and REGARCH models. The proposed models are evaluated on four financial stock index datasets, demonstrating significant improvements in prediction accuracy and computational complexity over standard time series and machine-learning models across various performance metrics. The model confidence set (MCS) test is used to validate performance.