Scholarlink

<p>Forecasting the volatility of financial assets—defined as the degree of their price variation over time—is a trending topic in financial research. Enhancing prediction accuracy is crucial in this field, as an asset’s volatility is widely used to assess the risk associated with its returns. In this context, we introduce a novel hybrid model that integrates traditional econometric techniques, specifically GARCH models, with the Temporal Fusion Transformer (TFT), a cutting-edge deep learning architecture. We designed such a model for analyzing Exchange Traded Funds (ETFs) composed of assets from the S&amp;P 500, a benchmark index tracking 500 large US companies therefore reflecting the overall health and trends of the stock market, across various sectors. Utilizing volatility proxies such as historical volatility (HV) and the Garman–Klass (GK) method, our study demonstrates that the hybrid GARCH-TFT model significantly outperforms alternative models in forecasting the GK proxy and achieves performance comparable to the stand-alone TFT model for HV, underscoring the potential of merging machine learning approaches with traditional econometric methods to enhance predictive precision in volatile financial markets.</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

A GARCH-temporal fusion transformer model for the volatility prediction of exchange traded funds

  • 期刊论文

  • 发表日期：

    2025年8月4日

查看全文
  • Lorenzo Petrosino,
  • Luca Bacco,
  • Giuliano Salvati,
  • Mario Merone,
  • Marco Papi

摘要

Forecasting the volatility of financial assets—defined as the degree of their price variation over time—is a trending topic in financial research. Enhancing prediction accuracy is crucial in this field, as an asset’s volatility is widely used to assess the risk associated with its returns. In this context, we introduce a novel hybrid model that integrates traditional econometric techniques, specifically GARCH models, with the Temporal Fusion Transformer (TFT), a cutting-edge deep learning architecture. We designed such a model for analyzing Exchange Traded Funds (ETFs) composed of assets from the S&P 500, a benchmark index tracking 500 large US companies therefore reflecting the overall health and trends of the stock market, across various sectors. Utilizing volatility proxies such as historical volatility (HV) and the Garman–Klass (GK) method, our study demonstrates that the hybrid GARCH-TFT model significantly outperforms alternative models in forecasting the GK proxy and achieves performance comparable to the stand-alone TFT model for HV, underscoring the potential of merging machine learning approaches with traditional econometric methods to enhance predictive precision in volatile financial markets.