EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

Return-Based Volatility Proxy Forecasting via a Residual-Correction LSTM–GARCH Hybrid Framework

Zhenxiang Zhang

Journal of Innovation and Development · 07.07.2026 · 16 (1) · 21-28

DOI: 10.54097/4wx6m038 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Innovation and Development, online erschienen am 07.07.2026. Verfasst von Zhenxiang Zhang. Die Arbeit nennt 15 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Forecasting return-based volatility proxies is challenging in financial markets due to volatility clustering, heavy tails, and nonlinear dynamics. This paper proposes a residual-correction LSTM–GARCH hybrid framework, where GARCH provides a robust econometric backbone, while LSTM learns the nonlinear residual in the log variance-ratio space and performs multiplicative correction. Experiments on a representative ETF show that the proposed RC-LSTM–GARCH achieves better RMSE, MAE, and QLIKE than the GARCH baseline. The proposed design aims to combine the interpretability and stability of GARCH with the nonlinear representation ability of LSTM, thereby improving calibration under volatility-specific losses, and the framework can be extended to multi-asset and multi-horizon risk forecasting.

Zitieren

APA

Zhang, Z. (2026). Return-Based Volatility Proxy Forecasting via a Residual-Correction LSTM–GARCH Hybrid Framework. Journal of Innovation and Development, 16(1), 21-28. https://doi.org/10.54097/4wx6m038

Vancouver

Zhang Z. Return-Based Volatility Proxy Forecasting via a Residual-Correction LSTM–GARCH Hybrid Framework. Journal of Innovation and Development. 2026;16(1):21-28. doi:10.54097/4wx6m038

BibTeX
@article{Zhang2026,
  author = {Zhang, Zhenxiang},
  title = {Return-Based Volatility Proxy Forecasting via a Residual-Correction LSTM–GARCH Hybrid Framework},
  journal = {Journal of Innovation and Development},
  year = {2026},
  volume = {16},
  number = {1},
  pages = {21--28},
  doi = {10.54097/4wx6m038}
}
Kostenlos starten →