Value-at-risk forecasting- based on textual information and a hybrid deep learning-based approach
The recent rise in deep learning and natural language processing (NLP) applications has notably improved productivity across different fields. This research aims to refine Value-at-Risk (VaR) model accuracy by leveraging text mining and deep learning. It first uses NLP to analyze online news sentime...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | en |
| Published: |
Elsevier
2025
|
| Online Access: | http://psasir.upm.edu.my/id/eprint/120553/1/120553.pdf http://psasir.upm.edu.my/id/eprint/120553/ https://linkinghub.elsevier.com/retrieve/pii/S1059056025005660 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The recent rise in deep learning and natural language processing (NLP) applications has notably improved productivity across different fields. This research aims to refine Value-at-Risk (VaR) model accuracy by leveraging text mining and deep learning. It first uses NLP to analyze online news sentiments, integrating these as variables to boost stock market risk forecasts and assess their effect on VaR accuracy. Additionally, the study combines predictions from four unique Generalized AutoRegressive Conditional Heteroskedasticity (GARCH)-type models into advanced Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN)-LSTM models to see if this boosts VaR precision. It also explores how textual data impacts VaR predictions over short and longer periods, using 7 and 20-day rolling windows. The analysis, using S&P500 (SPY), Dow Jones Industrial Average (DJI), and Nasdaq Composite (IXIC) data from 2012 to 2023 alongside news headlines, tests these approaches. The results confirm that incorporating textual information into the VaR model enhances its forecasting accuracy, highlighting the benefits of applying deep learning techniques in this process. |
|---|