---
title: Information Flow and Market Risk: Transformer-Based Evidence on Sentiment and Volatility Dynamics
url: https://www.ml-quant.com/papers/arxiv/2610.39636/
site: ML-Quant (https://www.ml-quant.com)
updated: 2026-10-30
license: Summaries CC BY 4.0; links go to the original sources
index: https://www.ml-quant.com/llms.txt
identifier: arXiv:2610.39636
source_url: https://arxiv.org/abs/2610.39636
featured: 2026-10-30
citations: unknown
topic: Econometrics & Forecasting
---


# Information Flow and Market Risk: Transformer-Based Evidence on Sentiment and Volatility Dynamics

Transformer-based news sentiment does not improve volatility forecasts beyond persistence and market indicators in walk-forward tests across eleven stocks and multiple horizons.

- Source: https://arxiv.org/abs/2610.39636
- Identifier: arXiv:2610.39636
- Released: 2026-10-30
- First featured: Quant Letter No. 137 (2026-10-30): https://www.ml-quant.com/issues/2026-10-30/
- Citations (Semantic Scholar): not tracked
- Published in: not yet
- Topic: Econometrics & Forecasting
- Authors: Szymon Lis, Robert Ślepaczuk, Paweł Sakowski

## Abstract (arXiv, CC0)

This study examines whether transformer-based financial news sentiment improves equity volatility forecasts beyond conventional market-based predictors. Using data for eleven US stocks, we compare volatility persistence models with specifications incorporating the VIX, AAII investor sentiment, and news sentiment extracted using ModernFinBERT. The forecasting target is annualized volatility constructed from future averages of trailing 252-session Yang-Zhang estimates, evaluated at horizons of 1, 5, and 22 trading days. Models are assessed through walk-forward testing with expanding and rolling estimation windows, fixed initial hyperparameter selection, and statistical comparisons accounting for multiple testing and serial dependence. Results show that a naive persistence benchmark achieves the lowest average QLIKE loss at all horizons. Although HAR-VIX produces the largest number of individual forecasting wins under expanding windows, it does not achieve the lowest mean loss. Adding text-based sentiment improves HAR forecasts in only two of 33 expanding-window comparisons, and neither improvement remains statistically significant under longer HAC bandwidths. A blinded audit of 2,700 news texts further characterizes the agreement and confidence patterns of ModernFinBERT classifications. The findings do not establish a general incremental forecasting advantage from transformer-based news sentiment. They highlight the importance of strong benchmarks, temporal alignment, robust inference, and careful evaluation of text-derived predictors.

## Related

- [Bridging Short- and Long-Term Dependencies: A CNN-Transformer Hybrid for Financial Time Series Forecasting](https://www.ml-quant.com/papers/arxiv/2504.19309/): The paper suggests a hybrid architecture combining Convolutional Neural Networks and Transformers for efficient short- and long-term financial time series data modeling, showing improved performance in intraday stock price prediction.
- [Cross-Modal Temporal Fusion for Financial Market Forecasting](https://www.ml-quant.com/papers/arxiv/2504.13522/): The article presents Cross-Modal Temporal Fusion (CMTF), a new transformer-based system that uses diverse financial data to enhance the precision of financial market predictions.
- [Hidformer: Transformer-Style Neural Network in Stock Price Forecasting](https://www.ml-quant.com/papers/arxiv/2412.19932/): The research explores the application of the Hidformer model, a Transformer-based neural network, in stock price forecasting, combining technical analysis and advanced machine learning for improved accuracy.
- [Causal Diffusion Transformers for Generative Modeling](https://www.ml-quant.com/papers/arxiv/2412.12095/): The article discusses Causal Diffusion, a framework that enhances diffusion models' performance and allows a seamless shift between autoregressive and diffusion generation modes, achieving top results on the ImageNet generation benchmark.
- [Systematic comparison of deep generative models applied to multivariate financial time series](https://www.ml-quant.com/papers/arxiv/2412.06417/): The study contrasts deep generative models (DGMs) and parametric models for creating financial time series, highlighting the advantages of DGMs in an implied volatility trading task.
- [SST: Multi-Scale Hybrid Mamba-Transformer Experts for Time Series Forecasting](https://www.ml-quant.com/papers/arxiv/2404.14757/): The paper introduces the State Space Transformer model for time series forecasting, which effectively captures global and local patterns, offering superior performance with less memory and computational cost.