Stock Price Prediction Using V-A Sentiment Divergence in Financial News

Β·

Introduction

Financial news serves as a critical source of real-time market insights, influencing investor sentiment and stock prices. This study explores the relationship between V-A (Valence-Arousal) sentiment divergence in financial news and sector-based stock price movements, proposing a predictive model that enhances accuracy by incorporating multidimensional emotional analysis.

Literature Review

Financial News Impact on Stock Markets

Research indicates that sentiment extracted from financial news significantly affects stock volatility. Studies by Xu Wei [6] and Zhao Cheng et al. [7] demonstrate improved prediction accuracy using sentiment indicators alongside traditional stock metrics.

Sentiment Analysis Limitations

Traditional methods relying on average sentiment intensity often overlook emotional diversity, leading to information loss. Siganos et al. [5] highlight the importance of sentiment divergence in capturing nuanced market reactions.

V-A Sentiment Model

The Valence-Arousal model quantifies emotions along two continuous dimensions:

Methodology

1. V-A Sentiment Calculation

A CNN-LSTM hybrid model analyzes news text to compute V-A scores:

  1. CNN Layer: Extracts local n-gram features.
  2. LSTM Layer: Captures long-range semantic dependencies.
  3. Linear Activation: Outputs continuous V-A values (Equation 1–2).

πŸ‘‰ Explore advanced sentiment analysis techniques

2. Sentiment Divergence Metric

Normalized V-A scores classify news as positive (V > 0) or negative (V < 0). Divergence is calculated using:

3. SVR Prediction Model

Inputs:

Experimental Results

Data Sources

Key Findings

ModelMSEMAE
EDS0.2030.383
ES [3]0.2410.421
TEDS0.2610.440

πŸ‘‰ See real-world trading applications

Conclusion

This study demonstrates that V-A sentiment divergence significantly improves sector-level stock predictions. Future work could integrate semi-supervised learning and additional textual features (e.g., syntactic patterns) for enhanced robustness.


FAQs

Q1: Why use V-A over discrete sentiment models?
A1: Continuous V-A scales capture nuanced emotions (e.g., mixed positive-negative states), reducing information loss.

Q2: How is sentiment divergence calculated?
A2: Daily entropy of sector-normalized V-A scores, aggregated from news articles (Equation 6).

Q3: Can this model predict individual stocks?
A3: Currently optimized for sectors; adapting toδΈͺθ‚‘ requires finer-grained news tagging.

Q4: What are the limitations?
A4: Dependence on labeled training data; future work may leverage unsupervised techniques.