Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MANA-Net: Mitigating Aggregated Sentiment Homogenization with News Weighting for Enhanced Market Prediction

Mengyu Wang, Tiejun Ma

TL;DR

The paper tackles Aggregated Sentiment Homogenization in news-driven market prediction by introducing MANA-Net, which uses a market-news attention mechanism to assign dynamic, trainable weights to individual news items before aggregation. By integrating weighted sentiment aggregation with the prediction model, MANA-Net optimizes sentiment representations directly for price-trend prediction and handles fluctuating news volumes. Empirical results on S&P 500 and NASDAQ 100 datasets (2003–2018) with TRNA data show consistent improvements in Profit & Loss and Sharpe Ratio over static baselines and several advanced methods, including GPT-based approaches, highlighting the value of market-informed weighting. The approach yields interpretable news weights, demonstrates robustness across horizons and models (notably ML PNs), and offers practical scalability for real-world deployment in finance.

Abstract

It is widely acknowledged that extracting market sentiments from news data benefits market predictions. However, existing methods of using financial sentiments remain simplistic, relying on equal-weight and static aggregation to manage sentiments from multiple news items. This leads to a critical issue termed ``Aggregated Sentiment Homogenization'', which has been explored through our analysis of a large financial news dataset from industry practice. This phenomenon occurs when aggregating numerous sentiments, causing representations to converge towards the mean values of sentiment distributions and thereby smoothing out unique and important information. Consequently, the aggregated sentiment representations lose much predictive value of news data. To address this problem, we introduce the Market Attention-weighted News Aggregation Network (MANA-Net), a novel method that leverages a dynamic market-news attention mechanism to aggregate news sentiments for market prediction. MANA-Net learns the relevance of news sentiments to price changes and assigns varying weights to individual news items. By integrating the news aggregation step into the networks for market prediction, MANA-Net allows for trainable sentiment representations that are optimized directly for prediction. We evaluate MANA-Net using the S&P 500 and NASDAQ 100 indices, along with financial news spanning from 2003 to 2018. Experimental results demonstrate that MANA-Net outperforms various recent market prediction methods, enhancing Profit & Loss by 1.1% and the daily Sharpe ratio by 0.252.

MANA-Net: Mitigating Aggregated Sentiment Homogenization with News Weighting for Enhanced Market Prediction

TL;DR

The paper tackles Aggregated Sentiment Homogenization in news-driven market prediction by introducing MANA-Net, which uses a market-news attention mechanism to assign dynamic, trainable weights to individual news items before aggregation. By integrating weighted sentiment aggregation with the prediction model, MANA-Net optimizes sentiment representations directly for price-trend prediction and handles fluctuating news volumes. Empirical results on S&P 500 and NASDAQ 100 datasets (2003–2018) with TRNA data show consistent improvements in Profit & Loss and Sharpe Ratio over static baselines and several advanced methods, including GPT-based approaches, highlighting the value of market-informed weighting. The approach yields interpretable news weights, demonstrates robustness across horizons and models (notably ML PNs), and offers practical scalability for real-world deployment in finance.

Abstract

It is widely acknowledged that extracting market sentiments from news data benefits market predictions. However, existing methods of using financial sentiments remain simplistic, relying on equal-weight and static aggregation to manage sentiments from multiple news items. This leads to a critical issue termed ``Aggregated Sentiment Homogenization'', which has been explored through our analysis of a large financial news dataset from industry practice. This phenomenon occurs when aggregating numerous sentiments, causing representations to converge towards the mean values of sentiment distributions and thereby smoothing out unique and important information. Consequently, the aggregated sentiment representations lose much predictive value of news data. To address this problem, we introduce the Market Attention-weighted News Aggregation Network (MANA-Net), a novel method that leverages a dynamic market-news attention mechanism to aggregate news sentiments for market prediction. MANA-Net learns the relevance of news sentiments to price changes and assigns varying weights to individual news items. By integrating the news aggregation step into the networks for market prediction, MANA-Net allows for trainable sentiment representations that are optimized directly for prediction. We evaluate MANA-Net using the S&P 500 and NASDAQ 100 indices, along with financial news spanning from 2003 to 2018. Experimental results demonstrate that MANA-Net outperforms various recent market prediction methods, enhancing Profit & Loss by 1.1% and the daily Sharpe ratio by 0.252.
Paper Structure (23 sections, 8 equations, 4 figures, 5 tables)

This paper contains 23 sections, 8 equations, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Financial news sentiment analysis
  4. News-based market prediction
  5. Attention Mechanism
  6. Preliminaries
  7. Problem Formulation
  8. Data
  9. Sentiment Aggregation Challenges
  10. Methodology
  11. Weighted News Aggregation
  12. Market Prediction and Training
  13. Experiment Settings
  14. Evaluation Metrics
  15. Baseline Methods
  16. ...and 8 more sections

Figures (4)

  • Figure 1: The kernel density estimate plot illustrates the distributions of sentiment scores for both individual news items and their daily average values.
  • Figure 2: The boxplot of positive sentiment scores for individual news items across 30 randomly sampled days. Green triangles represent the daily average scores, while red lines indicate the median values. The blue rectangle highlights the range of averaged sentiments.
  • Figure 3: An overview of MANA-Net. The market prices and news sentiments data are as defined in Section \ref{['sec:formulation']}. Blue arrows show the feed-forward process and grey arrows show the back-propagation process.
  • Figure 4: The kernel density esimate plot of MANA-Net's news weights on the test set.