Mining Longitudinal Network for Predicting Company Value

Yingzi Jin, Ching-Yung Lin, Yutaka Matsuo, Mitsuru IshizukaIJCAI 2011 · OpenReview

Abstract

Real-world corporate landscapes are inherently dynamic, characterized by continuous collaborations, strategic alignments, and competitive shifts. Traditional financial forecasting often relies on static snapshots or self-reported data, failing to capture the real-time evolution of these relationships. This study introduces a system to infer large-scale evolutionary company networks from public news archives (1981-2009). By integrating longitudinal network metrics with historical financial data, we demonstrate a significant improvement in profit prediction accuracy, outperforming models based solely on financial or network data.

Introduction

Predicting a company’s future value is a cornerstone of investment strategy and industrial analysis. Conventional approaches typically utilize financial analysis (balance sheet metrics) or technical analysis (stock price trends). While Social Network Analysis (SNA) has previously explored how industry positioning affects performance, these studies often rely on static datasets. However, the corporate world is a "messy" ecosystem where mergers and collaborations occur daily. This research addresses the challenge of tracking these shifting relationships over decades to determine if the evolution of a company's network serves as a reliable forecast for its financial health.

Methodology

The researchers developed a two-stage pipeline to transform unstructured news into predictive insights:

  1. Network Extraction: Using a corpus of New York Times articles from 1981 to 2009, the system identifies "impact" links between companies based on co-occurrence. A weighted scoring system, similar to TF-IDF, prioritizes links where companies appear together in articles with few other entities, suggesting a more direct strategic relationship.
  2. Feature Generation & Prediction: For each company, a 72-dimensional "network effect" vector was generated, capturing metrics such as connectivity, betweenness, and distance. Crucially, the model calculates "Delta Changes"—the rate at which these metrics shift year-over-year. These features, combined with historical financials, were processed using a Support Vector Regression (SVR) model to predict future profit growth.
Evolutionary company networks extracted from news articles, showing shifts in IBM's industry relationships over time.

Key Insights

  • Impact Over Sentiment: The study suggests that the density and structure of relationships are more predictive than sentiment. Even high competition indicates a company's significant industry impact.
  • Leading Indicators: While financial data is a "lagging" indicator of past performance, network shifts extracted from news serve as "leading" indicators of strategic changes that have yet to manifest on balance sheets.
  • The Power of Change: The "delta-change" in network position was identified as one of the most critical features, as sudden shifts in media mentions often precede financial volatility.

Results and Evaluation

The joint model (Network + Financials) demonstrated superior accuracy, beating network-only models by 150% and financial-only models by 34%. The system was particularly effective for established industry leaders. For instance, the model captured 88.8% of the variance in IBM’s profit trends over several decades ($r^2 = 0.888$).

Profit prediction results for IBM (a) and Intel (b), comparing real values against model predictions.

Conclusion

This research represents the first longitudinal network-mining approach to company performance analysis. By treating the corporate world as a dynamic, evolving network, the model provides a more nuanced view of company value. Future work will focus on refining co-occurrence logic to distinguish between different types of strategic events and moving toward real-time processing to capture value shifts as news breaks.

Summary updated

Abstract

Real-world social networks are dynamic in nature. Companies continue to collaborate, align strategically, acquire, and merge over time, and receive positive/negative impact from other companies. Consequently, their performance changes with time. If one can understand what types of network changes affect a company's value, he/she can predict the future value of the company, grasp industry innovations, and make business more successful. However, it often requires continuous records of relational changes, which are often difficult to track for companies, and the models of mining longitudinal network are quite complicated. In this study, we developed algorithms and a system to infer large-scale evolutionary company networks from public news during 1981-2009. Then, based on how networks change over time, as well as the financial information of the companies, we predicted company profit growth. This is the first study of longitudinal network-mining-based company performance analysis in the literature
BibTeX
@inproceedings{jin2011mining,
  author = {Yingzi Jin and Ching-Yung Lin and Yutaka Matsuo and Mitsuru Ishizuka},
  title = {{Mining Longitudinal Network for Predicting Company Value}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2011},
  url = {https://openreview.net/forum?id=By4bxmM_WB},
}
Figure