Sentence transformers improve topic models for financial risk analysis

From Word Counts to Context: Topic Models for Asset Pricing

Computational Engineering, Finance, and Science

Summary

News articles contain information that can signal risks for financial markets. The authors compare traditional topic modeling methods like LDA with newer context-aware techniques using sentence transformers to better understand these risks. They find that sentence transformers produce more coherent topic groups and slightly better financial results, although the improvements are not conclusive. The work suggests that adding context to analyzing news text could help build better tools for financial decision-making.

What this means in practice

  • For quantitative finance teams: Create improved financial risk factors by using sentence-transformer-based topic models on large news datasets for portfolio construction.
  • For news analytics developers: Develop tools that generate more coherent topic summaries from financial news by applying context-aware sentence embedding techniques.

Tested on one dataset.

Authors

Kevin Foley, Jonathan Hartadi, Shivesh Prakash, Swapnil Vatsal

Abstract

News may reveal systematic risk, but whether its context enhances the construction of systematic risk factors is still unclear. We seek to test whether utilizing a sentence transformer represents an improvement over techniques such as Latent Dirichlet Allocation (LDA) in the coherence of topic term lists generated from unstructured text data. To test this, the same collection of unstructured text data comprising of 394,661 articles and the same downstream financial portfolio construction pipeline were applied with the text layer differing, including the length of article text each model used and how topic terms were ranked: we benchmark LDA against a frozen sentence transformer with k-means clustering. We find that the sentence transformer branch had higher observed scores both in terms of coherence (measured by NPMI) as well as financial performance (measured by Sharpe), although the available tests do not establish outperformance. Further exploratory specifications such as utilizing spherical clustering and multi-horizon exposures had an observed excess-return Sharpe of 1.03 for the combined model. We believe that there is some promise in applying context-aware techniques on unstructured news text, but stricter tests using only information available at each date and broader datasets may be required to enhance the confidence in the observed performance.