Hidden neighbours: extracting industry momentum from stock networks

This paper introduces an innovative method for constructing industry momentum portfolios by leveraging two stock networks: one based on stock price correlations and the other on corporate text similarity. We find that these networks capture different aspects of company relationships, motivating us to combine them and form a portfolio that exploits less visible industry momentum. Our Hidden Neighbours portfolio, analysed from 2013 to 2022, delivered an annualised return of 18.16% with a Sharpe ratio of 0.85, outperforming the S&P 500 and other traditional momentum strategies. Factor decomposition attributes returns primarily to the idiosyncratic factor \(\alpha\) $\alpha$ . Our study employs interdisciplinary methods, merging network analysis and Natural Language Processing (NLP) techniques for portfolio construction. Utilising advanced text embedding models, we enhance portfolio construction by integrating textual insights from corporate disclosures into stock networks. The paper offers a comprehensive strategy across diverse data and the interdisciplinary approach, uniting financial theory, network science, and NLP, advances both theory and practice of portfolio management.