Investment Universe Complex Network: A Framework for Optimizing Asset Selection in Dynamic Financial Markets
Machine Learning, vol.115, no.8, 2026 (SCI-Expanded, Scopus)
- Publication Type: Article / Article
- Volume: 115 Issue: 8
- Publication Date: 2026
- Doi Number: 10.1007/s10994-026-07124-9
- Journal Name: Machine Learning
- Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Aerospace Database, Applied Science & Technology Source, Compendex, INSPEC, MathSciNet, zbMATH, Engineering Source (EBSCO), Pharma Collection (ProQuest), Technology Collection (ProQuest)
- Keywords: Complex networks, Dynamic network modeling, Financial networks, Portfolio allocation
- Galatasaray University Affiliated: Yes
Abstract
Financial systems are dynamic and complex structures influenced by numerous factors. Identifying an optimal portfolio allocation that maximizes returns within this evolving environment is a crucial challenge. In this study, we introduce InUCon, a network-science-based framework designed to address this problem. InUCon models assets as nodes in a network, where interactions between nodes change over time. By identifying communities within this dynamic network, InUCon selects optimal asset subsets. Our approach constitutes a methodological synthesis that integrates several techniques whose combined application in financial network–based portfolio construction has not yet been fully explored: (i) establishing asset relationships by harmonizing company description and price changes, (ii) tracking their temporal evolution to identify relevant communities, and (iii) selecting risk-minimizing assets using network topological metrics. Our key contribution lies in providing a systematic method for applying network science principles to real-world financial systems. Experimental results indicate that InUCoN, particularly when based on hybrid similarity of company descriptions and price changes, achieves higher average returns–exceeding benchmark methods by over 23%. However, risk-adjusted performance remains statistically comparable to the benchmarks, suggesting no significant superiority in terms of Sharpe ratio. Instead, the results highlight that InUCoN exhibits distinct distributional characteristics, indicating a different return–risk profile rather than a uniformly dominant performance.