At the International Conference in Banking and Financial Studies (ICBFS) 2025, held 11–12 September at the Catholic University of the Sacred Heart, Piacenza, Italy, Salford Business School’s Dr Zeeshan Ali Syed, Programme Director for Financial Technology (FinTech) and Lecturer in Finance, presented new research on unsupervised machine learning for portfolio management.
“Markets change fast. By letting the data tell us which assets truly move together—and then optimising within those groups—we can help investors stay diversified, responsive, and understandable in real time,” said Zeeshan.
About the paper
Co-authored with Dr Mohammad Rahman, Cynthia Akiotu, and Rasol Eskandari, the paper “How to manage portfolios in different clusters? An unsupervised machine learning approach” shows how clustering techniques can help investors build portfolios that adapt to changing market conditions, rather than relying on static, sector-based allocations.
“In an era characterised by dynamic and often volatile markets, there is an urgent need for innovative investment strategies,” the authors argue. “K-means clustering is increasingly reshaping how asset allocation is approached within portfolios.”
The study presents evidence that algorithmic classification of financial assets (e.g., equities) can improve a portfolio’s risk-return profile. Using a decade of market data, the team compared conventional diversification with a data-driven approach that:
- Uses K-means clustering to identify natural groupings of assets with similar return patterns.
- Blends Monte Carlo simulation with deterministic optimisation (SLSQP – Sequestional Least Squares Programming) to search for stable, implementable weights; and
- Evaluates outcomes with familiar, investor-friendly metrics (Sharpe, Sortino, expected return, minimum variance).
Why it matters
A key strength of the approach is that it can be re-estimated as market regimes evolve, updating automatically as new data arrive—offering a practical route to adaptive and explainable portfolio construction. Beyond the conference paper, Ambarin highlighted broader work linking high-frequency financial data to portfolio decision-making, where unsupervised learning can rapidly surface patterns that are hard to see with traditional tools.
As artificial intelligence (AI) becomes integral to finance, explainable and adaptive methods are vital for both practitioners and educators. This research brings together robust statistics, optimisation, and clear, actionable outputs that can be taught in the classroom and applied by investment teams.
