Multi-time scale anomaly detection and dynamic portfolio optimization with machine learning
- 13 June 2026
- journal article
- Published by World Scientific Pub Co Pte Ltd in International Journal of Financial Engineering
- Vol. 13 (03)
Abstract
Using the financial parameters of firms listed on the Borsa Istanbul BIST 30 index from 2004 to 2024, this study aims to identify anomalies in various time periods and calculate dynamic portfolio weights based on these findings. Beyond typical portfolio techniques based on fixed assumptions, a strategy responsive to market conditions has been created in light of the turbulent character of financial markets. In addition to fundamental financial measurements, anomaly scores were developed annually for each business in the research and utilized in portfolio development. The Random Forest technique was utilized to ascertain the variable significance levels in this case, whereas the Isolation Forest approach was employed for anomaly identification. The dynamic portfolio outperformed the traditional equal-weighted approach in the majority of years, according to comparative studies utilizing criteria like return, Sharpe ratio, and maximum value loss. Specifically, it was discovered that anomaly scores contributed to the attainment of risk-return equilibrium. The weights assigned to debt and market valuation ratios in the model have influenced portfolio performance, and their effects on returns are now clearly visible. The application findings suggest that portfolio management can benefit from judgments based on risk indicators that fluctuate over time in addition to historical data.Keywords
This publication has 15 references indexed in Scilit:
- A novel approach for dynamic portfolio management integrating K-means clustering, mean-variance optimization, and reinforcement learningKnowledge and Information Systems, 2025
- A Machine Learning Approach to Detecting Financial Anomalies in Large Global Companies2024
- Dynamic portfolio rebalancing through reinforcement learningNeural Computing and Applications, 2021
- Generalized isolation forest for anomaly detectionPattern Recognition Letters, 2021
- Anomaly detection using machine learning techniquesMalaya Journal of Matematik, 2020
- An Improved Data Anomaly Detection Method Based on Isolation Forest2017
- A survey of anomaly detection techniques in financial domainFuture Generation Computer Systems, 2016
- Isolation-Based Anomaly DetectionACM Transactions on Knowledge Discovery From Data, 2012
- Isolation Forest2008
- Data Mining techniques for the detection of fraudulent financial statementsExpert Systems with Applications, 2007