CEU eTD Collection (2025); Takacs, Karoly: Portfolio Cash Liquidity Forecasting with Machine Learning

CEU Electronic Theses and Dissertations, 2025
Author Takacs, Karoly
Title Portfolio Cash Liquidity Forecasting with Machine Learning
Summary This project investigates the application of machine learning (ML) to cash liquidity forecasting for UCITS-regulated mutual fund portfolios at a leading European asset management firm. Leveraging anonymized daily data from equity, fixed income, and multi-asset portfolios, the study compares traditional statistical models (ARIMA) with advanced ML techniques (Random Forest, XGBoost) to predict next-day (T+1) liquidity. Key features include portfolio cash positions, investor flows, liquidity metrics, and macroeconomic variables. The models were evaluated using strict rolling-window validation, with interpretability enhanced by SHapley Additive exPlanations (SHAP) and distilled linear approximations. Results show that ML models, particularly Random Forest, significantly outperform ARIMA in predictive accuracy, especially for equity funds. It underscores the necessity of interpretability and data quality for successful adoption in regulated finance. These findings provide a foundation for further enhancement of ML-driven liquidity forecasting, emphasizing their role as decision-support tools.
Supervisor Arino de la Rubia Eduardo
Department Economics MSc
Full texthttps://www.etd.ceu.edu/2025/takacs_karoly.pdf

Visit the CEU Library.

© 2007-2025, Central European University