Please use this identifier to cite or link to this item: https://hdl.handle.net/10419/312901 
Year of Publication: 
2025
Series/Report no.: 
Graduate Institute of International and Development Studies Working Paper No. HEIDWP05-2025
Publisher: 
Graduate Institute of International and Development Studies, Geneva
Abstract: 
Forecasting banking system liquidity is crucial for the effective monetary policy implementation. This study investigates the effectiveness of various econometric and machine learning models in predicting the autonomous factors of banking system liquidity. The research compares widely used econometric models such as SARIMA, Exponential Smoothing, and Prophet alongside machine learning models like Random Forest, applying various preprocessing techniques, including power transformations, scaling, and trend-cycle decomposition. Moreover, ensemble methods, like weighted blending and stacking, were used to improve accuracy. Experimental results indicate that SARIMA was the best individual model, but ensemble with Prophet and Random Forest further improved forecast performance. Neural network models underperformed potentially due to challenges in optimizing their architectures. Future research intends to explore multivariate and structural models, as well as advanced neural architectures, to enhance predictive accuracy.
Subjects: 
Monetary Policy
Time-Series Models
Model Evaluation and Selection
Forecasting and Other Model Applications
Payment Systems
JEL: 
E52
C22
C52
C53
E42
Document Type: 
Working Paper

Files in This Item:
File
Size





Items in EconStor are protected by copyright, with all rights reserved, unless otherwise indicated.