Alam, Mohammad Kowshik and Shuvo, Md Sabbir Hossen and Fahad, Md Lutfur Rahman (2023) Liquidity Withdrawal Dynamics in SME Working Capital Lending: A Random Forest–Based Stress Simulation Using Probability Threshold Shifts and Approval Rate Contraction Metrics. American Journal of Economics and Business Management, 6 (10). pp. 327-350. ISSN 2576-5973

Text AJEBM_Liquidity+Withdrawal+Dynamics (1).pdf - Published Version Download (1MB)

Abstract

The Small and Medium Enterprises (SMEs) are important stakeholders in the economic development process as they are contributing to employment creation, industrial productivity and innovation in the business world. But SMEs are very sensitive to working capital financing to carry on their daily operations and therefore they are very much affected when the banking industry is under financial pressure and expecting liquidity withdrawals. The study explores the dynamics of withdrawing liquidity from the SME working capital loan in the context of stress simulation under the framework of the machine learning-based classification model (Random Forest). The study is designed to test the effect of tightening the credit conditions on the credit accessibility of SMEs by using probability threshold shifts and contraction of approval rates metrics. The data set for the research includes financial characteristics of the borrowers, attributes of the loan, repayment history, employment information, and past default information. A baseline model is trained to estimate the probabilities of SME loan default, using a Random Forest (RF) model. The probability of borrowers defaulting is artificially raised to reflect greater liquidity pressure and uncertainty in underwhelming conditions. At the same time, lending approval criteria are adjusted to take account of tougher credit policies that financial institutions are likely to follow in times of liquidity shortage. The study then compares the approval rate and volume of loans, before and after the stress implementation. Results indicate that the rise in the level of default and the improbability that a loan would be approved have a significant negative impact on both the number of SME loan applications approved and the amount of working capital financing made available. The approval rates under stressed scenarios are significantly lower, showing the sensitivity of SME lending activity to shrinkage of liquidity. The findings also suggest that AI-powered predictive models are suitable tools that can be used to assist banking institutions in their stress test, credit risk assessment, and proactive liquidity management. This research combines the power of machine learning with stress simulation methods, enhancing the field of financial analytics and offering a valuable practical approach to assessing lending resilience under economic stress. The proposed method provides guidance to financial institutions, policy makers, and researchers to enhance the stability of the SME financing market in liquidity stress.

Item Type:	Article
Subjects:	A General Works > AI Indexes (General)
Depositing User:	admin eprints
Date Deposited:	08 Jun 2026 17:20
Last Modified:	13 Jun 2026 07:00
URI:	http://eprints.umsida.ac.id/id/eprint/16550

Actions (login required)

View Item