Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2410.13100 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Table of Contents:
- We develop and evaluate a family of discrete-time logit-link (LLink) models (including fixed-effects and frailty extensions) to capture latent heterogeneity in repayment behaviour and quantify the effects of socio-temporal factors in microfinance. Our findings highlight the importance of unobserved borrower risk, revealing that simple random intercept structures are sufficient to model latent heterogeneity in this context. Additionally, socio-temporal variables--such as festive seasons and long school breaks--consistently associate with delinquency transitions, offering key insights into repayment dynamics. While LLink models provide clear interpretability, tree-based methods outperform them in predictive accuracy, making them suitable for multistate classification tasks. Building on this, we propose an optimised classification strategy based on the Matthews Correlation Coefficient to enhance next-state prediction. Overall, our results highlight the benefit of combining interpretable risk modeling with advanced machine learning to support robust, data-driven decision-making in microfinance operations.