金融研究arXiv q-fin12 Oct, 12:00 pm SGTSIGNAL 15E61B

Enhancing Credit Risk Scorecards with Transactional Behavioural Features

ORIGINAL / Enhancing credit risk scorecards with transactional-based behavioural features

This study targets transaction narrative text, which credit risk modelling has long overlooked, and proposes an explainable method to convert it into behavioural features. Its novelty lies in using natural language processing and regular expressions to build a categorisation algorithm that maps unstructured transaction descriptions into interpretable spending categories, and in validating their incremental predictive value within a traditional scorecard framework. For quantit

01 ABSTRACT

Using transaction data from small and medium enterprises at a South African financial institution, the study applies natural language processing and regular expressions to transform unstructured transaction descriptions into interpretable spending-category behavioural features. The authors report that the new categorisation algorithm achieves high classification accuracy, and that adding these features to a credit risk scorecard significantly increases the Gini compared with a scorecard without them. The authors conclude that text-based transactional information can be systematically extracted, categorised, and operationalised to improve credit risk model performance; this conclusion reflects the authors' view based on their experimental setting and still requires external validation before deployment.

02 KEY FINDINGS

The study focuses on payment narrative text in SME transaction data as an alternative data source for credit risk modelling.
Methodologically, it combines natural language processing and regular expressions to develop and evaluate a structured transaction categorisation algorithm that produces explainable spending-category behavioural features.
The authors report high classification accuracy for the algorithm and a significant Gini increase when these features are added to a scorecard versus a version without them.

03 RELEVANCE

If the findings are reproducible, financial institutions could convert long-idle raw transaction descriptions into explainable behavioural variables without relying on black-box models, supplementing traditional credit history and repayment behaviour indicators. This may improve risk discrimination for SME borrowers and provide a reference for compliant use of alternative data in scorecard systems.

04 EVIDENCE BOUNDARY

The study relies on SME transaction data from a single South African financial institution, so sample and regional specificity may limit generalisation to other markets, customer segments, or institutions. The reported categorisation accuracy and Gini improvement are results under the authors' setting; it is not stated whether they were validated across time or out-of-sample. The abstract provides no specific values, benchmark model details, or ablation experiments, so the robustness and economic significance of the gains cannot be independently assessed. Feature interpretability depends on the categorisation scheme design, and results may change if category definitions change.

Return to the primary source

AI GENERATED SUMMARY / DISCOVERED BY ARXIV Q-FIN

Read original ↗