Facing unprecedented challenges during the COVID-19 pandemic, Kashf Foundation sought to understand the shifting repayment behaviors of its clients. Recognizing Statistique’s proven expertise in data-driven insights and risk assessment, they partnered with us to gain actionable recommendations for navigating these uncertainties.
Project Overview:
Client: Kashf Foundation, a pioneering microfinance institution in Pakistan.
Objective: To analyze client repayment patterns and develop an SVM-based credit scoring model to optimize loan disbursement decisions by identifying repayment risks. The project encompassed data from multiple branches, focusing on categorizing loans into two risk groups,
- Secure: Payments made on time
- Not secure: Payments delayed or overdue
Our Approach:
Data Analysis and Feature Selection: Measured salient changes in client behavior to identify trends and patterns that might be advantageous for the organization to facilitate risk management. Key indicators like loan amount, loan cycle number and family size were identified as significant predictors of repayment behaviors. Outdated factors, such as business duration and prior delinquencies, were excluded due to reduced relevance in the current dataset.
Predictive Credit Scoring Model: Created an SVM model designed to assess repayment risks by analyzing features like net disposable income (NDI), repayment history and loan utilization ratio. Credit scores were calibrated to align with default probabilities, providing actionable insights for loan approval decisions.
Key Insights:
- Leveraging Existing Model Stability: Given the volatility of the recent data, instead of retraining the model immediately, we recommended allowing time for client behavior to stabilize post-pandemic. Premature retraining could disrupt the stability and learning of previous models. It was likely that these shifts were only temporary.
- Granular Sampling for Better Insights: Refined the branch sampling technique to capture a more detailed picture of high-risk loans. Expanding the dataset by considering loans overdue over shorter timeframes would allow for better identification of branches with higher delinquency rates, enhancing predictive accuracy.
- Observe for Normalization: We tracked whether these features showed any sign of normalization as the market stabilized (with the easing of lockdowns and other changes). During future retraining efforts, we anticipated the data to either return to its original distribution or the emergence of new significant factors that will aid in correctly identifying the risk category.
Why Statistique?
In times of unprecedented challenges, businesses require more than standard solutions. Statistique offers a distinctive approach, combining deep analytical expertise with industry-specific knowledge to deliver insights that drive impactful decisions. Our commitment to understanding the unique needs of each client ensures that we provide customized strategies, enabling organizations to navigate complexities and achieve sustainable success.
