Vikram Dewangan’s Post

View profile for Vikram Dewangan, graphic

Analyst@Genpact | Looking for opportunities in Credit Risk Modelling (PD,LGD,EAD), Model risk management | IFRS9 | BASEL | Scorecards | IRB

🎉Excited to share a project on end-to-end scorecard modelling. 📊 Project Overview: This project encompasses the modelling of loss components, including Probability of Default (PD), Loss Given Default (LGD), Exposure at Default (EAD), and the calculation of Expected Loss. Here’s a closer look at what we covered. Topics Covered: 1. PD Modelling Using Logistic Regression: - Implemented logistic regression to predict the probability of default for each borrower. 2. Weight of Evidence (WoE) and Information Value (IV): - These techniques were used for feature selection. 3. Fine Classing and Coarse Classing: - Applied these methods to create bins for independent variables. 4. Finding Credit Score Using PD: - Credit scores were derived based on the calculated probabilities of default. 5. Model Validation: - Validated the model using various metrics such as the Confusion Matrix, ROCAUC, Gini Coefficient, and Kolmogorov-Smirnov. 6. Model Monitoring Using Population Stability Index (PSI): - PSI was used to ensure model stability over time and detect any significant shifts in the population. 7. LGD Modelling: - Employed a two-stage modeling approach: Logistic regression for cases with a recovery rate of 0 and linear regression for cases with a recovery rate above 0. 8. EAD Modelling Using Linear Regression: - Linear regression was used to model the exposure at default, providing insights into the potential exposure risk. 🛠️ Future Improvements: 1. Data Segmentation: - By segmenting the data using various variables, we can develop more tailored models for different segments, leading to better predictive accuracy. 2. Advanced ML Algorithms: - Implementing advanced machine learning techniques like bagging and boosting could enhance model performance, but at the cost of model interpretation. 3. Beta Regression for LGD and EAD: - Considering beta regression for LGD and EAD modelling could be beneficial, as these values typically lie between 0 and 1. 4. Reject inferencing: - Currently, our data set only includes approved loans. Access to rejected loan data could improve model accuracy. 🚀 Explore the Project: You can find more details and explore the code on GitHub. (https://lnkd.in/giy6rxyt) #creditriskmodelling #datascience #creditrisk #scorecard #pd #ecl #modelvalidation

To view or add a comment, sign in

Explore topics