Advanced Credit Risk Modeling for Basel and IFRS 9 using R, Python and SAS

📅 June 20-21, 9am-5pm
🌍 English

Introduction

This comprehensive advanced training to practical credit risk modeling provides a targeted training guide for risk professionals looking to perfect, validate and stress test their probability of default (PD), loss given default (LGD) or exposure at default (EAD) models in a Basel or IFRS 9 context.  Combining theory with practice, this course starts with a brief recap on PD, LGD and EAD modeling.  Various modeling approaches for dealing with Low Default Portfolios are discussed.  We then elaborate on model validation and discuss backtesting and benchmarking of PD, LGD and EAD models together with qualitative validation.  Stress testing is extensively covered by reviewing various approaches to understand the behavior of your credit risk models under adverse economic circumstances.  The last section of the course is based upon our recent research on credit risk modeling and reviews some advanced modeling topics.

Throughout the course, the instructor(s) extensively report on their recent scientific findings and international consulting experience.  The course provides a sound mix of both theoretical and technical insights, as well as practical implementation details.  These are illustrated by several real-life case studies, reference models, state of the art research insights and benchmarks.

Outline

Day 1 (9am -5pm)

• Recap (short)
  • Modeling PD
  • Modeling LGD and EAD
  • The Basel Capital Accord
  • IFRS 9
• Low Default Portfolios
  • Basic Concepts
  • Dealing with skewed data sets
  • Varying the sample window
  • Undersampling
  • Oversampling
  • Synthetic Minority Oversampling Technique (SMOTE)
  • Adjusting posterior probabilities
  • Cost sensitive learning
  • Mapping to an external rating agency
  • Ordinal logistic regression
  • Confidence level based approach
  • LGD and EAD for Low Default Portfolios
• Model Validation
  • Regulatory perspective
  • Basic concepts of validation
  • Defining validation
  • Common validation issues
  • General validation principles
  • Developing a validation framework
  • Quantitative validation
  • Data set split-up
  • Challenges
  • Backtesting PD Models
  • Backtesting PD at level 0 (data stability): system stability index
  • Backtesting PD at level 1 (model discrimination): AUC, Gini index
  • Backtesting PD at level 2 (model calibration): Brier score, binomial test, binomial test with default correlation, Hosmer-Lemeshow test, Vasicek test with asset correlation
  • Data aggregation
  • Risk philosophy
  • Traffic light indicator dashboard for PD backtesting
  • Action plan
  • Backtesting LGD and EAD Models
  • Backtesting LGD and EAD at level 0: system stability index
  • Backtesting LGD and EAD at level 1: discrete ROC analysis, Spearman rank correlation, Kendall’s tau, Bootstrapping
  • Backtesting LGD and EAD at level 2: scatter plots, Student t-test, Wilcoxon signed rank test, Loss Shortfall, Exposure weighted measures
  • Benchmarking
  • Qualitative validation
  • Use testing
  • Data quality
  • Model design
  • Documentation
  • Corporate governance and management oversight

Day 2 (9am - 5pm)

• Stress testing
  • Purpose of stress testing
  • Types of stress tests
  • Sensitivity-based stress testing
  • Univariate sensitivity tests
  • Multivariate sensitivity tests
  • Scenario-based stress testing
  • Portfolio-driven versus event-driven scenario stress testing
  • Historical versus hypothetical scenario stress testing
  • Post-GFC stress testing
  • Challenges in stress testing
  • Stress testing governance
  • Stress testing from a Basel perspective
  • Asset correlations and worst-case default rate
  • Downturn versus stressed LGD/EAD
• Advanced Modeling Topics
  • Alternative data sources for credit risk modeling: Telco data (Call Detail Records), Social Media Data (Facebook, Twitter, etc.), Google StreetView and Satellite Data, Google Trends and nowcasting
  • Optimized Feature Engineering for credit risk modeling: Ratio features and continuity correctness, Trends and time series features, Box-Cox/Yeo-Johnson transformation for saturation effects, Ordinal features (thermometer coding), Social network features
  • Deep learning for credit risk modeling: Multilayer Perceptrons, Deep belief networks, Convolutional nets, 
  • Ensemble methods for credit risk modeling: Random forests, XGBoost
  • Bayesian methods for credit risk modeling: Bayesian versus frequentist approach, Bayesian networks
  • Profit based performance evaluation for credit risk modeling: Expected Maximum Profit Measure (EMP), Profit analytics, Risk-Based Pricing, RAROC analysis
  • Ethical perspective
  • Privacy and Security