Credit Risk Modeling for Basel and IFRS 9 using R and Python

📅 November 29th - 30th 2018, 9am - 5pm
🌍 English

This course is fully booked. Registration is no longer possible.

Introduction

This comprehensive training to practical credit risk modeling provides a targeted training guide for risk professionals looking to efficiently build in-house 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 walks you through the fundamentals of credit risk modeling and shows you how to implement these concepts using both R and Python software, with helpful code provided. 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

Introduction to Credit Scoring

• Credit scoring for retail
  • application scoring
  • behavioral scoring
  • profit scoring
• Credit scoring for non-retail
  • prediction approach
  • expert based approach
  • agency ratings approach
  • shadow ratings approach
• Big Data for credit scoring (social media data, call detail records, web scraping)
• credit bureaus
• credit ratings and rating agencies
• privacy and ethics

The Basel Accords and IFRS 9

• regulatory versus economic capital
• Basel Accords
• PD versus LGD versus EAD
• standard approach versus IRB approaches for credit risk
• expected loss (EL) versus unexpected loss (UL)
• Merton/Vasicek model
• IFRS 9 (CECL)
• code examples: calculating Basel regulatory capital in R/Python

Data Selection, Sampling and Data Preprocessing

• sample selection
• variable types
• missing values (imputation schemes)
• outlier detection and treatment (box plots, z-scores, truncation, etc.)
• exploratory data analysis
• categorization (Chi-squared analysis, odds plots, etc.)
• variable transformation: weight of evidence (WOE), Box-Cox, Yeo-Johnson
• variable selection (information value, Cramer’s V)
• reject inference
• oversampling, undersampling, SMOTE
• data quality and data governance
• code examples: preprocessing credit risk data in R/Python

Day 2

Developing PD Models

• Level 1: Discrimination
  • logistic regression and decision trees
  • discrete time and continuous time hazard models
  • recent techniques: SVMs, random forests, XGBoost, deep learning (briefly)
  • measuring scorecard performance
  • splitting up the data: holdout sample, cross-validation, bootstrapping
  • ROC curve, CAP curve, and KS statistic
  • Expected Maximum Profit (EMP) measure
  • code examples: developing a PD scorecard in R/Python; calculating EMP in R/Python
• Level 2: Ratings and Calibration
  • defining ratings: supervised versus unsupervised methods
  • monotonicity constraints
  • rating philosophy: Point-in-Time (PIT) versus Through-the-Cycle (TTC)
  • migration matrices
  • stability metrics
  • calibration methods
  • calibration uncertainty (sampling uncertainty, economic volatility)
  • Lexis diagrams

Developing LGD Models

• Level 0: Data
  • default definition
  • LGD definition
  • Basel versus IFRS 9 perspective
  • choosing the workout period
  • dealing with incomplete workouts
  • setting the discount factor
  • calculating indirect costs
  • drivers of LGD
• Level 1: Discrimination
  • modeling LGD
  • segmentation (expert based versus regression trees)
  • (transformed) linear regression
  • fractional logistic regression
  • beta regression
  • two-stage models
  • measuring the performance of LGD models
  • code example: developing and evaluating an LGD model in R/Python
• Level 2: Ratings and Calibration
  • defining LGD ratings
  • calibrating LGD
  • default weighted versus exposure weighted versus time weighted LGD
  • economic downturn LGD

Developing EAD Models

• Level 0: Data
  • defining exposure at default (EAD): conversion measures, credit conversion factors (CCF)
  • regulatory perspective
  • defining CCF (cohort, fixed time horizon, variable time horizon, momentum method)
  • risk drivers for CCF
• Level 1: Discrimination
  • modeling CCF using segmentation and regression approaches
  • CAP curves for LGD and CCF
  • code example: developing and evaluating an EAD model in R/Python
• Level 2: Ratings and Calibration
  • defining EAD ratings
  • calibrating EAD
  • correlations between PD, LGD, and EAD