


Credit Risk Modeling Using SAS (CREDITRISK)

Presented by Bart Baesens, Ph.D. or Christophe Mues, Ph.D., Assistant Professors at the School of Management of the University of Southampton (UK)
In this course, students learn how to develop credit risk models in the context of the recent Basel II guidelines. The course provides a sound mix of both theoretical and technical insight, as well as practical implementation details. These are illustrated by several reallife case studies and exercises.
 develop probability of default (PD), loss given default (LGD), and exposure at default (EAD) models
 validate, backtest, and benchmark credit risk models
 stress test credit risk models
 develop credit risk models for low default portfolios
 use new and advanced techniques for improved credit risk modeling.
Anyone who is involved in building credit risk models, or is responsible for monitoring the behavior and performance of credit risk models
Before attending this course, you should have business expertise in credit risk and a basic understanding of statistical classification methods. Previous SAS software and SAS Enterprise Miner experience is helpful but not necessary.
Introduction to Credit Scoring
 application scoring, behavioral scoring, and dynamic scoring
 credit bureaus
 bankruptcy prediction models
 expert models
 credit ratings and rating agencies
Review of Basel I, Basel II, and Basel III
 Regulatory versus Economic capital
 Basel I, Basel II, and Basel III regulations
 standard approach versus IRB approaches for credit risk
 PD versus LGD versus EAD
 expected loss versus unexpected loss
 the Merton model
Sampling and Data Preprocessing
 selecting the sample
 types of variables
 missing values (imputation schemes)
 outlier detection and treatment (box plots, zscores, truncation, etc.)
 exploratory data analysis
 categorization (chisquared analysis, odds plots, etc.)
 weight of evidence (WOE) coding and information value (IV)
 segmentation
 reject inference (hard cutoff augmentation, parceling, etc.)
Developing PD Models for Basel II
 basic concepts of classification
 classification techniques: logistic regression, decision trees, linear programming, knearest neighbor, cumulative logistic regression
 input selection methods, such as filters, forward/backward/stepwise regression, and pvalues
 setting the cutoff (strategy curve, marginal goodbad rates)
 measuring scorecard performance
 splitting up the data: single sample, holdout sample, crossvalidation
 performance metrics, such as ROC curve, CAP curve, and KSstatistic
 defining ratings
 migration matrices
 rating philosophy (PointinTime versus ThroughtheCycle)
 mobility metrics
 PD calibration
 scorecard alignment and implementation
Developing LGD and EAD Models for Basel II
 modeling loss given default (LGD)
 defining LGD using market approach and workout approach
 choosing the workout period
 dealing with incomplete workouts
 setting the discount factor
 calculating indirect costs
 drivers of LGD
 modeling LGD
 modeling LGD using segmentation (expert based versus regression trees)
 modeling LGD using linear regression
 shaping the Beta distribution for LGD
 modeling LGD using twostage models
 measuring performance of LGD models
 defining LGD ratings
 calibrating LGD
 time weighted versus default weighted versus exposure weighted LGD
 economic downturn LGD
 modeling exposure at default (EAD): estimating credit conversion factors (CCF)
 defining CCF
 cohort/fixed time horizon/momentum approach for CCF
 risk drivers for CCF
 nodeling CCF using segmentation and regression approaches
 CAP curves for LGD and CCF
 correlations between PD, LGD, and EAD
 calculating expected loss (EL)
Validation, Backtesting, and Stress Testing
 validating PD, LGD, and EAD models
 quantitative versus qualitative validation
 backtesting for PD, LGD, and EAD
 backtesting model stability (system stability index)
 backtesting model discrimination (ROC, CAP, overrides, etc,)
 backtesting model calibration using the binomial, Vasicek, and chisquared tests
 traffic light indicator approach
 backtesting action plans
 throughthecycle (TTC) versus pointintime (PIT) validation
 benchmarking
 internal versus external benchmarking
 Kendall's tau and Kruskal's gamma for benchmarking
 use testing
 data quality
 documentation
 corporate governance and management oversight
Low Default Portfolios (LDPs)
 definition of low default porfolios
 undersampling versus oversampling
 likelihood approaches to LDPs
 rating mapping approaches to LDPs
Stress Testing for PD, LGD, and EAD Models
 overview of stress testing regulation
 sensitivity analysis
 scenario analysis (historical versus hypothetical)
 examples from industry
 Pillar 1 versus Pillar 2 stress testing
 macroeconomic stress testing
New Techniques to Develop PD, LGD, and EAD Models
 review of traditional techniques for scorecard development
 neural networks: the neuron model, multilayer perceptrons (MLPs), training an MLP
 opening up the neural network black box
 twostage models
 support vector machines: the SVM classification model and building scorecards using SVMs (short)
 case study: using logistic regression and support vector machines to develop a country rating system
Survival Analysis for Credit Risk Modeling
 example applications (predicting time to default, time to early repayment, etc.)
 the censoring problem
 survival curves versus hazard curves
 Kaplan Meier analysis
 parametric survival analysis
 proportional hazards regression
This course addresses SAS Enterprise Miner.
 The deadline of the application: 22. August 2013.
 Cancellation is possible at the latest: 22. August 2013. In case of a later cancellation, 35% of the course fee is going to be invoiced.


