LGD and EAD prediction using SAS
This two
day course offers a summary of today’s most commonly used average and
cohort-based LGD calculation methodologies, together with detailed
instructions on how to move toward more flexible statistical prediction
models, including regression trees, beta regression, two-stage scenario
prediction models, generalized additive neural networks and LGD
scorecards.
The course also includes the most important definitions for setting up EAD modelling for facilities with explicit limits.
The
course includes extensive and fully-scripted exercises that let the
student create modelling process flows step by step and discover the
rich possibilities and benefits of SAS Enterprise Miner.
All
custom SAS code used in the process is documented, including dedicated
LGD validation and bootstrapping code. Exercises are done on real
mortgage LGD data.
Dr Hendrik Wagner
For 9 years Dr Wagner was Product Manager Data Mining Solutions at the SAS Institute covering Europe, Middle East and Africa.
He
introduced scorecard development functionality into SAS' flagship data
mining solution Enterprise Miner and made it the market leading solution
for inhouse scorecard development. He also led the creation of an
end-to-end model development, deployment and monitoring solution and
defined specific functionality for building internal rating systems for
Basel 2 -PD and LGD modeling, pooling and backtesting.
After
writing SAS' first Risk Weighted Assets calculation code, he helped
launch SAS' market leading Credit Risk Management solution.
He
became a consultant in 2006 providing credit risk and internal audit
departments with advisory and implementation services, such as readiness
assessment, model development and rating system auditing.
Clients
include inter alia GHB bank, Thailand, (Housing Loan Application
Scorecard), Samlink, Finland, (Behavioral PD Model),Maybank Malaysia(
Corporate PD Model Validation), National Australia Group UK, (Retail PD,
LGD and EAD Model Validation for Basel2 IRB Approval), Deutsche
Telekom Germany (PD model validation and development, early warning
system, profit scoring) and BHW Bausparkasse ( PD and LGD model
validation of a home loans portfolio).
Dr Wagner holds a doctorate in Computer Sciences. Duration: 2 days
The course covers the following topics in detail:
Part1: LGD Basics
- Why LGD
- Individual Workout LGD Calculation
- LGD Averaging
- LGD Cohorts
- Structural LGD Models
Part2: Account Level Models- Development Sample Definition
- Reference Dates
- Input Variables
- Data Pre-Processing and Exploration
- Data Partitioning
- Treatment of Missing Values and Outliers
- Multi-Plot
- Segmentation
- Table of Averages
- Regression Trees
- Linear Regression
- Normal-Theory Linear Regression
- Beta Regression
- Input Standardization and Normalization
- Two-stage Modelling
- Non-Linear Regression
- Grouped Variable Regression
- Generalized Additive Neural Networks
- LGD Scorecards
- Validation
- Residual Analysis
- Correlation and Error Measures
- Stability Analysis
- Pseudo-Power Measures
- Bootstrapping
- LGD Pooling
- LEQ and CCF Modelling (EAD)
|