CREDIT RISK MODELS WITH DATA MINING TOOLS

introducTION
GETTING STARTED WITH SAS ENTERPRISE MINER
1.1 STARTING SAS ENTERPRISE MINER
1.2 ASSIGNING A LIBRARY LOCATION
1.3 DEFINING A NEW DATA SET
SAMPLING AND DATA PRE-PROCESSING
2.1 INTRODUCTION
2.2 SAMPLING AND VARIABLE SELECTION
2.2.1 Sampling
2.2.2 Variable selection
2.3 MISSING VALUES AND OUTLIER TREATMENT
2.3.1 Missing values
2.3.2 Outlier detection
2.4 DATA SEGMENTATION
2.4.1 Decision trees for segmentation
2.4.2 K-Means clustering
PROBABILITY OF DEFAULT MODELS
3.1 OVERVIEW OF PROBABILITY OF DEFAULT
3.1.1 PD models for retail credit
3.1.2 PD Models for corporate credit
3.1.3 PD Calibration
3.2 CLASSIFICATION TECHNIQUES FOR PD
3.2.1 Logistic regression
3.2.2 Linear and quadratic discriminant analysis
3.2.3 Neural networks
3.2.4 Decision trees
3.2.5 Memory based reasoning
3.2.6 Random forests
3.2.7 Gradient boosting
3.3 MODEL DEVELOPMENT (APPLICATION SCORECARDS)
3.3.1 Motivation for application scorecards
3.3.2 Developing a PD Model For Application Scoring
3.3.3 Overview
3.3.4 Input Variables
3.3.5 3.3.2.3 Data Preparation
3.3.6 Model Creation Process Flow
3.3.7 3.3.2.5 Known Good Bad Data
3.3.8 Data Sampling
3.3.9 Outlier Detection and Filtering
3.3.10 Data Partitioning
3.3.11 Transforming Input Variables
3.3.12 Variable Classing and Selection
3.3.13 Modeling and Scaling
3.3.14 Reject Inference
3.3.15 Model Validation
3.4 MODEL DEVELOPMENT (BEHAVIORAL SCORING)
3.4.1 Motivation for behavioral scorecards
3.4.2 Developing a PD model for behavioral scoring. Overview
3.4.3 Input Variables
3.4.4 Data Preparation
3.4.5 Model Creation Process Flow
3.5 PD MODEL REPORTING
3.5.1 Overview
3.5.2 Variable worth statistics
3.5.3 Scorecard strength
3.5.4 Model performance measures
3.5.5 Tuning the model
3.6 MODEL DEPLOYMENT
3.6.1 Creating a model package
3.6.2 Registering a model package
LOSS GIVEN DEFAULT LGD MODELS
4.1 OVERVIEW OF LOSS GIVEN DEFAULT
4.1.1 GD models for retail credit
4.1.2 LGD models for corporate credit
4.1.3 Economic variables for LGD estimation
4.1.4 Estimating downturn LGD
4.2 REGRESSION TECHNIQUES FOR LGD
4.3 MODEL DEVELOPMENT
4.3.1 Motivation for LGD models
4.3.2 Developing an LGD model
4.3.3 Model Creation Process Flow
4.3.4 LGD Data
4.3.5 Logistic Regression Model
4.3.6 Scoring Non-Defaults
4.3.7 Predicting the Amount of Loss
4.3.8 Model Validation
4.4 CASE STUDY: BENCHMARKING REGRESSION ALGORITHMS FOR LGD
4.4.1 Data set characteristics
4.4.2 Experimental set-up
4.4.3 Parameter Settings and Tuning
4.4.4 Ordinary Least Squares with Box-Cox Transformation (BC-OLS)
4.4.5 Regression Trees (RT)
4.4.6 Artificial Neural Networks (ANN)
4.4.7 Results and discussion
EXPOSURE AT DEFAULT EAD MODEL
5.1 INTRODUCTION
5.2 TIME HORIZONS FOR CCF
5.3 DATA PREPARATION
5.4 CCF DISTRIBUTION – TRANSFORMATIONS
5.5 MODEL DEVELOPMENT
5.5.1 Input selection
5.5.2 Model methodology
5.5.3 Ordinary Least Squares
5.5.4 Binary and Cumulative Logit Models
5.5.5 Performance metrics
5.6 MODEL VALIDATION AND REPORTING
5.6.1 Model validation
5.6.2 Reports
5.6.3 Strength Statistics
5.6.4 Model Performance Measures
5.6.5 Tuning the Model
STRESS TESTING
6.1 OVERVIEW OF STRESS TESTING
6.2 PURPOSE OF STRESS TESTING
6.3 6.3 STRESS TESTING METHODS
6.3.1 Sensitivity testing
6.3.2 Scenario testing
6.3.3 Historical Scenarios
6.3.4 Hypothetical Scenarios
6.3.5 Categories of Hypothetical Scenarios
6.3.6 Stress Testing using Macroeconomic Approaches
6.3.7 Regulatory stress testing
MODEL REPORTS
7.1 SURFACING REGULATORY REPORTS
7.2 MODEL VALIDATION
7.2.1 MODEL PERFORMANCE
7.2.2 Model stability
7.2.3 Model calibration
7.3 7.3 SAS MODEL MANAGER EXAMPLES
7.3.1 Create a PD report
7.3.2 Create a LGD report