Credit Risk Models August 24 26, 2010
AGENDA 1 st Case Study : Credit Rating Model Borrowers and Factoring (Accounts Receivable Financing) pages 3 10 2 nd Case Study : Credit Scoring Model Automobile Leasing pages 11 20 3 rd Case Study : The Validation of Internal Rating Systems pages 21 28 Credit Risk Model : What is Credit Risk Model? page 29 What Properties to be expected! page 29 Application page 30
3 1 st Case Study Credit Rating Model Borrowers and Factoring (Accounts Receivable Financing)
Credit Rating Model.. (1) 4 Model Development Portfolio Analysis Data Analysis Calibration and Mapping Data Preparation Performance Test Implementation Recommendation Data Cleansing Data Partition Accuracy Blind Test Univariate Correlation Multivariate
Credit Rating Model.. (2) 5 Portfolio Analysis Distribution of Financial Statements and Default Data -Year; Business Type / Group / Concentration; Size Performance Window Number of Accounts Coverage Bad Definition
Credit Rating Model.. (3) 6 Data Preparation Data Cleansing Data Partition Existing and Completeness of Data Variables -Financial Ratio -Qualitative Data Sample Grouping Dimension of Sampling Correction of Data Loan Status compared to Financial Performance In Sample / Out-of-Sample Product Type Logic Out-of-Time Asset Size Out-of-Universe Business Type
Credit Rating Model.. (4) 7 Data Analysis Univariate Correlation Multivariate Logistic Regression Screening Criteria Multiple Logistic Regression Quantitative Qualitative Make Business Sense Powerful Understandable / Intuitive Enough observation to develop and validate model
Credit Rating Model.. (5) 8 Performance Test Accuracy Blind Test
Credit Rating Model.. (6) 9 Calibration and Mapping Calibration Mapping
Credit Rating Model.. (7) 10
11 2 nd Case Study Credit Scoring Model Automobile Leasing
Credit Scoring Model 12 Scorecard Development Portfolio Analysis Model Development Portfolio Distribution Vintage Analysis Data Gathering Model Building Test on Similarity and/or Differentiation of Bad Rates Data Cleansing Preliminary Variable Selection Purposes of analyzing portfolio: -To understand the overall picture of portfolio -To know the portfolio s default rates in terms of Marginal and Cumulative Default Rates -To help set the sample group to be collected for model development Variable Classing
Model Development.. (1) 13 Data Gathering Performance Window Bad Definition Sample Size Variables Independent Variables Dependent Variables Borrower s Characteristics Collateral s Characteristics Facility s Characteristics Loan Status
Model Development.. (2) 14 Data Cleansing Data Exploration Problem and Solution Number of Fields Number of Complete Cases Missing Value Error from Data Transformation Reliability of Data Missing Data Extreme Value Vague / Unclear Data Discrepancy of Data
Model Development.. (3) 15 Variable Classing Classing is process of automatically and / or interactively binning and grouping interval, nominal, or ordinal input variables in order to Manage the number of attributes per characteristics Improve the predictive power of the characteristics Select predictive characteristics Make the Weight of Evidence and thereby the number of points in the scorecard vary smoothly or even linearly across the attributes
Model Development.. (4) 16 Variable Selection Preliminary Statistical Missing Value Consistency Univariate data analysis Multivariate data analysis Completeness and relationship with other variables
Model Development.. (5) 17 Model Building Why Logistic Regression! -It can handle discrete variable or qualitative variable. -The dependent variable need not to be normally distributed. -The dependent variable need not to be homoscedastic for each level of the independents. -Normally distributed error terms are not assumed. -It does not require that the independents be interval.
Model Development.. (6) 18 Model Building Sample Accuracy Training Sample Out-of-time Sample Type I & II errors K-S Statistics Testing Sample Discriminatory Power of the model : The maximum difference between the cumulative percent good distribution and the cumulative percent bad distribution
Model Development.. (7) 19 Criteria for Model Selection Accuracy : Type I & II errors; K-S statistics Consistency of testing results among Training, Testing and Out-of-time Samples Number of Variables in the model Including or excluding Policy Indicators in the model Example of Gain Table
Model Development.. (8) 20 Scorecard Implementation and Application
21 3 rd Case Study The Validation of Internal Rating Systems
Scope of Work and Validation Aspects 22 Scope of Work Analyze the discriminatory power of rating models Analyze the stability of rating models Analyze the connection between PD and Grade Analyze the models design Analyze the rating process Validation Aspects
Validation Method 23 Study and Analyze Data from Documents Statistical and Mathematical Tests Default Probability Model Validation Interview and Site Visit
Validation Results 24 Quantitative : Discriminatory Power Type I & II errors in theory Type I error of each model Relative Frequency Density Function for Good Cases Density Function for Bad Cases Cut-Off Point Project Financing <15M; Hire Purchase; Leasing Project Financing <15M Hire Purchase and Leasing Rating Class Type I error Type II error Project Financing >= 15M Factoring
Validation Results 25 Quantitative : Discriminatory Power (ROC Curve) Project Financing < 15M; Hire Purchase; Leasing Project Financing >= 15M Factoring
Validation Results 26 Quantitative : Stability Project Financing < 15M; Hire Purchase; Leasing Project Financing >= 15M Factoring 2007 2006 Overall 2008 % PD from Actual Default Rate : ADR (%)
Validation Results 27 Quantitative : Calibration PF >= 15M S&P s Default Rate PF < 15M; HP; Leasing PF >= 15M HP: LS PF < 15M HP: LS Implied PD PF < 15M; HP; Leasing PF < 15M Actual Default Rate (%) of each model compared to Implied PD (%) by CQC Grade Actual Default Rate (%) of each model compared to S&P s Default Rate by Rating
Executive Summary 28
Credit Risk Model.. (1) 29 What is Credit Risk Model? A tool used to evaluate the level of risk associated with applicants or borrowers. It consists of a group of characteristics, statistically determined to be predictive in separating good and bad accounts. It provides statistically odds or probability that an applicant or borrower with any given rating or score will be good or bad. What Properties to be expected! Understandable Powerful Calibrated Empirically validated
Credit Risk Model.. (2) 30 Application Origination Decisions Given the risk and a fixed price, is the asset worth taking? Given the risk, what price is required to make the asset worth buying? Portfolio Optimization To reduce the portfolio s risk, concentrations of risk and how the risk can be diversified must be known. Capital Management To set capital, the loss level is needed. Credit Process Management To gain the efficiencies of application processing that comes through automation. To gain control and consistency in lending practices for the entire credit portfolio. To identify the variables which are important in the credit evaluation process To improve delinquency statistics while maintaining desired approval rates
Credit Risk Model.. (3) 31 Credit Risk Model only classifies and predicts risk; It does not tell the lender how to manage it.