Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

RMPG Learning Series CRM Workshop Day 3

507 views

Published on

This is being made available for Risk Management Practice Group on Linkedin.

RMPG Learning Series CRM Workshop Handouts: File 7 of 9

Published in: Economy & Finance
  • Be the first to comment

  • Be the first to like this

RMPG Learning Series CRM Workshop Day 3

  1. 1. Agenda for Day 3 Credit Rating Models Lunch Break Case Studies Open Session/ Q&A IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 1
  2. 2. Introduction to credit risk modeling – What is a model Risk Score = Co-eff1*Leverage + Co-eff2 *Current Ratio +……. Co-eff6 *Integrity +….. Co-eff8 *Industry Phase…. IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 2
  3. 3. Credit Risk Models - Some Examples Altman’s Z - score model (Multiple Discriminant) Merton model Judgmental Hybrid IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 3
  4. 4. Altmans’s Z Score Model Z = 0.012 X 1 + 0.014 X 2 + 0.033 X 3 + 0.006 X 4 + 0.999 X 5 Where, • X 1 = Net Working Capital / Total Assets •X 2 = Retained earnings / Total Assets •X 3 = PBIT/ Total Assets •X 4 = Market value of equity/ Book Value of Total Liabilities •X 5 = Sales/ Total Assets IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 4
  5. 5. Altmans’s Z Score Model Z = 0.012 X 1 + 0.014 X 2 + 0.033 X 3 + 0.006 X 4 + 0.999 X 5 < 1.81 - Failing Zone Z 1.81 to 2.99 - Ignorance Zone > 2.99 - Non-failing Zone IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 5
  6. 6. Merton Model Expected Default Frequency - is calculated using 3 stepsStep 1: Estimate asset value and asset volatility from equity value andvolatility of equity returnStep 2: Calculate distance =Asset Value - Default point to default (DfD) Asset Value * Asset VolatilityStep 3: Calculate expected default frequency IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 6
  7. 7. Calculating distance to default: Merton The market value of a firm’s assets and its historical volatility imply a distribution of future firm value Given today’s obligations (debt), we can calculate the probability that the market value of assets will be lower than the firm’s obligations one year from now (i.e., default) Distance to default is mean value minus debt, normalized by S.D. in •Amount IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 7
  8. 8. The quantitative model would derive its strength from the Bank’s data and the human expertise and experience of CO Industry Firm Standing Management…. Convert into proxiesProfessional Judgement Check forfor weights consistency Construct indices Statistically explanatory set of variables IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 8
  9. 9. The benefit of the model Reduces the dimensionality of space of the credit officer IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 9
  10. 10. Banks need different risk scoring models for differentcredit segments Corporate Small Bank Retail Loan Credit Business ExposuresQuality offinancial Reasonably Less Partial Morestatements Reliable Reliable Information Reliable Global, Global,Market Regional Local National or National orSituation or Local Regional regionalType High value & Lower value & Low value & High Value & Low Numbers Higher Numbers High Numbers Low numbers IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 10
  11. 11. No. of rating models/ borrower categories in new systems The number of rating models should be determined: Based on the current portfolio of the bank Based on business strategy and focus areas of the bank A good thumb rule, is that 80-85% of the bank’s credit portfolio should be risk rated. For the remaining portfolio, the bank could use pool-based approach Banks use the following models: Corporate Segment: Large, SME and Small Business models; Retail Segments: Home Loan, Personal Loan & Credit Card models; Commercial Segment: Bank and NBFC models; Project Models: Infrastructure, Green-field and Brown-field models IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 11
  12. 12. Data Collection - What Type of Data is required to be collected Accounts (On which data is being collected) Performing Accounts Non - Performing Accounts This sample of accounts has to be representative of the Bank’s portfolio IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 12
  13. 13. Data Collection - What Type of Data is required to be collected Financial Information – Balance Sheet, Profit and Loss, Cash FlowData(Historical) Management Qualitative Data Industry Firm Standing Conduct of Account IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 13
  14. 14. Why do we need to collect this data ?• Historical Data is the basis of estimating the model equation (along with expert opinion)• What is the model ? Risk Score = A*Leverage + B* Current Ratio +C*Sales/Total Assets……• The Data would be the basis for both deducing the predictor variables and the coefficients of the model equation (along with expert opinion)• In other words, the fact that Leverage is to be chosen in the model and the A (coefficient of Leverage) is both coming from the Bank’s historical data IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 14
  15. 15. Data Collection – The criticality of this exercise The model is only as good as the data used to construct it• The Data sample used to estimate the model should be representative of the Bank’s portfolio• The Data sample has to be accurate IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 15
  16. 16. The Broad Model Construction PhilosophyPhase I Phase II Phase IIIParameter Selection Modeling Technique Risk Grading Implied probabilitiesChoose Universal set of (Output of the Statistical Limit/Filter parameters Technique)Risk DriversQualitative variables Risk Grading Transform Parameters (by probability)Index constructionShortlist Predictive Adjustment for accountParameters Statistical Technique - Operations (DA, LR, Probit etc) (Modified Borrower risk score) IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 16
  17. 17. Choosing of predictor parameters – The art and science of it How are financial ratios related to default ?• There exists a correlation between select ratios and default• The relation is non-linear (at no point is default certain)• Default would depend upon other predictor variables of the account IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 17
  18. 18. Choosing of predictor parameters – The art and science of it Analysis Univariate relation of predictor parameters to default (Financial Ratios) Aid the modeler in answering• The curve – Shape of the relationship between the predictor variable and default (In essence, what default probability corresponds to what parameter values)• What are the most potent ratios (What profitability ratio is the most potent predictor• How do correlations affect the coefficients in a multivariate model framework IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 18
  19. 19. Forward selection process• Start with variables with the highest univariate correlation with default and add more until additional variables have no additional importance• Ensure that variables selected do not suffer from “multicollinearity” (The wrong sign problem, inflated variances of coefficients, poor out of sample performance) The essence of the activity •Selection done based on suggestion of univariate power •Validation done in a multivariate framework IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 19
  20. 20. The Broad Model Construction Philosophy Most critical processes in model constructionPhase I Phase II Phase IIIParameter Selection Modeling Technique Risk Grading Implied probabilitiesUniversal set of (Output of the StatisticalPredictor Parameters Limit/Filter parameters Technique)Qualitative variables Risk Grading Transform Parameters (by probability)Index constructionChoose Predictive Adjustment for account Statistical Technique -Parameters Operations (DA, LR, Probit etc) (Modified Borrower risk score) IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 20
  21. 21. Transformations applied to Predictor Parameters Why is there a need to apply transformations?? Movement of Leverage from 1-2 The idea behind applying is not at as risky as a movement transformations is to mimic this analysis from 2-3 happening in the credit officers mindMovements of values in predictor variables result in non-linear Credit Risk profileis highly non-linear. We need to transform predictor variables to factor this IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 21
  22. 22. The Borrower Risk Score will be adjusted for risk impact ofaccount operations Financial Risk Account Operations* Industry Risk Borrower Adjusted Score Borrower Score Management * For existing accounts Risk Firm Standing IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 22
  23. 23. The monitoring parameters will be set in consultation with themanagement and will be an input for deriving modified risk grade Factors on which monitoring levels are to be set are as follows: 1. No. of days delay in receipt of principal/interest instalments 2. Submission of progress reports 3. Compliance with sanctioned/disbursement conditions 4. Key employees turnover 5. Comments on operations/assets during site visits 6. Change in accounting period during the last five years 7. No. of times rescheduling/relief obtained from lending institutions IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 23
  24. 24. The weightages of the various components – Concept ofDynamic Weights A Linear Rating Model 40 % Financial Risk 15% Industry Risk Borrower Score Management 15 % Risk 10% Firm Standing Account 20 % Operations IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 24
  25. 25. The weightages of the various components – Dynamic Weights Credit Risk is highly non-linear. • Borrower scoring low on integrity will not be accepted irrespective of scores on other parameters • Borrower with a leverage of 10 would not be accepted irrespective of scores on other parameters It is critical that the risk-scoring model mimics this non – linear thinking of a experienced credit risk officer IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 25
  26. 26. The weightages of the various components – Dynamic WeightsCase Study – Consider a account which got the following scores in Management RiskParameter Risk Score Integrity----------------------------------------------------------------------------- 4 Diversion of Funds-----------------------------------------------------------------4 Business Commitment-------------------------------------------------------------3 Payment Record of Group companies-------------------------------------------4 Internal Control---------------------------------------------------------------------4 Succession Planning----------------------------------------------------------------4 The scale is defined such that 1 is the best and 4 is the worst IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 26
  27. 27. The weightages of the various components – Dynamic Weights• The Borrower has a very high management risk. The Credit officer automatically recognizes this and would not lend no matter how impressive the financials or business• The credit risk model has to adjust accordingly to mimic this non-linear analysis happening in the credit officer’s mind. It cannot be churning out a safe risk-grade for such an obviously high risk account• The solution is the dynamic weights concept where the importance of every parameter would depend on the value allotted to it by the Credit officer IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 27
  28. 28. Model Calibration – The Process LR Output Model Output Account 1 – 0.001 Account 2 – 0.002 Account 3 – 0.004 RG1 0.00 – 0.05 Account 4 – 0.007 ………………….. RG2 0.05 – 0.08 …………………… …………………… RG3 0.08 - 0.12 ………………….. ………………. …………………… Model Calibration ……………………. Process ………………. ……………………. ………………. …………………….. ……………………. ………………. Account 347 – 0.97 Account 348 – 0.98 ………………. Account 349 – 0.99 RG10 0.85 – 1.00 IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 28
  29. 29. Model Calibration Process – What are the guidelines ofthe process For Basel II IRB compliance, each risk grade is to be mapped to a unique PD - No overlap of risk There should be no undue concentrations of borrowers in any one risk grade Number of Risk grades and interpretation desired is decided apriori and the spreading is done based on this Ensure that the statistical PD estimates for every risk grade follow a desired trend IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 29
  30. 30. Model Calibration Process – What are the guidelines of the process 90.00% 4.0% 80.00% 3.5% There should be no 70.00% 3.0% overlap of PDs by Probability of Default Probability of Default 60.00% 2.5% grade 50.00% 2.0% 40.00% 30.00% 1.5% 20.00% 1.0% 10.00% 0.5% 0.00% 1 2 3 4 5 6 7 8 9 10 11 0.0% 0 1 2 3 4 5 6 7 Risk Grade R isk Rating 45% Reduce 40% concentrations in 39% 35% any one rating gradePercent of Borrowers 30% 25% 23% 20% 20% 15% 13% 10% 5% 2% 2% 1% 0% 1 2 3 4 5 6 7 Risk Rating IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 30
  31. 31. Entry and exit criterion 100% 100% 1. At an operating level, an 90% 89% entry grade of RG 6 or 80% 82% better would roughly 70% 73% correspond to the credit 63% Exit 60% acceptance levels based on 50% 52% Entry risk appetite. 40% 40% 41% 30% 32% 2. The exit criteria (in case 20% 20% 27% 23% this means exiting from the 10% 11% 0% 0% 0% 9% portfolio to other banks) 0% may be set slightly lower at RG1 RG2 RG3 RG4 RG5 RG6 RG7 RG8 RG9 RG 7 % defaults % portfolio Relative Risk of Default 3. The monitoring intensity 1 3 5 7 10 may be set depending on the •Gr 1 •Gr 2 •Gr 3 •Gr 4 •Gr 5 •Gr 6 •Gr 7 to •Gr 9 grades , which need to beStrong Credit Quality annually re-evaluatedLow Risk Green Risk scores between 1 & 3 Good quality credit Zone Yellow Risk scores > 3 & up to 5 No immediate concern Zone Amber Risk scores > 5 & up to 7 Requires intensive Zone monitoring IM aCS 2010•High Red NPA/ Could turn NPA Risk scores greater than 7Classroom discussion only For Printed 11-M ay-11 Page 31 Zone over the medium term•Risk
  32. 32. Risk Scoring Model - the end product NPA/ Could turn NPA Risk over the medium term Requires intensive No monitoring Good quality immediate credit concern Risk Scale 1 2 3 4 5 6 7 8 9 IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 32
  33. 33. The criticality of model calibrationA Model may be powerful (able to distinguish between good and bad) BUT It maybe be incorrectly calibrated IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 33
  34. 34. Model Validation Results- Cumulative Accuracy Profile (CAP)Plots CAP Plot Perfect Model 120% Percentage reduction in NPA 100% 80% Rating Model 60% 40% Random Model 20% 0% 0% 20% 40% 60% 80% 100% Percentage of Proposals accepted IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 34
  35. 35. CAP curve metric to assess Model Power – The GINIcoefficient • The Gini Coefficient of the CAP plot is defined as the ratio of the area between the model curve and the random plot and area between the perfect model and random plot. Consequently the closer the AR of the model is to one the better the discriminatory power of the model is. • Gini Coefficient (AR) = Area between model curve and random plot / Area between Perfect model and Random plot IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 35
  36. 36. Classification Matrix Classification Matrix Classification Results Predicted Group Membership Total Default Non Default Count Default 43 13 56 Non Default 78 323 401 Percentage Default 76.79 23.21 100 Non Default 19.45 80.55 100 80.1% of original grouped cases correctly classified. Error Type Matrix Number of Accounts Percentage Type 1 Error 13 2.84 Type 2 Error 78 17.06 IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 36
  37. 37. Firm defaulted Graphical Back Testing at this point Movement of R isk Grade (N PA Account)Model signalled 9default well in 8advance of the event7 6 Risk Grade 5 4 3 2 1 0 1999 2000 2001 2002 2003 Ye ar • Ability of the model to signal default before the actual occurrence • Critical attribute of a robust credit risk model as a signal in advance gives the Bank time to take precautions (sell of the asset) IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 37
  38. 38. Definition of Probability of Default (PD) PD is the greater of One-year PD associated with the internal borrower grade to which that exposure is assigned, OR 0.03% per annum PD of borrowers assigned to a default grade(s) is 100% IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 38
  39. 39. Methods to generate Probability of Default – Basel II recommended techniques Every Risk Grade of the model has a unique Probability of Default Probability of DefaultBased on Internal Mapping to Statistical ModelDefault experience external data Estimates (LR) IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 39
  40. 40. Method 1 – Internal Default Experience Static Pool of Borrowers Transition of Borrower Risk Grades over Time Horizon – Transition MatrixRG1 RG2 RG3 RG4 RG5 RG6 RG7 RG1 0.04RG2 RG2 0.1RG3 RG3 0.2 Probability ofRG4 RG4 0.3 Default estimatesRG5RG6 RG10 0.98RG7 IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 40
  41. 41. Method 2 – Mapping to external ratings 2 2 R = 0.4991 R = 0.631 Mappings R2 = 0.631 2 R = 0.5048 10 2 R = 0.5533 Mapping the Internal 9 2 R = 0.6309 8 7 2 R = 0.5994 Ratings to Risk Grades Risk Scores Series1 6 Expon. (Series1) 5 4 Linear (Series1) of select External Credit Log. (Series1) 3 Power (Series1) 2 Poly. (Series1) Rating agencies 1 Poly. (Series1) 1 2 3 4 5 6 7 8 9 10 Poly. (Series1) Ratings IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 41
  42. 42. Method 2 – Mapping to external ratings IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 42
  43. 43. Method 2 – Mapping to external ratings IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 43
  44. 44. Method 2 – Mapping to external ratings IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 44
  45. 45. Method 3 – Statistical Probability of Default estimates Probability of Default based Account LR Model on the estimating equation Calibration Scale PD Table Calibration Scale RG1 - 0.025 RG1 0.00 – 0.05 RG2 - 0.075RG -> 3 Average PD estimates RG2 0.05 – 0.08 RG3 - 0.10PD -> 0.1 ………………. for every RG RG3 0.08 - 0.12 ………………. ………………. ………………. ………………. ………………. ………………. ………………. ………………. ………………. RG10 - 1.00 RG10 0.85 – 1.00 IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 45
  46. 46. Where does this model fit in to the IRB(F) approach Regulator LGD EAD estimator M • RAROCCorporate Business • Provisioning PD estimates • Expected LossSegment Model • Unexpected Loss • Pricing • Economic Capital for Credit Risk • Investor Transparency • Regulatory Transparence • Securitisation IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 46
  47. 47. IMaCS LGD Calc – An overview Categories of CRM Collateral Guarantee Structure (asset) Haircuts and other deductions Estimated Net Realisable Value of CRM Claims by senior lenders & adjustments with pari passu claims Value of CRM available to YBL IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Loss Given Default Page 47
  48. 48. Characteristic of a good risk scoring model Ability of the model to distinguish “good” borrower from a “weak” borrower Ability of the model to “measure change” in the credit quality of a borrower on a time series Ability of the model to “predict defaults” IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 48
  49. 49. DISCUSSIONS IM aCS 2010 Printed 11-M ay-11For Classroom discussion only Page 49
  50. 50. All the contents of the presentation are confidential andshould not be published, reproduced or circulated without the written consent of IFC, Bangladesh Bank and IMaCS. IM aCS 2010 Printed 11-M ay-11 For Classroom discussion only Page 50

×