1、Modeling Credit Risk in Asian Context:The Building Block ApproachMay 2002,Introduction to Credit Risk Modeling,Introduction: Recent Developments in Credit Risk Modeling,Regulatory New Basel Accord sets regulatory acceptance of internal-rating based models (foundation and advanced approach),Competiti
2、onAsian crisis revealed vulnerability of concentration loans (large loans) to default, inadequate margin to cover credit riskGeneral move to lending to consumers and SMEs, higher volume business, need to balance risk management and credit processing cycle time,TechnologyMore robust credit risk model
3、ing technologyIntegration of data management, statistical analysis and decision support technology,注意,Data issueHistorical portfolio information often not availableApplicability of external data is (highly) questionable Quality of data is highly questionable,Introduction: Challenges in Implementing
4、Credit Risk Models in Asia,Comparability of Financial StatementsLack of transparencyNo publicly traded debtDisparity in accounting standards,Unpredictable recovery ratesTax and regulatory rulesBankruptcy laws underdeveloped and incomparable,注意,Credit Risk Modeling:Conceptual Frameworkand Building Bl
5、ock Approach,Credit Risk Modeling: Conceptual Framework,Credit risk modeling has to be enabled by effective credit risk infrastructure where critical historical risk factors for modeling can be stored. To effectively transform credit risk management of a bank, it is important to integrate credit ris
6、k models and infrastructure into a transformed credit risk process.,注意,Modeling Credit Risk: The Building Block Approach,Phase 1 Component,External Component,Phase 2 Component,Phase 3 Component,方法,Phase 1:Designing and Implementing an Effective Credit Risk Infrastructure,ObjectivesRisk-adjusted pric
7、ingPortfolio risk diversificationRisk capital planningBIS 2 compliance (global best practice)Cross-country analysis,Block 1& 2: Determine Credit Risk Modeling Scope,ExclusionsEmployeesDeceasedSpecific industry,Modeling Credit Risk:Phase 1 - Credit Risk Infrastructure,Realistic implementation plan, t
8、ypically not more than 10 credit risk models for first attempt.,Modeling Credit Risk:Phase 1 - Credit Risk Infrastructure,Block 3: Business Environment Study,Credit Risk InfrastructureDesign,Credit Risk Management ToolsDesign,Credit Risk ProcessIntegration Design,Evaluate Existing Credit Management
9、Systems and IT strategyDevelop IT blue-print for Credit Risk Solution Architecture,Evaluate Existing Credit Risk ModelsDevelop Definition for Exposure, Exposure Aggregation (Related-Party), and Customer, Industry and Product SegmentationFeasibility Study of Empirical Scorecard vs. Subjective Judgmen
10、tal Rating-Based Approach,Map Existing Credit Risk Management ProcessesIdentify Key Credit Controls,No. Past Due 90d,% 90d PD within Industry,No. of Past Due 90d,% of 90d PD within Industry,Distribution of No. of Corporate Client within 90d Past Due by Industry:,Modeling Credit Risk:Phase 1 - Credit
11、 Risk Infrastructure,Block 3: Business Environment Study (I),Wholesale & Retail,Manufacturing,Real Estate,Financial Services,Construction,Professional Services,Social Services,Telecommunication & IT,Utilities,Fishery & Farming,Feasibility Cut-off,注意,Determination of Credit Risk Modeling Approach bas
12、ed on Business Composition Study for CorporateLoan Portfolio:,Modeling Credit Risk:Phase 1 - Credit Risk Infrastructure,Block 3: Business Environment Study (I),Modeling Credit Risk:Phase 1 - Credit Risk Infrastructure,Block 3: Business Environment Study (II),18.9%approved loan were below cut-off,Exa
13、mple of Evaluation of an Internal Subjective Scorecard Model of an Asian Bank Developed for Unsecured Consumer Loan:Excessive subjective intervention to the use of scorecard were found,17.5%above cut-off score were rejected,Subjective Cut-off,注意,Modeling Credit Risk:Phase 1 - Credit Risk Infrastruct
14、ure,Block 3: Business Environment Study (II),Score distribution revealed that model was not able to discriminate between good and bad accounts,Subjective Cut-off,注意,Modeling Credit Risk:Phase 1 - Credit Risk Infrastructure,Block 3: Business Environment Study (III),Example of Definition of Small and
15、Medium Sized Enterprises (SMEs) where banks internal client information is not available:,Underlying assumption: SMEs are non-listed and non-issuer of public debtSegmentation approach: Model listed companies on the stock exchange, determine 5% tail of smallest listed companies as cut-off criteria an
16、d model non-listed companies, determine 25% tail of large non-listed companies. Take lowest cut-off of the 2 cumulative distributionSegmentation criteria: Gross revenue, total asset and total capital,Gross Revenue cut-off at HK$10 million,Gross Revenue cut-off at HK$20 million,注意,Modeling Credit Ris
17、k:Phase 1 - Credit Risk Infrastructure,Block 3: Business Environment Study (III),Total asset cut-off at HK$50 million,Total asset cut-off at HK$30 million,Total asset cut-off at HK$50 million,Total asset cut-off at HK$40 million,Credit risk data is a key driver for developing credit risk management
18、capability. Credit risk data can be effectively used for testing internal rating based models as well as to build statistical models for scoring credit risk. CRDS framework is a comprehensive framework that define the data requirements for developing credit risk databases to support model building a
19、nd testing.CRDS framework for commercial loan comprises of seven building blocks:,Credit Risk Data Store (CRDS) Framework,Obligor Data,Financial Data,Default Status Data,Qualitative Data,FacilityData,InternalRatingData,Recovery Data,Modeling Credit Risk:Phase 1 - Credit Risk Infrastructure,Block 4:
20、Identify Credit Risk Factors,Example of Credit Risk Data Store (CRDS) Framework for Corporate Loan:,Modeling Credit Risk:Phase 1 - Credit Risk Infrastructure,Block 4: Identify Credit Risk Factors,CRDS framework for consumer loan comprises of five building blocks:,Credit Risk Data Store (CRDS) Framew
21、ork,Up-to-date demographics of borrowers and guarantors= Update from host system periodically, e.g. monthly,Loan specific information including account opening date, interest rate, collateral information etc. and periodic account performance information, including monthly payment, outstanding, delin
22、quency status etc.,Written-off accounts data and their recovery information should be stored for future analysis.,Default information at customer level,Borrower/GuarantorData,Default Status Data,Loan Data,Recovery Data,ScoreData,Consumer scoring data history,Modeling Credit Risk:Phase 1 - Credit Ris
23、k Infrastructure,Block 4: Identify Credit Risk Factors,Quantifiable Risk FactorsCurrent / Quick RatiosNPBT / AssetsNPBT YOY GrowthInterest CoverageSize (Sales or TA)Debt Service CoverageInventories / COGSSales YOY GrowthTL / TARE / AssetsNumber of Past dues / ExcessesIndustry Specific Risk Factors e
24、.g. CIDB grade,Non-Quantifiable Risk FactorsAudited / Qualified?Financials submitted within 6 months of FYEManagement ExperienceSuccession RiskFX RiskCountry RiskConcentration RiskCommodity RiskSupply RiskIndustry RiskImplied Access to Capital,Modeling Credit Risk:Phase 1 - Credit Risk Infrastructur
25、e,Example of Risk Factors for Private Firms:,Block 4: Identify Credit Risk Factors,注意,Demographic DataResidence (rent or own)Years at current addressMarital StatusOccupationEmployment HistoryYears on current jobFinancial DataBorrowers earnings stabilitySize of the income cushion represented by debt
26、ratios / savingsBorrowers sources of incomeOther credit cards / loans,Example of Risk Factors for Consumer:,Block 4: Identify Credit Risk Factors,Collateral DataSize of a borrowers equity in a homeAge of the mortgageGeographical locationType of propertySize of propertyLoan to value ratio,Modeling Cr
27、edit Risk:Phase 1 - Credit Risk Infrastructure,Common Issues:Consistency of risk factors captured (i.e. Same calculation of ratios / haircuts assumed etc)Legacy system only captures approved loanNo history, rejected cases not storedSufficiency of default data for model building,Assess Electronically
28、 Stored Information,If Available:Recommend data extraction and storingapproach (warehousing),If NOT Available:Recommend manualdata capture or purchase,If Critical:Re-design Credit InformationManagement Process,Review Risk Factor Requirement,If Non-Critical:Remove from RiskFactor Requirement,Block 5:
29、 Assess Data Availability,Modeling Credit Risk:Phase 1 - Credit Risk Infrastructure,Modeling Credit Risk:Phase 1 - Credit Risk Infrastructure,Block 6: Design Credit Risk Solution Architecture,Loan Servicing Systems,Back-end host,Presentation,Decision Support,Credit risk modeling engine,Data Store an
30、d Warehouse,Credit Risk Decision Support System (Score/Rating Application),Credit Risk DSS Database,Phase 2:Model Selection, Buildingand Testing,Models available in the market:,Statistical approach using regression techniques (logit, probit, linear), neural network, decision-tree to model the probab
31、ility of default. A dependent variable is explained by a set of independent variables. Work best with limited data.,Discriminant Models,Rule-Based orExpert Models,Model against a structured process that an experienced analyst uses to arrive at the credit decision. Model is characterized by set of de
32、cision rules. Assumptions made are not tested and allocation of weights for decision variables is subjective. Good as first attempt.,Market Models,Models calibrate probability of default based on market prices, e.g. equity prices or bond spreads. Models are limited to firms with sufficient liquid ma
33、rket prices and where financial markets are efficient. Limited application in Asia (ex-Japan).,Agency Rating-BasedModels,Models based on rating published by rating agencies or internal rating, incorporating loss given default and default probability per rating scale, probability matrices of rating m
34、igration and expected recovery per rating scale. Model is limited to rated institutions and is used at portfolio-level.,Modeling Credit Risk:Phase 2 - Model Selection, Building & Testing,Block 8: Model Selection,Partially Applicable,Fully Applicable,Not Applicable,Models available in market:,Modelin
35、g Credit Risk:Phase 2 - Model Selection, Building & Testing,Block 8: Model Selection,注意,Integrated Credit Risk Measurement:,Block 8: Model Selection,Modeling Credit Risk:Phase 2 - Model Selection, Building & Testing,Expected Loss = EAD x LGD x POD,注意,Block 8: Model Selection,Modeling Credit Risk:Pha
36、se 2 - Model Selection, Building & Testing,Key Credit Risk Measurement Development Process:,Internal rating-based model, estimating expected losses and generating credit value-at-risk are three distinct stages of credit risk measurement development process,Modeling Credit Risk:Phase 2 - Model Select
37、ion, Building & Testing,Block 9: Acquire Software & Data,Andersens system selection services:,Modeling Credit Risk:Phase 2 - Model Selection, Building & Testing,Modeling Credit Risk:Phase 2 - Model Selection, Building & Testing,Modeling Credit Risk:Phase 2 - Model Selection, Building & Testing,Model
38、ing Credit Risk:Phase 2 - Model Selection, Building & Testing,Logical TestApplicable for discriminant model onlyRelationship for independent variables must be validated, e.g. higher leverage results in higher probability of defaultNeed to decide to drop independent variable(s),Block 12: Model Testin
39、g,Stability TestAccuracy Ratio TestStatistical Robustness Test, e.g. Maximum Likelihood Test for Logit Models,Model Robustness Test,A) Stability TestDescription: A routine to test consistency of the estimated coefficients of the model constructed. Typically, a sample for model build is regressed 500
40、 times.Measurement: 1) Order Test - the order of estimated coefficients must be consistent in all 500 run. Number of run with inconsistent order of estimated coefficients are logged.2) Standard Error Test - each estimated coefficient from the 500 runs is measured for the standard error.,Modeling Cre
41、dit Risk:Phase 2 - Model Selection, Building & Testing,Block 12: Model Testing,B) Accuracy Ratio TestDescription: Ideally, a credit risk model should reject 100% of bad credits from sample(population) and 0% of good credits from sample(population). Accuracy ratio is to test the performance on the mo
42、del based on cut-off point determined.Measurement: 1) First Accuracy Ratio is defined as the rejection rate of bad credits less rejection rate of good credits at a given optimized cut-off point.2) Second Accuracy Ratio is applied to out-of-test sample based on estimated coefficients from model build
43、. Degradation of the Second Accuracy Ratio is monitored.,Modeling Credit Risk:Phase 2 - Model Selection, Building & Testing,Block 12: Model Testing,注意,B) Accuracy Ratio TestAccuracy Ratio degrade from 73% to 54% after applying out-of-sample data-set.,Modeling Credit Risk:Phase 2 - Model Selection, B
44、uilding & Testing,Block 12: Model Testing,?,C) Maximum Likelihood TestDescription: Maximum likelihood test is a test against the significance of hypothesis given a data-set used for model build.Measurement: 1) The Log of Likelihood Function is defined as Log(P). Given S is the number of successes (Y
45、t=1) observed in I observations, then for the logit model, the maximum value of the log likelihood function under the null hypothesis can be defined as:A test of null hypothesis that all the slope coefficients are zero can be carried out using the likelihood ratio (LR) test. LR test statistic is def
46、ined as 2Log(P) - Log(0). If the maximized likelihood under null hypothesis (H0), Log(0), is much smaller that the unrestricted maximized likelihood, Log(P). Therefore, a LR test statistic of more than zero would indicate an evidence against the null hypothesis.,Modeling Credit Risk:Phase 2 - Model
47、Selection, Building & Testing,Block 12: Model Testing,?,Phase 3:Model Deployment and Process Integration,Credit Acquisition,Account Maintenance,Collections,Credit Risk Processing,Credit Risk DSS,Write-off,Product Planning & Policy Integration,Application ScoreAcquisitionPricing,Behavioral ScoreAttri
48、tionCross-sellFraud detection,Recovery ScoreCollectionWorkout,Modeling Credit RiskPhase 3 - Deployment & Integration,Block 13: Credit Risk Decision Support System (CRDSS),ALCO Integration - Volume, Rate, MixRAROC,Modeling Credit RiskPhase 3 - Deployment & Integration,Block 13: Credit Risk Decision Support System (CRDSS),CRDSS stores and integrates the credit policy, credit models and business rules into bank-wide loan origination process.,Modeling Credit RiskPhase 3 - Deployment & Integration,Block 13: Credit Risk Decision Support System (CRDSS),
