毕博上海银行咨询Final Deliverables technicalfinal1.ppt

Table of Contents,1Executive Summary2CRMS Project Review3CRMS Project Commercial Process Improvements4HAVICS System Overview 5HAVICS System Technical Components,Table of Contents,1Executive Summary2CRMS Project Review3CRMS Project Commercial Process Improvements4HAVICS System Overview 5HAVICS System Technical Components 5.1Rating Methodology Overview5.2Rating Decision Support Parameters5.3Pricing Methodology Overview5.4Pricing Decision Support Parameters5.5Limits Methodology Overview5.6Limits Decision Support Parameters,Hanvit Commercial Credit Rating Model Development Support,Risk Rating Foundation SourcesThe Hanvit Risk Rating System incorporates:“Best Practices”noted in Korean,U.S.and International banks;Risk issues that are specific to Hanvit and Korean business structure as identified by KPMG and management of Hanvit;andFSS guidelines.,PurposeThe principal objectives for this module is to design a standardized,effective risk rating system for the commercial loan portfolio based on FSS and International Best Banking Practices.Risk ratings are the primary summary indicator of risk for Hanvit banks individual credit exposures.Specific uses of Hanvits risk ratings include:a guide for the loan origination process;portfolio monitoring and management reporting;analysis of the adequacy of loan loss reserves;loan pricing analysis;andan input to the portfolio management module,Distinction Between Default and LossThe overriding principal of risk rating is that the individual ratings are first aligned by borrower with risk of default.This risk of default measure is produced at the borrower level due to the presumption that a borrower will default on all obligations if it defaults on any.Individual borrower ratings are useful in that they:facilitate the understanding to the Bank of the mechanics of risk in the portfoliosupport pricingsupport marketing initiativesThe facility rating adjusts the original borrower rating based on the fraction of the loans value that is likely to be lost in the event of default.These adjustments vary by facility structure such as collateral type.The resulting final facility rating,representing expected and unexpected loss,corresponds to the product of the two concepts of default and loss in the event of default.,Borrower Grade,12345678910,Facility Info,1 2345678910,Facility Grade,Adjusts for LIED,Virtually no risk,Higher risk of default,Very little risk of loss,Definite risk of loss,with,gives,EDF,LIED,ELR,*,=,Hanvit Commercial Credit Rating Model Development Support,Borrower Scoring Model TheoryBased upon past modeling experience and test results from Korean audited company data,the Merton approach to default estimation forms the basis of borrower scoring.The general form of this model assumes that default occurs when a firms asset value falls sufficiently below the face value of its debt.,Hanvit Commercial Credit Rating Model Development Support,Borrower Scoring Model SpecificationAlthough the Merton approach is ideal for large corporate and middle market public companies,most loans made by the bank do not fall into this category.To create a default scoring model with broad applicability,proxy measures are used in place of asset value.Empirical knowledge of traditional credit analysis combined with testing has shown that historical average cash flow divided by the current value of debt is a good proxy for the idealized leverage ratio.In alternate arithmetic form,this ratio is represented by cash flow adjusted by subtracting out interest expenses.A highly leverage firm may have similar default risk to a less leveraged one,if the more leveraged firm has lower volatility.In broad terms,the ratio of the leverage term defined above to the volatility of cash flow is termed“cash flow default distance”.Not surprisingly,we find that the leverage used by volatile,small-capitalization firms generally falls short of that used by stable,large-capitalization firms.,Cash Flow Default Distance,Average Interest Expenses,Volatility of Cash Flow,Average Cash Flow,Hanvit Commercial Credit Rating Model Development Support,Borrower Scoring Model Specification(continued)In accordance with revised FSS guidelines,default distance focuses on future cash flow.A review of default situations in the Korean market has shown that balance sheet factors can also play a significant role in determining default status.Specifically,assets are sometimes used to supplement cash flow to continue operations for a period in an attempt to return to a more healthy cash flow situation.Equivalent to the cash flow default distance measure,a balance sheet default distance is used as a supplement to help predict default status.Testing for historical significance has shown that a geometric form of the equation,using the logarithm of assets to debt ratio,holds the most predictive power.,Balance Sheet Default Distance,Volatility of Assets as a percentage of Average Assets,Assets-Equity(Debt Equivalent),Assets,Hanvit Commercial Credit Rating Model Development Support,Initial Borrower Scoring ModelIf a borrower is a member of a traditional Korean business group,then both cash flow and balance sheet default distances are also calculated for that group.Their influence is used to help determine initial default probability.,Default DistanceCalculations,1-year financial default probability,Company Financials,Borrower Scoring Model,GroupFinancials,Hanvit Commercial Credit Rating Model Development Support,Combining Judgmental Factors with Borrower ScoringAlthough default distance is a reliable indicator of future default,there are compelling reasons for looking to other explanatory variablesMany large international banks use statistical models as an element of the rating process,but those banks generally believe that the limitations of statistical models are such that properly managed judgmental rating systems deliver more accurate estimates of riskWhen considering the banks larger exposures where an inaccuracy by a statistical model could place the bank in jeopardy,the benefits of higher accuracy outweigh the higher cost of a judgmental system When properly structured to reflect the opinion of how the future performance of a company will be affected,subjective factors can increase discrimination in the mid-section of the default probability curve where default distance is less discriminatory than at the endsGiven the role of subjective judgment in the rating process,the Bank will pay careful attention to the internal incentives and the internal rating and review control systems to avoid introducing bias.Since subjective factors that were captured in the past have proven not to be very reliable indicators of future default,it is anticipated that the initial rating model will likely rely on less weight for the subjective factors than if a full history were available,Hanvit Commercial Credit Rating Model Development Support,Combining Judgmental Factors with Borrower Scoring(continued)A final benefit of incorporating subjective measures into the rating process may not be recognized in the immediate future but will eventually provide a substantial return.By beginning the collection of meaningful subjective measures within a bias free process,the Bank will eventually have a reliable set of data for use in a continual model improvement and calibration process.Subjective factor weights can then be derived with a more solid statistical basis.Five main judgmental categories are used to summarize the factors.,Hanvit Commercial Credit Rating Model Development Support,Combining Judgmental Factors with Borrower Scoring(continued)To determine how meaningful the judgmental factors are,and thereby determining how much weight should be placed on them when combined with all other variables including default distance,these non-financial variables are included in a logistic equation along with default distance to compute the probability of default.,New companies,small companies and numerous non-audited medium size companies lack sufficient financials to generate meaningful results in any at all.In these situations,the borrower rating model then simply incorporates only the non-financial variables.,Hanvit Commercial Credit Rating Model Development Support,Combining Judgmental Factors with Borrower Scoring(continued)In its final form,the probability of default of a company relies on 1)both company and group measures of cash flow and balance sheet default distance;2)all the judgmental factors;and 3)industry groupings.A logistic regression is used to determine the parameters for converting default distance and judgmental factors to a probability of default.,e(l0+(l1*CCFDD)+(l2*GCFDD)+(l1*CBSDD)+(l2*GBSDD)+(l3*NF1)+.)1+e(l0+(l1*CCFDD)+(l2*GCFDD)+(l1*CBSDD)+(l2*GBSDD)+(l3*NF1)+.),PD=,CCFDD=Company Default DistanceGCFDD=Group Default DistanceCBSDD=Company Default DistanceGBSDD=Group Default DistanceNF=Non-Financial and Industry Factorsl=Equation Intercept and Weights,Facility Grading InputsGiven a Borrower Risk Grade and other inputs,the loan valuation algorithm(explained in the loan pricing documentation)is used to rank order facilities based on loan type and collateral.An expected loss rate is mapped over to a facility grade.It is anticipated that all inputs required by the algorithm will be input as part of the loan pricing exercise,leading to an automated facility rating.The only exceptions to this automation are facilities requiring manual override or part of the classified rating process.,EDF*LIED*Exposure,AssumptionsRecovery Rates by collateral typeNormal LIED,User InputsBorrower RatingLoan TypeExpected UsageCollateral Ratio,Expected Loss Rate,1 2 3 4 5 6 7 8 9 10,Hanvit Commercial Credit Rating Model Development Support,Data LimitationsThe availability of history that is long and clean enough to support the desired coverage of the rating process may be limited.Where there are data difficulties,these are typically areas where less predictive accuracy is required since the majority of the exposure risk to the bank is located within the Large Corporate Audited segment.,Large Audited,Medium andSmall Audited,Small non-Audited,Group,Data Limitations,Default PredictionAccuracy Generally Required,20 bps,+,_,100 bps,+,_,300 bps,+,_,Few,Occasional non-contiguous data history creates difficulty in gathering time series,Many difficulties collecting financial history,Where a default distance calculation can be done,the standard rating process is combined with the judgmental overlay to arrive at the final rating.Otherwise,the judgmental process is used exclusively.,Judgmental process used exclusively.Retail scorecard used for very small business.,Table of Contents,1Executive Summary2CRMS Project Review3CRMS Project Commercial Process Improvements4HAVICS System Overview 5HAVICS System Technical Components 5.1Rating Methodology Overview5.2Rating Decision Support Parameters5.3Pricing Methodology Overview5.4Pricing Decision Support Parameters5.5Limits Methodology Overview5.6Limits Decision Support Parameters,Hanvit Commercial Credit Rating Model Development Support,Rating ScaleThe most important thing about the number of grades is that there are sufficient gradations to permit accurate characterization of the underlying risk profile of a loan,or a portfolio of loans.The scale to be used starts with a reference point suggested by the FSC which gives guidelines as to the number of pass grades,seven.There are two competing notions to the number of risk grades.On the one hand,it is desirable to have more grades than less due to the ability to more finely distinguish risks(and its associated costs).However,when judgment plays a role in grading,the number of distinct grading characteristics that can be described limits the number of grades.Finally,a goal of the rating scale is to distinguish the areas where the most exposure lies(typically rating systems that are not sufficiently distinguished tend to“bunch”exposure in only a few of the Pass grades).,Rating Time HorizonThe scoring system generates a default probability that is based on one year.With the addition of forward looking non-financial variables the time horizon remains one year because the weights within the default probability equation are derived from the regression which is lagged by one year.,Hanvit Commercial Credit Rating Model Development Support,Rating Scale(continued)Specific problems with ratings scales that do not include a sufficient number of risk grades include:Margins in the lower end of the acceptable grades are relatively wideThe bank is susceptible to adverse selection by overpricing lower risk segments and underpricing higher risk segments,Risk Grade,.,.,.,.,Equity Allocation,“Average”equity allocation for each risk grade,Below average riskoverpriced,Above average riskunderpriced,Risk will increase over time,Hanvit Commercial Credit Rating Model Development Support,Rating Scale(continued)The rating scale addresses objectives of both the“Pass”and“Problem”portfolio segments.,PASS PORTFOLIO,PROBLEM PORTFOLIO,RISK RATINGSProblem Portfolio Pass Portfolio,Ratings Align with FSS Guidelines,Ratings Align with EDF and LIEDcombination,RISK OF DEFAULT/RISK OF LOSS ANALYSIS,Transactional IssuesPricingStructurePortfolio IssuesConcentrationsRisk Acceptance Criteria,Regulatory IssuesTemporary Impairment(Allowance)Permanent Impairment(Loss Recognition),Hanvit Commercial Credit Rating Model Development Support,Rating Scale(continued)There is no need to have different scales for the borrower rating and the facility rating.The notched risk grade scheme to be used(shown below)is based on a traditional and familiar risk scale while adding necessary degrees of risk distinction.,Risk Grade,12345678910,+-+-+-+-+-,Pass Portfolio,Problem Portfolio,Hanvit Commercial Credit Rating Model Development Support,Expected Default Frequency MappingA compromise among three factors was used to develop the definition of Expected Default Frequency(EDF)bin widths.This compromise divides up the Pass grades and determines the relative widths of all bins.1.Equal probability bins2.Similar bin widths3.Convenient mapping to S&P gradesThere is obvious conflict among these three conventions.Therefore,judgement must be used to determine the most practical solution.,Hanvit Commercial Credit Rating Model Development Support,Expected Default Frequency Mapping(continued)Provision requirements have been mandated by the FSS.Those provision amounts are equivalent to Expected Loss values.Equivalent EDF values can be calculated from those provision percentages.Based on simulation results,approximately 50%of the Banks current large audited portfolio would fall below the Pass criteria after initial default model scoring.,Risk Grade,12345678910,ApproximateEquivalent EDF*,Required Provision fromFSS Requirements(EL),1.1%,4.4%,44.4%,100.0%,0.5%,2.0%,20.0%,50.0%,100.0%,100.0%,*Created using the assumption of 45%LIED,Nearly 50%of portfolio falls below Pass,Approximately only 50