Credit Scoring In The Leasing Industry · A Knowledge-Based Credit Decision System FASTER & BETTER...
Transcript of Credit Scoring In The Leasing Industry · A Knowledge-Based Credit Decision System FASTER & BETTER...
1
Credit Scoring In The Leasing IndustryCredit Scoring In The Leasing Industry
Vernon GeretyVernon GeretySenior Vice PresidentSenior Vice President
PBDSPBDSPredictiveMetricsPredictiveMetrics
[email protected]@predictivemetrics.com
732732--530530--29802980
ELA CreditELA Credit & Collection Management ConferenceCollection Management Conference
June 9 June 9 –– 11, 200211, 2002
2
Discussion pointsDiscussion points
nn Leasing OverviewLeasing Overview
nn Scoring 101Scoring 101
nn Application of Scoring in the Leasing IndustryApplication of Scoring in the Leasing Industry
nn ROI of ScoringROI of Scoring
nn Q&AQ&A
3
Leasing Industry & ScoringLeasing Industry & Scoring
4
Leasing IndustryLeasing Industry
uu Quick, cost effective decisions are criticalQuick, cost effective decisions are critical
uu Competes with banks, specifically the credit card industry, Competes with banks, specifically the credit card industry, for small businesses working capital needsfor small businesses working capital needs
uu Scoring provides a cost effective risk management toolScoring provides a cost effective risk management tool
uu Custom and Generic scores are prevalentCustom and Generic scores are prevalentØØ Custom for high volume operationsCustom for high volume operations
ØØGeneric for Bank / Middle Market Lessors and BrokersGeneric for Bank / Middle Market Lessors and Brokersvv D&B Leasing Industry ModelD&B Leasing Industry Modelvv FICO Liquid Credit FICO Liquid Credit vv Experian BIS Experian BIS –– NEW Lease Decision ScoreNEW Lease Decision Scorevv PayNet Lease ScorePayNet Lease Score
The question is NOT The question is NOT ““Do I use scoring?Do I use scoring?”” but but ““Which score should I use?Which score should I use?””
5
CUSTOMERS DEMAND MORE APPROVALS…FASTERCUSTOMERS DEMAND MORE APPROVALS…FASTER
1) 1) Say Yes . . . Say Yes . . . ALWAYSALWAYS
2) Tell Me . . . 2) Tell Me . . . QUICKLYQUICKLY
3) Fund Me . . . 3) Fund Me . . . NOWNOW
Small Ticket LeasingSmall Ticket LeasingCase StudyCase Study
Customer / Vendor / Dealer / Broker RequirementsCustomer / Vendor / Dealer / Broker Requirements
6
Score Plus Leasing Case StudyScore Plus Leasing Case StudyA KnowledgeA Knowledge--Based Credit Decision SystemBased Credit Decision System
FASTER & BETTER DECISIONSFASTER & BETTER DECISIONS
D AnalystAnalystReviewReview
ValidatedValidatedPolicyPolicy+Score PlusScore Plus = ScoringScoring
ModelsModels
A,B,C AutoAutoApproveApprove
F AutoAutoDeclineDecline
All Applicants
CreditScore
PolicyRules
Credit Knockout Credit Knockout Analyst ReviewAnalyst Review
7
Score Plus Leasing Case StudyScore Plus Leasing Case StudySmall Ticket Commercial Lease FinancingSmall Ticket Commercial Lease Financing
Use of Score Plus To Help Risk Rate The Portfolio Will Use of Score Plus To Help Risk Rate The Portfolio Will Improve The Ability To Manage New & Existing CustomersImprove The Ability To Manage New & Existing Customers
“Through The Door” ApplicationsDistribution Of Applications
A B C D FFailure Rates For All Applications
4 Years
A B C D F
Approved And Booked Deals
% Loss/Collections For Funded Deals4 Years
Distribution Of Approved Deals
A B C D F
A B C D F
8
Scoring 101Scoring 101
9
Statistical Model DevelopmentStatistical Model DevelopmentPictorial Pictorial OverviewOverview
HistoricalHistoricalDataData
Statistical Statistical AnalysesAnalyses
Scoring Utilizes Past Experiences To Scoring Utilizes Past Experiences To Statistically Predict Future EventsStatistically Predict Future Events
PredictPredictFuture EventsFuture Events
PBDS ScoresPBDS ScoresABC Corp = 100ABC Corp = 100XYZ Corp = 71XYZ Corp = 71JKL Corp = 45JKL Corp = 45DEF Corp = 23DEF Corp = 23
BadBad GoodGood
YourYourCompanyCompany
Statistical models provide a superior risk tool by; Statistical models provide a superior risk tool by; 1.1.Picking the most significant predictors of risk from Picking the most significant predictors of risk from
100’s of possibilities.100’s of possibilities.2.2.Determining the relevant importance of each Determining the relevant importance of each
predictive variable (aka Weight).predictive variable (aka Weight).
10
Statistical Model Development Statistical Model Development Types Of Scoring ModelsTypes Of Scoring Models
New Application ModelsNew Application ModelsAssess the creditworthiness of new lease applicants
nn Credit Bureau Data: Commercial & ConsumerCredit Bureau Data: Commercial & Consumeruu Dun & Bradstreet & Experian BISDun & Bradstreet & Experian BIS
ss Trade Data, Public Records, Business TenureTrade Data, Public Records, Business Tenureuu Consumer Bureau For Business Principals Consumer Bureau For Business Principals
ss Small Office / Home Office and StartSmall Office / Home Office and Start--upsups
nn Application DataApplication Datauu Unique ID Information for Matching Unique ID Information for Matching
ss Name, Street Address, City, State, ZIP, TelephoneName, Street Address, City, State, ZIP, Telephoneuu Information Specific To DealInformation Specific To Deal
ss Equipment Type: Office, Computer, Construction,…Equipment Type: Office, Computer, Construction,…ss Channel: Vendor, Broker, Direct, Private Label,…Channel: Vendor, Broker, Direct, Private Label,…ss Deal Size: Micro, Small, MidTicket,…Deal Size: Micro, Small, MidTicket,…
11
Statistical Model Development Statistical Model Development Types Of Scoring ModelsTypes Of Scoring Models
Behavior ModelsBehavior ModelsMonitor the credit risk of existing customersMonitor the credit risk of existing customers
uu New Lease AuthorizationsNew Lease Authorizations
uu Dynamic Portfolio Monitoring: Loss ReservesDynamic Portfolio Monitoring: Loss Reserves
uu Early Stage CollectionsEarly Stage Collections
nn Internal Customer DataInternal Customer Datauu Profile Data Profile Data
ss Tenure, Equipment Type, Exposures…Tenure, Equipment Type, Exposures…
uu Accounts Receivable Data Accounts Receivable Data
ss Payment Detail, Aging, WritePayment Detail, Aging, Write--offs….offs….
nn Bureau DataBureau Datauu Similar to New Application Model But Significantly Less ImportanSimilar to New Application Model But Significantly Less Importantt
ss “Hit” Versus “No Hit” Model“Hit” Versus “No Hit” Model
12
Statistical Model DevelopmentStatistical Model DevelopmentSteps In Developing Scoring ModelsSteps In Developing Scoring Models
Identify Business Objective: The Goal Of The ModelIdentify Business Objective: The Goal Of The Modelnn Type Of Model: “New Accounts … Existing Customers” Type Of Model: “New Accounts … Existing Customers”
nn Data Available: “What do you need?” … “What do you have?”Data Available: “What do you need?” … “What do you have?”
nn Behavior to be Predicted: “What Is A GOOD / BAD Account?”Behavior to be Predicted: “What Is A GOOD / BAD Account?”
Database DesignDatabase Designnn Specify TimeframesSpecify Timeframes
nn Pull DataPull Data
nn Develop Model Development SampleDevelop Model Development Sample
Conduct Segmentation AnalysisConduct Segmentation Analysisnn Maximize Model's Predictive PowerMaximize Model's Predictive Power
nn Apriori Versus Statistically DeterminedApriori Versus Statistically Determined
13
Preliminary Model Development Preliminary Model Development •• Create Data Attributes (100’s)Create Data Attributes (100’s)
•• Identify Predictive BiIdentify Predictive Bi--Variate Candidates (30 Variate Candidates (30 -- 75)75)
•• Determine Statistically Significant Model Variables (10 Determine Statistically Significant Model Variables (10 –– 25)25)
Holdout Validation And Finalize ModelHoldout Validation And Finalize Model•• Development Vs. Holdout Sample. Preferably Forward Sample.Development Vs. Holdout Sample. Preferably Forward Sample.
•• Other Statistical Methods: KS & Predictive Index. Flawed StatistOther Statistical Methods: KS & Predictive Index. Flawed Statistics.ics.
Implementation And TestingImplementation And Testing•• Programming Of Technical Specifications Into Product SystemProgramming Of Technical Specifications Into Product System
•• Testing Up To 1,000 cases. Production Versus Model CodeTesting Up To 1,000 cases. Production Versus Model Code
Statistical Model DevelopmentStatistical Model DevelopmentSteps In Developing Scoring ModelsSteps In Developing Scoring Models
14
Credit Model DevelopmentCredit Model DevelopmentDatabase DesignDatabase Design
Historical WindowHistorical Window11 Months11 Months
ObservationObservationWindowWindow1 Month1 Month
Performance WindowPerformance Window24 Months24 Months
6/996/99 6/006/00 7/007/00 6/026/02
Time Period Time Period --1212 --1111 --1010 --99 ** --33 --22 --11 00 11 22 33 ** 2121 2222 2323 2424
Cust A Cust A GG GG GG GG ** GG GG GG GG GG GG GG ** GG GG GG GG
Cust BCust B G G GG GG GG GG G G GG GG GG GG GG ** GG BB BB BB
Cust CCust C GG GG ** GG BB BB BB GG GG GG ** GG GG GG GG
Cust ACust A GG ** GG GG BB BB
GG: Account currently in Good standing : Account currently in Good standing BB: Account currently in BAD standing.: Account currently in BAD standing.
§§ Behavior Models typically require 24 Behavior Models typically require 24 -- 36 months of customer history 36 months of customer history including a 6 including a 6 –– 12 month Historical Window of A/R past payments.12 month Historical Window of A/R past payments.§§ New Account Models require a minimum of 12 New Account Models require a minimum of 12 -- 24 months and DO 24 months and DO
NOT have a Historical Window since no history exists with the fiNOT have a Historical Window since no history exists with the firm.rm.
15
Segmentation RationaleSegmentation RationaleIncrease Scores Performance By Finding TheIncrease Scores Performance By Finding The
Optimal Number of Homogeneous GroupsOptimal Number of Homogeneous GroupsTo Develop ModelsTo Develop Models
0
20
40
60
80
100
0 10 20 30 40 50 60 70 80 90 100
Percentile
Fu
ture
Ba
d %
SegmentationNo SegmentationRandom
16
Differences In Customer’s ProductDifferences In Customer’s Productnn True Lease vs. Finance LeaseTrue Lease vs. Finance Lease
Exploit Differences In Information ContentExploit Differences In Information Contentnn Current Payers vs. History of Late PaymentCurrent Payers vs. History of Late Payment
Account For Differences In Behavioral DriversAccount For Differences In Behavioral Driversnn Large vs. Small FirmsLarge vs. Small Firms
Account For Unique Characteristics Of Credit PolicyAccount For Unique Characteristics Of Credit Policynn Special Or Preferential Treatment for Large ExposuresSpecial Or Preferential Treatment for Large Exposures
Exploit External Data When AvailableExploit External Data When Availablenn Bureau Hit vs. No / Thin HitBureau Hit vs. No / Thin Hit
Segment IdentifiersSegment Identifiers
17
Predictiveness IndexPredictiveness IndexModel Screening EffectivenessModel Screening Effectiveness
Calculating The PBDS PICalculating The PBDS PI
Assuming a 10% BAD RateAssuming a 10% BAD Rate
A perfect model would screen out A perfect model would screen out 100% of the BADS in the worst 10%100% of the BADS in the worst 10%
The Predictiveness Index measures the The Predictiveness Index measures the Models “perfect” percentageModels “perfect” percentage
P = (Area X + Area Y) is the capture P = (Area X + Area Y) is the capture rate from perfect predictivenessrate from perfect predictiveness
M = Area Y is the capture rate of the M = Area Y is the capture rate of the estimated modelestimated model
PI = M/PPI = M/P
A PI of 55 suggests that the estimated A PI of 55 suggests that the estimated model captures 55% of the perfect areamodel captures 55% of the perfect area0
10 20 30 40 50 60 70 80 90 100
Percentile Rank Order From Highest To Lowest RiskPercentile Rank Order From Highest To Lowest Risk
0
20
40
60
80
100
Model
Random
%% Of Bad CapturedOf Bad Captured
Perfect
Area X
Area Y
18
Application of Scoring Application of Scoring in the Leasing Industryin the Leasing Industry
19
Statistical Model DevelopmentStatistical Model DevelopmentMethods Of Score EstimationMethods Of Score Estimation
Good
Better
BestCustom Statistical
Credit Scores
Generic or Industry Credit Scores
Rules Based Systems
20
New Application Model New Application Model Small to Mid Ticket LeasingSmall to Mid Ticket Leasing
Statistical decision tool that Statistical decision tool that measures future credit risk measures future credit risk
Generally used for Generally used for transactions under $100,000transactions under $100,000
AccuratelyAccuratelySpeedsSpeeds
The NewThe NewCredit Credit
DecisionDecisionProcessProcess
The Goal Is To EffectivelyThe Goal Is To EffectivelyAutomate The Majority Of Automate The Majority Of
Real Time Decisions AtReal Time Decisions AtTime Of ApplicationTime Of Application
Uses application, Uses application, credit bureaucredit bureau
21
New Application ModelNew Application ModelSmall Ticket LeasingSmall Ticket Leasing
0%
20%
40%
60%
80%
100%
0 10 20 30 40 50 60 70 80 90 100
Percentile
Bad
% CommercialConsumerBlendedRandom
Optimizing Data Sources Using Optimizing Data Sources Using Advanced Statistical MethodsAdvanced Statistical Methods
22
Commercial Lease Decision ScoreCommercial Lease Decision ScorePooled Leasing ModelPooled Leasing Model
Score Classes 1, 2 & 3 are designed to be Lo w to Score Classes 1, 2 & 3 are designed to be Lo w to Score Classes 1, 2 & 3 are designed to be Lo w to Score Classes 1, 2 & 3 are designed to be Lo w to M o d erate Risk, Score Class 4A is High Risk, Score Class M o d erate Risk, Score Class 4A is High Risk, Score Class M o d erate Risk, Score Class 4A is High Risk, Score Class M o d erate Risk, Score Class 4A is High Risk, Score Class 4B Very High Risk and Score Class 5 is Extre m e Risk.4B Very High Risk and Score Class 5 is Extre m e Risk.4B Very High Risk and Score Class 5 is Extre m e Risk.4B Very High Risk and Score Class 5 is Extre m e Risk.
* * * * Average Risk is set to 100. A value of 25 represents 1/4 of the Average Risk is set to 100. A value of 25 represents 1/4 of the average risk, a value of 450 is 4.5 times above the average riskaverage risk, a value of 450 is 4.5 times above the average risk..
2.4%
29.8%
44.3%
14.1%
7.8%
1.6%
0%5%
10%15%20%25%30%35%40%45%50%
1 2 3 4A 4B 5
20.841.6
80.3
158.3
279.2
450.5
0
100
200
300
400
500
1 2 3 4A 4B 5
23
Behavior Scoring ModelsBehavior Scoring ModelsMultiple Applications To Maximize ProfitsMultiple Applications To Maximize Profits
Assesses the risk that a delinquent Assesses the risk that a delinquent account will become severely account will become severely
delinquent or written offdelinquent or written off
Prioritizes early stage collection Prioritizes early stage collection activitiesactivities
Reduces collection costs and Reduces collection costs and lowers delinquency rateslowers delinquency rates
CollectionCollection RecoveryRecoveryEvaluates the recovery rate Evaluates the recovery rate
of severely delinquent or of severely delinquent or written off accounts and can written off accounts and can be used for pricing a portfoliobe used for pricing a portfolio
Rank orders accounts by Rank orders accounts by expected recovery dollarsexpected recovery dollars
Maximizes recovery ratesMaximizes recovery rates
AuthorizationAuthorizationAssesses the risk from Assesses the risk from
expanding a credit relationship expanding a credit relationship with an existing accountwith an existing account
Automation of credit decisions Automation of credit decisions and line managementand line management
Reduces credit costs and Reduces credit costs and increases revenuesincreases revenues
24
Greatest RiskGreatest Risk Least RiskLeast Risk
Leasing Behavior ModelLeasing Behavior ModelCustom Versus Generic ScoresCustom Versus Generic Scores
For the 30% of Highest Risk For the 30% of Highest Risk customers, the custom model customers, the custom model identifies 868 additional BADsidentifies 868 additional BADs
Custom Behavior ScoreCustom Behavior Score
KS = KS = 0. 380. 38
PI = 51.36PI = 51.36
Commercial ScoreCommercial Score
KS = KS = 0.150.15
PI = 19.32PI = 19.32
0
20
40
60
80
100
0 10 20 30 40 50 60 70 80 90 100
Percentile
Bad
%
Custom Score
Commercial Score
Random
25
0
50
100
150
200
250
Bad
Rat
eA B C D F
Score Class
Leasing Case StudyLeasing Case StudyBehavior Model PerformanceBehavior Model Performance
4.4%
30.3%
38.9%
16.5%
2.6%
0%
10%
20%
30%
40%
% o
f Acc
ount
s
A B C D F
Score Class
Distribution by Risk GradesDistribution by Risk Grades
Risk Grades determine Approved / Decline decision. A, B, C Risk Grades determine Approved / Decline decision. A, B, C are Auto Approved up to $75,000 Exposures. D requires are Auto Approved up to $75,000 Exposures. D requires further review by an analyst and F’s are Auto Declines.further review by an analyst and F’s are Auto Declines.
Loss Rate IndexLoss Rate IndexAverage = 100
26
Return On Invest m e nt Return On Invest m e nt Using ScoringUsing Scoring
27
Risk Grade
Old
Approval Rate
New
Approval Rate
Incremental Deals
Loss Rate
# Losses $ Loss Profit
A 98.0% 99.5% 42 0.8% 0 1,331$ 9,979$
B 95.0% 99.0% 764 2.5% 19 76,356$ 183,254$
C 92.5% 97.5% 1,225 5.5% 67 269,577$ 294,084$
D 75.0% 70.0% (750) 15.2% (114) (455,818)$ (179,928)$
F 20.0% 0.0% (328) 32.9% (108) (431,122)$ (78,624)$
Total 87.4% 89.0% 953 (135) (539,676)$ 228,766$
ROI Of Improved Credit DecisionsROI Of Improved Credit DecisionsBased Upon One Year Of VolumeBased Upon One Year Of Volume
Improved decisions using Credit Ratings will allow you to approvImproved decisions using Credit Ratings will allow you to approve e better credits (A, B & C) and avoid unprofitable credits (D & F)better credits (A, B & C) and avoid unprofitable credits (D & F)..
Loss Avoidance: Loss Avoidance: Avoiding losses from D & F credits.Avoiding losses from D & F credits.
135 additional net losses avoided at $4,000 per loss. Total Loss Savings = $539,676.
Approve More Deals:Approve More Deals: Approving more A, B & C credits.Approving more A, B & C credits.
953 additional deals booked. Incremental Profit = $228,766
TOTAL ANNUAL PROFIT INCREASE = ($539,676 + 228,766) = $768,442TOTAL ANNUAL PROFIT INCREASE = ($539,676 + 228,766) = $768,442
28
Thank you for your time and attentionThank you for your time and attention
Questions / Comments?Questions / Comments?