Managing Unsecured Lending Risk

Alan Greenspan:
President, Federal Reserve Board
May 1996

“… We should not forget that the basic economic

function of these regulated entities (banks) is to
take risk. If we eliminate risk taking in order to
reduce failure rates to zero, we will, by
definition, have eliminated the purpose of the
banking system.”
 Types of Lending Risk

Borrower
Customer fails
failstotopay
pay
Change in
Losing money market
Wrong Strategy prices

Processing failures and

frauds

Regulatory compliance
 The Universal Balancing Act
How can I efficiently manage resources while
meeting business and operating constraints? Losses
Profits

How can I create and re-create strategies in a very Unused

dynamic environment? Capacity

Profits
How can I achieve these benefits with minimal Attrition
change to current systems infrastructure?
 Everyday Questions
Balancing Marketing and Risk

Should I … with what Should I … at what How will I

target this message? approve? credit level? continue to
consumer? influence?

Will the Will the Will the Will the consumer…

consumer consumer consumer
hear it? apply? use it? pay as agreed?
attrite too early?
build large balances?
repeat purchase?
buy add-on services?

be profitable?
 Everyday Questions
Balancing Marketing and Risk
Net income
How will I
continue to
Costs
influence?

Portfolio size
• # accounts
VALUE • Receivables
Will the consumer…

pay as agreed?
attrite too early? Risk
build large balances?
repeat purchase?
buy add-on services?
Yield
be profitable?
Losses

Growth in each
 Everyday Questions
Balancing Marketing and Risk
Should I … with what Should I … at what How will I
target this message? approve? credit level? continue to
consumer? influence?

Will the Will the Will the Will the consumer…

consumer consumer consumer
hear it? apply? use it? pay as agreed?
attrite too early?
build large balances?
Increase value by improving repeat purchase?

decisions
buy add-on services?

be profitable?
Use BI to optimize multiple
objectives
 Business Decisions Driven by Analytics
• New segment: Senior Students of Well Known
Universities (e.g. Oxford, Cambridge, London, Leeds,
Manchester)
• No credit history / very poor record
• No steady income
• In need of credit
• Yet, about to graduate and get good, steady income
• Growth of income all through life
Developed a new product driven by analytics: ‘Students
Silver Card: BoS your friend in need, friend for life’
• Good interest rate, annual fee & adequate credit limit
• Charges decrease with change in life stage
• Credit limit keeps increasing, upgraded to Gold, Platinum
 Business Decisions Driven by Analytics

• Ratio: Previous statement balance / Current

month’s purchases
• Ratios: divide by zero problem
• Denominator recoded
• Millions of recoded records. Why?
• Good borrower suddenly stopped using the card.
Why?
 Business Decisions Driven by Analytics
• Ratio: Previous statement balance / Current month’s
purchases
• Ratios: divide by zero problem. Denominator recoded
• Millions of recoded records. Why?
• Good borrower suddenly stopped using the card. Why?
• Borrower with no intention to default in temporary
payment difficulty (illness, accident, job loss etc.)
Developed a new product: ‘Restructured Debt’
• Credit card balance converted to personal loan
• Longer term with lower charges
• Payment holiday
• Free financial planning advice
Led to much lower default on Credit Card borrowings,
happier & more loyal customers
 Decision areas
• Solicitations
• New applications
• Account management
– Credit line
– Authorization
– Collections
– Reissue
• Cross-sell
• Keep / sell
 Business Objectives
• Increase consistency of lending decisions
– Consistent & unbiased treatment of each applicant
• Customers with the same details get the same treatment
– Total management control over credit approval systems
• Allows for loosening or tightening of lending through credit cycles
• Potential increase in approvals
• Reduce operating costs
– Increase in automated processing
• Improve customer service
– Fast and consistent decisions at application point
– More appropriate limit and authorisation decisions
– Reduction in collection actions on low risk accounts
– Risk based allocation of credit limits and reissue terms
9/23/2018 12
 Business Objectives
• Improved portfolio management
– Manage credit portfolios more effectively and
dynamically
• Better prediction of credit losses
• Management ability to react to changes fast & accurately
• Ability to measure & forecast impact of policy decisions
• Quick and uniform policy implementation
– Improved Information
• Permits information gathering to assist business needs and
marketing activities
• Information gathered can be fed back into future decision
support systems’ developments, collection activities and
strategy optimization

9/23/2018 13
 The BI Solution: Models

DATA MODEL OUTCOME

 BI in the Consumer Credit Industry
• Numerous quantitative advances have emerged in the
consumer credit risk area to support business strategy
throughout the customer life cycle.
• At credit origination, analytical models are used to:
– Identify likely consumers who are likely to be profitable
– Predict propensity to respond to a credit offering
– Align consumer preferences with products
– Assess borrower credit worthiness
– Determine line/loan authorization
– Apply risk-based pricing
– Evaluate relationship value of the customer
 BI in the Consumer Credit Industry
• Throughout loan servicing, analytical methods are used to:
– Anticipate consumer behaviour (risk) or payment patterns
– Determine opportunities for cross-selling
– Assess prepayment risk
– Identify any fraudulent transactions
– Optimize customer relationship management
– Prioritize the collections effort to maximize recoveries in the event
of delinquency
• Analytical models are fast becoming the back-bone of
efficient consumer credit risk management.
• Consumer lending represents an analytically robust and
data-rich environment for credit risk and capital
measurement.
 Account Life-cycle Scoring Progression
ACCOUNT STATUS
ON TIME DELINQUENT LATE-STAGE RECOVERY
COLLECTIONS
Behavior score Bureau-based
recovery score
Custom collection score
Custom recovery
score
FICO score
Other bureau scores or
custom scores
Transaction score

TRIAD adaptive
Debt Manager-RMS RMS, BridgeLink
control system
Primary Decision: Secondary Decision: Specialty Risk Removing Credit:
Reduce Loss Risk-Related Assessment Additional Precision
 Risk Analytics in Consumer Lending
• Credit risk in consumer banking has been (traditionally)
driven by 3 C’s of lending (based on judgment):
1.Character – “willingness to re-pay debt”
2.Collateral – “incentive to re-pay debt”
3.Capacity – “ability to re-pay debt”
• The presence of a large number of consumers makes this
environment ideal for empirical modeling to predict
borrower behaviour as the basis for acquisition and
management of customers.
• Markets with robust credit bureaus further provide the
impetus to use models to predict borrower behaviour.
• Credit scores can summarize the details of credit report
and application data into a single actionable metric.
Evaluating the credit applicants: Judgment
Versus Scoring
CHARACTERISTICS JUDGMENT CREDIT SCORING

Time at present address + 12

Time at present job + 20
Residential status - 5
Debt ratio + 21
Bank reference + 28
Age N /A 15
Income - 5
• • •
•• •• ••
# of Recent inquiries - -7
% of Balance to avail. lines + 10
# of Major derogs. + 35
Overall + 212

Decision Accept Accept

Odds of repayment ? 46:1
 Basic Concept of Credit Scoring
• A statistical means of providing a quantifiable risk factor for a
given customer or applicant.
• Credit scoring is a process whereby information provided is
converted into numbers that are added together (hence it is an
example of ‘Generalized Additive Models’) to arrive at a score
(using a “Scorecard”).
• The objective is to forecast future performance from past
behaviour.
• Credit scoring developed by Fair & Isaac in early 1960s
– Widespread acceptance in the US in early 80s and UK
early 90s
– FICO scores make 75% of US Mortgage loan decisions
– Behavioural scoring, introduced later, has been accepted as
more predictive than application scoring
 National (US) Distribution of
FICO Scores
30% 28%

25%
% of Population

20% 19%

16%
15%
12%
11%
10%
8%

5%
5%

1%
0%
Up to 499 500-549 550-599 600-649 650-699 700-749 750-799 800+

FICO Score Range

Bad Rates of Major FICO Score
Ranges

Score Range Bad %

300-500 48%
500-599 30%
600-699 11%
700-850 1.50%
 Risk Scorecard - Example
Own Rent Other NI
Residence 25 15 10 17
Years at <.5 .5-2.49 2.5-6.49 6.5-10.49 >10.49 NI
address 12 10 15 19 23 14
Prof Semi-Prf Mgr Offc. Bl.Col Retired Other NI
Occupation 50 44 31 28 25 31 22 27
<.5 .5-1.49 1.5-2.49 2.5-5.49 5.5-12.49 >= 12.5 Retired NI
Years on job
2 8 19 25 30 39 43 20
Dept St / None Dept-St Maj-CC Both Several NI
Major CC 0 11 16 27 30 12
Bank Chq A/c Sav A/c Chq&Sav Other NI
reference 5 10 20 11 9
Debt ratio <15 15-25 26-35 36-49 50+ NI
22 15 12 5 0 13
No. of recent 0 1 2 3 4 5-9 No Rcrd
inquiries 3 11 3 -7 -7 -20 0
Years in file <.5 1-2 3-4 5-7 8+
0 5 15 30 40
# of Loan 0 1-2 3-5 6+
Accounts 5 12 8 -4
% Credit line 0-15% 16-30% 31-40% 41-50% >50%
utilization 15 5 -3 -10 -18
Worst No Rcrd Maj. Def. Unsatisf 1 Satisf 2 Satisf 3 Satisf
reference 0 -59 -14 17 24 29
 Sources of information
CREDIT BUREAU
REPORTS
• Credit reports
CREDIT
APPLICATION
• Application data
• Public records
• Prior experience
• Demographics
• Billing file
• Deal terms
 Application Scoring

• Application scoring is a statistical means of

assessing risk at the point of application for credit
– The application is scored once
• Application scoring is used for:
– Credit risk determination
– Loan / Credit card application approval
– Loan amount / Credit limit setting

Credit
Decision
 Behavioural Scoring
• Behavioural scoring is a statistical means of
assessing risk for existing customers through
internal behavioural data
– Customers/accounts scored repeatedly
• Behaviour scoring is used for:
– Authorisations
– Limit increase/overdraft applications
– Renewals/reviews
– Collection strategies
Debit $1344. 12
Debit
Debit $1344.
$234. 01 12
Debit $234. 01 12
$1344.
Debit $987.56
Debit
Risk
Debit $987.56
Debit $234. 01
Debit $6543.22 Grading
Debit
Debit$32423.11$987.56
$6543.22
Debit
Debit$32423.11
Debit $6543.22
Total $2556.00
Debit
Total $32423.11
$2556.00
Total $2556.00
 The Objective
• The objective of a scorecard is to use characteristics that
discriminate between Good and Bad accounts with sufficiently
high accuracy.
• The score is a measure of the probability of being a Good or
Bad performer.
• If the scorecard is a good one then the mean score of ‘Bads’ is
lower than the mean score of the ‘Goods’.
 Good/Bad Odds (probability)

• A scoring system does not individually identify a good

performer from a bad performer, it classifies an
applicant in a particular “Good/Bad odds” group.
• An applicant belonging to a 200 to 1 group, appears
pretty safe and profitable.
• If the applicant belongs to a 4 to 1 risk group, we would
no doubt find the risk unacceptable.
• There is a “cut-off” point where it is not profitable for
the bank to accept a certain Good to Bad ratio
• Based on the above, it is accepted that there will be
some “bads” above the cut-off level set, and some
“goods” below the cut-off level set.
Credit score = odds (risk)

SCORE ODDS

220
•
•
= 64 / 1
•
• •
•

180
•
= 16 / 1
•
• •
• •

140
•
= 4/1
• •
• •
•

100 = 1/1
 Scores are Calibrated (Aligned) to Odds
(on a Log Scale)

-1
Ln(Odds)

Actual Data
-2
Std Line

-3

-4

-5
505 515 525 535 545 555 565 575 585 595
Score

For any given score the probability of ‘Bad’ can be found

using the equation of the Log-odds (straight) line.
 How Is a Credit Scoring Model
Developed ?
• Analysis of a large set of consumers (>= 1
Million)
• Identification of common variables that
define behavior
• Statistical models are then built that assign
weights to each variable
• Adding all variables combines to make an
individual score
 Scorecard Construction
Outsourced
Data Gathering
•External Data Source
•Population Identification •Scorecard Vendor
•Data Availability
•Data Extraction
•Sampling Generic Scorecard

Statistical Analysis Validation

•Characteristic Analysis
•Characteristic Selection
•Multivariate model build Set cut-off Score
•Reject Inference
Implementation
Customised Scorecard
Scorecard Monitoring
 Population Identification
• The entire portfolio population is usually too
heterogeneous for a single model
• The population is segmented either based on
some broad criteria (e.g. based on age groups:
18-25, 25-45, 50-65, >65) or empirically, to
segments with widely different good / bad odds
• Scorecards need to be built for each segment
• Availability of data (both credit bureau data and
internal data) needs to be considered
 Observation / Performance Timeline

Observation Performance
Snapshot Snapshot PRESENT

PAST FUTURE

Observation Performance
Period Period
 Good-bad Definition
(example)
• The accounts are classed as ‘good’ / ‘bad’ /
‘indeterminate’ based on performance during the
performance period. For example:
• Bad = bankrupt or 3 or more payments missed
within 9 months
• Indeterminate = ever 2 payments missed or
always inactive or very low balance during the 9
month performance period
• Good = always up-to-date payments or ever 1
payment missed
 Good-bad evolution

Good-Bad Evolution

120.00%
100.00%
Cum. Bad Rate

80.00%
69% of accounts that go
60.00% bad in 18 months can be
identified after 9 months
40.00%
20.00%
0.00%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
Months since Observation
 Characteristic Selection
• Not all predictive characteristics are used in the model.
– An inter-correlation effect may exist between variables.
– For example, age may be correlated with time at current employment
and therefore only one is necessary in the model.
• Some credit bureau characteristics and some internal ones are
selected based on their ‘marginal contribution’ to the outcome, and
‘monotonicity’ of the odds:
12.00 70

60
10.00

50
8.00
Good-Bad Odds

40

Scores
G/B Odds
6.00
Scores
30

4.00
20

2.00
10

0.00 0
1-50 50-70 70-80 80-85 85-90 90-95 95-97 97-100 100-101 Other
Bands

Ratio of current balance to maximum lifetime balance (all credit cards)

9/23/2018 37
 Model Build

• Once the characteristics have been selected a

statistical model can be developed.
• Multivariate statistical methods include
– Linear Regression
– Logistic Regression
– Heuristics like Decision Tree / Neural Network

9/23/2018 38
 Model Build
 The model is built on dichotomous data. In this case a 1 for “Good”
customers and a 0 for “Bad” customers.

0.8

0.6

0.4

0.2

0
0 200 400 600 800 1000
 Logistic Regression
 The logistic regression fits the probability better than Linear
regression.

0.8

0.6

0.4

Good/Bad Probability
0.2 Logistic
Linear (Good/Bad Probability)

0
0 200 400 600 800 1000
 Models

• Logistic Regression has the following form:

exp   x 
k
 p 
ln     j 0  j x j
k j 0 j j

1  exp   x 
p
1 p 
k
j 0 j j
 Reject Inference and Validation
• Reject Inference
– Reject Inference is necessary for application scorecards
because there is no performance information for the
rejected applications
• Applications that are rejected should be included in the final
model
– Behavioural scorecards deal only in existing customers,
therefore do not require reject inference.
• Validation
– A randomly selected control group (hold out sample) or
proxy portfolio is used to test the model.
 Comparison of Scoring Systems

Scoring Cutoff

% APPLICANTS
System A
Bads Goods

10%
5%

Scoring Cutoff
% APPLICANTS

System B
Bads Goods

20%
5%
SCORE
 Measures of Discrimination
Scorecard performance can be judged on the level of discrimination
• The measure that can be used are:
Gini (or ROC) – the area between Lorenz Curve and random line

Lorenz Curve
1

0.9

0.8

0.7
Cumulative Bads

0.6

0.5

0.4

0.3

0.2

0.1

0
0 0.2 0.4 0.6 0.8 1
10%
Cumulative Goods
Gini = 62%
 Advantages of Scoring
– Defines degree of credit risk for
each applicant
– Ranks risk relative to other applicants
– Allows decisions based on degree of risk
– Enables tracking of performance
over time
– Permits known and measurable adjustments
– Permits decision automation

but BI in Consumer Lending goes much beyond ...