Lecture 2:

Part one: Credit scoring techniques

Reading: Sathye: Chapter Three

1
 Learning objectives
1. Explain the practical use of a credit scorecard
2. Summarise the history of credit scoring
3. Explain the growth in consumer credit
4. Discuss why retail credit scoring has become so
important
5. Explain the process of building a credit
application scorecard
6. Discuss the 4 R’s of credit scoring
7. Discuss how credit scoring might evolve in the
future 2
 Introduction
• Credit scoring uses statistical analysis to determine
the probable repayments of debts by consumers. A
score is assigned to an individual based on
information processed from a number of sources
(summarise the risk in single number)
• It has revolutionised the granting of credit generally
and consumer credit in particular
• An example of a credit application scorecard is
covered
• And also how to build a credit application scorecard

3
 Advantages of credit scoring
• Decision making process is consistent
• Mass of data can be condensed into a credit
figure
• Can be easily shown on computer files
• Little need to keep re-examining paper work
An example of a credit application scorecard

• A credit scorecard from the Home Loan Experts

website: [Link]/credit-
score-home-loan/credit-score-calculator/
• Suggested by HLE for use by prospective home loan
borrowers to generate a realistic credit score
• Similar models use 12 questions in total
• The higher the score the lower the
creditworthiness of the prospective borrower
• Refer to “Industry Insight” section for a detailed
coverage of the scorecard
5
• The key scoring factors in alternative credit
scorecards
• Bankrupt = 100 points
• Discharged bankrupt = 50 points
• Loan-to-valuation ratio greater than 95% = 50
points
• Liabilities more than assets = 40 points
• More than 6 credit enquiries = 30 points
• Missed debt repayments in last 6 months = 25
points
• Borrowing more than $1m = 25 points
• No genuine savings = 20 points
6
 Using the credit scorecard
– The worst possible borrower score = 328 points
– The best possible borrower score = -10 points
– Scores up to 35 points are classified as “low
risk”
– Scores between 35 and 55 points are “medium
risk”
– Scores above 55 points are classified as “high
risk”
– These cut-off scores can be changed by the
lender
7
 History of credit scoring
• The average adult in the US is currently being
credit scored once per week either on new or
existing accounts
• Three phases of in the history of credit
scoring:
– 1935 – 1959 Pioneers
– 1960 – 1979 The Age of Automation
– 1980 + The Age of Expansion
How does your occupation fit in with growth or
soon to be obsolete areas in the economy?
8
 How to build a credit application scorecard

• Credit application scorecards are the main type of

scorecard
• Other types:
• Behavioural scoring (Moving beyond basic credit scores
and predicting your future actions), e.g, Fair Isaac is
modelling the likelihood of taking your medication
• Collections scoring (Impact of missed invoice payments)
• Customer score (Assessing likelihood of paying invoices)
• Bureau score (Veda / Dun & Bradstreet – Formal credit
file)
9
 Growth in consumer credit
– The link between the Global Financial Crisis (GFC)
and the use of credit scoring in consumer credit
– 58% of bank lending in Australia is consumer
credit
– Probably all credit card applications in Australia
are credit scored
– Probably most housing loan applications are also
credit scored
– Credit providers in Australia (since July 2012) can
no longer send unsolicited invitations to
customers (Unless you are already a bank
customer) 10
Why has retail credit scoring become so
important?
– The choice between judgemental (possibly
biased) credit decisions versus credit scoring
– The move from relationship management to
transactional lending
– The desire of banks to reduce costs and
increase accuracy of lending decisions

11
 The overall goal
• Collect historical information on all individuals who applied
for credit over a 1 year period
• Usually start with 50 explanatory variables
• These are usually reduced to around 10 variables
• The dependent variable is the borrower’s performance in
the first year (“Good” – less than 3 months of missed
payments; “Bad” – more than 3 months of missed
payments
• Note that 3 months corresponds with APRA’s impaired
asset definition
• Unsuccessful applicants are usually included in this analysis
• Usually 1m+ applicants in a data set for a larger bank 12
 Other considerations
– Data sample
– Data validation and cleaning
– Data segmentation (by age, parental status etc)
– Development sample and validation sample
(Will removing outliers accurately reflect a
contemporary sample?)
– Reducing the number of variables in the model
– Coarse classifying characteristics (eg, religious
background)
– Regression analysis
13
Moving beyond credit application scorecards

• The 4 R’s of credit scoring

– Risk
– Response
– Retention
– Revenue
• The apparent link between insurance fraud
and credit default (which causes which?)

14
 The future of credit scoring
• Building better credit scoring models (very
dynamic over economic cycles and volatile if
rich/poor gap widens)
• Incorporating economics and market
conditions into credit scoring models
• Credit scoring models for Basel II
(incorporation of minimum cash reserves to
cover risk)

15
 Part two: Credit risk analysis

Reading: Sathye: Chapter four

16
 Learning objectives
1. Define credit risk
2. Analyse various approaches to credit risk analysis
3. Explain expert systems
4. Carry out a five Cs analysis
5. Ascertain credit risk from market-based spreads
6. Describe various econometric processes
7. Carry out a basic Altman analysis
8. Describe hybrid systems of credit risk analysis
9. Look at company data and carry out a basic
credit analysis
17
 Introduction
• The focus on credit risk has emerged over the last
10–15 years
• Two key questions:
– What is credit risk?
– How do we analyse it?
• The vast range of new credit analysis tools should
help minimise the dangers of institutional
collapses (Must include better auditing)

18
 What is credit risk?
• Subtle difference between credit risk and
default risk
• Credit risk:
• The risk of loss through the default on financial
obligations
• Default risk:
• The risk that the issuer (borrower) will not fulfill its
financial obligations to the investor/creditor in
accordance with the terms of the obligation

19
To define credit risk properly, the obligations
must be clear in the contract including:

• Defining the obligations of the

borrower
• Defining the obligations of the lender
• Defining the payment dates for interest
and principal
• Indicating the maturity date

20
• The lender is faced with three credit
scenarios with any loan:
1. The credit risk analysis occurring at the
time the loan is made
2. The assessment of the credit risk profile
during the term of the loan
3. The credit risk profile should the loan
become a problem.
– For any borrower, the credit risk often changes
over time (credit migration)

21
 How do we analyze credit risk?
• The tools used in credit risk analysis can vary
considerably in complexity
• Credit risk analysis tools can be grouped in four
main areas:
1 – Expert Systems (5 Cs, PARSER, Ratios, Industry analysis)
2 – Risk premium analysis (covering cost of risk)
3 – Econometric Methods
4 – Hybrid Systems

22
 (1)Expert systems
• Despite the grandiose title, expert systems tend to
provide relatively simple computerised support to the
decision- making process
• Often manually based and used to perform simple
financial calculations including financial ratios (quite
subjective, based on assessor’s opinion)
• Generally place much of the credit decision-making
on the individual lending officer
• Often used in smaller lending firms
23
Expert systems commonly used to
support 5 Cs analysis
1. Character: Is the borrower the type of person who would direct efforts to
repayment or avoiding obligations?
2. Capacity: Does the borrower have the capacity to enter into loan (e.g.
minor) and have they enough residual income to make repayments?
3. Cash/Capital: Does the borrower have sufficient ongoing cashflows to
support the loan repayments and how much of their own money can
they contribute?
4. Collateral: Can the assets supporting the loan be sold at a fair price in the
event of default and will it be enough to cover the loan?
5. Conditions: Where are we in the economic cycle? Industry conditions
and competitors?
24
One alternative to 5 Cs is PARSER
• Personal characteristics of borrower
• Amount required and why
• Repayment capacity
• Security
• Expedience of future profitable
opportunities
• Return from the loan

25
 (2) Risk premium analysis
• Measuring credit risk can be inferred from the risk
premium provided by the market on traded
securities or ratings agency
• Risk premium measured as
p (1  r )  1  i

where p = Probability of repayment

r = Interest rate on a corporate bond

i = Risk-free rate

26
• Probability of repayment is then:
1 i
p 
1 r

• The previous equations do not incorporate the

proportion of the loan that may be recovered in the
event of default, e, via collateral for example. As the
recoverable amount increases, the risk premium falls:

[e(1  r )x(1  p)]  [p(1  r )]  1  i

27
• An alternative way of expressing the risk
premium, pr, with recoverable assets is:
( 1  i)
pr   ( 1  i)
(e  p  pe)

• The previous equations have assumed a single period

default horizon, e.g. one year. To assess the
cumulative default probability for multi-period loans
(e.g. a three year loan) we can apply:

Cumulative Default Probability  1 - p1 x p 2 x p n

28
 (3)Econometric analysis
–Regression Analysis
–Applied in three forms:
• Regression Analysis
• Advanced Regression Analysis
• Discriminant Analysis (separation into
multiple groups)

29
 Multiple regression
• Aims to forecast into the future by identifying significant
variables that have previously had explanatory power
• Two types of data used. Firstly, the dependent variable, that is
the one the model is attempting to forecast. Secondly, the
independent (or explanatory variables) that provide
explanatory power for the dependent variable.
• Some technical econometric issues, such as variable selection,
limit the value of strictly applying the multiple regression
approach (usually where independent variables are similar)
• Which variables are relevant, e.g, upbringing & parental
attitude to debt?
30
Linear probability model (Probit)
• Variation on multiple regression analysis that divides
variables into two samples, e.g. default and non-
default, and assigning them values of 1 if the loan has
defaulted and 0 if non-default.
• The division can then be regressed to determine
significant variables as:
n
Pi  β
i 1
i X i  error

where pi = Probability of default

 = The estimate of importance of variable

Xi
31
Discriminant analysis
• This method highlights/discriminates against two groups,
e.g. financially sound or financially distressed
• Applied in lending by Altman (1968) where ratio analysis
applied to determine which firms in distress and which
were not
• Became known as Altman’s Z Score (Zeta Score) which
provided cut-off levels
• Z < 1.81 Firm likely to default (DISTRESS ZONE)
• 1.81< Z < 2.99 May or may not default (GREY ZONE)
• Z  2.99 Firm unlikely to default (SAFE ZONE)

32
Altman’s Z Score
– Altman’s Z Score given as
Z  1.2 X 1  1.4 X 2  3.3 X 3  0.6 X 4  1.0 X 5

where X1 = Working Capital / Total Assets

X2 = Retained Earnings / Total Assets

X3 = EBIT / Total Assets
X4 = Market Value Equity / Book Value of
Liabilities
X5 = Sales / Total Assets

33
 (4) Hybrid systems
• Despite the apparent rigor, the econometric
methods discussed above provide little
more than the statistical relationships
between variables
• More recent developments have drawn on
broader sources of information and theory
in assessing credit risk, e.g, behavioural
finance

34
 – Expected Default Frequency
• Drawing on option theory, the borrower’s payoff
resembles a call option and the risk premium can be
determined accordingly
• E.g, An option on a more volatile share is more
expensive to buy. By comparison, a lower rated
borrower is more risky, pays a higher interest rate,
and is more likely to default
– Mortality Models
• Examines real bond default data to establish the
proportion of bonds for a credit rating that actually
did default
Total Amount of Loans in a Credit Rating that Defaults
MMR t 
Total Amount of Loans Issued in that Credit Rating
35
 Putting it all together-Example
• Facts: Hypothetical Retailer
• $10m loan application
• Listed on ASX, share price $3.85
• ‘A’ rated with bonds trading at 7.5%, while
comparable Government bonds are trading
at 5.9%
• EBIT was $156,298,000 on sales of
$521,566,000
• Inventory currently $61,001,000

36
Item $

Total Current Assets 588,015,000

Total Noncurrent Assets 569,749,000

Total Assets 1,157,764,000

Total Current Liabilities 403,930,000
Total Noncurrent Liabilities 203,508,000
Total Liabilities 607,438,000
Shareholders’ Equity 550,326,000
Retained Earnings 278,893,000
Number of Shares on Issue 142,869,000

37
– 5 Cs:
• Character: ASX listed, world-class board
• Capacity: Verify minutes approving loan
• Collateral: First mortgage with valuation from
Registered valuer
• Conditions: Appear generally favourable
• Cash: Ratios show that
588,015,000
Liquidity : Current Ratio   1.45
403,930,000
521,566,000
Efficiency : Inv T/O   8.56
61,001,000
156,298,000
Profitability : EBIT/Sales   0.30
521,566,000
607,438,000
Leverage : Debt/Equity   1.10
550,326,000
38
 – Market-Based Premiums
• Probability of repayment

1 i 1.059
p    98.51%
1 r 1.075

• Risk Premium at 90% recovery rate

(1  i) 1.059
pr   (1  i)  - (1  0.059) 
(e  p  pe) [0.9  0.9851  0.9851(0.9)]

0.00158 or 0.16%

39
 – Altman’s Z Score
Z  1.2 X 1  1.4 X 2  3.3 X 3  0.6 X 4  1.0 X 5
Z  1.2(0.159)  1.4(0.241)  3.3(0.135)  0.6(0.906)  1.0(0.451)
Z  1.968

• Decision criteria:

• Z < 1.81 Firm likely to default (Distress Zone, not

credit worthy)
• 1.81< Z < 2.99 May or may not default (Grey Zone,
may or may not default)-should consider qualitative
factors
• Z 2.99 Firm unlikely to default (Safe Zone, credit
worthy)
• In this example Z score is 1.968,i.e Grey Zone
40
 Conclusion
Three methods give conflicting results

1) Expert system
• 5 Cs: Fair with mixed ratio results
2) Risk premium analysis
• Theoretical risk premium of 0.16% though market
premium is 1.6% (7.5%-5.9%)
3) Econometric analysis
Altman’ s Z Score of 1.968: In unclear default range