Models Credit Policy,, PDF
Models Credit Policy,, PDF
Models Credit Policy,, PDF
ABSTRACT – The aim of this paper is to present how credit scoring models can be used in
financial institutions, in this case in banks, in order to simplify credit lending.
Unlike traditional models of credit analysis, scoring models provides valuation based on numerical
score who represent clients’ possibility to fulfil their obligation. Using credit scoring models, bank can
create a numerical snapshot of consumers risk profile. One of the most important characteristic of
scoring models is objectivity where two clients with the same characteristics will have the same credit
rating.
This paper presents some of credit scoring models and the way that financial institutions use them.
Introduction
Banks are one of the most important financial institutions in the economy. Although their
function has changed, loan approval is one of the most important functions of these financial
institutions. For the most banks, loans form a half or more of their total asset and about 1/2
up to 2/3 of their incomes. Risk in the banking industry has tendency of concentration in
credit portfolio and when the bank has to deal with those financial problems, the causes
should look for in loans, primarily those which can not be paid because of bad management,
bad loan policies or unexpected economy reversal. This trend is present from the beginning
of world financial crises in 2008. Actually, the crisis has led to drastic decline in the economy
activity, increase of unemployment, inflation and drastic decline of currency values what
have resulted in huge abasement of financial system and increased risk for money
laundering and financing terrorism. Large number of banks have confronted with liquidity
problem and in the purpose to avoid this problem they have avoid procedures and
standards for money laundering presentation and accept „dirty“ money without any check.
Risk quantification is one of the main challenges in contemporary banking and finance.
Banks are trying to cope with them on different ways. One of the most frequent ways is to
assign different status to certain debtor. For example, default represents debtor status which
denotes impossibility of debtor to fulfil its contractual obligations. A probability of default –
PD denotes probability that debtor will not fulfil its obligations within one-year period. For
1
Institute of Economic Sciences, Zmaj Jovina 12, Belgrade, Serbia, e-mail: aida.hanic@ien.bg.ac.rs
Hanić, A., et al., Scoring Models of Bank Credit Policy, EA (2013, Vol. 46, No, 1-2, 12-27) 13
avoiding such situations, respectively to mitigate credit risk, banks manage with credit risk
in the way to do selection and loan approval to those clients which fulfil certain criteria. For
instance, banks are using different acts such as credit scoring as statistical derived tool which
assign numerical evaluation to each of input client characteristic and the sum of all numeric
evaluations will be compared with the set threshold. This is where the story about credit
scoring begins.
Credit scoring
(1)
2
Van Gestel, T. & Baesens, B. 2008. Credit Risk Management: Basic Concepts: Financial Risk Components,
Rating Analysis, Models, Economic and Regulatory Capital: Basic Concepts: Financial Risk Components,
Rating Analysis, Models, Economic and Regulatory Capital. Oxford University Press
14 Economic Analysis (2013, Vol. 46, No. 1-2, 12-27)
Pounders (0,012 to 0,010) are constant based on empirical experiments. Values (from X1 to
X5) are calculated as follows:
X1= Working Capital/Total Assets
X2 = Retained Earnings/Total Assets
X3 = Earnings before Interest and Taxes/ Total Assets
X4 = Market value of Equity/Book Value of Total Liabilities
X5 = Sales/Total Assets
Structure for assessment and quantification of credit rating of potential borrower
(company) is calculated based on following matrix:
Z score Zone
Z < 1,81 Distress Zone
1,81 < Z < 1,99 Grey Zone
Z > 2,99 Safe Zone
There are special Altman’s formulas for Z score depends on business of company.
Z score for private firms:
(2)
Z score Zone
Z < 1,23 Distress Zone
1,23 < Z < 2,90 Grey Zone
Z > 2,90 Safe Zone
(3)
Being first introduced as a handy tool for underwriting retail credit, such as residential
mortgages, credit cards, instalment loans, and small business credits; credit scoring is
nowadays being used to administer and follow-up default risk across the entire credit
portfolio of a financial institution covering firms, sovereigns, local authorities, project
finance, and financial institutions. Credit scoring is not only used by banks or financial
institutions. Insurance companies, telecommunication companies, companies who want to
find customers or who want to analyze their customer risk also can use credit scoring. Let us
consider a simple example by using three characteristics: residential status, age and loan
purpose. These are the values:
• A 20-year old, living with his parents, wishes to borrow the money to buy a
second-hand car. On the other hand there is a 40-year old house owner who
wishes to borrow the money for daughter's wedding. The bank analysis their
characteristics and gives the scores where a 20-year old will score 69 (14+22+33)
and a 40-year old will score 95 (36+34+25). Some lenders operate a very strict cut-
off policy. If the score is greater than or equal to the cut-off, the application is
approved, if not it is declined. From this very simple example we can say that
main advantages of scoring systems are:
• Quantification of risk as probability – instead of subjective judge of credit analyst
we have numerical score or rating of creditworthiness;
• Consistency - two clients with the same characteristics will have the same credit
rating for impartial assessment;
• Interpretability - it is possible to explain (showing the input variables of the
model) that each variable effects on increasing or decreasing the likelihood
default.
16 Economic Analysis (2013, Vol. 46, No. 1-2, 12-27)
Most familiar risk metric is often the adequacy of general and specific loan loss
provisions and the size of the general and specific loan loss reserve in relationship to the
total exposures of the bank.3 The most basic model of expected loss considers two outcomes:
default and non-default.
• In the event of non-default, the credit loss is 0.
• In the event of default, the loss is loss given default (LGD) times the current
exposure (EAD)
Credit loss distributions tend to be largely skewed as the likelihood of significant losses is
lower than the likelihood of average losses or no losses. Active loan portfolio management
embracing diversification of exposures across industries and geographic areas can reduce the
variability of losses around the mean. Unexpected loss represents the minimum loss level for
a given confidence level an alpha UL(a) is the maximum loss a bank will suffer a% of the
time.4
Default LGDxEAD PD
Expected Loss= (1-PD)x0 + PDxLGDxEAD = PDxLGDxEAD
The aim of the credit score model is to build a single aggregated risk indicator for a set of
risk factors (Bolton, 2009). Data about scoring should be safer in order to prevent
3
Credit risk management. 2013. The GARP Risk Series http://www.garp.org (20.01.2013.)
4
Credit risk management. 2013. The GARP Risk Series http://www.garp.org (20.01.2013.)
5
Credit risk management. 2013. The GARP Risk Series http://www.garp.org (20.01.2013.)
Hanić, A., et al., Scoring Models of Bank Credit Policy, EA (2013, Vol. 46, No, 1-2, 12-27) 17
6
Sajter, D. 2009. „Pregled osnovnih metoda i istraživanja poslovnih poteškoća uz predviđanje stečaja“, Osijek
18 Economic Analysis (2013, Vol. 46, No. 1-2, 12-27)
• Age of decisions - decisions must be old enough to make the analysis to have the
sense.
• Sample size - sample must be sufficiently large and diversified that the analysis
made sense.
• Costs - include the development, implementation and maintenance of the model.
Typically range from 30,000 to 80,000 Euros for customized models as generic
models do not have these costs because they are already developed but have high
fees for use.
Today in the market there are over 50 generic credit scoring systems that contain over 100
different credit scoring models. Credit scoring models have greatly facilitated work of banks.
In this context, it is important to mention scoring model advantages and disadvantages.
Credit scoring models have the following advantages (Šaralija, 2008):
• scoring models are objective and consistent,
• if they are well designed, they can eliminate discriminatory practices,
• they are relatively cheap,
• relatively simple and easy to interpret,
• institution is able to provide better service to its customers with faster approval or
rejection of the application.
Disadvantages of credit-scoring models are (Šaralija, 2008):
• they can just automate existing practice of bank loans, but there is a little work on
the elimination of bias in the process created in the past,
• models may degrade over time if the population to which the model is applied
changes in relation to the original population by which the model is designed.
(4)
I the logit specification we use the logistic distribution function to specify probabilities,
i.e. we assume that
(5)
7
Lando, D. 2004. Credit Risk Modeling: Theory and Applications
Hanić, A., et al., Scoring Models of Bank Credit Policy, EA (2013, Vol. 46, No, 1-2, 12-27) 19
The model then assumes that the outcomes of different firms are independent. In a probit
specification we use the distribution function Φ of a standard normal random variable to
transform the regression into the unit interval,
(6)
At this point, we will focus on the logit specification because of its connection with
discriminant analysis and because we have an interpretation of the β-coefficients in terms of
log-odds ratios: if two firms have covariate vectors xi and xj and we let pi and pj denote their
probabilities of default as given by a logit specification, then we have
(7)
We let yi denote the response of the ith firm and think of yi = 1 as default, then we may
express the likelihood function as
Where we define
(9)
Logistic regression is not used so much in earlier studies related to default risk is
probably due to computational limitations. With modern computers and statistical software,
the maximization is simple when the number of regressors is not too large.
The basic assumption in a discriminant analysis is that we have two populations which
are normally distributed with different means. For the purpose of default modeling we think
of one group as being the firms which will survive over a relevant period of time and the
other group as being those which will default.
Hence the logic is somewhat reversed compared with a logistic regression. In a logistic
regression, we have certain firm characteristics which influence the probability of default.
Assume that we are given a “training sample”
20 Economic Analysis (2013, Vol. 46, No. 1-2, 12-27)
(10)
(11)
(12)
and assigns the firm with characteristics x to group 0 if and to 1 if d(x) < logK,
where
. (13)
(14)
and
(15)
A second approach (see Anderson 1984 for details) for classifying a new observation x is
to use a maximum-likelihood approach. Given our sample of no defaulting firms and their
characteristics and the new observation x first compute MLEs of μ0, μ1 and Σ under the
hypothesis that the observation x is added to the sample of no defaulting firms, and then
subsequently compute MLEs under the hypothesis that x is added to the sample of
Hanić, A., et al., Scoring Models of Bank Credit Policy, EA (2013, Vol. 46, No, 1-2, 12-27) 21
defaulting firms. The decision with the highest likelihood wins. The outcome of this exercise
is not completely satisfactory in modern risk management in which we want to be able to
assign probabilities of default.
While the efficiency depends on model assumptions, a fundamental problem with the
discriminant analysis is that the assumption of normality seems unrealistic for many types of
characteristics that we observe. Furthermore, it is hard to imagine that if we had a very large
sample of firms, we would see a two-point mixture distribution of normal. In practice,
characteristics do not follow such a simple distribution. Furthermore, we cannot reasonably
make default probability estimates using the model unless we are willing to specify an
“overall” default probability rate. And, finally, the model is static and does not include the
important information on how long a firm survives with a set of characteristics. All of this
can be remedied using methods of survival analysis.
FICO skoring
Payment history
10%
10% Amount of charged
35%
The lenght of credit history
15%
New credits
30%
Writing about scoring and not to mention FICO is almost impossible. Specifically FICO
score is still the most complete and the most common scoring for individuals, primarily in
the United States. FICO model is decomposed into the following parts:
• 35% - Payment history loan account,
• 30% - the amount of debt that a customer has with its creditors,
• 15% - length of credit history or how long a person is the credit user,
• 10% - New credit (if the customer got some credits in the preceding month was :
• 10% - types of credit scoring.
Hanić, A., et al., Scoring Models of Bank Credit Policy, EA (2013, Vol. 46, No, 1-2, 12-27) 23
The strongest accent is on the payment history (35%). This is logic because the first and
elementary thing which a creditor wants to know about the potential debtor is how he/she
services its obligations. Inside of payment history the greatest focus is on delay analysis, type
and length of delay etc. FICO inside the scoring has very diversified evaluation system of
every specific datum so that delay of 90 days will be more sanctioned than that with 60 days.
Amount of charge (30%) is next important factor. Here is a thing about total debt in the
country on any basis in regard to total income. FICO will take in consideration not only total
ability to service plus new request, but also debt frequencies, average usage of approved
request, debt ratio which is in every moment on the account. In this way, better results will
have a person with higher discharged debt on one card than a person with less discharged
debt on the 5 or 6 cards. The length of credit history (15%) is the next factor regarding the
importance. The longer is the credit history, the better will be FICO result. FICO make
distinction between those who have very long credit history, but in the last time they don’t
use loans, and those using loans for shorter time, but more frequently. New credit has
pondered 10% and in this case FICO observed new required debt and its characteristics and
bring it into connection with historical frequencies of debt. For the example, a person who
has in comparatively short period required debt on the several places is considered a risky
client. The last important factor is type of credit scoring (10%) and the crucial question is:
does the person has health credit mix by the purpose and amount. In FICO model those
information which are predictors of future behaviour in the repayment of the loans are only
used. If the score is larger, the risk is lower, i.e. if the score is lower, the risk is larger. It is
necessary to define marginal value which divides “good” consumers from “bad” ones. FICO
credit score rank is 300 to 850. 723 is the average FICO credit score and on the USA market,
the average FICO score is 678. The Score of 620 or lower means that your credit is
“subprime” what means that the credit will be repaid by the larger interest rate than by the
one it was borrowed (Baselinemag. 2013).
25,00
20,00
15,00 October
2005
apr.08
10,00
apr.12
5,00
0,00
300-499 500-549 550-599 600-649 650-699 700-749 750-799 800-850
The first two years of the recession (2008-2009) moved the scores for millions of people
into the lowest (300-499) and the highest (800-850) segments of the FICO Score range. This
flattening of the distribution curve peaked in 2009-2010 and has since slowly been reversing.
However, the latest numbers suggest two unusual patterns in this recovery. First, the
quantity of people with very low scores has continued to drop and is now well below pre-
recession volumes. In 2005, 14.6% of consumers had scores at the bottom of the score range
(300-549). In 2012, the corresponding figure is 14.2%, which is 0.4% lower. This means that
about 800,000 fewer people have such low scores today. At the highest part of the score
range (800-850), we see a similar pattern. Since 2010, the quantity of people in this range has
continued to increase instead of returning to pre-recession levels. The percentage of people
(18.6%) in this scoring segment is now 0.7% higher than it was in 2010, representing
approximately 1.4 million more people (Bankinganalystic, 2013).
Scoring models form of small and medium firms have several specificities and the most
important are:
• The combination of personal loans rate of entrepreneur and financial report of his
firm;
• Credit ability of small entrepreneur directly connected with his financial profile of
firm owner;
• Desire and ability of firm owner to pay his personal credit is collinear with ability
and desire of firm to repay business credit.
The greatest number of scoring models for firms includes financial indicators. When we
talk about huge and public firms, they have structured financial reports about their
operations, ventures and finances. In the SME situation is a little bit different. Actually, in the
SME sometimes is hard to use financial ratios because of fact that personal activity of the
owner and business activity of the firm are combined. Empirical researches Fair, Isaac and
Hanić, A., et al., Scoring Models of Bank Credit Policy, EA (2013, Vol. 46, No, 1-2, 12-27) 25
Co. Inc. show that data which are detailed investigate and take into consideration at
traditional way of evaluation don’t have to be involved in future repayment definition when
we talk about small firm. One of the reasons is that a smaller firm doesn’t have obligations of
regular reporting, and when they publish their financial reports, these don’t have to be
revised. Also business results of SME vary because only one huge order can completely
change financial picture of business for certain period (Bohaček, Z., Šarlija, N. & Benšić, M.
2008). Necessary condition for beginning of credit rating process over SME segment is
gathering and structuring data about:
1. Default of SME clients
2. Data from balance sheet and income statement – construction of financial ratio
Default variable or target variable represents key information for assessment of default
probability. Based on this variable, the key information about consumer behaviour is
acquired. Scoring process for SME segment can be presented as:
1. Making development sample
2. Detailed analyze and variables explanation
3. Transformation input variables
4. Estimation models’ parameter
5. Review of scoring model performances
6. Implementation and monitoring of scoring model
In the process of credit activation the conditions for loan approval will be automatically
checked and then the process of client request evaluation will start. As result of this
procedure the total credit limit of client will be settled, as well as maximum period of the
loan length, minimum percentage of insurance and maximum discount on the interest. Risk
manager is responsible for request resolving and making final decision. If the request has
been approved, the process will be continued with ordinary processing loan procedure;
otherwise the request will be settled at state denied. Until the moment of loan request
realization, request can be withdrawn. Principle “two pairs of eyes” (one operator entries
request and/or confirms request acceptance and automatically starts scoring process and
another evaluates scoring results and makes decision or gives recommendations) will be
gained with programmatic control. Requests which have not satisfied, scoring conditions can
be forwarded to resolving with positive recommendation. Scoring process is modelled by the
criteria defined by the loan administrator, for each type of loan. Based on these settled
criteria, scoring procedure analyzes answers to the questions given through the loan request
and the data obtained from information system and based on that passes adequate
assessment (Antegra, 2013). Ratios which are considered during the analysis are:
• A - activity
• C - cash flow
• G - growth
• L - leverage
• Q - liquidity
• P - profitability
• S - size
• - other
26 Economic Analysis (2013, Vol. 46, No. 1-2, 12-27)
Conclusion
Standardization and better risk management is one of the purposes of scoring models.
There are several reasons why banks use scoring models: first, they provide an increase in
revenue because faster and more efficient process of bank lending increases competitiveness
and allows a greater volume of sales; second, they are cost- efficient because of the
automation of the loan approval process, it reduces the need for the number of people who
work on processing the loan and finally, credit scoring models allows a better quality of
portfolio and implies lower costs of provision for credit losses. Seen from this point of view
we could say that scoring models have achieved the goal. Credit scoring models reduces the
time it takes for a decision, on whether to approve the loan or not, from previous 12 hours to
15 minutes. This clearly indicates the importance and necessity of application scoring models
in banks and also in other financial institutions which make decisions on ranking clients.
Although scoring models have some deficiencies, benefits of using scoring models have
greatly exceeded those deficiencies, so that in future we can expect the traditional method of
granting loans to be completely replaced by scoring models.
References
Van Gestel, T. & Baesens, B. 2008. Credit Risk Management:Basic Concepts: Financial Risk
Components, Rating Analysis, Models, Economic and Regulatory Capital: Basic Concepts: Financial
Risk Components, Rating Analysis, Models, Economic and Regulatory Capital. Oxford University
Press.
Bohaček, Z., Šarlija, N. & Benšić, M. 2008. Upotreba kredit skoring modela za ocjenjivanje kreditne
sposobnosti malih poduzetnika.
Bolton, C. 2009. "Logistic regression and its application in credit scoring." PhD University Pretoria.
Sajter, D. 2009. „Pregled osnovnih metoda i istraživanja poslovnih poteškoća uz predviđanje stečaja.“
Osijek.
Šaralija, N. 2008. „Upravljanje kreditnim rizicima, Ekonomski fakultet u Osijeku“.
Šehić, S. 2009. “Menadžment rizika u standardiziranim poslovnim aktivnostima: magistarski rad”,
Sarajevo.
Lando, D. 2004. Credit Risk Modeling: Theory and Applications.
Hanić, A., et al., Scoring Models of Bank Credit Policy, EA (2013, Vol. 46, No, 1-2, 12-27) 27
Credit risk management. 2013. The GARP Risk Series http://www.garp.org (20.01.2013.).
Econ. 2013. http://www.econ.upf.edu/~bozovic/master/Rizik-7.pdf (29.01.2013.).
Baselinemag. 2013. http://www.baselinemag.com/c/a/Business-Intelligence/Bad-Credit-Could-
Cost-You-A-Job-210188/ (05.02.2013.).
Bankinganalystic. 2013. http://bankinganalyticsblog.fico.com/2012/09/fico-scores-reflect-slow-
economic-recovery-.html (07.02.2013.).
Antegra. 2013. http://www.antegra.com/download/antegra_sr.pdf (10.02.2013.).
REZIME – Osnovni cilj rada jeste predstaviti kako kredit skoring modeli mogu biti upotrebljeni u
finansijskim institucijama, u ovom slučaju u slučaju u bankama, u cilju pojednostavljenja
odobravanja kredita.
Za razliku od tradicionalnih modela kreditne analize, kredit skoring modeli obezbeđuju ocenu
kreditne sposobnosti klijenta na osnovu numeričkog skora koji predstavlja verovatnoću ispunjena
obaveza klijenta. Upotrebom kredit skoring modela, banke mogu kreirati numerički prikaz rizičnosti
klijenta. Jedna od najvažnijih karakteristika skoring modela jeste objektivnost što znači da će dva
klijenta sa istim karakteristikama imati isti kreditni rejting.
Ovaj rad će predstaviti neke kredit skoring modele i način na koji ih finansijske institucije koriste.