You are on page 1of 6

The Profiles of Late-Paying Consumer Loan Borrowers: An Exploratory Study: Note

Author(s): James S. Ang, Jess H. Chua and Clinton H. Bowling


Source: Journal of Money, Credit and Banking, Vol. 11, No. 2 (May, 1979), pp. 222-226
Published by: Ohio State University Press
Stable URL: http://www.jstor.org/stable/1991836 .
Accessed: 15/01/2015 01:11

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

Ohio State University Press is collaborating with JSTOR to digitize, preserve and extend access to Journal of
Money, Credit and Banking.

http://www.jstor.org

This content downloaded from 128.235.251.160 on Thu, 15 Jan 2015 01:11:14 AM


All use subject to JSTOR Terms and Conditions
The Profilesof Late-PayingConsumerLoan
Borrowers:An ExploratoryStudy
A Note by JamesS. Ang, Jess H. Chua,and
ClintonH. Bowling

1. Introdaction
Empiricallyderivedcredit-scoringmodels have been extensively examinedin the
literature,e.g . [ 1]. The attentionreceivedby the models will probablyincreasesince
the EqualCreditOpportunityAct (ECOA) recognizesthese credit-scoringmodels as
a basis for nondiscriminatoryloan grantingdecisions, but stipulatesthatthe models
must be "demonstrablyand statistically sound."
One commonly acknowledgedproblemwith existing credit-scoringmodels is the
treatmentof late-payingloans. Some models have simply ignoredlate-payingloans
and excluded them from the empirical samples. Others have arbitrarilyclassified
late-payingloans as good loans. Therefore,they have eitherfailed to accountfor the
existence of late-payingloans or failed to recognize the considerablecosts involved
in late-paymentnotices, follow-ups, and collections. Since late-paying loans are
more commonly encounteredthan outrightdefaults, it might be arguedthat exist-
ing credit-scoringmodels have misplacedtheir emphasisas far as creditevaluation
is concemed.
Credit-scoringmodels that can also discriminatebetween late-payingand good
loans are needed. However, before we plunge into the process of constructing
credit-scoringsystems, it is worthwhileto see if statisticallysignificantdifferences
can be found to differentiatelate-payingloans. This is the purpose of the present
study. The resultsought is an understandingof the profilesof late-payingborrowers
ratherthan a predictive model. It is hoped that such an understandingwill help
providea list of the relevantvanables thatmustbe includedin a predictivemodel. It
may also indicate ie type of predictive model at would be most appropriate.

2. Data
A random sample of 180 late-paying loans was drawn from the file of three
branches of a major California bank. They are all consumer automobile loans

JAMESS. ANG is professoroffinunce,OklahomaStateUniversity.JESS


H. CHUAis assis-
tantprofessoroffinance,Oklahoma StateUniversit. CLINTONH. BOWLINGis loanofficer,
FirstNationalBankof OklahomaCity.
0022-2879179/0579-0222$00.5010 t 1979OhioStateUniversityPress
JOURNALOFMONEY,CREDIT,ANDBANKING,vol. 11, no. 2 (May 1979)

This content downloaded from 128.235.251.160 on Thu, 15 Jan 2015 01:11:14 AM


All use subject to JSTOR Terms and Conditions
NOTES, COMMENTS,REPLIES : 223

granted in the period 1970-73. The personal characteristicsand payment rec-


ords used were constrllctedfrom computerprintouts, individual credit files, and
loan applications.
Thepurposeof the studyis to characterizelate-payingloans. Therefore,a measure
of "lateness" is necessary. Since the loans in the sample involve different total
numbersof loan payments, the numberof delinquentpayments is not an appro-
priatemeasure. Instead, the percentageof total numberof paymentsthat were late
was used.
The variables available in the data Ellesto characterizelate-payingloans were:
gross amountof loan, age, sex, maritalsttus, numberof dependents,years lived at
residence, monthlytake-homepay, monthlytake-homepay of spouse, own or rent
residence,othermonthlyincome, total monthlypaymentson all debts, type of bank
accounts, numberof credit referenceslisted, years on job, total family income per
month, debt to income ratio, total number of payments on the loan, and annual
percentageinterest on the loan.

3. Methodology
Conventional clustering techniques were not used because of several possible
complications.First, many of the variablesare categorical(maritalstatus, sex, rent
or own residence)or measuredin discreteand unequalintervals(under$400 or over
$2,000 in monthly take-homepay). Second, there may be nonlinearrelationships.
Forexample, theremaybe "threshold-type" relationships.Latepaymentsmay take
a quantumjump when certainvalues for some of the variablesare obtained.Third,
theremay be considerableinteractioneffects. Certainvariablesmay be predictorsof
late payment only in the presence of other specific variables. To allow for these
possibilities, the Automatic InteractionDetector (AID) [2] was used.

4. Results
The AID technique was applied to the 180 borrowerssampled. The results are
presentedin Figure 1. The late-payingloans were classified into fourteen groups
from the lowest average percentagepayments late (7 percent) to the highest (66
percent).The techniquefound seven variablesrelevantin characterizinglate-paying
loans . These are:residenceownership,debt/incomeratio, age of borrower,number
of payments, numberof years at residence, total monthly income, and numberof
years employed at present job. Altogether, 39.3 percent of the total variance is
explainedby these seven variables. TheF-value of 6.06 indicatedthatthe predictors
are significantat the 0.001 level. A breakdownof the portionof varianceexplained
by each variable is presented in Table 1.
The characteristicsof each final group can be obtained by tracing the relevant
branch of the tree backward. For instance, the borrowers with the lowest late
paymentrate(group1)had the following profile:A loan withover 31 payments, over
29 years of age, on the job over 5 years, and lived at the same residence for over 10
y ears. On the otherhand, the borrowerswith the highestpercentageof late payments

This content downloaded from 128.235.251.160 on Thu, 15 Jan 2015 01:11:14 AM


All use subject to JSTOR Terms and Conditions
Group 2

G roup e

Clusler 1

Fig. 1. BorrowerProfiles as Predictorsfor Repayment.(N = numberof loan;m = averagepercentof


late payment in the sample).

This content downloaded from 128.235.251.160 on Thu, 15 Jan 2015 01:11:14 AM


All use subject to JSTOR Terms and Conditions
REPLIES : 225
NOTES,COMMENTS,

TABLE 1

THE PROPORTION OF VARIATION IN THE PERCENTAGE OF LATE PAYMENT EXPLAINED BY THE pREDIcroRs
(Sample Size = 180)

Portion of
hedicts Total Variance Explained
Resldence ownership (own or rent) 0.119
Debt to income ratio 0.063
Age of boITower 0.061
Number of payments 0.058
Number of years at residence 0.036
Total monily family intome 0.032
Number of years employed at present job 0.026

r2 0.393

(group14) hadbeen on theirjobs over 2 years, lived at theirpresentresidencesover


2 years, rented their homes, and had a loan of 31 or fewer payments.
Closer examinationof the results reveals nonlinearand interactiverelationships
concerningevery predictor.We shall discuss those of age in detail because of their
implications for the ECOA. The ECOA allows age to be used as a predictorin
credit-scoringmodels as long as it is based on experience and statisticalevidence.
However, it prohibitsthe assignmentof adverse scores to applicantsover 62 years
old on the basis of their ages. This policy would be rationalonly if late-payment
and defaultrates exhibit nonlinearrelationshipswith age; i.e., within a certainage
range, age is relevant, but outside this range, it is not.
First, observe that, when loans involve 31 or less payments, age is not a relevant
variableat all. When loans involve more then 31 payments, borrowerswere par-
titionedintothose aged 30 or over andthose less than30. The olderones were shown
to have a lower average late-paymentpercentage. The younger group was further
split into 18-24 years old and 25-29 years old. Note the surprisingresult, showing
that the 25-29 age group actually had a higher frequency of late payments. More
importantly,the techniquedid not furtherpartitionthe older groupaccordingto age
again, as it did the younger group. This shows thatas long as the borrowersare 30
or over, differences in late-paymentrates are due to factors other thanage. There-
fore, at least in this sample, the ECOA stipulationsconcerning age are rational
and not restrictive.
Otherresults that have implicationsfor the ECOA are those on sex and marital
status. Neitherone was a relevantvariablefor characterizingdelinquentborrowers.
Since dataon race, color, andreligionwere notcollected, therewas no way the study
could determinetheir relevance.

5. Suggestionsfor FurtherResearch
The results of AID analysis have shown that the relationships between late
paymentsandpredictorvariablesarehighly nonlinear.The existence of interactions
cause combinationsof variables to produce irregularlate-paymentpatterns.This
implies thatlinear credit-scoringmodels will probablybe inadequatein accounting
for ie differencesin borrowerquality. We would suggest the following approach.

This content downloaded from 128.235.251.160 on Thu, 15 Jan 2015 01:11:14 AM


All use subject to JSTOR Terms and Conditions
226 : MONEY, CREDIT, AND BANKING

Since linearcredit-scoringmodels are still the most convenientto constuct, they


may still be used. However, several models will have to be used to accountfor e
nonlinearities. For example, at least two models (one for loans with over 31
paymentsand one for loans wi 31 or less payments) are indicatedby the results
here. The marginalimprovementin predictiveability of two or more linear models
over one universal model should be measured.

LITERATURECITED

1 . Orgler, Yair E . "A Credit-ScoringModel for CommercialLoans. " Journal of Money,


Credit, and Banking, 2 (November 1970), 435-45.
2. Sonquist, John A., and James N. Morgan. The Detection of InteractionEffiects.Ann
Arbor:Institutefor Social Research, 1964.

This content downloaded from 128.235.251.160 on Thu, 15 Jan 2015 01:11:14 AM


All use subject to JSTOR Terms and Conditions