Professional Documents
Culture Documents
ABSTRACT
The purpose of this paper is to report the results of a study which
examines the relationship between the structural characteristics of
loans to minority small businesses and loan performance. The study
represents an extension of research to the analysis of small business
loan performance [2,5,7, 14]. As such it provides another perspective to
understanding the performance of such loans by considering the
relationships that exist between loan characteristics and loan perfor-
mance. In particular, the finding regarding the relationship between
the amount of the loan and loan performance provides important
additional evidence relevant to a debate in the literature between
Edelstein [5, 6] and Bates and Hester [3] about the effect of the loan
size variable.
INTRODUCTION:
EVOLUTION OF RESEARCH INTO SMALL
BUSINESS LOAN PERFORMANCE
The majority of the available literature examining the relationship
between business performance and loan performance has used
financial statement analysis and ratio analysis in the context of large
publicly held firms. Unfortunately this approach is often inappropriate
for small businesses since they frequently do not have the company
history and financial data required for such analysis. Furthermore, in
cases where these statements do exist they tend to fluctuate widely
from year-to-year because of the tenuous nature of the enterprise and
they may also be unaudited.
This had led to the development of alternative approaches to the
study of small business loan performance. The approaches have
focused on the use of demographic, firm and loan characteristics to
predict loan performance, A study by Bates [2] focused on an
evaluation of the Small Business Administration's (SBA) loan
programs to minority entrepreneurs. He found that the demographic
characteristics of a successful minority SBA loan applicant were
higher personal income and net worth, a better credit rating and more
managerial experience.
10
American Journal of Small Business, Volume IV, Number 1, July, 1979
A more recent study by Page, et. a1. [14] incorporated both loan
applicant demographics and firm characteristics into an analysis of
the performance of loans to minority small businessmen. The findings
of that study using a different population of minority small business-
men, corroborated Bate's findings with regard to the significance of
certain demographics in explaining loan performance. In addition, the
study found that there are important and useful relationships between
loan experience and certain firm characteristics. These variables
appear to be equally as important as the demographic variables in
discriminating between successful and unsuccessful loans.
In another study using the same SBA loan data, Bates [1]
examined the relationship between loan characteristics and loan
performance. He hypothesized and found "an inverse relationship,
ceteris paribus, between loan amount and loan repayment performance
and a direct relationship between loan maturity and loan repayment
performance [3, p. 1788]."
In a 1975 study Edelstein [5] also explored the relationship between
loan performance and certain demographic, firm and loan characteris-
tic variables in the context of a particular minority-oriented loan
program in Philadelphia. His findings with respect to demographic
and firm characteristic variables were quite consistent with those from
the Bates [1] and Page, et.a1. [14] studies. However, with regard to loan
characteristics, Edelstein found that there was a direct relationship
between loan size and performance whereas Bates found the nature of
this relationship to be inverse. This difference in findings as well as
differences in the methodologies employed has resulted in a debate
between the principal authors. [3, 6].
This paper extends the findings of the Page, et. a1. study by
considering the importance of loan characteristics to loan repayment
performance using the minority loan clients of the Cleveland OMBE
affiliate as the data base. It examines loan characteristic variables not
previously considered in the context of minority small business loan
performance. It also provides additional empirical evidence as to the
nature of the relationship between loan characteristics, and in
particular loan size, and loan performance.
11
American Journal of Small Business, Volume IV, Number 1, July, 1979
RESEARCH METHODOLOGY
The results reported in this study were based on data collected from
the Growth Corporation client files. Before the data collection process
began a tenative list of loan characteristic variables was prepared
incorporating as many structural dimensions of a loan as possible.
This list was subsequently revised based on the type of information
available in the client files. Of the 68 cases where a direct loan or
guarantee was extended 60 were complete enough for use in the
analysis reported here. Obviously, since the sample was limited in size
and locale any attempts at generalizing from the results of this study
should proceed with caution.
Table 1 enumerates the fourteen loan variables initially examined
in the study. Three of the variables were subsequently eliminated from
the analysis because of a lack of variation in the variable across the
12
American Journal of Small Business, Volume IV, Number 1, July, 1979
sixty cases. The mean value of each integer variable and the modal
value for each ordinal or nominal variable is included in the table to
provide a profile of the structural characteristics of the loans
represented in the data base. In addition, the a priori expected
relationship, based on the published literature and the authors'
feelings, between each loan variable and loan performance has been
indicated in the last column of Table 1. Throughout this research the
dependent variable is loan performance. The loan could either be paid
off, current (where payments for interest and principal were both being
made on time according to the terms of the loan agreement) or
defaulted. As in prior research, loans which were either paid off or
current are considered "good" or successful loans [2] [14].
TABLE 1
LOAN CHARACTERISTICS EXAMINED IN THE RESEARCH
AND THEIR HYPOTHESIZED RELATIONSHIP TO
LOAN PERFORMANCE
(N = 60)
therefore no analysis could be made between this variable and loan performance.
13
American Journal of Small Business, Volume nl, Number 1, July, 1979
The 60 loans studied were all granted during the period from 1967
to 1974 and the data was collected from the Growth Corporation files
during 1977. Therefore, even the most recent loans that were current in
1977 and classified as "good" loans for the purpose of this study were
in force for over two years.' As can be seen from Table 1, almost 57% of
the Growth Corporation's minority small business loans have de-
faulted. This loan default rate is almost identical to the 58% found by
Edelstein for a much larger set of minority loans made by eight
commercial banks in Philadelphia during the period of 1968-70 [5, p.
39].
STATISTICAL ANALYSIS
Because of the small size of the sample relative to the number of
variables contingency table analysis was used to determine which loan
characteristic variables should be included in a multivariate analysis.
The values for each of the eleven variables were grouped into two or
three natural categories on the basis of examination of their frequency
distributions. Table 2 summarizes the results of the contingency
analysis. An examination of Table 2 indicates three financial
characteristics - size of loan, maturity, and payment plan for loan
interest - which were found to be associated with loan performance at
a significant level of .10. The contingency tables indicated that for the
Growth Corporation loan population larger, longer loans of ten
thousand dollars principal or more and 5 years length or longer with
regular monthly interest payments are less likely to default. Only one
of the three significant variables did not move in the expected direction
indicated in Table 1. The size of loan variable was found to be
positively related to loan performance which is consistent with
Edelstein's [5] finding with respect to this variable.
The contingency table results are valuable in their own right but
they do not indicate how well the set of three variables can explain the
performance of loans by discriminating between successful and
unsuccessful loans. Questions also remain about the relative impor-
tance of each variable in the above profile since many loans will not
match the ideal profile along all three dimensions. Multiple discrimi-
nant analysia- was performed to determine the explanatory power of
the three loan characteristic variables.
The MULDIS program was used to perform the multiple discrimi-
nant analysis which produced the results described here [8]. MULDIS
was chosen over other possible discriminant programs because of two
unique features that were of particular value in this research. These
were 1) use of quadratic classification procedures in the event that the
variance - covariance matrices were different [11] and 2) use of the
Lachenbruch holdout method of classification for validation [12]. In
14
American Journal of Small Business, Volume IV, Number 1, July, 1979
Estimation Results
The results for the discriminant function estimated for the three
variables identified in the preliminary analysis are shown in Table 3.
The estimated discriminant function reflects a relatively high differen-
tiating power. The size of the Wilk's lambda is .815 and converting it to
an approximate F statistic indicates that the discriminating power of
the function is significant at the .009 level. Thus the null hypothesis
that the group of paid off and current loans and the group of defaulted
loans have the same group centroids can be rejected with confidence.
Table 3 contains the standardized discriminant coefficients for the
three independent variables. The larger the absolute value of the
variable coefficient, the better that variable discriminates between the
two groups. This is because a unit change in the variable has more
effect on the discriminant score, and the more a variable effects that
score, the better it discriminates. The ordinal rankings of the relative
15
American Journal of Small Business, Volume IV, Number 1, July, 1979
TABLE 3
STANDARDIZED DISCRIMINANT COEFFICIENTS AND
F-RATIO FOR
SUCCESSFUL vs, UNSUCCESSFUL LOANS
Loan Characteristic Discriminant % Discriminating
Variable Description Coefficierit Power" Rank
Principal amount
of loan .742
Payment plan for
loan interest -.535 38.7% 2
Length of loan .130 9.4'Y" 3
Wilk's Lambda .815
F·Ratio (a,56 D. of F.) 4.243
Significance level
of F·ratio .009
" The percentage of the total discriminating power. accounted for by each variable is
calculated by summing and weighting a scaled function.
Validation Results
The crucial role of validation in financial applications of multiple
discriminant analysis has been discussed by Joy and Tollefson [11].
Valid inference about the ability of a multiple discriminant function to
explain the difference between two groups depends on the use of a
separate validation sample. The well-known V 1 type of holdout sample
16
American Journal of Small Business, Volume IV, Number 1, July, 1979
discussed by Frank et.al. [9] could not be used for validation in this
study because of the small number of cases. Another approach was
needed to get a realistic picture of the true explanatory power of the
discriminant model by avoiding the model building bias inherent in
classifying the estimation sample.
The Lachenbruch holdout method (or V-method) available in
MVLDIS represents a solution to the problem of validation for small
samples [4]. In this method each case is classified using a classification
rule formed from the other N-l cases and which is therefore
independent of the case being classified. In essence, 60 separate
holdout samples of size one are classified, each classification being
based on a classification rule computed for the other 59 cases. The
"confusion" matrix produced for validation actually represents the
combined results of 60 separate classifications rather than the results
of one classification of all 60 cases.
The Lachenbruch holdout method of classification was used to
determine the explanatory power of the group of independent variables
by empirically measuring their success in discrimination. A Box Test
of the hypothesis of the equality of the group dispersions could not be
rejected at the .05 level [8]. Therefore, linear procedures were used for
classification.
The results for the Lachenbruch classification procedure are
presented in Table 4 in the form of a classification matrix.' The values
on the main diagonal indicate the number of loans in each group
correctly classified by the discriminant function. The probability of a
loan being correctly classified was computed to be 73.3% for this
validation run.?
How good is the observed classificatory efficiency? The discrimi-
nant function must outperform some standard of comparison to be
Useful in explaining loan performance. The appropriate chance
criterion here is a proportional chance criterion, Cpro which in this
study is 50.9% correct classification [13].
17
American Journal of Small Business, Volume IV, Number 1, July, 1979
TABLE 4
VALIDATION RESULTS
Predicted Group Membership
Actual
Group Defaulted Paid-off &
Membership Loans Current Loans Total
Defaulted Loans 26 8 34
(76.5%) (23.5%) (100%)
Paid-off &
Current Loans 8 18 26
(30.8%) (69.2%) 100%)
(26+18)
Total Classificatory Efficiency • • 73.33%
60
C
Pro • ( 160)
~ -' 2 • 50.9%
.733- .509
t
.509(1-.509)
60
DISCUSSION OF RESULTS
The univariate and multivariate analyses indicate that there is a
significant relationship among principal amount, loan maturity,
interest payment plan and loan performance. Among loans that are
granted; larger, longer loans with regular interest payment plans
appear less prone to failure. Why is this the case?
The reason larger loans appear more successful is probably a result
of:
1) lending institutions tending to be more conservative when
larger sums of money are involved as suggested by Edelstein
[5, p, 44]. As a result, they do a more thorough credit analysis
on these clients and reject a larger percentage of the larger
loans requested.
18
American Journal of Small Business, Volume IV, Number 1, July, 1979
CONCLUSION
While the results reported here should be considered with caution
because of the small, localized nature of the sample some of the
findings of this study support those of prior studies while using a
different set of data on minority small business loans. Furthermore, the
finding with respect to loan size and repayment performance tends to
corraborate Edelstein's despite using a methodology advocated by
Bates.
The findings reported here should also be of particular interest to
small businessmen and loan officers of independent or bank sponsored
Small Business Investment Corporations. They provide further
understanding of the factors that can explain small business loan
performance as well as provide an additional perspective which they
can use to evaluate potential loans to small businesses. The results
19
American Journal of Small Business, Volume IV, Number 1, July, 1979
FOOTNOTES
I The fact that current loans are put into the good or successful group even though
they are not yet paid off is not felt to seriously bias the results since the Growth
Corporation has observed that for its client population, default on a loan invariably
occurs within the first year of the term of the loan.
2 Following the tradition developed in the finance literature and advocated by Bates
and Hester [3) for analyzing classification type problems multiple discriminant analysis
was used here for the multivariate analysis rather than a least squares approach as
employed by Edelstein [5).
a Because the co-variances among the three explanatory loan characteristic
variables were small it is felt the standardized discriminant coefficients accurately
represent the relative power of the three variables.
4 For this validation run the prior probabilities were set equal to the relative group
freq uencies.
5 A separate validation run on the entire sample resulted in a somewhat inflated
classification rate of 75.0%.
REFERENCES
1. Bates, Timothy M., "Financing Black Enterprise." Journal of Finance, 29, No.2:
747-761, June 1974.
2. Bates, Timothy M., "Government as Financial Intermediary for Minority Entrepre-
neurs: An Evlauation." Journal of Business 48, No.4: 541-557, Oct. 1975.
3. Bates, Timothy M. and Hester, Donald D., "Analysis of a Commercial Bank
Minority Lending Program: Comment." Journal of Finance, 32, No.5: 1783-1790,
Dec. 1977.
4. Crask, M. and Perreault, W.D., "Validation of Discriminant Analysis in Marketing
Research." Journal of Marketing Research, 14 No.1: 60-68, Feb. 1977.
5. Edelstein, Robert H., "Improving the Selection of Credit Risks: An Analysis of a
Commercial Bank Minority Lending Program." Journal of Finance, 30, No.1: 37-55,
March, 1975.
6. Edelstein, Robert H., "Improving the Selection of Credit Risks: An Analysis of a
Commercial Bank Minority Lending Program - A Rely." Journal of Finance 32,
No.5: 1790-1794, DIlC. 1977.
7. Edmister, Robert 0., "An Empirical Test of Financial Ratio Analysis for Small
Business Failure Prediction." Journal of Financial and Quantitative Analysis, 7:
1477-1493, March 1972.
8. Eisenbeis, Robert A., and Avery, Robert B., Discriminant Analysis and Classifica-
tion Procedures, Lexington, Mass: D.C. Heath and Co., 1972.
20
American Journal of Small Business, Volume IV, Number 1, July, 1979
9. Frank, Ronald E., Massy, William F., and Morrison, Donald G., "Bias in Multiple
Discriminant Analysis." Journal of Marketing Research 2: 250-258, Aug. 1965.
10. Hester, Donald D., "An Empirical Examination of a Commercial Bank Loan Officer
Function." Yale Economic Essays 24: 3-57, Spring, 1962.
11. Joy, O.M. and Tollefson, John 0., "On the Financial Applications of Discriminant
Analysis." Journal of Financial and Quantitative Analysis, 10: 723-739, Dec. 1975.
12. Lachenbruch, P.A., and Mickey, M.R., "Estimation of Error Rates in Discriminant
Analysis." Technometrics, 10: 1-11, Feb. 1968.
13. Morrison, Donald G., "On the Interpretation of Discriminant Analysis." Journal of
Marketing Research, 6, No.2 156-163, May 1969.
14. Page, Albert L., Trombetta, William L., Werner, Charles and Kulifay, Marilyn,
"Identifying Successful Versus Unsuccessful Loans Held by the'Minority Small
Business Clients of an OMBE Affiliate." Journal of Business Research 5, No.2: 139-
154, June 1977.
21