Discrimination in Hire/Purchase Lending
By Retailers of Consumer Durables
In Apartheid South Africa
Douglas H. Graham,
Gerhard Coetzee, and
Published in Development Southern Africa, Vol. 13, No. 6, pp. 847-860
A double-hurdle partial observability model of hire/purchase lending is specified and estimated to
test for racial discrimination by retailers of consumer durables during apartheid. Discrimination is
defined as supplying no loans or less desirable loans to two borrowers who do not differ with
respect to creditworthiness but who do differ with respect to race. There is strong evidence of
discrimination. In particular, black households are 13 percentage points more likely to desire a
hire/purchase loan but not to have one supplied to them than are other households equivalent in
all ways except race. Although the statistical test cannot determine whether race affected lending
because lenders were bigoted or because race is correlated with unobserved characteristics
correlated with creditworthiness, increased access to formal loans for all South Africans could be
promoted by relaxing the Usury Act and by removing information from loan applications that
could reveal an applicant’s race.
The cash price is the equivalent of the principal of a loan, and the monthly rental
payments are the equivalent of the installment payments of a loan. The monthly effective interest
rate is the monthly discount rate that would make the present value of the inflow of the loan
principal followed by the outflows of the loan installments be zero. The annual effective interest
rate is the monthly effective interest rate multiplied by 12.
I. The Importance and Structure of Hire/Purchase Lending
One goal of the Reconstruction and Development Programme (RDP) is to provide access
to financial markets to all South Africans. After savings deposits, the most common formal
financial product used by black South Africans are hire/purchase agreements between households
and retailers of consumer durables such as televisions or furniture (Schreiner, Graham, and
Coetzee, 1996). These loans are common among all racial groups in South Africa.
Of all households with access to loans from formal lenders, the poorest households are
those with hire/purchase agreements. Therefore, understanding the structure of hire/purchase
lending may inform attempts under the RDP to extend the frontier of formal lending.
Even though hire/purchase agreements are legally structured as rental agreements, they are
equivalent to conventional installment loans. For example, a borrower may agree to make rental
payments of 120 Rands per month for 12 months for a wardrobe with a cash price of 1,000
Rands. Although there is no explicit interest rate, the implicit effective interest rate is about 73%
If the borrower misses a payment, then the rental contract is terminated and the lender
may repossess the wardrobe. Unless a fee is paid, the borrower loses the wardrobe and the
accumulated equity implicit in any payments made before falling into arrears.
The structure of hire/purchase loans is such that lenders are willing to supply them to
relatively poor households and that borrowers from relatively poor households are willing to
The Usury Act (No. 73 of 1968) limits the interest rate charged on loans larger than
6,000 Rands to 10 percentage points above the prime rate. There is no limit on the interest rate
that may be charged on loans smaller than 6,000 Rands. Even if interest rates exceed legal
ceilings, they are not necessarily usurious (e.g., Adams, Graham, and Von Pischke, 1984).
Making small, short loans to poor households is costly, and cost-coverage requires high interest
rates on small loans. Most of the costs of lending do not vary with the size of the loan, and thus
the average cost of lending a rand increases are loan size decreases.
demand them. On the supply side, there are at least two reasons why lenders are willing to incur
the relatively high transactions costs of supplying small, short loans. First, hire/purchase loans
generate their own collateral and are therefore relatively well-secured. If the borrower falls into
arrears, then the lender can repossess the consumer durable purchased with the loan. Second,
hire/purchase loans generate high yields. As illustrated above, the effective interest rates earned by
the lender can easily exceed the limit of imposed by the Usury Act.
On the demand side, there are at least two reasons why households accept the stringent
terms and relatively high interest rates of hire/purchase loans. First, there is a strong demand for
the consumer durables that may be financed by hire/purchase loans. Nearly every household uses
or would like to use the consumer durables that may be financed with hire/purchase loans.
Second, many households that cannot qualify for other types of loans nevertheless have cash
flows sufficient to qualify for hire/purchase loans.
Discrimination is defined as providing smaller loans and/or providing loans with more
stringent terms to borrowers who are identical with respect to creditworthiness but who differ
with respect to characteristics unrelated to creditworthiness, such as race. Even under apartheid,
there are at least two reasons to suspect that retailers might not have discriminated in
hire/purchase lending for consumer durables. First, a high proportion (14 percent) of black
households had hire/purchase debt; indeed, most hire/purchase loans went to black households
(Schreiner, Graham, and Coetzee, 1996, pp. 23-24). Second, profit-maximizing retailers of
consumer durables may have based lending decisions purely on creditworthiness.
II. Possible Policy Implications of This Investigation
This study investigates if, during the waning months of apartheid, a potential customer’s
race affected hire/purchase lending. There are policy implications associated with each of the two
possible outcomes of the investigation.
The first possible outcome is that there is not any statistical evidence that race affected
hire/purchase lending. This would imply that hire/purchase lending embodies a technology and a
market structure that reaches relatively poor households regardless of race. There are at least two
policy implications of a lack of evidence that race affects hire/purchase lending. First, it implies
that if lenders can make well-secured loans and can circumvent interest-rate ceilings and so earn
effective interest rates high enough to cover costs, then the natural forces of competition and the
market may circumvent artificial political institutions such as apartheid. Second, it implies that
profits can induce lenders to incur the costs of supplying small, short loans to relatively poor
households. Hire/purchase loans are profitable precisely because their structure skirts the Usury
Act and provides for high effective yields.
Of course, many hire/purchase loans would not be subject to the Usury Act even if they
were conventional loans because the implicit loan principal is less than 6,000 Rands. The high
effective interest rates on hire/purchase loans are not immediately obvious, and thus the structure
of hire/purchase loans not only enable them to avoid the Usury Act when applicable, but it also
shields them from the public disapproval. Relaxing or repealing the Usury Act would reduce the
stigma attached to high interest rates and thus encourage other providers of consumer credit to
This is often labeled statistical discrimination.
supply small, short loans.
In summary, no statistical relationship between race and hire/purchase lending would
imply that lenders will supply loans to the poor if it is profitable. If there is competition as well as
non-discrimination, then the profits from lending to the poor do not imply exploitation. If lenders
are competitive, the extension of formal loans to more households could be facilitated by
removing any legal limitations of social stigmas that discourage lenders from covering the costs of
extending small, short loans. This could increase the availability of credit cards or bank loans for
small consumer purchases, further increasing competition, lowering interest rates for the
borrower, and increasing the number of households with access to consumer loans.
The second possible outcome is that there is some statistical evidence that race affected
hire/purchase lending. The policy implication of this outcome depends on the specific channel
through which race affected hire/purchase lending. On the one hand, race may have affected
hire/purchase lending not because lenders were bigots but because race may be correlated with
economic characteristics which are unobserved by the lender but which are correlated with
In this case, the solution is to develop technology so that it is more profitable
for the lender to directly observe all characteristics correlated with the creditworthiness of a
potential borrower rather than taking race as a proxy for unobserved characteristics.
On the other hand, race may have affected hire/purchase lending because lenders were
bigots and, because competition was weak and the desire for profit was weaker than the
prejudice, market forces were not the remedy. In this case, the solution is more difficult.
Economic incentives clearly are not enough; if they were enough, lenders would not have
Of course, the retail outlet must somehow record the name and address of the potential
borrower, but this information need not be transmitted to the office where evaluation occurs.
discriminated in the first place. Although a bigoted lender profited from loans to black
households, the lender could have made more loans and more profits by lending to more black
There are at least two ways to combat prejudice, given that hearts and habits are not
completely subject to laws nor to economic incentives. The first way to combat prejudice is by
relaxing or repealing the Usury Act, making lending for the purchase of consumer durables more
attractive to issuers of credit cards and to banks, the competitors of hire/purchase lenders.
Eventually, competition will drive bigots out of business (Becker, 1971).
The second way to combat prejudice is by removing all traces of the race of an applicant
from the written credit application. Although race is not explicitly recorded on written
applications for hire/purchase loans, applications do include the applicant’s address, the
applicant’s name and, in many cases, the names of parents and relatives. Given the linguistic and
cultural/ethnic history of South Africa as well as the geographic implications of apartheid, a
person’s race is often easily and accurately guessed from an address or a surname. Although
written applications are accepted at retail outlets, evaluation usually occurs somewhere else.
Thus, the officer in charge of accepting or rejecting the loan application does not meet the
potential borrower and does not have firsthand knowledge of the applicant’s race.
In summary, reducing the legal limitations and the social stigma associated with high
interest rates would encourage lenders to charge interest rates high enough to make lending small
amounts to poor households profitable would eventually drive bigoted lenders out of business by
Other households encompasses coloured, Indian, and white households.
creating competition from other forms of consumer lending, such as bank loans or credit cards.
Even in the absence of discrimination, relaxing or repealing the Usury Act would also increase
competition and ultimately benefit poor households. Finally, removing obvious clues to a
borrower’s race from written applications would make being bigoted more difficult.
It turns out that there is strong evidence that hire/purchase lenders discriminated against
black households. Controlling for economic factors that affect creditworthiness (and thus supply)
and for factors that affect desire for consumer durables (and thus demand), black households are
13 percentage points more likely to desire a hire/purchase loan but not to have one than are other
Even though black households have more hire/purchase loans than do other households,
some black households without loans are just as creditworthy as some other households with
loans. The RDP could facilitate the extension of the frontier of formal loans to black households
by relaxing or repealing the Usury Act, thereby increasing the competition faced by the suppliers
of hire/purchase loans, and by removing obvious clues to a potential borrower’s race from the
written loan application.
III. Fundamental Characteristics of Loan Markets
and Their Implications For Detecting Discrimination
A model used to detect discrimination should incorporate at least five fundamental
characteristics of loan markets. First, observed debt depends on both demand by borrowers and
on supply from lenders (Maddala and Trost, 1982). To observe a hire/purchase loan, borrowers
must desire a consumer durable, be willing to borrow if they are unable to pay cash or desire not
to pay cash even if they can pay cash, and the lender must be willing to supply the loan. If the
proportion of black households with hire/purchase debt is relatively high even in the face of
discrimination because black households have an even higher demand, then a model ignoring
demand could misleadingly find no evidence of discrimination.
The second fundamental characteristic of loan markets is that, even in equilibrium, loans
are rationed (Stiglitz and Weiss, 1981). For this paper, rationing is defined as when the price and
contractual terms do not fully adjust to equalize supply and demand. The main cause of rationing
in loan markets is, of course, default; prudent lenders will not meet the demand of non-
creditworthy borrowers. In addition, asymmetric information about a potential borrower’s
creditworthiness may lead to signaling, learning, and enforcement costs so high that the potential
transaction never takes place. It also may be prohibitively costly for suppliers of hire/purchase
loans to tailor interest rates, transactions costs, and other contractual terms to the individual
circumstances of individual borrowers.
On the one hand, rationing implies that if the demand of the borrower is satisfied, then the
lender probably would have liked to have lent more. On the other hand, rationing implies that if
the lender has lent all the lender would desire, then the demand of the borrower probably is not
satisfied. Rationing is not the exception but the rule; it is highly unlikely that borrower and lender
would desire exactly the same hire/purchase loan at a given price.
The third fundamental characteristic of loan markets is that both borrower and lender have
veto power. Either may ration the other. Under demand rationing, the demander is unwilling to
borrow as much as the supplier would like to lend. Supply rationing is the converse. Both types
of rationing are not infrequent with hire/purchase loans. For example, the customers whom the
lender is likely to perceive as creditworthy are exactly those most willing and able to pay cash.
The fourth fundamental characteristic of loan markets is that rationing by either borrower
or lender may take two forms. Under loan rationing, no loan is transacted although one party
would have preferred a transaction. Under amount rationing, a loan is transacted but the loan is
smaller than one party would have liked. In the case of hire/purchase loans, there is loan rationing
because some customers pay cash even though the retailer would like to lend to them and because
some customers are denied credit even though they would like to borrow. There is also amount
rationing because, in general, the loan is not for the amount one party would have liked.
The fifth and final fundamental characteristic of loan markets is that not all creditworthy
households wish to hold debt at all times. Some households that are creditworthy and are willing
to borrow to finance consumer durables will not always hold hire/purchase debt simply because
they buy consumer durables infrequently and thus may have spells without hire/purchase debt.
The econometric model designed to detect discrimination must account for all five of these
factors. Rationing must be careful distinguished from discrimination, and rationing may originate
either with the borrower or with the lender and may take the form of vetoing the entire
transaction or a reduction in the amount of the loan. Finally, even creditworthy households usually
will not desire to always be indebted, and so lack of debt at any specific time must be
distinguished from inability to acquire debt.
There is a correspondence between supply rationing and discrimination. Discrimination is
supply rationing that is correlated with race. Supply rationing that is random or supply rationing
correlated with economic criteria is not discrimination, but rationing correlated with non-
economic criteria is discrimination.
The authors thank Simon Mpele and Dudley Horner of the South African Labour and
Development Research Unit at the University of Cape Town for providing access to the data.
For black households, non-hire/purchase debts were usually informal debts with
shopkeepers, while the most common sources of debt for other households were bank loans or
Discrimination may take two forms. The first possible form of discrimination is supply-
loan rationing correlated with race. That is, lenders may refuse to supply a loan to black
households who are otherwise equivalent to other households who are supplied with a loan.
The second possible form of discrimination is supply-amount rationing correlated with
race. That is, lenders may supply smaller loans to black households who are otherwise equivalent
to other households who are supplied with larger loans.
IV. Hypothesis and Data
The null hypothesis is that hire/purchase lenders did not discriminate against black
households. Independent variables were constructed from comprehensive demographic and
economic measurements from a nationwide random sample of 8,848 households surveyed during
August-December of 1993, the last months of apartheid (May et. al., 1995; Project For Statistics
On Living Standards and Development, 1994).
The survey included questions concerning the sources of the household’s outstanding
debts, the amount outstanding, and the amount repaid monthly. Even though the data were
obtained from households, the model includes all the variables that lenders observe.
The dependent variable was taken as the monthly payment for hire/purchase debt.
ideal dependent variable would have been the cash price of the consumer durable or the average
amount of implicit principal outstanding over a loan’s life, but the survey did not collect these
A table of the means of all independent variables for black households, other households,
all households, and all households classified by monthly expenditure quintile is available from the
data. The monthly payment for hire/purchase debt is appropriate as a dependent variable because
the matching of installment payments to a household’s cash flow is very important for both the
borrowing and lending decision.
Independent variables may affect only demand, only supply, or both demand and supply.
The first two columns of Table I classify the independent variables by inclusion in the supply
equation, in the demand equation, or in both equations.
On the demand side, some variables influence demand because they influence repayment
ability and thus creditworthiness without influencing supply because they are not observed by the
lender. In particular, the lender cannot observe the amount of monthly payments to informal
creditors, the existence of other formal debts, nor the employment status of non-applying
household members in the primary or secondary labor markets. These economic variables affect
demand but not supply.
Several demographic variables influence demand but not supply because they proxy for the
life-cycle stage of the household and thus for the desire of the household to acquire consumer
durables and to desire finance for them. These variables do not influence supply because the
lender does not care why a household wants to buy a consumer durable; the lender only cares
about ability and willingness to repay. In particular, the age of the head of the household, the size
of the household, and the recent migratory status of the household are likely to influence demand
because younger, larger, and newer households are likely to have higher demands for consumer
durables. While these demographic variables influence demand, they do not influence willingness
and ability to repay and are thus unimportant to the lender.
Some economic characteristics influence supply but do not influence demand because they
affect the ability to signal repayment capacity without necessarily affecting actual repayment
capacity. In particular, applications for hire/purchase loans gather information on the employment
status of the borrower and enough information to enable checking for past defaults with a credit
bureau. Thus the employment of the household head and home ownership, which may proxy for
having a formal credit history, are likely to affect the borrower’s ability to signal creditworthiness,
even though these variables may not affect actual creditworthiness.
Finally, some monthly expenditure influences both supply and demand because it
influences the ability to repay, which in turn influences both supply and demand. For example,
large monthly expenditures generally indicate large cash flows and thus large repayment capacity.
The location of a household’s residence affects transactions costs and thus influences both
demand and supply. In particular, it is more costly for a lender to lend to a rural household than to
an urban household because increased geographic distance increases both the level of asymmetric
information and the cost of reducing the asymmetry. Delivering the consumer durable is also more
expensive when the household is rural. The costs of borrowing also increases for the rural
household because distance increases the costs of shopping and of making payments.
The sex of the household head may influence demand because female-headed households
are usually single-parent or older households and thus are structurally different than two-parent or
younger households. Thus households headed by females may have different demands for
consumer durables than do households headed by males. Suppliers may also discriminate on the
basis of sex, although the issue is not addressed here.
Race may affect supply through discrimination, and it may affect demand through cultural
habits or other effects of the legacy of apartheid.
V. Specifying a Model to Test For Discrimination
One way to test for discrimination via supply-loan rationing correlated with race is with a
partial observability model. The model is termed partial observability because even though it may
be observed that a household does not have any monthly payment for hire/purchase debt, the
reasons for this absence are not observed. In particular, the observed absence of debt at the time
of a survey could be explained by any of the following unobserved events:
! No rationing. Neither borrower nor lender desires a transaction, or the borrower is
! Demand-loan rationing. Households are unwilling to borrow, but retailers are willing to
! Supply-loan rationing. Retailers are unwilling to lend, but households are willing to
! Purchase infrequency. The household is willing to borrow and the retailer is willing to
lend, but it happens that the household did not buy a consumer durable recently enough
that it had not completed payments by the time of the survey..
Of course, if it is observed that a household does have a monthly payment for
hire/purchase debt, then it is known that the household was willing to borrow, the retailer was
willing to lend, and the household bought a consumer durable via a hire/purchase agreement
recently enough that it had not completed payments at the time of the survey.
Thus, the first possible form of discrimination, supply-loan rationing correlated with race,
may be detected with a partial observability model where the presence or absence of a monthly
payment for hire/purchase debt is observed, but the reasons behind the absence of a monthly
payment are not observed.
The second possible form of discrimination, supply-amount rationing correlated with race,
This model is usually called a disequilibrium model (Maddala and Nelson, 1974). It is
formally equivalent, however, to a partial observability model.
also requires a partial observability model.
If a monthly payment for hire/purchase debt is
observed, then the level of the payment is also observed, but it is not observed whether this level
results from constraints on the demand side or from constraints on the supply side. That is, the
observed monthly payment cannot be greater than what the borrower is willing to pay, nor can it
be greater than what the lender is willing to accept in payment.
Thus, the observed monthly payment is the minimum of borrower demand and lender
supply. Either supply or demand is unobserved, and, for a given household, it is not known which
is observed and which unobserved. The observed amount of monthly payments could result from:
! Supply-amount rationing. Retailers are unwilling to lend as much as households demand;
! Demand-amount rationing. Households are unwilling to borrow as much as lenders
These considerations lead to the specification of a double-hurdle model (Cragg, 1971)
which incorporates a partial observability model at each of the two hurdles. The first hurdle is a
hire/purchase loan occurs or not; the second hurdle is the amount of the monthly repayment. Each
hurdle is itself a partial observability model in which the observed outcome is the minimum of
unobserved supply and demand (Fair and Jaffee, 1972). It is assumed that the decision of whether
to transact or not is independent of the decision of the amount to transact, given that both sides
desire a transaction.
be a household’s unobserved demand for a hire/purchase loan, where the
household demands a loan if DL
is unity. Let DA
be the households unobserved demand for a
level of monthly payments. Define SL
analogously for supply. Let QA be the observed
This formulation is discussed in Schneider (1993) and is a version of the standard
disequilibrium model with limited dependent variables. The model of Abowd and Farber (1982)
assumes uncorrelated error terms, while Poirier (1980) allows for correlation. For the model
studied here, the Poirier model and the Abowd-Farber model did not differ significantly on the
basis of a likelihood ratio test.
1 if â
1 if â
QL' min (DL
Prob(QL' 1)' Prob(DL
' 1) Prob(SL
amount of monthly payment. QL, the observed existence of a hire/purchase loan, is unity if QA is
positive and zero otherwise.
, with I=D, S, are vectors of independent variables influencing supply or demand. X
includes the regressors common to both demand and supply and the regressors unique to demand.
includes the regressors common to both demand and supply and the regressors unique to
supply. Identification requires that X
. The elements of X
are listed in section IV
above and in the first two columns of Table I.
Finally, assume that supply and demand have random elements that are independent and
normally distributed and represented by å
, and SA
In the first hurdle, there is partial observability because QL is observed to be zero if either
are negative, but QL is observed to be unity only if both DL
In words, the probability of observing a monthly payment for hire/purchase debt is the
product of the probability that the borrower demanded a loan recently enough not to have
' 1) Pr(SL
' 1) * Black household] ! [Pr(DL
' 1) Pr(SL
' 1) * Other household
completed payments at the time of the survey and the probability that the lender was willing to
supply a loan.
For the first hurdle, the probability of supply-loan rationing is Prob(DL
that is, the probability that the household demands a loan and that the lender is unwilling to supply
one. Discrimination is estimated as the influence of race on the supply-loan rationing. This is
simply the difference between the probability of supply-loan rationing for a black household and
the probability of supply-loan rationing for an other household, all else held equal:
In the second hurdle, there is partial observability because the observed monthly payment
is the minimum of the payment demanded and the payment supplied. The second hurdle is relevant
only when the first hurdle has been passed, that is, when households have hire/purchase debt and
thus QA is positive:
QA' min (DA
) * Black household] ! [min (DA
) * Other household].
In words, the observed monthly payment is the minimum of the payment demanded and
the payment supplied. For the second hurdle, any supply-amount rationing is estimated as the
maximum of zero and the difference between the estimated SA
and the estimated DA
Discrimination is estimated as the influence of race on the supply-amount rationing. This is simply
the difference between the amount of supply-loan rationing for a black household and the amount
of supply-loan rationing for an other household, all else held equal:
This specification is believed to be unique in the literature in that, using cross-section
survey data, it considers both supply and demand, the disequilibrium nature of loan markets, the
existence of rationing, and the distinction between the loan decision and the amount decision. The
models of racial discrimination in lending of Munnell et. al. (1996), Leece (1995), Duca (1993),
Avery (1981), and Maddala and Trost (1982) use different data or omit one or more of these
At each hurdle, there are two ways to consider the significance of discrimination. First,
discrimination is statistically significant if race has a statistically significant effect on the estimated
probability of supply-loan rationing or on the estimated amount of loan-amount rationing. Second,
discrimination is economically significant if race has a meaningfully large estimated effect on
It is not the estimated coefficient associated with race itself that matters, but rather the
estimated effect of race on supply-loan rationing and supply-amount rationing. These effects are
not equal to the estimated coefficient itself.
The asymptotic standard errors of the changes in the probabilities were calculated by the
delta method (Greene, 1993, pp. 297, 645). The distribution of the ratio of the average estimated
effect to its average standard error is unknown and so statements about significance are not
supply-loan rationing or on supply-amount rationing.
VI. Results and Conclusions
The first-hurdle partial observability model was estimated as a Bivariate Probit with
LIMDEP (Greene, 1991, pp. 463-472). The estimated coefficients and measures of statistical
significance appear in Table I. Most coefficients, in particular those on race and expenditure, are
highly statistically significant, and all statistically significant coefficients have the expected sign.
Table I also contains the estimated effects of each regressor on the probability of supply-
loan rationing. The average effect of race on the probability of supply-loan rationing over all
observations in the data set is 13 percentage points. The size of this effect suggests that
discrimination is economically significant. The average standard error of this estimated effect is 7
percentage points. The ratio of the estimated effect of race on the probability of supply-loan
rationing to its estimated standard error suggests that discrimination is also statistically
The estimated effects of other important regressors on supply-loan rationing have
plausible signs and also seem statistically and economic significant. In particular, employment of
the head of the household reduces the probability of supply-loan rationing by 3 percentage points,
and changing from the first expenditure quintile to the fourth changes the change in probability of
supply-loan rationing (relative to that of the first quintile) from 23 percentage points to 6
Results that are not presented here but which are available from the authors indicate that
the probability of supply-loan rationing is 75 percent for black households and 14 percent for
other households. Of the 75 percentage points for black households, 13 percentage points can be
attributed to race and 61 percentage points to economic factors such as the correlation between
poverty and race.
The second-hurdle partial observability model could not be estimated for technical
reasons. The maximum likelihood techniques of Maddala and Nelson (1974) failed because the
log-likelihood function is so highly multi-modal that different maxima were achieved for every
different set of starting values. In addition, as noted in Maddala (1983), the likelihood function for
this model can be unbounded for certain parameter values. Explorations of the parameter space
using a genetic algorithm (Dorsey and Mayer, 1995) have thus far produced unreasonable
parameter values. Research into these technical estimation problems will continue.
Black households are 13 percentage points more likely to demand a hire/purchase loan but
not to have one than are otherwise identical other households. The evidence strongly suggests
that retailers in South Africa during apartheid discriminated against blacks when supplying
hire/purchase loans. Still, the model cannot rule out the possibility that not all of this result is due
to bigotry rather than to any correlation of race with characteristics correlated with
creditworthiness but unobserved by lenders.
Whatever the motivations behind supply-loan rationing, it can be reduced by enabling
other forms of consumer credit to compete more effectively with hire/purchase loans. The best
way to do this is by relaxing or repealing the Usury Act so that small, short loans can be profitable
and non-disgraceful to those lenders that would compete with hire/purchase lenders. Small, short
loans are costly, but they are the loans demanded by those households beyond the current frontier
of formal lending in South Africa.
If bigotry accounts for at least some of the supply-loan rationing associated with race,
then formal loans may be extended to more households by removing obvious clues to a potential
borrower’s race from the written loan application for hire/purchase loans. In particular, there is no
reason why a loan evaluator would need to know an applicant’s surname or address, but these
data reveal an applicant’s race. This facilitates bigotry, even if the written application never
explicitly asks for an applicant’s race.
Abowd, John M.; and Henry S. Farber. (1982) “Job Queues and the Union Status of Workers”,
Industrial and Labor Relations Review, Vol. 35, No. 3, pp. 354-367.
Adams, Dale W; Graham, Douglas H.; and J.D. Von Pischke. (1984) Undermining Rural
Development With Cheap Credit. Boulder, Colorado: Westview Press.
Avery, Robert B. (1981) “Estimating Credit Constraints by Switching Regressions”, in Charles F.
Manski and Daniel McFadden, eds. Structural Analysis of Discrete Data With
Econometric Applications, Cambridge: MIT Press, pp. 435-472.
Becker, G.S. (1971) The Economics of Discrimination, Second Edition. Chicago: University of
Cragg, John G. (1971) “Some Statistical Models For Limited Dependent Variables With
Application to the Demand For Durable Goods”, Econometrica, Vol. 39, No. 5, pp. 829-
Dorsey, Robert E.; and Walter J. Mayer. (1995) “Genetic Algorithms for Estimation Problems
With Multiple Optima, Non-differentiability, and Other Irregular Features”, Journal of
Business and Economic Statistics, Vol. 13, No. 1, pp. 53-66.
Duca, John V. (1993) “Borrowing Constraints, Household Debt, and Racial Discrimination in
Loan Markets”, Journal of Financial Intermediation, Vol. 3, pp. 77-103.
Fair, Ray C.; and Dwight M. Jaffee. (1972) “Methods of Estimation For Markets In
Disequilibrium”, Econometrica, Vol. 40, pp. 497-514.
Greene, William H. (1991) LIMDEP, Version 6.0: User’s Manual, Bellport, N.Y.: Econometrics
_____. (1993) Econometric Analysis, Second Edition, New York: MacMillian.
Leece, David. (1995) “Rationing, Mortgage Demand, and the Impact of Financial Regulation”,
Oxford Bulletin of Economics and Statistics, Vol. 57, No. 1, pp. 43-61.
Maddala, G.S. (1983) Limited-dependent and qualitative variables in econometrics, Cambridge:
Cambridge University Press.
_____; and Forrest D. Nelson. (1974) “Maximum Likelihood Methods For Models of Markets In
Disequilibrium”, Econometrica, Vol. 42, No. 6, pp. 1013-1030.
_____; and Robert P. Trost. (1982) “On Measuring Discrimination in Loan Markets”, Housing
Finance Review, Vol. 1, No. 3, pp. 245-268.
May, Julian; Carter, Michael; and Dori Posel. (1995) “The Composition and Persistence of
Poverty In Rural South Africa: An Entitlements Approach to Poverty”, Johannesburg:
Land and Agricultural Policy Centre.
Munnell, Alicia H.; Tootell, Geoffrey M.B.; Browne, Lynn E; and James McEneaney. (1996)
“Mortgage Lending in Boston: Interpreting HMDA Data”, American Economic Review,
Vol. 86, No. 1, pp. 25-53.
Poirier, Dale J. (1980) “Partial Observability In Bivariate Probit Models”, Journal of
Econometrics, Vol. 12, pp. 209-217.
Project For Statistics On Living Standards and Development. (1994) South Africans Rich and
Poor: Baseline Household Statistics, University of Cape Town: South African Labour and
Development Research Unit.
Schneider, H. (1993) “Binary discrete choice under asymmetric restrictions”, Statistical Papers,
Vol. 34, pp. 43-57.
Stiglitz, Joseph E.; and Andrew Weiss. (1981) “Credit Rationing In Markets With Imperfect
Information”, American Economic Review, Vol. 71, pp. 393-410.
Schreiner, Mark; Graham, Douglas H.; and Gerhard Coetzee. (1996) “Household Debt in South
at the End of Apartheid: Results From a National Survey”, report submitted to the
Commission of Inquiry Into the Provision of Rural Financial Services (Strauss
Table I: Partial Observability Model For Joint Desire to Borrow/Lend:
Estimated Coefficients and Statistical Difference From Zero,
and Average Estimated Change in Probability of Supply Loan Rationing
Caused By a Change In an Independent Variable and Average Standard Error
Variable Demand Supply
Supply Loan Rationing:
Effect on Pr(D=1)Pr(S=0)
Coef. p-value Coef. p-value Ave. Effect Ave. S.E.
Common to Black (dummy) 0.96 0.00 *** -1.07 0.00 *** 0.13 0.07
Demand and Female Head (dummy) 0.28 0.04 ** -0.16 0.02 ** 0.03 0.02
Supply Rural (dummy) 0.26 0.15 0.03 0.67 0.03 0.02
1 Exp. Quintile (Poorest) (dummy) 1.91 0.06 * -4.44 0.00 *** 0.23 0.09
2 Exp. Quintile 1.09 0.02 ** -1.04 0.00 *** 0.17 0.09
3 Exp. Quintile 1.04 0.00 *** -0.64 0.00 *** 0.17 0.08
4 Exp. Quintile 0.32 0.01 *** -0.26 0.05 * 0.06 0.03
Regressors Constant -1.53 0.00 ***
Unique to Adults in Primary Labor Market 0.12 0.03 ** 0.01 0.01
Demand Adults in Secondary Labor Market 0.29 0.03 ** 0.03 0.02
Log (Monthly Informal Debt Payment) 0.03 0.00 *** 0.00 0.00
Non-hire/purchase Formal Debt (dummy) 0.10 0.31 0.01 0.01
Household Head Age$65 (dummy) -0.93 0.00 *** -0.10 0.05
Household Head 36<Age#64 (dummy) -0.47 0.00 *** -0.04 0.02
Childless (dummy) -0.10 0.27 -0.01 0.01
Adult Equivalents 0.33 0.00 *** 0.01 0.01
Migrated in past 5 years (dummy) 0.15 0.17 0.01 0.01
Regressors Constant 0.78 0.00 ***
Unique to Employed Head (dummy) 0.17 0.01 *** -0.03 0.01
Supply Non-mortgage, Non-hire/purchase Formal Debt -0.00 1.00 0.00 0.02
Log (Equity in Home) -0.02 0.03 ** 0.00 0.00
Mortgage (dummy) 0.20 0.12 -0.04 0.02
Own Home (dummy) 0.12 0.15 -0.02 0.01
Log-likelihood: -2513.2, n=6018
*** Significant at .01, ** Significant at .05, * Significant at .10