A Theory and Test of Credit Rationing

American Economic Association
A Theory and Test of Credit Rationing

Author(s): Dwight M. Jaffee and Franco Modigliani
Source: The American Economic Review, Vol. 59, No. 5 (Dec., 1969), pp. 850-872
Published by: American Economic Association
Stable URL: http://www.jstor.org/stable/1810681
Accessed: 23-04-2018 18:59 UTC
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.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms
American Economic Association is collaborating with JSTOR to digitize, preserve and extend
access to The American Economic Review
This content downloaded from 200.130.19.157 on Mon, 23 Apr 2018 18:59:23 UTC
All use subject to http://about.jstor.org/terms
A Theory and Test of Credit Rationing
By DWIGHT M. JAFFEE AND FRANCO MODIGLIANI*
Nonprice credit rationing by commerical in rationing over time and can these
banks and other intermediaries has at- variations be accounted for?
tracted a good deal of attention in recent
years, in part because of the role assigned With respect to the second question,
to this phenomenon by the Availability empirical research has been hampered by
the almost insolvable problem of directly
Doctrine as developed by Robert Roosa
measuring credit rationing. This has led to
[9] and others in the years immediately
following World War Il1. It is by now the use of proxy variables in the form of in-
generally agreed that credit rationing, if dicators of tight money such as interest
rate levels or changes in interest rates. It
it could be shown to be empirically wide-
spread, would have inmportant implications
is difficult to obtain conclusive results with
for an assessment of the effectiveness and such variables, however, since one cannot
then really differentiate credit rationing
timeliness of monetary policy as well as
for our understanding of its modus operandi.
from other symptoms of tight money. In
this study, by contrast, we are able to
But significant disagreement still exists
whether credit rationing is consistent with derive and exhibit an operational proxy
rational bank behavior and whether it is an for credit rationing based explicitly on a
important empirical phenomenon. These theory of rational lender behavior.
two issues essentially define the goal of this The major contributions concerned with
study; namely, to provide affirmative the first question have been provided in a
answers to the following questions: series of complementary studies by Donald
Hodgman [3], Merton Miller [8], and
(1) Is it rational for commercial banks Marshall Freimer and Myron Gordon [1].2
to ration credit by means other than Although criticism of a technical nature
price? has been raised with respect to the first
(2) Can credit rationing be measured? two studies, the most recent of these works
If so, are there significant variations by Freimer and Gordon, provides a com-
plete statement of the model. Consequent-
* The authors are, respectively, assistant professor ly, it may seem surprising that even the
of economics at Princeton University and professor of
economics and industrial management at Massachusetts
rationality of credit rationing is still con-
Institute of Technology. The study was carried out in sidered a debatable issue. The source of
conjunction with the Federal Reserve-Massachusetts this paradox, we believe, is essentially that
Institute of Technoloay econometric model sponsored
by the Social Science Research Council with support
the authors have not addressed themselves
from the Board of Governors of the Federal Reserve to the relevant question. Before elaborat-
System. The authors also wish to express their thanks ing on this point, however, it is important
to other participants in. the MIT-FRB project, and to
Franklin M. Fisher, Stephen M. Goldfeld, Burton G.
to precisely define credit rationing.
Malkiel, and Gerald A. Pogue for helpful criticism and In line with the generally accepted ter-
suggestions. The views expressed in the paper are those minology, we propose to define credit ra-
of the authors alone.
I Credit rationing is discussed in this study only 2 This list is not inclusive and references to other
within the institutional framework of the commercial works can be found in the studies cited. A second aspect
banking industry. Nonetheless, large sections of the of the mechanism of credit rationing based on "cus-
paper, and particularly the theoretical model, could be tomer relationships" is developed by Hodgman [31 and
applied equally well to other financial intennediaries. Kane and Malkiel [6] and is discussed further below.
850
JAFFEE AND MODIGLIANI: CREDIT RATIONING 851
tioning as a situation in which the demand It should be clear that information on the
for commercial loans exceeds the supply of supply curve alone will generally not be
these loans at the commercial loan rate sufficient to derive implications about
quoted by the banks. Thus credit rationing credit rationing. For this reason, the de-
is an excess demand for commercial loans velopment of our theoretical model of
at the ruling commercial loan rate. In credit rationing integrates the demand for
addition, it is helpful to distinguish two loans and the determinants of the loan rate
forms of credit rationing depending on the with the supply of loans. The development
status of the commercial loan interest rate. of this model is given in Sections I and II
Equilibrium rationing is defined as credit of the paper. In Section I, analytic propo-
rationing which occurs when the loan sitions concerning rationing are derived
rate is set at its long-run equilibrium level. under alternative assumptions about mar-
Dynamic rationing is defined as credit ket competition. In Section II, these re-
rationing which may occur in the short sults are then interpreted in the light of the
run when the loan rate has not been fully institutional structure of competition in
adjusted to the long-run optimal level. the commercial bankingf industry of the
This definition serves to bring into focus United States to provide the complete
the basic challenge which must be met in theory. Section III describes how our the-
order to establish the rationality of equi- ory of credit rationing can be used to de-
librium rationing, namely: can it ever be rive an operational credit rationing proxy.
rational for the bank to limit the loan to Section IV contains empirical tests of the
less than the amount demanded by the theory as set out in Sections I and II
borrower when both the loan rate and the using the credit rationing proxy derived
loan granted are chosen optimally, that is, in Section III.
to maximize profits. The problem of dy-
namic rationing differs only in that the I. Some Analytic Propositions
commercial loan rate is set at levels con-
1. The Bank's Optimal Loan Offer Curve
sistent with short-run profit maximiza-
tion.3 In either case, three elements enter The first set of propositions to be de-
into the problem: the demand for loans, veloped are concerned only with the bank's
the supply of loans, and the determinants supply curve or offer curve for commercial
of the commercial loan rate. The short- loans. This offer curve is derived by gen-
coming of the earlier studies already cited eralizing the results obtained by Freimer
is that they concentrate on the deter- and Gordon [1] for a rectangular density
minants of the quantity supplied by lenders function of possible returns to the bank on
while neglecting the other two elements.4 commercial loan contracts. We shall then
proceed to include a demand function and
3 Previous discussions of credit rationing have studied
the two forms of rationing independently: the theoreti-
derive the implications of alternative forms
cal justification of credit rationing was directed only atof competition.
the equilibrium form; the empirical significance was To start, consider a banker facing a
considered only with respect to dynamic rationing. In
our analysis we endeavor to integrate the two because
large number of customers each wishing to
our theory and the related empirical tests stress their finance its investment projects. As in the
common origin. earlier literature, we define the outcome
4The papers by Hodgman [31 and Miller [81 and the
discussion of "weak credit rationing" in Freimer and
of the customer's projects as the firm's end
Gordon [11 essentially just omit any reference to de- of period value, denoted by x. We further
mand and the determinants of the rate. The discussion assume that the bank views x as a random
of "strong credit rationing" by Freimer and Gordon
states the question properly but then assumes the
variable and summarize in the density
answer, as shown in fn. 12. functionfi[x] the bank's subjective evalua-
852 T HE AMERICAN ECONOMIC. REVIEW
tion of the probability of different out- The first term in this expression represents
comes. In general, the density function the gross receipts of the bank if the out-
fi[x] will be affected by the size of the come x is sufficiently favorable to enable
customer's investment which, in turn, may the firm to repay the agreed amount RiLi
be expected to depend on the size of the in full. The second term denotes the re-
loan granted. For expository convenience, ceipts if the outcome of the project falls
our formal analysis here will proceed under short of the contracted amount. In this
the restrictive assumption that the size of case, we assume the bank receives the en-
the project is fixed and therefore fi[x] can tire outcome x, whatever it might be.6 The
be taken as independent of the loan size last term represents the bank's opportunity
granted. This assumption implies that the cost, where I= I+j and j is the oppor-
firm has alternative means of finance which tunity rate. The rate is for the moment
can be used to complement the bank credit. assumed constant and independent of the
It can be shown, however, that all the loan contract on the premise that the bank
major conclusions reached here can be has unlimited access to a perfect capital
generalized to the case where the size of market.7
the project is not independent of the loan The expected profit function (1) can be
granted, provided the investment oppor- further simplified by adding and subtract-
tunity of the customer is subject to de- ing
creasing (expected) returns.'
r RiLj
The expected profit of the bank from RiLi f,[x]dx
the ith customer loan, Pi, is a function of
the size of the loan made Li, the loan rate
and then integrating the second term of
ri, and the density functionfi. It is helpful that expression by parts. This yields:
to assume the existence of a sure minimum
outcome ki and a maximum possible out- (2) Pi = Pi[R;L,]
come Ki for the projects such that:
f RiL;
f [x] = O for x < k, or x > K, = (R - I)L - Fi[x]dx
Now let Ri= 1+ri be the interest rate

where
factor and thus RiLi will be the total
amount of the contract repayment (in-
F4A] =Jf,xA
terest plus principal). Then the contribu-
tion of Li to the bank's expected profits can
be written as the difference between the o In such cases of at least partial default, it is reason-
total expected repayment by the firm and able that the bank would incur some collections which
the total opportunity cost, i.e.: should be deducted from the outcome x. While this re-
finement is neglected here, it can be shown that the
character of our conclusions would not change if it
Pi = Pi[RiLi] = RSLSJ' fi[x]dx were taken into account. See Jaffee 15] and Miller [8].
+;Li A similar remark applies to the cost of making and
servicing a loan.
I The existence of well-developed markets in Federal
+ ,} xfi[x]dx -IL, Funds and Certificates of Deposits (CDs) make this
assumption reasonable in normal periods. If these
I Freimer and Gordon [1] conclude that there is a markets cease to operate in tight money periods, for
significant difference between the fixed size investment example when the Regulation Q ceiling hinders the
and variable size investment assumptions. However, issue of new CDs, then the shadow price of funds would
this conclusion is based on an inconsistent definition of have to replace the market indicator as the opportun-
opportunity cost. In Jaffee [51 the essential equivalence ity cost. This may lead to dynamic effects, and these
of these two assumptions, using a proper definition of are considered below in the discussion of dynamic
opportunity cost, is derived. See also fn. 11. rationing (Section II.3) and in the empirical tests.
is the probability that x will be less than a. PROPOSITION 1. The optimal loan offer
An optimal loan to a customer is de- curve defined by condition (3) (and
fined as the loan size which maximizes the drawn in Figure 1 has the following
bank's expected profits from that customer properties:
for a given loan interest rate.8 The bank's
(1 ff 1) Zi = O for Ri < I
offer curve for that customer is then the set
of optimal loans corresponding to alterna- (1.2) O? L, < k,/I for R, = I
tive possible loan rates. This offer curve (1.3) R,Li < K, for all Ri
can be derived from the first-order con-
(1.4) lim L, = 0
dition for the maximization of expected
Ri X- 0
profit:
Proposition (1.1) follows from the non-
negativity condition and condition (3).
(3) OPj[R L] = Ri(1 - F[R;L,]) - I = 0
aL. Since the marginal expected profit of an
additional loan (dP,/dL;) is negative for
This condition can be usefully rewritten
all positive loan sizes whenever R; <I, the
in the form:
best the bank can do under this condition
is to extend no loan at all. Similarly, when
Fi[R;L, = 1 _ I ,-j R,= I, condition (3) can hold only if
Ri 1 +r;
Fi[RjLi]=0; this implies RjL,=ILj<k,
Since the quantity Fi[RiL.] is precisely
which is the
equivalent to proposition (1.2).
probability of default, (3) admits the fol- This means that the offer curve is a vertical
lowing simple interpretation: the optimal line segment when Ri = I. In fact, in the
loan is such that the probability of default case with no uncertainty in which Fi is
is equal to the excess of the loan rate over identically zero, the offer curve is nothing
the opportunity cost, normalized by the more than this vertical line.
loan rate factor Ri= 1+ri.9 The logic of proposition (1.3) is that for
The offer curve can now be defined as any given interest rate factor Ri, the bank
the implicit solution to (3) for Li in terms
will not receive additional income from
of Ri subject to the nonnegativity con- extending loans in amounts which exceed
dition L>O. We will denote this solu-
the solution of R,Li= Ki, because Ki is the
tion by Li= Li[Ri]. From (3) we can de- maximum amount the firm could con-
duce several properties of this offer curve ceivably earn. Furthermore, loans cost the
which will be used in developing our bank the opportunity rate I, and hence
argument. all solutions must satisfy the condition of
8 The formulation in terms of expected profits as- proposition (1.3). Thus the optimal loan is
sumes, of course, a linear utility function for the bank. finite no matter how high the interest rate
This makes it unnecessary to take into account higher
order moments of the distribution of outcomes for the
offered.10 From proposition (1.3) follows
firm's projects and covariance of profits between cus- immediately proposition (1.4). It implies
tomers. The rationale for this assumption is that banks that as the loan rate grows larger and
service a large number of relatively diverse customers.
In addition, a utility function with explicit risk aversion
larger, the optimal loan does not follow
would leave unclear whether the existence of credit course; to the contrary, at least after some
rationing in the model arises from the structure of the point, the optimal loan will begin to de-
model or simply from the utility function.
9 Since the empirically observed spread between the
cline as the rate grows and will eventually
loan rate and the opportunity rate at which banks can
secure or invest funds is typically small, this condition 10 Although the maximum is finite, there may be
implies that banks should tend to assume quite modest multiple local maxima for the loan offer curve. In the
default risk. This conclusion seems to be in agreement discussion which follows, we abstract from this compli-
with bank loss experience on commercial loans. cation since it does not affect the results.
854 THE AMERICAN ECONOMIC REVIEW
approach zero as the rate grows beyond But then,

bounds. The common sense of this sur-
prising implication can be understood from dP4RiLj = RiL,fi[RiL'] (L, + R, dL
the following considerations: (i) by making
the size of the contracted repayment RiLi and by implicitly differentiating (3)
sufficiently large, default becomes virtually
certain, and hence the bank can count on Li + R1 dL 1 - FJRiL,]
becoming the owner of all the net ac- dR, Rifi[R,L i
tivities of the firm; (ii) as R, rises, the
and thus
amount Li that the bank needs to invest to
achieve this result grows smaller and dPiJRiL'i
smaller and approaches zero as Ri tends to - = L(l - Fi[RiL,J) > 0
(6) dRi
infinity."
Two other properties of the offer curve for ki < RiLi < Ki
are also worth developing.
which proves the proposition. An obvious
PROPOSITION 2. For a given interest and reasonable implication of this propo-
factor Ri, expected profits decrease sition is that the bank will obtain its
monotonically as the loan size varies maximum potential profits when allowed
from the optimal size in either direc- to charge an infinite interest rate. Of
tion. course, this only serves to emphasize that
the offer curve is defined independently of
The proof follows directly from (3),
the demand for loans and thus many
since
points on the offer locus may not prove
a Pi[RiLi] 2 feasible for the lender.
(4) (4)
OLI8t2
~ ' ~Rifi[RiLis]
2. The < 0 as a Discriminating Monopo-
Banker
i
list
for ki < RiL, < K;
The optimal loan offer curve just derived
This also shows that the solution given by
can be interpreted as the bank's supply
(3) is, indeed, a global maximum.
curve for the bank customer. It is interest-
PROPOSITION 3. Expected profits in- ing that this supply curve is "backward
crease along the offer curve for suc- bending" as shown in Figure 1. In fact,
cessively higher interest rate factors. Freimer and Gordon [1] base part of their
argument for credit rationing on this char-
To derive this result, first substitute (3)
acteristic of the supply curve alone.12 Our
into (1), which allows us to write the profit
function along the offer curve as: 12 Reference is to the Freimer and Gordon [1J case of
weak credit rationing. The terminology is misleading
R1L, because one would normally not equate a backward
(;) P,[2,Li]- r rfLi bending supply curve with rationing. Freimer and
Gordon also consider what they call "strong credit
rationing," which conceptually is closely akin to our
U A similar result holds for the variable size invest- equilibrium rationing. For this case they argue that a
mnent case. Although the optimal loan offer remains bank sets a conventional interest rate (6 percent to be
positive as the interest rate approaches infinity, the exact) and then grants loans to all customers up to the
loan size approaches a finite asymptote. In contrast, amount indicated by their respective bank offer curves.
Freimer and Gordon [1, p. 407] conclude that the opti- This analysis is inadequate for two reasons. First, the
mal loan approaches infinity as the interest rate goes to optimality of a 6 percent rate is essentially just as-
infinity, because they only consider projects with an sumed; banks do not charge rates below 6 percent
expected value exceeding the opportunity cost; but this because of convention; banks do not charge rates above
contradicts the meaning of opportunity ccost; see fn. 5. 6 percent because their numerical examples suggest
banks for the customer's business is dis-

cussed further in Section II.
The conditions under which rational
credit rationing will occur can now be seen
Di-Dj[RI)
with the aid of Figure 1. The ith firm's de-
mand curve and the bank's offer curve to
that customer are assumed to intersect at
the interest rate factor W,.13 If the bank
chooses to charge an interest factor greater
than Ti, rationing will not occur since the
kI ' Li _iRil
loan demand is less than the loan offer
at such an interest factor. In fact, if the
firm were willing to accept a larger loan
(which it is not by definition of the de-
mand curve), the bank would increase its
expected profits by providing such a loan
1 I Ri Rj I, Rt
(see proposition 2). On the other hand, if
FIGURE I the bank chooses to charge an interest
factor less than Rk, then rationing will
definition of credit rationing, however, ob-
occur since in this region the bank's opti-
viously necessitates taking into account
mal loan offer is less than the amount de-
the demand for loans by the bank cus-
manded by the customer. The bank would
tomer. Such a demand curve for the ith
only reduce its expected profits by in-
bank customer, denoted hereafter by Di
creasing the loan offer to meet the demand.
=Di [Ri1, has been drawn in Figure 1. We
Thus the question of the rationality of
assume a downward sloping demand curve,
credit rationing can be reduced to a con-
which vanishes for a sufficiently high in-
sideration of the optimal rate factor to be
terest rate factor and is finite even for a
zero interest rate (or interest factor equal
charged by the bank, and its relation to
Ri.
to unity). The boundary conditions for the
Two critical variables enter into the
demand curve are based on the charac-
bank's selection of the loan rate. First
teristics of the underlying investment proj-
there is the question of the time horizon:
ects. At a zero interest rate the firm's de-
in the long run, a rational banker would
mand for loans is bounded by the size of
select the rate which maximizes his ex-
the investment projects, while the ex-
pected profits; but other constraints may
istence of some alternative means of fi-
preclude immediate full adjustment in the
nancing suggests that at some sufficiently
short run. This leads, of course, to our
high interest rate the demand will become
distinction between the cases of equilibrium
zero. The negative slope of the demand
rationing and dynamic rationing as al-
curve can be derived from the assumption
ready defined. We start by considering
that the firm has only limited access to
only the equilibrium case since it has been
these alternative means of finance. The
implied limit on competition between 13 The existence of at least one point of intersection is
assured by the boundary conditions on the demand
that customer utility maximization occurs at lower rates. curve and proposition (1.4). Throughout the remainder
Secondly, even if the 6 percent rate is accepted as op- of the paper we shall assume that, in fact, only one such
timal, we have no way of knowing from their analysis point exists. Equivalenit results can be obtained for the
whether rationing actually occurs, because the demand case of multiple intersections, but only at the cost of
curve is never shown. more complexity in the analysis.
the center of the theoretical discussion

profitable. and
Indeed, as in the standard the-
since the short-run dynamic case is easily ory of monopoly under certainty, at the
derived from the equilibrium case. rate R' the bank would be glad to lend
The second important variable influenc- more than the customer is prepared to
ing the bank's choice of rate is the nature take.
of market competition. Because the de-
3. The Banker Must Charge Alt Customers a
gree of competition turns out to be a
Uniform Rate of Interest
critical factor in determining the existence
of rationing, it is worthwhile considering Now suppose that the banker is con-
several different regimes from a purely strained to charge all customers the same
analytic standpoint. rate though he can choose that rate freely
The first regime considered is the simple and can also decide on the size of the loan
case of a discriminating monopolist. It is to be granted each customer. We will show
assumed that the bank maximizes its ex- that under these conditions credit ration-
pected profits with respect to each custo- ing may (and very frequently will) be
mer separately and is free to charge each profitable, i.e. at the common optimal in-
customer a different interest rate. Thus, terest rate, for some customers the most
we can take the bank's solution for the ith profitable loan for the bank to supply is
customer as typical: less than the amount demanded.
To establish this proposition it is con-
PROPOSITION 4. Let R* be the rate
venient to deal first with a subsidiary
factor which maximizes the bank's
problem. Suppose that the bank faces only
expected profits when the bank is
two customers and, for the moment, rule
acting as a discriminating monopolist.
out credit rationing by requiring that the
Then:
bank must satisfy the customers' demand
(4.1) Ri 2 Ri, which implies at the chosen common rate.
(4.2) Credit rationing is not profitable The bank's expected profits under these
for a banker acting as a discrim- conditions can be written as:
inating monoplist.
The proof of this proposition is easily
(7) P = P1[RD1] + P2[RD2]
derived from Figure 1 and the properties of where R is the common rate factor charged
the offer curve. The bank will always both customers and the constraint of sat-
charge an interest factor at least as high isfying both demand functions is implicit
as Ri since; (i) expected profits increase in the notation. By differentiating this
along the offer curve for successively higher profit function with respect to the interest
interest rates (proposition 3) and (ii) factor, we can obtain the first-order con-
dition for the optimal interest factor, say
the loan Li(Ri) is feasible by the definition
of Wi. We next observe that, in view of R*.
proposition 2, for any rate equal to, or An important property of R* can be ob-
larger than Rt, the bank's profit will be tained, however, without explicit reference
higher the closer the loan is to the cor- to this first order condition. Let Rr and
responding point on the offer curve. But R* be the optimal rate factors a dis-
since the demand curve places a ceiling on criminating monopolist banker would
the feasible loan size, the bank will find charge customers 1 and 2, respectively, and
the optimal rate R, by maximizing profit assume, without loss of generality, that
along the demand curve constraint, which R* <Rg. It follows from this definition
means that credit rationing will not be that the expected profit from each cus-
JAFFEE AND MODIGLIANI: CREDIT RATIONING
tomer must be a concave function of R in (6.2) R2 ?sR* > X2 implying that

the neighborhood of R* and R*, respec- R*< R2 is possible, in which
tively. We shall go somewhat further and case, credit rationing is prof-
assume that the concavity of each ex- itable.
pected profit function holds for all R > 1.14
We can then establish: Proposition (6.1) follows directly from
PROPOSITION, 5. If R* is the common rate propositions 5 and 4. As for establish-
factor that maximizes the bank's ex- ing (6.2), we need only exhibit a concrete
pected profit, we must have: example in which the condition holds. That
it is, in fact, quite easy to construct such
RI <R <R2 examples can be seen from the following
considerations. First, going back to prop-
This result can be verified through use of osition 5, one can readily establish that
a proof by contradiction. The assumption the position of R* within the range R* to
of concave expected profit functions im- R2 depends on the relative size of the two
plies that expected profits decrease mono- customers (as measured, say, by the size
tonically as the absolute value of the of the loan demanded for aniy R in the cri-
spread between the actual rate to a cus- tical range), and on the elasticity of the
tomer and the discrimination monopolist two demand curves. In particular, R* can
rate increases. Thus, if the bank chose a be made arbitrarily close to R* by assum-
common rate factor R* that was less than ing that customer 1 is sufficiently larger
both R* and R2, it would find that ex- than customer 2, and/or by assuming that
pected profits could be increased by in- the demand curve for customer 1 is suffi-
creasing the rate factor at least to the ciently inelastic in the range of rates above
R*. By the same token, R2 can be made
level of R*, thus contradicting the assump-
tion that the original rate factor was arbitrarily close to R* by assuming a
optimal. Essentially the same argument sufficiently elastic demand curve for cus-
shows that R* > R2 also leads to a con- tomer 2. Thus, by appropriate choice of
tradiction. these functions, one can readily construct
We can now relax the restriction that situations where R* is lower than W2. The
the bank must satisfy both customers' above construction also provides an inter-
demand function and show that, under esting interpretation of proposition 6.
these conditions, it will never pay to ration The constraint of charging both customers
customer I but it may very well pay to the same rate R* forces the bank to
ration customer 2. Specifically, we can charge customer 1 a rate which is too high
establish: relative to R* and hence the customer is
not rationed. On the other hand, the bank
PROPOSITION 6. For the common rate
is forced to charge customer 2 a rate which
regime
is too low relative to R*. If the rate is
(6.1) R* > R* 2 >R implying it is not
sufficiently low relative to W2, that is R* <
profitable to ration customer
R2, then the second customer will be
1.
rationed.
14 The assumption of global concavity is not a neces-

The possibility of credit rationing also
sary condition for proposition 6, to be proven below; adds another dimension to the problem. In
the proof can be carried out with weaker assumptions. the case in which rationing does not occur,
We have chosen to assume global concavity, however,
because it leads, via proposition 5, to a particularly
R* is the optimal common rate charged by
interesting demonstration of proposition 6. the bank and the customers receive loans
of D1 [R*I and D2 [R*], respectively.

common rate factor, But
it is important to
when customer 2 can be profitably ra- recall from proposition 1 that the bank's
tioned at R*, the very existence of ration- offer curve for a risk free customer is a
ing changes the conditions of the problem vertical line at R = I. Furthermore, it must
and R* need no longer be the optimal rate. be true that R* (or R&) is greater than I if
Instead, there will exist a more general the bank's expected profits are to be posi-
optimum optimorum rate, say R, which tive. But this implies that R*> R for the
yields the maximum expected profits after risk free customer, which on the basis of
allowing for credit rationing of customer 2. proposition 6 rules out the possibility of
R will just equal R* in cases in which ra- credit rationing.
tioning is not profitable, but will generally We can now proceed to generalizations
differ from R* when rationing is profitable. of the propositions developed in this part.
The fact that Ri may differ from R* creates First, consider a barker facing n customers
some difficulties in deriving the compara- and constrained to charge all n customers
tive static properties of the model, as will the same loan rate. Again, let R1 (i=
be apparent below. However, it does not 1,2, . . , n) denote the rate the bank would
affect our basic conclusion; proposition 6 charge if acting as a discriminating monop-
remains valid even if R* is replaced with olist and let R be the common "optimum
the true optimum P. For, as one can optimorum" rate the bank charges when
readily verify, R* < W2 implies R?< W2, and allowing credit rationing. Let us number
conversely.'5 customers in ascending order with respect
Before proceeding to the generalizations to the monopolist rate; that is R? 2R-
of the propositions developed so far, it is for i= 2,3, . . ,n. Proposition 5 can then
worthwhile considering the special case of a be generalized to:
risk free firm, that is a customer for whom
PROPOSITION 5'. The optimal common
the bank's subjective evaluation of default
rate R charged all customers must lie
risk is zero.
between the rate charged customer 1
PROPOSITION 7. Neither a banker acting and the rate charged customer n when
as a discriminating monopolist, nor a the bank is acting as a discriminating
banker charging all customers the monopolist. Formally, there exists an
same rate, will ration a risk free cus- integer j, 2 <? <n, such that R1 > R >
tomer. Rj1I
The result of proposition 7 for the case Similarly, proposition 6 can be general-
of a discriminating monopolist is simply a ized to:
special case of proposition 4 and follows PROPOSITION 6'.
directly from that proposition. To prove (6.1') For any customers i such that
the proposition for a banker charging a R > R* we have iR? > R imply-
ing that rationing of these cus-
1" We have previously established that R*>R2 im- tomers is not profitable.
plies that R* is already the optimum optimorum or
R*=Rk; hence R*>R2a+*R>Rs. Similarly R?R2 im- (6.2') In the case of customers for
plies that even when rationing is allowed, it is not whom Ri<RR* we have R1<R
profitable and therefore R must coincide with R*, the implying that rationing of some
maximum subject to the constraint that rationing is not
permissible; that is f>R2R-4R =R* and therefore of these customers may be prof-
R> ?2R,-R*>R2. But these two propositions together itable.
imply R*>?7-R?>R2, which in turn implies R*<Rt
4*>gy. The interpretation and proof of these
JAkFFEE AND MODIGLIANI: CREDIDT RATIONING 859
propositions follows directly from the in a class will be positively related to the
corresponding propositions for the case of heterogeneity of the R* of the customers in
two customers, and accordingly are not that class and hence the likelihood of ra-
repeated here. tioning will be inversely related to the size
of m. Indeed, if mn is allowed to be as large
4. Generalization to m Separate Customer as the number of customers, that is, n, then
Classes
the bank will be in the position of a dis-
It is now easy to extend our conclusions criminating monopolist and credit ration-
about the rationality of credit rationing to ing will not occur.
the case in which the bank can assign cus- There remains now to draw the implica-
tomers to any one of, say, m classes, where tions of these results by combining the
within- each class the bank must charge a propositions developed so far with a num-
single uniform rate. The principles which ber of considerations arising from the na-
govern the assignment of customers to ture of competition in the banking in-
classes and the choice of the rate for each dustry.
class can be readily inferred from the
IJ. Competition in Banking and
previous analysis:
Credit Rationing
(i) If the profit functions are all concave
in the relevant range, then the optimal 1. The Nature of Competition in Com-
classification will be achieved by dividingmercial Banking and its Implications for
the entire range of the set of R* (the rateCredit Rationing.
charged customer i when the bank acts as We have shown in Section I that a single
a discriminating monopolist) into m inter- bank, free to discriminate between bor-
vals and assigning to the same class all rowers by charging each customer its
monopolist rate R*, would not ration
customers whose R* falls in a given inter-
val. If the profit functions are not concave, credit. A similar conclusion holds even if
the principle for optimal classification be- there are many banks, as long as they act
comes more complex; but in any event, it collusively to maximize joint profits, rely-
is clear there will exist a set of optimal ing if necessary on side payments. If all
group rates and a corresponding optimal banks share the identical subjective evalu-
classification for the customers, and that ations of the profitability of borrowers'
furthermore, each class will contain cus- investment projects, then clearly the op-
tomers with different R's, as long as the timum rate R* to be charged to the
number of customers exceeds the number ith customer would be the same no matter
of classes. which bank served him. Furthermore, even
(ii) The optimum rate for any given allowing for differences in the subjective
class j, say Rj, must fall somewhere be- evaluation of borrower risk and assuming
tween the smallest and the largest R* of an arbitrary initial distribution of cus-
the customers in that class. tomers between banks, the device of buy-
(iii) It will not be profitable to ration ing and selling customers would allow each
customers whose R* is smaller than the bank and the industry as a whole to maxi-
group rate Rj, but it may pay to ration mize profits. In this way a banker's
those for whom R* exceeds Rj. In par- Pareto optimum would be reached with
ticular, rationing will occur whenever each bank charging its customers the
Ri> Rj. monopolist rate, and thus, again, no credit
(iv) The likelihood that it will be rationing would occur.
profitable to ration at least some customers In this section we propose to argue that
this solution is in fact not feasible, at least The inducement to adopt a classification
in the present American economy. We scheme of the type described is likely to be
suggest, instead, that banks can best ex- greatly strengthened when we take into
ploit their market power, while remaining account the fact that banks cannot openly
within the bounds set by prevailing in- collude, although they share a common
stitutions, by classifying customers into a desire to maintain rates as close as feasible
rather small number of classes within each to the collusive optimum. In order to
of which a uniform rate is charged, even prevent, or at least minimize, competitive
though the membership of each class will underbidding of rates they would need
exhibit considerable heterogeniety in terms tacit agreement as to the appropriate rate
of R:. structure for customers, and thus a classi-
First, even if there were but a single fication scheme based on readily verifiable
monopoly bank or a perfectly collusive objective criteria would appear as an
banking system, the mere existence of efficient and effective device. Furthermore
usury laws would lead toward the indi- to make the whole arrangement manage-
cated solution. Such laws would prevent able, the number of different rate classes
the banker from charging any rate R* would have to be reasonably small. Finally,
which is greater than the legal limit. Thus one can also readily understand how
all customers for whom R* is larger than such tacit agreement on the structure
the ceiling would be classified together in of class rates could be facilitated by tying
the category with the ceiling rate. Since these rates through fairly rigid differ-
the monopolist rate Ri for each customer entials, to a prime rate set through price
in this class would equal or exceed the uni- leadership.
form rate set for the class, namely the If we now superimpose the impact of
usury ceiling, it is apparent from the re- usury ceilings along with the other legal
sults of Section I that many, if not all, and social constraints, it is clear that the
customers in this class would be profitably entire structure of rates would tend to be
rationed. compressed within narrower limits than
Even aside from usury ceilings, the would otherwise be optimal. This means,
pressure of legal restrictions and considera- in particular, that the rate for each class
tions of good will and social mores would would tend to the lower limit of the R*
make it inadvisable if not impossible for spread appropriate for the customers in
the banker to charge widely different rates that class, with the possible exception of
to different customers. A banker would the lowest class rate reserved for the risk-
tend, instead, to limit the spread between less or nearly riskless prime customers.
the rates and to justify the remaining dif- The result is that widespread rationing
ferentials in terms of a few objective and would occur, particularly in the higher
verifiable criteria such as industry class, rate classes.
asset size, and other standard financial Finally, we may observe that the oli-
measures. An effort would no doubt be gopolistic price setting pattern outlined
made to choose the criteria for classifica- above is likely to lead to a very sluggish
tion so as to minimize the difference be- and somewhat jerky adjustment of the
tween the optimal classification of cus- entire rate structure as changes in under-
tomers into rate classes and the categories lying conditions generate changes in the
dictated by the objective criteria, but a optimal level and structure of rates. The
close approximation might be difficult to considerations relevant here are well
achieve. known from the literature on oligopolistic
market structure and price leadership. It It is worthwhile to consider briefly the

follows that when conditions are changing comparative static properties of equili-
rapidly we might expect to find that the brium rationing in our model. The total
entire structure of rates would lag behind, amount of equilibrium rationing, E, can be
and thus, for a while, would tend to be written as the difference between loan
higher or lower than the optimal structure, demand and loan supply for those cus-
depending on the direction of the change. tomers of the bank experiencing rationing:
These considerations provide the key to
n
dynamic rationing to be elaborated below.

(8) E = max [Di(h) - L,(Ai), 0]
2. Long-Run Equilibrium Credit Rationing
We are now in a position to assemble the The excess supply of loans to those cus-
complete theory of credit rationing. We tomers to whom the bank would like to
first take up equilibrium credit rationing extend additional loans does not, of course,
which occurs when the loan rate is set at offset the rationing to other customers.
its optimal level. In Section I it was shown The offer curve to a firm depends on the
that credit rationing will be profitable, opportunity cost of the bank, I, and the
even in long-run equilibrium, as long as firm's density function of possible out-
there is uncertainty of loan repayment and comes, fi, and hence E depends on these
banks cannot discriminate perfectly be- two factors and the firms' demand func-
tween customers. In Section II.1 both of tions. Consequently, for purposes of com-
these conditions were verified as features parative static analysis, oile may consider
of the commercial banking industry, and the impact of changes in demand, risk, and
thus we conclude that equilibrium ration- opportunity cost on the amount of equili-
ing is consistent with rational economic brium rationing. Unfortunately, the eval-
behavior. uation of the impact of these changes
This conclusion would remain basically turns out to be a very difficult process. In
unchanged if one recognizes that ;the particular, the rationed or nonrationed
bank's expected return and cost is af- status of customers may change because of
fected by other contract terms such as loan the change in the underlying parameter.
maturity and compensating balance re- To at least illustrate the possible out-
quirements. To be sure, if a bank could comes, however, we shall outline the case
discriminate freely between customers of a change in the opportunity cost I.
with respect to such factors, there might Two important results for the compara-
be no occasion for equilibrium rationing. tive static analysis are given in the follow-
But while the empirical evidence is scanty, ing propositions:
one would surmise that banks are limited PROPOSITION 8. A ceteris paribus iincrease
in their power to discriminate with respect in the opportunity cost causes the
to these terms as well as the nominal optimal loan offer curve to shift down-
interest charge. Similarly, our analysis ward at every interest rate. Indeed,
can incorporate the benefits of maintain- by implicit differentiation of equation
ing long-run "customer relationships" (cf.
contract terms may serve to modify the importance of
Hodgman [4], Edward Kane and Burton
uncertainty in our theory of credit rationing, the bank's
Malkiel [6]) without invalidating our basic
inability to discriminate perfectly in setting these terms
conclusions about the rationality of ra- is still critical. The need for such an element can be seen
clearly in the attempts by Hodgman [4, p. 2651 and Kane
tioning.18
and Malkiel [6, p. 1231 to rationalize credit rationing
X Althouwh the customer relationship and nonprice w'ithout appeal to such imnerfect discrimination.
(3) we obtain: demand curve D1, the customer would

move from a rationed to a nonrationed
09Li - 1
(9) _ f < R status, while for the case illustrated by the
AIl fsR 2 demand curve D2, the opposite conclusion
holds. Cases in which the rationing status
PROPOSITION 9. The common optimal
remains unchanged are, of course, equally
rate factor R* (applicable if the bank
possible. Thus the effect of an increase in I
were constrained to satisfy all de-
on the size of rationing must remain
mand functions) is positively related
uncertain. This result, however, should not
to the opportunity cost I.
be regarded as disturbing. What it implies
Proposition 8 is self-evident. Proposi- is that in the long run, no systematic rela-
tion 9 is in line with the standard theory of tionship exists between the extent of equilib-
the firm in that an increase in marginal rium rationing and the absolute level of
cost necessitates an increase in marginal interest rates.
revenue which takes the form of an in-
crease in R*. Because the very act of 3. Dynamic Credit Rationing
changing the parameter I may lead to a
We define dynamic rationing as the
change in the rationed or nonrationed
difference between equilibrium rationing
status of customers, as demonstrated be-
and the volume of rationing that arises
low, special cases can arise in which iR (the
when the actual rate charged customers,
common optimal rate allowing for ration-
R, differs from the long-run equilibrium
ing when profitable) and I are not posi- rate, R. By this definition, dynamic ration-
tively related, contrary to the relationship
ing can be positive or negative, and we
between R* and I. Since such cases are shall show that its magnitude will be
distinctly abnormal, the obvious and rea- positively associated with the spread, Ri-R.
sonable situation being a positive relation-
Furthermore, we have already suggested,
ship between I and R, we shall proceed on in view of the oligopolistic structure of the
the premise that proposition 9, is equally
banking industry, that R is likely to adjust
applicable to f.17
slowly to changes in R', thus lending
With the aid of Figure 2, we can survey
the impact of an increase in the oppor- ,D \
tunity cost from Io to h1. From proposition

8, it is known that the offer curve will
shift downward from the locus Lo to Li;
and from proposition 9 it is known that
the common optimal rate factor will in-
crease, say from Ro to Ri. Thus the net im-
pact of a change in the opportunity cost on
the maximum loan offered is the result of a
shift in the offer curve and a movement
along this curve, and the amount of
equilibrium rationing may either rise or Igto
fall. These two alternative outcomes are
illustrated in Figure 2. In the case of the
I0 Ro I Rr I . A
"I The abnormal conditions which lead to
FiGuRE
mal result are discussed further 2 [5, p.
in
credence to the empirical importance of crease in I will lower all offer curves and
dynamic credit rationing. In fact, there is also increase Ri. Since the quoted rate re-
evidence that banks may tend to rely on mains at Ro and all the demand curves are
some objective signal, such as changes in unchanged, there must be an aggregate in-
the Federal Reserve's discount rate, in crease in rationing. Our model also provides
determining the timing of loan rate some information about the incidence of
changes, and hence, dynamic credit ration- this increase. At one extreme, the risk free
ing would valy significantly depending firms will still not experience any rationing
upon Federal Reserve policy.'8 because the offer curve for them is a ver-
It is helpful to begin by considering a tical line.19 At the other extreme, firms
system that starts in long-run equilibrium already rationed in the initial equilibrium
with a rate, R0 and then receives a shock will be rationed even more as the offer
that changes Ro to R, while the quoted curve shifts down. Finally, among the
rate remains unchanged at Po. As we have risky firms initially not rationed, some will
seen, there are three main market forces remain unrationed while others will ex-
that can change Rf: a change in market perience new rationing, with the amount
interest rates leading to a change in the depending upon the extent of the shift.
opportunity cost I; a shift in customer Thus on the whole, the loan portfolio
demand schedules; and a change in risk as will shrink and the funds released by
may be indicated by a shift in F[x]. We rationing will be shifted into other assets
shall first consider the effects of changing whose yield has increased (or be used to
each of these factors, one at a time. Then repay the now more costly borrowed
by combining the results of these ceteris funds). However, the opportunity cost I is
paribus experiments, we can examine more now likely to fall relative to the market
realistic situations in which several factors rate rM because loans will be a smaller per-
change simultaneously. centage of the total portfolio. This will
1. Consider first a change in market tend to moderate, though not eliminate,
rates of return which, for convenience, the initial increase in rationing.
may be summarized by some representa- 2. Consider next the effect of a down-
tive rate, rM. a rise in rm will directly in- ward revision of the anticipated distribu-
crease the opportunity rate for funds in- tion of outcomes; operationally, we may
vested in the loan portfolio. From proposi- think of the initial distribution F[x] being
tions 8 and 9, we know that such an in- replaced by a new one G[x] with G[x]<
18 Empirical tests performed in an earlier work by one
F[x]. As can be verified from (3), such a re-
of the authors [5, Ch. 41 confirms the hypothesis that, in
the period covered by our data, changes in the Federal '9 This conclusion would not hold if the shift in I
Reserve discount rate were a major factor influencing were so large as to exceed Ro. In this case, the best
the timing of changes in the commercial loan rate. It course of action would be to cut off loans even to the
should be stressed, though, that such a relationship prime firms, and a fortiori to all customers. But this
might not continue to hold in the future unless the case can be disregarded for the banks could be counted
Federal Reserve continues to operate the discount upon to respond promptly by raising their quoted rate,
window in the customary fashion. Should the recent i.e., Ro would not remain unchanged in these circum-
Federal Reserve proposal [111 to keep the discount rate stances. More generally in the usual theory of monopoly
closely in line with market rates be enacted, it is likely or oligopoly, if the equilibrium price rose because of a
that the commercial loan rate would adjust more shift in either the demand or the cost function, and, for
quickly toward its desired level. In this case, dvnamic some reason, the market price was prevented from ris-
rationing in response to changes in market conditions ing, rationing would occur only if the shift were such
would tend to die out faster than in the past. Thus dis- that the marginal cost would exceed the price. It ap-
count rate policy is at least one way in which the Fed- pears, therefore, that dynamic rationing, just as equilib-
eral Reserve can influence the amount of credit ration- rium rationing, is intimately related to the uncertainty
ing. about the outcome of the loan.
864 - THE AMERICAN ECONOMIC REVIEW
vision results again in a downward shift in offered at the unchanged rate, Ro; at the
the offer curve, and in an increase in the same time, the quantity demanded at that
R" and hence in R. Since the quoted rate rate rises, and if I were unchanged, the
has not changed, the maximum loan of- effect on X, as well as on rationing, would
fered at that rate must tend to decline. depend on the relative amount of the two
With the demand unchanged, rationing shifts. Even on this basis, one would antic-
must therefore increase for some of the ipate an increase in rationing because the
customers, and the incidence of the in- demand curve is likely to shift further than
creased rationing is entirely analogous to the offer curve, reflecting an increase in the
that of case 1. Once more, as loans shift optimism of firms relative to that of the
out of the now less remunerative loan banks. But more fundamentally, I must
portfolio, the opportunity cost I may rise, both because with I constant, more
decline somewhat, mitigating the initial funds would flow into the loan portfolio
effects. causing I to rise relative to rM, and be-
3. Much the same conclusion can be cause rM itself will be rising. Hence the
seen to hold for the case of a shift in all final outcome will tend to be the same as in
demand schedules, rm and F[x] constant, the previous three cases. The extent of in-
except that in this case, it is the demand creased rationing will depend on the rela-
curve which shifts while the offer curve tive shift in F[x], in the quantity de-
remains unchanged. If I remains un- manded at the unchanged rate, and in rM,
changed, then the initially rationed cus- and on whether, on balance, these shifts
tomers will experience more rationing be- will cause funds to flow in or out of the
cause their increased demand is not satis- loan portfolio. It is apparent, however, that
fied at all; the risk free customers, in con- the rise in I and in rationing will tend to
trast, will receive larger loans and remain be greater the smaller the elasticity of
unrationed; and the initially unrationed supply of funds to the banking system. The
risky customers will also obtain larger effect might be particularly severe if the
loans, but possibly not enough to match ability of banks to attract funds were ac-
the increase in their demand. Furthermore, tually reduced, as happened in some recent
as funds flow into the loan portfolio (be- episodes in which the Certificates of De-
cause some of the increased demand is posit rates in secondary markets pierced
satisfied), the opportunity cost of loans the Regulation Q ceiling.
will tend to increase giving rise to addi- We may thus conclude quite generally that
tional rationing of the type under (1). as R rises relative to R&o, rationing will tend
Normally, an increase in loan demand to increase; and the incidence of the in-
will tend to occur in periods of buoyant creased rationing will tend to fall most
economic activity and hence will be asso- heavily on customers who would be rationed
ciated with a rise in rM (partly reflecting in equilibrium, and will tend to affect the
the higher demand in all markets and least, if at all, the riskless, or nearly riskless,
partly causing, in turn, a higher demand prime customers.20 One implication of this
for bank funds), and also with anrincrease
in the anticipated profitability of the in-
30 As we have argued in the text, the amount of dy-
vestment projects, the latter producing a
namic rationing tends to be positively related to the
decrease in risk as measured by F[x]. The spread between the equilibrium rate and the quoted
outcome is then a combination of the rate. Actually, though this relationship must hold in
the large, in some special circumstances it may fail to
ceteris paribus results under (1), (2), and hold in the small. The special cases arise only when the
(3). The shift in F[x] tends to raise the loan slope of the demand curve exceeds the slope of the offer
result is worth stressing since it provides defined by E in equation (8). The degree or
the key to our operational measurement of relative incidence of credit rationing could
credit rationing set forth in the next sec- then be measured by the ratio of the vol-
tion. Suppose that we were to classify all ume of rationing to the potential demand
customers into two broad classes, the of rationed customers, or:
prime customers and all others. We should
E D2-L2
then expect that as the gap between JR and (10) H = ~
R widens and dynamic rationing becomes E + L2 De
more severe, loans to the riskless cus- where D2 denotes the demand of rationed
tomers will tend to represent a growing
customers and L2 the volume of loans ac-
share of the total loan portfolio.
tually granted to them.
This result can be given the following
Unfortunately, the direct measurement
useful interpretation, which also serves to
of E and L2, the components of A, requires
bring to light the common nature of equi- information on the ex ante customer de-
librium and dynamic rationing. In the
mand and bank supply which is unlikely
presence of risk as to the outcome of the
to be available, even in the future. The
loan, reducing the size of the loan will in-
analysis of Section 11.3, however, points to
crease the expected rate of return, by re-
a possible, operational, proxy measure of
ducing the expected loss from insolvency
the degree of dynamic credit rationing. As
of the firm. It is therefore quite under-
shown there, our model suggests that there
standable that a bank faced with a higher should be a positive association between
opportunity cost (whether from a rise in variations in dynamic credit rationing and
the market rate or in lending oppor-
variations in the proportion of the total
tunities) and unable to raise the return by loan portfolio accounted for by the risk
raising rates, will find it profitable to raise
free prime customers. Let us then denote
its'return at least byVupgrading the quality
by Li-DI, the volume of loans granted to
of its portfolio through a reduction in risk; these customers, and by L2 and D2, re-
the upgrading may take the form of shift-
spectively, the loan granted and the loan
ing funds toward less risky customers,
demanded by all other customers. Our
and/or of reducing loans made to risky proposed operational proxy for the non-
customers, depending on the nature of the observable iH is either of the following two:
shift in underlying conditions.
III. A Measure of Credit Rationing (1 1.a) H1 Li

L1 + L2
In this section we shall develop an opera-
or
tional measure of credit rationing which is
based on the theory as developed in Sec- 1 L1+ L2
tions I and II and which is used in the test (11l.b) 152 = H-=
H1 L
of the theory presented in Section IV be-
low. In principle, the volume of credit The first measure, H1, is simply the per-
rationing should be measured by the centage of total loans which are granted to
difference between the loan demand and the risk free customers, and hence is posi-
bank supply for rationed customers as tively related to the degree of rationing.
The alternative proxy is its reciprocal and
curve. Since this can occur only at relatively high in- hence, is negatively related to the degree
terest rates, indicating default risks well above levels
bankers would consider acceptable, the case would seem
of rationing.
to have little empirical relevance. To see the relation between either proxy
and the ideal measure Hf, let concerning the comparative static prop-
erties of the model and the classification
B D2 of customers suggest that, at least in
Di principle, these variables would not be
correlated.
Then from (11) we obtain:
Some data problems are encountered
I even in attempting to measure the proxy
(12.a) H = 1 + B(1-B) developed in equations (11). Our source of
data is the Federal Reserve's "Quarterly
(12.b) H231+B(t-H-)
Interest Rate Survey" which records the
since L1=Di. volume of new loans granted by rate and
Differentiating (12) with respect to 4, size class during the first two weeks of the
yields: last month of the quarter. With this data,
the percentage of loans granted to risk free
,11, B firms (proxy H1) can be at least approxi-
(13.a) >0 mated by the percentage of loans granted
at the prime rate (and perhaps secondarily
by the percentage of loans which were
(13.b) -B < O
large in size). Unfortunately, in several in-
stances the prime rate changed during the
The equations (12) show that for a given period of the survey, with the effect that
value of B, our proxies are monotonic
one cannot distinguish between loans made
functions of the ideal measure ft, and (13) at the new prime rate and loans made at
confirms that the relation is in the direction this same rate while the old prime rate
expected. The functions relating either was still in effect. To circumvent this prob-
proxy to f involve as a parameter the
lem and secure a more reliable measure,
relative demand factor, B, essentially be- we smoothed these quarters as well as
cause we have replaced the nonobservable
possible. Following John Hand [2] who
D2 with the observable D1. This of course first used the data for this purpose, we
implies that any change in B will give rise combined the smoothed series with three
to a variation in H1 or 2 which does not other measures based on the distribution
correspond to variations in H. Hence, if
of loans by size through principal compo-
there were sizable variation in B over nents analysis. More specifically, we cal-
time, our proxy measures of ft could be culated the first principal component of
subject to appreciable errors in measure- the following four series:
ment. Note, however, that even in this
a) The proportion of total loans granted
event, as long as H, or 12 are used as
at the prime rate.
dependent variables in a statistical test as
b) The proportion of total loans over
we shall do below, these errors will not
$200,000 in size.
tend to generate bias in the estimated
c) The proportion of loans over
coefficients unless B happens to correlate
$200,000 in size granted at the prime
with the behavioral determinants of .21
rate.
Fortunately, the conclusions of Section II
d) The proportion of total loans $1,000-
21 Because (12a) is nonlinear in H, this statement $10,000 in size.
will be true only for a linear approximation of 11. If
either proxv were being used as an independent variable,
the problem of bias would be more serious. See, for
The first three series should enter posi-
example, Malinvaud [7, Ch. 101. tively into the principal component and
the fourth negatively, of course. The IV. A Test of the Model Using the
principal component has a mean of zero Credit Rationing Proxy
and standard deviation of unity and the
In this section we propose to use the
four factor loadings were, respectively:
credit rationing proxy to test the implica-
(a) .988 tions of our model as to the forces con-
trolling variations in time in the extent of
(b) .968
credit rationing. This test, relying on
(c) .939 quarterly time series data for the years
(d) -.959 1952 to 1965, will thus serve to shed light
on three aspects of our problem: our theory
This indicates that each series enters prom-
of credit rationing, the effectiveness of the
inently and about equally into the
proxy variable based on the theory, and
principal component.
the existence of rationing as an empirically
The principal component thus derived
significant phenomenon.24
corresponds to the IL measure since the
The principal implication of our theoret-
series (a) is analagous to H1. Exploratory
ical model is that the main source of
calculations indicated that the results
systematic variations in credit rationing is
would not be significantly affected by rely-
to be found in changes in dynamic ration-
ing on the alternative H2 measure derived
ing; and that these changes in turn are
from the reciprocal of series (a) to (d).22
positively associated with the spread be-
The solid line in Figure 3 is a plot of the
tween the long-run optimal or equilibrium
seasonally adjusted principal component H
loan rate denoted hereafter by r*, and the
to be used in the tests of the following
rate actually prevailing, rL.25 If we further
section. The pattern of rationing indicated
assume that, to a first approximation, this
by the proxy seems quite credible through-
association cail be formulated as a linear
out the period. Note in particular how the
relationship within the empirically rel-
most recent pattern is consistent with
evant range, we are led to
what we might have expected, rising very
high in the second and third quarters of
(14) H = ao + al(rL-rL) + e
1966 and then falling off somewhat in the
fourth quarter. Unfortunately this series Recall that, according to our model, w
cannot be computed beyond 1966 at the the commercial loan rate is at its long-r
present time, since the information pro- desired level, so that the second term of
duced by the loan survey beginning in the (14) is zero, we have only equilibrium
first quarter of 1967 is not strictly com-
parable with the earlier information.23 ponent in the four series used to compute H, sufficient
observations for the new survey period must be ob-
tained before the new seasonal component can be re-
22 We estimated a number of equations using the
rationing model developed in Section IV for the specifi- liably determined.
cation of the independent variables, and the individual 24 It is important to note that success in this test will
series (a) to (d), the principal component, and the re- confirm the value of the proxy as a variable to be used
ciprocal of each as separate dependent variables. The in testing for the impact of credit rationing on the real
principal component yielded the best fit as expected, sectors of the economy. Indeed, the first uses of the
but comparable results were obtained with the other proxy for this purpose have been made in [21 and [5].
measures. We are grateful to John Hand for making his 25 rL denotes an empirical approximation to the
data on these series readily available. theoretical construct R developed above. In particular,
23 The difficulty arises from the fact that with the rL must stand proxy for the spectrum of optimal rates
first quarter of 1967, the period of the survey was corresponding to the respective risk classes. Similarly,
changed from the first two weeks of the last month of rL stands for the spectrum of actual rates and is mea-
each quarter to the first two weeks of the second month sured as the average rate on commercial loans compiled
of each quarter. Because of the strong seasonal com- from the "Quarterly Interest Rate Survey."
co + ct{rr + b6[(DEP/TB) - 1] - b2D62 + b2'C + b3L/(A - L)

+ b4A[L/(A - L)]}
rationing. On the microeconomic level, that level at which the marginal proceeds
equilibrium rationing depends on the from a commercial loan, after adjustment
specific parameter values for the demand for risk, would just equal the opportunity
functions, density functions, and oppor- cost. We also know that this relationship
tunity cost. But our investigation in Section will be valid for all other assets in the
11.2 indicated that no systematic relation- bank's portfolio. This means that the
ship existed between the degree of ration- optimal commercial loan rate will tend to
ing and changes in these parameters. Ac- equal the market yield on any other asset
cordingly in (14), the constant term ao may held in the bank's portfolio after adjust-
be thought of as a measure of equilibrium ment for risk, maturity, liquidity, and,
rationing up to a stochastic error term possibly, any expectations concerning fu-
which is included in the overall error term ture levels of that rate. We are free, then,
e. When the quoted rate is above the to choose as the standard of comparison,
equilibrium rate, the amount of dynamic any security which is widely held by
rationing can be considered negative in the banks, the obvious criterion being practical
sense that rationing will be reduced below expedience. Our choice, on this basis, is the
its equilibrium level, even though the bank's holdings of Treasury bills.
actual amount of rationing can never be On the basis of the above considerations
less than zero by definition. The dependent we are led to equation (15) as the specifica-
variable as measured by our principal tion for the desired commercial loan rate,
component proxy will, in fact, take on where the notation will be defined as we
negative values because the zero point for proceed.
H is chosen arbitrarily. The arbitrary Consider first the bank's return on
choice of origin is, of course, reflected in Treasury bills. It may be regarded as con-
the constant ao, and consequently one can- sisting essentially of two components. The
not distinguish between the level of equili- first component, the Treasury bill rate,
brium rationing and the scaling effect in rT, is straightforward. The second com-
the constant.26 ponent is the liquidity value of Treasury
We now turn to the important task of bills, which is more involved and, in fact,
specifying the equilibrium commercial loan accounts for the second, third, and fourth
rate, r*. Our theory indicated that the terms in the brackets. Our basic premise is
desired commercial loan rate would be at that the liquidity yield of Treasury bills
should decrease as the bank's holdings of
26 Equation (14) can be rewritten such that the spread
between the desired commercial loan rate and the rate these bills (TB) rises relative to its deposit
actually quoted is a linear function of the degree of liabilities (DEP). Our specification for this
credit rationing. In addition, it has already been sug-
liquidity term takes the form b1 [(DEP!
gested that the timing of adjustments in the commercial
loan rate depends on changes in the Federal Reserve TB)-1] where bi (>0) is an estimated
discount rate. These two factors can be combined into a parameter.27
testable partial adjustment model of the determinants
of the commercial loan rate in which the size of the de- 27 The legal requirement that banks must maintain
sired adjustment depends on the degree of rationing government securities in their portfolio as collateral for
while the speed of adjustment depends on changes in the government deposits raises one conceptual problem. To
discount rate. See fn. 18. the extent that this requirement necessitates holding
The liquidity value of Treasury bills has of loans to assets (or liabilities) grows and
been reduced since about 1962, however, thus measure this effect as b3(L/(A-L))
by the development of a broad and active where b3 (>O) is an estimated coefficient,
market in Certificates of Deposits (CD's), A is total loans and investments, and L is
which affords the banks an important the commercial loan portfolio of the banks.
means for increasing their liquidity at Finally, we should also consider changes in
short notice. The effect of the existence of the liquidity ratio, since a dynamic short-
this market may be measured by a simple run increase in loans which is beyond the
shift parameter or dummy variable and this control of the bank would have additional
is included as the term -b2 (D62) where (although only transitional) liquidity cost,
D62 is a dummy variable which is unity and this effect is accounted for by the last
starting in 1962-I and zero before then.28 term in (15).
The value of the CD market is of course We have now almost completed the task
severely limited when the Regulation Q of specifying the desired commercial loan
ceiling on CD interest rates is binding. This rate. To obtain more generality we have
countervailing effect may be specified by formulated the desired loan rate as a linear
an additional dummy variable which takes function of the terms just summarized.
the value one in those quarters, if any, in The need for the linear function arises be-
which the secondary market rate for CD's cause we have not yet formally accounted
exceeds the ceiling rate. One would expect for differences in risk and maturity be-
this dummy variable, denoted by C, to tween commercial loans and Treasury
have a coefficient opposite in sign and of bills. The basic equation to be estimated
the same order of magnitude as the CD can now be derived by substituting equa-
dummy variable D62. tion (15) into (14), which yields:
The next to last term in equation (15) H = (ao + alco - d1b1) - al(rL)
measures the share of the loan portfolio in
total assets which, as suggested in Section + dl(rT) + d1b1(DEP/TB)
II, should tend to affect the opportunity (16) - d1b2(D62) + d1b'C
cost of funds for loans relative to market
+ d1bs[L/(A - L)]
rates. In addition, this variable may be
visualized as an adjustment of the required + d1b41[L/(A - L)]
rate on loans for their relative illiquidity. where di=aic1.
We anticipate that this illiquidity should The coefficients of (16) were estimated
increase at an increasing rate as the ratio using ordinary least squares from observa-
tions for the period 1952-II to 1965-IV.29
Treasury bills of some required amount, the correct
variable for our analysis would be the free bills; that is,
(The year 1966 was omitted to enable us to
the bills held above the required amount. It has been carry out extrapolation tests reported be-
suggested that, at least in 1966, such a restriction was a low.) The results are as follows:
restraining influence on bill holdings. Legally, however,
a wide variety of Federal and Local Government se- 29 The source of data for the independent variables is
curities are acceptable as collateral and it is unlikely the Federal Reserve Bulletin with the exception of com-
that more than a few banks, probably centered in New mercial loans. Commercial loans are the sum of indus-
York City, held Treasury bills only to satisfy collateral trial and commercial loans (from an unpublished Federal
requirements, even in 1966. Reserve series) and nonresidential mortgage loans of
28 We have also experimented with incorporating the the commercial banks (from the Federal Reserve Bul-
CD term in the form of a multiplicative factor operating letin). The mortgage loans are included since they are
on the Treasury bill ratio itself. Since the two estimates made to the same customers as the shorter maturity
are nearly identical, only the results for the linear form commercial loans. All dollar magnitudes are seasonally
will be shown. adjusted and interest rates are measured as a percent.
(17) H = - 3.48 - 1.16rL + .296rT CD's. The coefficient of the CD dummy,

(-7.0) (-5.3) (2.2) D62, is both very significant and large
(since the dependent variable has by con-
+ .007DEP/TB - 1.27D62 struction unit variance). It suggests that
(1.2) (-7.8) the newly acquired ability of banks to at-
tract and shed funds through CD's con-
tributed appreciably to a reduction of
+ .614C + 26.7[L/(A - L)]
(1.6) (8.0) rationing. The contribution of the ceiling
rate dummy C, on the other hand, is harder
to assess because the ceiling was binding,
+ 9.OA[L/(A-L)j
and hence the dummy was one, only in the
(.92)
last quarter of the sample, 1965-IV.80 Its
coefficient has the expected positive sign
Se = .350 R2 = .84 D.W. = 2.02
(T statistics shown in parentheses) but its magnitude, which we had expected
to be roughly equal to that of the coeffici-
ent of D62, is instead only half as large.
These estimates confirm the implica- Yet this result makes good sense when we
tions of our model in that all coefficients recall that the ceiling rate was effective
have the correct sign and the goodness of only to December 6, which is just about in
fit is respectable, taking into account the the middle of the two week period during
noise in the dependent variable which was which the loan survey is taken. Indeed, it
discussed above. Note in particular the suggests that if the ceiling had been effec-
very significant negative coefficient of the tive throughout the period, then its effect
commercial loan rate. This supports one of would have come close to offsetting totally
the most distinctive implications of our the contribution of the D62 dummy as
model, to wit, that a rise in the commer- expected.
cial loan rate given the optimal level of this The effectiveness of our model in ex-
rate as measured by the remaining vari- plaining the behavior of the rationing
ables in the equation, tends to reduce ra- proxy can also be judged from the plot of
tioning as it reduces the demand and in- the values of H computed from equation
creases the supply. Similarly, the oppor- (17) which is shown by the dotted line in
tunity cost as measured by the Treasury Figure 3. The equation appears to track
bill rate, and the loan illiquidity variable, the broad movements of the actual series
measured by the share of loans in the as well as the major turning points with
bank's portfolio, appear as the most signif- leads or lags not exceeding one quarter.
icant factors tending to increase ration- The largest errors correspond to a number
ing, given the commercial loan rate. Al-
30 The multiplicity of CD maturities and the thinness
though the Treasury bill liquidity term of the secondary market make it difficult to obtain a
and the change in the commercial loan reliable indicator of those periods in which the ceiling is
binding. In addition, an aggregation problem arises
illiquidity term are not significant, this is
because the ceiling may be binding only for some banks
probably due to multicollinearity since in or for some regions. The available information sug-
the short run, with the level of total assets gests, though, that until the last quarter of 1965 the
ceiling rate was promptly raised whenever it threatened
essentially fixed, banks may have to sell
to become a significant hindrance to the issuing banks.
Treasury bills to meet unexpected loan In October and November of 1965, however, the secon-
demand. dary rate rose unequivocably above the ceiling and re-
mained there until December 6th when the ceiling was
The remaining variables are the
again raised. See Willis [10] for further discussion of the
dummies intended to measure the effect of secondary market for CD's.
JAFFEE AND MODIGLIANI: CREDIT RATIONIN'G 871
Credit Rationing, 1952- II to 1966- IV
Actual Value (H)

.......... ... fitted Value, through
----, Extrapolatecl Value, Coefficient of C=O
"--" Extrapolated Value, Coefficient of C=1.27
tQ
*-.~~~~~~~~~~~~~~.
0.0~~~~~..
m I E I G I I 8 X X I I In E
FIGURE 3
of one quarter spikes in H which may well represented in Figure 3 by the dashed line,
reflect mostly noise in that series. in which the dummy variable C was as-
In Figure 3 we also present some extra- signed the value of one. In addition, since
polations of our equation to the year 1966, the ceiling rate was effective throughout
which marks the end of the period for these quarters, it was also assumed that
which usable data on the rationing proxy the coefficient of C was equal numerically
are presently available.3' Unfortunately to that of D62. Stated differently, the
extrapolations of (17) to 1966 run into alternative extrapolation assumes that a
rather formidable difficulties because binding ceiling has the effect of undoing
throughout the last three quarters the the loosening effect of CD's measured by
ceiling rate on CD's fell short of the secon- the dummy D62. It is seen that this alter-
dary market rate. Nonetheless we feel it native fits the observations remarkably
worthwhile to exhibit these extrapolations well.
because of the tentative light they shed on These results, if taken at face value,
the working of the ceiling rate. If one have rather interesting implications for the
extrapolates (17) as though the ceiling rate modus operandi of ceiling rates as a tool of
dummy C were zero, one obtains the com- monetary policy. In particular they would
puted values represented by the dotted support the view that, by allowing the
dashed line and as expected, this extra- ceiling rate to become a real hindrance to
polation very much underestimates the the ability of banks to attract new CD
extent of rationing in the last three quar- funds, the Federal Reserve could reduce
ters. In order to allow for the ceiling effect, significantly the availability of funds to
an alternative extrapolation of (17) was the commercial bank customers of the
carried out for the last three quarters, banking system. This reduction would
presumably occur to the benefit of cus-
'1 Cf. fn. 23. tomers of other intermediaries and/or of
those firms able to raise funds directly in 5. D. M. JAFFEE, Credit Rationing and the
the market. It should be recognized, how- Commercial Loan Market, unpublished
ever, that our evidence in support of this doctoral dissertation, M.I.T. 1968.
inference is at the moment rather limited 6. E. J. KANE AND B. G. MALKIEL, "Bank
Portfolio Allocation, Deposit Variability,
and hence, quite tentative, until it can be
and the Availability Doctrine," Quart. J.
confirmed by further experience under
Econ., Feb. 1965, 79, 113-34.
similar circumstances. Of course, by the
7. E. MALINVAUD, Statistical Methods of
time the system is again exposed to similar Econometrics, Chicago 1966.
circumstances, it may have learned ways 8. M. H. MILLER, "Credit Risk and Credit
of evading or by-passing, at least partially, Rationing: Further Comment," Quart.
the constraint imposed by the ceiling. J. Econ., Aug. 1962, 76, 480-88.
9. R. RoOSA, "Interest Rates and the
REFERENCES
Central Bank," Money, Trade, and
1. M. FREIMER AND M. J. GORDON, "Why Economic Growth in Honor of John Henry
Bankers Ration Credit," Quart. J. Econ, Williams, New York 1951.
Aug. 1965, 79, 397-416. 10. P. B. WILLIS, "The Secondary Market
2. J. HAND, The Availability of Credit and for Negotiable Certificates of Deposits,"
Corporate Investment, unpublished doc- paper prepared for the Committee for
toral dissertation, M.I.T. 1968. the Fundamental Reappraisal of the
3. D. R. HODGMAN, "Credit Risk and Discount Mechanism, appointed by the
Credit Rationing," Quart. J. Econ, May Board of Governors of the Federal Re-
1960, 74, 258-78. serve System.
4. , "The Deposit Relationship and 11. BOARD OF GOVERNORS OF THE FEDERAL
Commercial Bank Investment Behav- RESERVE SYSTEM, Reappraisal of the
ior," Rev. Econ. Statist, Aug. 1961, 43, Federal Reserve Discount Mechanism,
257-68. Washington 1968.

A Theory and Test of Credit Rationing

Enviado por

Dados do documento

Descrição original:

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

A Theory and Test of Credit Rationing

Enviado por

Direitos autorais:

Formatos disponíveis

American Economic Association

A Theory and Test of Credit Rationing

Now let Ri= 1+ri be the interest rate

approach zero as the rate grows beyond But then,

banks for the customer's business is dis-

the center of the theoretical discussion

tomer must be a concave function of R in (6.2) R2 ?sR* > X2 implying that

14 The assumption of global concavity is not a neces-

of D1 [RI and D2 [R], respectively.

market structure and price leadership. It It is worthwhile to consider briefly the

dynamic rationing to be elaborated below.

(3) we obtain: demand curve D1, the customer would

tunity cost from Io to h1. From proposition

III. A Measure of Credit Rationing (1 1.a) H1 Li

co + ct{rr + b6[(DEP/TB) - 1] - b2D62 + b2'C + b3L/(A - L)

(17) H = - 3.48 - 1.16rL + .296rT CD's. The coefficient of the CD dummy,

Credit Rationing, 1952- II to 1966- IV

Actual Value (H)

Você também pode gostar