Professional Documents
Culture Documents
Journalof: Banking & Finance
Journalof: Banking & Finance
BANKING &
FINANCE
E~E~ER Journal of Banking & Finance 21 (1997) 251-271
Abstract
This paper explores the determinants of optimal bank interest margins based on a simple
firm-theoretical model under multiple sources of uncertainty and risk aversion. The model
demonstrates how cost, regulation, credit risk and interest rate risk conditions jointly
determine the optimal bank interest margin decision. We find that the bank interest margin
is positively related to the bank's market power, to the operating costs, to the degree of
credit risk, and to the degree of interest rate risk. An increase in the bank's equity capital
has a negative effect on the spread when the bank faces little interest rate risk. The effect of
rising interbank market rate on the spread is ambiguous and depends on the net position of
the bank in the interbank market. Our findings provide alternative explanations for the
empirical evidence concerning bank spread behavior.
Keywords: Bank interest margins; Credit risk; Interest rate risk; Risk aversion
1. Introduction
i Bank interest margin, or net interest margin as it is commonly called, is usually defined as the
difference between interest revenue and interest expense expressed as percentage of average earning
assets. Spread, on the other hand, is the difference between the yield rate on average interest earning
assets and the cost rate on interest bearing funds, with both elements expressed in percentage terms.
Olson and Simonson (1982) show that bank interest margin and spread need not be identical unless
there are zero noninterest bearing funds. We are a bit informal here and use these two terms
synonymously.
2 Since we ponder over a situation with multiple sources of risk (both credit risk and interest rate
risk), the usual Arrow-Pratt theory of risk behavior may be too weak to yield intuitive results. As will
be shown below, two stronger notions of risk behavior, namely, the Ross (1981) characterization of
decreasing absolute risk aversion and the Kimball (1993) standard risk aversion, are needed to resolve
certain ambiguous results.
K.P. Wong / Journal of Banking & Finance 21 (1997) 251-271 253
3 In Zarruk and Madura (1992), there is an analytical mistake which invalidates their results. Zarruk
and Madura model loan defaults by a random variable lying between zero and the total amount of loans
granted by a bank. The total amount of loans, on the other hand, is a function of the loan rate. This
implies that the support of random loan defaults moves as the bank changes its loan rate. Zarruk and
Madura overlook this moving support factor and thus fail to obtain the correct first-order condition.
When this factor is taken into consideration, none of the comparative statics considered by Zarruk and
Mudura are determinate.
254 K.P. Wong/ Journal of Banking & Finance 21 (1997) 251-271
2. T h e m o d e l
L+B=D+K, (1)
4 The required minimum capital-to-deposits ratio can be made risk-based as in Zarruk and Madura
(1992) (i.e., K is an increasing function of L). Adding this complexity affects none of the qualitative
results.
5 Loan-rate setting behavior by banks is well documented by Stovin and Sushka (1983) and Hancock
(1986).
6 Throughout the paper, a tilde ( ~ ) always signifies a random variable.
K.P. Wong/ Journal of Banking & Finance21 (1997)251-271 255
RD = R o ( 0 ) + ~, (3)
7 The bank can in principle hedge against interest rate risk through the trading of interest rate
futures. The existence of basis risk in futures markets prevents a perfect hedge, and thus some interest
rate risk is always present.
8 Sealey (1980) argues that this assumption "can be justified on the basis of business cycle
movements in loan defaults and deposits" (p. 1145).
9 Using the technique introduced by Wong (1996), most of our qualitative results are unaffected if
we replace the nonlinear regression model by a more general dependence structure of 0 and Ro known
as regression dependence.
10 It is easy to show that the correlation between 0 and Ro is positive given (3) and R'o > 0. The
two random variables are independent only when Ro(O) is invariant to all 0 ~ [0, 1]. In fact, R'o >_0
is more than necessary for our qualitative results. What we really need is for R'o > - RLL(RL)/D at
the optimal loan rate. That is, as long as RD and 0 are not too negatively correlated, all of our
qualitative results remain intact.
II The marginal administrative cost of deposits and the fixed costs are omitted for simplicity because
they will have the same qualitative effect on the optimal bank interest margin as the marginal
administrative cost of loans.
12 The capital requirement constraint will be binding as long as the interbank market rate is
sufficiently higher than the expected deposit rate.
256 K.P. Wong/ Journal of Banking & Finance21 (1997)251-271
Before any uncertainty is resolved, the bank chooses its loan rate, RL, SO as to
maximize the expected utility of its end-of-period profit:
3. The o p t i m a l b a n k interest m a r g i n
where 0 = E(0), R E is the optimal risk-netural loan rate, and ~/n is the reciprocal
of the interest rate elasticity of loan demand evaluated at R~. Condition (8)
implies that the bank sets its optimal loan rate, R E, at the point where the expected
marginal profit of loans, (1 - 0)(1 - ~/n)R~ - C, equals the interbank market rate,
R, which is a well-known result by Klein (1971) for risk-neutral banks.
On the other hand, under risk aversion (i.e., U" < 0), we have 14
E[U'(qr*)lO] >(<)E[U'(qr*)[O] for 0>(<)5, (9)
where E(.[0) is the expectation operator with respect to G conditional upon
= 0. Since M' < 0, multiplying M(O) - M(O) to both sides of inequality 9 and
taking expectations with respect to F yields
Proposition 1. The optimal bank interest margin is larger when the bank is risk
averse than when the bank is risk neutral.
According to Ross, utility function V is strongly more risk averse than utility
function U if, and only if, there is a positive constant A such that
V"(~) V'(~2)
- - > A > - - for allTr 1 andzr 2. (12)
Vtt(q./.1) -- __ U t ( , . B . 2 )
Simply rearranging definition (12) reveals that V is more risk averse than U in
the Arrow-Pratt sense. Ross provides an example to show that the converse is not
true and thus definition 12 does indeed produce a strictly stronger ordering than
the Arrow-Pratt one. Ross further shows that definition 12 is equivalent to
V(~) = AU(~) + / 4 ( ~ ) , (13)
where H is a positive function with H ' < 0 and H" < O. Directly comparing the
optimal bank interest margin under V with that under U yields the following
proposition.
Proposition 2. The optimal bank interest margin increases as the bank becomes
stongly more risk averse in Ross' sense.
OO re=O--
Om
L(RL) >0.
RLE(RL)
That is, an increase in m increases the reciprocal of the interest rate elasticity of
loan demand at all loan rates. Thus, m can be viewed as a market power shift
parameter which keeps the market size of the bank constant at R L . We refer to an
increase in m as a size-preserving increase in market power of the bank.
Implicitly differentiating Eq. (7) with respect to m and evaluating at m = 0
yields
dC A A
(15)
The first term on the right-hand side of Eq. (15) can be interpreted as the
substitution effect, while the second term can be interpreted as the income
effect. 17
~7The decompositionof (15) into the substitution effect and the income effect can be formally done
via the method of compensation proposed by Davis (1989).
260 K.P. Wong/ Journal of Banking & Finance 21 (1997) 251-271
Proposition 4. I f the bank's utility function exhibits DARA in Ross' sense, then
an increase in the marginal administrative cost of loans will increase the optimal
bank interest margin.
Proposition 4 reveals that the bank passes the burden of rising operating
expenses to borrowers by widening the bank interest margin. This result is
consistent with the empirical findings of Ho and Saunders (1981) that actual bank
interest margins are positively related to operating costs.
thus it charges a higher loan rate to reduce lending. The income effect in this case
is therefore positive and reinforces the substitution effect. This implies that a net
borrower in the interbank market should raise its loan rate in response to an
increase in the interbank market rate. On the other hand, if the bank is a net lender
in the interbank market (i.e., B * > 0), the income effect becomes negative and
works against the substitution effect. This makes the total effect on R~ indetermi-
nate. To summarize, we have the following proposition.
Proposition 5. If the bank's utility function exhibits DARA in Ross' sense, then
the effect of an increase in the interbank market rate on the optimal bank interest
margin is either positive or ambiguous depending on whether the bank is a net
borrower or a net lender in the interbank market, respectively.
Ho and Saunders (1981) find that the relationship between the interbank market
rate and the actual bank interest margins is not statistically different from zero for
their full sample, which does not distinguish between banks according to their net
positions in the interbank market. Proposition 5 is consistent with this empirical
observation. Furthermore, according to Allen et al. (1989), large money-center
banks appear to be net Fed funds purchasers while smaller banks tend to be net
Fed funds sellers. An indirect implication of Proposition 5 is that large banks
should react positively to an increase in the interbank market rate, whereas small
banks' reactions are not known a priori.
E[U"(~-*)M(0)(0- 0)]R;L(RZ)E(R;)
+ A (17)
The first term on the right-hand side of Eq. (17) is the substitution effect which
is unambiguously positive. To see this, note that
( 0 - 0)(1 - n ) R L = M ( 0 ) - M(0). (18)
Using this and Eq. (7), the first term on the right-hand side of Eq. (17) becomes
E[ U'('~" * )] M ( O ) E ( R L )
>0,
A
262 K.P. Wong/ Journalof Banking & Finance21 (1997)251-271
where the inequality follows from condition 11. Roughly, the substitution effect
arises because greater credit risk increases the relative riskiness of loans vis-h-vis
lending in the interbank market, prodding the bank to steer away from issuing
risky loans even when compensated to remain at an unchanged level of expected
utility.
The second term on the right-hand side of Eq. (17) is the income effect which
is also positive given DARA in Ross' sense. To see this, rewrite the term using
Eq. (18) to yield
E[U"( ~r*)M( O)]M('O)L( R~ ) E ( R; )
( 1 - 'r/* ) d
E l U"(~'* ) M ( ~)2] L( R L ) E ( R; )
- >0,
(1--~*)A
where the inequality follows from E[U"(~r)*)M(O})] > 0 and condition 11. The
income effect arises because greater credit risk reduces the attainable expected
utility under risk aversion; at the greater level of risk, the bank has to be
compensated with additional income to attain the former (lower level of risk)
expected utility level. Taking away the compensation gives rise to the income
effect. With DARA, the sign of this income effect is not surprisingly positive.
As the income effect reinforces the substitution effect, the total effect of a
mean-preserving increase in credit risk on the optimal bank interest margin is
unambiguously positive, which is consistent with the empircal findings of Angbazo
(1995). 18
Proposition 6. If the bank's utility function exhibits DARA in Ross' sense, then a
mean-preserving increase in credit risk will increase the optimal bank interest
margin.
4.5. Increases in interest rate risk
side of Eq. (19) represents the income effect. Unfortunately, D A R A in Ross' sense
fails to ensure its sign. To get a concrete result, we use an additional strong yet
canonical notion of risk behavior.
Recently, Kimball (1993) has introduced the theory of standard risk aversion to
deal with situations of multiple risk bearing. He defines 'prudence' as U" > 0 and
the precautionary premium ~b as the quantity satisfying
U ' [ T r - ~0(~, #r)] = E [ U ' ( ~ - + ~)],
where ~ is a noise term. 19 He shows that $ is proportional to the index
- U ' ( ~ ) / U " ( T r ) , which he denotes as the degree of absolute prudence. A utility
function is said to have standard risk aversion if " e v e r y risk that has a negative
interaction with a small reduction in wealth also has a negative interaction with
any undesirable, independent r i s k " (see Kimball, 1993, p. 589), where "negative
interaction" denotes a utility loss. Kimball shows that this condition holds if, and
only if, the utility function exhibits both D A R A in the A r r o w - p r a t t sense and
decreasing absolute prudence (DAP).
Equipped with the concept of standard risk aversion, we show in the following
proposition that the effect of a mean-preserving increase in interest rate risk on the
optimal bank interest margin is unambiguously positive.
Proposition 7. If the bank's utility function exhibits DARA in Ross' sense and
standard risk aversion, then a mean-preserving increase in interest rate risk will
increase the optimal bank interest margin.
Ho and Saunders (1981), McShane and Sharpe (1985) and Angbazo (1995) all
find empirical evidence that bank interest margins are positively related to the
degree of interest rate risk. Thus, their findings lend support to Proposition 7.20
dK KA
19In essence, ~b is analogous to Pratt's risk premium, but for U' rather than U.
2o The effect of increases in the correlation between credit risk and interest rate risk on the optimal
spread is not reported in the paper because it is a priori indeterminate. The reason is that an increase in
this correlation, say, through a multiplicative shift parameter attached to the function RD(O), will
increase the interest rate risk at the same time. To restore the variance of /~o, we have to reduce the
magnitude of ~. An increase in RD(O)reduces the bank's profit and thus, with DARA, the bank has an
incentive to raise its loan rate. However, by Proposition 7, a decrease in ~ (so as to keep the variance
of /~o fixed) induces the bank to lower its loan rate. These two forces work against each other and
make the overall impact indeterminate.
264 K.P. Wong/Journal of Banking & Finance21 (1997)251-271
The right-hand side of Eq. (20) is the income effect which can be further
decomposed into two parts:
+ KA (21)
The first term on the right-hand side of Eq. (21) is negative given D A R A in
Ross' sense since Ro < R for the capital requirement constraint (2) to be binding.
In the absence of interest rate risk (i.e., /~o = RD), a one dollar increase in equity
capital increases the bank's profit by R + (Ro - R ) / K in all possible states. With
DARA, the bank becomes less risk averse and thus is willing to grant more risky
loans by lowering the loan rate. However, the second term on the right-hand side
of Eq. (21) is likely to be positive and thus works against the first term. 21 This
countervailing force arises because the enlarged capital base allows the bank to
issue more variable rate deposits while still satisfying the capital adequacy
requirement. This increases the variability of the bank's profit which in turn
induces the bank to take on less risky loans by raising the loan rate. Hence, the
aggregate income effect is a priori indeterminate unless interest rate risk is not a
significant factor for the bank spread decision. To summarize, we establish the
following proposition.
Proposition 8. If the bank's utility function exhibits DARA in Ross' sense and if
the interest rate risk is not severe, then an increase in the bank's equity capital
will decrease the optimal bank interest margin. Otherwise, the effect is ambiguous.
Ho and Saunders (1981) find that interest margins of small and large banks are
significantly different with the average spread of small banks approximately
one-third of one percent more than that of large banks. Ho and Saunders attribute
this difference to the fact that small banks are usually more regionally oriented and
thus possess greater market power as compared to large money-center banks. 22
Proposition 8 says that even if the market power factor is controlled for, a large
bank is likely to have a narrower spread as long as the bank can hedge away most
of the interest rate risk.
21 The second term on the right-hand side of (21) is positive, say, when Ro(O) is linear in 0. See the
appendix.
22This reasoning is consistent with Proposition 3.
K.P. Wong/ Journal of Banking & Finance 21 (1997) 251-271 265
We have assumed hitherto that credit risk is independent of the loan rate
charged by the bank. As argued by Stiglitz and Weiss (1981), this is a rather
strong assumption when incentive problems either in the form of adverse selection
or in the form of moral hazard are present in the loan market. With adverse
selection, an increase in the loan rate is likely to attract only higher risk borrowers
who preceive a lower probability of repaying the loan. It is because these
borrowers are less affected by the higher interest rate as compared to credit-worthy
borrowers. This implies that the average 'riskiness' of those who borrow increases
with an increase in the loan rate. Similarly, with moral hazard, the behavior of
borrowers is likely to change as the loan rate changes. For instance, a higher loan
rate induces firms to undertake projects with lower probability of success but
higher payoffs when successful. This again implies that credit risk increases with
an increase in the loan rate.
To model the incentive problems h la Stiglitz and Weiss (1981) into our
framework, we assume that the CDF of 0 shifts in the sense of first-order
stochastic dominance as the loan rate changes. Specifically, we consider a family
of CDFs of 0 indexed by R L, {/~(.[RL)}, such that
< 0, (22)
OR L
for all 0 ~ [0, 1]. Condition (22) says that an increase in R L increases the
likelihood that higher values of 0 will be realized and thus gives rise to a
deterioration in borrowers' credit-worthiness.
Replacing F with /~(. IRL), the first-order condition (7) can be written as
oE[u( * *)l
OR L
+ f of ,. ,. . . . . . L(R )
OP(OIRi *)
+ O) O] dOd ( ) = 0, (23)
OR L
at R~ and assume that F(O)= F(OIR~ ) (a ceteris paribus assumption), then the
first term on the left-hand side of Eq. (23) vanishes by Eq. (7). This implies that
0E[U(~" *)]/aR r < 0 . Thus, we have R~ * <R~ which follows from the
second-order condition of the new maximization problem taking the incentive
problems into account.
OP(OIR~ *)
+ R ' o ( 0 ) D] d 0 d G ( , ) > 0, (24)
ORL
where z~ = 0 2 E [ U ( ~ * * )]/OR 2 < 0 by the second-order condition. Thus, condi-
tion 24 creates a countervailing force which works against the original income
effect (given DARA in Ross' sense). However, as long as the incentive problems
are not severe in the sense that OF(OIRL)/OR L is small, we should expect the
comparative static results of the previous section to remain intact under imperfect
information.
6. Conclusions
funds rate) on the optimal spread is ambiguous and depends on the bank's net
position in the interbank market. Furthermore, the bank's equity capital is nega-
tively related to the spread when interest rate risk is trivial. These results are
largely supported by the empirical evidence of Ho and Saunders (1981), McShane
and Sharpe (1985) and Angbazo (1995) concerning bank spread behavior.
Acknowledgements
Appendix A. Proofs
A. 1. Proofof Proposition 2
Let V be the bank's new utility function. Then, V can be represented as Eq.
(13). Evaluating the first partial derivative of E[V(#)] with respect to R L at R L ,
we get
OE[V(~*)]
ORL =E{[AU'(~r*)+H'('7r*)IM(O)E(R;)
where the second inequality follows from J < 0, E < 0 and condition 11. The
desired result then follows directly from the first- and second-order conditions
under V. []
According to Ross (1981), U exhibits DARA in Ross' sense if, and only if,
there exists a positive constant y such that
(A.3)
By DARA in Ross' sense, condition A.2, we have
V " [ ~ * (0, ~)]
>Y. (A.4)
Multiplying -U"[zr *(0, E)] to both sides of inequality A.4 and taking expecta-
tions with respect to G yields
E[U"(~r*)fO] > - T E [ U " ( ~ r * ) I O ] . (A.5)
Again, from condition A.2, U exhibits DARA in Ross' sense implying that
V"[~* (0, , ) ]
- U ' [ T r * ( 0 , , ) ] -<Y" (A.6)
Multiplying -U'[zr*(O,e)]/E[U'(~r*)rO] to both sides of inequality A.6 and
taking expectations with respect to G yields
E[U"(~*)I0]
- E[U'(~*)]0] < y" (A.7)
Multiplying both sides of inequality A.14 by -U"['n" *(0, E)]e and taking expec-
tations on both sides with respect to G yields
>_
>--
270 K.P. Wong / Journal of Banking & Finance 21 (1997) 251-271
where the second inequality follows from inequalities A. 1 land A.13. Hence, the
expression inside the curly brackets of Eq. (A.10) is positive.
Since Z'(O) < O, we have
Z(O)<(>)Z(O) for 0>(<)3, (A.15)
where 0 solves M(t~)=0. Multiplying E[U'(#*)IO]M(O) to both sides of
inequality A.15 and taking expectations with respect to F yields
where the second equality follows from Eq. (7). The desired result then follows
from Eq. (19). []
To prove this proposition, it suffices to show that the second term on the
right-hand side of Eq. (21) can be positive. Suppose that Eq. (3) is a linear
regression model, i.e., R p(O)~ oz + tO, where a and /3 are positive constants.
Then, E[U"(~" * )M(O)(R o - Ro)] can be expressed as
/3 E[ U"(~'* )M((9) ((9- (9)] +/3 E[ U"(~r* ) M ( 0)] ((~- 0)
References
Allen, L., 1988, The determinants of bank interest margins: A note, Journal of Financial and
Quantitative Analysis 23, 231-235.
Allen, L., S. Peristiani and A. Saunders, 1989, Bank size, collateral, and net purchases behavior in the
federal funds market: Empirical evidence, Journal of Business 62, 501-515.
Angbazo, L.A., 1995, Commercial bank net interest margins, default risk, interest-rate risk and off
balance sheet banking, Journal of Banking and Finance, forthcoming.
Davis, G.K., 1989, Income and substitution effects for mean-preserving spreads, International Eco-
nomic Review 30, 131-136.
Finn, W.T. II and J.B. Frederick, 1992, Managing the margin, ABA Banking Journal 84, 50-53.
Hancock, D., 1986, A model of the financial firm with imperfect asset and deposit elasticities, Journal
of Banking and Finance 10, 37-54.
Ho, T.S.Y. and A. Saunders, 1981, The determinants of bank interest margins: Theory and empirical
evidence, Journal of Financial and Quantitative Analysis 16, 581-600.
K.P. Wong / Journal of Banking & Finance 21 (1997) 251-271 271