Size and Performance of Banking Afirms - Testing The Predictions of Theory
Size and Performance of Banking Afirms - Testing The Predictions of Theory
Size and Performance of Banking Afirms - Testing The Predictions of Theory
of Monetary
Economics
31 (1993) 47-67.
North-Holland
of banking
firms
of theory
1. Introduction
In recent years, two important literatures on the theory of the banking firm
have developed. Both predict relationships
between the size of banking firms
and their performance. This study tests predictions of the theories.
One substantial
literature deals with deposit insurance and the effect that it
has on bank decisions. A fundamental
finding is that the U.S. system of deposit
insurance produces an incentive for insured banking firms to take risk. Theoretically, in fact, this distortion pushes them to corner solutions taking as much risk
as they can (for example, through the use of financial leverage). With this
approach
banks are viewed. essentially, as portfolios of risky claims. Their
production technologies are unimportant,
and size plays no role in the theory. If
regulatory treatment were the same for insured banks of all sizes, this theory
would predict no relationship
between size and performance.
Correspondencr
to. John H. Boyd. Research Department,
P 0. Box 291, Minneapohs,
MN 55480-0291. USA.
Federal
0304-3932~93,$05.00
m(_1993-Elsevter
Sctence Pubhshers
highly levered. Their greater use of leverage is consistent with one prediction of
deposit insurance theory, a nonmarket distortion. Finally, we find no evidence
of a positive relationship between size and market valuation as represented by
Tobins 4. Such a relationship is predicted by both theories, due to either cost
efficiencies (modern intermediation theory) or a size-related subsidy (deposit
insurance theory). In fact, during the second half of the sample period, 1981-90,
size and Tobins y are significantly inversely correlated.
The empirical tests employ data for 122 banking holding companies over the
period 1971.-90. The sample is restricted to firms that are large by industry
standards, those whose shares are listed and actively traded. This is not a representative sample, and admittedly our findings could be different for small
banking firms.
The rest of the study proceeds as follows. Section 2 discusses the two theory
literatures. Section 3 examines the relation between this study and the literature
on economies of scale in banking. Section 4 considers issues in measurement,
explains the performance and risk indicators used, and describes a conjectured
industry equilibrium. Section 5 presents the empirical results. Section 6 considers whether our results may be influenced by systematic differences in market
power among sample firms. Section 7 concludes.
2. Theory
Flannerys (1989) model exhtbtts decreasmg returns to rusk-takmg, due to regulatory feedback m
the form of capital reqmrements. ln thts particular model, the incentive to mcrease the risk of farlure
may be hmtted. depending on parameter values There IS also some evidence that banks with high
charter values (for example. because of then ability to earn monopoly rents) may be relatively less
willing to take risks so as to exploit the deposit insurance subsrdy [Keeley (199O)J Benveniste, Boyd,
and Greenbaum
(1989) argue that thts constramt
on moral hazard may have been what kept the
FDIC intact unttl the late 1970s. when mcreased competitton
substanttally
reduced charter values.
than it is for others. For brevity, we do not include a formal proof, but the logic
is simple: too big . . . banking firms receive free insurance on their (technically)
uninsured deposits and other liabilities. Other banking firms dont. This asymmetric treatment is defended on the grounds that banking authorities fear the
possible macroeconomic
consequences
of permitting
a large banking firm to
default on its liabilities. OHara and Shaw (1990) find that public announcement
of commitment
to such a policy has had a favorable effect on the share prices of
the too big .
banks.
Ceteris might not be pwibus with respect to regulation of very large banking
firms. The authorities have repeatedly stated that they are especially concerned
about disruptions
of the largest banks, since these may result in systemic effects
(negative externalities). In fact, such concerns are the rrrison d&re of the too big
to fail policy. Based on these statements, one might logically expect regulators
to be more conservative
in setting risk constraints
on large banking firms ~ for
example, by requiring less risky asset holdings and less financial leverage. Then
if the true equilibrium
were one in which all firms are at regulatory
corner
solutions as predicted by the deposit insurance theory, empirical tests would
reveal an inverse relationship
between size and failure risk. Such regulatory risk
constraints
would also reduce the subsidy component
in deposit insurance so
that the net effect of differential treatment by size would be unclear. Even so,
deposit insurance theory does make a testable prediction:
Prediction 1. Either large (too big to fail) banking firms are less likely to fail
than small ones, or they are more highly subsidized per dollar of assets by
government
insurance, or both.
Obviously,
Prediction
1 does not preclude the possibility that large banking
firms will be less likely to fail than small ones urrd less subsidized. The conjecture
about regulatory risk constraints
is also testable.
Co?qecture. As a result of differential
less likely to fail than smaller banking
_.a.
7 7 Moderri intertmhltiorl
regulation,
firms.
large banking
firms will be
theor!.
J.H. Boyd and D.E. Runkle. Size and performance of banking firms
51
Prediction 2. Large banking firms will be less likely to fail and more costefficient than small banking firms.
This issue is
52
of scale literature
of bankmgjrtns
53
p(E<
-E)=p(r<k)=
$(r)dr.
(1)
Note that under this definition unq banking firm, whether or not it is too big to
fail, has failed when f < k. That is what we shall mean by failed from here
onward. The risk indicator is an estimate of the probability of failure which, like
Tobins q, depends on the 4(Y) distribution.
If Y is normally distributed, eq. (1)
may be rewritten as
p(r<k)=
5 N(O,l)d=,
z = (k - p)/a,
(2)
(3)
54
J.H
inhstry
equilihriw?l
55
4.2. Market
data estinutes
where Em = market value of equity (average price per share times average
number of shares outstanding)
and LB = accounting
(book) value of total
liabilities. Profits are estimated as
D,)/P,-
1)
where P = price per share of common stock, D = dividends per share, t = time
period, and N = average number of shares outstanding.
Our estimate of the rate
of return on assets is simply R = nm/Am.
The Z-score estimates depend critically upon the volatility of returns. For
these estimates, market return measures are also employed, since it seems sure
that BHCs accounting
profits are highly smoothed.
Finally, Tobins q is
estimated by Am/(LB + EB), where EB is the accounting
value of equity. Any
attempt to measure Tobins q is subject to measurement
error because of
difficulties in evaluating both the total market value of the firms assets and the
The assumed mdustry structure IS obvtously very sample. But It is consistent with some stylized
facts m that tt results m an equthbrmm distrtbutton
of BHC stzes. depending on srze of market. Of
course, if T > 0 and S > 0, thts structure predtcts there will be one BHC per geographrc market,
which is inconsistent
with the facts. However. the observed structure
of the Industry may be
significantly influenced by anti-trust policies.
For each firm m the sample we computed the standard deviation of the rate of return on equtty
with both accounting and market data. The mean (median) standard deviation of the rate of return
on equity for all firms was 0.051 (0.035) wtth accountmg data and 0.289 (0.278) wtth market data.
Clearly, market returns are much more volattle than accounting returns. Z-scores estimated with the
accounting
data are implausible, predicting farlure rates of essenttally zero. This is entirely attributable to the low standard deviations with the accounting
data. Simtlar evtdence of smoothing
accounting
profits has been reported elsewhere [e g.. Greenawalt
and Sinkey (1988)].
replacement cost of those assets. However, these problems are likely to be less
severe for banks than for manufacturing
firms for two reasons. First, banks issue
little long-term debt. The overwhelming
majority of their liabilities are shortterm deposits. For such deposits, book value is a close approximation
to market
value. Therefore, for banks, the sum of the market value of equity and the book
value of liabilities is likely to be a good approximation
to the market value of
total assets. Second, relatively few bank assets are plant and equipment. Therefore, the major deviation of asset book value from replacement cost is likely to
occur in the loan and bond portfolios. To the extent that asset values (especially
loan values) are inaccurate at any date, this will be reflected in sample estimates
of Tobins q. Note, however, that generally accepted accounting procedures can
only drfer capital gains and losses; they cannot make them vanish. Eventually,
these must be realized. and when that occurs, accounting
asset values are
adjusted accordingly.
Here we investigate
relative values of q over a long
(20-year) period. This surely helps to reduce, if not totally eliminate, measurement error from this source. Measurement
errors are potentially
much more
severe in estimating
the risk indicators
S and Z-score. Estimates
of those
statistics, however, rely strictly on market data.
4.3. Sample bank lzolditzy companies
The annual data in this study include the years 1971-90 and come from
Standard
and Poors COMPUSTAT.
This source provides both accounting
data and stock prices for publicly traded firms. We do not include all the BHCs
that are in the COMPUSTAT
data base. For one thing, not all BHCs have data
in all sample periods, and we require that sample firms have a minimum
of
five consecutive years. In addition, we exclude the smallest BHCs in the COMPUSTAT data base. those with average total assets below $1 billion. BHCs in
this size range typically are not publicly traded, and the COMPUSTAT
BHCs
in this size range are too few to provide reliable statistics. Finally, for those
BHCs that would have failed except for government assistance, we eliminated all
data from the first year of such assistance and thereafter. We justify this action
on the grounds that we are interested in market behavior. There are nine such
firms in the sample, including some very large ones. These are too big to fail
BHCs which actually did fail in the sense of eq. (1)
The so-called survivor bias is a widely recogmzed problem with the COMPUSTAT
data set.
Firms which are acquired. fail, or have their securities dehsted are dropped from COMPUSTAT.
At
any pomt m time. therefore. the firms mcluded m this data set are not necessarily representative
of
their Industry, but are the survtvors. Admittedly, that is true of our sample, but the special regulatory
treatment afforded to bankmg firms helps to attenuate the problem The FDIC has almost never
liquidated banks or BHCs as large as those in our sample. Instead, it infuses them wtth new capital
and reorgamzes
them. When the reorgamration
mvolves an acqutsition
by a stronger firm. the
failing bank loses its identity and will subsequently
be removed from COMPUSTAT.
Often.
0.137
(0.139)
on equtty, R,
8. Return
0.140
(0.136)
0.0056
(0.0057)
0.066
(0.063)
0.019
(0.017)
4.05
(4.21)
0.148
(0.149)
0.0072
(0.0059)
0.07 1
(0.067)
0.019
(0.017)
4.61
(4.19)
1.006
(1.003)
3.16
(3.13)
38
Cross-section
averages (medians) of tndivldual firms statisttcs.
bSignificance level from xi test. Test is against null hypothesis of no significant
0.0055
(0.0047)
on assets, R
0.061
(0.060)
0.017
(0.016)
4.31
(3.99)
7. Return
-K
devtatton of R, S
5. Standard
6 Equttyiassets,
(R + K)/S
4. Z-score,
1.004
(1.001)
I .ooz
11.QW
3. Tobms 9
24
1.98
(1.991
9.73
(3.86)
2. Assets ($ bk), A
_. .._
companies,
1971-90
(full sample period)
-..
_-.
1
sample means (medtans
difference
0.128
(0.132)
in group
0.0037
(0.0039)
0.053
(0.052)
0.015
(0.015)
4.44
(3.96)
0.995
(0.993)
5.61
(5 37)
27
means.
0.129
(0.132)
0.0039
(0.0030)
0.052
(0.045)
0.016
(0.013)
3.99
(3.78)
1.003
tt.OW
26.3
(15.2)
33
(IV) Over
$8 bil.
--
Table
of sample bank holding
122
All
firms
___-.
of firms
1. Number
Variable
-___-
Performance
0.603
0.009
O.OOO
0.054
0.282
0.014
Sigmficance
1eveP
in parentheses).
0.151
(0.162)
0.0065
(0.0064)
(~~~
0.016
(0.015)
4.90
(4.52)
(1,001)
1.005
13.93
(5.31)
122
All
firms
-..--.-___
1981-90
(second half)
~____
-~
58
After deletions, 122 BHCs remain. This is a small sample for an industry that
includes thousands of firms, and it does not include any firms with total assets of
less than $1 billion. Admittedly, findings could be different if these smaller firms
were included. Fortunately
there are still great differences in the sizes of firms
included: the largest sample BHC has total assets about 100 times greater than
the smallest. Table 1 shows the distribution
of sample firms, grouped into four
size classes. The largest size category (IV) includes 33 firms with average assets of
$26.3 billion. These are some of the largest BHCs in the United States, and
presumably most of them have been viewed as too big to fail by the authorities.
At the other extreme, the smallest size category (I) includes 24 firms with average
assets of about $2 billion. Based on official statements, these BHCs are not in the
too big to fail category.
Also shown in table 1 are performance and risk statistics for the sample firms,
grouped in several ways. The first column shows sample means and medians for
all firms in the entire 20-year period, 1971-90, followed by a breakdown by size
class. The last column shows sample statistics for all firms in the second half,
1981-90.
Table 2 shows the results of tests in which the performance and risk indicators
are regressed on BHC size, represented by In(A), the natural logarithm of total
assets. Both cross-section and panel results are displayed. However, the statistics
S and Z-score can only be obtained intertemporarily.
This is done once for each
firm, using the full sample period or whatever smaller number of observations
is
available.
Then the individual
firm statistics are regressed against In(A) in
cross-section. Panel data tests include dummy variables for time periods, but for
brevity those coefficients and t-statistics are not reported. In the panel data tests,
t-statistics
are adjusted to account for random firm-specific
effects. Crosssection regressions employ a correction for conditional
heteroskedasticity
using
Whites (1980) method.
J.H. Boyd and D.E. Runkle, Size and performance qf banking firms
59
Table 2
Performance
Dependent
Slope coefficient
of size. In(A)
variable
1. Tobins 4
2. Z-score,
(R + K)$
3. Standard
deviation
4. Equity/assets,
regressed
of R. S
-K
5. Return
on assets. R
6. Return
on equity,
R,
data,
1971-90,
122 firms.
t-statistic
Sample
sizeb
Type of
regression
- 0.0012
- 0.0004
1.07
0.36
122
2029
Cross-section
Panel
- 0.0904
n,a
0.77
n/a
122
n/a
Cross-section
Panel
~ 0.0022
n/a
4.02**
n/a
122
n/a
Cross-section
Panel
- 0.0082
- 0.0075
5 40**
5 21**
132
2029
Cross-section
Panel
- 0.0011
- 0.0007
3.63**
1.39
122
1907
Cross-section
Panel
- 0.0111
- 0.0073
2.47*
1.06
122
1907
Cross-section
Panel
t-statistics
are against the null hypothesis
that the slope coefficient is zero. **(*) indicates
significantly different than zero at 99% (95%) confidence.
bin panel data tests, sample size is smaller with R or R, the dependent variable than it is with
Tobms 4 dependent. Computation
of returns requires differencing, and that results in the loss of the
first sample date.
Panel data regressions include dummy variables for the time period. For brevity, these coefficients and t-statistics are not reported. In these regressions, r-statistics are also adjusted to account
for random firm-specific effects. Cross-section
tests employ a correction
for conditional
heteroskedasticity
using Whites (1980) method.
60
Row 1 in table 2 shows a negative relationship between size and Tobins q. not
significantly different than zero at even 90% confidence.13 Row 2 suggests that
there is no meaningful
relationship
between size and Z-score. However, row
3 displays an inverse relationship
between size and S, the standard deviation of
the rate of return on assets, which is significantly different than zero at a high
confidence level. The ratio of equity to assets, -K, is negatively related to size at
a high significance level, and that result is obtained in both cross-section
and
panel data tests (row 4). The rate of return on assets, R, is also negatively related
to size, and significance levels are high in the cross-section regression, but not in
the panel regression (row 5). However, the cross-section results may be spurious
because, on average, the smaller banks have fewer observations
from the first
half of the sample, during which R was lower, on average, for all banks. Finally,
the rate of return on equity, R,, appears to be weakly negatively related to size
(row 6).
Table 3 shows the results of the same regressions as in table 2, but estimated
with data from the last half of the sample period, 198 l-90. There are two reasons
to look at the subperiod.
First, a considerable
number of firms entered the
sample well after 1971. Thus, the second subperiod has many more firms (20
more) and less missing data than the first. Therefore, inference with this sample
is more robust. In addition, it is worth investigating
whether the regression
results are sensitive to choice of time period. In general, comparison
of the
regressions in table 2 and table 3 suggests thats not so. Results in both tables are
very similar when the dependent
variable is the Z-score (row 2), equity/assets
( - K) (row 4), and rate of return on assets (R) (row 5). As previously mentioned,
the probable cause for these differences is that small banks have many missing
observations
in the first half of the sample, when the average value of several of
the dependent variables was substantially
different from the average value in the
second half of the sample.
Both tests also suggest a negative relationship between size and S, the standard
deviation of R (row 3). However, it appears that this relationship got stronger over
time, in terms of both the regression coefficient and the r-statistic. The same is true
of the negative relationship between size and Tobins q. In the case of Tobins q,
the change is quite marked. The fact that the results generally differ little between
the two time periods shows that the results are robust with respect to sampling
difference. These sampling differences also explain why size does not explain
return on equity in the cross-section during the second half of the sample.
However. this IS entirely due to the difference between intertor size group III and the other groups,
and can be dtsmtssed as economtcally
unimportant.
(The other stze groups dtsplay mean and medtan
values of Tobins y whtch are almost tdenttcal in table 1.)
r3Keeley (1990) also reports findmg no stgmficant relattonship between size and Tobins q in tests
conducted with a stmrlar sample of large BHCs. However, his regresstons include several addttional
explanatory
vartables bestdes ttme pertod and firm stze.
61
Dependent
regressed
Slope coefficient
of size, In(A)
variable
1. Tobins 4
(R + K)/S
3. Standard
deviation
4. Equity/assets.
-K
5. Return
on assets, R
6. Return
on equity,
R,
1981-90,
121 firms.
Sample
size
Type of
regressionb
3.39**
2.13*
122
1160
Cross-section
Panel
0.17
n/a
122
n/a
Cross-section
Panel
- 0.0032
n/a
6.45**
n/a
122
n/a
Cross-section
Panel
- 0.0101
- 0.0088
6.36**
5.25**
122
1160
Cross-section
Panel
- 0.0010
- 0.0013
2.3-P
2.14*
122
1142
Cross-section
Panel
- 0.0056
- 0.0130
0 71
1.38
122
1142
Cross-section
Panel
0.03 16
nla
of R, S
data,
t-statisti?
- 0.0038
- 0.0028
2. Z-score.
at-statistics
are against the null hypothesis
that the slope coefficient is zero. **(*) Indicates
sigmficantly different than zero at 99% (95%) confidence.
Panel data regressions mclude dummy variables for the time period. For brevity, these coefficients and t-statistics are not reported. In these regressions, f-statistics are also adlusted to account
for random firm-specific effects. Cross-section
tests employ a correction
for conditional
heteroskedasticity usmg Whites (1980) method.
in a moment.
First, we examine
some data on
JMon-
62
Table 4
Bank failure rates by size class. 1971-91.
Asset size of banksh
Small (assets less than $1 bilhon)
Large (assets of $1 btllton or more)
1971-80
0.56
2 74
1981-91
1971-91
9.9 I
IO.45
1026
16.15
Table 4 shows that the large-bank failure rate was consistently higher than the
small-bank
failure rate. That is true over the full period 1971-91 and over both
subintervals.
For both size classes, failure rates were much higher in the 1980s
than in the 1970s; this is not surprising since the banking industrys problems in
the 1980s have been widely documented. ls
identification
of BHC afhhation of failed banks Where atlihation is obvious. we treat multiple bank
failures as a smgle fatlure This 15 done m Just two cases, both m Texas. This adJuatment (mvolvmg large banks) tends to understate
the large-bank
failure rate relative to the small-bank
fatlure
rate
Work by Kuester and OBrien (1990) supports some of the results reported here Uamg data for
225 BHCa, they find that the standard deviation ofasset returns is negatively related to size. at a high
significance
level They also directly estimate the value of government
msurance.
usmg the
BlackScholes
put-pricmg equation, In their tests, the option value of government
msurance is not
related to size at any reasonable sigmficance level. A study by OHara and Shaw (1990). however,
suggests that equitv mvestors may value too big to fail status. They examme the announcement
of
the too big to fali pohcy by the Comptroller
of the Currency m 1984 Usmg an event testmg
approach, they find that the announcement
resulted m positive short-run wealth effects for shareholders of some large BHCs
63
64
J.H
qf barking firm
65
Table 5
Size class comparisons.
means (medians
m parentheses)
Variable
--~
~-
Number
of firms
(I) Under
$1.5 bd.
total assets)
~__
(III) $2 bil.$3.5 bd.
~___.
(IV) Over
$3 5 bll
-
27
35
21
1.12
(1.13)
I .80
(1.76)
2.56
(3.34)
Tobins q
0.99 1
(0.987)
0.990
(0.986)
0.997
(0.996)
1000
(0 996)
Return
0 0050
(0.0041)
0.0020
(0.0022)
0.0022
(0.0020)
0.0025
(0.0030)
Total assets. A
on assets. R
Second half
Size class (average
Variable
____
1. Number
~-~
(I) $1 b1l.m
$3 bll
of firms
27
(II) $3 bll$5. bd
29
28
17.40
(8.85)
(1981-90)
total assets)
(III) $5 b&
$15 bll
35
~~~__
~_~
(IV) Over
$15 bll.
~__~__
30
2. Total assets. A
2 39
(2.47)
3.84
(3 6.5)
3. Tobms q
1.010
(1.008)
(1.007)
(0.999)
1002
(0 998)
0.0070
(0.0072)
0.0066
(0.0066)
0.0049
(0.0038)
4 Return
on assets. R
__
0.0070
(00061)
__
1007
~___~~~~_
8.24
(7.57)
1001
40.6
(25.5)
7. Conclusion
66
ofharking.firms
References
Allen. Frankhn. 1990, The market for mformatton and the origin of financial mtermediation,
Journal
of Fmancial Intermediation
1. no 1, May.
Benveniste, Larry. John H Boyd, and Stuart I. Greenbaum,
1989. Bank capital regulatton (ASSOCIatton of Reserve City Bankers, Washington,
DC).
Bernanhe. Ben S and Mark Gertler, 1989. Agency costs, net worth, and busmess fluctuations,
American Economic Review 79. 14-31.
Boyd, John H. and Stanley Graham.
1991. Investigating
the banhmg consolidatton
trend. Federal
Reserve Bank of Minneapolis
Quarterly
Review 15, 3-l 5.
Boyd, John H and Edward C Prescott, 1986. Fmancial intermediary-coalitions.
Journal of Economtc Theory 38, 2 1 l-232.
Chart. Yuk-Shee. Stuart Greenbaum.
and AnJan Thakor. 1992. Is fatrly priced depostt insurance
posstble?, Journal of Fmance 47, 127-245.
Clark, Jeffrey A, 1988. Economies of scale and scope at depository financtal mstitutions: A review of
the literature, Federal Reserve Bank of Kansas City Economic Revtew 73. 16-33.
Diamond. Douglas W., 1984. Fmancial Intermediation
and delegated monitormg.
Revtew of Economtc Studies 51. 393414.
Federal Reserve Bank of New York, 1974, Annual report (New York. NY).
Flannery. Mark J . 1989, Capttal regulation and Insured banks choice of mdividual loan default
risks. Journal of Monetary Economtcs 24. 2355258
Frtedman. Milton and Anna J. Schwartz. 1963, A monetary history of the United States, 186771960
(National Bureau of Economtc Research,Princeton
Umverstty Press. Prmceton. NJ)
Greenawalt.
M.B. and J.F. Smkey. Jr. 1988. Bank loan-loss pro&tons
and the mcome-smoothmg
hvnothesis
An empirical analvsis. 1976-84, Journal of Fmancial Services Research 1, 301-318.
Greenbaum.
Stuart and Anjan Thakor, 1991, Contemporary
financtal intermedtatton,
Manuscript
(Northwestern
Umversity. Evanston. IL).
Humphrey, David B.. 1990. Why do estimates of bank scale economies differ?. Federal Reserve Bank
of Richmond Economic Review 76. 38-50.
Kareken, John H. and Neil Wallace, 1978. Deposit msurance and bank regulatton
A partialequilibrium exposition, Journal of Busmess 51. 413438.
Keeley, Michael C.. 1990. Deposit insurance,
risk, and market power m banking.
American
Economic Revtew 80, 11X3-1200.
Kuester, Kathleen A. and James M. OBrien. 1990. Market-based
deposit msurance premiums: An
evaluation.
Manuscrtpt
(Board of Governors,
Federal Reserve System. Washington,
DC).
Merton, Robert C.. 1977. An analytic derivation
of the cost of deposit insurance
and loan
guarantees: An apphcatton
of modern option pricmg theory. Journal of Banking and Fmance 1.
3311
OHara. Maureen and Wayne Shaw. 1990, Deposit msurance and wealth effects: The value of bemg
too big to fad, Journal of Fmance 45. 1587-1600.
Ramakrtshnan,
Ram and Anjan V. Thakor, 1984. Informatton
reliability and a theory of financtal
mtermedtatton,
Revtew or Economtc Studies 51. 415431.
Sharpe. Wtlham F., 1978. Bank capital adequacy, deposit insurance and security values. Journal of
Financial and Quantttative
Analysts 13, 701-718.
Smirlock, Michael, Thomas Gilligan, and William Marshall. 1984. Tobms q and the structureperformance
relationshtp,
American Economic Review 74, 1051~~1060.
Stevens. Jerry L.. 1990, Tobins q and the structureeperformance
relationshtp
Comment. American
Economtc Revtew 80, 618-613
Whtte. Halbert, 19X0, A heteroskedasttctty-conststent
covartance matrix estimator and a dtrect test
for heteroskedasttctty.
Econometrtca
48, X17-838.
Wtlhamson. Stephen D.. 1986. Costly monttormg, financtal mtermedtatton.
and equihbrtum
credtt
rattonmg. Journal of Monetary Economtca 18. 1599179