Profitability and Marketability of the Top

55 U.S. Commercial Banks

Lawrence M. Seiford • Joe Zhu
Department of Mechanical & Industrial Engineering, University of Massachusetts at Amherst,
Amherst, Massachusetts 01003
Department of Management, Worcester Polytechnic Institute, Worcester, Massachusetts 01609
seiford@ecs.umass.edu • jzhu@wpi.edu

U tilizing recent developments in data envelopment analysis (DEA), this paper examines
the performance of the top 55 U.S. commercial banks via a two-stage production process
that separates profitability and marketability. Substantial performance inefficiency is uncov-
ered in both dimensions. Relatively large banks exhibit better performance on profitability,
whereas smaller banks tend to perform better with respect to marketability. New context-
dependent performance measures are defined for profitability and marketability which
employ a DEA stratification model and a DEA attractiveness measure. When combined with
the original DEA measure, the context-dependent performance measure better characterizes
the profitability and marketability of 55 U.S. commercial banks. The new approach identifies
areas for improved bank performance over the two-stage production process. The effect of
acquisition on efficiency and attractiveness is also examined.
(Data Envelopment Analysis (DEA); Profitability; Marketability; Efficiency; Banks)

1. Introduction puts, has proven itself as a metric for measuring bank

Performance analysis of financial institutions, partic-
ularly commercial banks, has received increased atten-
tion over the past several years. As a result, the focus
has moved from attempts to characterize performance
in terms of simple ratios (ROA or ROI) to a multidi-
mensional systems perspective. Although accounting
and financial ratios provide important and useful
information for benchmarking a bank’s financial per-
formance, there are, in fact, many factors relative to
bank performance, e.g., assets, revenue, profit, market
value, number of employees, investments, and cus-
tomer satisfaction, etc., and, obviously, a better under-
standing of the relationship among these various
factors would provide the key for improving bank
productivity.
Data envelopment analysis (DEA), a mathematical
programming approach for characterizing the rela-
tionships among multiple inputs and multiple out-
performance. Numerous applications of DEA have
appeared in the bank performance literature. For
example, Sherman and Gold (1985) use DEA to eval-
uate the performance of branches of a U.S. savings
bank. Other studies on bank branch efficiency include
Vassiloglou and Giokas (1990), Oral and Yolalan
(1990), and Sherman and Ladino (1995), who evaluate
bank branch efficiency in Greece, Turkey, and the
United States, respectively. Other DEA studies have
compared performance among different banks. For
example, Charnes et al. (1990) provide an illustrative
application of DEA to U.S. commercial banks in the
time period 1980 –1985, and Barr et al. (1993, 1994)
examine U.S. bank failures over the period 1984 –1988.
Also, the quality, profitability, and productivity of
banking services have been studied (Grifell-Tatje and
Lovell 1999, Soteriou and Zenios 1999) and models
proposed for assessing the market and cost efficiency

0025-1909/99/4509/1270$05.00
Management Science/Vol. 45, No. 9, September 1999 Copyright © 1999, Institute for Operations Research
pp. 1270 –1288 and the Management Sciences
 SEIFORD AND ZHU
Profitability and Marketability of the Top 55 U.S. Commercial Banks
of large-scale bank branch networks (Athanassopou-
los 1998). Berger and Humphrey (1997) provide a
review on 130 studies of financial institution including
banks, bank branches, savings and loans, credit
unions, and insurance companies. Finally, see Harker
and Zenios (1999) for recent studies on efficiency of
financial institutions.
Most bank performance studies employ labor, inter-
est (non-interest) expenses, office space, and number
of accounts as inputs, and number of transactions,
interest (non-interest) income, total loans, and depos-
its as outputs. While these inputs and outputs can
characterize a bank’s operational performance, they do
not reflect the market valuation or performance of a
bank’s stock. To evaluate stock marketability, one may
include such additional factors as market value, earn-
ings per share, and return to investors in a DEA
analysis. By incorporating some new factors in a
two-stage approach, the current paper explores prof-
itability and (stock) marketability of 55 U.S. commer-
cial banks in 1995.
A two-stage production process is defined that
generates profit in the first stage, and market value in
the second stage. The effect of bank size on profitabil-
ity and marketability is revealed by evaluating both
technical and scale efficiencies. Increasing returns to
scale (IRS) are found among relatively small banks—
banks with less than $2,000 million in assets in our
data set—and decreasing returns to scale (DRS)
among the larger banks. However, some of the larger
banks exhibit IRS in profitability. An examination of
input congestion indicates that a reduction in current
input levels may actually increase revenue and profit
levels. Several bank acquisitions occurred in early
1996. This allows a performance study on the effect of
bank acquisition on profitability and marketability. It
is shown that marketability is, at most, slightly af-
fected by bank acquisition. The present study also
indicates serious inefficiencies in both profitability
and marketability and offers a procedure to improve
the bank production process by focusing on profitabil-
ity and scale efficiency. In addition to a standard DEA
analysis, the paper also measures context-dependent
bank performance for different efficiency levels. This
context-dependent DEA model allows (i) a bank to
Management Science/Vol. 45, No. 9, September 1999
benchmark itself against its competitors; (ii) the mea-
surement of the relative attractiveness in profitability
and marketability of each bank against the back-
ground of a group of banks operating at a different
efficiency level; and (iii) the further study of the effect
of bank acquisitions on the performance of other
banks.
The remainder of the paper is organized as follows.
Section 2 defines the bank production process and its
corresponding production factors (inputs and out-
puts). Section 3 describes the DEA tools employed and
the resulting empirical findings for the 55 banks. Our
conclusions appear in § 4 along with suggestions for
future investigation.
2. Bank Production Process
Fifty-five U.S. commercial banks appear in the Fortune
1000 (Fortune April 29, 1996). 1 Obviously, ranking by
revenue while necessary for inclusion in the Fortune
1000 does not adequately characterize the perfor-
mance of these banks. Fortune also provides other
factors: e.g., number of employees (employees), assets,
stockholders’ equity (equity), market value (MV), total
return to investors (TRI), and earning per share (EPS)
to further characterize the performance of the top
companies. 2 As pointed out by Zhu (1999), any single
performance measure based on these eight factors will
be unsatisfactory in its performance characterization;
with several measures one gets conflicting reports and
it is difficult to determine which company exhibits a
better overall.
While Fortune’s analysis is based on each single
1
The banks are available from the authors or from Fortune April 29,
1996.
2
Revenues are interest and non-interest revenues and are for the
fiscal year ended on or before January 31, 1996. Profits are after
taxes, after extraordinary credits or charges, and after cumulative
effects of accounting charges. Assets are the company’s year-end
total. Equity is the sum of all capital stock, paid-in capital, and
retained earnings at the company’s year-end. MV is obtained by
multiplying the number of common shares outstanding by the price
per common share as of March 15, 1996. EPS for each company is the
primary earnings per share that appear on the income statement.
TRI includes both price appreciation and dividend yield to an
investor in the company’s stock. For more detailed explanations, see
Fortune April 29, 1996.
1271
 SEIFORD AND ZHU
Profitability and Marketability of the Top 55 U.S. Commercial Banks
Figure 1 Bank Production Process
measure, the current study illustrates the value of a
performance measure that operates simultaneously
with multiple dimensions. To fully reconcile these
eight factors and have a more complete portrayal of
the performance of these 55 banks utilizing these
different factors, Figure 1 describes a bank production
process based on these eight factors.
The process is divided into two stages and the eight
factors are expressed as inputs and outputs in each
stage. The first stage (Stage 1) measures profitability,
i.e., a bank’s ability to generate the revenue and profit
in terms of its current labor, assets, and capital stock.
The second stage (Stage 2) measures (stock) market-
ability, i.e., a bank’s performance in the stock market
by the revenue and profit it generates. It can be seen
that revenue and profit serve as intermediate factors in
the sense that they are outputs from the first stage and
inputs to the second stage.
The DEA inputs and outputs selected are based on
Fortune’s original choice of factors for performance
characterization. However, we will note that other
factors, e.g., market value to book value, price to
earnings ratio, turnover ratio (ratio of traded value to
year-end market value), etc., may also be used to
capture marketability. Unfortunately, Fortune does not
1272
provide data on these factors. Therefore, we restrict
our analysis to the eight Fortune factors as described in
Figure 1. These are sufficient to illustrate our approach
to explaining bank performance while attempting to
disentangle marketability from profitability. Valuable
insights result from such a change in perspective.
3. Measuring Profitability and
Marketability of Commercial
Banks
In this section, we analyze the profitability and mar-
ketability of 55 U.S. commercial banks. To actually
characterize the performance of these 55 banks, it is
necessary to extend the basic DEA methodology. For
an introduction to the basic DEA models and theoret-
ical extensions, readers are referred to Ali and Seiford
(1993b) or Charnes et al. (1994).
3.1 Characterization of Bank Performance
The output-oriented (CCR) DEA model is employed
to measure efficiency in profitability and marketability
(Charnes et al. 1978).
冉冘 冊
冘s
m s
⫺ ⫹
max ␾ ⫹ ␧t
o s ⫹
i r t ⫽ 1, 2
i⫽1 r⫽1
Management Science/Vol. 45, No. 9, September 1999
 SEIFORD AND ZHU
Profitability and Marketability of the Top 55 U.S. Commercial Banks

whereas most of the CCR-efficient DMUs for the Stage

冘 ␭x ⫹s
n
⫺
s.t. j ij i ⫽ x io i ⫽ 1, 2, . . . , m;
j⫽1
冘 ␭y
n
j rj ⫺ s⫹
r ⫽ ␾ o y ro
t
r ⫽ 1, 2, . . . , s;
j⫽1
␭ j, s ⫺ ⫹
i , s r ⭓ 0,
where x ij and y rj are the amount of the ith input
consumed and the amount of the rth output produced
by the jth DMU (or bank), respectively. In Stage 1 (t
⫽ 1), we have n ⫽ 55 DMUs (banks), i ⫽ 3 inputs
⫺ employees, assets and equity, and r ⫽ 2 outputs
⫺ revenue and profit. In Stage 2 (t ⫽ 2), we have n
⫽ 54 DMUs (banks), i ⫽ 2, inputs ⫺ revenue and
profit, and r ⫽ 3 outputs ⫺ MV, TRI, and EPS. 3
Let ␾ o1* and ␾ o2* be the optimal values for Stage 1 and
Stage 2 models respectively. If ␾ o1* ⫽ 1 and all
input/output slacks are zero, then a bank is said to be
CCR-efficient in profitability. If ␾ o2* ⫽ 1 and all input/
output slacks are zero, then a bank is said to be
CCR-efficient in marketability.
Table 1 reports the CCR efficiency scores. Only six
banks, namely, DMU1 (Citicorp), DMU5 (J. P. Morgan
& Co.), DMU10 (Bankers Trust New York Corp.),
DMU16 (Wells Fargo & Co.), DMU17 (Bank of New
York Co.), and DMU30 (MBNA), are CCR-efficient in
profitability. Six banks, namely, DMU10, DMU13
(PNC Bank Corp.), DMU40 (Fifth Third Bancorp),
2 marketability model are among the last 16 banks.
The average CCR scores for the first 30 banks and the
remaining 25 banks are respectively 1.28 and 1.46 in
profitability versus 1.69 and 1.34 in marketability. This
may indicate that larger banks are better at generating
profits, while the smaller banks are more accurately
(1) valued.
Overall, only 11% of the banks were efficiently
operating under profitability and marketability. Fig-
ure 2a– b give the distributions of CCR efficiency
scores for Stage 1 and Stage 2, respectively. Potential
improvements of 20% to 60% in revenue and profit are
indicated for nearly 71% of the banks, and a range of
20% to 60% improvement in MV, TRI, and EPS is
expected for approximately 45% of the banks. On
average, banks had a relatively better performance
with respect to profitability.
Scale efficiency has long been recognized in the
banking literature as an important issue. However, the
CCR Model (1) assumes constant returns to scale
(CRS). In order to determine the scale efficiency of
these 55 banks, we employ the output-oriented BCC
DEA model which allows variable returns to scale
(VRS) (Banker et al. 1984).
max ␥ ot ⫹ ␧ 冉 冘s ⫹冘s
m

i⫽1
⫺
i
s

r⫽1
⫹
r 冊 t ⫽ 1, 2

DMU48 (First Security Corp.), DMU52 (First Empire

冘 ␭x ⫹s
n
State Corp.), and DMU54 (BayBanks), are CCR- ⫺
s.t. j ij i ⫽ x io i ⫽ 1, 2, . . . , m;
efficient in marketability. A single bank, DMU10 ex- j⫽1
hibits best practice in both dimensions. Since the

冘 ␭y
DMUs are sorted by revenue, the first 31 banks are n

among the Fortune 500 with assets greater than $2,000 j rj ⫺ s⫹

r ⫽ ␥ o y ro
t
r ⫽ 1, 2, . . . , s;
million (with one exception—DMU30 –MBNA). It can j⫽1

be seen that all CCR-efficient DMUs for the Stage 1

冘 ␭ ⫽1
n
profitability model are among the first 30 banks,
j
j⫽1
3
EPS data for one bank (Bank 34, Southern National Corp.) was
negative. Therefore this bank was excluded from the DEA analysis
␭ j, s ⫺ ⫹
i , s r ⭓ 0. (2)
of Stage 2 resulting in 55 DMUs in Stage 1 and 54 DMUs in Stage 2.
Although one could invoke the translation invariance property of
DEA (Ali and Seiford 1990) to allow negative values, such use is
Let ␥ o1* and ␥ o2* be the optimal values for (2) in Stage
restricted to the convex case, and some properties, e.g., returns to 1 and Stage 2, respectively. Define a scale efficiency
scale (RTS), are not translation invariant. measure by

Management Science/Vol. 45, No. 9, September 1999 1273

 SEIFORD AND ZHU
Profitability and Marketability of the Top 55 U.S. Commercial Banks

Table 1 Efficiency Results

Profitability Marketability Adjusted

DMU No. Bank Name CCR BCC RTS CCR BCC RTS CCR

1 Citicorp 1.00000 1.00000 CRS 2.22629 1.00000 DRS 1.96759

2 BankAmerica Corp. 1.46572 1.19242 DRS 2.02745 1.00000 DRS 2.67492
3 NationsBank Corp. 1.25852 1.23178 DRS 2.01639 1.04762 DRS 2.24049
4 Chemical Banking Corp. 1.14565 1.07208 DRS 2.22441 1.17976 DRS 2.24530
5 J. P. Morgan & Co. 1.00000 1.00000 CRS 1.88763 1.07401 DRS 1.62636
6 Chase Manhattan Corp. 1.19998 1.18324 IRS 1.97889 1.16520 DRS 2.13593
7 First Chicago NBD Corp. 1.26820 1.26812 IRS 1.99399 1.24469 DRS 2.22424
8 First Union Corp. 1.34203 1.26789 DRS 1.74746 1.01410 DRS 2.10488
9 Banc One Corp. 1.42400 1.17498 DRS 1.75796 1.05878 DRS 2.24649
10 Bankers Trust New York Corp. 1.00000 1.00000 CRS 1.00000 1.00000 CRS 1.00000
11 Fleet Financial Group 1.34885 1.34882 IRS 1.35231 1.09968 DRS 2.09335
12 Norwest Corp. 1.41056 1.13822 DRS 1.62002 1.09350 DRS 2.15575
13 PNC Bank Corp. 1.46868 1.44546 IRS 1.00000 1.00000 CRS 2.50310
14 KeyCorp 1.40077 1.33349 DRS 1.95201 1.40564 DRS 2.43276
15 Bank of Boston Corp. 1.13527 1.09816 IRS 2.04048 1.19025 DRS 2.06496
16 Wells Fargo & Co. 1.00000 1.00000 CRS 1.53265 1.00000 DRS 1.33729
17 Bank of New York Co. 1.00000 1.00000 CRS 1.79339 1.23352 DRS 1.54366
18 First Interstate Bancorp 1.24357 1.17181 DRS 1.39599 1.00000 DRS 1.50272
19 Mellon Bank Corp. 1.33620 1.13285 DRS 1.82595 1.24302 DRS 2.14184
20 Wachovia Corp. 1.32595 1.31020 IRS 1.58118 1.23491 DRS 1.82137
21 SunTrust Banks 1.52661 1.52475 DRS 1.42458 1.15455 DRS 1.89208
22 Barnett Banks 1.48538 1.42508 DRS 1.81902 1.44995 DRS 2.34092
23 National City Corp. 1.55531 1.43373 DRS 1.92081 1.66344 DRS 2.63457
24 First Bank System 1.14793 1.10864 IRS 1.35522 1.06841 DRS 1.33801
25 Comerica 1.35078 1.35024 DRS 1.75552 1.33088 DRS 1.93687
26 Boatmen’s Bancshares 1.57613 1.56623 IRS 1.66734 1.39892 DRS 2.29205
27 U.S. Bancorp 1.52131 1.46983 IRS 1.36872 1.20952 DRS 1.77726
28 CoreStates Financial Corp. 1.29265 1.28448 IRS 1.55757 1.33732 DRS 1.74606
29 Republic New York Corp. 1.21513 1.00000 IRS 1.79015 1.69502 DRS 1.82236
30 MBNA 1.00000 1.00000 CRS 1.10393 1.03421 DRS 1.00000
31 State Street Boston Corp. 1.26003 1.25722 IRS 1.28135 1.26093 DRS 1.39079
32 First of America Bank Corp. 1.62366 1.61784 DRS 1.55994 1.49311 DRS 2.01993
33 Northern Trust Corp. 1.28003 1.09901 IRS 1.34935 1.28464 DRS 1.31137
34 Southern National Corp. 1.49424 1.31987 IRS
35 Firstar Corp. 1.51298 1.43026 IRS 1.29746 1.29659 IRS 1.07937
36 Huntington Bancshares 1.36592 1.19881 IRS 1.41447 1.37485 DRS 1.58138
37 SouthTrust Corp. 1.45287 1.33217 IRS 1.56641 1.56620 DRS 1.74594
38 Crestar Financial Corp. 1.51322 1.30351 IRS 1.33815 1.33491 DRS 1.44925
39 Amsouth Bancorp. 1.41893 1.17595 IRS 1.36419 1.33486 DRS 1.34549
40 Fifth Third Bancorp 1.18494 1.00000 IRS 1.00000 1.00000 CRS 1.00467
41 Mercantile Bancorp. 1.41364 1.21397 IRS 1.56240 1.50667 DRS 1.72484
42 Meridian Bancorp 1.38069 1.14637 IRS 1.03884 1.01601 DRS 1.00000
43 Marshall & Ilsley Corp. 1.61165 1.45907 IRS 1.51110 1.49380 IRS 1.38028
44 First Tennessee National Corp. 1.33328 1.00000 IRS 1.37990 1.37936 IRS 1.00167
45 Banponce Corp. 1.67449 1.51418 IRS 1.67664 1.66202 IRS 1.36842
46 UJB Financial 1.51301 1.23857 IRS 1.43904 1.43747 IRS 1.06925
47 Integra Financial 1.50398 1.19804 IRS 1.03590 1.00000 IRS 1.00000
48 First Security Corp. 1.65955 1.43364 IRS 1.00000 1.00000 CRS 1.19419
49 Regions Financial 1.51619 1.19075 IRS 1.50995 1.46555 IRS 1.34299
50 Signet Banking Corp. 1.28587 1.00000 IRS 1.52596 1.00000 IRS 1.64113
51 Old Kent Financial Corp. 1.42787 1.04675 IRS 1.36141 1.23668 IRS 1.23456
52 First Empire State Corp. 1.36812 1.00000 IRS 1.00000 1.00000 CRS 1.00000
53 Bancorp Hawaii 1.52981 1.00000 IRS 1.43523 1.00000 IRS 1.26220
54 BayBanks 1.60314 1.21369 IRS 1.00000 1.00000 CRS 1.00000
55 Union Planters Corp. 1.51155 1.01584 IRS 1.50077 1.00000 IRS 1.00000

1274 Management Science/Vol. 45, No. 9, September 1999

 SEIFORD AND ZHU
Profitability and Marketability of the Top 55 U.S. Commercial Banks
Figure 2a CCR Efficiency Distribution in Stage 1
Figure 2c BCC Efficiency Distribution in Stage 1
␾ ot*
␲ ot ⫽ , for t ⫽ 1, 2
␥ ot*
Obviously, ␲ ot 肁 1. If ␲ ot ⫽ 1, a bank is called
scale-efficient; otherwise, if ␲ ot ⬎ 1, a bank is called
scale-inefficient. Table 1 gives ␥ ot* for each bank when
we employ (2) to measure profitability and market-
ability. Twelve and thirteen DMUs were BCC-efficient
in Stage 1 and Stage 2, respectively. Figures 2c–d give
the distribution of the BCC scores ( ␥ ot* ) for each stage.
Note that ␲ ot ⫽ 1 if and only if ␥ ot* ⫽ ␾ ot* . In our case,
within the 55 banks, only those banks that were CCR
efficient are scale-efficient. A paired-difference t-test
was applied to CCR and BCC scores in each stage. The
results of the t-test were significant, indicating that a
serious scale inefficiency was present for the 55 com-
mercial banks in both profitability and marketability.
(For profitability and marketability, t values are equal
to 7.6 and 6.8, respectively.)
We next determine whether increasing or decreas-
Management Science/Vol. 45, No. 9, September 1999
Figure 2b CCR Efficiency Distribution in Stage 2
Figure 2d BCC Efficiency Distribution in Stage 2
ing returns to scale (IRS or DRS) is the primary cause
of scale inefficiency. As shown in Banker (1984), the
optimal solution for ␭ *j ( j ⫽ 1, . . . , n) in (1), i.e., the
magnitude of ¥ j⫽1 n
␭ *j , contains the information for RTS
classification. However, because of the possibility of
multiple optimal lambda solutions for (1), DMUs may
be misclassified with respect to RTS by ¥ j⫽1 n
␭ *j (see
Banker and Thrall 1992). To avoid the misclassifica-
tion, we use the recent result of Zhu and Shen (1995)
to determine the RTS classification. 4 That is, let DMU o
be a bank under evaluation and ␭ *j be an optimal
solution to (1) associated with ␾ ot* , then CRS prevail
for DMU o if and only if ␥ ot* ⫽ ␾ ot* , i.e., ␲ ot ⫽ 1;
otherwise, if ␥ ot* ⫽ ␾ ot* , i.e., ␲ ot ⬎ 1, then IRS prevail for
4
An alternative approach is to use Banker and Thrall’s (1992) RTS
interval obtained by calculating two auxiliary BCC-dual-like linear
programming problems. However, the two methods are equivalent
and our approach is easier. See Seiford and Zhu (1999a) for the
discussion on alternative RTS approaches.
1275
 SEIFORD AND ZHU
Profitability and Marketability of the Top 55 U.S. Commercial Banks
DMU o if and only if ¥ j⫽1 n
␭ *j ⬍ 1, and DRS prevail for
DMU o if and only if ¥ j⫽1 ␭ *j ⬎ 1. Using this method,
n
one does not have to worry about possible misclassi-
fication errors from multiple optimal solutions for ␭ *j ,
and the RTS classifications are readily obtained from
the optimal solutions to (1) and (2).
Table 1 indicates that, for profitability, the BCC-
efficient but not scale-efficient banks were operating on
an IRS frontier. For marketability, four BCC-efficient
banks were operating on each of the IRS and DRS
frontiers, respectively. Of the BCC-inefficient banks, 64%
and 20% were in the IRS region in Stage 1 and Stage 2,
respectively. As economists have long recognized, an
IRS frontier firm would generally be in a more favorable
position for expansion, compared to a firm operating in
a CRS or DRS region. Furthermore, on the basis of Stigler
(1966), U.S. commercial banks were more economically
viable in profitability than marketability.
Note that the concept of RTS may be ambiguous
unless a bank is on the BCC-efficient frontier, since we
classified RTS for inefficient banks by their output-
oriented BCC projections. Thus, a different RTS clas-
sification may be obtained for a different orientation,
since the input-oriented and the output-oriented BCC
models can yield different projection points on the
VRS frontier. Thus, it is necessary to explore the
robustness of the RTS classification under the input-
oriented DEA method.
Note that an IRS DMU (under the output-oriented
DEA method) must be termed as IRS by the input-
oriented DEA method (see Appendix A for the proof).
Therefore, one only needs to check the CRS and DRS
banks in the current study. Using Zhu and Shen’s (1995)
input-oriented approach, we discover that only three
DRS banks in profitability (DMUs 21, 25 and 32) and
twelve DRS banks in marketability (DMUs 23, 26, 28, 29,
31, 32, 33, 36, 37, 38, 39 and 41) are termed as IRS DMUs
for the input-oriented RTS method. These results indi-
cate that (i) in general, the RTS classification under
profitability is independent of the orientation of DEA
model; and (ii) there are serious output deficiencies in
marketability at the current revenue and profit levels.
The findings from most previous empirical research
on scale efficiency in banking indicate that IRS occurs
only among relatively small banks, and DRS among
1276
larger banks. However, the recent study of MacAllis-
ter and MacManus (1993) demonstrates that these
previous results were biased by problems in the
statistical techniques employed and by the fact that
the models ignored an important input—financial
capital. Our current study shows that for profitability,
some relatively large banks with assets greater than
$2,000 million exhibited IRS, whereas for marketabil-
ity, all of these large banks were in the DRS region.
3.2 Input Congestion in Profitability and
Marketability
Input congestion, an economic concept that originated in
the areas of transportation and agriculture, refers to the
situation where reductions in the usage of a proper
subset of inputs, holding the usage of other inputs
constant, generate an increase in one or more outputs.
For example, “too much fertilizer applied to a given plot
could easily reduce overall output” (Byrnes et al. 1984).
After an inefficient bank is projected onto the frontier by
a proportional (radial) increase of outputs or a propor-
tional decrease in inputs, the input congestion measure
provides information about the effect on output im-
provement of further individual input reduction.
In DEA, input congestion is usually investigated
under assumptions of strong and weak input dispos-
abilities. However, as shown in Ray et al. (1998), input
congestion can be measured by nonzero input slacks
when all frontier DMUs are extreme-efficient. 5 Since
the CCR Model (1) and BCC Model (2) respectively
yield unique optimal solutions for efficient DMUs, all
of the CCR and BCC frontier banks are extreme-
efficient. Thus, input congestion can be determined by
the following linear programming problem. 6
冘s
m
c
max i
i⫽1
s.t. 冘 ␭x ⫹s ⫽␪ x
j僆E
j ij
c
i
t*
io i ⫽ 1, . . . , m;
5
A complete classification of efficient and inefficient DMUs is
defined in Charnes et al. (1991).
6
See Cooper et al. (1999) for the situation when some frontier DMUs
are weakly efficient.
Management Science/Vol. 45, No. 9, September 1999
 SEIFORD AND ZHU
Profitability and Marketability of the Top 55 U.S. Commercial Banks
冘 ␭y
j僆E
j rj ⭓ y ro r ⫽ 1, . . . , s;
␭ j ⭓ 0, j 僆 E,
where E is the set of all CCR-extreme-efficient banks
and ␪ t* represents the optimal value for the input-
oriented CCR model (or, equivalently, the reciprocal
of the optimal value ␾ ot* in (1)). If the optimal value to
(3) is equal to zero, input is not congested; otherwise,
if the optimal value to (3) is not equal to zero, input
congestion is present. Furthermore, if a specific s ic*o
⫽ 0, then the i o th input is a cause of input congestion
and s ic*o gives the amount of congestion.
Model (3) measures input congestion for a CRS
technology. If one appends the additional constraint,
¥ j僆E ␭ j ⫽ 1, to (3), an input congestion measure for a
VRS technology is obtained. In this case, ␪ t* should be
replaced by the optimal value to the input-oriented
BCC model and E should represent the set of all
BCC-extreme-efficient banks.
Table 2 reports the nonzero input slacks, s ic* . It is
obvious from Table 2 that a serious input congestion
exists in profitability. Between a third to a half of the
banks exhibited input congestion under CRS and VRS
technologies. The input employee was not a factor in
causing input congestion for the CRS technology.
Thus, labor does not appear to be overutilized under a
CRS assumption. However, for a VRS technology, the
amount of employee congestion is 18.36% of the
corresponding employee input level.
In marketability, a very different picture is portrayed.
With the exception of revenue congestion for four large
banks, input congestion is essentially absent.
3.3 Process Improvement
The analyses of the previous section indicated which
banks were efficient and which were inefficient. For
each stage, inefficient banks are able to improve their
performance and the DEA projections provide a pre-
scription for improvement. For example, DMU1 (Citi-
corp) was CCR-inefficient in Stage 2; however, it can
move its performance to best-practice by either (a)
increasing its MV, TRI, and EPS (proportionally by
123%), or (b) decreasing its revenue and profit (pro-
portionally by 55% and some additional nonzero
profit slack). If Citicorp chooses to improve its perfor-
Management Science/Vol. 45, No. 9, September 1999
mance with Plan (b), then it may no longer be efficient
in Stage 1 profitability. In addition, a bank may not
have direct control over all Stage 2 outputs, i.e., its
(3) performance in the stock market. Thus, to improve
performance over Stage 1 and Stage 2, we focus on
maximizing revenue and profit. However, increases in
revenue and profit may affect the scale efficiency in
Stage 2. Figures 3a– b illustrate this with employees
(input) and profit (output) in Stage 1, and profit
(input) and market value (output) in Stage 2.
If maximizing profit is a major goal, then a bank will
increase its current profit level to point S (Figure 3a). As
mentioned earlier, being positioned in an IRS region is
ideal for economic viability. Therefore, one should avoid
the situation (illustrated in Figure 3) that a profit increase
(AS) in Stage 1 would move the bank into a DRS region
in Stage 2, i.e., AS should not be greater than UV. If AS
⬎ UV, then one may wish to investigate alternative
approaches to performance improvement in Stage 1. For
example, one may move the bank onto the best practice
point G by increasing the profit to P and then by
reducing the number of employees to G.
We formalize this preference for maintaining viabil-
ity while improving a bank’s profitability as:
Process Improvement Rule: The increase in revenue
and profit for Stage 1 improvement should not move
an IRS bank in Stage 2 into a DRS region.
It can be seen that to implement the process im-
provement rule, we must determine an IRS stability
region which preserves the IRS positioning of a bank.
As in Seiford and Zhu (1999b), we formulate the
following linear programming problem for an IRS
bank, DMU o , in Stage 2
冘 ␭˜
n
共 ␶ *o 兲 ⫺1 ⫽ max j
j⫽1
冘 ␭˜ x ⭐ x
n
s.t. j ij io i ⫽ 1, 2, . . . , m;
j⫽1
冘 ␭˜ y
n
j rj ⭓ ␾ o2* y ro r ⫽ 1, 2, . . . , s;
j⫽1
␭˜ j ⭓ 0. (4)
where ␾ o2* is the optimal value to (1) for DMU o in
Stage 2.
1277
 SEIFORD AND ZHU
Profitability and Marketability of the Top 55 U.S. Commercial Banks

Table 2 Input Congestion in Profitability and Marketability

Profitability (Stage 1) Marketability (Stage 2)

Employees Assets Equity Revenue Profit

DMU
No. Bank Name BCC CCR BCC CCR BCC BCC CCR BCC

2 BankAmerica Corp. 13721.28 203.77 1961.03

3 NationsBank Corp. 121.40 599.93 17239.53 1425.96
4 Chemical Banking Corp. 5587.14 666.65
5 J. P. Morgan & Co. 3089.34
6 Chase Manhattan Corp. 369.55 387.67 1043.07
7 First Chicago NBD Corp. 6580.5 2431.26
8 First Union Corp. 1626.79 5363.77 12454.39
9 Banc One Corp. 10760.95 892.05
12 Norwest Corp. 15138.72 9931.92 4881.1
13 PNC Bank Corp. 55.83 47.15
14 KeyCorp 499.50 2017.48
15 Bank of Boston Corp. 49.93 41.47
16 Wells Fargo & Co. 50.14
18 First Interstate Bancorp 4746.91 8021.54 5964.45
19 Mellon Bank Corp. 4089.54 322.81 655.12
22 Barnett Banks 559.73 946.44
23 National City Corp. 755.2 1660.45
25 Comerica 486.12
27 U.S. Bancorp 39.33 31.72
28 CoreStates Financial Corp. 630.92
29 Republic New York Corp. 303.88
31 State Street Boston Corp. 3622.47 4908.75
32 First of America Bank Corp. 732.86
33 Northern Trust Corp. 703.57
34 Southern National Corp. 52.19 50.53
35 Firstar Corp. 648.81
37 SouthTrust Corp. 2032.32 2239.8
40 Fifth Third Bancorp 259.82
42 Meridian Bancorp 60.96 26.61
43 Marshall & Ilsley Corp. 1683.31 75.26
44 First Tennessee National Corp. 1907.96
45 Banponce Corp. 612.87 182.83
46 UJB Financial 265.04
47 Integra Financial 27.11 1.32
48 First Security Corp. 1324.28
49 Regions Financial 11.95
50 Signet Banking Corp. 12.29
51 Old Kent Financial Corp. 2.26
52 First Empire State Corp. 805.11
53 Bancorp Hawaii 4.54
55 Union Planters Corp. 40.20

Because DMU o exhibits IRS in Stage 2, the optimal tion continues to hold for ␩ 僆 { ␩ : 1 聿 ␩ ⬍ ␶ *o }, where
value to (4) must be less than one, i.e., ␶ *o ⬎ 1. As ␩ represents the proportional increase of all inputs, x̂ io
shown in Seiford and Zhu (1999b), the IRS classifica- ⫽ ␩ x io (i ⫽ 1, . . . , m) and ␶ *o is defined by (4). (See

1278 Management Science/Vol. 45, No. 9, September 1999

 SEIFORD AND ZHU
Profitability and Marketability of the Top 55 U.S. Commercial Banks

Figure 3 Process Improvement

Appendix A for the proof.) Table 3 reports the value of Table 3 RTS Sensitivity Results for IRS Banks in Stage 2
␶ *o for the 11 IRS banks in Stage 2.
DMU No. Bank Name Change Rate
By this sensitivity analysis of RTS classification, we
can implement the process improvement rule as follows. 35 Firstar Corp. 1.01630
Step 1. For an IRS bank, DMU o , in Stage 2 which is 43 Marshall & Ilsley Corp. 1.12606
CCR-inefficient in Stage 1, calculate (4). First Tennessee National
(i) If ␾ o1* 聿 ␶ *o , improve profitability performance by 44 Corp. 1.00675
45 Banponce Corp. 1.15406
setting
46 UJB Financial 1.01941

冦
⫺*
Employees Stage1 ⫽ employees ⫺ s employees 47 Integra Financial 1.36827
⫺*
Assets Stage1 ⫽ assets ⫺ s assets 49 Regions Financial 1.14642
⫺* 50 Signet Banking Corp. 1.51128
Stage 1 Equity Stage1
⫽ equity ⫺ s equity
⫹* 51 Old Kent Financial Corp. 1.25299
Revenue Stage1
⫽ ␾ o revenue ⫹ s revenue
1*
53 Bancorp Hawaii 1.29549
⫹*
Profit Stage1
⫽ ␾ o profit ⫹ s profit
1*
55 Union Planters Corp. 1.03753

where (*) represents the optimal value in (1);

(ii) If ␾ o1* ⬎ ␶ *o , then improve the profitability
performance by computing the following input-

冘 ␭ Assets ⫹ s
oriented CCR model 55
⫺
⫽ ␪ o Assets o
␪ *o ⫽ min ␪ o
j j assets
j⫽1

冘 ␭ Employees ⫹ s 冘 ␭ Equity ⫹ s
55 55
⫺
s.t. j j employees ⫽ ␪ o Employees o j j
⫺
equity ⫽ ␪ o Equity o
j⫽1 j⫽1

Management Science/Vol. 45, No. 9, September 1999 1279

 SEIFORD AND ZHU
Profitability and Marketability of the Top 55 U.S. Commercial Banks

聿 ␶ *o (1.28587 ⬍ 1.51128) is only satisfied by DMU50

冘 ␭ Revenue ⫺ s
55
⫹
j j revenue ⫽ ␶ *o Revenue o in 11 IRS banks. Therefore, we apply (ii) in Step 1 to
j⫽1 the remaining 10 IRS banks and apply (i) to DMU50
and all other banks. The last column of Table 1 reports

冘 ␭ Profit ⫺ s
55
⫹
the adjusted CCR scores in Stage 2 after all banks
j j profit ⫽ ␶ *o Profit o
become CCR-efficient in Stage 1.
j⫽1
Table 4 reports the adjusted efficient input-output
␭ j ⭓ 0, j ⫽ 1, . . . , 55. levels for three selected banks. DMU2 is a DRS bank in
Stage 2, therefore we use the original CCR results to
In effect, we first increase DMU o ’s current output project the current levels of employees, assets, equity,
levels by ␶ *o (moving the bank onto point P as in Figure revenue, and profit in Stage 1. Then, in Stage 2, we use
3a), then decrease the three input levels by an input- the new CCR results to adjust the levels of MV, TRI, and
oriented CCR model (moving the bank onto point G in EPS. Because no input slack is detected during CCR
Figure 3a). analysis in Stage 2, the adjusted revenue and profit levels
Finally, we obtain an efficient input-output level in in Stage 2 are the same as those in Stage 1.
Stage 1 by setting DMU35 is an IRS bank in Stage 2, and the sensi-
tivity analysis of RTS classification indicates that the

冦
⫺*
Employees Stage1 ⫽ ␪ *o employees ⫺ s employees
⫺* original CCR score is greater than ␶ *o (1.51298
Assets Stage1
⫽ ␪ *o assets ⫺ s assets
⫺*
Equity Stage1 ⫽ ␪ *o equity ⫺ s equity ⬎ 0.01630). Therefore the levels of revenue and
Stage 1
Revenue Stage1 ⫹*
⫽ ␶ *o revenue ⫹ s revenue profit are increased proportionally by 0.01630, and
Profit Stage1 ⫹*
⫽ ␶ *o profit ⫹ s profit we use the input-oriented CCR model to move
DMU35 onto the best-practice frontier in Stage 1.
Step 2. Apply case (i) to other banks and obtain the Finally, we move this Stage 1 frontier bank onto the
corresponding efficient input-output levels. best-practice frontier in Stage 2.
Step 3. Calculate (1) again with the new Revenue Stage1 DMU50 is an IRS bank in Stage 2 with ␾ o1* ⬍ ␶ *o .
and Profit Stage1 levels obtained from Steps 1 and 2 as Therefore we apply the same procedure as for
new input levels in Stage 2. We then obtain the DMU2.
following efficient input-output levels for DMU o in
3.4 The Effect of Acquisition on Profitability and
Stage 2:
Marketability

冦
⫹*
MV Stage2 ⫽ ␾ o2* MV ⫹ s MV Chase Manhattan Corp. (DMU6) and First Interstate
⫹*
TRI Stage2
⫽ ␾ o TRI ⫹ s TRI
2*
Bancorp (DMU18) were acquired by Chemical Banking
⫹*
Stage 2 EPS Stage2
⫽ ␾ o EPS ⫹ s EPS
2*
Corp. (DMU4) (March 31, 1996) and Wells Fargo & Co.
⫺*
Revenue Stage2
⫽ Revenue Stage1
⫺ s revenue (DMU16) (April 1, 1996), respectively.7 Employing vari-
⫺*
Profit Stage2
⫽ Profit Stage1
⫺ s profit ous expense ratios, Rhoades (1993) shows that during
The above steps improve the profitability and 1981–1986, horizontal bank mergers did not yield effi-
marketability and satisfy the process improvement ciency gains. For our current empirical study, we exam-
rule. Note that the only difference between the ine the potential effect of acquisitions on these banks’
Revenue Stage1 (Profit Stage1 ) and Revenue Stage2 (Profit Stage2 ) performance over profitability and marketability.
is possible nonzero input slacks in Stage 2. How- Two hypothetical DMUs, namely Bank A (DMU4
ever, on the basis of complementary slackness, only ⫹ DMU6) and Bank B (DMU16 ⫹ DMU18), are
one of revenue and profit has a possible nonzero 7
NBD Bancorp acquired First Chicago Corp. and changed its name
slack for each bank in Stage 2. From our empirical
to First Chicago NBD Corp. Also, First Union Corp., Fleet Financial
application, only a nonzero revenue slack of $53 Group, and PNC Bank Corp. acquired First Fidelity Bancorp,
million was found for DMU13 (PNC Bank Corp.). Shawmut National Corp., and Midlantic Corp., respectively. How-
From Table 3, we observe that the condition of ␾ o1* ever, data before acquisition are not available for the current study.

1280 Management Science/Vol. 45, No. 9, September 1999

 SEIFORD AND ZHU
Profitability and Marketability of the Top 55 U.S. Commercial Banks

Table 4 Adjusted Efficiency Levels for Selected Banks respectively. For example, the number of employees in
Bank A (72443) is the summation of number of em-
Adjusted Level
ployees in DMU4 (39078) and DMU6 (33365). TRI and
DMU2 BankAmerica Corp. Original Level Stage 1 Stage 2 EPS are relative numbers. To obtain an aggregated
number for TRI or EPS, we need information on total
DRS in investment and profit after taxes. Unfortunately, the
Stage current study does not have access to data on total
2 Employees 95288.00 95288.00
investment. Therefore, we use the profit levels to
Assets ($ millions) 232446.00 229124.67
Equity ($ millions) 20222.00 19923.33
combine the TRIs and EPSs respectively.
Revenue ($ millions) 20386.00 29880.00 29880.00
Profit ($ millions) 2664.00 3904.68 3904.68
TRI acquisition
j ⫽ ␮ 1j TRI j1 ⫹ ␮ 2j TRI j2 , j ⫽ A and B.
Market Value ($ millions) 27148.60 72620.33
Total Returns to Investors (%) 69.40 190.48 where ␮ 1j ⫽ (Profit j1/Profit j1 ⫹ Profit j2) and ␮ 2j ⫽ (Profit j2/
Earning per share ($) 6.49 17.36 Profit j1 ⫹ Profit j2). Specifically, we consider the convex
combination of the levels of TRIs for the acquired banks.
Adjusted Level
In this case, we have ␮ 1A ⫽ 0.61, ␮ 2A ⫽ 0.39, and ␮ 1B ⫽ 0.54,
DMU35 Firstar Corp. Original Level Stage 1 Stage 2 ␮2B ⫽ 0.46. Note that the profit levels are “profits after
taxes.” Thus, we may derive the EPS for banks A and B
IRS in by
Stage
2 Employees 9263.00 6222.16 Profit j1 ⫹ Profit j2
Assets ($ millions) 19168.30 12875.76 EPS acquisition
j ⫽ , j ⫽ A and B
Profit j1 Profit j2
Equity ($ millions) 1524.80 1024.24 ⫹
Revenue ($ millions) 1740.00 1768.36 1768.36 EPS j1 EPS j2
Profit ($ millions) 228.90 232.63 232.63
Market Value ($ millions) 3393.4 3662.74
Table 5 provides the input and output data on the
Total Returns to Investors (%) 53.50 57.75 two hypothetical banks.
Earning per share ($) 3.00 3.24 We examine the effect of acquisition on profitability
and marketability for two cases. (Note that DMU16 is
Adjusted Level
a CCR-efficient bank in Stage 1.) In Case I, we keep the
DMU50 Signet Banking Corp. Original Level Stage 1 Stage 2
original CCR-frontier fixed, i.e., we keep the CCR-
efficient DMUs (including DMU16) in the reference
IRS in set. In Case II, we exclude DMU16 from the reference
Stage set. This affects the CCR-frontier in Stage 1 but not in
2 Employees 4485.00 4485.00 Stage 2, since DMU16 is a CCR-inefficient DMU in
Assets ($ millions) 11012.90 11012.90
Stage 2. Thus, for marketability, we only need to
Equity ($ millions) 886.70 870.90
Revenue ($ millions) 1145.30 1472.71 1472.71 consider Case I. Table 5 reports the results. In Case I,
Profit ($ millions) 114.30 169.85 169.85 the CCR scores of the two hypothetical banks are
Market Value ($ millions) 1454.50 2387.02 almost the same. In both stages, no obvious efficiency
Total Returns to Investors (%) 31.00 50.88 gain results from the acquisition. (Note DMU16 is one
Earning per share ($) 0.53 1.80
of the referent banks.)
In Case II, removal of DMU16 results in better profit-
ability for both banks A and B. In both cases, DMU30 is
created to represent the result of the two acquisitions.
in the reference set for the evaluation of Banks A and B.
Additivity is assumed in acquisition, i.e., the input
and output levels in Bank A and Bank B are the 3.5 Context-Dependent Performance
summations of the associated input and output levels We have identified best-practice/performance for
for DMU4 and DMU6, and DMU16 and DMU18, profitability and marketability and examined/pre-

Management Science/Vol. 45, No. 9, September 1999 1281

 SEIFORD AND ZHU
Profitability and Marketability of the Top 55 U.S. Commercial Banks

Table 5 The Effect of Acquisition on Performance

Case I Case II
BANK Chemical Banking Corp.
A (plus) Chase Manhattan Corp. New Level Stage 1 Stage 2 Stage 1

Employees 72443.00
Assets ($ millions) 304099.00 CCR Efficiency CCR Efficiency CCR Efficiency
Equity ($ millions) 21046.00 ⫽ 1.19712 ⫽ 2.12843 ⫽ 1.18188

Revenue ($ millions) 26220.00 Referent banks Referent banks Referent banks

Profit ($ millions) 2970.00 Bank 5 (1.41) Bank 13 (2.55) Bank 5 (1.34)

Market Value ($ millions) 29587.70 Bank 16 (0.33) Bank 40 (6.70) Bank 17 (0.34)
Total Returns to Investors (%) 75.17 Bank 30 (3.94) Bank 30 (4.12)
Earning per share ($) 4.38

Case I Case II
BANK Wells Fargo & Co. (plus) First
B Interstate Bancorp New Level Stage 1 Stage 2 Stage 1

Employees 46900.00
Assets ($ millions) 108387.00 CCR Efficiency CCR Efficiency CCR Efficiency
Equity ($ millions) 8209.00 ⫽ 1.19137 ⫽ 1.58192 ⫽ 1.09935

Revenue ($ millions) 10236.50 Referent banks Referent banks Referent banks

Profit ($ millions) 1917.10 Bank 16 (1.96) Bank 13 (0.06) Bank 17 (1.51)

Market Value ($ millions) 23480.50 Bank 30 (0.75) Bank 40 (6.42) Bank 30 (2.05)
Total Returns to Investors (%) 78.52 Bank 52 (0.35)
Earning per share ($) 14.64

Note. The numbers in parentheses represent optimal lambda values.

scribed improvements for individual banks. Our sug-
gestions for improvement addressed the radial dis-
tance from an inefficient bank to the best-practice
frontier. However, we have not examined the influ-
ence of similar or closely performing banks. Research-
ers of consumer choice theory point out that consumer
choice is often influenced by context. For example, a
circle appears large when surrounded by small circles
and small when surrounded by larger ones. Similarly
a product may appear attractive against a background
of less attractive alternatives and unattractive when
compared to more attractive alternatives (Tversky and
Simonson 1993). Considering this influence within the
framework of the current study, one would ask “what
is the relative attractiveness of a particular bank when
compared to others in terms of profitability and mar-
ketability?” As in Tversky and Simonson (1993), one
agrees that the relative attractiveness of bank x , com-
pared to bank y depends on the presence or absence of
a third option, say bank z (or a group of banks).
Relative attractiveness is dependent upon the back-
grounds constructed from alternative options (banks).
We first develop the following DEA technique to
construct these different backgrounds for measuring
the attractiveness of a particular bank.
Define J 1 ⫽ {DMU j , j ⫽ 1, . . . , n} (the set of all n
DMUs (banks)) and interactively define J l⫹1 ⫽ J l ⫺ E l
where E l ⫽ {DMU k 僆 J l 兩 ␾ *(l, k) ⫽ 1, and all slacks on
inputs and outputs are zero}, where ␾ *(l, k) is the
optimal value to the following linear programming
problem:
␾ *共l, k兲 ⫽ max ␾ 共l, k兲
␭ j , ␾ ,共l,k兲
s.t. 冘
j僆f共J l 兲
␭ j y rj ⭓ ␾ 共l, k兲y rk r ⫽ 1, . . . , s;

1282 Management Science/Vol. 45, No. 9, September 1999

 SEIFORD AND ZHU
Profitability and Marketability of the Top 55 U.S. Commercial Banks
冘
j僆f共J l 兲
␭ j x ij ⭐ x ik i ⫽ 1, . . . , m;
␭j ⭓ 0 j 僆 f共J l 兲,
where x ik and y rk represent the amounts of the ith
input and the ith output of DMU k , and j 僆 f(J l ) means
DMU j 僆 J l , i.e., f⵺ represents the correspondence
from a DMU set to the corresponding subscript index
set.
When l ⫽ 1, Model (5) becomes the original output-
oriented CCR model and E 1 consists of all the efficient
DMUs. When l ⫽ 2, Model (1) gives E 2 the set of
efficient DMUs after exclusion of the original efficient
DMUs. And so on. We call E l the level-/best-practice
frontier (BPF). In this manner, we identify several
levels of BPFs. 8 The following algorithm accomplishes
the identification of these BPFs by Model (5).
Step 1. Set l ⫽ 1. Evaluate the entire set of DMUs, J 1
by (1) to obtain the Level 1 efficient DMUs, set E 1 (the
Level 1 BPF).
Step 2. Exclude the efficient DMUs from future DEA
runs. J l⫹1 ⫽ J l ⫺ E l .
Step 3. Evaluate the new subset of “inefficient”
DMUs, J l⫹1 , by (1) to obtain a new set of efficient
DMUs E l⫹1 (the new BPF).
Step 4. Go to step 2.
Stopping Rule. J l⫹1 ⫽ A, the algorithm stops.
Thus employing Model (5) iteratively gives a strat-
ification of the entire set of DMUs. From the algo-
rithm, it is obvious that l goes from 1 to L, where L is
determined by the stopping rule, i.e., J L⫹1 ⫽ A. It is
easy to show that the set of n DMUs has the following
properties:
Property 1. J 1 ⫽ 艛 L
l⫽1 E l and E l 艚 E l⬘ ⫽ A for l
⫽ l⬘;
Property 2. The DMUs in E l⬘ are enveloped by the
DMUs in E l if l⬘ ⬎ l;
Property 3. Each DMU in set E l is efficient with
reference to the DMUs in set E l⫹l⬘ for all 0 ⬍ l⬘ 聿 l
⫺ L.
As described in Seiford and Zhu (1996), the sets E l
serve as evaluation backgrounds (contexts) for DMUs
8
The larger the value of l, the lower the efficiency level.
Management Science/Vol. 45, No. 9, September 1999
in E l⬘ , where l⬘ ⬍ l. The attractiveness score of a
specific DMU q ⫽ ( x q , y q ) in a specific level E l o , l o
僆 {1, . . . , L ⫺ 1} is defined as the optimal value to the
(5)
following linear programming problem
⍀ qlo共 ␣ 兲 ⫽ max ⍀ q 共 ␣ 兲 ␣ ⫽ 1, . . . , L ⫺ l o
␭ j ,⍀ q 共 ␣ 兲
冘
s.t. ␭ j y rj ⭓ ⍀ q 共 ␣ 兲y rq r ⫽ 1, . . . , s;
j僆f共E l o⫹ ␣ 兲
冘
j僆f共E l o⫹ ␣ 兲
␭ j x ij ⭐ x iq i ⫽ 1, . . . , m;
␭j ⭓ 0 j 僆 f共E l o⫹ ␣ 兲, (6)
where the integer ␣ gives the degree of attractiveness.
The optimal value ⍀ qlo (␣) is called the (output-oriented)
␣-degree attractiveness of DMUq. Obviously, the smaller
the value of ⍀ qlo (␣), the more attractive DMU q is relative
to the other DMUs. (See Seiford and Zhu 1996 for a more
complete discussion of the properties on ⍀ qlo (␣).) 9 Note
that ⍀ qlo (␣) depends on the particular evaluation back-
grounds E lo ⫹␣, and Model (6) provides a measure of
context-dependent performance.10
Tables 6a– 6b report the different levels of BPF and
the first-degree attractiveness scores (␣ ⫽ 1) for prof-
itability and marketability. First-degree attractiveness
evaluates how distinctive a bank’s performance is
compared to banks in the next lower efficiency level.
For example, first-degree attractiveness for the banks
in E 1 is determined in the context of the banks in E 2.
There are six and ten levels of BPFs for profitability
and marketability respectively.
Nine banks (DMUs 2, 34, 38, 45, 46, 47, 53, 54, and
55) are in the last BPF level for profitability. How-
ever, the original CCR scores for these last-level
banks are not all among the worst. For example,
9
Model (6) can be regarded as a CCR model in which the particular
DMU under evaluation is excluded from the reference set. This is
frequently referred to as the superefficiency DEA model. As a result,
either ⍀ qlo ( ␣ ) ⬍ 1 or (6) is infeasible. However, as shown in Zhu
(1996a) and Seiford and Zhu (1999c), (6) is always feasible.
10
It can be seen that (6) is different from the model used in ranking
efficient DMUs in (Andersen and Petersen 1993). Model (6) mea-
sures attractiveness of the efficient DMUs against inefficient DMUs,
whereas the Andersen and Petersen model for ranking efficient
DMUs measures an efficient DMU against all other DMUs including
the remaining efficient DMUs.
1283
 SEIFORD AND ZHU
Profitability and Marketability of the Top 55 U.S. Commercial Banks

Table 6a Context-Dependent Performance on Profitability

Level 1 Level 2 Level 3 Level 4 Level 5

Bank Score Bank Score Bank Score Bank Score Bank Score

1 0.84150 4 0.78011 3 0.96562 8 0.88833 13 0.95788

5 0.65790 15 0.88827 6 0.80313 9 0.88447 21 0.95007
10 0.67591 18 0.89194 7 0.90253 11 0.90669 26 0.96626
16 0.81779 19 0.88524 12 0.86922 14 0.93219 27 0.9516
17 0.80435 24 0.82672 20 0.87435 22 0.93993 32 0.92688
30 0.58251 29 0.58218 28 0.84498 23 0.94184 35 0.93535
31 0.89216 43 0.93006 25 0.90558 37 0.90979
40 0.81706 50 0.91427 33 0.87066 39 0.91815
44 0.94487 36 0.87859 42 0.93191
41 0.90757 48 0.92328
52 0.9195 49 0.92211
51 0.9523
Average 0.72999 0.83428 0.88802 0.90685 0.93713

Table 6b Context-Dependent Performance on Marketability

Level 1 Level 2 Level 3 Level 4 Level 5 Level 6 Level 7 Level 8 Level 9

Bank Score Bank Score Bank Score Bank Score Bank Score Bank Score Bank Score Bank Score Bank Score

10 0.75259 16 0.87268 11 0.86496 21 0.92260 12 0.92975 17 0.87346 5 0.93545 6 0.61235 2 0.61271

13 0.76095 18 0.87263 24 0.84421 27 0.92055 20 0.89468 22 0.67554 8 0.89467 14 0.55573 3 0.52007
40 0.66196 30 0.80994 31 0.90959 33 0.99889 28 0.88067 25 0.75769 9 0.86775 23 0.48203 4 0.44036
48 0.76172 42 0.73030 35 0.93232 36 0.93673 29 0.61548 26 0.88301 15 0.46555 7 0.43889
52 0.19053 47 0.75900 38 0.78661 41 0.73560 32 0.59321 19 0.52931
54 0.48098 55 0.74557 39 0.89101 44 0.87010 37 0.52722
45 0.64466 46 0.85899 43 0.72346
49 0.85018 50 0.76527
51 0.82074
53 0.75326
Average 0.60146 0.79835 0.82975 0.89192 0.74122 0.79743 0.73855 0.55004 0.50301

DMU48, with a CCR score of 1.65955, is in Level 5, tiveness score for DMU29 indicates a large difference
whereas, DMU2, with a CCR score of 1.46572, is in relative to the other banks in Level 2 BPF.
the last level. We observe that CCR scores alone do Similarly, for marketability, DMU52, DMU42,
not entirely characterize the performance of these DMU45, DMU41, DMU37, DMU22, DMU15, DMU23,
banks. The stratification which results from DEA and DMU7 are the most attractive banks in each of the
Model (5) provides a better indication of the struc- first 9 levels of BPF. Among the original CCR-efficient
ture of the set of banks. DMUs, DMU52 has the lowest first-degree attractiveness
For profitability, DMU30, DMU29, DMU6, DMU33, score—almost one third of the average. Relative to the
and DMU37 are the most attractive banks for each of the banks in the Level 2 BPF (evaluation background),
first 5 levels of BPF. However, the first-degree attractive- DMU52 exhibits distinctive performance on marketabil-
ness scores for the banks in Level 3, Level 4, and Level 5 ity, compared to the other CCR-efficient banks.
BPFs are very similar. In contrast, the first-degree attrac- The average first-degree attractiveness scores in

1284 Management Science/Vol. 45, No. 9, September 1999

 SEIFORD AND ZHU
Profitability and Marketability of the Top 55 U.S. Commercial Banks

Table 7a Effect of Acquisition on Attractiveness Measure for Profitability

1st-Degree Attractiveness With Acquisition 1st-Degree Attractiveness Without Acquisition

Level 1 Change
Bank Score Referent Banks Score Referent Banks (%)

1 0.81261 A (0.01), 15 (4.00), 29 (0.52), 40 (1.69) 0.84150 18 (0.54), 24 (0.18) ⫺0.0343

5 0.65790 29 (3.18) 0.65790 18 (1.16) 0
10 0.65976 A (0.01), 15 (0.46), 29 (1.00) 0.67591 15 (0.11), 18 (0.33), 19 (0.06) ⫺0.0239
17 0.80435 29 (0.38), 40 (2.18) 0.80435 4 (0.03), 24 (0.74) 0
30 0.58251 15 (0.08), 19 (0.23) 0.58251 4 (0.03), 24 (0.72) 0
Average 0.70343 0.71243

Notes. 1. Bank 24 becomes the new CCR efficient DMU due to the removal of Bank 16. The 1st-degree attractiveness score for Bank 24 is 0.86337.
2. The banks in level 2 BPF (with acquisition) are DMUs A, B, 15, 19, 28, 29, 31, 40, and 44.
3. The numbers in parentheses represent the optimal lambda values in the attractiveness measure (6).
Table 7b Effect of Acquisition on Attractiveness Measure for Marketability

2nd-Degree Attractiveness With Acquisition 2nd-Degree Attractiveness Without Acquisition

Level 1 Change
Bank Score Referent Banks Score Referent Banks (%)

10 0.67681 11 (0.21), 31 (0.35) 0.67681 11 (0.21), 31 (0.35) 0

13 0.68671 11 (0.49), 31 (0.44) 0.68671 11 (0.49), 31 (0.44) 0
40 0.55193 B (0.06), 35 (0.49) 0.60281 24 (0.33), 51 (0.33) ⫺0.0844
48 0.59269 53 (0.98) 0.59269 53 (0.99) 0
52 0.16317 49 (0.15), 53 (0.87) 0.18788 45 (0.84) ⫺0.1315
54 0.43956 49 (0.52), 53 (0.39) 0.44063 45 (0.09), 49 (0.36), 51 (0.44) ⫺0.0024
Average 0.53426 0.54938

Notes. 1. The banks in Level 3 BPF (with acquisition) are DMUs B, 11, 21, 31, 35, 38, 39, 41, 49, 51, and 53.
2. The numbers in parentheses represent the optimal lambda values in the attractiveness measure (6).

each level may be used as a measure of average radial
distance between successive BPFs. For example, the
average value of 0.72999 for Level 1 in Table 6a
indicates that on average, the two output levels (rev-
enue and profit) may be decreased approximately by
73% before the Level 1 BPF reaches the Level 2 BPF.
Thus the averages in Tables 6a and 6b portray average
performance differences of profitability and market-
ability between successive BPFs.
If we consider the second-degree attractiveness for
banks in Level 1 BPF (Stage 1), i.e., we set l 0 ⫽ 1 and
␣ ⫽ 2, then we have the optimal values to (5), 0.77653
(DMU1), 0.38302 (DMU5), 0.55309 (DMU10), 0.67744
(DMU16), 0.65317 (DMU17), and 0.54656 (DMU30).
Therefore, if Level 3 BPF is chosen as evaluation
background, DMU5 is the most attractive one in
profitability, and then DMU30. Whereas if the Level 2
BPF is chosen as evaluation background, we obtain the
opposite result on DMUs 5 and 30, i.e., DMU30 is the
most attractive bank, and then DMU5. This indicates
that different evaluation backgrounds may yield dif-
ferent results on context-dependent performance.
Note that in Model (1), deleting or adding some
inefficient DMUs does not alter the efficiencies of the
existing DMUs. However, such actions will obviously
affect the results from the attractiveness measure (6).
In § 3.3, we examined the effect of acquisition on the
performance of the banks involved in the acquisition.
Now, by (6), one is able to study the effect of acquisi-
tion on the relative attractiveness of the other banks.
As described in § 3.3, Banks 4 and 6, and Banks 16
and 18, are respectively replaced by two hypothetical
banks, A and B. From Model (5), we obtain that, in
Stage 1, A and B are two new members in the Level 2

Management Science/Vol. 45, No. 9, September 1999 1285

 SEIFORD AND ZHU
Profitability and Marketability of the Top 55 U.S. Commercial Banks
BPF. The second and third columns in Table 7a
respectively report the first-degree attractiveness
scores and the referent banks in the Level 2 BPF for the
banks in Level 1 BPF after the acquisition. For conve-
nience, Table 7a also reports the corresponding infor-
mation in Columns 4 and 5 when there is no acquisi-
tion. The last column in Table 7a gives the change in
the attractiveness scores. It can be seen that the attrac-
tiveness scores for DMUs 5, 10, and 30 remain un-
changed, while scores for DMUs 1 and 17 are slightly
decreased. Note also that the difference between
DMU1’s and DMU17’s scores becomes smaller after
the acquisitions. The acquisitions alter the context and
cause the performances of DMUs 1 and 17 to be less
distinct from each other.
In Stage 2, Model (5) concludes that Bank A belongs
to the Level 3 BPF. Therefore we use (6) to measure the
second-degree attractiveness of the banks in the Level
1 BPF, i.e., the banks in the Level 3 BPF serve as the
referent set. Table 7b reports the results. It can be seen
that the second-degree attractiveness scores for DMUs
10, 13, and 48 remain unchanged while the scores for
DMUs 40, 52, and 54 are decreased. Our interpretation
is that the acquisition of Bank 18 by Bank 16 resulted
in an increase in the attractiveness of these banks in
terms of marketability.
Finally, we point out that because of the acquisition,
i.e., deleting and adding of DMUs, the structure of the
BPFs is changed. For example, in Stage 2, DMUs 24
and 45, which were originally in Level 3 BPF, are now
members of the Level 2 BPF. As a result, the first-
degree attractiveness score for DMUs 40, 52, and 54
decreases, i.e., these banks become more attractive
compared to the other banks in the Level 1 BPF.
4. Conclusions and Future Research
The paper analyzes the profitability and (stock) mar-
ketability of the top 55 U.S. commercial banks in 1995
using an innovative two-stage DEA model. Close to
90% of the banks were inefficient in both profitability
and marketability. Further, most large banks exhibited
DRS in marketability, while some of them exhibited
IRS in profitability. This suggests that bank size may
have a negative effect on marketability. Having exam-
ined performance at both the institutional level and
1286
market level, the current study develops a procedure
to improve a bank’s profitability and marketability,
moving it onto the best-practice frontier while main-
taining a viable IRS positioning across both stages.
Our newly developed context-dependent DEA model
allows one to measure the attractiveness of a bank
against the background of its competitors. Finally, our
examination of bank acquisition for the data set indi-
cates that such activity did not affect the performance
of the banks involved in the acquisition, but did affect
the attractiveness of other banks.
The current study does not attempt to incorporate
judgment as was done in Charnes et al. (1990). The
incorporation of value judgment, e.g., introduction of
weight bounds into the dual of (1), will sharpen DEA
scores and rule out possibly unreasonable values.
Since the current study did not have access to such a
priori information and the number of efficient banks is
very small (only six out of fifty-five), we did not
employ a cone ratio or assurance region approach.
However, with the appropriate information, such an
approach would be possible. For example, one would
employ the DEA/preference structure model of Zhu
(1996b) for the two-stage bank production process
improvement procedure if a preference structure over
various production factors is available. This method
has improved flexibility for target setting since it
allows increases on inputs and decreases on outputs.
As an alternative approach to performance evalua-
tion, the current paper employs DEA to analyze the
financial performance for the Fortune 1000 (banking)
data as published by Fortune magazine. The current
study did not have access to such additional factors as
asset quality, capital adequacy, and liquidity which
could provide additional insight into the financial
performance of commercial banks. The relationships
of these dimensions to the profitability and market-
ability of banks deserves to be further studied.
In addition, at the time of this study, data for
succeeding years had not been published and the data
for prior years are incomplete. However, in future
studies, we do expect to examine performance over
time with window analysis and the Malmquist pro-
ductivity change index techniques. Such an approach
Management Science/Vol. 45, No. 9, September 1999
 SEIFORD AND ZHU
Profitability and Marketability of the Top 55 U.S. Commercial Banks

would allow a dynamic view of the banks’ profitabil- oriented DEA method, then DMU o exhibits DRS under output-
ity and marketability over time. 11 oriented DEA method.

Theorem 2. (RTS sensitivity analysis for output-oriented DEA meth-

11
The authors appreciate the helpful comments and suggestions od.) The IRS classification continues to hold for ␩ 僆 { ␩ : 1 聿 ␩ ⬍ ␶ *o },
provided by two anonymous referees. This research was partly where ␩ represents the proportional increase of all inputs, x̂ io ⫽ ␩ x io (i
supported by the Canadian Imperial Bank of Commerce. ⫽ 1, . . . , m) and ␶ *o is defined in (4).

Proof. Suppose DMU o ⫽ ( ␩ x o , y o ) and DMU o exhibits CRS or

Appendix A: Proofs ៮ ⫽ 共 ␩ x , ␩y ) must also exhibit CRS or DRS
DRS. Then DMU o o o

Theorem 1. (Relationship between the output-oriented and input- under an output-oriented DEA method.
oriented RTS classifications.) Suppose DMU o exhibits IRS under an Furthermore, we have
output-oriented DEA method. Then DMU o exhibits IRS under an

冘
n
input-oriented DEA method.
␭ *j x ij ⭐ ␩ x io i ⫽ 1, 2, . . . , m;
Proof. Since DMU o exhibits IRS under an output-oriented DEA j⫽1

method, for any optimal solution ␭ *j and ␾* to the following

output-oriented CCR model (B.1), we have ¥ j⫽1 ␭ *j ⬍ 1.
冘
n n

␭ *j y rj ⭓ ␾ * ␩ y ro r ⫽ 1, 2, . . . , s;
max ␾ j⫽1

冘 冘
n
n

s.t. ␭ jx j ⭐ x o ␭ *j ⭓ 1,
j⫽1
j⫽1

៮ .

冘
n where ␾* is the optimal value of ␾ ot in (B.1) when evaluating DMU o

␭ jy j ⭓ ␾ y o Obviously, ( ␭ *j / ␩ ) is a feasible solution to (4). Thus (¥ j⫽1 ␭ *j / ␩ )

n

j⫽1 肁 (1/ ␩ ) ⬎ (1/ ␶ *o ) violating the optimality of (4).

␭ j ⭓ 0. (B.1)
References
Now suppose DMU o exhibits CRS or DRS under input-oriented DEA Ali, A. I., Lawrence M. Seiford. 1990. Translation invariance in data
method, then for any optimal solution ␭ˆ j* and ␾ˆ * to the following envelopment analysis. Oper. Res. Letters 9 403– 405.
input-oriented CCR model (B.2), we have ¥ j⫽1
n
␭ˆ j* ⫽ 1 (or ⬎1). , . 1993a. Computational accuracy and infinitesimals in
data envelopment analysis. INFOR 31 290 –297.
min ␾ˆ , . 1993b. The mathematical programming approach to
efficiency measurement. Harold O. Fried, C. A. Knox Lovell,

冘
n Shelton S. Schmidt, eds. The Measurement of Productivity Effi-
s.t. ␭ˆ j x j ⭐ ␾ˆ x o ciency: Techniques and Applications. Oxford University Press,
j⫽1 New York.
Andersen, P., N. C. Petersen. 1993. A procedure for ranking efficient

冘
n
units in data envelopment analysis. Management Sci. 39 1261–
␭ˆ j y j ⭓ y o 1264.
j⫽1
Athanassopoulos, Antreas D. 1998. Nonparametric frontier models
␭ˆ j ⭓ 0. (B.2) for assessing the market and cost efficiency of large-scale bank
branch networks. J. Money, Credit, and Banking 2 172–192.
However, the following relationship holds for any optimal solutions Banker, R. D. 1984. Estimating most productive scale size using data
to (B.1) and (B.2). envelopment analysis. European J. Oper. Res. 17 35– 44.
, A. Charnes, W. W. Cooper. 1984. Some models for estimating
1 ␭ *j technical and scale efficiencies in data envelopment analysis.
␾ˆ * ⫽ and ␭ˆ *j ⫽
␾* ␾* Management Sci. 30 1078 –1092.
, R. M. Thrall. 1992. Estimation of returns to scale using data
Since ␾* 肁 1, therefore ¥ j⫽1
n
␭ˆ *j ⫽ (¥ j⫽1
n
␭ *j / ␾ *) ⬍ 1. A contradiction.
envelopment analysis. European J. Oper. Res. 62 74 – 84.
This indicates that DMU o exhibits IRS under input-oriented DEA
Barr, R. D., Lawrence M. Seiford, T. F. Siems. 1993. An envelopment-
method.
analysis approach to measuring the managerial efficiency of
Remark 1. Similarly, if DMU o exhibits DRS under input- banks. Ann. Oper. Res. 45 1–19.

Management Science/Vol. 45, No. 9, September 1999 1287

 SEIFORD AND ZHU
Profitability and Marketability of the Top 55 U.S. Commercial Banks
, , . 1994. Forecasting bank failure: A non-parametric
frontier estimation approach. Rech. Econom. Louvain 60 417– 429.
Berger, A. N., D. D. Humphrey. 1997. Efficiency of financial insti-
tutions: International survey and directions for future research.
European J. Oper. Res. 98(April) 175–212.
Byrnes, P., R. Färe, S. Grosskopf. 1984. Measuring productive
efficiency: an application to Illinois strip mines. Management Sci.
23 671– 681.
Charnes, A., W. W. Cooper, E. Rhodes. 1978. Measuring the efficiency
of decision making units. European J. Oper. Res. 2 429 – 444.
, , Z. M. Huang, D. B. Sun. 1990. Polyhedral cone-ratio DEA
models with an illustrative application to larger commercial
banks. J. Econometrics 46 73–91.
, , R. M. Thrall. 1991. A structure for classifying and
characterizing efficiency and inefficiency in data envelopment
analysis. J. Productivity Anal. 2 197–237.
, , A. Y. Lewin, Lawrence M. Seiford. 1994. Data Envelop-
ment Analysis: Theory, Methodology and Applications. Kluwer
Academic Publishers, Boston, MA.
Cooper, W. W., Lawrence M. Seiford, Joe Zhu. 1999. A unified
additive model approach for evaluating inefficiency and con-
gestion with associated measures in DEA. Forthcoming in
Socio-Economic Planning Sciences.
Grifell-Tatje, E., C. A. K. Lovell. Profits and productivity. Manage-
ment Sci. 45(9) 1177–1193.
Harker, P. T., S. A. Zenios. 1999. Financial Institutions: Efficiency,
Innovation, Regulations. In press, Cambridge University Press,
Cambridge, U.K.
Lewin, Arie Y., Lawrence M. Seiford. 1997. From Efficiency Calcula-
tions to a New Approach for Organizing and Analyzing Data: DEA
Fifteen Years Later. Baltzer Science Publishers, Bussum, The
Netherlands.
MacAllister, P. H., D. Macmanus. 1993. Resolving the scale effi-
ciency puzzle in banking. J. Banking and Finance 17 389 – 405.
Oral, H., R. Yolanlan. 1990. An empirical study on measuring
operating efficiency and profitability of bank branches. Euro-
pean J. Oper. Res. 46 282–294.
Ray, S., Lawrence M. Seiford, Joe Zhu. 1998. Chinese state-owned
enterprises behave as market entities: DEA and translog pro-
duction frontier approaches. OMEGA 26 263–278.
Accepted by Stavros A. Zenios; received April 1997. This paper has been with the authors for 2 revisions.
1288
Rhoades, S. A. 1993. Efficiency effects of horizontal (in-market) bank
mergers. J. Banking and Finance 17 411– 422.
Seiford, Lawrence M., Joe Zhu. 1996. Context-dependent data
envelopment analysis: Measuring attractiveness and progress.
SQPL Research Report 96-5, MIE, 219 ELAB, University of
Massachusetts, Amherst, MA.
, . 1999a. An investigation of returns to scale under data
envelopment analysis. OMEGA 27 1–11.
, . 1999b. Sensitivity and stability of the classification of
returns to scale in data envelopment analysis. J. Productivity
Anal. 12 55–75.
, . 1999c. Infeasibility of super-efficiency data envelopment
analysis models. INFOR 37(May) 174 –187.
Sexton, T. R., R. H. Silkman, A. J. Hogan. 1986. Data envelopment
analysis: Critique and extensions. Measuring Efficiency: An As-
sessment of Data Envelopment Analysis. Jossey Bass and American
Evaluation Association, San Francisco, CA. 73–105.
Sherman, H. D., F. Gold. 1985. Bank branch operating efficiency:
evaluation with data envelopment analysis. J. Banking and
Finance 9 297–315.
, G. Ladino. 1995. Managing bank productivity using data
envelopment analysis (DEA). Interfaces 25 60 –73.
Soteriou, A., S. A. Zenios. Operations, quality and profitability in
the provision of banking services. Management Sci. 45(9) 1221–
1238.
Stigler, G. J. 1966. The Theory of Price, 3 rd Ed. Macmillan, New York.
Tversky, A., I. Simonson. 1993. Context-dependent preferences.
Management Sci. 39 1179 –1189.
Vassiloglou, M., D. Giokas. 1990. A study of the relative efficiency of
bank branches: an application of data envelopment analysis. J.
Oper. Res. Soc. 41 591–597.
Zhu, Joe. 1996a. Robustness of the efficient DMUs in data envelop-
ment analysis. European J. Oper. Res. 90 451– 460.
. 1996b. Data envelopment analysis with preference structure. J.
Oper. Res. Soc. 47 136 –150.
. 1999. Multi-factor performance measure model with an appli-
cation to Fortune 500 companies. Forthcoming in European J.
Oper. Res.
, Z. Shen. 1995. A discussion of testing DMUs’ returns to scale.
European J. Oper. Res. 81 590 –596.
Management Science/Vol. 45, No. 9, September 1999