See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/388079776

Using accounting measures and data envelopment analysis to measure firm

performance: the effect of variability in prices

Article · January 2025

DOI: 10.1108/MEDAR-12-2023-2277

CITATIONS READS

0 11

4 authors:

Claire Cui Julie Harrison

Auckland University of Technology University of Auckland
6 PUBLICATIONS 1 CITATION 32 PUBLICATIONS 467 CITATIONS

SEE PROFILE SEE PROFILE

Frederick Ng Paul Rouse

University of Auckland University of Auckland
14 PUBLICATIONS 71 CITATIONS 91 PUBLICATIONS 2,113 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Claire Cui on 17 January 2025.

The user has requested enhancement of the downloaded file.

 The current issue and full text archive of this journal is available on Emerald Insight at:
https://www.emerald.com/insight/2049-372X.htm

Meditari
Using accounting measures and Accountancy
data envelopment analysis to measure Research

firm performance: the effect of

variability in prices
Claire Murong Cui Received 27 December 2023
Revised 20 July 2024
Department of Accounting, Auckland University of Technology, 7 October 2024
Auckland, New Zealand, and 9 December 2024
Accepted 9 December 2024

Julie A. Harrison, Frederick Ng and Paul Rouse

Department of Accounting and Finance, The University of Auckland,
Auckland, New Zealand

Abstract
Purpose – Recent accounting research using data envelopment analysis (DEA) measures firm performance using
accounting measures from annual reports, which are readily available from electronic databases (e.g. Demerjian et al.,
2013; Schwab, 2022). This approach differs from conventional DEA studies that analyse productivity and use internal
data about physical quantities of production inputs and outputs. Using accounting measures instead of physical
measures presents challenges as accounting measures aggregate physical quantities using unknown but fluctuating
prices. This raises the issue of what these DEA models measure. This study aims to examine how price variability
influences DEA results when measuring firm performance and identifies implications for future accounting research.
Design/methodology/approach – This study uses a Cobb–Douglas function to simulate physical data for
input and output quantities, which are then priced to form accounting measures that incorporate different levels
of price variability. These simulated accounting data are used to estimate DEA results. The results using
physical data and accounting data are compared to identify the impact of increasing levels of price variation
and sample size on the comparability of DEA results.
Findings – The study confirms the theoretical argument that accounting measures can be used in DEA to
measure productivity when prices are identical across a sample of firms. Moreover, where price variability is low,
large samples can also reliably estimate productivity when using accounting measures. This measure of
productivity fundamentally underpins financial performance and provides a new dimension of firm performance
that can be measured by accounting measures. However, where price variability is high, DEA using accounting
measures cannot estimate productivity and can only be used for benchmarking financial performance. In this case,
DEA provides an alternative measure for financial performance, which incorporates multiple dimensions and can
extend traditional financial analysis approaches by providing a more comprehensive measure.
Originality/value – Despite calls for investigation (Camanho et al., 2024; Färe et al., 2017; Zelenyuk,
2020), evidence has been scarce regarding the impact of price variability when using accounting measures in
DEA. Understanding this impact is key to understanding the nature of DEA results produced using accounting
measures, as this can affect the interpretation and use of those results. This study is the first to focus on the
impact of price variability on accounting measures within DEA and suggests new avenues for accounting
research using this performance measurement method.
Keywords Data envelopment analysis, Accounting measures, Firm performance
Paper type Research paper
Meditari Accountancy Research
The authors gratefully acknowledge the financial support received from the University of Auckland’s © Emerald Publishing Limited
2049-372X
Faculty Research Development Fund. DOI 10.1108/MEDAR-12-2023-2277
MEDAR 1. Introduction
The proliferation of large-scale databases that compile company accounting measures has
provided opportunities for accounting researchers to analyse firm performance with various
models. In response, there has been rapid growth in research using data envelopment
analysis (DEA) based on accounting measures. These DEA models provide a method that
focuses on identifying the top-performing firms, unlike regression analysis, which looks at
average performance. By comparing a firm to the best performers, DEA provides insights on
how a firm can improve its performance relative to its peers.
Leading this trend, Demerjian et al. (2012, 2013) used DEA to measure managerial
performance, and large numbers of accounting researchers have adopted this approach to
examine a variety of accounting issues [1]. Compared with conventional accounting research
tools such as ratio analysis and regressions, DEA can provide additional insights, such as
underlying productivity, which relate to firm performance and avenues for future
improvements (Schwab et al., 2022; Smith, 1990).
There is a long history of using accounting measures in DEA. When Charnes et al. (1978)
introduced DEA, based on the work of Farrell (1957), they restricted their study to physical
measures and focused on “technical efficiency”, being the “success in producing as large as
possible an output from a given set of inputs” (Farrell, 1957, p. 254). Over time, DEA studies
have used accounting measures as proxies for physical measures of production inputs and
outputs, given that accounting information is more often readily available. For example,
Sherman and Gold (1985) used rent as a proxy for physical square meters for banks when
measuring the banking production process.
The issue with accounting measures in DEA lies in the ambiguity of what is being
measured. Are these accounting measures merely proxies for physical measures, thereby
enabling physical productivity and technical efficiency to be assessed? For example, when
Schwab et al. (2022) selected variables to measure the effectiveness of tax planning, they
stated that “these inputs not only relate to the firm’s overall production function but also have
key tax planning opportunities associated with them.” (p. 417).
Alternatively, are accounting measures attributes of firm performance, thus providing
benchmarks of firm performance that incorporate additional information, such as input and
output quality or market conditions? For example, to examine the impact of fair-value-based
accounting as opposed to historical cost accounting, Rodríguez-Pérez et al. (2011) stated that
“with regard to the implications of the differences in the book values under historical cost
and fair value for the analysis of financial statements, a benchmark was obtained by running
our DEA model on the historical-cost-based observations” (p.75).
It is worth noting that accounting measures inherently differ from physical measures
because they aggregate physical measures of different kinds using prices. Accounting
measures represent the sum of various unit prices multiplied by corresponding physical
quantities. For example, salaries and wages comprise physical hours for different labour skill
levels aggregated using different pay rates. These prices reflect differences in quality, market
conditions and pricing strategies. A firm in a weak negotiating position might have to pay
higher prices than a firm with stronger negotiating power. Thus, DEA measures of firm
performance for a firm with a weak market position will be worse than DEA measures for
another with a stronger market position, assuming all other things are equal.
This limitation could be overcome if information were available regarding unit
prices for inputs used or outputs produced. However, unit price information is rarely
available to researchers. Prior empirical studies have sought to overcome this limitation
by using average unit prices, assuming that the unit price is the same for individuals. In
this way, they estimate the physical quantities using accounting measures to calculate
technical efficiency (Barros and Mascarenhas, 2005; Anderson et al., 2000). For Meditari
example, Ravanos and Karagiannis (2022) “assume that input and output prices are Accountancy
uniform” across the units being compared; therefore, “input and output data expressed Research
both in terms of quantities and terms of values (i.e. costs and revenues) can be used to
assess technical efficiency” (p. 210). However, average prices introduce additional
limitations given that they cannot capture price variability within a market and so may
be unrealistic under certain market conditions, such as imperfect markets.
Theoretically, prior literature has investigated whether aggregated accounting measures
can accurately measure technical efficiency. The seminal theoretical work by Färe and
Grosskopf (1985) demonstrated that DEA models using total costs generate equivalent
solutions to physical measures where all inputs are of the same type and have identical
prices. Tone (2002) noted that using aggregated accounting measures as inputs or outputs can
distort the measurement of technical efficiency. Similarly, researchers argue that aggregated
accounting measures can only accurately calculate technical efficiency when firms are
subject to static prices (Camanho et al., 2024; Färe et al., 2017; Färe and Grosskopf, 1985;
Froot et al., 2019; Zelenyuk, 2020).
This study uses simulations to isolate and control price variation within accounting
measures. This enables us to examine the impact on technical efficiency when the level of
price variation increases within our samples of simulated firms. We compare the DEA results
using accounting measures with corresponding results using physical measures to identify
differences. To generate this simulated data, we follow prior DEA simulation research
(Banker et al., 2017; Banker and Podinovski, 2017; Jradi and Ruggiero, 2019) and use a
Cobb–Douglas function to simulate physical measures of inputs and outputs. Next, we
generate price information with varying levels of variability and multiply these with the
corresponding physical measures to produce accounting measures for each simulated firm.
This approach allows us to know the underlying “truth” about price variability for each firm,
a detail often unavailable or proprietary in empirical studies. We use the simulated physical
and accounting measures to estimate DEA firm performance efficiency scores. Finally, we
compare these scores in the presence of increasing levels of price variation and sample size.
This enables the identification of the conditions under which DEA evaluations of
performance using accounting measures can proxy DEA evaluations using physical
measures, i.e. how well they estimate “true” technical efficiency.
This study has three key findings. First, when prices are uniform across all organisations,
the DEA model using accounting measures yields results identical to those of the model
using physical quantities. This finding aligns with the theoretical literature (Camanho et al.,
2024; Färe et al., 2017). Second, where there is low price variability across firms, the DEA
model using accounting measures to measure productivity has similar results to the model
using physical quantities. Third, when price variability is high across organisations, the DEA
model using accounting measures produces significantly different results than the model
using physical quantities. In this last case, the issue becomes what the DEA measures when it
is based on accounting measures. In this case, we argue that DEA provides a general
benchmarking measure of performance against best practices and not a measure of
productivity.
Performance measurement does not necessarily imply benchmarking, as it is often used to
examine trends over time or explain performance variability using explanatory factors in
second-stage multivariate analysis (e.g. Demerjian et al., 2012, 2013). A fundamental aim of
benchmarking is continuous improvement, where performance indicators or measures can
inform process improvements (Wireman, 2004). Hence, although there may be issues with
MEDAR high price variability, the insights provided by DEA about scores, targets and benchmarking
peers can be valuable for benchmarking purposes.
In this study, we make the following contributions to accounting research. First, we
demonstrate that when price variability is low, accounting measures are suitable for
measuring productivity. When accounting measures are proxies for underlying physical
measures underpinning financial performance, DEA results can align with the measurement
of underlying physical productivity. Researchers can, therefore, interpret the DEA results
using accounting measures in the same way as when using physical measures. Compared
with other models, such as ratios or regressions, DEA can capture two different aspects of
firm performance and physical productivity in addition to financial performance (Wang
et al., 2018).
Second, we show that when price variability is high, accounting measures are useful for
benchmarking financial performance but should not be relied on to measure the underlying
physical productivity. DEA can accommodate different dimensions of firm performance, as
the model allows for multiple combinations of inputs and outputs. In this case, DEA provides
a comprehensive measure to benchmark the financial performance of firms but does not
directly relate to the underlying physical productivity.
Third, we show that it is important that researchers consider price variability if comparing
across industries or sub-industries when assessing performance using accounting measures
in DEA. Market structures, such as demand and supply relationships, negotiation power and
pricing strategies, can lead to price variability being incorporated into accounting measures.
This is of particular importance to accounting researchers, given that large-scale panel data is
often used in empirical research (Demerjian et al., 2012, 2013; Schwab et al., 2022).
Fourth, we illustrate that internal accounting information, such as knowledge of price
variability of an organisation’s resources, can enable managers to better interpret and act on
DEA results for continuous improvement. It is important to note that DEA originates from
productivity theories and was designed to measure physical productivity using physical
quantities. Incorporating internal accounting data broadens the source of information and
scope of DEA for performance measurement, but ignoring the existence of price variability
and treating accounting measures the same as physical measures can lead to distortions in
interpreting results.
Fifth, we contribute to the practitioner’s accounting toolkit by outlining how financial
DEA performance metrics can inform performance evaluation where there are different
levels of price variation in accounting measures. We expand the traditional toolkit by helping
firms assess whether DEA results measure operational productivity and can be used to
identify specific priorities related to optimising production processes and minimising
financial inputs or maximising financial outputs. Alternatively, whether DEA results provide
benchmarks that identify best practices and can be used to identify general priorities related
to reducing financial resources or increasing financial outputs. For benchmarking, DEA
results can combine multiple dimensions into a single index, which can resolve interpretation
difficulties when using multiple financial ratios.
Finally, we suggest that simulation can be considered a suitable research method in
accounting research where access to ideal empirical data is constrained, especially if the aim
is to control and quantify the magnitude of variable changes.
The structure of the paper is as follows. Section 2 reviews the literature on DEA and its
application in the accounting field. Section 3 details the research objectives and the research
proposition. Section 4 explains the methodology for data generation and the research
method. Section 5 discusses the findings, and Section 6 summarises the principal results and
offers recommendations for future research.
2. Literature review Meditari
2.1 Data envelopment analysis fundamentals Accountancy
DEA is an optimisation tool used to measure the relative performance (efficiencies) of Research
organisations, including both for-profit and not-for-profit entities, which are termed decision-
making units (DMUs) (Charnes et al., 1978; Farrell, 1957). DEA is mathematically based on
linear programming, where weights are unknown variables applied to inputs and outputs in
their production process. DEA finds the best weights so that each unit being compared is
given an optimal efficiency score based on how it performs compared to all the other units in
the same group. The DMUs with the highest efficiency scores (100%) can be used to
construct a piece-wise frontier from this process, which is considered the efficient frontier.
Relatively inefficient DMUs are dominated by the efficient frontier, and the distance from
inefficient DMUs to the efficient frontier provides a measure of efficiency and the need to
find strategies for performance improvement. To be a better performer, inefficient DMUs
need to learn from their peers on the frontier who use similar resource allocations and have
similar strategic goals (Coelli et al., 2005; Cooper et al., 2011; Rouse et al., 2010).
DEA has several advantages over traditional accounting performance measurement tools.
First, unlike regression analysis, DEA is a frontier-based method and is not based on central
tendency. DEA identifies best practice performers being the DMUs on the frontier. These
efficient DMUs receive an efficiency score of 1 (i.e.100% efficient). In comparison, DMUs
within the frontier are relatively inefficient firms (e.g. 97% efficient). DEA offers strategic
insights for inefficient DMUs on how to improve their performance by learning from their
efficient peers (i.e. efficient DMUs whose production configuration is closest to the
inefficient DMU). The inefficient DMUs can either reduce the inputs used (resources such as
labour or capital) or increase the outputs produced (such as products or services) to move
towards the frontier and maximise their efficiency score (e.g. improve from being 97% to
100% efficient). DEA identifies areas where these resources can be reallocated in the
production process to increase efficiency and cut down on waste.
Second, DEA assigns weights based on the strengths and weaknesses of each DMU,
contrary to other performance measures such as financial ratios (Hassan et al., 2023; Sattar
et al., 2023) or composite measures that often use arbitrary or uniform weightings. DEA
calculates efficiency without imposing such a structure, allowing each firm to maximise its
performance based on what it does best. This approach also reduces opportunistic behaviour
that might arise if the weightings used to combine performance measures were
predetermined. Information about a DMU’s strategic preferences can also be inferred by its
weight profile (Podinovski, 2016). Third, DEA considers multiple inputs and outputs,
capturing the multifaceted nature of firm performance (Harrison et al., 2012). This addresses
issues in performance measurement based on financial ratios, as DEA can combine multiple
performance dimensions that are difficult to combine using traditional approaches, e.g. ROI
and gearing ratios. DEA provides a comprehensive composite performance score that is units
invariant, i.e. inputs and outputs can be combined irrespective of their measurement
dimensions.
Given these benefits, DEA has been applied extensively to measure firm performance in
diverse industries such as banking (Desta, 2016; Rouse and Tripe, 2016), health care (Färe
et al., 2007), agriculture (Färe and Grosskopf, 2006), education (Avilés-Sacoto et al., 2014)
and sustainability reporting (Zhou et al., 2018).
DEA can measure three types of efficiency related to productivity. Technical efficiency
(TE) uses physical measures and maximises (minimises) output (input) from a given set of
inputs (outputs) (Farrell, 1957, p. 254). Allocative efficiency (AE) takes prices into
consideration and measures a firm’s resource allocation responsiveness to specific price
MEDAR conditions (Farrell, 1957, p. 261). Overall efficiency (OE) combines TE and AE (Farrell,
1957, p. 255).
Figure 1, adapted from Coelli et al. (2005), illustrates how TE, AE and OE differ for five firms
(1–5 in Figure 1) for a two-input, one-output, input-oriented constant returns to scale (CRS) DEA
model [2]. The two inputs are denoted as x1 and x2, and the output is denoted as q. The firms on the
frontier (shown as the piecewise “curve” connecting Firms 2 and 5) are technically efficient;
Firms 1, 3 and 4 are technically inefficient as they lie above this frontier. All the firms face the
same prices, which are $1 and $3 for input x1 and x2, respectively. These prices form an isocost
line with a slope of −1/3 and on this line, firms are allocatively efficient. Among the five firms,
only firm 5 is overall efficient, being both technically and allocatively efficient.
The calculation of TE, AE and OE can be illustrated by Firm 3. The TE is measured as the
ratio of the distance from the origin to Point 3 0 divided by the distance from the origin to 3,
with TE being 0.833. The AE is calculated as the ratio of the distance from the origin to 3″
divided by the distance from the origin to 3 0 and AE is 0.9. The OE is the ratio of the distance
from the origin to 3″ divided by the distance from the origin to 3, with OE being 0.75. The
values of OE, AE and TE follow OE = AE × TE (0.75 = 0.9 × 0.833).
When using accounting measures in DEA to measure firm performance, the distinction
among the three efficiencies (OE, TE, AE) is blurred because of the inherent nature of firm-
level accounting measures, which aggregate physical quantities using various corresponding
prices. Coelli et al. (2005, p. 183) note: “[i]f price data are available and a behavioural
objective, such as cost minimisation or revenue or profit maximisation, is appropriate, then it
is possible to measure allocative efficiencies as well as technical efficiencies”. If both
quantities and corresponding price information for inputs and outputs are available, TE, AE
and OE can be calculated. Without specific price information, efficiency measures can
become ambiguous from a productivity perspective (Farrell, 1957).
Notwithstanding, technical efficiency can be estimated if accounting measures are
proxies for physical quantities, with unknown prices treated as measurement errors affecting
the technical efficiency estimation. Alternatively, DEA could measure overall efficiency and
allocative efficiency if average prices are used to enable physical quantities to be estimated
from the accounting information (Harrison and Rouse, 2016).

Figure 1. Technical efficiency, allocative efficiency and overall efficiency from Coelli et al. (2005)
 This traditional view of DEA is closely tied to productivity theory in economics. In Meditari
contrast, Cook et al. (2014, p. 2) emphasise that while DEA is rooted in production theory, it Accountancy
can also be used for benchmarking. In this context, firms are compared to each other to assess Research
relative performance, with inputs representing “less-the-better” performance attributes and
outputs representing “more-the-better” attributes. These performance attributes can go
beyond physical quantities and incorporate additional information, e.g. the quality of inputs/
outputs. Hence, benchmarking using DEA based on accounting measures can be used to
measure firm performance against “best practice” and to establish performance targets and
identify appropriate peers.

2.2 Use of accounting measures in data envelopment analysis

In the early 1980s, a majority of inputs and outputs used in DEA studies were based on
physical measures, aligning with DEA’s origin in economics (Färe et al., 1985; Farrell,
1957). For example, Charnes and Cooper (1980) and Charnes et al. (1981) measured the
performance of public schools using physical variables [3]. Later, in the 1980s, a mix of
accounting and physical measures began to be used. For instance, Sherman and Gold (1985)
used a mix of physical and accounting measurements for the performance of banks [4],
noting that “[t]here may be some difficulty with using rent in the analysis. Rent will vary due
to the desirability of locations” (p. 305). They also noted that rent may not be a good proxy
for the occupancy area.
Over the years, the use of accounting measures in DEA has increased across a range of
industries, particularly in the banking sector where a significant number of measures are
monetary (e.g. Benston, 2004). Production resources are often proxied by dollars when
measuring technical efficiency, as the physical measures are generally proprietary
information and not available. For example, an input such as labour is proxied by personnel
costs (Leightner and Lovell, 1998) and materials can be proxied by operating expenses
(Giokas, 2008). On the output side, banking transactions and the number of services can be
proxied by sales and revenue (Noulas et al., 2008). Alternatively, accounting measures can
be direct measures of an intermediation production process. To measure efficiency, inputs
can be deposits and equity, with outputs being loans and investments (Athanassopoulos,
1997).
Smith (1990) was the first study to use only accounting data in DEA for financial
statement analysis. The author “extends the traditional ratio analysis to permit the
incorporation of any number of decisions of performance, using Data Envelopment
Analysis” (Smith, 1990, p. 131). The study used interest and tax payments, debt and equity as
inputs to generate net income as the output. Other studies have used both accounting and
physical measures (Andreou et al., 2016; Gong et al., 2018).
Recently, there has been a surge in the use of accounting data in DEA models led by
Demerjian et al. (2012, 2013), who used DEA to calculate the financial efficiency of
generating revenues from resources [5]. These types of DEA models based on accounting
measures have been used to examine various accounting topics, such as auditing (Demerjian
et al., 2013, 2020; Krishnan and Wang, 2015), financial statement analysis (Ahn et al., 2020),
corporate governance (Baghdadi et al., 2018; Baik et al., 2011, 2020) and the change of
accounting regulations (Banker et al., 2021; Rodríguez-Pérez et al., 2011). More recently,
Schwab et al. (2022) used DEA to examine tax issues by developing a measure for the
efficiency of maximising after-tax returns given firms’ operating, investing and financing
decisions [6].
However, the use of accounting measures in DEA without proper consideration of how
the aggregation of different price and quantity information can impact the types of
MEDAR performance evaluation that can be made (Färe et al., 2017; Färe and Grosskopf, 1985;
Portela and Thanassoulis, 2014; Zelenyuk, 2020). At the micro level, price variability can
represent differences in quality, such as various skill levels of auditors (Banker et al., 2007;
Knechel et al., 2009). At the macro level, price variability can represent the variation in
supply and demand relationships in various markets, industries or countries (Kapelko and
Oude Lansink, 2017).

3. Research objective and research proposition

In this study, we aim to provide a better understanding of DEA results when accounting
measures are used to measure performance. This study focuses on price variability and its
critical role when using accounting measures in DEA. We identify the level of price
variability as the condition under which accounting measures used in DEA allow for a
reliable estimation of technical efficiency and productivity, and conditions under which DEA
using accounting measures can only be used for benchmarking purposes.
A core accounting assumption is that accounting representations act as surrogates for
underlying economic reality (Bedford, 1968; Ijiri, 1967) [7]. Accordingly, we postulate an
underlying physical production process being measured using DEA and accounting
measures, where aggregated accounting data serve as proxies for physical measures. Thus,
DEA using accounting measures can be used to estimate productive efficiency. Our
investigation is premised on the understanding that accounting measures typically represent
the product of physical quantities and their associated prices. For example, labour cost (an
accounting measure) represents the number of labour hours (a physical quantity) multiplied
by wage rates (price). While accounting measures can be affected by factors such as the
aggregation of various prices (e.g. different hourly rates for senior and junior labour) and
accounting choices (such as the allocation of labour costs to specific cost centres or
products), our study primarily examines the influence of price variation on DEA results.
When only accounting measures are available, without specific prices or quantities,
certain conditions, such as the Law of One Price, [8] must be satisfied to reliably calculate
productive efficiency (Camanho et al., 2024; Färe et al., 2017). However, when specific
prices are unknown and vary across firms, accounting measures can introduce measurement
errors. This prompts us to investigate whether the estimation of productive efficiency in DEA
is affected by price variability and the size of the variability. Hence, the proposition to be
tested is that price variability within accounting measures influences DEA results compared
to the productive efficiency measured using physical quantities.

4. Research method
This study uses simulated data to assess the impact of price variability on the measurement of
technical efficiency. Simulated data are used because unit price information is usually
unavailable for researchers because of confidentiality reasons, making the use of empirical
data impractical for our purpose. The advantage of using simulated data to compare DEA
results based on accounting measures to DEA results based on corresponding physical
quantities is that we can identify “true” productive efficiency. This enables us to determine
the extent to which estimates of DEA based on accounting measures vary from that truth
(Olesen and Ruggiero, 2022). However, the disadvantage of simulated data is that it relies on
simplifying assumptions regarding production and economic relationships that may not
capture the complexities of the real world.
4.1 Physical data generation process Meditari
4.1.1 The underlying physical production process. Following simulation research in DEA Accountancy
(Banker and Chang, 1995; Jradi and Ruggiero, 2019), we used a simulation built around a Research
Cobb–Douglas function in two stages. We first use the Cobb–Douglas function to simulate
physical quantities of outputs from two inputs. The general form of the Cobb–Douglas
function is as follows:

y = xα1 xβ2

where y and x denote output and input variables, respectively, and α and β modify the shape
of the frontier by allowing for returns to scale.
By construction, the output is efficient, so inefficiency is introduced by reducing the
Cobb–Douglas output quantity using a randomised measure of inefficiency. These simulated
physical inputs and outputs are used in a DEA model to produce efficiency scores for the
base model.
To simulate the underlying physical process, we followed the common procedure around
two inputs (X1, X2) with exponents set to 0.5 and one output (Y) (Khezrimotlagh, 2022):

Y = ðX1 Þ0:5 ðX2 Þ0:5 (1)

As shown for the base model in Table 1, the inputs (X1, X2) were generated from a normal
distribution with a mean of 7.5 and a standard deviation of 0.75 (Banker and Chang, 1995;
Harrison et al., 2012). The output (Y) was calculated using equation (1). As the sum of the
two exponents is one (Oh and Shin, 2015), the model is a CRS production function. To
illustrate the application of equation (1), if X1 = 7.5 and X2 = 7, the n Y = (7.5)0.5 (7)0.5 = 7.25.
4.1.2 Generating inefficiencies. The inputs and outputs generated by the Cobb–Douglas
function in Section 4.1.1 are all fully efficient. Following prior literature, inefficiencies were
generated from a half-normal distribution with a mean of 0 and a standard deviation of 0.2
(Harrison et al., 2012; Jradi and Ruggiero, 2019). These parameters were used to adjust the
efficient output to obtain (Yadj) for 75% of the DMUs, leaving 25% of the DMUs fully
efficient (Adler and Yazhemsky, 2010). Using the numerical example above, if the
inefficiency parameter at random was 0.1, the output of 7.25 would be reduced using this
parameter. These physical values of the two inputs and adjusted output were used in all
scenarios and iterations. They form the base model used to compare with the DEA results
under seven different price variability scenarios, described below. A constant-returns-to-
scale and output orientation DEA model was chosen consistent with prior simulation
research that uses the Cobb–Douglas function (Ruggiero, 1998).

4.2 Price variation scenarios – constant to volatile

Seven different assumptions were made regarding prices for the scenarios (Scenarios 1–7)
shown in Table 1. For the seven price variability scenarios, accounting measures for the
inputs and outputs were calculated by multiplying the physical quantities with their
respective prices in the different scenarios. In Scenario 1 (constant prices), the market prices

for each input and output were
not subject to variability with the price of input one Wx 1 c
equal to 10, input two Wx 2 c to 15 and the price of the output ðWyc Þ equal to 30.
This scenario represents an ideal case where the Law of One Price holds and theory
suggests price-based aggregated data can accurately estimate the physical productive
efficiency (Färe et al., 2017).
 MEDAR

Table 1. Research design parameter specification

Coefficient
Scenarios Input Price distribution of variation Output

Base model X1, X2 ∼ n(7.5,0.75) – – Yadj

Scenario 1 X1 Wx1c ; X2 Wx2c Wx1c = 10; Wx2c = 15; Wyc = 30 – YadjWyc
Scenario 2 X1 Wx1_s1 ; X2 Wx2_s1 Wx1_s1 ∼ N ð10; 0:1Þ; Wx2 _s1 ∼ N ð15; 0:15Þ; Wy_s1 ∼ N ð30; 0:3Þ 0.01 Yadj Wy_s1
Scenario 3 X1 Wx1_s2 ; X2 Wx2_s2 Wx1_s2 ∼ N ð10; 0:5Þ; Wx2 _s1 ∼ N ð15; 0:75Þ; Wy_s1 ∼ N ð30; 0:3Þ 0.05 Yadj Wy_s2
Scenario 4 X1 Wx1_s3 ; X2 Wx2_s3 Wx1_s1 ∼ N ð10; 0:1Þ; Wx2 _s3 ∼ N ð15; 1:5Þ; Wy_s3 ∼ N ð30; 0:3Þ 0.1 Yadj Wy_s3
Scenario 5 X1 Wx1_s4 ; X2 Wx2_s4 Wx1_s4 ∼ N ð10; 1:5Þ;Wx2_s4 ∼ N ð15; 2:25Þ; Wy_s4 ∼ N ð30; 4:5Þ 0.15 Yadj Wy_s4
Scenario 6 X1 Wx1_s5 ; X2 Wx2_s5 Wx1_s5 ∼ N ð10; 2:0Þ; Wx2 _s5 ∼ N ð15; 3:0Þ; Wy_s5 ∼ N ð30; 6:0Þ 0.2 Yadj Wy_s5
Scenario 7 X1 Wx1n ; X2 Wx2n Wx1_s6 ∼ N ð10; 2:5Þ; Wx2 _s6 ∼ N ð15; 3:75Þ; Wy_s6 ∼ N ð30; 7:5Þ 0.25 Yadj Wy_s6

Note: Variable definitions can be found in the Supplementary Table 1

Source: Table by authors
 In the narrow scenario (Scenario 2), the market prices for each input and output were Meditari
generated from a relatively narrow distribution band. This scenario aims to simulate a market Accountancy
with minimal price variation. For example, in a market where firms have relatively limited Research
buying and selling power and where the products are of similar quality, the prices of inputs
used in the production process and the prices of outputs sold in the market are expected to be
highly similar (Porter, 2011). In such a market, the prices may exhibit some diversity, but this
is limited. This scenario was simulated using a coefficient of variation of 0.01, resulting in a
narrow price variation. The price of input one ðWx1_s1 Þ followed a normal distribution with a
mean of 10 and a standard deviation of 0.1. The price of input two ðWx2_s1 Þ followed a normal
distribution with a mean of 15 and a standard deviation of 0.15. The price of the output
(Wy−s1) followed a normal distribution with a mean of 30 and a standard deviation of 0.3
(Olesen and Ruggiero, 2022).
In the broad scenario (Scenario 7), market prices exhibit relatively high levels of price
variability. In markets where firms have significant buying and selling power, different firms
can price a product quite differently, which can be attributed to variations in product features
or quality (Porter, 2011). A coefficient of variation of 0.25 was chosen to simulate this
scenario, which is 25 times larger than the narrow scenario (Scenario 2). The price of input
one ðWx1_s6 Þ was generated from a normal distribution with a mean of 10 and a standard
deviation of 2.5. The price of input two (Wx2_s6) was generated from a normal distribution
with a mean of 15 and a standard deviation of 3.75. The price of the output (Wy_s6) was
generated from a normal distribution with a mean of 30 and a standard deviation of 7.5.
In addition to the narrow and broad scenarios (Scenarios 2 and 7), four intermediate
scenarios (Scenarios 3–6) were constructed to demonstrate the effect of a gradual change in
price variation. The coefficients of variation for these scenarios were set to 0.05, 0.1, 0.15
and 0.2, respectively (Olesen and Ruggiero, 2022).

4.3 Sample sizes

The research also examined the impact of price variability for different sample sizes. Four
sample sizes were tested, ranging from 6 to 384, and each test was repeated 1,000 times using
the Monte Carlo method (Jradi et al., 2021). The smallest sample size chosen was six, based
on the DEA ‘rule of thumb’ that the number of firms (or DMUs) should be no less than 2 ×
(m + s) = 2(2 + 1) = 6, where m is the number of inputs and s is the number of outputs (Golany
and Roll, 1989; Khezrimotlagh et al., 2021). The largest sample size was comparable to the
largest industries examined in Demerjian’s studies (2012, 2018).

4.4 Evaluation criteria for data envelopment analysis models with accounting measures
Following criteria proposed in prior literature (Banker et al., 1993; Jradi and Ruggiero, 2019;
Olesen and Ruggiero, 2022; Pedraja-Chaparro et al., 1999), the DEA models were evaluated
using four criteria (a–d) and nine tests (1–9) to determine the similarity of the estimated
efficiencies for each of the seven scenarios (accounting measures) compared to the base
model (physical measures). The detailed mathematical calculations for these criteria are in
Supplementary Material 1.
Criterion (a) compares the ranks in the performance of firms between the DEA results for
the seven scenarios and the base model using (1) Pearson correlations and (2) Spearman’s
rank correlations.
Criterion (b) estimates the efficiency of the entire sample using the following three tests to
measure the deviation of the DEA results for the seven scenarios from the base model using
(3) mean absolute deviations (MAD), which measure how close the DEA results for the
scenarios are to the base model (Banker et al., 1993; Santín and Sicilia, 2017); and (4) total
MEDAR sum squared errors (TSS) and (5) mean squared errors (MSE), which measure the deviation
of the DEA results for the scenarios to the base model (Khezrimotlagh, 2022).
Criterion (c) identifies inefficient firms from their efficient peers using two tests for the
scenarios compared to the base model using (6) TOP, which compares the top proportion
(most efficient) of the DEA results for the scenarios to those of the base model; and (7)
INEFF, which compares the bottom proportion (most inefficient) of the DEA results for the
scenarios to those of the base model (Kohl and Brunner, 2020).
Criterion (d) estimates efficiencies and compares model improvements using two tests
applied to the seven scenarios and the base model using (8) CORRI, which compares the
estimated efficiency scores within certain corridors for the scenarios to the efficiency scores
from the base model (Kohl and Brunner, 2020); and (9) DIFF%, which measures the
percentage of the efficiency scores generated by DEA for the scenarios that differ from the
base model (Harrison et al., 2012).

5. Results and discussion

The descriptive statistics of the simulated variables are consistent with the research design
parameters shown in Table 1, which indicates that the simulation fits the research design and
that the 1,000 iterations were sufficient to mitigate the effects of randomisation [9]. Table 2
reports the DEA scores, and Table 3 reports the results of the four tests when comparing the
DEA scores using the seven scenarios (accounting measures) and nine criteria to the base
model (physical measures) for sample sizes 6 and 384. The results for sample sizes 24 and 96
are provided in Supplementary Material Table 4 and Table 5.

Table 2. Descriptive statistics for efficiency scores

MEAN 100.00% 99.00% RANGE MIN Q1 MED Q3 SD CV

Panel A: n = 6, iteration = 1,000

Physical 0.9228 0.0018 0.4507 0.5189 0.4811 0.8655 0.9692 1.0000 0.0981 0.1064
Scenario 1 0.9228 0.0015 0.4507 0.5189 0.4811 0.8655 0.9692 1.0000 0.0981 0.1064
Scenario 2 0.9220 0.0018 0.4358 0.5207 0.4793 0.8640 0.9685 1.0000 0.0983 0.1066
Scenario 3 0.9049 0.0008 0.3603 0.5270 0.4730 0.8373 0.9399 1.0000 0.1068 0.1180
Scenario 4 0.8731 0.0017 0.3337 0.6130 0.3870 0.7792 0.9057 1.0000 0.1308 0.1498
Scenario 5 0.8403 0.0007 0.3200 0.7537 0.2463 0.7220 0.8739 1.0000 0.1583 0.1884
Scenario 6 0.8065 0.0015 0.3153 0.8025 0.1975 0.6660 0.8369 1.0000 0.1883 0.2335
Scenario 7 0.7765 0.0007 0.3210 0.9068 0.0932 0.6078 0.8151 1.0000 0.2175 0.2802
Panel B: n = 384, iteration = 1,000
Physical 0.8942 0.0019 0.2816 0.5999 0.4001 0.8249 0.9177 1.0000 0.1045 0.1169
Scenario 1 0.8942 0.0017 0.2816 0.5999 0.4001 0.8249 0.9177 1.0000 0.1045 0.1169
Scenario 2 0.8733 0.0001 0.0465 0.6131 0.3869 0.8053 0.8961 0.9641 0.1028 0.1177
Scenario 3 0.7876 0.0001 0.0144 0.6757 0.3243 0.7164 0.7989 0.8658 0.1064 0.1351
Scenario 4 0.6878 0.0000 0.0104 0.7500 0.2500 0.6026 0.6859 0.7700 0.1225 0.1781
Scenario 5 0.5992 0.0001 0.0099 0.8588 0.1412 0.4996 0.5888 0.6874 0.1403 0.2342
Scenario 6 0.5159 0.0000 0.0096 0.9706 0.0294 0.4058 0.4993 0.6073 0.1547 0.2999
Scenario 7 0.4403 0.0000 0.0100 0.9894 0.0106 0.3239 0.4171 0.5304 0.1652 0.3752

Notes: (a) The columns include the mean value (MEAN), percentage of the number of efficient DMUs to the total number
of DMUs in the sample set (100%), the percentage of the number of DMUs that are at least 99% efficient to the total number
of DMUs a the sample set (99%), the range of efficiency scores (RANGE), the minimal value (MIN), 25th percentile (Q1),
50th percentile (MED), 75th percentile (Q3), the standard deviation (SD) and the coefficient of variation, which is the ratio
of the standard deviation to the mean (CV); (b) all efficiency scores are generated from the CRS model; (3) the sample sizes
are n = 6, 24, 96, 384. This table reports the results of n = 6 and 384; the rest of the results are reported in Supplementary
Table 4
Source: Table by authors
Table 3. Variation in DEA scores Meditari
Accountancy
Scenario Pearson Spearman’s MAD TSS MSE TOP INEFF CORRI DIFF%
Research
Panel A: n = 6, iteration = 1,000
Scenario 1 1.0000 0.9207 0.0000 0.0000 0.0000 0.4210 1.0000 0.7500 0.0000
(0.0000***) (0.0320**)
Scenario 2 0.9899 0.8876 0.0071 0.0008 0.0001 0.4210 0.9470 0.7499 0.0081
(0.0009***) (0.0518*)
Scenario 3 0.8419 0.7286 0.0391 0.0203 0.0034 0.3620 0.7350 0.6363 0.0441
(0.0669*) (0.1604)
Scenario 4 0.6229 0.5416 0.0795 0.0769 0.0128 0.3100 0.5590 0.4825 0.0887
(0.2127) (0.3030)
Scenario 5 0.4471 0.3794 0.1186 0.1626 0.0271 0.2510 0.4170 0.3899 0.1315
(0.3304) (0.4143)
Scenario 6 0.3569 0.3120 0.1547 0.2714 0.0452 0.2340 0.3640 0.3350 0.1712
(0.3843) (0.4399)
Scenario 7 0.2880 0.2538 0.1860 0.3966 0.0661 0.2390 0.3120 0.3110 0.2050
(0.4274) (0.4807)
Panel B: n = 384, iteration = 1,000
Scenario 1 1.0000 0.9840 0.0000 0.0000 0.0000 0.5996 1.0000 1.0000 0.0000
(0.0000***) (0.0000***)
Scenario 2 0.9934 0.9688 0.0213 0.2263 0.0006 0.5291 0.9519 0.9964 0.0238
(0.0000***) (0.0000***)
Scenario 3 0.8712 0.8320 0.1076 5.5190 0.0144 0.3901 0.7802 0.4617 0.1203
(0.0000***) (0.0000***)
Scenario 4 0.6670 0.6379 0.2076 19.8696 0.0517 0.3154 0.5974 0.1518 0.2322
(0.0000***) (0.0000***)
Scenario 5 0.5080 0.4904 0.2965 39.5957 0.1031 0.2741 0.4728 0.0742 0.3316
(0.0000***) (0.0000***)
Scenario 6 0.3972 0.3893 0.3798 63.5544 0.1655 0.2482 0.3871 0.0438 0.4248
(0.0000***) (0.0000***)
Scenario 7 0.3186 0.3199 0.4557 89.7053 0.2336 0.2335 0.3304 0.0296 0.5097
(0.0000***) (0.0000***)

Notes: (a) DEA models are CRS; (b) P-values are presented in parentheses; ***significance level < 0.01,
** significance level < 0.05, * significance level < 0.1; (c) the criteria for the impact on DEA scores results
are the Pearson Correlation (Pearson); the Spearman’s Ranking Correlation (Spearman’s); the Mean
Absolute Deviation (MAD); TSS: total sum squared errors; MSE: Mean squared errors; TOP: considers the
top 15% DMUs identified by DEA; INEFF: considers the bottom 15% DMUs identified by DEA; CORRI:
the mean value over 3 corridors of size 0.05; DIFF%: percentage DEA measure differ from the base model;
(d) the sample sizes are n = 6, 24, 96, 384. This table reports the results of n = 6 and 384; the rest of the
results are reported in Supplementary Table 5
Source: Table by authors

Where prices remain constant across firms (Scenario 1), accounting measures accurately
estimate physical productive efficiency (base model). However, as price variability is
incrementally introduced in Scenarios 2–7, the scores gradually deviate from the underlying
productive efficiency. Figure 2 illustrates this using box plots of the distributions for the base
model and seven scenarios.

5.1 Physical measures (base model) and constant prices (Scenario 1) produce identical
results
The results for the base model and constant price scenario are consistent with the theoretical
literature (Camanho et al., 2024; Färe et al., 2017; Tone, 2002; Tone and Tsutsui, 2007).
When prices are static, DEA results using accounting measures generate identical results to
MEDAR

Figure 2. Box plot for the distribution of efficiency scores

those using physical measures, and both measure physical productive efficiency (Table 2 and
Table 3). This is because of the theoretical characteristics of DEA (Banker et al., 1984), [10]
where if a constant value is applied to all the firms’ physical measures of an input or output,
the shape of the frontier and the relative rankings of the firms under evaluation remain the
same.
From an economic and accounting perspective, under perfect competition where all the
firms’ products are of identical quality (Law of One Price), the price for a resource (or
output) is the same and DEA can estimate productive efficiency using accounting measures.
In practice, these conditions may be rare. However, markets close to perfectly competitive
may approximate these conditions.

5.2 Effects of incremental price variability on data envelopment analysis results

When using accounting measures, efficiency scores in scenarios with high price variability
(e.g. Scenarios 5–7) exhibit larger deviations from the underlying productive efficiency
compared to those with less variability (e.g. Scenarios 2–4).
For example, in Table 2, Panel A, the mean efficiency scores for the physical (base model)
and constant prices (Scenario 1) are 0.9228. However, for Scenarios 2–7, which have
increasing price variability (the price variation grows from 1% of the average price to 25% of
the average price), the average efficiency scores decrease from 0.9220 to 0.7765. This
indicates that where the standard deviation of price variation is 1% of the average price, Meditari
the efficiency scores are highly comparable with the base model. By comparison, where Accountancy
the standard deviation of price variation is a quarter of the average price, the efficiency scores Research
are less comparable.
Similarly, in Table 3, Panel A, the results of scenario 1 closely align with those of the base
model. Specifically, Pearson and Spearman’s correlation coefficients are high (1.0000 and
0.9207, respectively) and statistically significant (P-values of 0.0000 and 0.0320,
respectively). The Mean Absolute Deviation (MAD) indicates no deviation from the base
model (MAD = 0.0000). Additionally, the Total Sum of Squares (TSS) and Mean Squared
Error (MSE) confirm the absence of squared errors in Scenario 1 compared to the base model
(TSS = 0.0000; MSE = 0.0000). Scenario 1 also demonstrates the highest accuracy in
identifying the top and bottom DMUs (TOP = 0.4210, INEFF = 1.0000). Compared to other
scenarios, scenario 1 has the highest proportion of accurately identified DMUs relative to the
base model (CORRI = 0.7500)and the lowest percentage of DMUs differing from the base
model (DIFF% = 0.0000).
In contrast, in Table 3, Panel A, the results of Scenario 7 diverge from the base model.
Specifically, Pearson and Spearman’s correlation coefficients are low (0.2880 and 0.2538,
respectively) and statistically insignificant (P-values of 0.4274 and 0.4807, respectively).
The Mean Absolute Deviation (MAD) indicates the largest deviation from the base model
(MAD = 0.1860). Furthermore, the Total Sum of Squares (TSS) and Mean Squared Error
(MSE) show the highest levels compared to the base model (TSS = 0.3966; MSE = 0.0661).
Scenario 7 also has the lowest accuracy in identifying the top and bottom DMUs (TOP =
0.2390, INEFF = 0.3120). Among all scenarios, Scenario 7 has the lowest proportion of
DMUs correctly identified relative to the base model (CORRI = 0.3110) and the highest
percentage of DMUs that differ from the base model (DIFF% = 0.2050).
This pattern is depicted in Figure 2, where scenarios with wider price variability show
broader ranges and the box plots are positioned lower than those with less variability. When
significant price variations exist, DEA results diverge more from the base model.
Conversely, when prices are relatively stable, DEA results using accounting measures reflect
the underlying productivity more accurately.
In practice, where the market structure is close to perfect competition or products are of
similar quality with homogenous prices, DEA can accurately estimate underlying
productivity with accounting measures. This explains why using accounting measures in
DEA flourishes in the banking industry and is often used to measure technical efficiency
(Section 2.2). This is particularly appropriate when competition among branches is in
densely populated areas, as it is arguably close to perfect competition. Moreover, the banking
sector is closely overseen by regulatory bodies to ensure financial stability and to protect
depositors (Bhattacharya et al., 1998). Bank branches provide similar products to the
customers, with similar quality and prices (interest rates) (Oral and Yolalan, 1990).
Accordingly, the level of price variability is expected to be low.
Given the impact of price variability, one strategy accounting researchers could employ
would be to sort their sample data according to pricing strategies, such as cost leaders and
differentiators (Porter, 1980). Although this information is not often directly available from
annual reports, certain measures, such as the gross or operating profit margin, could be used
to identify these strategies and categorise firms into groups. In addition, if panel data across
multiple industries is available, classifying the data according to industry categories or by
financial years could be considered as a remedy to improve the price comparability within
sub-samples (Demerjian et al., 2012, 2013; 2018; Schwab et al., 2022). The homogeneity
MEDAR because of the operating environment for firms could also be improved by this treatment
(Dyson et al., 2001).

5.3 Larger sample size – limited improvement

Increasing the sample size brings the results with accounting measures closer to the
underlying productivity when price variability is low. However, increasing the sample size
does not help where there are high levels of price variability in the accounting measures. For
example, in Table 3, in Scenarios 2–4, where price variability is less than 10% of the average
price, increasing the sample size from 6 to 384 brings the results with accounting measures
closer to the underlying productivity measures (p-values significant at 0.0000). By
comparison, when the price variability is high, such as in Scenario 5–7, where price
variability is 15% to 25% of the average price, increasing the sample size from 6 to 384 can
improve significance levels (p-values at 0.0000), but the coefficients have limited
improvement to around 0.3 in Panel B, which is similar to Panel A (0.2880 for Scenario 7,
person correlation). This finding is also evident for other testing criteria, such as the Mean
Absolute Deviation (MAD) (Table 3).
The impact of sample sizes is in line with the findings of Demerjian (2018), who provides
guidelines for accounting researchers who use DEA for performance measurement. Our
findings extend Demerjian (2018) by specifically focusing on the impact of price variability.
In particular, we found that higher price variability leads to a decline in the efficiency scores
for the same size of samples. In contrast to Demerjian (2018), we also found that this can be
amended by increasing sample sizes when the price variability is low (e.g. in Scenario 2–4,
where the price variability is less than 10% of the average price). However, the discrepancy
cannot be reduced by increasing sample sizes when the price variability is high (e.g. in
Scenario 5–7, when the price variability is higher than 10% of the average price).

5.4 Guideline for future researchers: a four-by-four diagram and reflection on prior
literature
Figure 3 provides a four-by-four diagram to inform accounting researchers and practitioners
using DEA and accounting measures to measure firm performance. In all cases, DEA based
on accounting measures can be used for general firm performance evaluation against best

Figure 3. Four-by-four diagram

practices. When there is no price variation (Scenario 1), accounting measures can be used in Meditari
DEA to estimate underlying productivity. With low price variability (Scenarios 2–3), Accountancy
moderately large sample sizes (n = 24, 96 or 384) can provide reliable estimates of Research
productivity. When price variability is moderate (Scenarios 4–5), large sample sizes (n = 96
or 384) can also provide reasonable estimates of productivity. With high price volatility
(Scenarios 6–7), sample size does not improve estimates of productivity and DEA results
based on accounting measures can be used for benchmarking purposes only.
Our findings offer an alternative perspective for interpreting previous DEA studies that
use accounting measures. For instance, Schwab et al. (2022) apply DEA to assess how
efficiently a firm maximises its after-tax return on equity, given factors such as research and
development expenditures, gross property, plant and equipment, tax haven operations,
intangible assets, inventory and leverage. The authors chose DEA because of its common use
in operations research for examining the production of physical outputs (Schwab et al., 2022,
page 414). However, without addressing the potential price variation in their study, their
interpretation of the results is more of a benchmarking perspective related to the use of
financial resources, rather than operational physical resources. This is still useful since it
provides an evaluation of the relative performance of DMUs against their industry peers,
focusing on maximising after-tax return on equity while minimising the use of resources
such as research and development and property, plant and equipment.
Similarly, Demerjian et al. (2012, p. 1230) “form an efficient frontier by measuring the
amount and mix of resources used to generate revenue by the firms within each industry”.
Figure 3 suggests an alternative benchmarking interpretation, particularly because price
variation is not explicitly considered given the large-scale accounting data used. A new
interpretation could treat the results as benchmarks of firm performance whereby the
performance metric provides general priorities for reducing financial resources used, such as
the seven inputs used by Demerjian et al. (2012).
Our findings align with prior DEA literature that argues DEA results can also be useful
for benchmarking, even though DEA originates from production theory. In this case, firms
can be compared to assess relative performance, where inputs are seen as “less-the-better”
and outputs as “more-the-better” (Cook et al., 2014). Given that accounting data comprise
not only physical quantities but also prices, the DEA performance scores reflect efficiency in
both physical resource utilisation and a firm’s ability to manage market prices.

6. Conclusion
The use of DEA with accounting measures by accounting researchers has rapidly increased
in popularity in recent years, partly because of the increasing availability of large-scale
databases containing accounting measures. This study investigates the ambiguity introduced
by using accounting measures in DEA. We examined when accounting measures are proxies
for physical measures, thus allowing DEA to provide a more comprehensive construct than
traditional financial performance measures.
When price variation is low, accounting measures can be used as proxies for physical
measures. In this case, DEA provides measures for the underlying physical productivity,
which underlies the financial performance. Alternatively, when price variation is high,
accounting measures can be regarded as attributes of firm performance and thus, DEA
provides additional options for benchmarking firm performance. The multi-dimensional
nature of DEA can enhance traditional financial measures of firm performance, such as
accounting ratios, to provide a multi-dimensional financial performance tool.
Theoretical literature suggests that aggregated accounting measures can only accurately
measure productivity and technical efficiency when the underlying prices of inputs and
MEDAR outputs are identical (Camanho et al., 2024; Färe et al., 2017; Färe and Grosskopf, 1985). We
extend this research by using simulations to isolate price variability within accounting
measures and, thus, identify the impact of different levels of price variability on DEA
estimations of firm performance. We compared DEA results based on simulated accounting
measures with DEA results based on simulated physical measures. This approach allowed us
to know the “true” price variability for our simulated firms. From this, we could identify
when DEA results can be used to measure physical productivity (e.g., technical efficiency)
and when they are better suited to general benchmarking evaluations only.
The key findings of our study are, first, when prices are uniform across all organisations,
the DEA model using accounting measures yields results identical to those using physical
quantities, consistent with the theoretical literature (e.g. Färe and Grosskopf, 1985).
Second, with low price variability across firms, DEA based on accounting measures
produces results similar to those from DEA using physical quantities. This suggests that
accounting measures can still be used in DEA to accurately estimate productivity under
certain conditions.
Third, when there is high price variability across organisations, DEA using accounting
measures results in significantly different performance estimates than DEA using physical
quantities. Accordingly, in these circumstances, DEA based on accounting measures should
be considered more as a benchmarking tool that can provide a more comprehensive measure
than traditional accounting measures.
This study contributes to the use of DEA in accounting research by providing a simple
four-by-four guide to using accounting measures that can inform the reliability of research
results, thereby assisting with a more meaningful interpretation of results. The implications
of our findings for accounting research relate to the key differences between productivity
measurement and benchmarking. Measures of technical efficiency are relative productivity
measures that address the maximisation (minimisation) of outputs (inputs) for a given level
of inputs (outputs) through the estimation of a productive frontier (Farrell, 1957). They are
specific performance measures used to improve production processes through better resource
usage and waste elimination.
In contrast, benchmarking measures are general performance measures of firm
performance against best practices. They can incorporate multiple performance attributes
beyond production inputs and outputs (Cook et al., 2014), such as accounting ratios and non-
production resources. Benchmarking is used to identify performance targets and best-in-
class peers and does not necessarily connect to underlying production processes. We argue
that current accounting literature focuses more on benchmarking against peers than
measuring physical productivity. Accordingly, when the financial DEA literature includes no
consideration of price variation in the construction of DEA samples, the results should
be interpreted in the same way that traditional accounting financial ratios are treated
(i.e. financial benchmarking perspective). In contrast, if DEA samples are constructed to
minimise price variation, then the results can inform improvements in underlying physical
production processes (i.e. an operational productivity perspective).
This study makes multiple contributions to accounting research. First, we demonstrate
that accounting measures are suitable for both physical productivity measurement and
benchmarking performance when price variability is low. However, when price variability is
high, accounting measures should be used only for benchmarking financial performance
rather than productivity measurement.
Second, we identify that researchers need to consider price variability when comparing
industries or sub-industries. Market structures, such as demand and supply relationships,
negotiation power and pricing strategies, can lead to price variability being aggregated into
accounting measures. These additional performance attributes will affect how DEA results Meditari
can be interpreted and what additional analyses (such as second-stage regression) are needed Accountancy
to isolate performance controllable by firms and/or their managers. Research
Third, we have also contributed to the performance measurement toolkit of accounting
practitioners. We have explained how DEA can provide a valuable enhancement for
performance measurement when using accounting measures over traditional financial
measurement tools. Depending on the extent of price variability, it can be used to measure
productivity in firms using accounting measures and/or for benchmarking where information
about best practice peers and target values is readily provided by the DEA results. For
accounting practitioners, DEA can add value by identifying areas of both inefficiency and
efficiency to inform efforts for improvement both within an organisation and across
organisations. For example, identifying potential areas of audit risk for internal branches or
benchmarking against leaders in an industry. Accounting practitioners’ knowledge of their
organisations and price behaviour offers opportunities to use accounting measures in DEA
modelling and this research enables them to evaluate the robustness of the results.
There are multiple avenues for further investigation. These results could be extended to
empirically investigate the influence of pricing strategies or market conditions on DEA
results when using accounting measures. This would address the key limitation of this study,
which is its reliance on simulated data. Simulated data enabled us to isolate the impact of
price variability, but it is subject to a number of simplifying assumptions. Owing to the
scarcity of individual unit prices for empirical research, this would be an avenue for future
researchers (Barros and Mascarenhas, 2005; Anderson et al., 2000; Ravanos and
Karagiannis, 2022). Extending this study to empirical data may allow the examination of
specific pricing strategies, such as differentiation or cost leadership, or specific market
conditions applying to different industries.

Notes
1. As of June 9, 2024, Demerjian et al. (2013) has been cited 1,429 times, according to Google
Scholar.
2. There are two types of DEA orientation: input-orientation and output-orientation. Input
orientation focuses on the efficiency of consuming inputs while holding the outputs constant.
Output orientation focuses on the efficient generation of outputs while inputs are fixed (Charnes
et al., 1978). There are two types of returns to scale: constant returns to scale (CRS) and variable
returns to scale (VRS) (Charnes et al., 1978, Banker et al., 1984) where the latter considers the
effect of differences in DMU size.
3. The inputs were (1) educational level of the mother; (2) highest education of a family member;
(3) number of parent visit to the school; (4) time spent with the child on school-related topics; (5)
time spent with the child on school-related topics and (6) the number of teachers at the school.
The outputs are (1) reading scores, (2) mathematical scores (3) self-esteem scores.
4. The inputs were (1) full-time equivalent personnel per branch; (2) rent paid for each branch; and
(3) supply inputs. The outputs are (1) number of transactions of each type processed by the
branch (there are four types of transaction groups, therefore, four output variables).
5. This study included the cost of goods sold, sales, general and administrative expenses, net
property, plant and equipment, net operating leases, research and development expenses,
goodwill and other intangible assets.
6. This study used six inputs, research and development expenses, gross property, plant and
equipment, the number of countries with disclosed subsidiaries in tax havens, intangible assets,
inventory and total debt, to produce the output of after-tax return.
MEDAR 7. See also IFRS Foundation (2018, A26).
8. Identical goods sold in different locations must sell for the same price.
9. Supplementary Material Table 1 summarises variable definitions. Supplementary Material
Table 2 reports the descriptive statistics for sample sizes 6 and 384, and Supplementary Material
Table 3 reports the results for sample sizes 24 and 96.
10. In DEA, the minimum extrapolation axiom dictates that the frontier be estimated as the smallest
function encompassing all observed DMUs.

References
Adler, N. and Yazhemsky, E. (2010), “Improving discrimination in data envelopment analysis: PCA-DEA
or variable reduction”, European Journal of Operational Research, Vol. 202 No. 1, pp. 273-284,
doi: 10.1016/j.ejor.2009.03.050.
Ahn, H., Choi, S. and Yun, S.C. (2020), “Financial statement comparability and the market value of cash
holdings”, Accounting Horizons, Vol. 34 No. 3, pp. 1-21, doi: 10.2308/horizons-18-008.
Andreou, P.C., Philip, D. and Robejsek, P. (2016), “Bank liquidity creation and risk-taking: does
managerial ability matter?”, Journal of Business Finance and Accounting, Vol. 43 Nos 1/2,
pp. 226-259, doi: 10.1111/jbfa.12169.
Athanassopoulos, A.D. (1997), “Service quality and operating efficiency synergies for management
control in the provision of financial services: evidence from Greek bank branches”, European
Journal of Operational Research, Vol. 98 No. 2, pp. 300-313.
Avilés-Sacoto, S.V., Cook, W.D. and Güemes-Castorena, D. (2014), “Competitiveness among higher
education institutions: a two-stage Cobb-Douglas model for efficiency measurement of schools
of business”, Journal of CENTRUM Cathedra (JCC): The Business and Economics Research
Journal, Vol. 7 No. 1, pp. 91-115, doi: 10.7835/jcc-berj-2014-0097.
Baghdadi, G.A., Bhatti, I.M., Nguyen, L.H.G. and Podolski, E.J. (2018), “Skill or effort? Institutional
ownership and managerial efficiency”, Journal of Banking and Finance, Vol. 91, pp. 19-33, doi:
10.1016/j.jbankfin.2018.04.002.
Baik, B., Choi, S. and Farber, D.B. (2020), “Managerial ability and income smoothing”, The Accounting
Review, Vol. 95 No. 4, pp. 1-22, doi: 10.2308/ACCR-52600.
Baik, B., Farber, D.B. and Lee, S.S.(. (2011), “CEO ability and management earnings forecasts”,
Contemporary Accounting Research, Vol. 28 No. 5, pp. 1645-1668, doi: 10.1111/j.1911-
3846.2011.01091.x.
Banker, R.D. and Chang, H. (1995), “A simulation study of hypothesis tests for differences in
efficiencies”, International Journal of Production Economics, Vol. 39 Nos 1/2, pp. 37-54.
Banker, R.D. and Podinovski, V.V. (2017), “Novel theory and methodology developments in data
envelopment analysis”, Annals of Operations Research, Vol. 250 No. 1, pp. 1-3, doi: 10.1007/
s10479-017-2413-7.
Banker, R.D., Chang, H. and Natarajan, R. (2007), “Estimating DEA technical and allocative
inefficiency using aggregate cost or revenue data”, Journal of Productivity Analysis, Vol. 27
No. 2, pp. 115-121, doi: 10.1007/s11123-006-0027-1.
Banker, R.D., Chang, H. and Zheng, Z. (2017), “On the use of super-efficiency procedures for ranking
efficient units and identifying outliers”, Annals of Operations Research, Vol. 250 No. 1,
pp. 21-35, doi: 10.1007/s10479-015-1980-8.
Banker, R.D., Charnes, A. and Cooper, W.W. (1984), “Some models for estimating technical and scale
inefficiencies in data envelopment analysis”, Management Science, Vol. 30 No. 9,
pp. 1078-1092.
Banker, R.D., Gadh, V.M. and Gorr, W.L. (1993), “Theory and methodology: a Monte Carlo Meditari
comparison of two production frontier estimation methods: corrected ordinary least squares and Accountancy
data envelopment analysis”, European Journal of Operational Research, Vol. 67 No. 3,
pp. 332-343. Research
Banker, R.D., Huang, R., Li, Y. and Zhao, S. (2021), “Do accounting standards matter for
productivity?”, Production and Operations Management, Vol. 30 No. 1, pp. 68-84, doi: 10.1111/
poms.13257.
Barros, C.P. and Mascarenhas, M.J. (2005), “Technical and allocative efficiency in a chain of small
hotels”, International Journal of Hospitality Management, Vol. 24 No. 3, pp. 415-436, doi:
10.1016/j.ijhm.2004.08.007.
Bedford, N.M. (1968), “The foundations of accounting measurement”, Journal of Accounting Research,
Vol. 6 No. 2, pp. 270-282.
Benston, G.J. (2004), “What is special about banks?”, Financial Review, Vol. 39 No. 1, pp. 13-33, doi:
10.1111/j.0732-8516.2004.00065.x.
Bhattacharya, S., Boot, A.W.A. and Thakor, A.V. (1998), “The economics of bank regulation”, Journal
of Money, Credit and Banking, Vol. 30 No. 4, pp. 745-770, doi: 10.2307/2601127.
Camanho, A.S., Silva, M.C., Piran, F.S. and Lacerda, D.P. (2024), “A literature review of economic
efficiency assessments using data envelopment analysis”, European Journal of Operational
Research, Vol. 315 No. 1, pp. 1-18, doi: 10.1016/j.ejor.2023.07.027.
Charnes, A. and Cooper, W.W. (1980), “Auditing and accounting for program efficiency and
management efficiency in not-for-profit entities”, Accounting, Organizations and Society, Vol. 5
No. 1, pp. 87-107.
Charnes, A., Cooper, W.W. and Rhodes, E. (1981), “Evaluating program and managerial efficiency: an
application of data envelopment analysis to program follow through”, Management Science,
Vol. 27 No. 6, pp. 668-697, doi: 10.1287/mnsc.27.6.668.
Charnes, A., Cooper, W. and Rhodes, E. (1978), “Measuring the efficiency of decision-making units”,
European Journal of Operational Research, Vol. 2 No. 6, pp. 429-444.
Coelli, T.J., Rao, D.S.P., O’Donnell, C.J. and Battese, G.E. (2005), An Introduction to Efficiency and
Productivity Analysis, 2nd ed., Springer.
Cook, W.D., Tone, K. and Zhu, J. (2014), “Data envelopment analysis: prior to choosing a model”,
Omega, Vol. 44, pp. 1-4, doi: 10.1016/j.omega.2013.09.004.
Cooper, W.W., Seiford, L.M. and Zhu, J. (2011), “Handbook in data envelopment analysis”.
Demerjian, P. (2018), “Calculating efficiency with financial accounting data: data envelopment analysis
for accounting researchers”, available at: https://ssrn.com/abstract=2995038
Demerjian, P., Lev, B. and McVay, S. (2012), “Quantifying managerial ability: a new measure and
validity tests”, Management Science, Vol. 58 No. 7, pp. 1229-1248, doi: 10.1287/
mnsc.1110.1487.
Demerjian, P., Lev, B., Lewis, M.F. and McVay, S.E. (2013), “Managerial ability and earnings quality”,
The Accounting Review, Vol. 88 No. 2, pp. 463-498, doi: 10.2308/accr-50318.
Demerjian, P., Lewis-Western, M. and McVay, S. (2020), “How does intentional earnings smoothing
vary with managerial ability?”, Journal of Accounting, Auditing and Finance, Vol. 35 No. 2,
pp. 406-437, doi: 10.1177/0148558X17748405.
Desta, T.S. (2016), “Are the best African banks really the best? A Malmquist data envelopment
analysis”, Meditari Accountancy Research, Vol. 24 No. 4, pp. 588-610, doi: 10.1108/MEDAR-
02-2016-0016.
Dyson, R.G., Allen, R., Camanho, A.S., Podinovski, V.V., Sarrico, C.S. and Shale, E.A. (2001), “Pitfalls
and protocols in DEA”, European Journal of Operational Research, Vol. 132 No. 2, pp. 245-259,
doi: 10.1016/S0377-2217(00)00149-1.
MEDAR Färe, R. and Grosskopf, S. (1985), “A nonparametric cost approach to scale efficiency”, The
Scandinavian Journal of Economics, Vol. 87 No. 4, pp. 594-604.
Färe, R. and Grosskopf, S. (2006), “Resolving a strange case of efficiency”, Journal of the Operational
Research Society, Vol. 57 No. 11, pp. 1366-1368, doi: 10.1057/palgrave.jors.2602109.
Färe, R., Grosskopf, S. and Lovell, C.A.K. (1985), “The measurement of efficiency of production”.
Färe, R., Grosskopf, S., Karagiannis, G. and Margaritis, D. (2017), “Data envelopment analysis and its
related linear programming models”, Annals of Operations Research, Vol. 250 No. 1, pp. 37-43,
doi: 10.1007/s10479-015-2042-y.
Färe, R., Grosskopf, S., Lundström, M. and Roos, P. (2007), “Evaluating health care efficiency”,
Evaluating Hospital Policy and Performance: contributions from Hospital Policy and
Productivity Research.
Farrell, M.J. (1957), “The measurement of productive efficiency”, Journal of the Royal Statistical
Society, Vol. 120 No. 3, pp. 253-290.
Froot, K.A., Kim, M. and Rogoff, K. (2019), “The law of one price over 700 years”, Annals of
Economics And Finance, Vol. 20 No. 1, pp. 1-70.
Giokas, D.I. (2008), “Assessing the efficiency in operations of a large Greek bank branch network
adopting different economic behaviours”, Economic Modelling, Vol. 25 No. 3, pp. 559-574, doi:
10.1016/j.econmod.2007.10.007.
Golany, B. and Roll, Y. (1989), “An application procedure for DEA”, Omega, Vol. 17 No. 3,
pp. 237-250.
Gong, Y., Zhu, J., Chen, Y. and Cook, W.D. (2018), “DEA as a tool for auditing: application to Chinese
manufacturing industry with parallel network structures”, Annals of Operations Research,
Vol. 263 Nos 1/2, pp. 247-269, doi: 10.1007/s10479-016-2197-1.
Harrison, J. and Rouse, P. (2016), “DEA and accounting performance measurement”, in Hwang, S.N.,
Lee, H.S. and Zhu, J. (Eds), Handbook of Operations Analytics Using Data Envelopment
Analysis, Springer, pp. 385-412.
Harrison, J., Rouse, P. and Armstrong, J. (2012), “Categorical and continuous non-discretionary
variables in data envelopment analysis: a comparison of two single-stage models”, Journal of
Productivity Analysis, Vol. 37 No. 3, pp. 261-276, doi: 10.1007/s11123-011-0239-x.
Hassan, M.K., Lahyani, F.E. and Elgharbawy, A. (2023), “Political connections, media coverage and
firm performance: evidence from an emerging market”, Meditari Accountancy Research, Vol. 31
No. 6, pp. 1634-1653, doi: 10.1108/MEDAR-09-2021-1439.
Ijiri, Y. (1967), The Foundation of Accounting Measurement, Prentice-Hall, Englewood Cliffs, N.J.
Jradi, S. and Ruggiero, J. (2019), “Stochastic data envelopment analysis: a quantile regression approach
to estimate the production frontier”, European Journal of Operational Research, Vol. 278 No. 2,
pp. 385-393, doi: 10.1016/j.ejor.2018.11.017.
Jradi, S., Parmeter, C.F. and Ruggiero, J. (2021), “Quantile estimation of stochastic frontiers with the
normal-exponential specification”, European Journal of Operational Research, Vol. 295 No. 2,
pp. 475-483, doi: 10.1016/j.ejor.2021.03.002.
Kapelko, M. and Oude Lansink, A. (2017), “Dynamic multi-directional inefficiency analysis of
European dairy manufacturing firms”, European Journal of Operational Research, Vol. 257
No. 1, pp. 338-344, doi: 10.1016/j.ejor.2016.08.009.
Khezrimotlagh, D. (2022), “Simulation designs for production frontiers”, European Journal
of Operational Research, Vol. 303 No. 3, pp. 1321-1334, doi: 10.1016/j.
ejor.2022.04.004.
Khezrimotlagh, D., Cook, W.D. and Zhu, J. (2021), “Number of performance measures versus number
of decision making units in DEA”, Annals of Operations Research, Vol. 303 Nos 1/2,
pp. 529-562, doi: 10.1007/s10479-019-03411-y.
Knechel, W.R., Rouse, P. and Schelleman, C. (2009), “A modified audit production framework: Meditari
evaluating the relative efficiency of audit engagements”, The Accounting Review, Vol. 84 No. 5, Accountancy
pp. 1607-1638, doi: 10.2308/accr.2009.84.5.1607.
Research
Kohl, S. and Brunner, J.O. (2020), “Benchmarking the benchmarks – comparing the accuracy of data
envelopment analysis models in constant returns to scale settings”, European Journal of
Operational Research, Vol. 285 No. 3, pp. 1042-1057, doi: 10.1016/j.ejor.2020.02.031.
Krishnan, G.V. and Wang, C. (2015), “The relation between managerial ability and audit fees and going
concern opinions”, Auditing: A Journal of Practice and Theory, Vol. 34 No. 3, pp. 139-160, doi:
10.2308/ajpt-50985.
Leightner, J.E. and Lovell, C.A.K. (1998), “The impact of financial liberalisation on the performance of
thai banks”, Journal of Economics and Business, Vol. 50 No. 2, pp. 115-131, doi: 10.1016/
S0148-6195(97)00073-8.
Noulas, A.G., Glaveli, N. and Kiriakopoulos, I. (2008), “Investigating cost efficiency in the branch
network of a Greek bank: an empirical study”, Managerial Finance, Vol. 34 No. 3, pp. 160-171,
doi: 10.1108/03074350810848045.
Oh, S.C. and Shin, J. (2015), “The impact of mismeasurement in performance benchmarking: a monte
Carlo comparison of SFA and DEA with different multi-period budgeting strategies”, European
Journal of Operational Research, Vol. 240 No. 2, pp. 518-527, doi: 10.1016/j.ejor.2014.07.026.
Olesen, O.B. and Ruggiero, J. (2022), “The hinging hyperplanes: an alternative nonparametric
representation of a production function”, European Journal of Operational Research, Vol. 296
No. 1, pp. 254-266, doi: 10.1016/j.ejor.2021.03.054.
Oral, M. and Yolalan, R. (1990), “An empirical study on measuring operating efficiency and
profitability of bank branches”, European Journal of Operational Research, Vol. 46 No. 3,
pp. 282-294.
Pedraja-Chaparro, F., Salinas-Jime Ânez, J. and Smith, P. (1999), “On the quality of the data
envelopment analysis model”, Journal of the Operational Research Society, Vol. 50 No. 6,
pp. 636-644, available at: www.stockton-press.co.uk/jor
Podinovski, V.V. (2016), “Optimal weights in DEA models with weight restrictions”, European Journal
of Operational Research, Vol. 254 No. 3, pp. 916-924, doi: 10.1016/j.ejor.2016.04.035.
Portela, M.C.A.S. and Thanassoulis, E. (2014), “Economic efficiency when prices are not fixed:
Disentangling quantity and price efficiency”, Omega, Vol. 47, pp. 36-44, doi: 10.1016/j.
omega.2014.03.005.
Porter, M.E. (1980), “Techniques for analysing industries and competitors”.
Porter, M.E. (2011), Competitive Advantage of Nations: Creating and Sustaining Superior
Performance, Simon and Schuster.
Ravanos, P. and Karagiannis, G. (2022), “In search for the most preferred solution in value efficiency
analysis”, Journal of Productivity Analysis, Vol. 58 Nos 2/3, pp. 203-220, doi: 10.1007/s11123-
022-00645-0.
Rodríguez-Pérez, G., Slof, J., Solà, M., Torrent, M. and Vilardell, I. (2011), “Assessing the impact of
fair-value accounting on financial statement analysis: a data envelopment analysis approach”,
Abacus, Vol. 47 No. 1, pp. 61-84, doi: 10.1111/j.1467-6281.2011.00331.x.
Rouse, P. and Tripe, D. (2016), “Allocative and technical efficiency of New Zealand banks”, Meditari
Accountancy Research, Vol. 24 No. 4, pp. 574-587, doi: 10.1108/MEDAR-02-2016-0035.
Rouse, P., Harrison, J. and Chen, L. (2010), “Data envelopment analysis: a practical tool to measure
performance”, Australian Accounting Review, Vol. 20 No. 2, pp. 165-177, doi: 10.1111/j.1835-
2561.2010.00090.x.
Ruggiero, J. (1998), “Theory and methodology a new approach for technical efficiency estimation in
multiple output production”, European Journal of Operational Research, Vol. 111 No. 2,
pp. 369-380.
MEDAR Santín, D. and Sicilia, G. (2017), “Dealing with endogeneity in data envelopment analysis
applications”, Expert Systems with Applications, Vol. 68, pp. 173-184, doi: 10.1016/j.
eswa.2016.10.002.
Sattar, M., Biswas, P.K. and Roberts, H. (2023), “Private firm performance: do women directors
matter?”, Meditari Accountancy Research, Vol. 31 No. 3, pp. 602-634, doi: 10.1108/MEDAR-
03-2021-1233.
Schwab, C.M., Stomberg, B. and Williams, B.M. (2022), “Effective tax planning”, The Accounting
Review, Vol. 97 No. 1, pp. 413-437, doi: 10.2308/TAR-2019-0020.
Sherman, H.D. and Gold, F. (1985), “Bank branch operating efficiency”, Journal of Banking and
Finance, Vol. 9 No. 2, pp. 297-315, doi: 10.1016/0378-4266(85)90025-1.
Smith, P. (1990), “Data envelopment analysis applied to financial statements”, Omega, Vol. 18 No. 2,
pp. 131-138.
Tone, K. (2002), “A strange case of the cost and allocative efficiencies in DEA”, Journal of the
Operational Research Society, Vol. 53 No. 11, pp. 1225-1231, doi: 10.1057/palgrave.
jors.2601438.
Tone, K. and Tsutsui, M. (2007), “Decomposition of cost efficiency and its application to Japanese-US
electric utility comparisons”, Socio-Economic Planning Sciences, Vol. 41 No. 2, pp. 91-106, doi:
10.1016/j.seps.2005.10.007.
Wang, W.K., Ting, I.W.K., Kuo, K.C., Kweh, Q.L. and Lin, Y.H. (2018), “Corporate diversification and
efficiency: evidence from Taiwanese top 100 manufacturing firms”, Operational Research,
Vol. 18 No. 1, pp. 187-203, doi: 10.1007/s12351-016-0259-4.
Wireman, Y. (2004), “Benchmarking or performance measurement: which is right for your plan?”,
Plant Engineering, Vols 54/56.
Zelenyuk, V. (2020), “Aggregation of inputs and outputs prior to data envelopment analysis under big
data”, European Journal of Operational Research, Vol. 282 No. 1, pp. 172-187, doi: 10.1016/j.
ejor.2019.08.007.
Zhou, H., Yang, Y., Chen, Y. and Zhu, J. (2018), “Data envelopment analysis application in
sustainability: the origins, development and future directions”, European Journal of Operational
Research, Vol. 264 No. 1, pp. 1-16, doi: 10.1016/j.ejor.2017.06.023.

Supplementary material
The supplementary material for this article can be found online.

Corresponding author
Claire Murong Cui can be contacted at: [email protected]

For instructions on how to order reprints of this article, please visit our website:
www.emeraldgrouppublishing.com/licensing/reprints.htm
Or contact us for further details: [email protected]
View publication stats