See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/220196852

Detecting evolutionary financial statement

fraud

Article in Decision Support Systems · February 2011

DOI: 10.1016/j.dss.2010.08.007 · Source: DBLP

CITATIONS READS

40 878

2 authors, including:

Wei Zhou
ESCP Europe
81 PUBLICATIONS 726 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Big Data & Business Analytics View project

All content following this page was uploaded by Wei Zhou on 24 September 2015.

The user has requested enhancement of the downloaded file.

 Decision Support Systems 50 (2011) 570–575

Contents lists available at ScienceDirect

Decision Support Systems

j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / d s s

Detecting evolutionary financial statement fraud

Wei Zhou a,⁎, Gaurav Kapoor b
a
Information Systems and Technologies, ESCP Europe, 75543 Paris cedex 11, France
b
Information Systems and Operations Management, University of Florida, Gainesville, Florida 32611, USA

a r t i c l e i n f o a b s t r a c t

Available online 24 August 2010 A fraudulent financial statement involves the intentional furnishing and/or publishing of false information in it
and this has become a severe economic and social problem. We consider Data Mining (DM) based financial
Keywords: fraud detection techniques (such as regression, decision tree, neural networks and Bayesian networks) that
Financial statement fraud help identify fraud. The effectiveness of these DM methods (and their limitations) is examined, especially when
Data mining technique new schemes of financial statement fraud adapt to the detection techniques. We then explore a self-adaptive
Neural networks framework (based on a response surface model) with domain knowledge to detect financial statement fraud.
We conclude by suggesting that, in an era with evolutionary financial frauds, computer assisted automated
fraud detection mechanisms will be more effective and efficient with specialized domain knowledge.
© 2010 Elsevier B.V. All rights reserved.

1. Introduction misapplication of accounting principles, policies, and procedures used

Since the booming of the Internet and the invention of other
modern technologies, there has been a dramatic increase in fraudulent
schemes associated with all facets in the business world. Some of these
commonly observed schemes include credit card fraud, financial
statement fraud, e-commerce transaction fraud, insurance fraud,
money laundering, computer intrusion fraud, telecommunications
fraud, and subscription fraud. Statistic and machine learning based
technologies have been shown to be an effective way to deter and
detect fraud, but fraudsters are adaptive and are usually able to find
ways to circumvent them. Existing fraud detection techniques for most
of the situations involving fraud usually share very similar data mining
principles, but they can differ in many aspects with specialized domain
knowledge [5].
Financial statement fraud in particular has cast rapidly increasing
adverse impact not only on individual investors but the overall
stability of global economies. Although there are minor variations in
its definition, a financial statement fraud is defined by the Association
of Certified Fraud Examiners as “The intentional, deliberate, misstate-
ment or omission of material facts, or accounting data which is
misleading and, when considered with all the information made
available, would cause the reader to change or alter his or her
judgment or decision.” In practice, financial statement fraud might
involve: (1) manipulation of financial records, (2) intentional
omission of events, transactions, accounts, or other significant
information from which financial statements are prepared, or (3)
to measure, recognize, report, and disclose business transactions [20].
Many techniques based on data mining have been investigated
and implemented to detect financial statement fraud, including
regression, decision trees, neural networks and Bayesian belief
networks [12]. These techniques have been shown to be successful
in their early stages. However, there is no agreement on which data
features and techniques are best for detection. Also, while supervised
learning techniques have been among the dominant methods used
for detecting financial statement fraud, a majority of related
implementations do not keep track of new variations in the methods
designed for committing fraud. Moreover, financial fraud is becoming
more and more difficult to detect using the current detection
techniques. A CEO who is truly knowledgeable and wants to really
commit a crime has the necessary resources to easily outwit the
system and is able to fool any detection mechanism [7].
Despite the increased of time and effort that has been spent to
detect the same, the number of detected frauds1 and the detection
rate2 have largely decreased [7]. When the executives who are
involved in financial fraud are well aware of the fraud detection
techniques and software, which are usually public information and
are easy to obtain, they are likely to adapt the methods in which they
commit fraud and make it difficult to detect the same, especially by
existing techniques. There exists an urgent need for new methods that
is not only efficient but effective to catch up with these probable
newly emerged or adaptive financial shenanigans. We (1) consider
existing detection techniques based on data mining, (2) provide an

⁎ Corresponding author.
E-mail addresses: wzhou@escpeurope.eu (W. Zhou), grkapoor@ufl.edu (G. Kapoor).
1
In 2003, the SEC issued 77 financial statement fraud AAERs (Accounting and
Auditing Enforcement Releases) relating to its registrants. It was 44 in 2006.
2
Average time elapsed between the initiation date and the date the SEC issued
AAERs increased by one-third, to 5.6 years in 2006 from 4.1 years in 2001.

0167-9236/$ – see front matter © 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.dss.2010.08.007
 W. Zhou, G. Kapoor / Decision Support Systems 50 (2011) 570–575
overview of existing financial shenanigans and their trend, and (3)
suggest a new framework to detect evolutionary financial statement
fraud.
The remainder of this paper is organized as follows. We review the
application of regression, decision trees, neural networks and
Bayesian belief networks in financial statement fraud detection in
the next section. We survey the history and trend of contemporary
financial fraud in Section 3. We analyze the effectiveness and
limitation of existing fraud detection technologies in Sections 4 and
5 and then suggest a framework that addresses the problem when
emerging financial fraud is evolutionary. Section 6 concludes the
paper with a brief discussion on the insights garnered and possible
future research.
2. Review on detection techniques
Classification has been the most popular and the only way used so
far to identify fraudulent financial statements [8]. Most financial
statement fraud (FSF) auto-detection programs use supervised
machine learning methodologies [1,3,4,9–11,14,18,21,22] that usually
have a two-stage procedure, where in the first stage a model is trained
by using a training sample. The training sample is organized in tuples
and attributes, with the class label attribute containing values
indicating the pre-defined class to which each tuple belongs. In the
second stage, objects are classified through the model obtained from
the first stage. After reviewing relevant research in data mining-based
financial statement fraud detection literature, we observe that the
following five methods have been used so far in this general area.
These methods include regression, decision trees, neural networks,
Bayesian networks and support vector machines.
Regression is the most widely used method to detect financial
statement fraud [1,3,4,11,18,21,22]. Transformations of variables in
regression models have also been studied in the context of fraud
detection, including logit, stepwise-logistic, multi-criteria decision aid
method and exponential generalized beta two. For example, Spathis
[21] used a collection of data from 76 firms that include 38 fraudsters
and 38 non-fraudulent firms in Greece. They use ten financial
variables and logistic multivariate regression to identify the relation-
ship among factors associated with financial statement fraud. A total
of ten financial ratios such as the net profit to total assets ratio, the
ratio of total debt to total assets, financial distress, the inventories to
sales ratio, and the working capital to total assets ratio are selected for
examination as potential predictors of FSF. The results indicate that
companies with high inventories with respect to sales, high debt to
total assets, low net profit to total assets, low working capital to total
assets and high financial stress are more likely to manipulate financial
statements.
A neural network is another popular data mining technique that
has been successfully used to detect financial statement fraud
[6,9,14,16,24]. Neural network doesn't assume an attribute's inde-
pendence and is capable of mining inter-correlated data and is a
suitable alternative for problems where some of the assumptions
associated with regression are not valid. White [24], nonetheless, has
shown that feed-forward neural networks, which require no pre-
specified functional form, perform the same stochastic approximation
as nonlinear regression. Back propagation neural network allows the
network to adapt and has become one of the most popular techniques
for prediction and classification problems. The back propagation
learning process works in small iterative steps that continuously make
small changes to the weights in each neural network layer, which are
calculated to reduce the systematic error. The iteration is repeated
until the overall error value drops below some pre-determined
threshold [13]. Drawbacks of implementing neural networks to
discover FSF is that neural network is not accurate if the data is
volatile or if the causal functionality evolves in a direction that is not
pre-defined.
571
The objective of decision trees is classification by dividing
observations into mutually exclusive and exhaustive subgroups by
properly selecting attributes that best separate the sample. Koh and
Low [13] construct a decision tree to predict the hidden problems in
financial statements by examining the following six variables: quick
assets to current liabilities, market value of equity to total assets, total
liabilities to total assets, interest payments to earnings before interest
and tax, net income to total assets, and retained earnings to total assets.
The above mentioned data mining techniques have generally been
shown to be effective in detecting financial statement fraud. However,
they are not without limitations. For example, while these techniques
are well developed for predictive modeling, they are not as well
developed for effect assessment. In particular, test statistics for
assessing the effects of independent variables on dependent variables
have not yet been constructed for some data mining techniques.
Incidentally, this shortcoming also demands with the challenge to
develop more effective mechanisms especially in an adaptive
economic environment where financial fraudsters learn to circumvent
existing automated detection systems.
3. Detection with domain knowledge on FSF
Financial statement fraud, including motivations, opportunities,
and rationalizations for management to commit such fraud, has been
extensively studied by researchers in finance. Loebbecke et al. [15]
suggest a model consisting of three variables that may explain
financial statement fraud: (C) the degree to which conditions are such
that a financial fraud could be committed, (M) the degree to which the
management has a reason or motivation to commit financial fraud,
and (A) the degree to which the management has an attitude or set of
ethical values such that they would allow themselves to commit
management fraud. These three variables together form the assess-
ment model such that the possibility of having financial statement
fraud (FSF) can be described as a function.
P ðFSF Þ = f ðC; M; AÞ ð1Þ
where if C or M or A = 0, then P(FSF) = 0.
According to Rezaee [19], fraud accomplishment can be explained
by three variables: (1) conditions, (2) corporate structure and (3)
choice. The sufficient incentives and opportunities for a company to
commit financial statement fraud can be interpreted by the pattern
exhibited by these three variables. We, however, believe that certain
combination of the variables mentioned in this model also leads to
certain suitable fraud strategies and, by measuring the pattern among
the variables and integrating the findings in auto-detection heuristics,
we should be able to pinpoint the unique shenanigans and their
underlying dynamic to commit fraud.
We first review possible variables that can be utilized in an auto-
detection system. In Rezaee's[19] 3C's model as shown in Fig. 1,
“conditions” refers to the economic and financial pressures that a
corporation faces. Financial pressures, such as pressure to meet
analysts' earning estimates, can be a key factor stimulating earnings
management and resulting in financial statement fraud. The principle
underlying this variable is that financial statement fraud will most
probably occur if the benefits for fraudulent management outweigh
the associated costs. Management compares the benefit, in terms of
possible increase in the company's stock price or the possible savings
related to preventing share price from decreasing, with the possible
cost of committing fraud in terms of probability and consequences of
detection. Financial pressures, such as the inability to meet analysts'
earning estimates or decline in quality and quantity of earnings, are
often motivations for management commitment in financial frauds.
“Capital structure” refers to the existence of an effective corporate
governance mechanism (such as internal control structure and audit
committees) that could discourage management from committing
572 W. Zhou, G. Kapoor / Decision Support Systems 50 (2011) 570–575
Fig. 1. Rezaee's 3C financial statement fraud model.
fraud. The role of corporate governance devices can also be discussed
in relation to other social and economic characteristics of different
countries where fraud can be committed. It is also important to
explain to what extent an effective corporate governance system can
help to prevent and detect fraud.
“Choices” refer to the management's option, which could either be
ethical strategies of continuous improvements of earnings or illegal
earning manipulations, to deal with various financial situations.
Financial statement fraud is simply one of the choices. Regardless of
the corporate structure or environmental pressure, management
could use financial statement fraud simply as a strategic tool
according to its own characteristics in terms of aggressiveness or
lack of moral principles.
Both the CMA model [15] and the 3Cs model [19] mentioned above
explain the financial statement fraud by examining selected para-
meters, such as the motivations, conditions of pressure, corporate
structure, management's attitude and the choices. These variables
may each function separately, and careful selection and combination
of variables, however, are more likely to help explain the degree and
pattern of financial statement fraud, thus providing insightful and
immediate reference for an FSF auto-detection system.
The strategy that the management utilizes to commit financial
statement fraud is influenced by the economic circumstances as well as
many other variables that have been taken into consideration by the
fraudster. Schilit [20] provides an extensive examination of the
financial statement fraud techniques and categorizes them into
seven groups: recording revenue too soon or of questionable quality,
recording bogus revenue, boosting income with one-time gains,
shifting current expenses to a later or earlier period, failing to record
or improperly reducing liabilities, shifting current revenue to a later
period, and shifting future expenses to the current period as a special
charge. The COSO report [2], by analyzing 204 cases of fraud presented
in the SEC's Accounting Auditing Enforcement Releases (AAERs) from
1987 to 1997, lists common financial statement fraud techniques in the
following categories: improper revenue recognition, overstatement of
assets other than accounts receivable, understatement of expenses/
liabilities, misappropriation of assets, inappropriate disclosure and
other miscellaneous techniques.
4. Limitations of current detection techniques
Despite increasingly stringent legislation such as the Foreign
Corrupt Practices Act and the Sarbanes–Oxley Act aimed at combating
fraud – and despite increasingly growing number of fraud auto-
detection systems – financial statement fraud is becoming a more and
more severe public concern. Doloitte [7] reviews around 1300
Accounting and Auditing Enforcement Releases (AAER) that the SEC
released from January 2000 to December 2006 and sorts each company
into one of the following nine industries: aviation and transport
services manufacturing; consumer business; public sector; energy and
resources; real estate; financial services; technology, media, and
telecommunications; life sciences; and health care. Doloitte [7] reports
that the number of financial statement frauds that the SEC issued has
decreased from 77 in 2003 to 44 in 2006. Although the number of
reported frauds has decreased dramatically after the Sarbanes–Oxley
Act was introduced in 2002, it is arguable whether the real number has
decreased by that much. In reality, we have seen more fraudulent
behaviors since 2008 when the economy has been going towards the
lowest point since the great depression in 1929.
It leads us to believe that fraud types and industry patterns
changed over time. It is important to understand how fraud schemes
have evolved and it is more important to predict by any means
possible, the direction of the change and keep the automatic fraud
detection techniques up-to-date. The findings may have significance
to develop more robust business processes as well as adaptive fraud
detection mechanism for managing/deterring/detecting the risk of
fraud. The efficacy of such processes will depend on knowing the fraud
schemes typically committed and industry-by-industry differences in
those schemes.
A majority of data mining-based financial statement fraud
detection techniques are primarily based on classification, which is
the process of explaining and differentiating data classes in order to
predict the class of objects whose class label is unknown. As compared
to association rule mining that deals with existing data items, the
classification rule mining deals with attributes and its values, clusters
the attributes, and uses time-series mining or outlier detection to
recognize the new mode for financial statement fraud detection.
Despite the early success of these pioneering FSF auto-detection
systems and algorithms, the rate of successful detection has
continuously decreasing over the past several years. Although it is
certainly possible that ethical standards have greatly improved since
2000, given the fact that the severity and quantity of law suits for
financial statement frauds have increased recently, we are more
inclined to believe that managements have adapted certain ways to
avoid being identified by automated detection systems.
Although according to the No-Free-Lunch theorem, existing data
mining search techniques, including random search, should perform
equally well on average, some techniques would have more accurate
estimation than others when appropriate domain knowledge can be
identified and integrated. In the next section, we propose a
framework that utilizes domain knowledge to facilitate the detection
of financial statement fraud in a constantly evolving economic
environment.
5. Adaptive FSF detection framework and methodologies
5.1. Framework to detect adaptive financial fraud
Similar to traditional classification procedure, in general we consider
two stages in our framework (Fig. 2). In the first stage, relevant external
and internal variables that differentiate the industries, economic
conditions, management's choice, timing considerations and any
other factors that have the potential to form domain knowledge are
selected and experimented. In the second stage, the financial data of the
firm in question is analyzed based on this domain knowledge learned
from the previous step. Certain detection strategy is formed accordingly
and the data is further analyzed using data mining techniques.
To realize the general model described above and to catch the real
world dynamics and possible new methods to commit financial
statement fraud, we propose an adaptive learning framework (Fig. 3).
We consider, but are not limited to, exogenous parameters described
in existing literature, such as capital structure, conditions, choices,
management attitudes, etc. The mechanism works as follows: based
 W. Zhou, G. Kapoor / Decision Support Systems 50 (2011) 570–575
Fig. 2. Proposed two-stage framework.
on external and internal economical circumstances, management
chooses their action on whether or not to commit financial statement
fraud at the year end. Financial data and statements of concern are
audited and examined by a fraud detection unit. Resulting audit
reports are further evaluated and learned by a self-adaptive module to
collect relevant patterns and trends of each company in different
industries. In the meanwhile, an adaptive fraud discovery module
keeps evolving with exogenous parameters to discover unknown but
possible pattern of financial statement fraud. New discoveries were
also evaluated and learned to prepare the knowledge base for future
fraud detection.
To increase detection relevancy and to reduce computational
complexity, we also propose adaptive feature selection that fits
specific domain to choose the proper parameters for companies with
similar internal and external financial environment [13]. Once relevant
parameters are selected, we opt to proper methodology and data
mining technique to detect financial fraud that evolves. As discussed in
the previous section, no single data mining-based FSF detection
technique is perfect and each of them is subject to its own handicaps.
We propose response surface methodology to construct the founda-
tion in order to find the right data mining-based detection technique.
5.2. Adaptive financial fraud detection with RSM
Response Surface Methodology (RSM), a method for constructing
global approximations to system behavior based on results calculated
at various points in the design space [23], is a natural fit to estimate
the relationships between the variables and the financial statement
fraud techniques. RSM provides statistically validated predictive
models that can be manipulated for finding the probability of different
forms of possible financial statement frauds.
RSM is extensively applied in various situations where output
performance or service quality, which is called response, is influenced
Fig. 3. FSF detection strategy selection to detect adaptive financial.
573
by a list of several input variables, which may or may not be complete
and are called independent variables. Independent variables are
subject to the control of the experiment designer. Approximation of
the relationship between the response and independent variables can
be visualized by RSM, which consists of three factors: (1) the
experimental strategy for exploring the space of independent
variables, (2) empirical statistical modeling to develop an appropriate
approximating relationship between the response and the indepen-
dent variables, and (3) optimization methods for finding the values of
the process variables that produce desirable values of response.
Theoretically, the appropriate approximating model between the
response y and independent variables x1, x2,…xn can be constructed as
y = f(x1, x2,…xn) + ε, where the form of the true real-time response
function f is unknown and may be complex. ε, which usually includes
measurement error on response and other unpredictable noise, is a
random term that represents the variability not caught by the
response function. If we assume that it has a normal distribution
with zero mean, Y = E(y) = f(x1, x2,…xn). With the form of the true
response function f kept unknown, the designer has to approximate it
and further utilize it to locate the possible response with discovered
independent variables.
Feature selection can be implemented to identify a relevant list of
variables that the designer may further utilize to construct the
response surface. Once we have a firm in question, we are able to find
the possibility of certain financial statement frauds based on the
response surface estimation. Then, we select data mining techniques
that suit the profile that we have learned from the previous step.
RSM provides statistical tools for analysis of historical data and
selection of variables aimed at better prediction. The objectives for
using RSM in the context of financial statement fraud detection are to
find the optimum response and to understand how the response
changes in a given direction by adjusting the design variables [17].
When there are constraints on the design data, then the variable
574 W. Zhou, G. Kapoor / Decision Support Systems 50 (2011) 570–575
selection and experimental design has to meet requirements of the
constraints. In general, the response surface can be visualized
graphically and the graph, if in fewer than three dimensions, is helpful
to navigate over the response surface to reach the desired outcome.
In a simple form, a function f (x1, x2) can be plotted versus the levels
of x1 and x2, and this three-dimensional graph forms a response surface
plot (Fig. 3). This figure shows an adaptive learning framework to
detect evolutionary financial statement frauds using a response
surface method (Fig. 4). We proceed by first selecting relevant
variables that could either be the three parameters in the CMA
model [15], the parameters in a 3C's model [19], a mixture from both
models, or any other parameters that have some causal relationship to
financial statement frauds. The selected variables are denoted as
Va ; Vb ; :::Vn
respectively, so the probability of committing financial statement
fraud in the form of k can be described as Eq. (2) that is subject to Eqs.
(3) and 4(4). We assume that the first n forms of financial statement
fraud have been discovered and that the number of possible forms is
infinity, such that
P ðFSFk Þ = f ðVa ; Vb ; :::Vn Þ
which is subject to
n
∑ P ðFSFk Þb1:
k=1
Fig. 4. Adaptive learning and detecting financial statement fraud using a response surface method.
6. Conclusion
In recent years, data mining has gained widespread attention and
increasing popularity in the financial world. Successful data mining
applications have been reported and recent surveys have found that
data mining has grown in usage and effectiveness. Professional
accounting bodies have also identified data mining as an important
technology for the new century. Implementing straightforward data
mining techniques to discover financial statement fraud, however, has
many disadvantages and limits of usage. After exhaustive literature
review, we find that a majority of existing data mining techniques to
detect financial fraud have their domain of usage and limitations.
Furthermore, when financial statement fraudsters have found ways
to circumvent the automatic detection programs, there is an urgent
need for a mechanism that is able to learn and use the industrial domain
ð2Þ
knowledge to facilitate the data mining techniques. We propose such a
framework that is based on the response surface method to
automatically pivot the program in accordance with the unique
circumstances of the firm in question. Unlike traditional financial
fraud detection techniques that are based on historical financial data,
we further propose an innovative way, the active discovery module that
evolves ahead of possible fraudsters. Preparing the intelligent detection
system in anticipation before any unknown or future fraud happens
enables us to effectively detect financial statement frauds that adapt.
ð3Þ Future research is needed to design the active discovery module that
is both effective and efficient. Furthermore, although we have suggested
the response surface method to extract the domain knowledge and to
adaptively learn the changes from fraudsters, there may exist
alternatives that have equal or better performance than the RSM.
ð4Þ Research is also needed to examine the circumstances under which our
suggested framework performs better than other techniques.
 W. Zhou, G. Kapoor / Decision Support Systems 50 (2011) 570–575
References
[1] M.S. Beasley, An empirical analysis of the relation between the board of director
composition and financial statement fraud, The Accounting Review of Finance 71
(4) (1996) 443–465.
[2] M.S. Beasley, J.V. Carcello, D.R. Hermanson, Fraudulent Financial Reporting
1987–1991: An Analysis of U.S. Public Companies, The CoSo Report, 1999.
[3] T.B. Bell, J.V. Carcello, A Decision aid for assessing the likelihood of fraudulent
financial reporting, Auditing 19 (1) (2000) 169–184.
[4] R.A. Bernardi, Fraud detection: the effect of client integrity and competence and
auditor cognitive style, Auditing: A Journal of Practice & Theory, Supplement 13
(1994) 68–84.
[5] R.J. Bolton, D.J. Hand, Statistical fraud detection: A review, Statistical Science 17
(3) (2002) 235–249.
[6] W.-S. Chen, Y.-K. Du, Using neural networks and data mining techniques for the
financial distress prediction model, Expert Systems with Applications Part 2 36
(2) (March 2009) 4075–4086.
[7] Deloitte, Ten things about financial statement fraud: a review of SEC enforcement
releases 2000–2006, Deloitte Forensic Center, June 2007.
[8] Y. Dianmin, W. Xiaodan, W. Yunfeng, L. Yue, C. Chao-Hsien, A review of data
mining-based financial fraud detection research, International Conference on
Wireless Communications, Networking and Mobile Computing, 2007,
pp. 5519–5522.
[9] K. Fanning, K.O. Cogger, R. Srivastava, Detection of management fraud: a neural
network approach, 11th Conference on Artificial Intelligence for Applications,
1995, pp. 220–223.
[10] E.H. Feroz, M.K. Taek, V.S. Pastena, K. Park, The Efficacy of red flags in predicting
the sec's targets: an artificial neural networks approach, International Journal of
Intelligent Systems in Accounting, Finance & Management 9 (2000) 145–157.
[11] J.V. Hansen, J.B. McDonald, W.F. Messier, A generalized qualitative-response
model and the analysis of management fraud, Management Science 42 (1997)
1022–1032.
[12] E. Kirkos, C. Spathis, Y. Manolopoulos, Data mining techniques for the detection of
fraudulent financial statements, Expert Systems with Applications 32 (4) (2007)
995–1003.
[13] H.C. Koh, C.K. Low, Going concern prediction using data mining techniques,
Managerial Auditing Journal 19 (3) (2004) 462–476.
[14] K.O.C., Kurt M. Fanning, Neural network detection of management fraud using
published financial data, Intelligent Systems in Accounting, Finance & Management
7 (1) (1998) 21–41.
View publication stats
575
[15] J.K. Loebbecke, M.M. Eining, J.J. Willingham, Auditors' experience with material
irregularities: frequency, Nature, and Detectability, Auditing: A Journal of Practice
and Theory 9 (1989) 1–28.
[16] M.D. Odom, R. Sharda, A neural network model for bankruptcy prediction, IJCNN
International Joint Conference on Neural Networks, 1990, pp. 163–168.
[17] G.W. Oehlert, Design and Analysis of Experiments: Response Surface Design, W.H.
Freeman and Company, New York, 2000.
[18] O. Persons, Using financial statement data to identify factors associated with
fraudulent financing reporting, Journal of Applied Business Research 11 (1995)
38–46.
[19] Z. Rezaee, Financial Statement Fraud-Prevention and Detection, John Wiley &
Sons, Inc., 2002.
[20] H.M. Schilit, Financial Shenanigans, McGraw-Hill, New York, NY, 2002.
[21] C. Spathis, Detecting false financial statements using published. data: some
evidence from Greece, Managerial Auditing Journal 17 (4) (2002) 179–191.
[22] C. Spathis, M. Doumpos, C. Zopounidis, Detecting falsified financial statements: a
comparative study using multicriteria analysis and multivariate statistical
techniques, European Accounting Review 11 (3) (2002) 509–535.
[23] N.S.R.T.H., W.J. Roux, Response surface approximations for structural optimization,
1998, pp. 517–534.
[24] H. White, Learning in artificial neural networks: a statistical perspective, Neural
Computation 1 (1989) 425–464.
Wei Zhou is an Assistant Professor in the Information and
Operations Management department at ESCP Europe. His
research interests include Internet advertising, RFID —
enabled item-level information visibility, data mining and
its application in supply chain management. His work has
appeared in Decision Support Systems, European Journal of
Information Systems, European Journal of Operational
Research, IEEE transaction on Geosciences and Remote
Sensing, International Journal of Electronic Commerce and
Optical Engineering, among others.
Gaurav Kapoor received his Ph.D. in Information Systems from the University of
Florida. His research interests include RFID systems.