consequence, the fraudulent nancial statements can corrode public condence on the
reliability of nancial reporting as a means to assess a rm’s future prospects.
Financial frauds, unlike other hard crimes such as murder and rape, are harder to prove
in the court of law. A study by the Association of Certied Fraud Examiners (ACFE) (2012)
concluded that nancial frauds normally take between three and six years to detect. Usually,
by the time nancial frauds are detected, their related evidences have either been removed or
distorted. There are also no scientic tests, such as DNA or ngerprint, that can be
conducted. Just like any other serious illnesses, fraud is best “prevented than cured”.
Due to its seriousness, many fraud detection methods had been especially implemented to
help the auditors to detect, or better yet, to predict, fraudulent nancial reporting because
they are the rst layer of defence that could prevent fraud activity from spreading. Some of
the most commonly used methods of analysis are trend analysis, common-sized nancial
statement analysis, nancial ratio, Beneish model and Altman Z-score. Of late, researchers
have also proposed the use of data mining techniques to predict the occurrence of fraud in
Data mining is a data processing tool that uses sophisticated data search capabilities and
statistical algorithms to discover patterns and correlations in large data. Some of the
proposed data mining algorithms that can be used to predict fraudulent nancial reporting
are logistic regression, probit regression, decision trees and Bayesian networks and articial
neural networks (ANNs) (Kirkos et al., 2007;Kotsiantis et al., 2006;Liou, 2008). Nevertheless,
the use of these newer techniques is either limited or not adequately reported and
documented in literature.
Premised on the purported advantages of the use of these techniques, this study explores
the use of ANNs to predict fraudulent nancial reporting in small market capitalization
companies in Malaysia. This study focuses on fraudulent nancial statements, which is also
known as management fraud (Kranacher et al., 2011;Zimbelman and Albrecht, 2012).
Management fraud involves top management’s deceptive manipulation of nancial
statements. The reason the top management commits the fraud is to make the company look
better than it is. This is the most expensive type of fraud. The common examples of
management fraud are Enron, WorldCom, Waste Management, Parmalat and Satyam,
among others. After examining two sample types, namely, fraud companies which were
charged by the Securities Commission for presenting falsied nancial statements and
matched with non-fraud companies which were selected among small market cap
companies, this study found that ANN could give a higher prediction result (94.87 per cent)
on fraudulent nancial reporting model compared to traditional statistic, linear regression
(92.4 per cent) and other techniques as in previous study.
Previous studies have focused mainly on large capitalization companies (Abbott and
Parker, 2000;Klein, 2002;Lin and Hwang, 2010;Lin et al., 2006), with little attention having
been paid to the study of mid-cap and small-cap companies (Kang et al., 2011). Regulators,
analysts, investors and the public have continued to focus their attention on big companies
because of the effect of failure of big companies towards them (Arnott and Wu, 2012),
meaning that lower levels of scrutiny provide the opportunity for the small-cap companies to
commit fraud. Thus, it is crucial to study on small market-cap companies.
This study is signicant in the Malaysian context, as there is limited study on adapting
ANN in predicting fraudulent nancial reporting, especially using small market cap as the
sample. Nevertheless, there is limited exploration on specically using multilayer feed
forward neural network (MLF) in the Malaysian accounting eld, especially in detecting and
predicting fraudulent nancial reporting. Instead of helping in extending the literature on
both ANN and small market-cap companies as well as the practitioners, this study is also