Detecting corporate fraud and assessing the relative risk factors have been significant issues confronting the auditing profession for decades. This study therefore aims to apply a neural network system to predict fraud litigation for assisting accountants on audit strategy making. The empirical results show that neural network provides not only a promising predicting accuracy, but also a better detecting power and a less misclassification cost comparing with that of a logit model and auditor judgments. This suggests that an artificial intelligence technique is quite well in identifying a fraud-lawsuit presence, and hence could be a supportive tool for practitioners. Further, a remarkable finding related to the greater effects of management’s capability on fraud commitments acquires an attentive investigation of ethic issues in emerging markets where contribute the most important force in the global economy nowadays.
Since Enron and WorldCom’s collapse, a series of financial statement frauds have not only surprised local and international capital markets, but also awaked the authorities to have a prime interest in the problem. SAS No. 82 emphasizes auditor’s responsibility and practically provides auditors with “how to” guidance on fraud detection whereas SAS No. 99 steps further into a brainstorming session that requires auditors to interact with audit team members to discuss fraud and to document the discussion. Also, researchers have largely addressed themselves to examine various aspects of the exploration of fraud models (Eining et al., 1997 and Hansen et al., 1996); the magnitude of “red flags” or “red-flag cues” evaluated (Apostolou et al., 2001 and Bell and Carcello, 2000); and audit decision support system implementation (Galderon & Cheh, 2002). Unfortunately, one thing holds true for the results is that financial scandal is still an everlasting hazard.
Noteworthy is that the professional literature makes it clear that failure in financial statement fraud detection rests with auditor’s insufficient capability and decision aid system’s inherent limits. Audit judgment consistency has been witnessed with being subject to auditor’s work experience (Messier, 1983) as well as knowledge and ability of problem solving (Bonner & Walker, 1994), so that leads the audit decisions encountered in today’s complex business environment to cover with a layer. Besides, models recently developed in fraud prediction have also been questioned about their intrinsic restrictions. The statistical assumptions such as linearity, normality and independence among variables of multi-variate discriminant analysis and regressions (Ohlson, 1980 and Zmijewski, 1984) have constrained their applications whereas binary models such as logit and probit (Gessner, Kamakura, Malhortra, & Zmijewski, 1988) have been criticized for the subjective determination of cutoff points. Further, a recent trend to develop the artificial intelligence has brought a new alternative. The expert system can embed the past experience into the system; fuzzy logic can describe the problem in a way that is close to the human reasoning process and accommodate the inaccuracy and uncertainty associated with the data. However, their difficulty with the acquisition of the knowledge base has been challenged. Accordingly, a call for improvement of early warning system has inspired this research.
Neural network, an approach has been applied to several audit fields: the assessment of material misstatements, the evaluation of management fraud, the issue of audit opinion, the prediction of financial crises, the assessment of internal control systems and the decision of audit fee (Deshmukh and Talluru, 1998 and Koskivaara et al., 2004). Its efficacy has been tested; it is limited, in most cases, in the USA. This study, therefore, is to adopt neural network to predict a fraudulent litigation crisis with data of Taiwan. Specifically, the proposed contributions of the study are two-fold: as a tool for eliciting knowledge of fraud risks and as a vehicle for supporting auditor’s learning. The following section reviews the framework of neural network. Next, the research design and the results are discussed. The final section summarizes the findings.
