Combining Belief Functions and Neural Networks to Assess the Likelihood of Fraud: The Case of Commercial Bank Audits

When assessing the likelihood of fraud in commercial banks, an auditor is faced with two related issues: determining significant red flags in the commercial banking industry, and combining red flags in a model (Decision Aid) based on weights (Values of uncertainties) assigned to them. Prior research largely ignores the first issue. Also, models developed in previous studies fail to provide objective methods to assign weights to red flags and to combine them. This study has two main objectives. The first objective is to identify red flags in the commercial banks context. The second objective is to develop a decision aid to assess the likelihood of fraud in commercial bank audits. To achieve the first objective, a questionnaire was directed to auditors in two big accounting firms in United States. Forty-four red flags were found to be valid. To achieve the second objective, a model combining belief functions and neural networks has been developed. This model is consistent with SAS No. 82, which requires the auditor to assess the likelihood of fraud as a cumulative process that should made throughout the audit.