Accounting Fraud Detection: Is it Possible to Quantify Undiscovered Cases?

Abstract:     
Accounting fraud is defined as intentional misstatement of financial reports, in violation of generally accepted accounting principles, with the objective of making certain people act in detriment to their best interests. It is possible to identify the determinants of fraud using an econometric model, but the dependent variable (occurrence of fraud in a given company) is vulnerable to misclassification: Not every case of fraud will be detected, thus false negatives are possible. This paper estimates the percentage of undiscovered frauds and also estimates a probit model to detect fraud in US companies. The dependent variable was built using the instances of fraud discovered by the Securities and Exchange Commission (SEC). The model was estimated only with frauds that occurred between 1998 and 2002 (since many cases of fraud from the last years are still unknown). The independent variables were chosen using the concept of fraud triangle. The financial statement data were obtained using Compustat. The results show that the likelihood of fraud is negatively related to the current ratio, to the cash change (scaled by total assets) and to the fixed assets (also scaled by total assets). Companies that changed their auditors or receive a qualified auditing report are more susceptible to fraud. The probability that a case of fraud is not detected was estimated as 97.61%; this means just a small part of fraud cases are discovered by the SEC.

Number of Pages in PDF File: 24

Keywords: Accounting fraud, AAER, Misclassification, Probit

working papers series