Detecting manager’s fraud risk using text analysis: evidence from Iran

In accounting and finance, researchers have used many ways to detect manager’s fraud risk. Until now, many researchers have used some data mining methods in these two fields to detect this risk. The purpose of this paper is to compare the precision of two data mining methods in detecting such a risk.

For this purpose, this paper analyzed the texts of board’s reports and used two methods including the convex optimization (CVX) method and least absolute shrinkage and selection operator (LASSO) regression method. In this way, the words of these reports, which have the greatest power in explaining the manager’s high fraud risk index, were identified. Using these words, this paper presented a model that could detect manager’s high fraud risk index in companies.

The results indicated that both methods can detect the manager’s high fraud risk index with a precision between 82.55 and 91.25 percent. The LASSO method was significantly more precise than the CVX method.

The lack of access to an official and reliable list of firms suspected to fraud and the lack of access to the Microsoft Word (MS Word) file of board’s reports were two of the most important limitations of this study.

Regulatory bodies and independent auditors can consider the proposed methods in this study for assessing the fraud risk for a firm or other legal parties.

This paper avoided using merely financial statements data to detect the manager’s fraud risk index and focused on texts of board’s reports for the detection process. The capabilities of data mining and text mining methods for detecting the manager’s fraud risk index using board’s reports were tested in this paper. By comparing CVX and LASSO results, this paper indicated that methods with a binary-dependent variable have more power and are more precise than methods with continuous-dependent variables for detecting fraud.