New research pinpoints intentional financial misreporting
Source: Sunny Rosen
UD's Nerissa Brown has conducted research to develop a textual algorithm that successfully spots purposeful accounting misstatements.
Nerissa Brown, associate professor of accounting and MIS at the University of Delaware, offers a new and more effective approach to identifying accounting fraud in her new research.
"There have been quite a few studies developing measures to help us predict which firms are fudging the numbers or trying to look better than they really are financially," said Brown. These studies examine various measurements in the hopes that they will help predict if a firm is manipulating its accounting.
But Brown's research uses a new variable to identify fraud: the number of topics that a firm discusses in its annual financial reports and how this number deviates from the average firm.
"You know how sometimes when people are lying they tend to talk a whole lot?" asked Brown. "In some ways, we're trying to pick up instances like that."
"Firms who are engaging in fraud tend to talk about issues that are not important when compared to other firms in the industry," added Brown. "They also underreport important risk factors."
Recent statements from officials at the Securities Exchange Commission (SEC) highlight that regulators are taking notice of firms' discussion of benign topics.
With this in mind, Brown worked with with Richard Crowley and W. Brooke Elliott from the University of Illinois to develop an algorithm designed to detect the number of topics discussed in an accounting firm's 10-K. The software examined more than 130,000 of these annual reports, assessing over three billion words total.
"The algorithm runs through all the text in all the 10-K's and picks out the topics itself," said Brown. "Does the number of topics that you're talking about predict whether you're trying to hide something?"
Brown and her colleagues found that their model was highly successful, performing 2,083 percent better than traditional financial measurements at predicting accounting fraud between 2008 and 2012.
"The topic measure predicts instances of fraudulent accounting much better than examining the financials alone," said Brown. "This has larger implications for regulators and practitioners who run these computer-generated tools. They could think about developing a measurement of topics when building their detection models."
The Securities and Exchange Commission is currently using a computer-based analytical model, dubbed "RoboCop" in the financial industry, which relies heavily on financial metrics.
Brown and her co-authors are working to expand their analysis to learn exactly what types of topics are most predictive of fraud, and to detect when firms are omitting important topics.
"If one firm in the industry seems to skirt an issue, maybe this should set off a red flag," said Brown. "Why is one firm not talking about this, but everyone else is?'"
This research blends accounting with data analytics, a currently expanding field at UD. A new doctoral program in Financial Services Analytics (FSAN) begins this year and is the first of its kind. This cross-disciplinary program is offered by the Alfred Lerner College of Business and Economics and the College of Engineering.
Brown hopes to work with this program in the future and its "students who are interested in using data analytics from an accounting point of view."
| }
|