In fact, the linguistic patterns are so common that they can be coded into a computer algorithm and may eventually be used to detect fraud automatically as part of the U.S. Securities and Exchange Commission’s Accounting Quality Model (AQM) — also known as “Robocop.”
That’s an important step for the AQM, which current focuses on mathematical indicators of financial misstatements and fraud but does not actually listen to what companies say.
“We have studied what companies talk about in the filings, such as 10-Ks, and look for linguistic patterns that are indicators of financial misreporting,” says Nerissa Brown, associate professor of accountancy at the University of Delaware and one of the authors of the study. “Instead of just looking at commonplace words associated with fraud, we measure the content and understand it in context.”
According to the study, so called “textual analysis methodology” can be used to scan company 10-K filings to predict which filers are more likely to issue a financial restatement that could be indicative of fraud.
Using heavy-duty computing power that studied 130,000 filings and 3 billion word combinations, the algorithm developed as part of the research created so-called “topics” that “provide significant incremental predictive power over financial and style variables for detecting intentional managerial misreporting.”
For example, the research states that “topics” where financial reporting problems are typically discovered include “liabilities”, “warranties”, “investment” and “debt repayments.” Those topics can then be matched with a larger word list and weighted for likely financial misrepresentation.
“It looks out for certain words and how they are connected together,” Brown says. “It also looks at style characteristics, the use of different words and measures them to build a more holistic model. Companies that are focused on misrepresenting themselves tend to use more words, more litigious words and are generally more careful about what they discuss. There are definite patterns.”
Although the research was just published, Brown says she hopes the study's findings will eventually make their way into the AQM. “Our motivation for the research stemmed from the development of Robocop. We feel it could be an important next step in its development.”