Key Scientific Paper Redux: How pervasive is corporate fraud?

At Deception And Truth Analysis we developed our algorithm based on deception science. That is, for many decades, deception scientists have researched, identified, and replicated their studies about the behavioral differences between deceivers and truth-tellers. We then use Natural Language Processing to look for those behaviors that lend themselves well to a NLP approach. Our method has resulted in, what we believe is, a general deception and truth algorithm that works well in many different domains.

That said, we have been curious about the success of our algorithm because we do not have, nor does anyone have, the resources to fact check every statement of a document to establish ground truth. Instead, we have to rely on ground truth datasets or established norms as an indirect test of the success of our algorithm. Enter the scientific research entitled, “How pervasive is corporate fraud?”[i]This meta-analysis not only sought to establish what level of fraud is caught by companies, their auditors, and regulators, but also the amount of fraud that remains un-caught. In other words, it sought to quantify both the tip of the fraud iceberg, as well as the entirety of the fraud iceberg.

Researchers in “How pervasive is corporate fraud?” sought to estimate the amount of undetected fraud. As a part of this analysis, they relied upon the amount of suspected malfeasance and fraud by four methods:

1. Auditor-detected
2. Accounting restatements
3. All securities fraud falling under SEC Rule 10b-5 [Note: this category are actual lawsuits brought by the Securities and Exchange Commission. Given that agency’s limited resources they are likely to only bring these cases where fraud is most easily proven. Therefore, the authors of the paper consider this category to be fraud, rather than just corporate malfeasance.
4. SEC Accounting and Auditing Enforcement Releases

Additionally, by using the techniques of meta-analysis, the researchers were able to develop a pooled measure of how much suspected fraud is discovered in the United States; that is, the “tip of the [suspected fraud] iceberg.”

Next, the researchers also relied on previous researchers’ work in estimating the fraud detection likelihood by using the above four methods. From these estimates, the authors of “How pervasive is corporate fraud?” were then able to estimate how much fraud goes undetected; that is, the rest of the [suspected fraud] iceberg. Last, using statistical methods a 95% confidence interval was placed around their estimates.

1. The sampling period for detecting malfeasance was during 30 November 2001 (i.e. Enron) through 2003. Covered is the period just after the fall of Enron when scrutiny around reporting malfeasance and fraud in the corporate sector of any kind was at a generational high. This presumably leads to a conservative estimate of the amount of total malfeasance and fraud, seen and unseen, given that the number of seen malfeasance and frauds was likely highest during this era.

2. For the four methods described earlier, the following are the estimates for the percentage of suspected malfeasance and frauds detected. Based on other researchers’ work about the likelihood of detecting malfeasance and fraud the researchers assumed that malfeasance and fraud pervasiveness follows a Poisson Distribution. Below then are the “tip of the iceberg” estimates for malfeasance and fraud pervasiveness:

   a. 29.4% = Auditor detected

   b. 33.7% = Accounting restatements

   c. 47.0% = All securities fraud, SEC 10b-5

   d. 52.0% = SEC Accounting & Auditing Enforcement Releases

3. For #2, above, the pooled estimate for the percentage of suspected frauds detected = 38.4%

4. The figures in #2, above, also imply an estimate for the % of undetected malfeasance, or “the hidden icebergs,” by simply subtracting the figures above from 100%, or:

   a. 70.6% = Auditor detected

   b. 66.3% = Accounting restatements

   c. 53.0% = All securities fraud as represented by SEC 10b-5 complaints

   d. 48.0% = SEC Accounting & Auditing Enforcement Releases

   e. 61.6% = pooled estimate for the number of suspected frauds undetected

5. For the four methods of fraud detection, here is the breakdown of suspected frauds detected during the sampling period:

   a. 21 = Auditor detected, or 0.81% of malfeasance detected in this way

   b. 168 = Accounting restatements, or 13.46% of malfeasance detected in this way

   c. 63 = All securities fraud, SEC 10b-5, or 3.39% of malfeasance detected in this way

   d. 59 = SEC Accounting & Auditing Enforcement Releases, or 2.64% of malfeasance caught in this way

6.  The above figures in #s 2 and 5, above, allow us to calculate the pervasiveness of malfeasance, as follows:

   a. Auditor detected = 0.81% / 29.5% = 2.76%

   b. Accounting restatements = 13.46% / 33.7% = 39.95%

   c. All securities fraud, SEC 10b-5 = 3.39% / 47.0% = 7.21%

   d. SEC Accounting & Auditing Enforcement Releases = 2.64% / 51.9% = 5.08%

7. The above figures in #s 5 and 6, above, allow us to calculate a weighted average “tip of the iceberg” estimate for all types of fraud of 5.68%

8. With all of this data we can derive a best estimate of how pervasive fraud is, how much is caught and uncaught. Specifically:

   a. Frauds caught by all methods = 5.68%

   b. Frauds uncaught = 18.51%

   b. Estimated total frauds, caught and uncaught = 24.19%

The authors utilize their numbers in other insightful ways, too as a check on their work. For example:

9. The authors conclude that the best estimate of detection likelihood is their pooled measure that includes both the “auditor detected” category of 29.4%, and “restatement measures” category of 33.7%. Their choice of these two methods is based on the idea that either the auditors or companies themselves are catching the potential frauds and announcing them and so likely there was some form of malfeasance present.

When a weighted average is taken of these two fraud detection methods then roughly 33% is their best estimate of the “tip of the iceberg” frauds. In other words, and crucially, 67% of frauds are never detected! 

10. Last among the insights contained in this paper, using insights of the estimates of the cost of corporate malfeasance and fraud from other researchers. Namely, that detected fraud (33% of cases as estimated in this paper) costs malfeasant and fraudulent firms 25% of their market cap, while undetected malfeasance and fraud (67% of cases) costs 10.9% of market cap, for a blended cost of 15.6% of market cap. In dollar terms this figure in 2021 was $830 billion.

The above figures derived by the authors of How Pervasive is Corporate Fraud? for the number of companies engaged is some form of misconduct is 24.19%, from 8.b, above. This figure comports very well with D.A.T.A.’s own figures of the percentage of companies that score as high risk or likely engaged in deceptive behavior (i.e. those that have negative DATA Scores) of 27.18%. 

Additionally, the authors’ figure of 5.68% of companies actually caught conducting fraud matches well the number of companies in D.A.T.A.’s DATAbase product whose D.A.T.A. Scores are two standard deviations below the average DATA Score, or 3.60%.

At D.A.T.A. we consider the work of the authors of “How pervasive is corporate fraud?” as a form of independent validation that our method of evaluating deception works accurately. This is because the researchers measured actual frauds and used statistical techniques to estimate how much fraud is undetected. Whereas, at D.A.T.A. we measure the amount of deceptive and truthful behaviors present in documents and transcripts based on a deep understanding of deception science. These two methods for estimated fraud pervasiveness are completely independent of one another. Yet, the figures roughly agree with one another.

[i]Dyck, Alexander, Adair Morse, and Luigi Zingales. “How pervasive is corporate fraud?” Review of Accounting Studies. 5 January 2023