Our project aims to detect potential earnings misstatements in company accounts filed with the Securities Exchange Commission (SEC). But what are financial misstatements? There is no set way to detect potential earnings manipulation, however there are a number of measures, or scores, that can flag company accounts for further investigation.
To make this simple, we are looking at numbers in public companies’ annual reports, that raise questions, and may no longer reflect the company’s day-to-day operations.
In the past there have been several kinds of misstatements. A 2011 paper published in Contemporary Accounting Research compiled all the misstatements reported by the SEC in what are called Accounting and Auditing Enforcement Releases (AAERs) between 1982 and 2013.
Financial misstatements in SEC’s data are under-reported as the commission has a limited budget. A lot of cases are not exposed. “Firms selected often have already admitted a “mistake” by restating earnings or having large write-offs,” researchers wrote in 2011. “Many firms that manipulate earnings are likely to go unidentified.”
The researchers also suspect a selection bias because the SEC might pursue cases that move the markets – “since the identifiable losses to investors are greater.” The researchers conclude that selection bias “is a general concern when analyzing the determinants of earnings manipulation.” For us it seems best to go with the official AAER data from the SEC to understand the areas of accounting that earnings manipulation is likely to affect.Based on the AAER data, the researchers broke down financial misstatements into 11 categories. In each case, the company does not properly report an accounting number, which can be:
Earnings Inspector uses a statistical model developed by professor Messod D. Beneish from the Kelly School of Business at the University of Indiana. In 1999 Beneish published his first paper about the M-Score. His model uses eight variables to detect potential earnings manipulations, focusing on extremely quick growth, deteriorating fundamentals and aggressive accounting practices.
For his calculations, Beneish used a sample of companies whose manipulations were publicly discovered. “Such companies probably represent the upper tail of the distribution of companies that seek to influence their reported earnings – successful and undiscovered manipulators undoubtedly exist – so the evidence should be interpreted in that light.”
It is important that journalists further investigate these companies since Beneish's model has a large rate of errors: 13.8% of non-manipulators are classified as manipulators. The distortions in the financial statements could also be “a material shift in the company’s value-maximizing strategy, or a significant change in the company’s economic environment.”
The model used financial information for publicly traded companies. It cannot be reliably used to study privately held companies. The model also can only be used to detect earnings overstatement and will not detect companies that are conducive to decreasing earnings.
One indicator of earnings manipulation is fast revenue growth. Sales growth is not negative in general but a company’s high growth in the past may increase its predisposition for manipulation. Other red flags include deteriorating gross margins and increasing administrative expenses. Manipulated financial statements can also show increasing debts and more soft assets other than plant, property and equipment. Other variables track possible aggressive accounting: receivables that are growing faster than sales, depreciation expenses that are slowing down and large income-inflating accruals.
Beneish's model compares several accounting variables from one year to the following, using this formula:
M-score = −4.84 + 0.920(DSR) + 0.528(GMI) + 0.404(AQI) + 0.892(SGI) + 0.115(DEPI) − 0.172(SGAI) + 4.679(Accruals) − 0.327(LEVI)
If the M-Score is above -2.22 this flags possible earnings manipulation. If it is greater than -1.78, the probability of manipulated earnings for this company is 76%.
Four of these variables show distortions that result from manipulation:
Earnings Inspector uses a statistical model based on Benford’s Law, also known as the First-Digit Law. The law is based on the frequency of distribution of digits in a dataset. It works by predicting the percentage of time a particular digit will occur as the leading digit in a set of numbers.
The model predicts that the number “1” is the most commonly occurring first digit in datasets, appearing 30% of the time. The likelihood decreases consistently to “9”, which is predicted to appear in 4.6% of cases.
This benchmark is then used to compare the distribution of digits in other datasets. In this case the dataset is the entire set of accounts provided by a company, as reported in our dataset.
If the frequency of distribution differs from that found in Benford’s Law, Earnings Inspector will flag that the accounts “are not consistent” with the rule. This means that it could be useful to look into the accounts further.
The ‘law’ was first noted in its most basic form by astronomer Simon Newcomb in 1881 and further developed later by physicist Frank Benford in 1938. However it is still used widely today. In the journalism world, the European datajournalism agency J++ developed a Benford's Law calculator in 2013.
Benford’s Law is used by the Internal Revenue Service (IRS) to detect problematic tax preparers and flagging those who are potentially misstating their tax returns for further investigation.
Earnings Inspector uses a statistical model developed in 1968 by professor Edward Altman, then Assistant Professor of Finance at New York University. The Z Score differs from the other metrics listed above insofar as it doesn’t detect earnings manipulation but the likelihood of bankruptcy.
Altman based his analysis on data for publicly held manufacturers. In 2012, Altman developed an updated version of the formula, which extends the Z Score to non-manufacturing companies as well as to private companies. Earnings Inspector currently only uses the original Z Score.
A video explanation can be found here.
The Altman Z Score compares seven variables using the following formula:
Z-Score = 1.2A + 1.4B + 3.3C + 0.6D + 1.0EWhere:
Companies with a score above 3.0 are considered unlikely to go bankrupt. A score between 3.0 and 1.8 is in the grey area and indicates that the company could be headed towards bankruptcy. A company with a score lower than 1.8 is likely to go bankrupt. The lower the score, the higher the risk of bankruptcy.