The Beneish M-Score is a statistical model that uses eight ratios taken from a company's financial statements to estimate whether its reported earnings are likely to have been manipulated. Each ratio compares the current year with the year before, and the weighted sum of the eight gives a single number. A score above a chosen cut-off marks the company as a possible manipulator that deserves closer examination.
The model was developed by Messod D. Beneish, a professor of accounting at Indiana University, and set out in a 1999 paper in the Financial Analysts Journal. It is used in forensic accounting, by auditors and equity analysts, and by investors engaged in short selling. It is often set beside other screens of quality of earnings, such as the accrual ratio.
Formula
The eight-variable version of the model is:
- , the days sales in receivables index, is receivables relative to sales this year divided by the same figure last year. A jump may point to revenue booked too early.
- , the gross margin index, is last year's gross margin divided by this year's. A value above 1 means margins have worsened, which raises the incentive to manipulate.
- , the asset quality index, tracks the share of assets other than current assets and property, plant and equipment. A rise can signal costs being deferred onto the balance sheet.
- , the sales growth index, is this year's sales divided by last year's. Fast growth is not manipulation, but growing firms are under more pressure to meet targets.
- , the depreciation index, compares the rate of depreciation in the two years. A value above 1 means assets are being written down more slowly.
- compares selling, general and administrative expenses relative to sales in the two years.
- , total accruals to total assets, is the gap between reported income and operating cash flow, scaled by assets.
- , the leverage index, compares the ratio of debt to assets in the two years.
In the original study a score greater than about -1.78 marked a likely manipulator. Later users often quote -2.22 as a more cautious cut-off. A simpler five-variable version drops the three indices that were not statistically significant.
Worked example
A company whose ratios are unchanged from the year before has every index equal to 1 and total accruals of zero. Its score is the constant plus the sum of the seven index weights: , comfortably below the cut-off.
Now suppose a company reports a receivables index of 1.4, a gross margin index of 1.1, a sales growth index of 1.5 and total accruals equal to 8 percent of assets, with the other indices at 1. Each departure from the neutral case adds to the score: for receivables, for margins, for growth and for accruals. The total addition is 1.241, giving . That is above -1.78, so the model flags the company.
How the model was built
Beneish compared a sample of companies that were known to have manipulated earnings, mostly firms subject to accounting enforcement actions by the Securities and Exchange Commission or forced to restate, with a much larger group of companies that had not. A probit regression produced the weights. The variables capture two ideas: distortions that manipulation leaves in the accounts, such as swelling receivables and accruals, and conditions that give managers a motive, such as shrinking margins and rising debt.
The model identified roughly three quarters of the manipulators in the sample while wrongly flagging a minority of honest firms. It became widely known after a group of business students at Cornell University applied it to Enron in 1998, several years before the company collapsed, and found a score in the manipulator range.
Use in practice
The score is a screening tool. Analysts compute it across many companies as part of financial statement analysis and then study the flagged firms in detail, looking at which components drove the result. A high receivables index leads to questions about revenue recognition, and a high accruals figure leads to a comparison of net income with cash generated. An auditor may use it in planning to decide where to direct testing. Later research by Beneish and co-authors reported that companies with high scores tended to earn lower stock returns afterwards, which drew the interest of quantitative investors.
Limitations
- False positives. Fast-growing companies, firms that have made acquisitions and businesses with changing sales terms often score high without any wrongdoing.
- Sample and period. The weights were estimated on US industrial companies in the 1980s and early 1990s. Accounting standards and business models have changed since.
- Financial firms. Banks and insurers were excluded from the sample, and the ratios have little meaning for them.
- Direction of manipulation. The model looks for overstated earnings. It does not detect companies that understate profit.
- Not proof. A score is a probability signal drawn from public figures. It cannot establish fraud, which requires evidence of intent.