The Beneish M-Score: Detecting Earnings Manipulation Before the Market Catches On

Messod Beneish, a professor at Indiana University, published his earnings manipulation model in 1999. It takes eight variables off the income statement and balance sheet, weights them, and spits out a single number that estimates how likely a company is cooking its earnings. The famous story is that a group of Cornell business students, working from Beneish's model, flagged Enron as a probable manipulator well before the company collapsed. The model has since been pointed at plenty of other blowups after the fact and lit up on most of them.

What always surprises me is how few individual investors actually run it. Forensic accountants use it. Short sellers use it. It shows up in academic and regulatory analysis. But most people evaluating a stock never bother, even though the downside of holding a company that turns out to be faking its numbers is close to total. You do not lose 20 percent on those. You lose the position. That asymmetry alone is worth a ten-minute screen.

How the model works

The M-Score is a weighted sum of eight indices. Here is the full formula so you have it in one place:

M-Score = -4.84 + 0.920(DSRI) + 0.528(GMI) + 0.404(AQI) + 0.892(SGI) + 0.115(DEPI) - 0.172(SGAI) + 4.679(TATA) - 0.327(LVGI)

Every index compares this year to last year, so the model reads changes in behavior rather than absolute levels. That is the whole idea. A company with steadily high receivables is not interesting. A company whose receivables suddenly balloon relative to sales is. Here is what each piece is watching for.

DSRI, Days Sales in Receivables Index. Compares this year's receivables-to-sales ratio against last year's. When receivables grow much faster than sales, you have to ask whether those are real sales or whether the company is booking revenue it has not collected and may never collect.

GMI, Gross Margin Index. Flags a deteriorating gross margin. The logic is about motive: management watching margins erode has a stronger reason to reach for aggressive accounting to prop up the earnings line.

AQI, Asset Quality Index. Tracks the share of assets that are neither current nor property, plant, and equipment. A rising AQI can mean the company is capitalizing costs it should be expensing, which quietly inflates both assets and earnings at the same time.

SGI, Sales Growth Index. Fast growth is not manipulation. But high-growth companies live under intense pressure to keep the streak going, and when growth naturally slows, the temptation to pull revenue forward gets real. The model treats rapid growth as a risk factor, not a verdict.

DEPI, Depreciation Index. A falling depreciation rate suggests the company may have stretched the useful lives of its assets, which shrinks depreciation expense and lifts reported earnings without anything real changing underneath.

SGAI, SG&A Expense Index. This one carries a negative coefficient, so rising overhead actually pulls the score down. The rough intuition is that firms managing earnings tend to report suspiciously restrained expense growth, so unusually low SG&A growth is the more worrying signal.

TATA, Total Accruals to Total Assets. The heavyweight, with by far the largest coefficient. High accruals relative to assets mean reported earnings are being driven by accounting entries rather than actual cash coming in the door. When accrual earnings and cash earnings drift apart and keep drifting, pay attention.

LVGI, Leverage Index. Measures the change in leverage. It also has a negative coefficient, so rising leverage slightly lowers the score. It is the lightest-weighted input and rarely the thing that moves your result.

Reading the output

You get one number. Beneish set the cutoff at -1.78. Score above -1.78 and the company is flagged as a likely manipulator. Score below it and manipulation is unlikely.

Beneish calibrated that threshold deliberately, leaning toward catching real manipulators even at the cost of some false alarms. In his original testing the model correctly identified roughly three-quarters of the manipulators in the sample while also flagging a chunk of clean companies. Those false positives are not noise to be annoyed about. They tend to be companies running aggressive but legal accounting, which is exactly the kind of thing you want a second look at before you buy.

What you need to calculate it

You need two consecutive years of financials, both statements. Specifically: revenue, cost of goods sold, total current assets, property/plant/equipment, total assets, depreciation, SG&A, net income from continuing operations, cash from operations, current liabilities, long-term debt, and accounts receivable. All of it sits in the 10-K and the prior year's 10-K, both free on EDGAR. There is no proprietary data here. Anyone with a spreadsheet and twenty minutes can build the whole thing.

If you screen a universe of stocks, run the M-Score across all of them and flag everything above -1.78. Do not auto-reject the flagged names. A company can score high for boring, legitimate reasons: it is growing fast, it entered a new market with longer customer payment terms, so receivables and the growth index both spike. The flag is a prompt to look, not a reason to walk.

A quick worked example

Two of the indices are easy enough to compute in your head, so they are worth walking through. Say a company reports revenue of 1,000 last year and 1,100 this year, with receivables of 100 last year and 200 this year. All numbers here are made up to show the mechanics.

DSRI is the receivables-to-sales ratio this year divided by the same ratio last year. This year that ratio is 200 over 1,100, or about 0.18. Last year it was 100 over 1,000, or 0.10. Divide the two and DSRI comes out near 1.8. In other words, receivables per dollar of sales nearly doubled while revenue barely moved. That is a large contribution to the score and exactly the kind of pattern that should make you open the filings.

TATA is total accruals divided by total assets, where a simple version of total accruals is net income minus cash from operations. Say net income is 90 but operating cash flow is only 20, against total assets of 800. Accruals are 70, so TATA is 70 over 800, roughly 0.09. It looks small until you remember TATA carries the 4.679 coefficient, so even a modest ratio can dominate the result. A company reporting 90 of profit while collecting 20 of cash is precisely what that heavy weighting is built to catch.

What to do when a company flags

A high M-Score is an invitation to investigate, nothing more. Here is the order I work in.

Find the driver. Decompose the score and see which index is doing the damage. If TATA is carrying it, go straight to the cash flow statement and line up net income against operating cash flow. A persistent, widening gap between reported profit and cash generated is the single most reliable tell in the whole exercise.

Read the footnotes. This is where it lives. Look for changes in revenue recognition, newly capitalized costs, tweaked depreciation methods, or odd accrual entries. Companies disclose this stuff, but they bury it in dense language deep in the notes, which is precisely why almost nobody reads it.

Compare to peers. If receivables are climbing across the entire industry, your DSRI flag is probably a sector story, not a fraud story. If your company is the only one showing the pattern, it moves up the worry list.

Check the auditor. Look for a qualified opinion, a recent change of auditor, or any mention of material weakness in internal controls. Each of those is an independent red flag on its own, and it means a lot more when it lands on top of a high M-Score.

Known catches and what the backtests show

Beyond Enron, the model has flagged a long list of companies that later restated earnings or drew SEC enforcement, WorldCom among the well-known ones. It is not magic, and it does miss cases, but the hit rate on genuine accounting frauds is good enough that ignoring it is hard to defend.

The backtesting result I find most useful is not about average returns. On average, screening out high M-Score names moves the needle only modestly. The value shows up in the tails. The stocks that fall the hardest in any given year are heavily overrepresented among high M-Score companies, so the model earns its keep by keeping you out of the catastrophes, not by adding a point or two to a good year. That is a trade most long-term investors should happily take.

One honest limitation: the M-Score does not separate outright fraud from accounting that is merely aggressive. Plenty of flagged companies are not committing fraud at all, just running earnings hotter than the underlying cash supports. Even in those cases the flag is telling you something real, which is that the reported earnings are probably less durable than they look.

Pairing it with the Z-Score and F-Score

The M-Score gets more powerful next to the Altman Z-Score and the Piotroski F-Score, because the three are watching different things. The Z-Score estimates bankruptcy risk, with the classic distress zone sitting below roughly 1.8. The F-Score, from Joseph Piotroski's 2000 paper, grades whether the fundamentals are actually improving or quietly rotting. The M-Score asks whether the earnings are honest. None of them overlaps much with the others.

A company that clears all three, low distress risk, improving fundamentals, no manipulation flag, is standing on far more solid ground than one that only passes one or two. When three independent methods agree, the odds that you have misread the situation drop sharply.

For an existing portfolio, run all three quarterly across your holdings. A slipping F-Score catches operational decline early. A rising M-Score catches accounting games before the market prices them in. A falling Z-Score catches financial distress before it turns terminal. Kept together, they give you a lead time you do not get from watching the stock price, which usually moves last.