Every stock screen, valuation model, and growth projection starts from the same place: the financial statements. If those numbers are unreliable, everything built on them is wrong. Messod Beneish published the M-Score in 1999 at Indiana University to address exactly this problem. The model computes eight financial ratios from two consecutive years of annual statements, combines them into a single probit score, and flags companies whose patterns match known manipulators. A composite score above -1.78 means the company is statistically likely to have manipulated its earnings. Below -2.22 is considered safe. The zone between is ambiguous.
This guide walks through the M-Score step by step: running the basic screen, breaking down individual indices, combining it with the Piotroski F-Score, tracking changes over time, and validating the model against historical cases. For the screener overview and comparison table, see the Beneish M-Score Screener.
Step 1: Run the basic M-Score screen
The composite score is the starting point. Claude computes all eight indices from the two most recent years of annual financials, applies Beneish's probit formula, and classifies each company into one of three zones: manipulation likely (above -1.78), grey zone (-2.22 to -1.78), or safe (below -2.22).
"From the last two years of annual financials, compute the Beneish M-Score for all US stocks above $1 billion market cap. The eight year-over-year indices: DSRI (receivables/revenue), GMI (gross margin), AQI (asset quality), SGI (sales growth), DEPI (depreciation rate), SGAI (SG&A/revenue), LVGI (leverage), TATA (accruals/total assets). Formula: M = -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. Show the score, the classification (above -1.78 = manipulation likely, -2.22 to -1.78 = grey zone, below -2.22 = safe), and the sector."
Expect 50 to 100 companies to flag at any given time, depending on market conditions. Most will be in technology, biotech, and other high-growth sectors. This is not because those sectors are more fraudulent. It is because the M-Score's Sales Growth Index (SGI) penalizes rapid revenue growth, which mimics the pattern left by fabricated sales. The next step separates real concerns from growth artifacts.
Step 2: Break down the individual indices
The composite score tells you a company flagged. The individual indices tell you why. There are eight of them: DSRI (receivables growing faster than revenue), GMI (declining gross margins), AQI (rising non-current assets outside PP&E), SGI (revenue growth rate), DEPI (slowing depreciation), SGAI (SG&A rising relative to revenue), LVGI (increasing total debt relative to assets), and TATA (accruals that never convert to cash). Each has its own red-flag threshold from Beneish's original research.
"For NVDA, compute each of the eight Beneish indices from the last two years of annual financials: DSRI = (receivables/revenue this year) / (receivables/revenue last year), GMI = (gross margin last year) / (gross margin this year), AQI = (1 - (current assets + PP&E)/total assets this year) / same last year, SGI = revenue this year / last year, DEPI = (depreciation rate last year) / (depreciation rate this year), SGAI = (SG&A/revenue this year) / same last year, LVGI = (leverage this year) / (leverage last year), TATA = (net income from continuing ops - operating cash flow) / total assets. Flag any index above its red-flag threshold: DSRI > 1.465, GMI > 1.193, AQI > 1.254, SGI > 1.607, DEPI > 1.077, SGAI > 1.041, LVGI > 1.111, TATA > 0.018."
NVIDIA will likely flag on SGI (revenue grew roughly 65% year over year) and AQI (massive data center infrastructure spending shifted asset composition). Both are genuine business changes, not manipulation. The individual breakdown is how you make that call. A company flagged by one or two indices driven by real growth is different from a company flagged across four or five indices with no obvious business explanation. TATA is the most reliable single index. A high TATA means reported earnings consistently exceed cash from operations, the definition of low-quality earnings regardless of whether manipulation is involved.
Step 3: Pair with Piotroski for forensic + strength screening
The Beneish M-Score and the Piotroski F-Score answer complementary questions. Beneish asks: are the financial statements reliable? Piotroski asks: is the company getting financially stronger? A stock flagged by both, suspicious earnings and deteriorating fundamentals, is a convergent negative signal. A stock flagged by Beneish but scoring 8 or 9 on Piotroski is almost certainly a false positive: strong, improving financials that happen to look unusual because of growth.
"From annual financials, compute the Beneish M-Score (M = -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) for US stocks above $2 billion market cap. Flag any with M-Score above -1.78 AND Piotroski F-Score below 5. Show both scores, trailing P/E, and sector."
This is the screen with the lowest false positive rate. Companies that flag on Beneish and also score poorly on Piotroski are both statistically suspicious and financially weak. The overlap is small, typically 10 to 20 companies above $2 billion. Each one is worth investigating further: check insider selling patterns, recent 8-K filings, and whether the company changed auditors.
Step 4: Track score changes over time
A single M-Score snapshot tells you the current state. The trend tells you more. A company that was safely below -2.22 last year but crossed above -1.78 this year shows a shift in how it reports its financials. Something changed, and whether the cause is manipulation, a legitimate business transition, or an accounting policy change, it is worth understanding.
"Compute the Beneish M-Score (M = -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) for each of the last three years of annual financials, for all US stocks above $500 million market cap. Find companies that crossed from below -2.22 (safe) to above -1.78 (manipulation likely) in the most recent year. Show both years' scores."
Cross-reference the results with insider trading data. If the M-Score deteriorated and executives were selling stock during the same period, the behavioral signal and the statistical signal are pointing in the same direction. Neither one alone is conclusive. Together they narrow the investigation.
Step 5: Validate against historical cases
The acid test for any model: does it work on data it was not trained on? Shibui has 20+ years of annual financials, so you can compute the M-Score at any point in the past and check whether flagged companies later restated earnings, suffered price declines, or faced regulatory action.
"Compute the Beneish M-Score (M = -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) as of fiscal year 2019 for all US stocks above $5 billion market cap. What percentage of flagged companies (score above -1.78) declined more than 30% over the following 12 months?"
Important limitations. Survivorship bias applies: companies that went bankrupt or were delisted are not in the dataset, so the worst outcomes are missing. The model was calibrated on pre-2000 data, and its accuracy on modern financials has been debated in academic literature. Different studies report detection rates ranging from 50% to 82%, depending on the sample and time period. The M-Score is better at confirming suspicion than generating it. If you already have a reason to investigate a company, it helps. As a blind screener across the market, expect noise.
What Shibui cannot do
The M-Score is computed from reported financial statements. It does not catch:
- Revenue recognition fraud that does not change the year-over-year ratios (consistent manipulation at a constant level).
- Off-balance-sheet liabilities or related-party transactions that never appear in the standard financial statements.
- Qualitative red flags: auditor changes, restatement history, management turnover, or whistleblower reports. These require SEC filing analysis, not ratio screens. See the 13D/13G filing tracker for ownership-based signals.
- Banks and insurers with incomplete data. Cost of revenue is not applicable to financial companies, so the Gross Margin Index cannot be computed for them.
Frequently asked questions
What financial data does the Beneish M-Score need?
The Beneish M-Score requires two consecutive years of annual financial statements. It computes eight indices from revenue, receivables, gross profit, total assets, current assets, PP&E, depreciation, SG&A expenses, long-term debt, current liabilities, net income, and operating cash flow. Shibui has all of these fields for 10,000+ US stocks going back 20 or more years, so Claude can compute the M-Score for any company or screen the entire market.
Why do high-growth companies flag on the Beneish M-Score?
High-growth companies trigger the M-Score because rapid revenue growth inflates the Sales Growth Index (SGI), and large capital investments shift the Asset Quality Index (AQI). These patterns look identical to the patterns left by companies inflating revenue or capitalizing expenses they should not. A company that doubled revenue legitimately and one that fabricated the growth show similar year-over-year ratio shifts. The M-Score cannot distinguish motive, only pattern. This is the most common source of false positives.
Can the Beneish M-Score detect fraud before it is announced?
In some cases, yes. Beneish's original research demonstrated that the model would have flagged Enron before the fraud became public. Academic studies have found detection rates around 76% for known manipulators. However, the false positive rate is high. Many flagged companies turn out to be legitimate high-growth businesses or companies going through unusual but honest transitions. The M-Score is a probability screen, not proof of fraud. It narrows the list of companies worth investigating.
How do you combine the Beneish M-Score with other screens?
The most useful combination is Beneish plus Piotroski. A company flagged by Beneish (earnings manipulation likely) and scoring poorly on Piotroski (financially weak) shows convergent negative signals. A Beneish flag with a high Piotroski score is usually a false positive from growth. You can also pair Beneish with insider selling data: executives selling stock while the M-Score is flagged adds a behavioral signal to the statistical one. On Shibui, you describe any combination in plain English and Claude runs it in one query.