Beneish M-Score
What it measures¶
The Beneish M-Score is a screen for earnings manipulation. It combines eight indices — each this year's ratio divided by last year's, so 1.0 means no change — with weights Messod Beneish estimated in 1999:
| Index | What rises when it rises |
|---|---|
| receivables to sales | money owed by customers grows faster than sales |
| gross margin | the margin fades (last year's divided by this year's) |
| asset quality | assets that are neither current nor property and equipment grow |
| sales growth | sales grow |
| depreciation | equipment is written down more slowly |
| selling and administrative costs | those costs grow faster than sales |
| total accruals | profit not backed by cash, against assets |
| leverage | liabilities grow against assets |
The weighted sum, plus a constant, is the M-score; a more negative score looks less like the manipulators in his study.
Why it is on the page¶
It is the best-known published screen for manipulated accounts. The Fundamentals page shows every index it can compute and refuses the total when one is missing — a partial score would be a different number wearing the same name. The condition summary's Manipulation screen check uses the same score.
Thresholds, their source and the population they were fitted on¶
The row's bar has three zones:
| Zone | M-score |
|---|---|
| below −2.22 | does not resemble the manipulators |
| possible | −2.22 to −1.78 |
| above −1.78 | resembles them — Beneish's published line |
- Source: Beneish (1999), "The detection of earnings manipulation".
- Fitted on: 74 companies charged with manipulation by the US securities regulator, 1982–1992, against a matched sample. At −1.78 it missed 26% of manipulators and wrongly flagged 13.8% of the others.
- Evidence grade: peer-reviewed and replicated.
A worked example¶
Celestica, as the Fundamentals page showed it on 9 August 2026: the full score read insufficient data, because one index needs a balance sheet from fiscal 2021 that the data provider left empty; the page showed the six indices it could compute and no total. Apple, as the condition summary computed it on 25 October 2026: −2.32, below −2.22.
What “No verdict” means here¶
The row carries no verdict of its own; the panel header shows the verdict of Earnings quality & red flags. The total reads insufficient data — {the missing input} whenever an index cannot be computed.
What the verdict is and is not¶
A score above −1.78 is a reason to look closer, not a finding of manipulation: fast-growing companies trip it often for innocent reasons, and about one honest company in seven did in Beneish's own study. It is not a forecast and not an instruction to trade the shares.