Beneish M-score — 8-variable academic model, fully transparent per item
Accounting problems usually leave statistical traces before they blow up. The Beneish M-score is academia's standard earnings-manipulation detector: 8 financial-statement variables weighted into one score, with M > -1.78 flagged red. We recompute it deterministically from official filings, disclosing every variable — and honestly imputing neutral values where an input is unavailable.
What the 8 variables watch
The model probes growth quality: receivables outpacing sales (DSRI), deteriorating gross margins (GMI), declining asset quality (AQI), abnormal sales growth (SGI), rising leverage (LVGI), and more. No single variable is conclusive — the flag fires only when the composite exceeds the threshold. We return every variable with its availability status; neutrally-imputed inputs are labelled as such.
A red flag is not an accusation
The M-score is a statistical screen. A high score means the filing pattern resembles historical manipulation cases more closely than average, which is a reason for further diligence, not an audit conclusion and not a claim that a company did anything wrong.
Computation is deterministic and the formula is public (Beneish, 1999), so the same filings always produce the same score and a backtest reproduces exactly. Where Taiwan filing granularity leaves an input unavailable, we impute a neutral value and label it in the response rather than hiding the gap inside the score.