What the model measures
The Beneish M-Score is a formula published in 1999 that combines eight indices, each comparing this year's financial statements with last year's: receivables against sales, gross margin, asset quality, sales growth, depreciation, selling and administrative expense, accruals, and leverage. Each index equals 1.0 when nothing has changed.
The composite is compared against a cutoff of −1.78. Above it, the model classifies a company as warranting a closer look. This is a deterministic formula on public data, not a judgement.
Which market the cutoff was fitted on
Beneish estimated the model on 74 US companies found to have manipulated earnings between 1982 and 1992, matched against 2,332 peers. On his own holdout sample, the −1.78 cutoff misclassified roughly 7.2% of non-manipulators.
Any threshold is a trade-off between missed cases and false alarms, and where that trade-off sits depends on the distribution of the market it was fitted to. Nothing guarantees it carries to a different one.
What happens when it is applied here
We computed the M-Score for all 1,460 Vietnamese stocks the model can legitimately score, from company-published financial statements collected via a third-party data provider. Measured on 25 August 2026: 412 cross the −1.78 cutoff, or 28.2%.
That is nearly four times the 7.2% rate the same model produces on the market it was fitted to. A threshold crossed by more than a quarter of an exchange has no discriminating power left: it cannot separate the unusual from everything else, because everything else is inside it too.
Nothing in this figure says Vietnamese companies report less honestly. It says the distribution of accounting ratios here differs from the distribution the threshold was drawn on.
Why the formula is sensitive to growth
Two of the eight indices measure directly what a fast-growing company has: SGI is literally the sales-growth ratio, and DSRI rises when receivables expand faster than sales. On a small revenue base both move violently.
This is a mechanical property of the formula, not an empirical finding about growth companies. The practical consequence: a rapidly expanding and entirely ordinary business can score high, and that implies nothing about the honesty of its accounts.
What we publish instead
The figure vnvalue publishes is a percentile against other listed companies in the same year, alongside the year-over-year change. A percentile answers a question that can be answered: how unusually did this company's accounting figures move, against the norm of this market rather than another one.
The percentile bands we use are our own cuts on a continuous distribution; no paper prescribes them. The year-over-year change is a secondary cue rather than an independent detector: measured on our data, consecutive-year M-Scores correlate at only 0.064.
We do not publish the raw M-Score per company. An absolute number calibrated on a different market, set beside the name of a real business, implies a precision it does not have.