This is the plain-language version you use in interviews. If you can say all of this naturally, you can defend the project.
"Companies report their own financials, and there's no regulator checking every number. I built a scanner that uses two established forensic tools to flag companies whose numbers look statistically suspicious. Tool one is the Beneish M-Score — it scores how likely a company is cooking its books. Tool two is Benford's Law — it checks whether the first digits of a company's numbers look natural or invented. I ran it on 15 large-cap NSE companies."
A professor, Messod Beneish, studied companies the SEC caught manipulating earnings and found 8 financial ratios that reliably separate manipulators from honest firms. He combined them into one number:
M = -4.84 + 0.92(DSRI) + 0.53(GMI) + 0.40(AQI) + 0.89(SGI)
+ 0.12(DEPI) - 0.17(SGAI) + 4.68(TATA) - 0.33(LVGI)
The threshold: M above -1.78 means "possible manipulator." Below means clean. (Roughly 1 in 60 honest firms trips the test — it's a screen, not a verdict.)
The 8 ratios, in plain English:
| Ratio | What it actually checks |
|---|---|
| DSRI | Receivables growing much faster than sales? (Booking fake sales) |
| GMI | Gross margin suddenly worse than last year? (An excuse for weak profits later) |
| AQI | Growing share of "soft" assets instead of real plant/machinery? |
| SGI | Unusually fast sales growth? (Growth firms get pressured to fake) |
| DEPI | Depreciation slowing down? (Extending asset life to boost profit) |
| SGAI | SG&A costs jumping vs sales? (Layoffs to hit targets, then fake) |
| LVGI | Debt piling up? (Desperation signals) |
| TATA | Reported profit much higher than actual cash in the door? (The biggest tell) |
My actual result: none of the 12 non-bank companies crossed the -1.78 line. Titan came closest at -2.04, then ITC at -2.08. That's the honest answer — big caps rarely trip it, and I report that honestly instead of forcing a result.
In naturally occurring datasets, the first digit follows a known curve:
P(d) = log10(1 + 1/d)
So ~30% of numbers start with 1, ~17.6% with 2, down to ~4.6% with 9. Real measured numbers follow this. People inventing numbers tend to pick "nice" digits, so the distribution breaks. I compared each company's reported numbers against this curve, and graded the deviation with Nigrini's Mean Absolute Deviation (MAD): below 0.006 = close, 0.006–0.012 = acceptable, 0.012–0.018 = marginal, above 0.018 = nonconforming.
My actual result: Sun Pharma conformed most closely (MAD 0.0068); most others landed in "marginal to nonconforming."
- Banks are excluded from the M-Score. Banks have no cost of goods sold and no conventional working capital, so the model doesn't apply. I flagged them "n/a" instead of forcing a number — that's the intellectually honest choice.
- Benford flags honest firms too. Accountants round numbers, so real financial statements often fail Benford. It's a screening tool — it raises questions, it doesn't prove fraud.
- A clean score isn't proof of honesty. It means "nothing suspicious here using these two tests."
- Sample is small and large-cap only. These are blue chips with auditor scrutiny; the tool is more interesting on smaller firms (a natural next step to mention).
- "Why did you pick these two tools?" → They're the standard, published, non-subjective methods; one (Beneish) is a model, the other (Benford) is a simple statistical law. Both are explainable in two sentences.
- "Where did the data come from?" → Annual statements via Yahoo Finance's fundamentals feed, as-reported values. Handled field-name variations across companies with an alias map.
- "What's the most interesting thing you found?" → Titan is the closest large cap to the manipulation threshold; Sun Pharma is the most Benford-conforming; banks can't be scored at all because of how their statements are structured.
- "What would you do next?" → Apply it to small-caps / delisted / restated companies, and cross-check flagged firms against actual SEBI/auditor actions.