Piotroski, Altman and Beneish scores
Three academics each squeezed the financial statements into a single number. One looks for companies close to failing, one for cheap stocks whose finances are improving, and one for profits that might have been dressed up.
Altman's Z-score
Edward Altman, at New York University, published the Z-score in The Journal of Finance in 1968. He took 66 US manufacturers, half of which had gone bankrupt, and used a statistical method called discriminant analysis to find the mix of ratios that best told the two groups apart. On his own sample it sorted about 95% of companies correctly a year before failure, with accuracy falling off quickly further out.
In plain words: working capital is the short-term cushion, and retained earnings show how much profit the company has kept over its life, so young firms score low. EBIT over assets, the heaviest term, is how hard the assets work. The market value term asks how far the shares could fall before the liabilities swallowed them.
Notice the fourth term. Halve the share price and Z falls by a third of a point with nothing changing in the accounts. The market's mood is built into the score.
Altman later published variants: Z' for private companies, which uses book equity in place of market value, and Z'' for non-manufacturers and emerging markets, which drops the sales term. They come with their own coefficients and cut-offs, so don't mix their thresholds with the original's.
Piotroski's F-score
Joseph Piotroski, then at the University of Chicago, started from a problem every value investor knows. Cheap stocks (he used a high book-to-market ratio) beat the market on average, but many of them are cheap for a reason. He wanted a quick way to separate the ones getting better from the ones getting worse.
His answer was nine tests worth a point each. Here's Rowan again, this year against last.
The first four tests are about profitability, the next three about debt and liquidity, the last two about efficiency. Rowan scores 7: profits are rising and backed by cash, while debt crept up and the short-term cushion slipped.
In the paper, buying the high scorers among cheap stocks and selling short the low scorers returned about 23% a year from 1976 to 1996. That was a backtest, before costs, and the effect was strongest among small, little-followed companies.
Beneish's M-score
Messod Beneish, at Indiana University, asked a different question: do the accounts show the fingerprints of a company bending its profits? He compared 74 firms known to have manipulated earnings with thousands that hadn't, and published the model in the Financial Analysts Journal in 1999.
The M-score combines eight ratios, most of them comparing this year with last:
- DSRI, days' sales in receivables: are customers owing more relative to sales? Sales booked early show up here.
- GMI, gross margin: a falling margin puts pressure on management to find profit elsewhere.
- AQI, asset quality: a growing share of long-term assets other than plant and equipment can mean costs are being parked on the balance sheet.
- SGI, sales growth: innocent in itself, though it brings pressure to keep the streak going.
- DEPI, depreciation: a slowing rate of depreciation lifts profit on paper.
- SGAI, selling and admin costs relative to sales.
- LVGI, leverage: rising debt brings covenants worth dodging.
- TATA, total accruals: profit that isn't matched by cash, scaled by total assets.
Two of them do a lot of the work in practice. If Rowan's sales rose 10% (from $1,000 million to $1,100 million) while receivables rose 45% (from $100 million to $145 million), DSRI would be (145 ÷ 1,100) ÷ (100 ÷ 1,000) = 1.32. And if it reported $80 million of net income on $20 million of operating cash flow, accruals would be 6% of its $1,000 million of assets. Both push the score up.
Beneish's 1999 paper puts the cut-off for the eight-variable model at −1.78: above it, the risk of manipulation is higher. Many analysts add a stricter line at −2.22 and treat the space between the two as a grey zone. A score in the grey zone or above it is a reason to read the notes. Proving fraud takes far more than a ratio.
Old samples, blunt tools
The Z-score was fitted to manufacturers from 1946 to 1965. A software company with few physical assets, or one that has spent its retained earnings on buybacks, can score oddly in either direction. For banks and insurers it means nothing at all.
The F-score is binary. A gross margin up by 0.1 of a point earns the same point as one up by five. It was built for cheap stocks, and its record on expensive ones is weaker.
The M-score trips on honest companies that grow fast or buy other businesses, because acquisitions swell receivables and intangible assets in a single year. Expect false alarms.
Where the three scores appear
The Summary tab of each company page has a section called The Examination with all three. The F-score is shown out of 9, labelled Strong at 7 or more and Average from 4 to 6. The Altman Z sits on a bar with the 1.81 and 2.99 lines marked. The Beneish M-score is worked out from the last two annual reports and shown in three zones: clean below −2.22, grey between −2.22 and −1.78, and a red flag above −1.78.
Only one of the three feeds the Fortress Score. The Altman Z is a small part of the Hidden & quality wall, 3 of its 10 points, with full credit at 3 or above. It's left out for REITs and utilities, where a formula fitted to manufacturers makes no sense. Piotroski and Beneish appear on the Summary tab and don't touch the Fortress Score.
Takeaways
Sources for this lesson
- Edward I. Altman, "Financial Ratios, Discriminant Analysis and the Prediction of Corporate Bankruptcy", The Journal of Finance 23(4), 1968, pp. 589–609
- Joseph D. Piotroski, "Value Investing: The Use of Historical Financial Statement Information to Separate Winners from Losers", Journal of Accounting Research 38 (Supplement), 2000, pp. 1–41
- Messod D. Beneish, "The Detection of Earnings Manipulation", Financial Analysts Journal 55(5), 1999, pp. 24–36
- Edward I. Altman, Edith Hotchkiss and Wei Wang, Corporate Financial Distress, Restructuring, and Bankruptcy (4th edition, 2019), on the Z' and Z'' variants