What the Z-Score actually measures
The Z-Score measures how closely a company's financial profile resembles firms that went bankrupt. Edward Altman built it in 1968 by running multiple discriminant analysis on 33 manufacturers that filed for bankruptcy between 1946 and 1965 and 33 solvent manufacturers matched by industry and asset size. He started with 22 candidate ratios and kept the five that separated the two groups best when used together.
That last phrase carries the whole idea. No single ratio identifies a failing company reliably — a highly levered firm can be perfectly safe if it earns well, and a debt-free firm can still run out of cash. Discriminant analysis finds the weighted combination of ratios that pulls the two groups furthest apart, and the weights fall out of the statistics rather than out of judgement. That is why the coefficients look arbitrary: 3.3 on EBIT/assets is not an opinion about profitability, it is what the 1968 sample produced.
Altman reported that the model classified 95% of his sample correctly one year before failure and 72% two years before. Accuracy decays fast beyond that, which tells you what the score is for: it is a near-term screening device, not a forecast of where a business will be in five years.
Credit officers use it to rank a portfolio and decide where to spend review time. Auditors use it as one input to a going-concern assessment. Analysts use it to sanity-check a story that looks too good — a company reporting record profit with a Z-Score of 1.4 deserves a much closer look at its accruals and its cash conversion.
The five ratios and why each one is in the model
Each ratio scales a different failure mechanism by total assets, which makes the score comparable across company sizes.
X1, working capital over total assets, is the liquidity term. Working capital is current assets minus current liabilities, so a negative X1 means the next twelve months of obligations exceed the assets expected to cover them. You can compute it separately with the working capital calculator.
X2, retained earnings over total assets, is cumulative profitability — and implicitly age. A company that has earned and kept profits for twenty years scores well here; a five-year-old company with the same current profitability does not. Altman was explicit that this makes the model harsher on young firms.
X3, EBIT over total assets, is the productivity of the asset base before financing and tax choices distort it. It carries the largest weight in the original model, at 3.3, because insolvency is ultimately the point where earning power stops covering obligations.
X4, equity value over total liabilities, asks how far asset values can fall before liabilities exceed assets. The original model uses market capitalisation, which pulls forward-looking information into an otherwise backward-looking score. It is also the reason the same balance sheet can produce two different Z-Scores in a month if the share price moves.
X5, sales over total assets, is capital turnover. On its own it discriminates poorly — Altman noted it ranked last of the five — but it earns its place through interaction with the other four. It is the ratio the non-manufacturer model drops, precisely because asset intensity varies so much between a steel mill and a consultancy that including it would mostly measure industry rather than distress.
Worked example: scoring a mid-size manufacturer
Take a listed manufacturer with these figures, all in thousands: total assets 1,250,000; current assets 620,000; current liabilities 380,000; retained earnings 410,000; EBIT 145,000; sales 1,480,000; total liabilities 690,000; market capitalisation 980,000.
- Working capital. 620,000 − 380,000 = 240,000. X1 = 240,000 ÷ 1,250,000 = 0.1920.
- X2. 410,000 ÷ 1,250,000 = 0.3280.
- X3. 145,000 ÷ 1,250,000 = 0.1160.
- X4. 980,000 ÷ 690,000 = 1.4203.
- X5. 1,480,000 ÷ 1,250,000 = 1.1840.
- Apply the weights. 1.2 × 0.1920 = 0.2304. 1.4 × 0.3280 = 0.4592. 3.3 × 0.1160 = 0.3828. 0.6 × 1.4203 = 0.8522. 1.0 × 1.1840 = 1.1840.
- Add them. 0.2304 + 0.4592 + 0.3828 + 0.8522 + 1.1840 = 3.11.
3.11 is in the safe zone, but only 0.12 above the 2.99 line — and the largest single contributor is not profitability, it is the market-value term at 0.85. Halve the market capitalisation to 490,000 and X4 falls to 0.7101, its contribution to 0.4261, and the score to 2.68: the same balance sheet, the same profit, now in the grey zone.
That sensitivity is the honest reading of the original model. Roughly a quarter of this company's score comes from what the equity market thinks, which is useful information and also the part that can evaporate in a quarter. Run the private-firm variant on the same numbers using book equity of 560,000 and you get 0.717 × 0.1920 + 0.847 × 0.3280 + 3.107 × 0.1160 + 0.420 × 0.8116 + 0.998 × 1.1840 = 2.30, inside that model's grey band of 1.23 to 2.90 rather than in its safe zone. The revised coefficients are smaller across the board, so the same company reads as more marginal once you stop crediting it with a market valuation.
How to read the score: zones, distance and direction
Read three things, in this order: the zone, the distance from the nearest cut-off, and the direction over the last four periods.
The zone is a classification, not a probability. A Z of 1.2 does not mean a 40% chance of bankruptcy; it means the company's ratio profile resembles the failed group in Altman's sample. The grey zone — 1.81 to 2.99 in the original model — exists because Altman found the misclassification cost minimised inside that band. Firms there are genuinely ambiguous, and no amount of decimal places will resolve it.
Distance matters more than the exact value. A score of 3.05 and a score of 2.95 describe the same company; a score of 5.5 and a score of 3.05 do not. Where you land close to a boundary, work out which ratio is doing the work — the contribution table above tells you — and stress that ratio.
Direction beats level. A company at 2.7 and rising for three years is a different credit from a company at 2.7 and falling for three years. Credit committees that use the Z-Score seriously plot it quarterly rather than quoting it once.
Cross-check anything alarming against the ratios the Z-Score handles indirectly or not at all: interest coverage for the ability to service debt out of current earnings, debt service coverage for principal as well as interest, and the current ratio for near-term liquidity. Nothing in the Z-Score sees a bullet maturity landing next quarter.
Coefficients and cut-offs for the three published models
| Term | Z — public manufacturer (1968) | Z′ — private firm (1983) | Z″ — non-manufacturer (1995) |
|---|---|---|---|
| X1 — working capital / total assets | 1.2 | 0.717 | 6.56 |
| X2 — retained earnings / total assets | 1.4 | 0.847 | 3.26 |
| X3 — EBIT / total assets | 3.3 | 3.107 | 6.72 |
| X4 — equity / total liabilities | 0.6 (market value) | 0.420 (book value) | 1.05 (book value) |
| X5 — sales / total assets | 1.0 (0.999 as published) | 0.998 | not used |
| Safe zone | above 2.99 | above 2.90 | above 2.60 |
| Grey zone | 1.81 to 2.99 | 1.23 to 2.90 | 1.10 to 2.60 |
| Distress zone | below 1.81 | below 1.23 | below 1.10 |
For emerging-market issuers Altman adds a constant of 3.25 to the Z-double-prime score so that it can be mapped onto US bond-rating equivalents.
Do not score a bank, an insurer or a REIT with this model
Altman drew his sample from manufacturers, and financial institutions break the model's assumptions outright. A bank's balance sheet has no meaningful working-capital split, its liabilities are its raw material rather than a burden, and its asset turnover bears no relation to a manufacturer's. Insurance companies and property vehicles fail the same tests for different reasons. Use issuer-specific frameworks for those: regulatory capital ratios for banks, risk-based capital for insurers, and coverage plus loan-to-value for real estate entities.
Early-stage companies are the other structural exception. A pre-profit business with negative retained earnings and negative EBIT will score in the distress zone even when it holds three years of cash, because the model has no term for a funded runway.
Mistakes that produce a wrong Z-Score
- Putting book equity into the public model. The 0.6 coefficient was fitted to market capitalisation. Substituting book value usually lowers the score and quietly invalidates the cut-offs. Use the Z-prime model instead, which was refitted for book value.
- Using net income for X3. The ratio wants EBIT, before interest and tax, so that financing and tax structure do not contaminate a measure of asset productivity. Using net income double-counts leverage, which X4 already handles.
- Mixing periods. Take EBIT and sales from the same twelve months as the balance-sheet date. A trailing-twelve-month EBIT against a year-old balance sheet produces a number that describes no company.
- Forgetting capitalised leases. Under ASC 842 and IFRS 16, operating leases sit on the balance sheet as both an asset and a liability. That inflates total assets and total liabilities, which mechanically lowers X1, X2, X3, X4 and X5. Scores computed before and after adoption are not comparable.
- Comparing scores across models. A 2.8 from the private model and a 2.8 from the public model mean different things, because the cut-offs differ. Always state which variant you used.
- Treating the score as a default probability. It is a discriminant score. If you need a probability, use a logit model such as Ohlson's, which was estimated to produce one.
- Scoring a single year in isolation. One observation cannot tell you whether a 2.4 is a company recovering from 1.7 or collapsing from 3.6.
Where the Z-Score sits among distress and quality models
The Z-Score is the oldest widely used model of its kind, and several later ones answer different questions.
Ohlson's O-Score (1980) uses logistic regression on a much larger sample and produces an actual probability of bankruptcy rather than a classification. Zmijewski's probit model (1984) uses only three ratios and corrects for the choice-based sampling that Altman's matched design introduces. Altman's own ZETA model, developed commercially in the 1970s, added ratio stability terms and extended the useful horizon. Merton-type distance-to-default models, including the commercial KMV implementation, drop accounting ratios almost entirely and infer default risk from equity volatility and the debt maturity structure.
Two adjacent models measure something related but distinct. The Piotroski F-Score rates fundamental improvement rather than failure risk, and it is a strength screen for cheap stocks, not a distress screen. The Beneish M-Score tests for earnings manipulation, which is a different question again — though a distressed company has more motive to manipulate, so the two are worth reading together.
For ongoing monitoring, pair the Z-Score with cash-based tests. The accruals ratio shows whether reported profit is turning into cash; debt to equity and asset turnover break out two of the Z-Score's own inputs so you can see which one moved. The score tells you where to look; those ratios tell you what you are looking at.