The Altman Z-Score

What the model is

Altman ran a discriminant analysis on 66 manufacturers — half of which had gone bankrupt — and kept the five ratios that best separated the two groups:

Z = 1.2·A + 1.4·B + 3.3·C + 0.6·D + 1.0·E

• A — working capital ÷ total assets (short-term liquidity)
• B — retained earnings ÷ total assets (cumulative profitability, and implicitly age)
• C — EBIT ÷ total assets (operating productivity of the asset base)
• D — market value of equity ÷ total liabilities (how much cushion the market's valuation gives creditors)
• E — sales ÷ total assets (asset turnover)

The bands

On the original sample the model called about 95% of bankruptcies correctly one year ahead, and about 72% two years ahead. Accuracy falls away quickly with horizon, which is the honest way to read it: useful, and not a prophecy.

The part people skip

The coefficients were fitted on public manufacturers, and Altman himself published different variants for private companies and for non-manufacturers because the same weights do not transfer. Leg D uses market value of equity, so Z moves when the share price moves even if nothing in the business changed — the model has a price input, which is easy to forget when reading it as a pure accounting measure.

Two consequences for how Pythia reports it:

• A Z-score is computed only when all five legs resolve. A partial Z is a different, unlabelled model, and printing it as "the Z-score" would be a quiet fabrication.
• Because leg D is priced at a moment, our figure can differ from another vendor's for the same company on the same day purely through market-cap timing. We state the basis on the surface rather than let the gap read as an error.