How the desk scores

Classic Z

Altman’s original industrial formula. Higher is safer. Distress is below 1.80, grey from 1.80 to 2.99, safe at 3.00 and above. Those cuts come from Altman’s manufacturing sample — they are not a default forecast.

Z = 1.2 X1 + 1.4 X2 + 3.3 X3 + 0.6 X4 + 1.0 X5

• X1 Working capital / assets. Short-term liquidity versus the size of the balance sheet.
• X2 Retained earnings / assets. Cushion from years of profits. Share buybacks can push this negative at otherwise healthy firms.
• X3 EBIT / assets. Operating return on assets. Highest weight in the original model.
• X4 Book equity / liabilities. Book equity versus the liability stack. Mega-caps often look greyer here than they would if X4 used market cap.
• X5 Sales / assets. Asset turnover — how much revenue the asset base supports.

Who gets which model

The desk routes by SIC. You cannot pick a model to rewrite the live rank — Rank-by on the universe tape only sorts the table.

Z″ = 6.56 X1 + 3.26 X2 + 6.72 X3 + 1.05 X4

• Manufacturers (SIC 20–39) stay on classic Z. Distress below 1.80, safe at 3.00.
• Other non-financials use Z″, which drops sales / assets so software and other asset-light names are not punished for low turnover. Distress below 1.10, grey to 2.60, safe above.
• Banks (SIC 60–61, and 6712 bank holding companies) use bank-z = (ROA + equity / assets) / σ(ROA), ranked only among banks. Higher is safer. There is no 1.80 distress line. Classic Z stays on the page as a teaching print only.
• Insurers, brokers, and REITs (SIC 62–67 except 6712) are unlabeled. Unscored is a policy choice, not missing data — they remain on the universe tape.

Zones

Colors on the desk follow the live model’s cuts. Banks do not use these Altman zones.

• Distress — classic Z below 1.80, or Z″ below 1.10. Elevated historical bankruptcy incidence in Altman’s original industrial sample.
• Grey — classic Z 1.80–2.99, or Z″ 1.10–2.59. Inconclusive. Treat as a watch, not a verdict.
• Safe — classic Z at 3.00 and above, or Z″ at 2.60 and above. The original “non-bankrupt” region.

O-score

A second industrial screen on the same annual print (Ohlson 1980). Higher O is worse. The desk screens on O itself, not the implied probability. Banks and REITs stay out.

O = −1.32 − 0.407 SIZE + 6.03 TLTA − 1.43 WCTA + 0.0757 CLCA − 1.72 OENEG − 2.37 NITA − 1.83 FUTL + 0.285 INTWO − 0.521 CHIN

SIZE is ln(assets in millions) — no GNP deflator. NITA and FUTL use EBIT, not net income or funds from operations. INTWO and CHIN compare this year’s ROA (X3) with the prior print when one exists.

Universe filters

The universe is one table. Filters stack: model, Z″ ceiling, live-score ceiling, O-score floor, and 1-year PD floor. Rank-by only sorts the column — it does not change who is scored or which model they sit on.

Live-score max uses the latest annual print, not a 10-Q TTM. PD only matches names that already have a priced tape. A cut such as “Z″ below 1.50 and 1y PD at least 5%” is a research screen, not a forecast.

Market distance-to-default

A market lens on top of the book score. Equity is last price times shares. The default point F is current liabilities plus half of long-term (KMV). σ is annualized equity vol. Drift r is a 4% discount — the stock’s sample mean is not used. One-year PD is N(−DD). Cached 12 hours. Never mixed into book-equity X4. Banks stay on bank-z, not Merton.

DD = [ln((E+F)/F) + (r − ½σ²) T] / (σ √T)

Prices come from Yahoo’s daily chart, with Nasdaq historical as fallback. Shares come from a stored print, else Nasdaq market cap / price, else EDGAR companyfacts on a single-name refresh. The printed DD is Bharath–Shumway naive; a KMV-style iteration runs on the same tape and is kept only if it would move the PD bucket (below 1%, 1–5%, 5%+). The universe is not priced in bulk — Price this page tapes up to 12 scored names at a time.

View as

On a non-financial company page, View as recalculates this filing under classic Z, Z″, O-score, or Merton PD. It is a display overlay: it does not rewrite the live score, the overview, or the peer set. Banks cannot be viewed as Altman Z — that is how a dummy score would sneak back in. The desk model is SIC routing, not a dropdown.

Overview

Opening a company page pulls live 10-Ks if they are stale, redraws the five-year path, prices the market tape when the model allows it, then writes an AI overview. Until that rewrite lands, a rule-based brief stays on screen so the card is never empty.

The rule brief names the model, the factor that sets the print, the move versus last year, Merton if a tape is loaded, and one caveat. Stamp is rule-v1. Enhance replaces the prose with llm-v1 in three sections — trend, industry positioning, and credit strategy — without changing the numbers. The cache key is cik, fiscal year, model, score, zone, prior print, source, and PD, so a new 10-K or tape rewrites it. The universe is not briefed on ingest.

Where the model lies

Classic Z was fit on manufacturing firms. The desk no longer applies it to banks — bank-z is a separate tape. Buybacks can still send retained earnings negative at otherwise healthy issuers (Apple is the usual example). X4 is book equity on every name, the same basis as the original filings extract. Mega-caps often print greyer here than they would if X4 used market cap. The desk does not mix the two.

Where the numbers come from

Live rank is the latest 10-K or 20-F. The original pipeline maps SEC DERA financial statement data (Assets, AssetsCurrent, Liabilities, OperatingIncomeLoss, Revenues, StockholdersEquity) into annual prints, then copies the latest year to the live row.

On first load, Refresh SEC universe pulls four DERA quarterly zips, keeps listed 10-K and 20-F filers, and writes the latest annual print — on the order of six thousand names, book-equity X4, SIC sectors. Sector leaders and the distress watch require at least $500M of assets so pre-revenue microcaps do not dominate those strips; they remain on the universe tape.

On a company page, the 10-K refresh pulls that issuer’s companyfacts file to backfill the FY 2016–present path. EDGAR does not publish market cap, which is why X4 stays book equity desk-wide. AI overviews are generated on demand and cached on the live row.