Every deep-value screen turns up cheap stocks. Some are cheap because the market overreacted. Others are cheap because the company is heading for default. The Altman Z-Score separates the two. Edward Altman published the model in 1968 at NYU, using five financial ratios to estimate the probability of bankruptcy. Working capital, retained earnings, EBIT, market cap, and revenue, each scaled by total assets or total liabilities, combine into a single score. Below 1.81 is the distress zone. Above 2.99 is safe.
This guide walks through the Z-Score step by step: running the basic screen, comparing the original and Z'' variants, combining it with the Beneish M-Score and Piotroski F-Score for a complete forensic suite, tracking trends over time, and validating the model against historical cases. For the screener overview and comparison table, see the Altman Z-Score Screener.
Step 1: Run the basic Z-Score screen
The composite score is the starting point. Claude computes all five ratios from the latest quarterly financials, joins with daily market cap for X4, applies Altman's weights, and classifies each company into one of three zones.
"From the latest quarterly financials, compute the Altman Z-Score for all US stocks above $1 billion market cap. The five ratios: X1 = (current assets - current liabilities) / total assets, X2 = retained earnings / total assets, X3 = EBIT / total assets, X4 = market cap / total liabilities, X5 = revenue / total assets. Formula: Z = 1.2*X1 + 1.4*X2 + 3.3*X3 + 0.6*X4 + 1.0*X5. Show the score, each ratio, and the zone (above 2.99 = safe, 1.81-2.99 = grey, below 1.81 = distress). Sort by Z-Score ascending."
Expect 30 to 60 companies above $1 billion in the distress zone at any given time. The majority will have negative retained earnings (X2 deeply negative) or negative EBIT (X3 negative). Both of these can reflect real distress or structural quirks. The next step tells you which.
Step 2: Compare the original Z-Score and the Z'' variant
Altman's original model was calibrated on manufacturing firms. Revenue divided by total assets (X5) varies enormously across industries. A grocery chain turns over its assets 3x per year. A software company turns them over 0.5x. Both could be perfectly healthy. In 1993, Altman published the Z''-Score, which drops X5 entirely and re-weights for non-manufacturing firms. The thresholds also shift: above 2.6 is safe (vs 2.99 in the original), below 1.1 is distress (vs 1.81).
"For AAPL, MSFT, BA, DPZ, and LUMN, compute both the Altman Z-Score (Z = 1.2*X1 + 1.4*X2 + 3.3*X3 + 0.6*X4 + 1.0*X5) and the Z''-Score (Z'' = 6.56*X1 + 3.26*X2 + 6.72*X3 + 1.05*X4) from the latest quarterly financials. X1 = working capital / total assets, X2 = retained earnings / total assets, X3 = EBIT / total assets, X4 = market cap / total liabilities. X5 = revenue / total assets (original only). Show each ratio, both composite scores, and both zones."
Domino's (DPZ) is the instructive case. Negative retained earnings from aggressive buybacks push X2 deeply negative, dragging both scores into distress. But the company generates strong free cash flow and covers its interest payments comfortably. The Z-Score flags capital structure risk, not operational risk. Comparing original vs Z'' shows whether the industry-specific X5 ratio is masking or amplifying the result. For service and tech companies, trust Z''. For manufacturers, use the original.
Step 3: Build a three-score forensic screen
The Altman Z-Score, Beneish M-Score, and Piotroski F-Score each answer a different question. Altman: is the company solvent? Beneish: are its earnings real? Piotroski: is it getting stronger? A stock flagged by all three is a convergent negative signal that no single score would produce.
"For US stocks above $2 billion market cap, compute the Altman Z-Score (Z = 1.2*X1 + 1.4*X2 + 3.3*X3 + 0.6*X4 + 1.0*X5, where X1=working capital/total assets, X2=retained earnings/total assets, X3=EBIT/total assets, X4=market cap/total liabilities, X5=revenue/total assets). Also include the Piotroski F-Score from the latest quarter. Flag any stock where Z-Score is below 1.81 AND Piotroski F-Score is below 4."
This screen combines solvency risk with financial weakness. A company in the distress zone that is also deteriorating on Piotroski's nine signals has convergent negative evidence. The overlap is typically 5 to 15 companies above $2 billion. Each one warrants a look at recent insider transactions and 8-K filings for additional context. Adding the Beneish M-Score as a third filter (prompt the formula as shown on the Beneish screener page) narrows further to companies that are distressed, weakening, and reporting suspicious earnings.
Step 4: Track Z-Score trends over time
A single Z-Score snapshot tells you the current state. The trend tells you more. A company that was safely above 2.99 two years ago but has crossed below 1.81 today is deteriorating. Whether that trajectory continues or reverses depends on factors the score does not capture, but the direction of travel matters.
"Compute the Altman Z-Score (Z = 1.2*X1 + 1.4*X2 + 3.3*X3 + 0.6*X4 + 1.0*X5, where X1=working capital/total assets, X2=retained earnings/total assets, X3=EBIT/total assets, X4=market cap/total liabilities, X5=revenue/total assets) for each of the last 8 quarters for US stocks above $5 billion market cap. Find companies where the Z-Score declined by more than 1.5 points over that period."
Quarterly Z-Scores are noisier than annual because revenue and EBIT fluctuate with seasonal patterns. A declining trend over 8 quarters smooths out the seasonality. If a company's Z-Score has fallen steadily over two years, the deterioration is structural, not a one-quarter anomaly. Cross-reference with the Beneish M-Score trend to see whether earnings quality has also shifted.
Step 5: Validate against historical cases
The question every model faces: does it work out of sample? Shibui has 20+ years of quarterly financials, so you can compute the Z-Score at any point in the past and check whether distressed companies later defaulted, declined, or recovered.
"Compute the Altman Z-Score (Z = 1.2*X1 + 1.4*X2 + 3.3*X3 + 0.6*X4 + 1.0*X5, where X1=working capital/total assets, X2=retained earnings/total assets, X3=EBIT/total assets, X4=market cap/total liabilities, X5=revenue/total assets) as of Q4 2019 for all US stocks above $5 billion market cap. What percentage of companies in the distress zone (below 1.81) declined more than 40% over the following 12 months?"
Important limitations. Survivorship bias is especially severe for bankruptcy prediction: companies that actually went bankrupt were delisted and are not in the dataset. The backtest can only show whether distressed companies declined in price, not whether they defaulted. Altman's original 1968 study reported ~95% accuracy on his training sample, but out-of-sample accuracy has ranged from ~72% to ~85% in later studies. The model works best on manufacturing firms and least well on financials and early-stage companies with negative retained earnings. It is a probability screen, not a crystal ball.
What Shibui cannot do
The Z-Score is computed from reported financial statements and market prices. It does not catch:
- Off-balance-sheet liabilities or contingent obligations not recorded in total liabilities.
- Covenant violations, credit facility drawdowns, or lender negotiations that signal distress before the ratios move.
- Cash runway for pre-revenue companies. The Z-Score was designed for operating businesses, not clinical-stage biotech or pre-revenue startups.
- Financial institutions. Banks, insurers, and broker-dealers have capital structures the model was not calibrated for. Their total liabilities are dominated by customer deposits and policyholder obligations, not debt.
Frequently asked questions
What is the difference between the original Altman Z-Score and the Z'' model?
The original Z-Score (1968) uses five ratios including revenue-to-total-assets (asset turnover), which varies dramatically across industries. A grocery chain turns over its assets 3x per year. A software company turns them over 0.5x. Both could be perfectly healthy. The Z'' model (1993) drops asset turnover and re-weights the remaining four ratios: Z'' = 6.56*X1 + 3.26*X2 + 6.72*X3 + 1.05*X4. The thresholds also shift: above 2.6 is safe (vs 2.99), below 1.1 is distress (vs 1.81). Use the original for manufacturing companies; use Z'' for everything else.
Why do some profitable companies have low Altman Z-Scores?
The most common cause is negative retained earnings. Companies like Verisign, Domino's, and Starbucks have accumulated deficits from years of share buybacks funded by debt, which drives retained earnings deeply negative. This drags X2 (retained earnings / total assets) below zero, pulling the entire Z-Score down even though the companies generate strong cash flow and have no near-term solvency risk. Airlines, REITs, and other capital-intensive businesses also score low because their business models require high debt by design. The Z-Score was not built for these capital structures.
Can AI screen for bankruptcy risk across the entire market?
Yes. On Shibui, you describe the Altman Z-Score formula in plain English and Claude computes it from quarterly financial statements and daily market cap for 10,000+ US stocks. The prompt includes the five ratios and their weights so Claude maps them to the right data fields. No spreadsheet entry, no per-company calculators. You can screen the full market, filter by sector or market cap, and combine the Z-Score with other forensic scores like the Beneish M-Score in one query.
How do you combine the Altman Z-Score with other forensic scores?
The strongest combination is Altman Z-Score plus Beneish M-Score plus Piotroski F-Score. Altman checks solvency (can the company pay its debts?), Beneish checks earnings quality (are the financial statements reliable?), and Piotroski checks momentum (is the company getting stronger or weaker?). A company in the distress zone on Altman, flagged by Beneish, and scoring poorly on Piotroski shows convergent negative signals across three independent models. On Shibui, you describe all three in one prompt and Claude runs them together.