Knowledge Base › Comparisons › Ohlson O-Score vs Zmijewski Score
Comparison

Ohlson O-Score vs Zmijewski Score

Both estimate distress probability, but Ohlson weighs nine accounting factors through a logit model while Zmijewski uses three ratios through a probit.

The difference

Both models take a company's financial statements and return a model-estimated probability that the company is on the failing side of a historical sample, but they get there differently. The Ohlson O-Score, published by James Ohlson in 1980, is a nine-factor logistic regression spanning size, leverage, liquidity, profitability and recent performance, including dummy variables for negative equity and consecutive net losses; the fitted score O is converted with the logistic function, PD = 1 / (1 + e^(-O)). The Zmijewski Score, published by Mark Zmijewski in 1984, is a probit model built on just three ratios — return on assets, total liabilities to total assets, and the current ratio — with the raw score X converted through the standard normal CDF, PD = Phi(X). The practical consequence is that the two produce numbers on different scales with different cutoffs: Ohlson's classic cutoff is O greater than 0.38, while Zmijewski flags a probability above 0.5. Neither is the more accurate model; one asks how a broad panel of accounting signals stacks up together, the other asks whether a company earns enough on its assets to carry the liabilities it has piled on them.

Side by side

Ohlson O-Score compared with Zmijewski Score
AspectOhlson O-ScoreZmijewski Score
InputsNine factors: size, leverage, liquidity, profitability, two dummies, a change termThree ratios: return on assets, liabilities/assets, current ratio
Statistical formLogistic regression; O is a logit score, PD = 1 / (1 + e^(-O))Probit; raw X converts via the standard normal CDF, PD = Phi(X)
Classic cutoffO above 0.38 indicates elevated failure riskPD above 0.5 sits on the distressed side of the model's cutoff
OriginJames Ohlson, 1980Mark Zmijewski, 1984
Calibration sampleUS-listed industrial firms of the 1970s; coefficients and cutoff fitted to that era's accounting practices and failure base rates1980s NYSE and AMEX industrials, bankrupt group deliberately oversampled
Known weak spotDecades-old coefficients; misleads for banks, insurers, asset-light firmsLiquidity coefficient was insignificant and is transcribed inconsistently

Which one to use

Reach for Ohlson O-Score when…

Reach for the O-Score when a single ratio is hiding the picture. Its nine inputs weigh size, leverage, liquidity, profitability and recent performance together, and its dummy variables for negative equity and consecutive net losses pick up conditions that a small set of ratios can miss entirely. As the model's own page puts it, two companies can look similar on a single ratio and differ sharply once those factors are weighed together.

Reach for Zmijewski Score when…

Reach for the Zmijewski Score when you want a lens you can fully audit by hand. Three inputs means you can see exactly which ratio moved the score, and the question it answers is narrow and legible: does the company earn enough on its assets to carry the liabilities on them? That parsimony also helps when inputs are sparse and a nine-factor model cannot be populated cleanly.

The common mistake

The concrete mistake is comparing the two numbers as if they sat on the same scale. Ohlson's 0.38 cutoff applies to O, the raw logit score, which still has to be pushed through the logistic function before it means anything as a probability; Zmijewski's 0.5 cutoff is already a probability from the normal CDF. Read an O of 0.5 as "a 50% chance of bankruptcy," or hold it up against Zmijewski's 0.5, and you have mislabelled the number and measured it against the wrong threshold at the same time.

How Quintarthai uses them

Both models are covered in this Knowledge Base, and bankruptcy-model outputs are shown next to the underlying ratios that drive them in the financial-risk section of a deep analysis on /app/. They are presented as educational context on what historical accounting data looks like — not as a recommendation, a forecast, or a rating of any company.

FAQ

If the two models disagree on the same company, which one is right?
Neither, in the sense you probably mean. They read different inputs through different statistical forms, so disagreement is information rather than error: it usually means something the O-Score sees — negative equity, consecutive losses, size, or working capital — is not visible in Zmijewski's three ratios, or that the leverage term is dominating one model and not the other. Both were also fitted to decades-old samples — Ohlson to US-listed industrial firms of the 1970s, Zmijewski to a 1980s NYSE and AMEX industrial sample — so both carry calibration limits described on their own pages.
Are these scores forecasts that a company will go bankrupt?
No. Both are model probabilities describing what a company's historical accounting ratios look like relative to a decades-old estimation sample, not predictions about the future. Ohlson's coefficients and its 0.38 cutoff were fitted to the accounting practices and failure base rates of that era, and Zmijewski's absolute probabilities travel poorly to modern asset-light, financial, or non-US companies. Treat either output as one input to your own reading, not a verdict.
Read the full definitions
Explore the data behind the definitions
Quintarthai covers 7,500+ US and Canadian listings — screeners, company pages and live filings.
Open the app →