Ohlson O-Score: How the Nine-Factor Bankruptcy Model Works

The Ohlson O-score asks a blunt question of a balance sheet: how likely is this company to go bust? Not “is the stock cheap” or “will earnings beat estimates”, just whether the business is close to the edge. James Ohlson built it in 1980, about a decade after Edward Altman’s Z-score, and he took a different route to the answer. Instead of a score you compare with fixed zones, his model spits out a probability.

That difference matters more than it sounds. A probability is easier to reason about, easier to compare across companies, and harder to misread. We’ll go through the nine inputs, what each one is trying to capture, a full worked example in rupees, and the places where the model quietly breaks when you bring it to Indian listed companies.

Where the Ohlson O-score comes from

The model was published as “Financial Ratios and the Probabilistic Prediction of Bankruptcy” in the Journal of Accounting Research (Vol. 18, No. 1, Spring 1980). Ohlson used data from 1970 to 1976 and fitted what statisticians call a conditional logit model, estimated by maximum likelihood.

The sample was large for its time. Ohlson’s profile table lists 105 firms that went bankrupt and 2,058 that did not. Altman’s original work used a much smaller matched sample, which is one reason the O-score is often described as a sturdier follow-up rather than a rival.

Ohlson was also refreshingly honest about his choices. He wrote that no attempt was made to develop “new or exotic” ratios, and that simplicity drove the selection. The ratios are the ordinary ones any analyst already pulls from an annual report. That’s good news for us, because nothing in the formula needs data you can’t find in a company’s filings.

The nine inputs, in plain words

Ohlson estimated three versions of the model. Model 1, the one people mean when they say “O-score”, predicts bankruptcy within one year. Its nine variables and their coefficients, from Table 4 of the paper, are below.

Add up the constant and each coefficient multiplied by its variable and you get a number, call it T. That number on its own is not a probability. You convert it with the logistic function, which Ohlson sets out in the paper:

Probability of bankruptcy = e^T / (1 + e^T)

So a T of zero means a 50% probability. Negative T values pull the probability towards zero, positive ones push it towards one.

What the signs are telling you

Read the signs first and the model starts to make intuitive sense. Bigger companies (SIZE) are less likely to fail. More leverage (TLTA) raises the risk sharply; look at that 6.03 coefficient, the biggest in the table. More working capital relative to assets lowers risk. Profits (NITA) and operating cash relative to debt (FUTL) both lower it. Two straight years of losses (INTWO) raises it, and a falling profit trend (CHIN negative) raises it too, because a negative CHIN multiplied by -0.521 becomes a positive addition.

The odd one out is OENEG. When liabilities exceed assets, the company has negative book equity, and you’d expect that to push the score up. Instead its coefficient is negative. Ohlson explains it as a “discontinuity correction” for TLTA: once leverage goes above 100%, TLTA is already adding a huge amount to T, and OENEG trims the overshoot. In his words, the situation is “very bad indeed (due to TLTA), but not that bad.”

Leverage does most of the heavy lifting in the O-score. A company that funds 90% of its assets with liabilities starts the race several points behind, and profits have to work hard to claw that back.

A worked example: two hypothetical companies

Let’s run the numbers. The two companies below are made up, and the figures are illustrative only. Everything is in ₹ crore.

Company X is a steady mid-sized manufacturer having a weaker year:

• Total assets: 1,000; total liabilities: 720
• Current assets: 400; current liabilities: 350, so working capital is 50
• Net income this year: 15; last year: 40
• Funds from operations: 60

Company Y is in real trouble:

• Total assets: 1,000; total liabilities: 940
• Current assets: 300; current liabilities: 420, so working capital is -120
• Net loss this year: 60; net loss last year: 20
• Funds from operations: -10

The SIZE problem, handled honestly

SIZE is the awkward input. Ohlson defined it as the log of total assets divided by a US GNP price-level index with 1968 as the base of 100, using dollar figures. There’s no clean way to drop rupee balance sheets into that. For the example, we’ll simply assume a SIZE of 13.0 for both companies. That sits between the averages in Ohlson’s own data, where failed firms one year before bankruptcy averaged 12.134 and healthy firms averaged 13.26. Later we’ll show how much the answer moves if you change it.

Company X, step by step

Convert it: e^-2.2259 / (1 + e^-2.2259) works out to about 0.097, or roughly a 9.7% probability of bankruptcy within a year. That’s comfortably below the 0.5 line, but notice what’s driving it. Leverage alone adds 4.34 to T. The profit slump adds another 0.24 through CHIN. Company X isn’t in danger on these numbers, yet it isn’t spotless either.

Company Y, step by step

T is just above zero, so the probability is about 0.51, or 51%. On the classic 0.5 rule, Company Y gets flagged. Every single term except the constant and SIZE is now working against it: negative working capital, losses two years running, a deeper loss this year, and operating cash that doesn’t even cover itself.

How much does SIZE move the answer?

Quite a lot. Keep everything else about Company X the same and change only SIZE:

• SIZE of 12.0: T becomes -1.8189 and the probability rises to about 14.0%.
• SIZE of 13.0: about 9.7%, as above.
• SIZE of 14.0: T becomes -2.6329 and the probability falls to about 6.7%.

Each one-point change in SIZE shifts T by 0.407. That’s why we’d never treat an O-score computed on Indian numbers as a precise probability. It’s a ranking tool at best, and SIZE is the main reason.

Reading the result without fooling yourself

The common shorthand is that a probability above 0.5 signals likely failure. Ohlson did report his accuracy using that cutoff: Model 1 classified 96.12% of the sample correctly. But he immediately pointed out the catch. Only 105 of the 2,163 firms in his sample failed, so simply calling every firm a survivor would have been right 2,058 times out of 2,163, or about 95% of the time. He also wrote that there is “no apparent reason why .5 is an appropriate cutoff point”, since it assumes both kinds of mistake cost the same.

They don’t. For a shareholder, missing a company that goes bust is far more painful than wrongly worrying about a healthy one. So in practice, we’d treat any O-score probability that is high relative to its peers, or rising year after year, as a reason to dig, even if it is well below 0.5.

A 96% hit rate sounds impressive until you realise that guessing “nobody fails” would have scored about 95% on the same sample. The O-score earns its keep at the margins, not in the headline number.

Ohlson’s own conclusion is a useful checklist. He found four factors from financial statements that were statistically significant: the size of the company, its leverage, some measure of performance (net income to assets, funds from operations to liabilities, or both), and some measure of current liquidity. If you remember nothing else from the formula, remember those four.

Using the O-score on Indian companies

The model was fitted on US industrial companies in the 1970s. Bringing it to an Indian annual report needs some care.

• Skip banks, NBFCs and insurers. Their liabilities are deposits and borrowings by design, so TLTA will look alarming for a perfectly healthy lender. The model wasn’t built for them.
• Be consistent with “funds from operations”. Ohlson used funds provided by operations. Today, the closest practical stand-in is cash flow from operating activities in the cash flow statement. Use the same definition across every company you compare.
• Use consolidated numbers for groups. Standalone statements can hide debt sitting in subsidiaries.
• Watch one-off items. An exceptional gain can flatter NITA and CHIN for a year. Look at what’s recurring.
• Compare within a sector. A capital-heavy power or infrastructure company will carry more leverage than an IT services firm. The O-score doesn’t adjust for that, so peer comparison matters more than the absolute figure.

Our guide on how to analyse quarterly results of a company covers where to find most of these line items, and the interest coverage ratio is a natural partner check when TLTA looks high.

O-score vs Z-score, F-score and M-score

The O-score isn’t meant to stand alone. Each of the well-known scoring models answers a slightly different question:

• Altman Z-score: a weighted score for financial distress, read against fixed zones. The closest cousin to the O-score.
• Piotroski F-score: a nine-point checklist of whether a company’s fundamentals are improving or worsening.
• Beneish M-score: a screen for possible earnings manipulation.

When a company looks shaky on both the Z-score and the O-score, that’s a stronger signal than either one alone. When the M-score also flags it, we’d want to be very sure about the quality of the numbers going into the first two.

Limits you should keep in mind

The O-score is old, and it shows. The coefficients were estimated on a single country’s companies over seven years, with an accounting regime very different from Ind AS. The SIZE variable depends on a US price index. And the model only uses accounting data. Ohlson himself suggested that market-based information, like equity prices and their changes, “might prove to be most useful” and said he intended to test it.

There’s also a timing gap. Annual reports arrive months after the year ends. A company can deteriorate a lot between the balance sheet date and the day you run the numbers, so the most recent quarterly filings and any rating actions deserve a look too.

A quick checklist before you rely on it

1. Pull total assets, total liabilities, current assets, current liabilities and net income for the last two years from the consolidated statements.
2. Take cash flow from operations as your funds-from-operations input, and note that you did.
3. Pick one SIZE convention and apply it to every company you compare.
4. Compute T, convert it to a probability, and compare it with sector peers and with the company’s own history.
5. Cross-check with the Z-score and interest coverage before drawing any conclusion.

Used that way, the Ohlson O-score is a handy early-warning light. It won’t tell you when a company will fail, and it can’t read the footnotes for you. But it does force you to look hard at leverage, liquidity and the profit trend together, and most balance-sheet blow-ups show up in at least one of those well before the headlines do.

Disclaimer: This article is for information and education only and is not investment advice. The companies and figures in the worked examples are hypothetical and illustrative. Scoring models can misclassify companies, so please do your own research or consult a SEBI-registered adviser before making investment decisions.