Edward Altman Model vs Ohlson Model

Edward Altman's Z-Score (1968) and James Ohlson's O-Score (1980) represent the two dominant academic approaches to predicting corporate financial failure developed in the pre-machine-learning era. Despite both models being over four decades old, they remained widely cited in credit analysis, equity research, and academic studies on Indian corporate distress, partly because their interpretability and low data requirements made them practical for markets where detailed company disclosures were sometimes limited.

The statistical methodology differentiated the two models fundamentally. Altman used multiple discriminant analysis (MDA), which classified firms into groups — failed and non-failed — by finding a linear combination of variables that maximised the separation between groups. MDA required the assumption that the group covariance matrices were equal and that the predictors were multivariate normally distributed. In practice, financial ratios rarely met these statistical assumptions precisely, which introduced some theoretical weakness into the MDA framework.

Ohlson's approach used logistic regression (logit analysis), which modelled the probability of bankruptcy directly as a value between 0 and 1 without requiring distributional assumptions about the predictors. This made the O-Score statistically more robust and produced an intuitive output: the probability that a firm would become bankrupt within one or two years. The nine variables in Ohlson's model included total assets (size-adjusted), total liabilities to total assets, working capital to total assets, current liabilities to current assets, a dummy for negative net worth, net income to total assets, cash flow from operations to total liabilities, a dummy for two consecutive years of negative net income, and the change in net income.

When applied to Indian companies, each model exhibited characteristic strengths and weaknesses. The Altman model's use of market value of equity in the X4 variable introduced a feedback loop with stock price — a company whose stock was in distress already had a lower market cap, which pushed the Z-Score further into the distress zone regardless of whether the fundamental balance sheet had meaningfully deteriorated. The Ohlson model avoided this by using book value-based inputs exclusively (in most specifications), making it less circular in prediction.

In the Indian context, several academic studies — including dissertations from IIMs, XLRI, and the National Stock Exchange — compared both models on panels of Indian corporate defaults and found mixed results. The Ohlson model generally showed higher classification accuracy in Indian samples, partly because the logit framework handled the non-normal distribution of financial ratios more gracefully. However, neither model had been formally recalibrated using a large, representative sample of Indian corporate failures, meaning that the original US-derived coefficients were applied with caution as indicative signals rather than precise probability estimates.

Practitioners in Indian equity research typically used both models in parallel as a cross-validation exercise rather than selecting one over the other. A company flagging distress on both the Altman Z-Score and the Ohlson O-Score simultaneously — with both scores in their respective distress zones — provided a stronger combined signal than a single-model alert. This dual-framework approach was applied in credit risk assessment for equity holders in highly leveraged sectors: infrastructure, power generation, real estate, and textiles, where the IBC process of 2016 onwards created real consequences for equity value when companies crossed into insolvency.