Altman Z-score: what the model adds up and where it stops working

The Altman Z-score is a credit scoring model built from ratios of balance-sheet and income-statement figures: it takes a set of ratios, multiplies each by its own weight and adds them up into a single number that places a company in the safe zone, the grey zone or the distress zone. It is neither a valuation nor a share price forecast, but a statistical estimate of proximity to default, calibrated by Altman on a sample of industrial companies in a paper published in 1968. The model answers exactly one question: does the structure of a company's financial statements resemble that of companies which went bankrupt shortly after the reporting date?

What exactly the model adds up

In substance, the Z-score consists of ratios, each of which covers its own slice of risk, and all of them are scaled to assets or to liabilities, so that a large company and a small one can be compared.

• Net working capital to assets — the cushion of short-term liquidity. A negative value means that current liabilities are financing non-current assets.
• Retained earnings to assets — the ability to earn, accumulated over the company's history. This ratio systematically understates the score of young companies: they have simply not had the time to accumulate profit.
• Operating profit (EBIT) to assets — the return on assets before the effects of capital structure and tax. This is the core of the model: it is the operating capacity to service debt that sets the survivors apart.
• Market capitalisation to the book value of liabilities — the market buffer. It shows how far the valuation of equity has to fall before liabilities are no longer covered.
• Revenue to assets — turnover, that is, the industry-specific intensity with which assets are used.

The first ratios are read from the financial statements, while the market multiple requires an exchange quote: it can be calculated only for a company with a liquid listing, whose price can be seen in the stock data.

Why this article gives no weights or zone boundaries

The numerical values of the weights and the cut-off boundaries of the zones are deliberately not printed here: the site's editorial standard forbids typing market and calculated figures into the text by hand, and they are inserted by a directive from the database — for the ratios of the Altman model there is no such metric code in the database. They should be taken from Altman's original source or from later recalibrations, and it is essential to check which version of the model a given set of weights belongs to: mixing the weights of one version with the zone boundaries of another is a typical and costly mistake.

Versions of the model: public, private, emerging markets

The original formula applies to public manufacturing companies. For unlisted companies there is a modification in which market capitalisation is replaced by the book value of equity — and its weights are different, recalculated for that substitution. For non-manufacturing companies and emerging-market issuers there is a version from which the turnover ratio is excluded: it depends too heavily on the industry and would penalise retail and services for the structure of their assets. This version is applied to Russian issuers more often than the others, including in the analysis of issues on the corporate bond market.

Gathering the inputs from Russian financial statements

The main work in a Z-score is not the arithmetic but bringing the data into a consistent form. What to check in issuers' financial statements:

• Step 1. Choose the basis — IFRS or RAS — and do not mix them between periods. Under RAS, retained earnings and the structure of current assets read differently than under IFRS.
• Step 2. Build EBIT by hand rather than taking the "profit from sales" line: interest expense and tax have to be excluded, but operating impairments kept in.
• Step 3. Decide what to do about leases. After the move to lease capitalisation under IFRS, the assets and liabilities of retailers and transport companies grew severalfold, and the Z-score lost its comparability with earlier periods.
• Step 4. For the market multiple, take the capitalisation as of a date close to the reporting date, not the current one — otherwise the reported data and the market data describe different states of the company.

Where the model stops working

Banks, insurers and financial companies are excluded from its scope: their balance-sheet structure is fundamentally different, and the ratio of working capital to assets carries no real meaning for them. The model is blind to off-balance-sheet liabilities, to intra-group guarantees and to the debt repayment schedule — a company with an acceptable Z-score may fail to get through a refinancing peak. It does not see shareholder support or state ownership, which on the Russian market often determines the outcome more than the financial statements do. And it is static: the score refers to the reporting date, while events in the corporate calendar change the picture faster than the next set of statements comes out — it is more convenient to track them through the events calendar.

A separate matter is the risk of the procedure itself — what finance calls model risk: the Z-score gives one number, and that number tends to be read as a verdict, although it is only a measure of resemblance to a historical sample. The right use is screening and sorting candidates for in-depth analysis, not replacing that analysis.

What the Z-score is not

The Z-score gives neither the value of a company nor a required return. It contains no discount rate, and it cannot stand in for CAPM or for dividend models such as the Gordon model. In bond valuation it does not replace work with the curve either: the risk premium is calculated relative to the risk-free base, that is, OFZ yields, and the Z-score merely helps to rank issuers before that premium is estimated.