Edward Altman is a professor emeritus at New York University’s Stern School of Business. We spoke about the Z-Score model, which he created, and how to use it to assess the financial riskiness of a company. The original Z-Score is calculated and automatically updated in our Stock Investor Pro fundamental stock screening and research database program (click on the “Ratios” tab.)
—Charles Rotblut, CFA

Charles Rotblut (CR): What is the Z-Score designed to measure?

Edward Altman (EA): The Z-Score is used extensively in the financial community as an indicator of the probability of a company going bankrupt or having significant financial distress. The original model was built almost 50 years ago based on balance sheet, income statement and stock price data for publicly traded manufacturing companies. The newer models are for companies that are both public and private and cover the entire industrial sector area.

CR: Are there industries where it doesn’t work—say, for instance, financial companies, whose balance sheets are different than most companies?

EA: Yes. The model was not built for financial companies, including banks, insurance companies, brokerage firms, investment companies, mutual funds and the like. The original Z-Score model was built in 1968 for manufacturers. More recently, models like Z prime (Z′) and Z double prime (Z′′) were created for not only private companies, but companies in other industrial sectors like retail, wholesale, service, energy and even public utilities.

CR: When you originally created the model, it was designed to predict bankruptcy within a two-year period, correct?

EA: That’s correct. The two-year period was indicated because of the accuracy level of the model. The model could be used for longer than two years, but the caveat is that the accuracy goes down. It went down quite a bit as a long-range predictor when the actual event was longer than two years away. More recently, the later models that I built have longer, more accurate prediction accuracy, but still, I would say that the model is probably not all that useful for more than two years prior to the event.

CR: Let’s just talk about those five variables that make up the original indicator (Table 1). The first one is working capital (current assets minus current liabilities) divided by total assets. You have found that to be better than the current ratio or the quick ratio in terms of assessing risk?

EA: Yes. It’s still, of the five variables, probably the least helpful in predicting distress. I’d like to emphasize that this is a multivariate model, so each indicator, by itself, may or may not be very accurate as a single indicator. When they all act together, the accuracy levels go up quite a bit and the meaningfulness of the data increases.

The first variable, working capital to total assets, is a measure of liquidity. It proved to be more stable and predictive than the current ratio and the quick ratio, which is more well-known. Indeed, the current ratio and, to some extent, any liquidity ratio, can be a misleading indicator. This is because as a firm approaches bankruptcy, their current assets are bloated due to the fact that they cannot collect on their receivables or they have too much inventory that they can’t sell. So the current ratio sometimes actually goes up rather than down as bankruptcy approaches. By itself, the ratio of working capital to assets can be a misleading indicator. But we found it to be helpful in just about all our models, though not as helpful as some of the other indicators.

CR: Have you seen a scenario with these companies where cash is being built up simply because they’re cutting back on spending faster than the business is slowing?

EA: Yes, as a matter of fact, one of the ironic strategies that bankruptcy lawyers sometimes recommend to their clients is to go bankrupt when you have a lot of cash. Even though you don’t have enough cash to pay all your bills, it’s better to have more cash as you go into bankruptcy because you’ll need it during the restructuring period. To be clear, we’re talking about bankruptcy reorganization rather than liquidation. The timing of the bankruptcy filing is sometimes attuned to the cash cycle. This is a strategy that some very savvy bankruptcy lawyers tend to use.

Our model generally looks at data six months, one year, two years before bankruptcy, and these companies probably have not built up a lot of cash yet. It’s right before bankruptcy—maybe a couple months—that the bankruptcy lawyers tend to focus on cash levels. But, yes, in general, the working capital ratio may be a misleading indicator by itself because of this cash buildup and too much inventories and too much receivables, relative to their assets.

CR: The next indicator is retained earnings divided by total assets. This one favors older companies, right?

EA: That’s correct.

Retained earnings is the balance sheet figure, not the accounting statement figure. It’s the sum of all past earnings of a company minus their dividends and adjusted for any restructurings. So it’s a cumulative earnings figure rather than the last year’s retained earnings—just for clarification purposes.

Generally, the older the firm, the more time it’s had to build up retained earnings than, say, the younger firm that may have had losses or very low profits in the earlier years. As a result, the retained earnings ratio is a very good indicator. The young firms have a much higher probability of default and bankruptcy than older firms, and that’s what statistics have shown for many, many years.

In addition, the retained earnings figure—which I find to be an extraordinarily helpful individual indicator—is a kind of back-door way of looking at leverage. The firm that has built up its retained earnings as a contributor to help finance its asset growth has probably used less of what I call “OPM,” or other people’s money—debt and outside equity financing. As a result, they have more capacity going forward, should they need to raise debt, than a firm that has very low retained earnings. So a combination of cumulative profitability, age of the company and leverage usage are all implicit in the retained-earnings-to-total-assets ratio. It is a very good indicator in just about all my models.

CR: Earnings before interest and taxes, EBIT, divided by total assets, is essentially a return on assets number, isn’t it?

EA: It’s a return on asset number, absolutely. Some people wondered why I didn’t use cash flow as opposed to EBIT. The answer is that back when I built that original model, we didn’t have very good data on depreciation and amortization on many companies. So we went to EBIT. It turns out that when we’ve done studies of both EBIT and what’s called EBITDA, which includes depreciation and amortization, the accuracy levels and the discriminating power were just about the same. You don’t gain that much by using cash flow in a multivariate context.

CR: What about going to the cash flow statement? Would any measure from it lead to an improvement? Does it make a significant difference?

EA: Most people think it would improve the model. I do, too, when thinking about it. But it turned out that in our empirical studies, it didn’t make much difference. So the answer is, either one is good. I like the fact that the EBIT number is not impacted by interest and taxes. It’s much more of a pure earning power number than the earnings after interest and taxes, for example. I’m not saying that taxes and interest are not important, but we take care of that elsewhere in the model.

CR: Interesting. Moving on to market value of equity divided by total liabilities: In the original model, you used common stock market capitalization plus preferred stock. For private companies, you have said you use book value. For non-manufacturers, is the Z double prime using market cap or is it using the recorded book value?

EA: The Z double prime model is using book value rather than market value. The reason is that it was built for, originally, emerging market companies. We didn’t trust the efficiency of the stock market in those countries. The original Z model used the market value of the common stock plus preferred stock relative to total liabilities. It was the first time any researcher or analyst had used market equity as opposed to book equity as an indicator of financial distress. Market equity captures future earning power, expectations on the part of a market with respect to solvency and health.

That’s what the model’s all about, looking to the future rather than only picking up what’s happened in the past. So I’ve always liked market equity as an indicator, rather than book equity, if the firm is publicly traded and it’s in a liquid and efficient market. So, market equity was in the original model and proved to be an extraordinarily important variable. I would not say you should use it to the exclusion of the other factors in the model, however.

There are some models out there for predicting distress that are called structural models. Structural models are mainly based on equity valuation and volatility of equity. If I had to do it over again, I would include volatility of equity in the model. I’ve done it in more recent models. The original one did not include it; I did not have the insight at that time to use volatility of equity as well as the level of equity as an indicator of distress. Obviously, the firm that is more volatile in its equity, its sales, et cetera, is more risky than a firm that is less volatile for a given level of equity value. When you get to the Z double prime model, there and in Z prime, I used book value so it would be applicable for analyzing private companies, as well.

CR: You’ve previously written that the ratio of sales to total assets, a measure of asset turnover, is actually the second-highest contributor to the model. Why is that?

EA: What that shows is the importance of how each variable interacts with other variables. This picks up the correlation between variables. That variable by itself as an indicator of discrimination between bankrupt and non-bankrupt companies is not statistically significant. So you would normally throw that out, except that because of its interaction it entered the model and was very helpful.

The problem with this variable, however, is that it’s very industry-sensitive. As a result, I did not feel it was helpful for indicating, for example, the distress of retail companies, which have a very high sales-to-total-assets ratio compared to manufacturers. When I built the Z double prime model, which I wanted to be applicable to many other sectors, we eliminated that variable even though it was quite helpful in the original Z model for manufacturers. So that is why that variable entered the model and why, ironically, I’ve not used it in subsequent models.

CR: The newer Z double prime model is just the first four indicators weighted differently?

EA: That’s correct. I’d like to get to what the score means—what it meant then, what it means now and how we can improve upon the meaningfulness of the results of the model.

CR: Absolutely.

EA: Thank you for giving me the opportunity to do this, because it’s not that well-known even by people who use this model on a regular basis. There are many software programs that have the Z-Score model on it, including your own stock screening software program (Stock Investor Pro). From what you told me, you only use the original Z-Score model at this point, and you don’t include Z double prime. Hopefully, I’ll be able to get you to change, or at least to add the second model.

The Z-Score was not complete in terms of the information content of the scores. What I mean by that is, the original model gave guidance to the reader by stating that if the score was below 1.8, the firm was in a distress zone. If the score was between 1.8 and 2.99 or 3.0, it was in the gray zone. Some errors were observed in the model. If it was above 3.0, it was in the so-called “safe zone” and not likely to go bankrupt at least in the next two years.

It turns out that those were just guidelines based on the original sample of manufacturing companies that I used to build the model. The model was 100% accurate below 1.8 and 100% percent accurate above 3.0. However, it’s pretty clear that there’s not much difference between a company that has a score of 1.79 and a score of 1.82. They’re basically the same level of risk, but one is in one zone and the other is in another zone. Clearly the firm with the higher Z-Score has less financial vulnerability, according to the model, than a firm with a low score, but the zones were specific to the original sample 50 years ago!

There was no real way, when I first published the model, of establishing the probability of default. Over time, the levels of scores have changed. The average American company today is more risky in terms of the Z-score model than it was back in 1968 when we published the model for the first time.

To give you an indication, the average score today of a B-rated company according to Standard & Poor’s, for example, has a score of 1.66 (Table 2). This is an average score lying somewhat in the distressed zone. Some people interpret that as meaning the firm is likely to go bankrupt, but that would be incorrect. A B-rated company has, according to our analysis, about a 28% chance of going bankrupt within five years. Therefore, it has a 72% probability of not going bankrupt. My suggestion is to no longer use those zone ranges and to instead substitute other metrics.

Table 2. Median Z-Score by S&P Bond Rating for U.S. Manufacturing Firms (1992–2013)

One very useful metric is a bond rating equivalent. In other words, according to the model, what would be the firm’s bond rating, regardless of whether or not it has outstanding bonds or loans that are rated. You can give a bond rating equivalent to any company: private, public, large, small, etc. The bond rating equivalent changes over time as the level of risk changes over time. The beauty of a bond rating equivalent is that we can then translate that into a much more granular probability of default than just what zone the firm appears in, which is intuitively nice and simple, but it’s not telling the whole story.

CR: Interesting. Could you give a more thorough explanation of the Z double prime model?

EA: The Z double prime model was built in 1995 when I did some consulting for one of the large investment banks that was interested in getting a credit score for companies domiciled in emerging market countries like Argentina, Brazil and Mexico. These companies had issued bonds in the international market denominated in non-local currency, primarily dollars. They were, then, vulnerable to credit risk as well as devaluation risk and other types of risk such as industry risk. Most of these companies were not even rated by the large rating agencies, at least not in 1994. Because there weren’t any defaults yet in that market, I looked at a model that I could tweak from my original work, rerun it without the fifth variable (sales to assets) and without market value of equity (because it was dealing with emerging market firms), re-estimate the coefficients and then look at the bond rating equivalent [Table 1 shows the formula for the Z double prime model.]

Keep in mind also, Charles, that there’s a coefficient in this model, a constant term, 3.25, as well as the four variables and their weightings. I put that 3.25 in there arbitrarily, because I wanted to standardize the model so that a score below zero would be in default, a D-level rating. Anything above zero was likely to be CCC or B or whatever, but not D. That model has been used very successfully in emerging markets, and I also use it now in the United States and in other developed countries.

It is more robust and more useful for non-manufacturers, as I said, such as service firms, retailers, airlines, energy companies, et cetera. I recommend that your readers calculate both the Z and Z double prime, but particularly when it gets to non-manufacturers, I recommend emphasizing the Z double prime and its bond rating equivalent.

Finally, there are tables available in my book [“Corporate Financial Distress & Bankruptcy”(3rd ed., John Wiley & Sons, 2005)] that use the bond rating equivalent we can get out of the score to estimate a probability of default from one year to 10 years into the future. The estimation is based on what the original rating was of a company’s bonds and how frequently they defaulted one year, two years or three years afterward. We have aggregated all that to come up with this probability of default table, based on the original ratings. Similar types of cumulative default numbers can be accessed from the rating agencies that publish them every year. [Bond rating equivalents are shown in Table 3.]

Table 3. U.S. Bond Rating Equivalents Based on Z Double Prime Scores

So let me summarize by saying that the best way to use the model today, in my opinion, is to use the Z-Score, which is very easy to calculate yourself (or get from a software package) by multiplying each variable by its weighting and sum it all up to get the score. Then look at the firm’s bond rating equivalent. This allows you to gain more insight into the likely health of the company rather than simply putting it in one zone or another, which most people who use the model still do. I don’t mean it’s not interesting and helpful to observe the score, but it’s not as granular and not as relevant today as it was back in 1968 when we built the first model.

CR: If the Z-Score is low and the actual bond rating is in junk range, that would be a sign that a company is very risky. If the Z-score is low, but the actual bond rating is, say, a BB+, does that mean that you should be fine?

EA: That’s a good question. One of the applications of this model is to look for anomalies where the model says, for example, the company looks like a high-quality company—let’s say it has a BBB or BB or better bond rating equivalent—but the rating agencies are saying it looks like a B–…in other words, what I call “quality junk.” Or the reverse situation, where the model says that a company looks very bad, like a CCC-rated company, but the rating agencies are saying it looks like a BB or above (e.g., junk quality).

One way of using the model is to compare the bond rating equivalent in the model with the actual bond rating. In most cases they’re going to be fairly similar. When they differ, analysts should then sharpen their pencils and get more in-depth analysis of the company.

The model is only a model. It makes errors. It is not 100% accurate. It is an indicator for the analysts to do their homework a little bit more to determine the risk of holding these companies in their portfolios.

There will be times when it’s quite different than the rating agencies, and you will have to decide which makes more sense.

The conversation continues

In this online exclusive extension of our conversation, Altman discusses the helpfulness of looking at historical trends in the Z-Score and offers a suggestion for using the indicator to help make better business decisions.

CR: One other question: Is it worth looking back at how the Z-Score was, say, one or two or three fiscal years prior?

EA: Absolutely. The trend analysis of the model is also important. If you find that the company has consistently deteriorated over, say, five, or even 10 years—certainly three years—that’s more worrisome than if the firm drops a lot in one year into, say, a distressed realm with a relatively low score. That doesn’t mean you ignore the latest number if it’s low and it was high before. That’s also a problem, but not as much as when the number has consistently deteriorated.

Let me give you an example of a company that deteriorated for 10 years in a row but was still rated AAA by the rating agencies. The model said it looked more like a B-rated company after this 10-year deterioration. I’m referring to IBM Inc. (IBM) from the early 1980s to the early 1990s.

It was rated AAA by both rating agencies for the entire 10-year period, but our model said it was getting riskier for over the same period. In early 1992, and again in mid-1992, the rating agencies finally woke up and downgraded IBM from AAA to AA- and then to A. If the board hadn’t finally thrown out the old management and changed the business model, IBM could have gone bankrupt.

The model was saying IBM was getting increasingly risky. The rating agencies were not capturing that trend until finally it became blatantly clear to everyone that this was no longer a AAA company. So that’s another application of the model—predicting downgrades, and how trend analysis could help.

CR: Is there anything else that’s important that I haven’t asked you about?

EA: Well, there are a number of applications of this model that your readers might be interested in, in addition to the obvious one of answering the question, “does it fit into my portfolio?” While I realize that these models and their application may be more limited for the individual investor who, of course, makes up your main audience, than, say, a hedge fund or a private equity fund analyst would be interested in, another fantastic application of the Z-Score models is for firms themselves to use the model. Maybe some of your readers own their own firms or are senior officers of their own companies and are interested in one of the really fascinating applications that I’ve found over the years. That is, management itself can use the Z model as an indicator and a guide to a financial turnaround.

I actually worked on some case studies with companies, one in particular that I remember very well, where I knew the CEO (after the fact). He was assigned as an interim and then a permanent CEO to take over a company on the verge of bankruptcy. What he did was to say, “Okay, I going to make business decisions for improving the situation, but I’m only going to make those decisions where the Z-Score is improved.” In other words, he simulated the impact of a business decision, like selling assets, reducing debt, paying back debt, consolidating assets, but he only made carried them out if the score would improve based on his simulations or his expectations. Finally, the firm moved clearly out of the distressed area into the safer zone, above 3, 4, 5. He credited the model for being an extremely helpful, unemotional and objective indicator of just how serious the situation was and how to improve it. So I’d like to share that as a potential helpful indicator for your readers and anyone else, above and beyond its investment strategy issues, which I know is the main purpose of your educational effort.

Another thing I’ve learned over the years is how to answer the question, “how do you define distress with respect to a company?” I segregated bankrupt versus non-bankrupt companies to build the original Z model. I still recommend that, but in addition, I’ve added a number of other conditions that would qualify a company as distressed to include in the sample of firms when you’re building models, but also in terms of testing the accuracy of the model. So I would include such things as missing interest payments, but not actually defaulting, not actually going bankrupt. A lot of firms do that and then they either make the interest payment or they do—and this is another indicator—something called a distressed exchange. A distressed exchange is when the company makes a tender offer to their creditors, like bondholders, and asks them to accept some other security in lieu of the original security, change the contract, if you will, change the indenture with the bondholders. When that package of new securities is worth less than the original issuance value, that’s a distressed exchange. That qualifies also, in my opinion, as a default and something that you want to avoid if you are an investor, because you’re going to lose money, both on the stock and the bond.

Finally, one other last thing. That is, what I was surprised to find over the years is that there was more interest in the Z models from equity investors rather than debt investors. Yes, there is a lot of interest by debt investors, but I was really surprised when Bloomberg—the financial data company that provides a lot of financial information to professional investors—told me that they have Z-Score calculated for any publicly owned company in the world and their customers who uses the model do so more for equity investments rather than debt.

CR: Interesting. Any ideas why the equity investors are more interested than the bond investors?

EA: Well maybe “more interested” is too strong. The bond investors are certainly interested in it, because they are very much trying to avoid the downside, and there’s not much upside with bonds after you are promised a certain yield. Yes, you can improve if the credit quality of the company improves or interest rates drop, but in general, there’s not much upside beyond the promised yield. With a stock, of course, there’s an upside and a downside. In both cases, everyone’s interested in avoiding the big downside, the tail of the distribution, where you lose 70%, 80%, 90% of your money. In the case of equity, you could lose 100%. Because often, the equity is wiped out in a bankruptcy. So yes, I can understand why the equity guys are really interested in the model as well, because if it goes bankrupt, they’re going to be wiped out in most cases.

Related

Financial Statements

16 Financial Ratios for Analyzing a Company’s Strengths and Weaknesses

Related

Investor Professor

Corporate Bankruptcy and Your Investments