We all know the hazards of using data from the past to support assumptions about the future. Less widely discussed, however, are the challenges in using historical data to tell us about the past. My tendency to drive Portfolio123.com users crazy fussing over a metric I refer to as "Business Income" represents an effort to address this. It's still a work in progress. But efforts like this are important if we are to get databases to tell us what we think they're telling us.
One of the least discussed issues in contemporary stock evaluation is that of the misspecified model. This occurs when we use data that does not provide the information we're expecting it to provide.
This is not a matter of inaccuracy. In fact, modern financial data vendors do a pretty good job correctly translating financial disclosures, mainly 10-K and 10-Q documents, into electronic format. Actually, the problem is the opposite. Because they're so good at what they do, investors tend to get complacent and assume what they're giving us is what we need. All too often, it isn't.
For an example of how security analysis can wander down the wrong path even when our data is accurate, start with Figure 1, which is extracted from a 10-K filed by Constellation Brands (STZ) for the year ended 2/28/09.
What was the fiscal 2009 operating margin?
The database reports it as 0.63%, based on operating income of $23.0 million divided by revenues, which were $4.723 billion. This is what investors would get when they look on the web for the operating for fiscal year that ended 2/28/09. Those who wish to examine a three-year trend would also see margins of minus 7.3% for the 2/08 year, and 10.9% for the 2/07 year.
Putting it all together, we might think we're looking at a company that got badly hammered over the course of a year but now seems to be on the mend. Perhaps we should consider this a cyclical recovery play.
Unfortunately, that would be an incorrect answer.
The classic Graham and Dodd's Security Analysis (Cottle, Murray, Block, McGraw Hill, 1988) states as follows on page 138:
Analysis requires the ability to discriminate, to separate the ordinary from the unusual, and to detect change. These require disaggregation, enhancement, and reconstruction of accounting information to prepare it for further processing.
It's not enough to simply look at the past, even if we have good reason to assume the information we're seeing is correct. We have to pick and choose among those aspects of the past that are most likely to help us form rational assumptions about the future. (That's why we use such words as "analyze," "analysis," "analyst." Our examination of data must be an active one; we can't just look at it.)
Graham and Dodd discuss various ways analysts should recast the data. The one we're concerned about in this example, the one mentioned in the quoted language, involves separating the ordinary from the unusual.
The margins we should be using are 10.9% for the 2/09 year, 12.6% for the 2/08 year, and 14.5% for the 2/07 year. Rather than working with reported operating profit, we should have added back unusual charges. Doing so, we see now that we are not dealing with a cyclical collapse-and-recovery. Instead, we're seeing a gradual deterioration in margin that, at least at first glance, may not have anything to do with the economy. We don't know if this is likely to continue even as GDP improves. But we are being warned that it would probably be prudent to assume ongoing deterioration unless and until further investigation gives us reason to believe otherwise.
STZ would have passed a stock screen that required operating margin (or EBIT margin, pretax margin or net margin for that matter) to have improved in the trailing 12 month period. That would have been an example of a misspecified model. In truth, the 2/09 year (that aspect we need to consider if we are to properly analyze the stock) was lower than the one recorded in fiscal 2/08.
A Challenging New World
Back in the days before electronic databases speedily and automatically disseminated financial information, none of this would have been worthy of discussion. Anybody who knew how to evaluate stocks would have worked with operating margins of 10.9%, 12.6% and 14.5% for the last three years respectively.
In fact, that's what analysts do regularly today. If you look closely at how historical EPS is presented, you'll see important differences between the numbers on web pages that show financial statements and web pages that show the historical earnings used as a basis for computing earnings surprise. Look at the web site of your choice with STZ. You'll see it!
The analysts are restating historical results in the way described above and estimates are compared with these refined historicals in order to determine if there's been an EPS surprise. That's why historical EPS is recorded very differently depending on whether you're looking at a fundamentals database or an estimates database, even when both are produced by the same vendor.
There's no easy fix here. We can't just insist that the fundamental data get better. Notwithstanding all the politics and rhetoric surrounding Wall Street analysis over the past decade, the truth is that analysts are well trained in the process of deconstructing financials in ways that will support the process of forecasting. The historical numbers you see in an estimates database reflect thoughtful decisions on the part of people who have the wherewithal to make such judgments.
The historical numbers you see in a fundamentals database reflect copying (accurate copying from the best of sources, but copying nonetheless) on the part of people who may or may not know what stocks or stock analysis is and who, nowadays, tend to be based in offshore locations selected mainly to cut costs. As tempting as it is to wish those who input fundamental data would do so more thoughtfully, we might do best to remember the expression "be careful what you wish for." Who knows what we could wind up with if we were to demand such judgments on the part of people who haven't been trained for this sort of thing.
Disclosure: No positions in STZ