Peter St Onge

A few weeks ago I talked about how Big Data makes economic theory important in investing. This week we’ve got a nice illustration, courtesy of everybody’s favorite billionaire Mayor of New York.

Last week Bloomberg Business published an article that’s typical of the data-driven and theory-free fad of most financial analysis.

“American CEOs Most Bearish on Earnings Since 2008 Crisis” cites surveys, then traces out the main reasons for this bearishness. Which, it turns out, are one-off, ex-cycle events: the strong dollar and weak oil. (The mechanisms, by the way, are straightforward: the strong dollar reduces overseas revenue and profits, while weak oil hits the oil industry. And the oil industry disproportionately dominates large US indices.)

So what’s the problem? The article depends on an apples-to-apples comparison (“since 2008”) that’s entirely driven by 1-time events. A bit like seeing a balloon float away and worrying that earth’s gravitational field might be reversing.

This is exactly the problem with relying on data alone: how do you know what to ignore? If you hear that xyz happened, and last time it happened the world fell apart, how to tell when the world really is falling apart?

Without theory, anything at all becomes a respectable apples-to-apples indicator. Because statistics. Problem is, there are literally billions of correlations one could run. There’s even a site for such spurious correlations — Nicolas Cage films and swimming-pool drownings, for example, are suspiciously correlated. Of course, spurious correlations are awesome for media, since the more reasonable-sounding ones are great for a quick article on deadline. But the problem for theory-free analysts boils down to: how do you know what to ignore?

Ideally, what we’d do is strip out one-time events that are unrelated to the cycle. Without theory, though, how do we know what’s true signal and what’s noise?

So let’s start with the supposed signal from corporate profits. On their own, corporate profits are decent indicators of stock prices, but they suffer from a long and variable lag. In other words, they’re too early, and unpredictably so.

Let’s go to the data. Here’s a chart of corporate profits for the past 3 recessions. It’s log scale, so think of it as a percentage change (if you don’t use log then recent numbers look overly dramatic — this is why the world always seems spinning out of control for the log-challenged):

Let’s start with the early 1990’s recession. Bush Senior was in office, the Berlin Wall was falling down, and Madonna was making a come-back. Corporate profits started dipping in Q4 1988. Which was actually a full two years before the market started dropping. Now, it was a mild recession: stocks fell about 20% at worst. So that could explain the muted and too-early signal?

Or not. Next up: the dot-com bust, when dressing like the Matrix became uncool. This one was more serious: it knocked about 50% off the S&P starting mid-2000. Again, though, we’ve got about a 2-year delay: corporate profits started falling in Q1 1998. In fact, you’d have missed out on 50% gains if you’d bailed with the “profits are falling” signal.

Then, we’ve got everybody’s favorite, the 2008 crisis. Here, the lead-time was shorter: profits started leveling off in about Q4 2006. Stocks kept rising until October, 2007 before falling, rebounding then plunging in the wake of the Lehman Crisis.

This should worry you. Not only is the lag too variable, but, if you’d followed the 2-year-delay rule from the previous 2 recessions, then you’d have stayed in through the Lehman collapse, and you’d have actually sold at nearly the absolute bottom, in Q4 2008. You’d have missed the rebound and, it turns out, should’ve just spent that recession on a remote South Pacific island blissfully unaware of your portfolio performance.

So that’s the data itself. Mind this is all public information, no theory at all yet. We can talk about journalists’ reluctance to inform readers that a particular indicator is typically years early. But so far we’re just noting that profits do fall before recessions, but they fall too early so delay your reaction.

The second problem here is that even if profits do reliably foreshadow recessions, the years-long delay suggests the relationship is not causal. That is, falls in corporate profits are not themselves causing recessions. Why does this matter? Because if something is an indicator and not a cause, then it becomes very important to be able to sort signal from noise. A piano falling on your head is a cause, so you don’t really care why it’s falling. But rattling backdoors on windy nights is an indicator. Meaning it’s important to filter the noise before shooting up the porch.

So, if drops in corporate profit and recessions are both caused by some outside factor — the credit cycle, say — then corporate profits would only have value as an indicator of recessions. Meaning that the specific causes of profits revisions cited in the article — the strong dollar and oil drop — are too remote to the cycle to constitute signals. They are noise. And should be aggressively ignored. You certainly wouldn’t run around comparing them to 2008 without a lot of asterisks.

Given that other cycle indicators are benign, I’d interpret recent profits revisions as indeed no more than one-off noise. Both the strong dollar and weak oil are only obliquely related to the cycle, with too many intermediaries to do any more than confirm what we already know: that we’re in mid-to-late cycle but closing time’s still awhile yet. At most put a reminder on your calendar to check back in a year. But, really, it’s probably just the wind.

If you liked this article, please share it with a friend. And if you haven’t already, consider subscribing in for semi-weekly updates in the box below. I’m happy to hear suggestions for topics so if you’ve got one email me here.