Update: The charts below are based on reported earnings, which I converted to constant 2013 dollars using the CPI.
Back in May, I suggested the stock market was showing some signs of froth, although “maybe in the same way that froth was apparent in 1997 or 2006, several years before the respective peaks.”
Before making that statement, I checked a few valuation indicators that I use to gauge such things. I was especially interested in a comparison of three types of price-to-earnings (P/E) multiples, which I’ll share here.
The idea isn’t to forecast a market top, since P/E multiples are useless for predicting turning points. But they can tell us something about the risks we may face when the bull market finally runs out of steam (whether that’s what we’re witnessing this month or not).
I’ll start with the not-too-bad news.
The first chart shows a traditional P/E multiple (based on trailing four quarters earnings) and compares today’s valuation to the last eight bull market peaks:
You may be concerned by the proximity of today’s P/E to the peaks in 1966, 1968 and 1973. But the market pushed much higher before correcting in 1961, 1987 and 2000, and there’s still more than a full P/E unit separating today from the 2007 peak. Optimists can take some comfort from these results, and especially if they expect earnings to climb higher.
Which brings us to, well, earnings.
It’s important to keep three things in mind when we talk about earnings:
- In economic expansions, the consensus always calls for increasing earnings. Forecasters may expect a short-term dip if company guidance has been weak, but the consensus projection invariably reverses the dip a quarter or two out.
- Based on (1), the consensus never anticipates a prolonged drop in earnings.
- History shows that earnings often defy the consensus and fall further and longer than expected, reverting to a rising trendline.
Here’s a chart that demonstrates (3):
Especially in recent economic cycles, it’s clear that changing perceptions about earnings explain a substantial portion of the market’s volatility. Just as investors can easily forget that P/E doesn’t rise forever, they sometimes forget that earnings don’t climb forever. And when earnings are unusually high, traditional P/E multiples fail to capture the full risk of a correction.
Yale economist Robert Shiller had these challenges in mind when he suggested a different P/E measure over 15 years ago. Instead of using four quarters of data, he averages earnings over the last ten years to smooth the ups and downs and plugs the result into the denominator of the P/E multiple.
Here’s a different version of the first chart above, this one using Shiller’s P/Es:
The market looks more expensive today with Shiller’s approach than it does with traditional P/Es. But I’m not finished just yet. A third way to capture earnings risk is to divide price by a “normalized” earnings estimate that’s plucked from a trendline, such as the trendline in the second chart above. Although trendlines are notoriously unstable, I like this approach because it accounts for long-term earnings growth (and also because it’s visual).
To compare today’s P/E to past bull market peaks using this third method, I calculated a separate trendline for each peak. In each case, I used only information that was available prior to the peak. (For further details, see the “technical notes” link at the bottom of the article.)
Here are the results (please have a look, Jeremy Stein, and give us your thoughts on stock market froth as you did for the bond market in February):
The chart shows that today’s valuation may be comparable to what we were seeing a few years before the 2000 and 2007 bull market peaks. This is the conclusion I flagged in the May article, when I suggested investors may be experiencing another 1997 or 2006.
More ominously, we’ve blown past all of the other six bull markets in the 50+ year data set. In my opinion, this isn’t a welcome result. It marks valuation risk as unusually high, exceeded only by the truly extreme levels reached in the last two market cycles.
What Does Price-to-Trend-Earnings Tell Us About the Fed’s Policy Mix?
The price-to-trend-earnings multiples also raise questions about the Federal Reserve’s long campaign to prop up asset prices through unconventional policies. I’ve argued in other articles that the “wealth effects” sought by the Fed are mostly bringing forward gains that would have otherwise occurred in the future. They’re weakening tomorrow’s growth in return for a shot in the arm today.
Of course, policymakers would like you to believe their actions are stabilizing. But the last two decades suggest otherwise. And the chart above reinforces the risk that we’re stuck in a Groundhog Day-like loop of living through the same boom-bust cycle over and over. It shows that the next policy-induced bust may be gradually coming into view.
(Click here for technical notes about this article and a few more charts.)
Update 2: If anyone read the Horan Capital Advisors response to this post and wondered if their conclusions are accurate (as of right now, they’re not), please see the comment below from reader Lance Paddock and my reply.
Excellent piece, thx. The missing piece to complete the ensemble would be to show forward P/E (but I don’t know how difficult that’d be to calculate for past time periods). The comparison of those 4 methodologies would really help to show more people the shenanigans (yep, that’s right–I said “shenanigans” dammit!) being played by Wall St. to fuel further gains. Keep up the good work.
You’re right – great piece of work. For me, I’d rather see P/E based on trailing earnings – those are “in the bag”, so to speak. They are real, established data points. Forward P/E are based on a moving target, where companies might continue to change their guidance for a subsequent earnings report.
What I’d like to see, is this same approach using GAAP EBITDA.
In any event – this is terrific stuff, thanks!
Excellent article! Would be interesting to see the same analysis for the market bottoms following these peaks.
Thanks to all for reading and commenting – I’ll keep your suggestions in mind for the future. The post below shows a price-to-forward-earnings history (difficult to interpret, though, because of the way forecasters look through the earnings dips):
http://www.zerohedge.com/news/2013-07-29/faith-hope-and-pe-multiple-expansion
I would be interested in your thoughts on this blog post. http://disciplinedinvesting.blogspot.com/2013/08/are-s-500-earnings-above-or-below-trend.html
I hastily put together a few thoughts which I paste below. Possibly I am not thinking everything through correctly, but they are my first few thoughts.
“David,
This is an important issue, and your point about appropriate scales is well taken in general.
However, I noticed you struck out a sentence, which ties into my point. The scale being used isn’t an issue with the illustration you critique. I am glad you recognized that, but your striking of the sentence without clarification makes it seems as if it is an issue still.
Of more concern to me is whether we are comparing apples to apples. The exponential earnings trendline used by the original author starts in 1950. Yours starts in 1985 when earnings were below the trend since 1950 and end now when earnings are above the trend since 1950. This leads to an upward bias in the data it appears to me.
In addition, the period you chose has been one dominated by a long period of exceptionally high profit margins, while revenues have been relatively normal for the period as a whole. This has led to a period where earnings growth was higher than typical. Unless that can be repeated (continuing growth in profit margins) then future growth should track more closely revenues. The trend you illustrate in that case would be more misleading still.
That assumes profit margins average what they have in recent years. If not, and they revert to past levels, then we should expect profit growth to not resemble the more recent trend you illustrate by a wide margin over the next 10-20 years. In fact, over the next 10 years we would expect little to no real growth were that to happen, which would track closely the trendline of the original author. Adding in the starting and end point issues of using a shorter period such as you use would lead to a wild overestimation of potential profit growth based on the trend you illustrate.
Finally, you don’t say whether you are using as reported or operating earnings. While operating earnings have their uses, the two types of earnings have diverged more and more over time even if one regarded operating earnings as superior (I don’t, but that is another discussion.) Over time as companies have changed their practices in how operating earnings have been used has resulted in a growing divergence between the two series which would result in another upward bias to the earnings trend you illustrate that would not be replicable in the future unless such a divergence would grow at a similar rate over time. Obviously if it were to shrink the implications would be starkly different. So, if you are using that series I would suggest that makes the issues with the specific period chosen even larger. For consistency in reporting over time reasons measuring trend growth with operating earnings is just plain problematic even if you feel they are a more useful than as reported earnings now and in the future for other purposes.
Anyway, those are the first few questions I would like to see addressed.”
Thanks Lance,
I also reached out to the author of this (hence his striking out of some of his original post), although my response wasn’t as thoughtful as yours.
I made two points. First, an exponential trend line with a linear axis gives the exact same result as a linear trendline with a log axis, as far as the relative positions of the earnings data and the trendline. So, the author’s explanation for the difference in results just isn’t accurate.
Second, the explanation that I offered is that he’s using a forward earnings measure. I think your points are dead-on also, but I suspect the use of forward earnings is another big reason for the difference because it eliminates most of the big dips in recessions. You’re basically fitting a trendline to an always optimistic view of earnings that’s especially inaccurate in and before recessions, which is the part that biases the trendline upward.
With another look, though, your point about the ’85 starting point may be just as important. The forward earnings data is available back to about ’76 (per the Zero Hedge link in my earlier comment) and I think if you used the full history it would bend the trendline down as you say.
I’ll readily acknowledge that the “true” earnings trend (if there is such thing) could be a bit higher than the $61.60 figure from my 40 year regression, but I also believe that the Fed has brought future growth forward and inflated current earnings above anything that’s long-term sustainable. I think we’ll see earnings below the trendline once again in the next recession.
Also, I almost forgot to mention one more apples-to-oranges on my post vs. Horan – I inflation-adjusted the earnings data before calculating a trendline but the other post looks to be nominal.
Thanks again for your comment. I thought about writing a short post on the Horan trendline but didn’t think many people had seen it.