Bit late to discover flaw in Shiller’s CAPE

I have recently come across two separate arguments against the idea that equities have become dangerously overvalued, even though the S&P has more than tripled from its mid-crisis low in early 2009.  I will assess them, respectively, here and in a follow-up post at some point.

To avoid suspense, my very humble take is that they are consistent with my claim that equities are not demonstrably overvalued, but are instead roughly fairly priced to deliver very subpar returns over the coming years. This is a fairly novel configuration, in place for perhaps a couple years now,  and it is irritating.

In fairness to the arguments themselves, they could be pressed to a stronger conclusion: that equities are a  buy here and will have to make another leg higher to get to the pricing I suspect is already be in place. I would not exclude that interpretation, fwiw. But I do not have the stones for it.

The two arguments follow. First, the WSJ has recently published a piece arguing that the Shiller CAPE overstates the extent to which the cyclically adjusted P/E multiple is elevated.  And second, I noticed Brad Delong linking to a piece by Antonio Fatas claiming that equities are actually quite cheap, once we take account of the low risk-free return available in Treasurys.

The second of these two arguments seems the more substantial, controversial and important.  And I will put up a post with my take on it, once I have done a put together a set of (typically simple) charts on the issue.  In this post, I will try to handle the simpler case of the WSJ’s argument, which strikes me as not so much directionally wrong as late.

The WSJ article is breezy and accessible, in a way that journalists not paid to deal with the tricky bits can easily write. So if I were you, I would just go read it. It is not a huge investment of time.

But here is the money image from the piece:


As produced by Robert Shiller himself, the Cyclically Adjusted P/E multiple in the S&P500 is about 27. This is far above its historical average and near the levels it traded just prior to the Global Financial Crisis, which cut US equities in less than half.

However, according to the WSJ piece, Shiller’s CAPE is deeply flawed because the GAAP earnings on which it is based are inconsistent over time (due to accounting rule changes) and currently incorporate huge write-offs taken during 2008 and 2009.

According to WSJ, then, it is better than to use a GDP-based measure of multiples, which is consistent over time, is not affected by write-offs, and shows valuation much less stretched. The “correct” cyclically-adjusted P/E at just under 19, by WSJ’s calculations.*

I would raise three points about this argument, in what I take to be ascending order of importance.  First, the GDP-based measure of valuation incorporates earnings from S Corporations, which cannot be traded and whose lack of public-market capitalization should bias the multiple downward, perhaps particularly when the market is rich.  The WSJ article goes over this, but does not come to a strong conclusion on it.  I will leave that debate to others, as I find it uninteresting, for reasons I will get to in a moment.

Second, even if we take Shiller’s CAPE at face value, it is not clear that its current level is dangerously inflated. If economic volatility is lower than average, as it appear to be, then the P/E should arguably be higher than average.

Returns will be lower because of this, a point I would rather emphasize than bury. But that is different from saying that the P/E itself is too high and that returns will inevitably be crushed by it soon returning to normal.

Closely related, it is not necessarily dangerous if equities are trading at the same valuation now as they were before the crisis.  Unlike the unwinding of the tech bubble during the early 2000s, the GFC was not brought on by overvaluation in equities. Equities were themselves arguably fine, although housing finance clearly was not.

Third, the real problem with the WSJ argument is that it should have been made many years ago — and was, much more effectively, by “Jesse Livermore” who writes over at  Philosophical Economics. If you have not been reading Jesse for the past few  years, then you have been missing out on a quite rigorous and perceptive assessment of equity market valuation — along with much else.

So far as I can tell, Jesse is gainfully employed. And yet he seems somehow to come up with the time to produce and give away far higher quality research than I could muster even with all the time in the world! I linked to him in this piece setting  out my own way of looking at equity market valuation, but you might want to leave out the middle man.

Jesse is not dwelling much on valuation these days, because he thinks the bull case for equities is stale and he has moved on to other issues. But back in the day, he was a wonderful antidote to the CAPERS and MEAN REVERTERS, whose self-righteous cocksureness was exceeded only by their wrongness, now admittedly evident with the advantage of hindsight.

Livermore pointed out the flaw in the Shiller CAPE several years ago and suggested that to get around it we should just sub in operating earnings for GAAP during the (recent) periods for which the former are available.  My take is that this approach is not perfect, but that it has the extremely useful advantage of highlighting whether the bearish signal from the Shiller CAPE comes from cyclically adjusting the earnings OR from the choice of the underlying earnings measure itself.  The perma bears too often glossed over that issue and left some free information on the table, in my view.

Take a look at this picture, comparing GAAP earnings (unsmoothed) with operating earnings, both for the S&P500.screen-shot-2016-10-24-at-9-14-52-am

As the bears might point out, operating earnings are a bit of a fudge, in the sense that they are almost always above GAAP earnings.  So if you want compare historical GAAP earnings with operating earnings, available only for more recent periods, then you need to make some adjustment.

My solution is to adjust historical GAAP upwards earnings by 10% to get a rough proxy of what operating earnings would have been had they existed in the more distant past. You could also discount operating earnings, but it makes no difference so long as you interpret the valuation signal in terms of the deviation from the historical average.

Note in the panel at right that “adjusted” GAAP earnings have usually tracked operating earnings, except prominently during the write-down periods at the turn of the century and during the GFC.  Robert Shiller would respond that that is all fine and good but that write-downs have happened and that we ought not airbrush them out of history.

I would neither dismiss nor get into that debate. Rather, my point here is to quantify the importance of it and to point out the WSJ argument was made more effectively by Jesse Livermore (and perhaps others) several years ago when big gains in equities were still to be taken. It is a bit late to discover now that Shiller’s CAPE is inflated by write-downs, however you choose to deal with them.

A closely related and (to me) fun point, to which I alluded above, is that this “controversy” around the Shiller CAPE now has much more to do with the choice of underlying earnings series (GAAP vs Operating) than it has to with Shiller’s attempt at smoothing for the business cycle itself.  Shiller-smoothed operating earnings are no longer even below one-year trailing earnings.  I would pause to internalize that.

The bears may be right (even if for the wrong reasons) that profit margins are due to come down, but they can no longer rely on Shiller’s approach to cyclical adjustment to make this point.  The Shiller CAPE is “stretched” because of its use of GAAP, not because of its attempt (roughly) to take out the cycle.


I would like to offer a couple minor technical points about the charts above before moving on. When Shillerizing even operating earnings, I use the PCE deflator rather than the CPI to convert nominal to real, for all periods in which the deflator is available.  The effect of this is tiny, and I do it only to be consistent with other bits of research where I emphasize the deflator.

Second, and more importantly, I center the Shillerized earnings series (and the multiple derived from it), by multiplying the entire history by roughly 1.09, which is the historical historical real earnings growth rate compounded over five years. Again, this has no effect on the valuation signal, so long as the valuation signal is interpreted as the deviation from its own average.

But this adjustment is required for the left panel above to be coherent and for the resulting multiple to be intuitively comparable with more popular metrics such as price to trailing or price to forward.  Basically, I want to follow Shiller in taking out the cycle as he does, while leaving the rest intact.

This brings me to my final bit of evidence here.


The difference between my measure of price of Shillerized operating earnings and the CAPE that would arise by using GAAP earnings (while retaining the rest of my method) is not actually that large right now.  Moreover, (absent recession or crisis) it is going to disappear in the next three years, as the crisis drops out of the 10-year moving average of GAAP earnings. As of right now, the differences in the multiples can be converted to a 60 basis point difference in the associated earning yield. As I measure it, the Shillerized (operating) earnings yield is 4.7%. Per Shiller’s CAPE, the GAAP earnings yield is 4.1%.

The direction of this difference cuts in favor of the conclusion that would be drawn by WSJ. So there is that.  But I think it makes more sense to follow Jesse Livermore and to sub in operating earnings rather than GDP-based earnings when making this “correction.”

More to the point, Jesse reminded us to do this several years ago. And if you follow his approach — or my riff on his approach — then there would seem to be much bigger fish to fry than those 60 bps.

For example, what to make of the fact that the competitive risk-free asset is offering roughly nothing in real terms?  That is the much tougher question, and I really doubt I will give an authoritative answer to it. But in a couple days to a week, I might have some typically-simple charts to show you on that.

* I was able roughly to replicate them. I assume but have not confirmed that the main difference is my use of book, rather than CCA- and IVA-adjusted earnings.