BH Greatest Hits: PE10 predictive power

Discuss all general (i.e. non-personal) investing questions and issues, investing news, and theory.
Post Reply
Topic Author
grayfox
Posts: 5569
Joined: Sat Sep 15, 2007 4:30 am

BH Greatest Hits: PE10 predictive power

Post by grayfox »

Stocks have had a pretty good run and it seems that there is a lot of discussion about high valuations, and what that means for future returns. Of course, one of the favorite valuations to discuss at Bogleheads is P/E10, a.k.a CAPE or Shiller PE Ratio. You can find Shiller PE Ratio plotted here. Currently P/E10 is 25.17. The historical mean and median were 16.52/15.91, so 52%-58% above historical valuation.

What does this mean?

Well instead of starting a whole new discussion on it, the subject has been discussed in the past in this thread: PE10 predicitive [sic] power.

The thread starts with the scatter chart of 10-YR returns vs. P/E10, which shows what looks like some amount of correlation between 10-Year Return and P/E10. [I think there also used to be a 20-year return chart, but the image link is dead.]

This is followed by the astute comment that overlapping data reduces the statistical significance. It may look like hundreds of data points, but its really only about 6 independent observations for 20-year returns. Is it 13 for 10-year returns?

Another astute commenter casts doubt on the reliability of P/E10 mean reverting in your lifetime and trying to use P/E10 to forecast imminent market crashes.

But a poster makes a good point in this post that even without mean reversion, high valuations should logically lead to low returns.
The mean reversion refers to the change in P/E or Bogle’s speculative return. If P/E doubles or halves over 20 years, that adds or subtracts about 3.5% p.a. to returns. ... Even if P/E10 remains constant for the 20 years (no mean reversion), the Gordon equation says real returns will be higher for a lower P/E10.
But then the OP counters with "If a low PE10 predicts low earnings growth, then Gordon doesn't help the argument." and later remarks:
A low PE10 may mean we are paying less for future earnings (a bargain) or it may mean that the market is predicting low earnings and is paying a full price for them. There is no way to know ex ante.
But later camontgo posts a link to a 2008 paper by John Cochrane from the University of Chicago The Dog That Did Not Bark which concludes that the evidence shows that dividend growth has not been forecastable by the market. Therefore, returns must be forecastable.

In other words, Cochrane is saying thay high valuations forecast low returns, not high earnings and dividend growth. This would resolve the apparent dilemma in the previous quote.

Another interesting point of view is expressed in this post. The poster says:
I am confident that buying stocks at lower prices and selling them at higher prices is a good investment strategy. If the statistical evidence does not reject that hypothesis, that is further proof of it, not disproof. Inconclusive statistical evidence is not sufficient to convince me that the price paid for stocks does not matter.
This turns the tables. Instead of asking whether or not the data proves that P/E10 forecasts returns, the null hypothesis is:

H0: Lower/Higher valuations lead to higher/lower returns.

Now ask if H0 should be rejected based on the data we have available.

There are a lot of other good ideas presented in that thread, although it starts to get a little technical further on. A lot more discussion about overlapping data. But the thread is worthwhile reading before starting a new discussion on valuations.

There are probably other good threads about P/E10 which have probably already hashed though everything.
beammeupscotty
Posts: 187
Joined: Thu Jan 21, 2010 7:13 am

Re: BH Greatest Hits: PE10 predicitive power

Post by beammeupscotty »

No doubt the hypothesis is correct, but that doesn't mean you're likely to beat market returns by going to cash (or "tactical" re-allocation), waiting for a crash, and timing the move back in. Judging by some of the recent posts on this forum, many are already way behind by sitting out the last 3+ years due to the same concerns.

https://philosophicaleconomics.wordpres ... should-be/
LittleD
Posts: 227
Joined: Sun Jul 31, 2011 7:23 am

Re: BH Greatest Hits: PE10 predicitive power

Post by LittleD »

Not many here who use CAPE PE/10 go 100% to cash. Most that I have read just subtly move allocations from say 60/40 stocks-bonds to something with
less risk say 55/45 or 50/50. Some do a bit more if PE/10 goes over 30 which it is not today such at 40/60 or 30/70. This is nothing more than reallocating
the portfolio for expected lower returns in the future and the higher volatility that comes with it. None of us knows if the market will crash or go on a
sustained bear market sojourn given above normal valuations but we who do this loathe volatility and are willing to get less market return to avoid it.

In addition, some of us use a companion indicator from Buffett called the Total Market/GDP ratio which is also over extended right now and suggests
market overvaluation. I don't intend to move to 100% cash and most of those who think valuations matter do not either.

Now if the market does swoon we will be well positioned to move strong into stocks when "blood is in the water" and our allocation might again go to
70/30 or 82/20 stocks-bonds... Just sayin'...

Good Luck with your investments!!!
larryswedroe
Posts: 16022
Joined: Thu Feb 22, 2007 7:28 am
Location: St Louis MO

Re: BH Greatest Hits: PE10 predicitive power

Post by larryswedroe »

While the 113-year mean for the CAPE 10 is 16.5, it’s mean since 1960 is a much higher 19.6. This is important because there are logical reasons to believe that over time the risk premium investors demand from U.S. stocks should have fallen.
First, all else equal, the wealthier a nation, the lower the risk premium we should expect. Consider frontier markets, where capital is a very scarce resource. Economic theory tells us that the scarce resource earns the higher “economic rent.” In addition, those countries typically have weaker regulatory environments in terms of investor protections. And it’s often the case that foreign investors have even less protection than domestic ones. Thus, the cost of capital in such markets is very high — valuations are low and expected returns high. As those countries progress over time from frontier to emerging to developed, the cost of capital tends to fall as capital becomes less scarce and the regulatory environment is strengthened. Thus, it shouldn’t be a surprise that the mean CAPE 10 has migrated higher over the past 50 plus years.

There are also some other issues that have been raised about the use of the CAPE 10. One relates to the issue that far fewer companies pay dividends than was the case in the past. Today, something like 60 percent of U.S. stocks don’t pay dividends, and 40 percent of non-U.S. don’t pay them. In the U.S. that has resulted in the dividend payout ratio on the S&P 500 falling from an average of 52 percent from 1954 through 1995, to just 34 percent from 1995 through 2013. At least in theory, higher retention of earnings should result in faster growth of earnings as those retained earnings are reinvested. And that has been the case for this particular period as from 1954 to 1995 real EPS growth rate averaged 1.72 percent, and from 1995 to 2013 it averaged 4.9 percent.
The website Philosophical Economics has a blog on this subject. The author explains that in order to make comparisons between present and past values of the Shiller CAPE, we need to normalize for differences in payout ratios. Making the adjustment between a 52 percent payout ratio (the average of 1954-1995) and a 34 percent payout ratio (the average since 1995) corresponds to around 1 point worth of Shiller CAPE.
The second issue relates to the change in accounting rules regarding writing off goodwill. As Philosophic Economics explains:
“In the old days, GAAP required goodwill amounts to be amortized–deducted from earnings as an incremental non-cash expense–over a forty year period. But in 2001, the standard changed. FAS 142 was introduced, which eliminated the amortization of goodwill entirely. Instead of amortizing the goodwill on their balance sheets over a multi-decade period, companies are now required to annually test it for impairment. In plain english, this means that they have to examine, on an annual basis, any corporate assets that they’ve acquired, and make sure that those assets are still reasonably worth the prices paid. If they conclude that the assets are not worth the prices paid, then they have to write down their goodwill. The requirement for annual impairment testing doesn’t just apply to goodwill, it applies to all intangible assets, and, per FAS 144 (issued a couple months later), all long-lived assets.”
While FASB 142 may be a more accurate method of accounting, it has created an inconsistency in earnings measurements with the present values end up looking more expensive relative to the past than they actually are. And the difference is dramatic. While the CAPE 10 as now measured is about 24.9, adjusting for the accounting change would put it about 4 points lower.
If we combine the two adjustments of 1 for the lower dividend payout and 4 for the FSB 142 change, the current CAPE 10 at 24.9 which looks way above the mean, doesn’t look so overvalued at a now 19.9. In fact, that’s right about in line with it’s average since 1960. Which begs the questions:
a) Are stocks really overvalued or just highly valued (meaning returns are likely to be lower than historical levels, but that there is no reason to expect a major correction due to RTM)?
b) To what mean should the CAPE ratio revert: The 16.5 mean of the past 113 years, or the 19.6 mean since 1960?
These three points — that over time it’s logical to believe that the equity risk premium for U.S. stocks might have fallen, the accounting change regarding write offs, and the lower payout ratios — provide us with plausible explanations that the CAPE 10’s high level is not signaling a massive overvaluation of U.S. stocks
richard
Posts: 7961
Joined: Tue Feb 20, 2007 2:38 pm
Contact:

Re: BH Greatest Hits: PE10 predicitive power

Post by richard »

grayfox wrote:
A low PE10 may mean we are paying less for future earnings (a bargain) or it may mean that the market is predicting low earnings and is paying a full price for them. There is no way to know ex ante.
But later camontgo posts a link to a 2008 paper by John Cochrane from the University of Chicago The Dog That Did Not Bark which concludes that the evidence shows that dividend growth has not been forecastable by the market. Therefore, returns must be forecastable.

In other words, Cochrane is saying thay high valuations forecast low returns, not high earnings and dividend growth. This would resolve the apparent dilemma in the previous quote.
How does one prove dividend growth has not been forecastable by the market? Perhaps the price of risk has changed?

Assuming dividend growth is not forecastable, does that necessarily mean that returns are forecastable? Why?

Thanks for the trip down memory lane. It's interesting reading what I said four years ago.

These days I'd add this from a Vanguard research paper: "We confirm that valuation metrics such as price/earnings ratios, or P/Es, have had an inverse or mean-reverting relationship with future stock market returns, although it has only been meaningful at long horizons and, even then, P/E ratios have “explained” only about 40% of the time variation in net-of-inflation returns. Our results are similar whether or not trailing earnings are smoothed or cyclically adjusted (as is done in Robert Shiller’s popular P/E10 ratio)."
Topic Author
grayfox
Posts: 5569
Joined: Sat Sep 15, 2007 4:30 am

Re: BH Greatest Hits: PE10 predicitive power

Post by grayfox »

richard wrote: How does one prove dividend growth has not been forecastable by the market? Perhaps the price of risk has changed?

Assuming dividend growth is not forecastable, does that necessarily mean that returns are forecastable? Why?
I'm not familiar with the actual research, just the summary results that are reported by Cochrane in a 2011 paper "Presidential Address: Discount Rates". There was a whole thread devoted to it. I recall spending a lot of time studying that paper and posting about it.

I'm guessing that the researches must have looked at returns starting in times of high, average and low valuation. The high/low valuation would imply that either 1. dividend growth will be high/low or 2. returns will be low/high. It's got to be one or the other, or some combo.

The Constant Expected Return (CER) model would lead to 1.
So they must have observed that it was 2 that prevailed, which meant the end of the CER model to forecast returns.
richard wrote: Thanks for the trip down memory lane. It's interesting reading what I said four years ago.
I think those older threads made a lot of progress on understanding what valuations can and can not tell you. It seems that on more recent threads, everyone is back to square one. I think anyone can benefit by reading the old threads before starting a new one about P/E10.
User avatar
camontgo
Posts: 329
Joined: Tue Oct 27, 2009 1:46 pm
Location: Chicago, IL
Contact:

Re: BH Greatest Hits: PE10 predicitive power

Post by camontgo »

richard wrote:How does one prove dividend growth has not been forecastable by the market? Perhaps the price of risk has changed?

Assuming dividend growth is not forecastable, does that necessarily mean that returns are forecastable? Why?
Example:

Suppose I purchase the opportunity to draw a "payoff" from some distribution of possible outcomes. You purchase a similar opportunity to independently draw a "payoff" from the same distribution, but you pay a higher price for the opportunity.

Is my expected return not higher? Of course, in any particular instance your realized return might be higher...but not over many trials.

Now, it could be that the market pays a higher price for shares at some times because the market has some limited ability to forecast higher growth. In other words, the market has some knowledge about the distribution of likely outcomes. However, in that case we should see some relationship (perhaps very noisy) between valuations and subsequent earnings/dividend growth. We don't see this in stocks...or in many other asset markets.

So, there is no evidence that the market's high or low valuations are correlated with future changes in the income stream. If we don't see evidence that market valuations forecast changes in the income stream of the assets, then doesn't the situation looks similar to my example above?

Some references to other asset markets where growth hasn't shown correlation with valuation(though no links to papers):

http://www.bloombergview.com/articles/2 ... h-cochrane

Of course, the future may be different. Maybe the market will get better at forecasting growth...but the past data, across many types of assets, doesn't show much correlation between valuation and subsequent growth.

BTW, I agree that the price of risk changing may account for valuation changes...in which case PE10 may be a very poor timing tool. In fact, I think it is a poor timing tool. Lower expected returns don't necessarily mean that we should expect mean reversion.
"Essentially, all models are wrong, but some are useful." - George E. P Box
Topic Author
grayfox
Posts: 5569
Joined: Sat Sep 15, 2007 4:30 am

Re: BH Greatest Hits: PE10 predicitive power

Post by grayfox »

Here's an October 14, 2013 update on the chart from Brad DeLong
http://delong.typepad.com/sdj/2013/10/c ... -earn.html

(wait, the x-axis says Current Price Divided by a Thirty-Year Moving Average of Earnings, So Brad Delong is now using P/E30 ?!

Note: J. Bradford DeLong is an economist teaching at the University of California at Berkeley.
richard
Posts: 7961
Joined: Tue Feb 20, 2007 2:38 pm
Contact:

Re: BH Greatest Hits: PE10 predicitive power

Post by richard »

camontgo wrote:
richard wrote:How does one prove dividend growth has not been forecastable by the market? Perhaps the price of risk has changed?

Assuming dividend growth is not forecastable, does that necessarily mean that returns are forecastable? Why?
Example:

Suppose I purchase the opportunity to draw a "payoff" from some distribution of possible outcomes. You purchase a similar opportunity to independently draw a "payoff" from the same distribution, but you pay a higher price for the opportunity.

Is my expected return not higher? Of course, in any particular instance your realized return might be higher...but not over many trials.

Now, it could be that the market pays a higher price for shares at some times because the market has some limited ability to forecast higher growth. In other words, the market has some knowledge about the distribution of likely outcomes. However, in that case we should see some relationship (perhaps very noisy) between valuations and subsequent earnings/dividend growth. We don't see this in stocks...or in many other asset markets.

So, there is no evidence that the market's high or low valuations are correlated with future changes in the income stream. If we don't see evidence that market valuations forecast changes in the income stream of the assets, then doesn't the situation looks similar to my example above?<snip>
One major difference between the two cases is that in the first instance we had a constant underlying distribution while in the later case the underlying distribution could change over time.

The empirical result from Cochrane's paper is that p/e does not forecast earnings/dividend growth. The conclusion is that returns are forecastable. I continue not to see the conclusion as following from the empirical work as a matter of logical necessity. Is there another interim step or assumption I'm missing? EDIT: I may have answered this in the immediately following post?
Last edited by richard on Wed Apr 09, 2014 10:21 am, edited 1 time in total.
richard
Posts: 7961
Joined: Tue Feb 20, 2007 2:38 pm
Contact:

Re: BH Greatest Hits: PE10 predicitive power

Post by richard »

grayfox wrote:Here's an October 14, 2013 update on the chart from Brad DeLong
http://delong.typepad.com/sdj/2013/10/c ... -earn.html

(wait, the x-axis says Current Price Divided by a Thirty-Year Moving Average of Earnings, So Brad Delong is now using P/E30 ?!

Note: J. Bradford DeLong is an economist teaching at the University of California at Berkeley.
DeLong writes "This is a major, major, empirical win for Campbell and Shiller. This is why only fools say today that movements in market-wide price-earnings ratios are best interpreted as shifts in rational expectations of future earnings and dividend growth. Instead, they are best interpreted as due to "fads and fashions" in how much people are willing to nerve themselves to pay for a dollar of earnings today"

He's saying the only possible choices are (1) p/e predicts returns, (2) p/e predicts earnings/dividend growth, (3) the price of risk changes over time, and implies these are mutually exclusive. He picks (1) as an empirical matter, rejects (3) as an empirical matter and therefore rejects (2) as a matter of logic.

Cochrane appears to regard (1) and (2) as the only possible choices and assumes they are mutually exclusive. He rejects (2) as an empirical matter and therefore picks (1). He ignore (3).

Any disagreement with these interpretations?

I believe I agree with DeLong that 1, 2 and 3 are the only logical possibilities. I'm not convinced they are mutually exclusive.
Topic Author
grayfox
Posts: 5569
Joined: Sat Sep 15, 2007 4:30 am

Re: BH Greatest Hits: PE10 predicitive power

Post by grayfox »

My simple understanding is based on the Gordon Model.

Dividend Yield + Growth = Expected Return
D/P + g = r

Put some number on it. D/P = 4%, g = 3%, r = 4 + 3 = 7

If price doubles so that D/P = 2%, then either 1) g must increase or 2) r must fall, or 3) some combination.
The Constant Expected Return (CER) Model would say that g must increase to 5, i.e. the market is forecasting higher growth
2 + 5 = 7

Cochrane is saying that r must fall to 5. Expected return has fallen.
2 + 3 = 5
richard
Posts: 7961
Joined: Tue Feb 20, 2007 2:38 pm
Contact:

Re: BH Greatest Hits: PE10 predicitive power

Post by richard »

grayfox wrote:My simple understanding is based on the Gordon Model.

Dividend Yield + Growth = Expected Return
D/P + g = r
Current dividend yield plus future dividend growth = future return.
grayfox wrote:Put some number on it. D/P = 4%, g = 3%, r = 4 + 3 = 7

If price doubles so that D/P = 2%, then either 1) g must increase or 2) r must fall, or 3) some combination.
The Constant Expected Return (CER) Model would say that g must increase to 5, i.e. the market is forecasting higher growth
2 + 5 = 7

Cochrane is saying that r must fall to 5. Expected return has fallen.
2 + 3 = 5
Constant expected returns assumes away the major issues.

I believe I respond to Cochrane in the post immediately above yours.
User avatar
camontgo
Posts: 329
Joined: Tue Oct 27, 2009 1:46 pm
Location: Chicago, IL
Contact:

Re: BH Greatest Hits: PE10 predicitive power

Post by camontgo »

richard wrote: One major difference between the two cases is that in the first instance we had a constant underlying distribution while in the later case the underlying distribution could change over time.
Yes, it will change over time.

It could be that the market is more risky at some times than others, and this drives changes in valuation. I don't think that is inconsistent with Cochrane's argument. If the risk is changing, it would mean that market participants shouldn't be overly excited by low or high valuations.

However, it is still true that if there is no correlation between valuation and the average level of growth then there will be a correlation between valuation and average return.

Bogle uses this equation for expected returns:

Return = Yield + Growth + Speculative Return (change in valuation)

Over short to medium horizons the third term dominates, but over very long horizons this term becomes small and the other terms dominate. That's why PE10 only "works" (I'm not arguing it works well enough to market time) over very long horizons.

The first term is correlated with valuation...higher valuation = lower yield and vice versa. Returns will therefore be correlated with valuation unless there is some tendency for the other terms to offset the relationship between valuation and yield.

So, for there to be no correlation between returns and valuation, the other terms must, on average, have some offsetting correlation with valuation....but they have no correlation (growth term) or a weak correlation in the wrong direction (speculative return term).

Going back to my example of paying for the opportunity to draw a payoff from some distribution... Assume the variance in possible payoffs changes over time but the mean does not. In this case, it may be rational to pay more for to draw a payoff at some times than at others. But, the cases where you pay less for a draw will still have higher expected return. For the expected return to be constant at different price levels the mean of the distribution must vary with price....and that would be the same as valuation forecasting growth.
Last edited by camontgo on Wed Apr 09, 2014 11:01 am, edited 1 time in total.
"Essentially, all models are wrong, but some are useful." - George E. P Box
Longtimelurker
Posts: 471
Joined: Fri Dec 13, 2013 7:23 am

Re: BH Greatest Hits: PE10 predicitive power

Post by Longtimelurker »

The good news is that we are only 4 to 5 years away from PE10 dropping dramatically. The cause. `08 & `09 dropping from the data set. I am quite sure that all the PE10 fanatics can explain how something occurring 10 years in the past can suddenly make stocks "cheaper" - perhaps something to due to quantum theory or the Higgs Bozon?
Stay the course. If you can't resist greed, and fear is proven to be 2x as strong, you are doomed as an investor.
User avatar
grap0013
Posts: 1892
Joined: Thu Mar 18, 2010 1:24 pm

Re: BH Greatest Hits: PE10 predicitive power

Post by grap0013 »

larryswedroe wrote:If we combine the two adjustments of 1 for the lower dividend payout and 4 for the FSB 142 change, the current CAPE 10 at 24.9 which looks way above the mean, doesn’t look so overvalued at a now 19.9. In fact, that’s right about in line with it’s average since 1960. Which begs the questions:
a) Are stocks really overvalued or just highly valued (meaning returns are likely to be lower than historical levels, but that there is no reason to expect a major correction due to RTM)?
b) To what mean should the CAPE ratio revert: The 16.5 mean of the past 113 years, or the 19.6 mean since 1960?
These three points — that over time it’s logical to believe that the equity risk premium for U.S. stocks might have fallen, the accounting change regarding write offs, and the lower payout ratios — provide us with plausible explanations that the CAPE 10’s high level is not signaling a massive overvaluation of U.S. stocks
+1. Best post I've seen on the topic.
There are no guarantees, only probabilities.
User avatar
camontgo
Posts: 329
Joined: Tue Oct 27, 2009 1:46 pm
Location: Chicago, IL
Contact:

Re: BH Greatest Hits: PE10 predicitive power

Post by camontgo »

grap0013 wrote:
larryswedroe wrote:If we combine the two adjustments of 1 for the lower dividend payout and 4 for the FSB 142 change, the current CAPE 10 at 24.9 which looks way above the mean, doesn’t look so overvalued at a now 19.9. In fact, that’s right about in line with it’s average since 1960. Which begs the questions:
a) Are stocks really overvalued or just highly valued (meaning returns are likely to be lower than historical levels, but that there is no reason to expect a major correction due to RTM)?
b) To what mean should the CAPE ratio revert: The 16.5 mean of the past 113 years, or the 19.6 mean since 1960?
These three points — that over time it’s logical to believe that the equity risk premium for U.S. stocks might have fallen, the accounting change regarding write offs, and the lower payout ratios — provide us with plausible explanations that the CAPE 10’s high level is not signaling a massive overvaluation of U.S. stocks
+1. Best post I've seen on the topic.
Similar points are made in more detail in this blog post (it has been discussed in a previous thread):

http://philosophicaleconomics.wordpress ... 3/shiller/

It is a great critique which goes into a lot of detail on the adjustments.

On the other hand, this defense of Shiller PE is also worth reading:

http://www.aqr.com/Portals/1/ResearchPa ... Asness.pdf
"Essentially, all models are wrong, but some are useful." - George E. P Box
richard
Posts: 7961
Joined: Tue Feb 20, 2007 2:38 pm
Contact:

Re: BH Greatest Hits: PE10 predicitive power

Post by richard »

camontgo wrote:
richard wrote: One major difference between the two cases is that in the first instance we had a constant underlying distribution while in the later case the underlying distribution could change over time.
Yes, it will change over time.

It could be that the market is more risky at some times than others, and this drives changes in valuation. I don't think that is inconsistent with Cochrane's argument. If the risk is changing, it would mean that market participants shouldn't be overly excited by low or high valuations.

However, it is still true that if there is no correlation between valuation and the average level of growth then there will be a correlation between valuation and average return.

Bogle uses this equation for expected returns:

Return = Yield + Growth + Speculative Return (change in valuation)<snip>
I believe this maps to my interpretation of DeLong in http://www.bogleheads.org/forum/viewtop ... 8#p2023098 Do you disagree with that post?
camontgo wrote:Going back to my example of paying for the opportunity to draw a payoff from some distribution... Assume the variance in possible payoffs changes over time but the mean does not. In this case, it may be rational to pay more for to draw a payoff at some times than at others. But, the cases where you pay less for a draw will still have higher expected return. For the expected return to be constant at different price levels the mean of the distribution must vary with price....and that would be the same as valuation forecasting growth.
Again, given a constant underlying distribution. p/e appears to be somewhat bounded (it can't go below zero and, for example, 100 is not plausible at the moment), so a bounded underlying distribution may not be that far off.
User avatar
Random Musings
Posts: 6770
Joined: Thu Feb 22, 2007 3:24 pm
Location: Pennsylvania

Re: BH Greatest Hits: PE10 predicitive power

Post by Random Musings »

The most recent Hussman article has a chart that illustrates that market capitalization to GDP has the best predictive power of all.

Of course, there are no guarantees that this will work in the future.

RM
I figure the odds be fifty-fifty I just might have something to say. FZ
Jeff Albertson
Posts: 904
Joined: Sat Apr 06, 2013 7:11 pm
Location: Springfield

Re: BH Greatest Hits: PE10 predicitive power

Post by Jeff Albertson »

John Authers recently discussed valuation methods with Andrew Smithers, in a two part interview.
http://video.ft.com/3425406840001/Value ... torschoice
http://video.ft.com/3425374374001/Q-is- ... torschoice
By far the best measure, it turns out, is q. Graphed on a chart, q and hindsight value look almost identical. If you want to know whether hindsight will deem today’s market to have been cheap or expensive, you are best to look at q. Sadly, q now shows the S&P as 70 per cent overvalued.
For those who really do not want to believe this, they should now focus their energies on attempts to prove that the replacement value of assets is a less valid measure than it used to be. That involves proving that you know more than the government statisticians who draw up the data on which q is based.
Hindsight value is an exciting breakthrough. But one thing does not change. None of these measures, not even q, can help time the market. From any given valuation, human nature being what it is, it is always possible for stocks to grow even more wildly cheap or expensive.
User avatar
camontgo
Posts: 329
Joined: Tue Oct 27, 2009 1:46 pm
Location: Chicago, IL
Contact:

Re: BH Greatest Hits: PE10 predicitive power

Post by camontgo »

richard wrote:I believe this maps to my interpretation of DeLong in http://www.bogleheads.org/forum/viewtop ... 8#p2023098 Do you disagree with that post?
Following the link:
richard wrote:He's (DeLong) saying the only possible choices are (1) p/e predicts returns, (2) p/e predicts earnings/dividend growth, (3) the price of risk changes over time, and implies these are mutually exclusive. He picks (1) as an empirical matter, rejects (3) as an empirical matter and therefore rejects (2) as a matter of logic.

Cochrane appears to regard (1) and (2) as the only possible choices and assumes they are mutually exclusive. He rejects (2) as an empirical matter and therefore picks (1). He ignore (3).
I think over the long term, it is between the first 2:

(1) p/e predicts returns
(2) p/e predicts earnings/dividend growth

Number 3 seems like one of several potential reasons for a shift in valuation (i.e. speculative return)....but I think the reason for a shift in expected returns (rational, irrational, some of both) is a secondary question. I think valuation tells us a bit about expected returns, and I think it is reasonable to assume the expected value of the speculative return is zero..(though there may be some weak mean reversion). So, we really want to know if valuation is telling us something about (1) or (2).

I don't think that the two possibilities above are mutually exclusive, but I think there is a lot of historical evidence for 1, and very little for 2. So, if p/e has little to no correlation with subsequent growth, then p/e must have some correlation with subsequent returns. (holding valuation constant, or looking over long time periods where the annualized effect of valuation changes is very small).
"Essentially, all models are wrong, but some are useful." - George E. P Box
richard
Posts: 7961
Joined: Tue Feb 20, 2007 2:38 pm
Contact:

Re: BH Greatest Hits: PE10 predicitive power

Post by richard »

Going back to my exploration of this by-way, there is an implicit assumption that the price of risk, speculative return, etc. remains more or less constant or at least has little effect on return over the long term. This seems reasonable over the long term. There are therefore two main choices and empirical work leads us to return predictability rather than earnings/dividend growth predictability. Nonetheless, there is enough noise in the system (and other issues) that using p/e or its variants as any sort of market timing or tactical allocation tool is not likely to be a good idea.
User avatar
Clearly_Irrational
Posts: 3087
Joined: Thu Oct 13, 2011 3:43 pm

Re: BH Greatest Hits: PE10 predicitive power

Post by Clearly_Irrational »

PE10 is the best predictor we have, but it's still not all that great. It does tend to be more useful towards the extreme readings though. Personally I use it for two things:

1) To let me know when the market is getting far enough out of whack that I should be trying to figure out what's going on

2) As one variable in my crash indicator

I don't consider this unusual since the market more closely resembles a levy flight than a random walk and it has fatter tails than predicted by the normal distribution. These things suggest that we can have extreme events more frequently and more severely than would be predicted by a pure EMH model and it's only prudent to build some protection for that into your portfolio. Obviously not everyone agrees that information is reliable or actionable.
random_walker_77
Posts: 2212
Joined: Tue May 21, 2013 8:49 pm

Re: BH Greatest Hits: PE10 predicitive power

Post by random_walker_77 »

grap0013 wrote:
larryswedroe wrote:If we combine the two adjustments of 1 for the lower dividend payout and 4 for the FSB 142 change, the current CAPE 10 at 24.9 which looks way above the mean, doesn’t look so overvalued at a now 19.9. In fact, that’s right about in line with it’s average since 1960. Which begs the questions:
a) Are stocks really overvalued or just highly valued (meaning returns are likely to be lower than historical levels, but that there is no reason to expect a major correction due to RTM)?
b) To what mean should the CAPE ratio revert: The 16.5 mean of the past 113 years, or the 19.6 mean since 1960?
These three points — that over time it’s logical to believe that the equity risk premium for U.S. stocks might have fallen, the accounting change regarding write offs, and the lower payout ratios — provide us with plausible explanations that the CAPE 10’s high level is not signaling a massive overvaluation of U.S. stocks
+1. Best post I've seen on the topic.
agreed.

Somewhat relevant, I was wondering how cape10 compares against another arbitrary timeframe such as cape15 or cape5. Just the other day, I was curious and looked it up. The raw data is all in a spreadsheet at Shiller's website: http://www.econ.yale.edu/~shiller/data.htm

Adding a few columns for cape15/cape5/cape4 showed that the exact timescale doesn't change things all that much. cape5 and cape15 isn't all that different from cape10, especially when you consider the minor differences in their long term averages. Of course, they react to different extents on events like 2008, but there's a lot of correlation.

Adding cape4 to eliminate '08-'09 and it doesn't look so bad compared to the cape4 long term average, especially if you buy Swedroe's argument for adjusting for lower dividend payout and FSB 142.

With the data, you can average price over the last several years of earnings.

Code: Select all

			CAPE10	CAPE15	CAPE5	CAPE4
Long term 	16.5	17.3	15.8	15.6
average		

2012.03		22.0	22.9	22.3	24.4
2013.03		22.4	24.4	24.3	22.4
2014.03		25.4	28.2	25.6	21.7
edit: my version of the spreadsheet adding cape15/cape5/cape4 is here: https://drive.google.com/file/d/0BzAj8r ... sp=sharing
Topic Author
grayfox
Posts: 5569
Joined: Sat Sep 15, 2007 4:30 am

Re: BH Greatest Hits: PE10 predicitive power

Post by grayfox »

camontgo wrote:
Similar points are made in more detail in this blog post (it has been discussed in a previous thread):

http://philosophicaleconomics.wordpress ... 3/shiller/

It is a great critique which goes into a lot of detail on the adjustments.
I read that article. I didn't understand all the accounting rule stuff and adjustments, but generally he is saying what I was saying back in 2010

1. Current higher valuations imply lower expected returns
2. No law that says valuations must return to some historical mean value.
There is no question that the current stock market is more expensive than the averages of certain past eras–the 1910s, 1930s, 1940s, 1970s, 1980s, etc. Looking forward, long-term equity returns will obviously be lower than they were in those eras. But the market is not as expensive as the Shiller CAPE suggests. Moreover, there’s no reason to think that the valuations of those eras–distorted by world wars (1914-1918, 1939-1945, 1950-1953), gross economic mismanagement (1929-1938), and painfully high inflation and interest rates (1970-1982)–were somehow more “appropriate” than current valuations. The valuations in those eras were “appropriate” to the circumstances of those eras; we live in different circumstances.
Discount rates are just one of the state variables that describes the state of the world.
In 2014, discount rates for bonds and stocks are low. 20-year real rate = 1%. P/E10 = 25. They is what they is.
In 2024 P/E10 may be 25 or 30 or 15. We don't have the ability to accurately predict the future state of the world.
Topic Author
grayfox
Posts: 5569
Joined: Sat Sep 15, 2007 4:30 am

Re: BH Greatest Hits: PE10 predicitive power

Post by grayfox »

random_walker_77 wrote: Somewhat relevant, I was wondering how cape10 compares against another arbitrary timeframe such as cape15 or cape5. Just the other day, I was curious and looked it up. The raw data is all in a spreadsheet at Shiller's website: http://www.econ.yale.edu/~shiller/data.htm

Adding a few columns for cape15/cape5/cape4 showed that the exact timescale doesn't change things all that much. cape5 and cape15 isn't all that different from cape10, especially when you consider the minor differences in their long term averages. Of course, they react to different extents on events like 2008, but there's a lot of correlation.

Adding cape4 to eliminate '08-'09 and it doesn't look so bad compared to the cape4 long term average, especially if you buy Swedroe's argument for adjusting for lower dividend payout and FSB 142.

With the data, you can average price over the last several years of earnings.

Code: Select all

			CAPE10	CAPE15	CAPE5	CAPE4
Long term 	16.5	17.3	15.8	15.6
average		

2012.03		22.0	22.9	22.3	24.4
2013.03		22.4	24.4	24.3	22.4
2014.03		25.4	28.2	25.6	21.7
edit: my version of the spreadsheet adding cape15/cape5/cape4 is here: https://drive.google.com/file/d/0BzAj8r ... sp=sharing
All of these valuation measures = CurrentPrice divided by "something".
It answers the question how many times the value of "something" do you have to pay?

That something can be a moving average of earnings like E4, E5, E10, E15, E20, E30.
It could be trend earnings. It could be peak earnings. I could be dividends. It could be book value.
Hopefully, that "something" captures the value of what you are buying.

So the question is, which "something" best captures the value you are getting?
random_walker_77
Posts: 2212
Joined: Tue May 21, 2013 8:49 pm

Re: BH Greatest Hits: PE10 predicitive power

Post by random_walker_77 »

grayfox wrote:
All of these valuation measures = CurrentPrice divided by "something".
It answers the question how many times the value of "something" do you have to pay?

That something can be a moving average of earnings like E4, E5, E10, E15, E20, E30.
It could be trend earnings. It could be peak earnings. I could be dividends. It could be book value.
Hopefully, that "something" captures the value of what you are buying.

So the question is, which "something" best captures the value you are getting?
Exactly, and my understanding is that the standard CAPE uses 10 years b/c p/e over 1 year is inferior. And it's deemed inferior because earnings over 1 year are too "noisy"; there's too much variation from year to year. That p/e in different parts of the business cycle naturally vary. That 1 year earnings are subject to manipulation. By using 10 years, you average over the good and the bad parts of a business cycle (supposedly ~6 years).

I had in my head that the events from 2000 onwards have seemed exceptional, so I was surprised to see that E5/E15 versus E10 doesn't change things all that much. I haven't looked at E4 long enough to figure out if I believe it's nearly as good as E10 (this is all academic to me anyways -- "stay the course")
richard
Posts: 7961
Joined: Tue Feb 20, 2007 2:38 pm
Contact:

Re: BH Greatest Hits: PE10 predicitive power

Post by richard »

random_walker_77 wrote:<snip>Exactly, and my understanding is that the standard CAPE uses 10 years b/c p/e over 1 year is inferior. And it's deemed inferior because earnings over 1 year are too "noisy"; there's too much variation from year to year. That p/e in different parts of the business cycle naturally vary. That 1 year earnings are subject to manipulation. By using 10 years, you average over the good and the bad parts of a business cycle (supposedly ~6 years).

I had in my head that the events from 2000 onwards have seemed exceptional, so I was surprised to see that E5/E15 versus E10 doesn't change things all that much. I haven't looked at E4 long enough to figure out if I believe it's nearly as good as E10 (this is all academic to me anyways -- "stay the course")
The problem with p/e is one year is noisy. The problem with CAPE is it includes years which are no longer relevant. Vanguard found the two methods to have equal predictive power.

As I mentioned up-thread: "We confirm that valuation metrics such as price/earnings ratios, or P/Es, have had an inverse or mean-reverting relationship with future stock market returns, although it has only been meaningful at long horizons and, even then, P/E ratios have “explained” only about 40% of the time variation in net-of-inflation returns. Our results are similar whether or not trailing earnings are smoothed or cyclically adjusted (as is done in Robert Shiller’s popular P/E10 ratio)."

The full paper is at https://personal.vanguard.com/pdf/s338.pdf
User avatar
Clearly_Irrational
Posts: 3087
Joined: Thu Oct 13, 2011 3:43 pm

Re: BH Greatest Hits: PE10 predicitive power

Post by Clearly_Irrational »

richard wrote:The problem with CAPE is it includes years which are no longer relevant. Vanguard found the two methods to have equal predictive power.
I wouldn't call R^2 of 0.43 vs R^2 of 0.38 "equal". A better comparison would be to Price to Peak Earnings or Tobin's Q.
Rodc
Posts: 13601
Joined: Tue Jun 26, 2007 9:46 am

Re: BH Greatest Hits: PE10 predicitive power

Post by Rodc »

Clearly_Irrational wrote:
richard wrote:The problem with CAPE is it includes years which are no longer relevant. Vanguard found the two methods to have equal predictive power.
I wouldn't call R^2 of 0.43 vs R^2 of 0.38 "equal". A better comparison would be to Price to Peak Earnings or Tobin's Q.
While I think the comparison with Price to Peak Earnings or Tobin's Q would be useful and interesting, given the level of noise in the system I certainly consider R^2 of 0.43 vs R^2 of 0.38 "equal" within the bounds of the noise. Change time period and not only will you get a different pair of answers the direction could easily flip. The reality is they are both in the ballpark of "About 0.4" and the difference is likely just noise. Frankly I would not be surprised if an analysis of the confidence interval gave error bars greater than +- 0.1.
We live a world with knowledge of the future markets has less than one significant figure. And people will still and always demand answers to three significant digits.
User avatar
Clearly_Irrational
Posts: 3087
Joined: Thu Oct 13, 2011 3:43 pm

Re: BH Greatest Hits: PE10 predicitive power

Post by Clearly_Irrational »

Rodc wrote:While I think the comparison with Price to Peak Earnings or Tobin's Q would be useful and interesting, given the level of noise in the system I certainly consider R^2 of 0.43 vs R^2 of 0.38 "equal" within the bounds of the noise. Change time period and not only will you get a different pair of answers the direction could easily flip. The reality is they are both in the ballpark of "About 0.4" and the difference is likely just noise. Frankly I would not be surprised if an analysis of the confidence interval gave error bars greater than +- 0.1.
I understand what you're saying but I suppose it depends on what you want to use the results for. If your goal is to "predict" with some level of accuracy the returns over the next decade then PE1 is incredibly noisy and it's not something I'd feel comfortable using. For example in 2004, a year that wasn't particularly tumultuous, PE1 predictions for the next 10 years varied by 0.85% while PE10 predictions only varied by 0.35%. A half percentage point variance over ten years is a big deal in my opinion.

More importantly, when using the indicator as a danger signal PE1 tends to be late to the party. During the tech crisis PE10 peaked Dec 1999, the S&P 500 peaked Aug 2000 while PE1 didn't peak until Mar 2002. During the housing crisis PE10 peaked May 2007, the S&P 500 peaked Oct 2007 while PE1 didn't peak until May 2009. So when you need it most PE1 works the least, that makes it a bad indicator for this purpose.
Longtimelurker
Posts: 471
Joined: Fri Dec 13, 2013 7:23 am

Re: BH Greatest Hits: PE10 predicitive power

Post by Longtimelurker »

Clearly_Irrational wrote:If your goal is to "predict" with some level of accuracy the returns over the next decade then
... you are living in a fantasy land where you can imagine the unimaginable, predict the unpredictable, and secure yourself a false sense of security an knowledge that at best is irrational, at worst leads to behavioral investing issues.

Tune out ALL of the noise. Stop looking for patterns in the snow. Backtesting says nothing about the future. Nothing.
Stay the course. If you can't resist greed, and fear is proven to be 2x as strong, you are doomed as an investor.
User avatar
Clearly_Irrational
Posts: 3087
Joined: Thu Oct 13, 2011 3:43 pm

Re: BH Greatest Hits: PE10 predicitive power

Post by Clearly_Irrational »

Longtimelurker wrote:
Clearly_Irrational wrote:If your goal is to "predict" with some level of accuracy the returns over the next decade then
... you are living in a fantasy land where you can imagine the unimaginable, predict the unpredictable, and secure yourself a false sense of security an knowledge that at best is irrational, at worst leads to behavioral investing issues.

Tune out ALL of the noise. Stop looking for patterns in the snow. Backtesting says nothing about the future. Nothing.
If it says nothing about the future then we have to assume that stock returns are unknowable, for example there would be an equal chance of all stocks going to zero tomorrow as there would be of everything going to a million. That's both silly and wrong. If it were true I wouldn't get anywhere near the stock market as the risk/reward tradeoff would be terrible.

The academic research has pretty clearly demonstrated that the market is essentially a stochastic process (probabilistic) with a fairly understandable distribution curve. While it's exceedingly difficult to predict any one data point, we nevertheless have quite a bit of information about how things will turn out in general.
Topic Author
grayfox
Posts: 5569
Joined: Sat Sep 15, 2007 4:30 am

Re: BH Greatest Hits: PE10 predicitive power

Post by grayfox »

richard wrote: The problem with p/e is one year is noisy. The problem with CAPE is it includes years which are no longer relevant. Vanguard found the two methods to have equal predictive power.

As I mentioned up-thread: "We confirm that valuation metrics such as price/earnings ratios, or P/Es, have had an inverse or mean-reverting relationship with future stock market returns, although it has only been meaningful at long horizons and, even then, P/E ratios have “explained” only about 40% of the time variation in net-of-inflation returns. Our results are similar whether or not trailing earnings are smoothed or cyclically adjusted (as is done in Robert Shiller’s popular P/E10 ratio)."

The full paper is at https://personal.vanguard.com/pdf/s338.pdf
Vanguard says P/E10 explained 40% of 10-year return. That leaves 60% unexplained. But I also recall others have found that that R^2 for 20-year return was higher, about 0.60. I don't think that Vanguard reported anything other than 10-year returns.

I can't remember if I had ever run the numbers before. So I decided to look at some regressions myself. I found what I would call strong correlation between valuation and future return. However, the question of overlapping data makes one wonder if the correlation is real or not. If I use monthly data for 20-year returns, there might be as many as 1477 observations. But are they independent observations? If not, why not? Does lack of independence invalidate the results?

BTW, this is not a unique issue. I can google papers that discuss it and report that it appears to be standard practice in finance to use overlapping data sets and there are debates about whether or not is is valid, and what methods are better when using overlapping data.

That's more or less what prompted me to google P/E10 on bogleheads.org that brought up the referenced thread.
User avatar
fredflinstone
Posts: 2829
Joined: Mon Mar 29, 2010 7:35 am
Location: Bedrock

Re: BH Greatest Hits: PE10 predicitive power

Post by fredflinstone »

Looking over the Shiller PE10 chart at http://www.multpl.com/shiller-pe/, I see two time periods in which PE10 increased 6-fold. One was the period between 1920 and 1929. The other was the period between 1982 and 2000. Both were immediately followed by severe bear markets.

In Japan, there was a more than four-fold increase in PE10 just before the lengthy bear market that began in 1989. http://www.vectorgrader.com/indicators/ ... e-earnings

In short, a dramatic increase in PE10 appears to be a good warning sign of an impending bear market.
Rodc
Posts: 13601
Joined: Tue Jun 26, 2007 9:46 am

Re: BH Greatest Hits: PE10 predicitive power

Post by Rodc »

grayfox wrote:
richard wrote: The problem with p/e is one year is noisy. The problem with CAPE is it includes years which are no longer relevant. Vanguard found the two methods to have equal predictive power.

As I mentioned up-thread: "We confirm that valuation metrics such as price/earnings ratios, or P/Es, have had an inverse or mean-reverting relationship with future stock market returns, although it has only been meaningful at long horizons and, even then, P/E ratios have “explained” only about 40% of the time variation in net-of-inflation returns. Our results are similar whether or not trailing earnings are smoothed or cyclically adjusted (as is done in Robert Shiller’s popular P/E10 ratio)."

The full paper is at https://personal.vanguard.com/pdf/s338.pdf
Vanguard says P/E10 explained 40% of 10-year return. That leaves 60% unexplained. But I also recall others have found that that R^2 for 20-year return was higher, about 0.60. I don't think that Vanguard reported anything other than 10-year returns.

I can't remember if I had ever run the numbers before. So I decided to look at some regressions myself. I found what I would call strong correlation between valuation and future return. However, the question of overlapping data makes one wonder if the correlation is real or not. If I use monthly data for 20-year returns, there might be as many as 1477 observations. But are they independent observations? If not, why not? Does lack of independence invalidate the results?

BTW, this is not a unique issue. I can google papers that discuss it and report that it appears to be standard practice in finance to use overlapping data sets and there are debates about whether or not is is valid, and what methods are better when using overlapping data.

That's more or less what prompted me to google P/E10 on bogleheads.org that brought up the referenced thread.
FWIW:
http://home.comcast.net/~rodec/finance/ ... eriods.pdf
We live a world with knowledge of the future markets has less than one significant figure. And people will still and always demand answers to three significant digits.
letsgobobby
Posts: 12073
Joined: Fri Sep 18, 2009 1:10 am

Re: BH Greatest Hits: PE10 predicitive power

Post by letsgobobby »

fredflinstone wrote:Looking over the Shiller PE10 chart at http://www.multpl.com/shiller-pe/, I see two time periods in which PE10 increased 6-fold. One was the period between 1920 and 1929. The other was the period between 1982 and 2000. Both were immediately followed by severe bear markets.

In Japan, there was a more than four-fold increase in PE10 just before the lengthy bear market that began in 1989. http://www.vectorgrader.com/indicators/ ... e-earnings

In short, a dramatic increase in PE10 appears to be a good warning sign of an impending bear market.
Or the flip side, a dramatic decrease in PE10 appears to be a good time to buy a lot of stocks.

To my mind, the recent accounting changes in E10 are important considerations. I am a firm believer that extreme valuations in PE10 mean something and are actionable (ie, PE10 < 12 and PE10 > 25), but if today's PE10 25 is really yesterday's PE10 20 then that gives me pause. For that reason when PE10 went over 25 last fall, rather than cutting stocks I shifted from US small cap value to international, which had/has much lower PE10 values. If S&P500 PE10 exceeds 30 I will move more decisively to reduce US stock exposure.
User avatar
Random Musings
Posts: 6770
Joined: Thu Feb 22, 2007 3:24 pm
Location: Pennsylvania

Re: BH Greatest Hits: PE10 predicitive power

Post by Random Musings »

Of course, even if today's Oldsmobiles 25 is yesterdays 20, there still should be times in history when it moves below the average for a reasonable period of time. Historically, that is when stocks are truly despised and these PE 10 troughs last for a few years.

If we use Larry's metrics, the only time we hit a trough, and that being relatively quickly, was a brief moment from late 2008-early 2009.

RM
I figure the odds be fifty-fifty I just might have something to say. FZ
Topic Author
grayfox
Posts: 5569
Joined: Sat Sep 15, 2007 4:30 am

Re: BH Greatest Hits: PE10 predictive power

Post by grayfox »

So we have 100+ years of stock market return data. Now I know that some people question the veracity or relevance of some of the earlier data. But assume for the sake of argument that it is all good and relevant data. But if we are interested in 20-, 30-, 40-, 50-years returns, 100 years still not enough data to show anything with a high degree of statistical significance.

My answer to that is, "So what?"

Who needs historical data?
Suppose there was NO historical stock market data. Suppose there was NO public stock market where company shares traded, and NO historical prices.

You could use logic and simple arithmetic to forecast the return on an investment in a company.

Let's say a company holds some assets that generate a return.
Say the management has been able to generate a 5% return on those assets, i.e. ROA = 5%.
The company bought some of those assets with money from the owners (equity) and has also borrowed money (debt) to buy some of the assets.
The equation is Assets = Liabilities + Equity. Equity is another name for Book Value.
Suppose the management has been able to achieve a 10% return on equity, i.e. ROE = 10%.

Now you are going to buy the whole company.
If you pay 1x the Book Value, i.e. P/B = 1.0, and the company manages to continue the 10% ROE, what will your return be?
Clearly 10%. You are basically buying the equity at 1x its value and it generates a 10% return.

What if you pay 2x Book Value, i.e. P/B=2.0, and the company manages to continue the 10% ROE.
Will your return be more or less than the 10% return at P/B=1.0?
Obviously less. You are getting the same thing, but at twice the price.

What if you managed to buy the company at a bargain price, P/B=0.5, and the company manages to continue the 10% ROE.
Will your return be more or less than the return at P/B=1.0?
Obviously more.

It should be obvious that the more you pay, the lower your return.
Highly valuations -> lower return expectations.
So that is the null hypothesis.

(Now back to the real world where there is a stock market and historical data.)

Then someone comes along and says:
"But the market is efficient. The market puts a higher valuation on the company because it is expected to have a higher ROE. So the expected return doesn't go down with higher P/B."

I say prove it. Prove that when market P/B or P/E goes up it is forecasting higher growth.
Using historical data, show that the null hypothesis, H0: higher valuations -> lower return expectations, should be rejected.
The onus is on them because they are trying to disprove something obvious that is based on simple and sound logic.

Wait, up at the top we said that there was not enough data to show anything about long-term returns with much statistical significance.
So we can't reject the null hypothesis.

There is really no evidence that valuations do not forecast returns. In fact, whatever little evidence we have shows it does.
richard
Posts: 7961
Joined: Tue Feb 20, 2007 2:38 pm
Contact:

Re: BH Greatest Hits: PE10 predictive power

Post by richard »

grayfox wrote:<snip>It should be obvious that the more you pay, the lower your return.
Highly valuations -> lower return expectations.
So that is the null hypothesis.
That's not how statistics works under the standard definition of null hypothesis.

"In statistical inference of observed data of a scientific experiment, the null hypothesis refers to a general or default position: that there is no relationship between two measured phenomena" http://en.wikipedia.org/wiki/Null_hypothesis

Therefore, the null hypothesis is that there is no relationation between what you pay and your return. That's assuming we're in the realm of scientific experiment. The better argument seems to be that finance is not subject to scientific experiment.

If you want to go by logic, it's not at all obvious that the more you pay for past earnings, the lower your return will be in the future. As Warren Buffet famously said, if past history was all there was to the game, the richest people would be librarians. Another thought to consider is that you can't beat the market with information known to all. Or simply that we don't see a lot of people beating the market, even though p/e is one of the most widely known investment metrics.
grayfox wrote:(Now back to the real world where there is a stock market and historical data.)
Back in the real world we continue not to have enough independent data points.
Topic Author
grayfox
Posts: 5569
Joined: Sat Sep 15, 2007 4:30 am

Re: BH Greatest Hits: PE10 predictive power

Post by grayfox »

richard wrote: That's not how statistics works under the standard definition of null hypothesis.

"In statistical inference of observed data of a scientific experiment, the null hypothesis refers to a general or default position: that there is no relationship between two measured phenomena" http://en.wikipedia.org/wiki/Null_hypothesis

Therefore, the null hypothesis is that there is no relationation between what you pay and your return. That's assuming we're in the realm of scientific experiment. The better argument seems to be that finance is not subject to scientific experiment.
I was not aware of that point about statistical inference regarding the null hypothesis, but it makes sense. I might have to look in a statistics textbook to see if all the inference problems set up the null hypothesis that way.

So how about if I set up the null hypothesis H0: valuations /-> future growth. Then see if the data says I should reject H0. If I cannot reject H0, then we stick with the idea that valuation does not forecast growth. I think that is basically what Cochrane says in The Dog That Didn't Bark. There is a lack of evidence that high valuations lead to high growth.
Post Reply