1Year Berkshire Price Prediction - Inflation +17% From 10/22

I was fascinated by it when mungofitch posted, and it’s great to see your follow-up and generous sharing of results.

Technical discussion put aside for a moment, a general point that occurs to me is that the result seems to tie into ideas of ‘mechanical investing’, which has been struggling of late. The result demonstrates an empirical relation between being ‘cheap’ and future returns, at least for one company i.e. BRK. Many mechanical, value investing, approaches look to buy ‘good companies that are cheap’. Here, the ‘good’ part is pre-selected, it’s Berkshire after all (admittedly, what ‘good’ precisely means is not clear but clearly Buffett is good). What’s been shown for this pre-selected ‘good’ company is that buying (very) cheaply tends to have nice returns, while buying (very) expensively tends to have poor returns.

Anyway, one might conjecture, given the result, is that the hard part of value investing, i.e. buying ‘good’ companies on the ‘cheap’, isn’t so much defining ‘cheap’, but defining ‘good’.

10 Likes

100% agree. That’s why I tend to stick to the simple companies doing simple / understandable things for customers with an extremely heavy slant toward companies making or selling physical products a person could touch and feel.

I’ll forever miss out on the sky high returns asset light companies like google and Saul stocks but I’m a simple person needing simple investments.

13 Likes

I don’t have one offhand, but mungofitch mentioned it repeatedly on the old board. He apparently did regression of returns vs current book value and compared to regression of returns vs peak book value and found that peak book value was the better predictor.

As I remember, he explained this result as likely coming from the fact that current book deviates significantly from peak book primarily during heavy bear markets, when the stock portfolio takes a severe beating. But, those are the same times that Buffett has been able to capitalize on good deals available in those times of distress. I.e. the high interest loan made to Occidental during the Great Recession that came with oodles of warrants.

Also it presumes that the vast majority of Berkshire’s stocks and businesses are built for long term performance and won’t be significantly impaired long term by even a big economic dislocation.

So, when the stormy weather clears, Berkshire ends up in better position than when the bear struck and book value recovers nicely to its long term trend. If you were relying on current book value instead of peak book value, you might not have recognized a great buying opportunity.

13 Likes

Your summary matches my understanding of Jim’s findings as relayed to this board. He liked to use the word “transitory” to describe the drops in Berk’s book value, as though it was best to ignore them.

4 Likes

Thanks, I’ll have to look it up. It rings a bell now, but I think I tended to view it as an embellishment whereas I was still blown away by the main point.

There are of course paid data services that have historical data for company fundamentals. I may look into that, at least for a free trial period, but I did find the following way to pull up some free data.

google a ticker like this (note the spelling of the first word):
“discoverci BRK price book”
and it will (usually) pull up a link that has a chart for the ticker. 'Right click 'on the chart and you can download csv data, in this case, price to book value for BRK.

You can play around with pulling down other fundamental data in the same way by adjusting the search phrase. Some of the charts/data looks odd, perhaps they didn’t adjust for splits in ratios or something, you may well get what you pay for.

The reason the above search works is that it’s finding data from: www.discoverci.com
I know nothing about this site, so be careful. It also seems to have some interesting titles for articles in its blog link, but I haven’t read more than the titles yet.

If you find problems, give a shout?

3 Likes

I mean for other tickers e.g. GOOG.
A hunting ground for other ‘good’ companies (other than BRK) may be some of the companies that BRK holds … or not. Could be fun to play a bit.

4 Likes

I’ve been using wwwdotdatahelperdotcom and also the search engine on the new board to unearth some of Jim’s old posts on PriceToBookValue (PBV) for BRK, in order to record them on the new boards for posterity, or however long the new boards last. It wasn’t an exhaustive search, if others feel inclined to go gold mining then have at it!

I had an email exchange with Mark Wilcox who did datahelper, and if I understood correctly he has an archive of the old board and when the new boards abandon their history he’ll try to help out. But he commented that he doesn’t use the boards any more. Given that everyone has time constraints, it seemed worthwhile trying to preserve some relevant posts now i.e. below.

I wasn’t always able to capture dates in a copy/paste, but never mind, the following text was useful (at least to me).

#--------------------------------------------------------------------
Author: mungofitch Date: 08/02/2020 11:50 AM Msg: BH-255100
Subject: Observations Recs: 40
No, not observations in the sense of insights, just some numerical observations in the sense of a science experiment.

I was doing some fiddling about and thought others might find these numbers interesting.

Given the ratio of market price to “peak to date” book per share on any given date in the past, look at the average market price 1-2 years later.
So, it’s a look ahead period of average 1.5 years.
The stock prices and resulting returns are adjusted for inflation and annualized.
Starting dates 2002-01-02 through 2018-07-31

      Initial ratio of price to peak book   Avg 1.5yr return

1% of time P/peakB from 0.928 to 1.076 30.3%/year
1% of time P/peakB from 1.076 to 1.112 24.2%
1% of time P/peakB from 1.112 to 1.132 24.9% We are in this bucket right now
1% of time P/peakB from 1.132 to 1.144 24.0%
1% of time P/peakB from 1.144 to 1.154 23.5%
5% of time P/peakB from 1.154 to 1.194 21.5%
5% of time P/peakB from 1.194 to 1.248 17.4%
5% of time P/peakB from 1.248 to 1.304 11.6%
5% of time P/peakB from 1.304 to 1.339 7.7%
5% of time P/peakB from 1.339 to 1.361 12.2%
5% of time P/peakB from 1.361 to 1.380 10.9%
5% of time P/peakB from 1.380 to 1.400 7.5%
5% of time P/peakB from 1.400 to 1.432 8.3%
5% of time P/peakB from 1.432 to 1.456 9.0%
5% of time P/peakB from 1.456 to 1.476 8.6%
5% of time P/peakB from 1.476 to 1.499 9.8%
5% of time P/peakB from 1.499 to 1.525 8.9%
5% of time P/peakB from 1.525 to 1.551 3.6%
5% of time P/peakB from 1.551 to 1.604 0.8%
5% of time P/peakB from 1.604 to 1.646 1.9%
5% of time P/peakB from 1.646 to 1.689 0.5%
5% of time P/peakB from 1.689 to 1.741 -0.9%
5% of time P/peakB from 1.741 to 1.805 -1.5%
1% of time P/peakB from 1.805 to 1.818 2.3%
1% of time P/peakB from 1.818 to 1.830 -2.8%
1% of time P/peakB from 1.830 to 1.846 -0.3%
1% of time P/peakB from 1.846 to 1.882 -6.9%
1% of time P/peakB from 1.882 to 2.032 -8.0%
I used the “peak to date” book value for the simple reason that dips in book per share have always
been transient in the past and using the peak-to-date value gives a stronger predictor of forward returns.

A lot of the good returns from purchases come from mean reversion of P/B values.
The interesting things is that the mean reversion is not complete in the 1-2 year time frame.
On average the P/B ratio has moved towards the long run average in that time frame, but not nearly all the way.
Here is a table showing the typical price-to-peak-book ratio from the same buckets as the table above, and the average P/B in the ending period 1-2 years later.
I added a column showing the change in book value ratio, and a column for a formula fit to the change.

                Median        Average    Observed    Model
             Starting P/B   Ending P/B    Average   Fit of
              this bucket                 Change    Change

1% of time 1.044 1.334 +0.290 +0.321
1% of time 1.106 1.328 +0.222 +0.233
1% of time 1.130 1.351 +0.222 +0.203 We are in this bucket right now
1% of time 1.142 1.355 +0.213 +0.188
1% of time 1.152 1.355 +0.202 +0.177
5% of time 1.184 1.347 +0.163 +0.144
5% of time 1.224 1.325 +0.102 +0.107
5% of time 1.292 1.314 +0.022 +0.054
5% of time 1.325 1.312 -0.013 +0.032
5% of time 1.352 1.387 +0.035 +0.016
5% of time 1.372 1.396 +0.024 +0.005
5% of time 1.391 1.364 -0.027 -0.005
5% of time 1.413 1.384 -0.029 -0.016
5% of time 1.446 1.422 -0.024 -0.032
5% of time 1.466 1.447 -0.019 -0.041
5% of time 1.489 1.492 +0.004 -0.051
5% of time 1.514 1.489 -0.025 -0.062
5% of time 1.539 1.435 -0.105 -0.073
5% of time 1.574 1.435 -0.139 -0.089
5% of time 1.629 1.507 -0.122 -0.114
5% of time 1.663 1.507 -0.156 -0.132
5% of time 1.717 1.501 -0.217 -0.164
5% of time 1.786 1.560 -0.226 -0.212
1% of time 1.814 1.678 -0.137 -0.236
1% of time 1.828 1.603 -0.225 -0.248
1% of time 1.842 1.640 -0.202 -0.261
1% of time 1.863 1.538 -0.325 -0.282
1% of time 1.938 1.525 -0.413 -0.369

These aren’t predictions for what will happen in future, just observations of what has happened on average in the last 16.6 years.
The future will probably show the same general effect, but with different specific numbers.

Either way, there seems to be no particular reason to expect good returns within a couple of years if you’re adding shares at higher than about 1.55 times book.
And probably the 1-2 year outlook is pretty good starting at under 1.35 times peak book.
Those two cutoffs would split history in two three sections, with these historical results:

P/B from 0.97 to 1.35 inflation+ 16.7%/year (25% of the time 2002-2018)
P/B from 1.35 to 1.55 inflation+ 9.0% (45% of the time 2002-2018)
P/B from 1.55 to 2.03 inflation -0.4% (31% of the time 2002-2018)

Jim

#--------------------------------------------------------------------------------

Market valuation as a predictor
mungofitch
May 29
I was dusting off an old market prediction model I posted about ten years ago.
https://discussion.fool.com/cheap-stocks-at-a-predictor-or-filter-29492673.aspx

It’s pretty basic: How many of the VL stocks are cheap, i.e., P/E under 10?

Rather surprisingly it’s not a bad predictor of one-year-forward returns.

FWIW, 18.62% of VL stocks are notionally “cheap” on that definition at the moment.
That’s pretty high#somewhere around the 82nd percentile of history.
That is historically consistent with average one year forward returns of around 18.5% to 21.5% for an equally weighted VL stock portfolio, depending on the data smoothing you use.
(emphasis on “average”#an average can hide a lot of variation)

Normally the market valuation level is a terrible predictor of forward returns on time frames under a few years.
(with the exception that very cheap moments tend to lead to good results fairly quickly)
This metric seems to avoid that trap, perhaps specifically because the earnings figures being used are NOT cyclically adjusted.
It captures the apparent current earning power, rather than sustainable earning power: apparent earning power might matter more for price movements.

Incidentally, the predictor works with roughly similar variation and similar numbers among the S&P 500 set.
The percentage of “cheap” S&P 500 stocks right now is only 11.6% measured the same way.
So, to the extent that any of this reasoning holds water, the speculation would be that valuation
levels and forward one year prospects are worse among the S&P 500 set than among the rest of the VL set right now.
So, what would be the lesson? Go small or stay home?
Jim

#--------------------------------------------------------------------------------------------------
232427 of 284296 GoSubject: Cheap stocks at a predictor or filterDate: 8/21/2011 11:21 AM
P/E’s aren’t the very best predictor of individual stock performance or
of the broad market direction, because earnings vary wildly with the
business cycle. On the other hand, they aren’t totally useless.

Right now, earnings are well above trend, meaning stocks are expensive
relative to what they can sustain, but on the other hand stocks are
quite cheap as a function of current earnings which is better than nothin’.

So, I thought I’d look at a simple metric. How many of the Value Line
stocks have current earnings yields over 10%? (P/E under 10x).

Courtesy of the GTR1 backtester, we can see how the market has behaved
based on where this percentage likes. In this case, for “the market”
I took an equally weighted portfolio of all the VL stocks.
I looked at the fraction of stocks trading at a P/E of 10 or less
on any given date, bucketed those percentages, and looked at the
average one-year-forward total returns based on the starting percentage.

Starting percentile % of VL stocks Return
From 0 to 0.02 (0.012 to 0.021) -4.4%
From 0.02 to 0.05 (0.021 to 0.027) -1.7%
From 0.05 to 0.1 (0.027 to 0.036) 6.2%
From 0.1 to 0.2 (0.036 to 0.050) 9.9%
From 0.2 to 0.3 (0.050 to 0.062) 11.7%
From 0.3 to 0.4 (0.062 to 0.073) 9.4%
From 0.4 to 0.5 (0.073 to 0.088) 10.4%
From 0.5 to 0.6 (0.088 to 0.103) 14.8%
From 0.6 to 0.7 (0.103 to 0.121) 14.9%
From 0.7 to 0.8 (0.121 to 0.175) 23.2%
From 0.8 to 0.9 (0.175 to 0.239) 6.5%
From 0.9 to 0.95 (0.239 to 0.289) 18.9%
From 0.95 to 0.98 (0.289 to 0.344) 39.9%
From 0.98 to 1 (0.344 to 0.501) 57.5%

As you can see from the table, there is a clear correlation. If lots
of the stocks are trading at a P/E of 10 or less, the next year is
on average a whole lot better than average for market returns.

Today, there are 396 of 1700 stocks at least that cheap on current earnings.
That’s 23.3%, which is percentile .892 in the table above.
Top quartile in the table above suggests an average forward return of around 20.2% in the next year.

Again, this apparent undervaluation is probably partly a function of
today’s unsustainable earnings levels. But that has happened lots of times
in the past, too, so to a certain extent it’s baked into that projected return.

The trick would be to figure out which firms that are that cheap now on current
earnings will not have any meaningful slide in earnings with the next slowdown.
(not saying we’re about to see one, but there is always one coming).
Those firms should be very good buys right now, because the cheapness
is real, not just a cyclical artefact.

Anybody have a “cyclicality of earnings” metric for all the stocks?
Unfortunately VL’s earnings predictability rank is a short-run thing,
standard deviation of quarterly earnings over 8 years, not a cyclical metric.
But, starting with that, I poked around at a few Value Line reports.
A few that popped out at me were Medtronic, Torchmark, Constellation Brands,
Cracker Barrel, Entergy, Microsoft, perhaps less reliably (for varying
reasons) Best Buy, State Street, and my current fave Intel.
Exxon is a standout, cyclical but on a different kind of cycle.

The process I used is pretty simple: start with over 10% earnings
yield, sort by earnings predictability, pull up the stock report for each,
and look for (a) earnings that haven’t gone [much] down year-to-year in
the last two recessions, and (b) aren’t just flat either.

To cushion the fall a bit for the incorrect picks, I note that 58 of
the 396 firms at a P/E under 10 also have dividend yields of at least 4%.
15 of those are also rated Financial Strength A or better, 10 with A+ or better.
Those 10 comprise 3 oil, 3 guns, 2 drugs, one copper miner and Intel.

It’s quant, even if it’s not really backtestable.

Jim
#-----------------------------------------------------
mungofitch:
I guess the question you raise is: Are the current earnings actually realistic, or are they cooked because of changes in the corporate climate?
Is it possible to do a similar analysis with free cash flow? This would reduce the variance of earnings which may be deviating recently. This presupposes that earnings are not being evened out by looking at the market as a whole (which may or may not be true)

I think they’re real earnings, not fake or cooked, I just don’t think they’re sustainable.
Any economy has earnings cycles, and the US is at or extremely near the top of one.
Earnings were lower a couple/few years ago, and will be lower again a couple/few
years from now, so today’s on-trend level of earnings is lower than the current observed level.
The trend line ist on the order of $20 lower in earnings on the S&P 500 level;
somewhere around the low $60’s, not the low $80’s which we’re seeing.

This is true only in aggregate, however: some firms will be doing even better
in the next recession than they are now, some about the same, and some will go bust.
It’s probably good to think today about how you can distinguish those classes.

There’s nothing wrong with irregular or cyclical earnings per se, but
it’s even more critical that you value that kind of company on the long
run trend and its likely future trajectory, not on the current level.
Jim
#----------------------------------------------------------------------------------------

Is it possible to do a similar analysis with free cash flow? This would reduce the variance of earnings which may be deviating recently. This presupposes that earnings are not being evened out by looking at the market as a whole (which may or may not be true)

Free cash flow is actually more volatile than earnings. The whole purpose of the income statement is to give a smoother picture of cash flow (Matching revenues with expenses). For ex., capital expenditures for Boeing are erratic, with huge amounts of spending lumped together followed by very little spending for several years.

Curious if testing SP500 components might be a more reliable predictor due to size, earnings predictability, liquidity, # analysts, etc… opposed to smaller cap, less liquid companies
Author: mungofitch 8/22/2011 3:34 PM
I checked it.
Extremely similar results, using the count of S&P stocks > earnings yield
10% (rather than count of VL 1700 stocks) to predict average VL stock return.

About the only difference is that at the bottommost percentiles the forward returns aren’t negative.

Percentile range Forward 1yr ret Forward 1yr ret
Using VL Counts Using S&P Counts
From 0 to 0.02 -4.4% 1.9%
From 0.02 to 0.05 -1.7% 7.4%
From 0.05 to 0.1 6.2% 5.0%
From 0.1 to 0.2 9.9% 7.0%
From 0.2 to 0.3 11.7% 12.0%
From 0.3 to 0.4 9.4% 10.2%
From 0.4 to 0.5 10.4% 7.5%
From 0.5 to 0.6 14.8% 12.3%
From 0.6 to 0.7 14.9% 16.6%
From 0.7 to 0.8 23.2% 21.1%
From 0.8 to 0.9 6.5% 7.9%
From 0.9 to 0.95 18.9% 20.8%
From 0.95 to 0.98 39.9% 36.1%
From 0.98 to 1 57.5% 65.3%

The current level is at the 91st percentile.

Jim
#---------------------------------------------------------------
Author: mungofitch
At the moment, it’s just over 30% of the stocks with a stated “Current P/E” ratio listed.

Using the table from the old post…
Starting percentile % of VL stocks Return
From 0 to 0.02 (0.012 to 0.021) -4.4%
From 0.02 to 0.05 (0.021 to 0.027) -1.7%
From 0.05 to 0.1 (0.027 to 0.036) 6.2%
From 0.1 to 0.2 (0.036 to 0.050) 9.9%
From 0.2 to 0.3 (0.050 to 0.062) 11.7%
From 0.3 to 0.4 (0.062 to 0.073) 9.4%
From 0.4 to 0.5 (0.073 to 0.088) 10.4%
From 0.5 to 0.6 (0.088 to 0.103) 14.8%
From 0.6 to 0.7 (0.103 to 0.121) 14.9%
From 0.7 to 0.8 (0.121 to 0.175) 23.2%
From 0.8 to 0.9 (0.175 to 0.239) 6.5%
From 0.9 to 0.95 (0.239 to 0.289) 18.9%
From 0.95 to 0.98 (0.289 to 0.344) 39.9%
From 0.98 to 1 (0.344 to 0.501) 57.5%

…that suggests we’re between percentile 95 and 98, and returns close to 40% would be predicted in the next year.
Don’t hold your breath for that, but at least it’s a bullish sign of hope in these awkward times.

Jim
#------------------------------------------------------------------------------------------------------------------------------------
mungofitch
May 9

When the market drops, BRK’s book value also drops. So it would take lower price to reach 1.25 or 1.2.

True, but probably not particularly useful.
The ratio of price to “peak-to-date book per share” is a better metric of the company’s value.
Dips in book value at Berkshire tend to be transient.

Also, empirically it has also been a better predictor of forward returns in the past.
Buying at a medium multiple of depressed book value (low multiple of peak book value) has generally worked out very well.

Jim

#----------------------------------------------------------------------

do you find this site reliable or calculate by hand:
Berkshire Hathaway Price to Book Value
mungofitch
By eyeball, it doesn’t match mine. Generally yes, but in the finer details no.
At a guess, it seems (?) they are using P/B at (say) Dec 31 2019 being the book value on that date, even though it wasn’t known till a couple of months later.

Personally, I track “published book”: the most recently published figure known on a given date.
Technically speaking, book value is something that exists only on a day that a set of statements is prepared.
And knowable only after that team tells you that result.
This isn’t just price changes in the equity portfolio#there are many accounting adjustments that are done only when a set of books is prepared.
Jim

#----------------------------------------------

mungofitch
Here’s a really simple approach.
(a) If the valuation multiple is not unreasonably high, just invest any saved money immediately.
(b) If it is unreasonably high, wait till it is, letting the cash stack up for a while.
It’s up to you to decide what constitutes “unreasonably high”.
But FWIW, since the credit crunch, the average one year stock price return has been negative in real terms on purchase dates at or above 1.54 times peak-to-date published book per share.
The average one year forward return starting from P/B 1.50 has been inflation + 1.88%.
Jim
#--------------------------------------------------------------------
Knighted:
Jim, thanks for the thoughtful responses. I plan to study the option strategy you suggested further. In the meantime, I did some additional backtesting of the approach you stated. I tested two flavors, but unfortunately both seem to be flops. Any thoughts on why that may be? What you state makes logical sense, but I struggle to reproduce a consistent advantage in backtest. I assume that’s because the missed opportunity during periods where P/B remains stubbornly high offsets the advantage by holding cash during these periods and buying in at lower P/Bs. (Note: I haven’t tested your other idea yet of gradually ramping up the P/B thresholds as time passes)

Pick a single P/B threshold, say 1.5. Invest any newly available funds immediately at any P/B below this, and hold onto newly available funds in cash at any level above this, investing the lump sum the first time P/B reaches that 1.5 threshold . I ran this through solver to analyze P/B values from 1.00 to 2.00 from Y2000 and again from Y2005 to date. Surprisingly, this showed negligible total return benefit even with the best parameter: the best produced around 1.5% advantage -total- from 2005, not annual.

Pick two P/B thresholds, say 1.2 and 1.5. Invest any newly available funds immediately at any P/B below 1.5. But if P/B is above 1.5, hold the funds that would have been invested in cash until it reaches 1.2 or below (although future funds would still be automatically invested in the < 1.5 range). I again ran this through solver to analyze all P/B values from 1.00 to 2.00 for each threshold from Y2000 and also from Y2005 to date. The 2005 to date produced a max total return around 8% higher than the baseline for the two optimal P/B thresholds. Still not much of an advantage, but something (+0.4%/yr). I also plotted all runs on a surface in excel to look for the largest cluster of groupings (trying to avoid picking an outlier) - these two special P/B thresholds fell within that. But to stress test the idea, I then ran the same backtest on Y2000 to date data and ended up with very different results. That advantageous P/B from Y2005 to date yields worse results than the baseline from Y2000 to date. This shakes my confidence in the approach, but welcome thoughts on why it should not.
#--------------------------------
BenSolar

I struggle to reproduce a consistent advantage in backtest. I assume that’s because the missed opportunity during periods where P/B remains stubbornly high offsets the advantage by holding cash during these periods and buying in at lower P/Bs.

I expect that’s the case. Berkshire is a stock that generally doesn’t go through a lot of gyrations, and the company is always piling up value, so going ahead and investing at first opportunity is basically as good as the valuation based schemes being proposed.

#------------------------------------

mungofitch:
Bear in mind that I’m not assuming that a multiple of book value remains a pretty good yardstick.
There are many good reasons to assume that a given P/B ratio might not mean the same thing over time, so that would be a bad idea.
Mr Buffett’s warning not to rely on it was a good warning.

Rather, I’m stating that it remains a pretty good yardstick.
That result may be because of a bunch of coincidences, and that might change at any time, but so far there is no divergence between a dumb multiple of book valuation model and what I believe is a pretty sophisticated one taking into account operating earnings, the value of float, cyclical adjustments on various units, etc.
A given P/B multiple is (so far) as good as it ever was, and means about the same as it meant many years ago. There has been no reason to assume that this would be the case, but it is the case so far.

Many expect the fair multiple of book to drift upwards ever so slowly over time.
But if anything, to the extent there is divergence between P/B and other valuation models, based on my figures the fair P/B is either flat or has slid down just a teeny tiny bit in recent years.

There are quite a few moving parts determining the “fair” P/B and its evolution. Over/undervaluation of big equity positions, buybacks, big acquisitions driving the fair P/B down organic growth and capex within subs driving it up, and inflation driving it up.
So far, those things are in balance within my measurement error.

Jim
#---------------------------------------------------

“One-year Forward Returns based on [high] P/E Level” Brooklyn Investor

You probably shouldn’t have been thinking about doing that study. Waste of time.
Earnings are so volatile that the current P/E ratio is pretty darned weak as a metric of market valuation level,
and market valuation in turn is very weak (nearly useless except when dirt cheap) as a predictor of forward one year returns.
When earnings are cyclically low, for example, high P/E ratios make perfect sense: low earnings are temporary.

Look instead a smoothed earnings, versus forward returns over a period longer than a year.
Then it all makes sense.

e.g, even since 1995, ignoring the older data that most people use to define “normal”.
Below, forward real total return ~7 years into the future based on buckets of starting trend earnings yield.
The figures shown display the range of outcomes for start dates with trend-earnings valuations in the buckets indicated.
(ending periods are smoothed, the average index level in a period centred 7 years later, adjusted for dividends and inflation)

                           Lowest    Pctl 10   Average   Pctl 90   Highest

Most expensive 5% of time -4.1% -3.8% -3.2% -2.5% -2.3%
Next 10% of time -3.4% -3.1% -2.1% -0.8% -0.1%
Next 10% of time -2.4% -1.9% -0.1% 1.8% 2.2%
Next 10% of time -0.5% -0.1% 1.5% 3.6% 4.3%
[current situation is between these two rows, assuming net margin levels of the last 10-15 years continue]
Next 10% of time 0.6% 0.9% 2.8% 5.2% 5.7%
Next 10% of time 0.9% 1.2% 2.6% 3.5% 6.3%
Next 10% of time 1.4% 1.6% 3.1% 3.8% 7.0%
Next 10% of time 1.8% 2.0% 4.2% 7.5% 7.8%
Next 10% of time 2.1% 2.3% 4.8% 8.6% 9.1%
Next 10% of time 3.2% 3.7% 6.4% 10.3% 11.2%
Cheapest 5% of time 4.7% 5.0% 9.7% 13.2% 13.6%

So, the best return seen starting in the most expensive 5% of the time has been -2.3%/year.
The worst return seen starting in the cheapest 5% of the time has been +4.7%/year.

So, the reasonable conclusions are that:

  • Earnings need a cyclical adjustment before they have any utility at all in estimating market valuation levels.
  • Valuations at purchase date matter than anything else in predicting forward returns in period over a couple of years.
  • Trend earnings yields keep on working as a predictor even in the modern era.
  • There is not much evidence that valuation levels, even properly measured, tell you anything much about the next year.
  • To the best of my knowledge there is no useful market timing signal based on market valuations.
  • You can’t debunk the notion that market valuation levels predict forward returns by looking at a bad model.

Jim

17 Likes

Awesome!

Thank you, Ted.

I wish I could super like this.

1 Like

The stuff about how to get free PBV for other tickers seems useful, I’ll be checking some of them out. I also gather from what cwags02 said that he’s starting to get some results on other tickers.

The rest of the long post above is simply archiving on this new board some of the posts that Jim/mungofitch had on this topic on the old board.

5 Likes

A question on my mind for a while is: what features of a company makes measuring P/B value the most useful valuation metric compared to other metrics like P/E, P/S, etc?

We consider P/B to be useful for valuing Berkshire, but Jim has pointed out before that seeking low price/book can be a very poor screening criteria for many companies because it can flag companies with high levels of assets yet weak growth. Whereas very high price/book can be indicative of some really good companies that are able to earn a lot with very little assets (a great business).

In contrast, Jim’s stated before that he considers a good valuation metric for a company like Alphabet to be price/sales.

What I’d be fascinated to learn/read about is: what characteristics makes price/book a good valuation tool for one company, price/sales for some and good old fashioned P/E for others. Has anyone come across something like this?

3 Likes

The top quote of 8/2/2020 is very informative.

Jim would often share projected 1 year returns. I believe that post gives us some insight into his model. The idea is to first determine what P/peakB will be in 1 year, then back out the price. I surmise it goes something like this.

  1. Jim knows daily P (apply smoothing/averaging) and daily peakB. Use these time series to create a third daily time series: P/peakB.

  2. Use linear regression to determine a best-fit line to the peakB time series. This might work best using only the days book value is reported (i.e. “fresh” book value).

  3. Use the best-fit line to estimate peakB(today + 1.5yrs)

  4. To build the model he references in the table he shared, let D(t) be the time series
    D(t) = P/peakB(t+1.5yrs) - P/peakB(t). Create the set of points (P/peakB(t), D(t)). Use linear regression to determine a best-fit line to this set of points (Jim said he sometimes used a cubic model to get a little bend at the ends, and then might even average that with the line, but let’s just leave it at a line for now).

  5. Calculate today’s P/peakB.

  6. Use the regression line from (4) to estimate D(today).

  7. Estimate P/peakB in 1.5 years adding the results from (5) and (6):
    P/peakB(today+1.5yrs) = P/peakB(today) + D(today).

  8. Finally, estimate price in 1.5 years multiplying results from (7) and (3):
    P(today+1.5) = P/peakB(today+1.5) * peakB(today+1.5)

That gives an estimate 1.5 years out. I did that to match the data he shared. You’d have to replace 1.5 with 1 everywhere to get an estimate one year out.

Caveats: I’m leaving out adjustments for inflation and probably using log(peakB) in (2), which you would expect to be a line.

Sound feasible?

1 Like

@Knighted: I’m pretty much a dolt at fundamental analysis, so I bought a book (now in storage because I’ve moved, and I can’t remember the title, and apparently it wasn’t bought via Amazon) that was published relatively recently. IIRC, it went through different types of companies on a case by case basis, looking at the import of different fundamentals, and then had you try to analyze a new company (with answer later).
Maybe someone knows which book I mean?
Perhaps we have some knowledgable fundamental analysis folks who can chime in?

A little googling does provide some help e.g.
Investopedia

EXCERPT:
Capital intensive industries
The P/B ratio is only considered useful in practice when applied to capital-intensive businesses, such as energy or transportation firms, large manufacturers, or financial businesses with a significant amount of assets on their books.
For example, the bulk of Microsoft’s asset value is determined by its intellectual property rather than its physical property. As a result, Microsoft’s share value bears little relation to its book value.

Intangible Assets
Book value ignores intangible assets such as a company’s brand name, goodwill, patents, and other intellectual property. That means it does not carry much meaning for service-based firms with few tangible assets.
For example, the bulk of Microsoft’s asset value is determined by its intellectual property rather than its physical property. As a result, Microsoft’s share value bears little relation to its book value.

Debt Levels
Book value does not offer insight into companies that carry high debt levels or sustained losses. Debt can boost a company’s liabilities to the point where they wipe out much of the book value of its hard assets, creating artificially high P/B values. Highly leveraged companies, such as cable and wireless telecommunications companies, have P/B ratios that understate their assets. For companies with a string of losses, book value can be negative and, hence, meaningless.

Asset Values
Behind-the-scenes, non-operating issues can impact book value so much that it no longer reflects the real value of the assets.
First, the book value of an asset reflects its original cost, which is not informative when assets are aging. Second, the value of assets might deviate significantly from the market value if the earnings power of the assets has increased or declined since they were acquired. Inflation-or rising prices-alone may well ensure that the book value of assets is less than the current market value.
END OF EXCERPT

The way to pull down free historical fundamental ratio data that I mentioned earlier works for ratios other than P/BV.

FWIW, I suspect MSFT is a “good” company, i.e. it would have a good scatterplot if the right fundamental ratio for it is used.

3 Likes

@JohnIII
I didn’t interpret it that way, BWDIK?

I also took his description in the 8/2/2020 post;

“Given the ratio of market price to ‘peak to date’ book per share on any given date in the past, look at the average market price 1-2 years later. So, it’s a look ahead period of average 1.5 years.”

to mean (I’ll leave out ‘peak’ and ‘inflation adjusted’ to save space&clutter):

that if you were going to do a regression, then the dependent variable would be the average of the price over a date span covering “today plus one year” to “today plus two years” i.e. a smoothed variable and not a single price that’s 1.5 years out from date of present P/BV.
The independent variable would be today’s P/BV.
The regression gives a a modeled price that can be centered at 1.5 years out and you have today’s price, so then simply annualize that return. Or, could use point price at fixed time period out and not a smoothed price, doesn’t matter to the regression, and you could add a cubic in P/BV to the simple regression if desired.

I’m confused, how could you match the data he shared given your later caveat that you left out inflation correction and comment about using (or not using) a log. I must be misinterpreting something in what you wrote.

Anyway, a request could be sent via his site asking if he has time to clarify. Maybe he’d even have time to read some of the analysis here and could comment on it on his site, if so that’d be wonderful!

I thought his main privacy concern related to him posting on the new site. But if it also relates to reading (cookie tracking) then wouldn’t a burner account allow him to read, and could even do so within a virtual OS, and if it’s tracked then it’s of no consequence to him personally? Maybe not, he’s the guy who apparently has patents on tracking.

2 Likes

I’ve found the best source of free data is morningstar courtesy of my local library.

I can download historical P/E, P/S, P/BV, and P/Cash going back to the mid 1980s. The P can be daily, weekly, or, monthly and the E, S, BV, and Cash are quarterly.

I spot checked the morningstar data to the data I found using Yahoo, and the annual reports and found the results were very similar. I trust morningstar.

5 Likes

When I look on Morningstar it requires a subscription (I long ago up used the free trial period).
I guess it’s the library that has the subscription, so it’s now free to you. I’ll look into that, M* should be a reputable source.

2 Likes

Sorry if I wasn’t clear. The steps I outlined assumed access to the daily time series, which were not provided, but could be constructed easily enough IF you have the “P/B on announcement day” series. If you are fitting a line to the peakB series, I imagine you would want to adjust for inflation first. Also, you would expect peakB to follow an exponential curve over time, even after inflation adjustment, no? So fit a line to log(peakB).

I think the most important ingredient in any of these models is having a good BV series. I imagine using BV as reported by BRK would be okay, but the requirement that they follow GAAP introduces some undesired variability. For instance, Jim gave a haircut to stocks held by BRK that he felt were overpriced.

Jim regularly shared his BV calculation for BRK. It’s possible every quarter is available going back 15 years buried in the old message board. If we used those, then for consistency we would also need his BV calculations going forward. I imagine there’s a good chance he could be coerced into sharing those on his website. Is there another source, publicly available and that we would trust, from which we could download BV going back 15 years? Ideally it would have this property: Assume God has the correct BV series which of course no human has access to. Any BV series that deviates from God’s by the same constant multiple each quarter would work just as well. So if someone is consistently overestimating BV by 15% each quarter, that would work perfectly.

2 Likes

Also, when I said “match the data” I probably wasn’t speaking precisely. I just meant “match” in the sense that Jim was projecting 1.5 years out so the steps I wrote out did the same. I wasn’t matching in the sense of producing the same numbers. I couldn’t do that without working with the daily time series, which would include Jim’s BV estimates (which I assume is what he’s using). And even then, there are likely things Jim does that might be difficult to reproduce (e.g. a weighted sum of a linear and cubic model?)

2 Likes

Ultimately, for Berkshire, I want to re-create the five groves on a quarterly basis going back to 2018 when Buffett announced the methodology. I trust his reasoning.

For everything else, well it seems to depend - it looks like you, @tedthedog, and I are in the same boat.

Let’s take Carmax as an example. Carmax is an operating company so using something like P/BV seems to be a poor fit and P/E would be much better. Performing a similar analysis as what was done for Berkshire but with P/E instead of book value are shown in Figure 1 and Table 1 below. The analyses are not helpful and the results are not useful for projecting returns, the 2008/2009 time period really skews the data. The data boil down to, ‘invest when P/E is really low or really high but definitely not when it’s around it’s usual P/E’. Some sort of smoothing should really occur since earnings are so bumpy. Smoothing like what’s been mentioned by @tedthedog and @johnIII seem like very good ideas when using Carmax’s P/E data.

A different way to look at Carmax would be through their P/Sales ratio since sales are a whole lot less volatile. Sales did decrease thru the great recession but did not collapse like earnings. Performing the same analysis results in Figure 2 and Table 2 below. This is a lot more informative and seems useful if a person believes Carmax will be around in the future, the earnings will follow revenue, and the past is similar enough to the future. Just my kind of model. Interestingly, the P/Sales analysis shows this as a great and rare opportunity to buy Carmax.

I’m working through doing similar analysis for my favorite companies like GGG, FAST, ECL, and others (DLTR). Everything goes pretty quick once I get the data from morningstar.

I think there’s a very valid discussion to be had on how to smooth the data to create the approximate answers that are needed instead of the approximate answers that are wanted.

Direct image uploads didn’t work - I’m trying to get this sorted. Until it’s sorted, here are some links:

7 Likes

You are correct - my library has a subscription and I am a library member. I also have access to ValueLine but I can’t figure out how to download or extract the data. Their website is pretty clunky.

2 Likes