The Evolution of the “Hussman Chart”

Hardly a week goes by without an article like this one by the influential Henry Blodget — One smart stock market analyst thinks this is where we’re headed…(gulp). Mr. Blodget writes as follows:

But anyone who’s feeling comfortable after a strong week in the markets should at least understand that: 1. The macro environment most conducive to crashes is still in place (overvaluation + increasing risk aversion) and 2. The way the market is behaving now is exactly the way it behaved before the biggest crashes in history.

So, neither Hussman, nor I, nor you should be surprised if the market keeps on dropping and doesn’t bottom until it’s down 50% or more from the peak.

As Hussman noted last week in his usual depressing note, a 50% crash would not even be the worst-case scenario. It would just be a normal correction from valuations we reached in 2015.

Featured in the valuation articles is a chart purporting to show very low expected annualized returns for a multiple-year period. The implication for stock investors is clear: Little upside combined with huge risk. It has had a big impact both with individual investors and also my investment advisor colleagues.

Background

Last week, among several other illustrations of popular investment misconceptions, I included a version of what I will call the “Hussman Chart.” I suggested that if you did not understand the chart, you shouldn’t be using it for your investment decisions. My main point was that people blindly accept conclusions from intelligent sources who use sophisticated methods. Of the are dozens of possible illustrations, I included the Hussman Chart. I know that many people sold their stocks some time ago when their investment advisors warned them, producing one of the charts I discuss below.

My worst fears were confirmed! Out of the thousands reading the post, only two or three explained anything about the chart – how it was constructed, what it implied, how to think about it. Quite a few people repeated the author’s conclusion. Wow! They understood and accepted the conclusion without any evaluation of the reasoning. Others did not want to be challenged. They wanted me to explain what I thought was wrong with the chart.

Readers promptly ignored everything else in the article. Some even concluded (amazingly) that I was stating that I personally did not understand the chart. Jesse Felder, a fellow investment advisor and blogger stated this viewpoint explicitly. His conclusion (without any explanation of the chart):

Furthermore, this negative correlation between valuations and forward returns is statistically very high (greater than -90%) and backed by 65 years worth of data. The Buffett Yardstick, as Hussman demonstrates, has been nearly as good as his own version at forecasting forward returns and is backed by roughly 90 years worth of data. Both charts, and the data and reasoning behind them, clearly demonstrate and validate the concept that, “the price you pay determines your rate of return.”

Summary

Apparently I need to elaborate on the original theme. I will do so by providing examples of “the chart” over the years. The variables, adjustments and time periods change, but the conclusions are generally the same. Each chart has a documented method and stands on its own. Together one gets a different picture. While the method is continually “improved” and the time period changed, there is never a date with destiny. We do not know whether the early versions worked or not. There is no distinction between the time period used to create the method and the “out of sample” period that follows.

Here is a summary of the charts below.

Date Independent variable Starting point Length of Forecast Adjustments
Nov, 2008 Terminal multiple 1950 7 years
Oct, 2010 Terminal yield 1944 7 years
Aug, 2010 Adjusted forward earnings 1963 10 years Reducing margins
Jan, 2011 Normalized earnings 1928 10 years “Normalizing” earnings
Dec, 2013 CAPE 1932 10 years Mean-reverting margins
Feb, 2016 Non-financial Gross Value Added 1950 12 years World effect, excludes financials

The rest of this report will show the evolution of the approach and raise some specific concerns and points that you might wish to consider.

[I have never met Dr. Hussman, but I have a generally favorable impression of him. He taught for a bit at one of my schools. (Someday I might learn if he considers himself a “Michigan man.”) He is respected as a philanthropist. His approach is intended to be in the best interest of his investors. Updating his methods and conclusions is a natural part of investment management. He reports his thinking frequently and takes on issues directly. Were this not the case, a review like this one would not be possible. He has built a very successful business and earned a strong reputation. His articles are always among the most popular, especially among investment advisors].

Analysis – the Evolution of “The Chart”

First example — How Low, How Bad, How Long? November, 2008

http://www.hussmanfunds.com/wmc/wmc081110.htm

wmc081110h

Second example — No Margin of Safety, No Room for Error October, 2010

http://www.hussmanfunds.com/wmc/wmc101011.htm

wmc101011b

Third example — Valuing the S&P 500 Using Forward Operating Earnings August, 2010

http://www.hussmanfunds.com/wmc/wmc100802.htm

This quoted explanation illustrates why some might have trouble following the methodology:

The two main failures of standard FOE analysis are that 1) analysts assume a long-term norm for the P/E ratio that properly applies to trailing net, not forward operating earnings, and; 2) analysts fail to model the variation in prospective earnings growth induced by changes in the level of profit margins, and therefore wildly over- or underestimate long-term cash flows that are relevant to proper valuation. By dealing directly with those two issues, we can obtain useful implications about market valuation.

As I have frequently noted, it is not theory, but simple algebra, that the long-term annual total return for the S&P 500 over any horizon T can be written as:

Long term total return = (1+g)(future PE / current PE)^(1/T) – 1
+ dividend yield(current PE / future PE + 1) / 2

The first term is just the annualized capital gain, while the second term reasonably approximates the average dividend yield over the holding period. For the future P/E, one can apply a variety of historically observed P/E ratios in order to obtain a range of reasonable projections, but the most likely outcome turns out to be somewhere between the historical mean and median.

You have to get two things right: the “normal” future P/E and the prospective long-term earnings growth rate g. Standard FOE analysis misses on both counts. Very simply, looking out over a 7-10 year horizon, the proper historical norm for price-to-forward operating earnings is approximately 12.7. Moreover, one cannot simply apply the long-term operating earnings growth rate of 6.3% (0.063) as an unchanging measure of g. Rather, an accurate growth rate for the model has to reflect the level of profit margins at any point in time, since the current P/E multiple may reflect either depressed or elevated earnings. For a 10-year investment horizon, the proper value of g should take into account the gradual normalization of margins. Historically, the best estimate is approximately:

g = 1.063 x (0.072 / (FOE/S&P 500 Revenues))^(1/10) – 1

[Jeff]You should at least be able to understand that the earnings are “adjusted” by a method that is deemed to be appropriate.

wmc100802a

Fourth example — Borrowing Returns from the Future January, 2011

http://www.hussmanfunds.com/wmc/wmc110117.htm

wmc110117a

Fifth example — Does the CAPE Still Work? December, 2013

http://www.hussmanfunds.com/rsi/cape.htm

CAPEh

http://www.advisorperspectives.com/commentaries/20160221-hussman-funds-speculative-half-cycles-tend-to-be-completed-badly

wmc160418c

[Jeff] If you look at this chart and the two above, you will see that the big divergence in the late 80’s has disappeared.

Comments on the multi-year growth projections

These are points that would be discussed extensively if the research had a peer review.

  • It is necessary to explain carefully both variables, especially making clear when one can evaluate the relationship
  • There should be a sharp distinction between the portion of a chart which is a back test, or an idea fitted to past data, and the “out of sample” data that follows.
  • A multi-year projection has an eventual “date with destiny.” If you are one year away, you can calculate the return that would be needed to make the forecast correct. Think of it as a runner going for a world record in the mile. If he is five seconds off the pace with 100 yards to go, you may safely conclude that he will not break the record.
  • The concept might be extended to more years. If a very negative forecast is in place and the first year or two is strong, it might take a market crash for the forecast to come true.

Research Tests

This very brief summary is a glimpse of what a solid research design should include.

Necessary

There should be a hypothesis and a test of the hypothesis.

It should be possible to disprove the conclusion.

Stated results should not consume all of the data.

Desirable

It is best to share data, especially when not proprietary. This allows others to replicate the work. (One of the top economics books of the past year included a serious spreadsheet error, discovered because data were shared. It is fairly common in academic circles. Dr. Shiller shares his data, despite the great difficult in develop the historical earnings).

It is important to provide a complete description of the methodology. This should include paths not taken and variables that were rejected.

It is helpful to show the link (ideally with an update) of past research theses as more evidence emerges.

My Own Concerns about the Conclusions

Many have asked me why I have not followed this approach in my own investment management. I do not write about it very much because of the work required. Dr. Hussman has a great research budget and team. I have a small staff who are already fully-employed on stock picking and managing our programs. Going back to replicate one of the old charts would be a fair amount of work. I will share my concerns here, but only in abbreviated form.

  • We never seem to reach the point of evaluation. How did these approaches work in the past?
  • The methodology seems to include many of the classic overfitting problems. I am certainly not the first to note this. Philosophical Economics in late 2013 wrote Valuation and Stock Market Returns: Adventures in Curve Fitting.
  • There are adjustments that are not well explained. The earnings are adjusted for expected changes in profit margins, for example. What if this assumption is not accurate? Profit margins are an intense (and separate) debate.
  • The method for adjustment keeps changing – different approaches, coefficients, etc.
  • Over the years, the time frame for the forecast keeps moving, from seven, to ten, and to 12. If you go back to the original Shiller papers, he was using five years. His disciples keep experimenting with different choices.
  • The independent variables change with each new iteration. The overall model always seems to fit. Past discrepancies disappear.
  • The attribution of “bad patches” in results to market overvaluation or undervaluation. This seems backwards. Why is the market wrong and the model right?

I am especially bothered by what I see as exaggeration and distortion. What does it add to this discussion to call valuations “obscene?” I find especially distasteful the statement, “The CAPE Ratio id doing exactly what it has always done, which is to help investors anticipate the investment returns they should expect over the next decade. Those returns will very likely be in the low, single digits”.

The CAPE ratio is not some wise old friend that has been around for centuries. It was invented only recently and has not worked very well. The claim of historical validation is also completely wrong. What if I told you that the Packers always won at home after a double-digit away loss in a dome? (I made this one up, but you get the idea). It is historically accurate, but does not have any value for predicting the future. Since Dr. Shiller and Dr. Hussman made a lot of specific choices about measuring earnings, past time frames, use of inflation information, and future time frames, their conclusions should be described as a model, not some definitive historical record. It is rather easy to create a view of history that provides a vastly different conclusion. (see The Single Greatest Predictor of Future Stock Market Returns). It includes this impressive chart.

avginv11

[Jeff] Similar approach, vastly different result. This is not the only such example.

Implications for Investors

My most important point is a plea, repeated from last week’s post: Be careful about investing your money using analysis you do not really understand!

Whether you share my concerns or not, I recommend a deeper look into these issues, with one of three conclusions:

  1. If this leads you to agree with Dr. Hussman, his fund offerings that provide the best balance. I have written that his stock picking is excellent. Investing with him is better than going “all out” on your own because of fear.
  2. If the deeper look leads you to disagree, you might consider funds or advisors who take a different approach.
  3. If you are not sure, then hedge your investment “bets.”

5 thoughts on “The Evolution of the “Hussman Chart””

  1. Thanks for the thorough work. I have been an investor in Hussmans funds, but as always, I give it a trial period (with a small investment) and evaluate vs. my own methods. In the early 2000’s he got it right. Then he panicked and changed his methods in the middle of the 2008-2009 turmoil; which promptly killed the performance. As you say, I have been concerned about his missing/messing with time frame factors in his modelling. Old tried and true methods that are moderately simple (that I know he has access to due to public information I have found), have proven vastly superior to his set of variables.

    Another weakness he had was that he depended heavily (basically completely) on data sets since 1945. One of my more important market health variables has been useful and tested back to the 1920’s; and I am just an amateur. He has since indicated that his testing data go to prior to WWII.

    Thanks again.

  2. In all honesty I do not understand the trouble that people seem to have with Hussman’s conclusions, essentially because they are based on a very simple premise and that is that the price one pays for an asset determines the return one gets from that asset. Warren Buffet has stated that repeatedly and it squares with plain old common sense. No investor in the future cares about what you or I paid for an asset, nor does the price you or I pay impact the operating entity underlying that asset. So the price in the future of that asset is totally independent of the price we pay for that asset today. The higher the price one pays the lower the return one gets and the lower the price one pays the higher the return will be. That is the essence of any of these Hussman or CAPE or whatever models.

    If his work was based on seasonality, alignment of the planets, what team wins the Super Bowl and other such factors one could understand the skepticism, but his work is about as fundamentally driven as work of this nature could be.

    Your date with destiny comment is not appropriate. By example, let us assume that I buy the S&P500 index at a PE ratio of 50. My expected return using one of Hussman’s models would presumably be very low, in fact probably negative. However, if somehow seven years later (or 10 or whatever forecasting horizon the model uses) my returns have turned out to be higher than that assumed by his model what have we learned? Well, maybe some fundamental element changed, driving profitability higher than that assumed in his model. Hurray for me. Did I presciently know where his model underestimated expected profitability or was I just lucky? Or maybe some exuberance took hold, irrational or otherwise, which bid up stocks to 100 times earnings. Again, hurray for me. Should we throw out his model, or others of that nature, because they happened to not account for investors willing to pay ever higher prices for a stream of earnings? However, if we answer that question affirmatively we are then saying that for an investor to realize the same return over the ensuing 7 years that I just achieved over the last 7 (after all the Hussman model has just been proven to be garbage and the market returns what the market returns so why not expect a similar return in the future, right?) investors will need to be willing to pay 200 times earnings at that future point, and 7 years after that 400 times earnings, etc.

    Frankly, I find it very unfortunate the comment that you made about what you found distasteful, as that is a very strong statement. Models of this nature are meant to guide investors to understand what returns they should reasonably expect, given current asset prices. There will be the usual standard statistical errors associated with such a forecast and, as I stated before, returns may differ because of some fundamental cause (which we should probably assign an expected value of zero today, but if you have any reason why that should not be the base assumption we would benefit by knowing your reasoning) or because investors in the future will pay a different price for a stream of earnings than that assumed in the model. If you think that second factor should have an upward bias, relative to the assumptions inherent in his model, it would be helpful to the reader for you to explain why.

    I am not a cheerleader for Hussman, specifically. Frankly, it is much easier just to go to GMO’s quarterly 7-year expected returns for major asset classes to get the essence of what Hussman’s work would conclude. Of course, you may not agree with the conclusion GMO reaches either.

    Thanks for your good work. I read your Dash of Insight every Sunday.

  3. Great post! CAPE strikes me as useful for understanding the past, but a dubious base for conclusions about the future, especially since my sense is that the data has to be tortured to some extent to yield those precise results. However, I have not the time nor the inclination to really dig on this one, there are other arenas of research that seem more relevant to my investment approach.

  4. I was initially a big fan of Hussman’s back in the early 2000’s. However, his “prowess” turned out to be almost entirely the result of favorable dollar-cost averaging and growing his fund’s assets during the fall and recovery of the tech-stock bust. That accounts for his years of poor relative (and absolute) performance since.

    http://finance.yahoo.com/q/pm?s=HSGFX+Performance

    Honestly, I don’t understand how Hussman-defenders can get past this.

Comments are closed.