John Hussman and the Fed Model
John Hussman has written a piece about market valuation and the Fed model that breaks some new ground. It is sufficiently different from his past work to deserve some special attention.
Background
At "A Dash" we believe that thinking about valuation is important. Some idea of overall market valuation helps both individual investors and traders by providing context about risk and reward. Significant deviations in market prices from the "fair value" of one’s model provide a signal of what we call long-term sentiment, be it euphoria or a "negativity bubble."
Dr. Hussman agrees about the importance of valuation, since he writes about it on a regular basis. He is an outspoken critic of the Fed model, preferring his own approach. He uses something he calls peak earnings, referring to the highest past level of GAAP earnings. While he writes with the objective of explaining hedging strategy to his own investors, his work is widely cited.
In another element of common ground, the Hussman articles frequently cite those using an approach like the Fed model, but not naming it explicitly. The many market commentators who view current stock prices as cheap, comparing forward earnings to interest rates, are doing something akin to the Fed model, but without citing it explicitly. Our guess is that many commentators wish to avoid the debate about this approach–the intellectual baggage associated with naming it.
We have not been bashful about analyzing and naming the Fed model. We get frequent questions, from readers and our own investors, asking why our approach differs so much in its conclusion.
The Key Issues
While the purpose of this article is the current Hussman argument, readers can understand the problem better by understanding the following key issues:
- Looking backward. Hussman (and other noted commentators) prefer to look at documented past earnings rather than the forward projections of analysts. We prefer prediction based upon the forward look. Our basic rationale is that a backward view misses turning points in the market and that analyst estimates are much better and more honest now than in times past. We like to take the work of hundreds of specialists, trying to do their jobs. Our daily reading of these reports suggest that there is plenty of skepticism built into their work.
- Using interest rates. Most of Hussman’s work looks at interest rates OR earnings, not putting the two together. We believe this misses the major valuation point. When interest rates were at extreme levels, stocks deservedly had a low P/E (or high earnings return if one inverts the equation) to merit investor consideration. When interest rates are low, the opposite argument holds.
- Interest rate trend. Hussman feels that the trend in interest rates is an important element. We disagree. At any point in time, Mr. Market is offering the investor a choice between stocks and bonds. The investor can lock in a return via a bond purchase. The key element is NOT the direction of interest rates, but rather is investor skepticism about earnings estimates. One factor is known, the other is a matter of forecasting.
- Relevant time period. The Fed model uses data from 1980, because that is the time when forward earnings forecasts for the S&P 500 were available. Hussman believes that this period is atypical. Readers should read his work and consider his argument carefully. We believe that the time period includes a lot of interesting variation, and provides a lot of useful information. Like any researchers, we always wish we had more data, covering more time. Having said this, going back too far can also be a mistake. The investment world is much different in the time from the mid-70’s to now. The use of computers by all participants and the availability of information, and the growth of options and futures are all important changes.
The Current Question
Dr. Hussman reports research that is innovative in two important respects. First, he combines interest rates and expected earnings. Second, he uses the known period of forward earnings to create an intriguing model. He uses information about past earnings and economic trends to create a model for forward earnings. He then takes the model and applies it to the past – the period before forward earnings were readily available — to simulate what those forecasts would have shown had they been available.
He writes as follows:
Operating earnings are a “smooth” measure of earnings that, as it turns
out, can be cleanly and accurately approximated using variables that are
available historically – specifically, the highest level of S&P 500
(trailing net) earnings achieved to-date, the actual level of S&P
500 trailing net earnings, and the U.S. unemployment rate (which helps
to capture business cycle fluctuations). The ISM Purchasing Managers
Index provides slightly better accuracy than the unemployment rate, but
has a shorter history. The results below are not very sensitive to the
choice, so in the interest of data availability and to make it easy for
others to replicate and verify these results, I’ve used the
unemployment rate.
From this starting point, he derives the following regression equation:
forward operating earnings = k * record trailing net earnings to-date
where
k = 1.2721 + 0.3115 (current trailing net earnings / record trailing net earnings – 1)
– 0.5011 (unemployment rate / 16-month average of unemployment rate – 1)
– 0.5985 ( (record trailing net / record trailing net 5 years prior) ^ (1/5) – 1)
From this equation he develops a chart showing that in the pre-1980 era, going back to 1948, the Fed model did not have a good fit.
What the Equation Means
The essential problem is that very few readers have the methodological skills to interpret the Hussman equation (which is missing some parentheses, but we get the point). That means that people will focus uncritically on the conclusion.
At "A Dash" we believe in using experts whenever possible. Our own expertise is in choosing the right experts and in research methods. Let us try to interpret the Hussman findings.
The constant term of the model shows a "starting point" of 27% earnings growth. The next variable boosts that growth by a percentage calculated by looking at current trailing earnings versus record earnings. This is the hallmark Hussman variable.
The next variable considers the unemployment rate compared to the prior 16 months, giving earnings a reduction if the rate is higher. Fine.
The final variable looks at the "record trailing net" versus the record for the last five years and gives a major haircut to estimates if the record is higher. Hmm. We do not see the logic from this variable.
Perspective
Before stating our analysis of this research, the differing perspective should be made clear to readers. Dr. Hussman is a developer of a valuation method which he has used to guide his fund management. He has skin in the game, and frequently writes to investors to justify his approach. If his messages were private, we would have nothing to say. The problem is that his work gets broad exposure, including our investors and many people we are trying to help with our work.
If we believed that the Hussman method to be superior, we would adopt it in a heartbeat. Regular readers know that we follow a disciplined approach that reacts to market conditions. We represent the consumers of valuation methods, open to any approach.
Our Analysis
We have done thousands of regression models over the years, and taught the classes for grad students. What may make eyes glaze over for most is routine for us.
Here are some key thoughts, all of which come from old class notes:
- Think about the logic of the model. The key concept of the Fed model is that investors compare forward earnings to interest rates. In the pre-1980 days they did not have this information. More importantly, they did not have access to the Hussman model! The 1948 investors could not make the relevant comparison. They were not looking at past record earnings versus the prior five years or any of his other variables. We do not know what they were thinking. While we would love to have data about this era, we do not.
- Technique of model development. Let us suppose that we know about one period of time and wish to backtest a different period – the exact Hussman problem. We would divide the known period into two groups, developing the regression equation for one and testing it against the other. If the fit is not good, how can one make a forecast for the unknown period? Hussman did not do this, and we suspect that the correct method would generate a very different regression equation.
- The logic of each variable. Each variable should have an independent contribution, meaningful in theory as well as in improving the fit. In particular, we suspect the "record earnings" variables. If removing these variables creates wild swings — even swings in the sign –of the other variables, then the model lacks intellectual integrity. We suspect that this is true, but cannot prove it without the data.
- Optimizing parameters. Whenever we see something like "16 months" we say "hmm." The question is why this value was chosen and whether the variable is "robust."
Dr. Hussman’s work is not offered as a proprietary model. He is an advocate for his approach. Given this, he should be willing to share the data and invite alternative analysis.
Summarizing
Dr. Hussman’s quest is a good one. He is an excellent business person and advocate for his method. If he is confident of his results, he should be willing to have peer review of his work, they way it is done in academic circles.
Meanwhile, we do not see the advantages of this approach.
As a reader of both your blog and Hussman’s Weekly Market Comment, I was really looking forward to your response to Hussman’s criticisms of the Fed model over the last few months. Thank you very much!
Even if you use unadjusted Maximum Trailing Earnings as a proxy for Forward Operating Earnings (which, judging by the graphs, seems pretty correlated), rather than Hussman’s carefully fitted formula, I imagine you will come to similar conclusions as Hussman about the lack of fit of the Fed Model before 1980.
Maybe we truly can’t know if the Fed Model fit back then. Or maybe the investment environment was truly too different for the Fed Model, simple as it is, to work. It seems that both sides of the argument have valid points. Maybe this time things ARE different, but maybe not as different as some think.
Of course we can’t be sure one way or the other. Hussman makes many statements which seem at odds with a lot of prevailing notions. I don’t have the expertise to dissect some of his more controversial statements, e.g. the trend, not the level, of interest rates correlates (weakly) with future market returns. Or, the Fed is basically irrelevant (it’s government spending that affects the monetary environment)…?
I pay attention to Hussman because of his fund’s fantastic record at beating the market while maintaining the volatility and downside risk of a bond fund, something very unique for a mutual fund available to a layman like me. (Come to think of it, it would be very useful to see your overall risk-adjusted investment performance at NewArc.)
I’m anxious to see if Hussman will be able to maintain his absolute outperformance until the next bear market, and in an environment where growth possibly dominates over value. I have a feeling the bear market will be further off than he thinks. (I don’t know why some people can’t imagine that valuations could be permanently higher since the 90’s.)
Scott
Even though his regression fit well, there were two things amiss. One, how many models did he try before he published his model? Did he do a specification search? When I did my model, I did only two passes over the data, and the first was accidental because I didn’t have a lengthy corporate yield series. The Moody’s series is one of the few that goes back a long way, and Bloomberg did not carry it. I wanted to use BBB corporates from the start, but could not find a series, so I did one pass with Treasuries.
http://alephblog.com/2007/07/09/the-fed-model/
Second, after doing the analysis, the rest of his results rely on an extrapolation from the recent past to the further past. Dr. Hussman is the one who argues that the 80s are unique, but that is a large part of the data that he uses to estimate his backcast. No matter how good the fit, it is not safe to do extrapolations. Too many structural things change over time in capitalist economies.
I say these criticisms hesitantly, because I genuinely admire Dr. Hussman. We even live in the same town, but we have never met. So it goes.
When it comes to investing, many strategies work. In essence, the goal of the Hussman fund is appreciation with minimal drawdowns. As such, he hedges considerably. Nothing wrong with that if that’s what you want. As a long-term investor, a well-managed focused portfolio will outperform this strategy since hedging comes with a not so small price tag. Granted there are bigger swings, but as Buffett and others like him have stated, no one ever complains about volatility to the upside.
That said, I actually prefer Hussman’s method to dollar-cost averaging into the S&P, as Buffett suggests for the novice investor. The S&P takes some huge swings, and most investors simply do not have the stomach for it. Such people would probably be happier letting Hussman manage their money.
I do believe Hussman is not seeing a number of intuitive points about the Fed Model. As the Old Prof suggests, one of its most useful purposes is guaging investor sentiment. Imo, Hussman is a little bit of a numbers geek. The markets are as much of a psychological game as anything. Further, Hussman repeatedly says he isn’t trying to predict anything, which I believe is a mistake. It’s been my experience that the more one focuses on the future, the better one becomes at predicting what will come. Case in point, this housing debacle did not catch me or my clients offguard. (Though one wanted to hang onto his housing stocks when I advised him to sell them five months ago. They’re now down another 50%.)
Nice article. Hussman is obviously a very bright guy, but his dispatches are also obviously marketing material.
I also tend to suspect equations with non-obvious constants. For this reason, I don’t understand the appeal of his valuation method versus simpler ones like Andrew Smither’s model which is based on Tobin’s Q. This method calculates Q as the S&P 500 market cap versus replacement cost as provided by the Fed flow of funds data. Smithers’ model seems to avoid most of the criticisms that you make about Hussman’s model. It is simple, the data goes back a very long way, and Smithers claims it to be statistically predictive of future returns.
Given that Smithers’ and Hussman’s models seem to be given the same signal, I’d like to know whether you have looked into Smithers’ claims and what your opinion was.
As Chris above notes, Hussman has pointed out the similarity with Tobin’s results and therefore cannot be easily dismissed. At the simplest level, Hussman’s arguments are about mean reversion of earnings yields, something that even diehard bulls like Siegel admit to. Perhaps things are different since 1980 such as lower transaction costs, although the book value related arguments such as Q are quite compelling. With regards to the analysts and their expertise, I have here before me Malkiel’s book where he says that in a study by Sandretto and Milkrishnamurthi, analysts earnings estimates were off by an annual average of 31% over a 5-year period. This is also in agreement with the fact that forward PE averages to 11 while trailing reported PE averages ~15 as pointed out by Asness. Analysts estimates were further found to be even less accurate over one-year periods and including for stable earning industries such as utilities. Anecdotally, as senior management once told us — earnings visibility is very good for one quarter, not so good for two quarters and beyond that, it is just a roll of the dice. Maybe the analysts know something about companies that companies themselves do not.
How does the Fed model explain the performance of share prices in environments with extremely low interest rates? I’m thinking of Japan in the ultra-low interest rate environment of the last decade or so, or US share prices in the Great Depression and WWII period.
When long-term interest rates are 0.5%, doesn’t this imply a price-earnings ratio of 200 – yet the Nikkei never remotely approach this PE level. Similarly, when US long bond yields were 1.5% in the early 40s, the PE was no way near as high as the Fed model would suggest. Even if you add a risk premium to the bond yield, and assume zero earnings growth, the implied PE was still way above where the markets actually traded.
Unless I have missed something major, then the only conclusion I can reach is that the Fed Model doesn’t work. You cannot use a model which would have given horrendeously erroneous predictions for over a decade. And how well did the Fed model work from 2000-2003? Once again, its performance seems unsatisfactory.
Cutten –
You raise some good questions, which I think I have answered in prior articles in this series. Summarizing briefly, extremely low rates may be associated with concern over global deflation or some other event that threatens profits and security. No one believes that the fair value P/E multiple should be 100 if interest rates are 1%.
Determining the correct range to apply a model is a standard research problem. I used to give students a trick question where the correct answer was that the problem was outside the range of data used to develop the model.
Most importantly, if you read the articles on the series you will see that I see the major use of the Fed model has an objective way of spotting extreme investor sentiment.
Thanks for pointing out these concerns. Readers with a serious interest in this topic might enjoy putting “fed model” in our search window and looking at some older articles, including a review of criticisms of the model.
Jeff