Interesting column by Mark Hulbert in the Sunday NYT about an econometric model I have observed over the years and have been impressed with.
What makes it so interesting is the deceptively simple system for assessing risk versus reward in the market place. It does so by tracking just three items: Stock market Dividend Yield, interest on the 90-day T bill, and the median 3 Year Value Line analyst earnings projection.
Taken together, they forecast trouble for the next 6 quarters:
"The model arose from research conducted some 15 years ago by William
Reichenstein, a professor of investments at Baylor University, and
Steven P. Rich, a finance professor there. They reported their results
in an article in the summer 1993 issue of The Journal of Portfolio
Management.The model is quite simple, especially when compared with many
econometric models. It has just three ingredients: the stock market’s
dividend yield, the interest rate on 90-day Treasury bills and the
median of projections from analysts at Value Line, the investment
research firm, of how much the 1,700 stocks they monitor will
appreciate over the next three to five years. The first two numbers are
readily available at many financial Web sites, and the third is
published weekly in the Value Line Investment Survey. The formula for
combining the three pieces of data can be easily figured with a
spreadsheet program or a calculator.Despite the model’s simplicity, the professors found in back testing
over the period from 1968 through 1989 that its periodic readings had
done an impressive job of forecasting the stock market’s gains and
losses over the subsequent six calendar quarters."
What makes this so fascinating is that the three elements are unbiased and uncorrupted: the market’s dividend yield is real money paid by companies to shareholders, so it cannot be phonied up (even the dividend tax cut only had a minor impact on it); the 90 day T bill interest is also a market determined element. While it is impacted short term by the Fed, it is ultimately set by the bond market; Lastly, the median of projections from analysts at Value Line are free from Investment Banking and/.or Marketing pressures (and other stupidity). Value line has an outstanding track record, and they are completely unbiased and objective, untainted by sellside conflicts.
Here’s a chart of the timing model’s forecasting history:
>
click for larger graph
graphic courtesy of NYT
>
The model forecasts that "the market is likely to underperform garden-variety money market funds through the end of next year." That gloomy forecast is quite consistent with out expectations.
Hulbert notes that the market-timing model "while
not perfect, has had an impressive track record over the long run." He writes:
"To be sure, the model has had a mixed record in the 13 years since their study appeared. Though the model has performed well in the current decade, its record in the 1990’s was poor. Through much of that decade, it projected below-average performance for the stock market, thereby greatly underestimating equities’ actual returns.
The model’s failure in the 1990’s, however, may be the exception that proves the rule. In an interview, Professor Reichenstein contended that the stock market’s outsized returns in that decade were in large part attributable to investor "irrationality," and that the model should therefore not be faulted for failing to forecast them. The model aims to forecast what the market’s level would be if investors were rational, and "no model built on rational pricing is able to explain irrational behavior," he said.
A study by The Hulbert Financial Digest provides further support for the notion that the model’s failure during the 1990’s was an anomaly. The study focused on its performance from 1968 through 2006 — a period that includes the 22 years covered in the professors’ original study and the 17 years since. Even after the incorrect forecasts in the 1990’s are taken into account, the model’s overall record is good enough to be statistically meaningful and not likely to be mere luck."
My only issue is that the time period involved is kinda short and limited — it covers most of one bear market (1970-82) and all of the next bull market (1982-2000). Valueline started in 1931, so it would be interesting to see how this system did during the prior post WWII Bull market — especially given the model’s miss during the late 1990s moonshot.
That might provide some insight into whether the system misses these types of strong markets. Or, it might just suggest that the 1995-2000 period was extremely aberational. Either way, it could provide insight into that one predictive failure.
>
Source:
An Old Formula That Points to New Worry
MARK HULBERT, Strategies
NYTimes, June 18, 2006
http://www.nytimes.com/2006/06/18/business/yourmoney/18stra.html
It’s easy to see why the model “missed” the bull market of the 90’s if it doesn’t include any element of human behavior. When belief systems converge in a durable way the market can move in a direction not expected by quantitiative models like this.
The problem as illustrated by dramatic failures like LTCM is that these models work really well, until they don’t. Often the change is that beliefs of market players shift and hence break the math.
I remember at the peak in Mar. 2000, the adjusted Value Line forcast foresaw a flat market for the next 5 yrs (2000 – 2005). The subsequent round-trip was all too accurate. Actually, the proper use of the raw VL forecast is to subtract 55% from their 5 yr forecast to compensate for their analyst”s optimism.
Barry, I don’t know how these charts are supposed to be evaluated, but the divergence between the predicted and the actual seems significant in the late 1970s and parts of the 1980s. What’s the definition of a “major failure” here?
“…Professor Reichenstein contended that the stock market’s outsized returns in that decade were in large part attributable to investor “irrationality,” and that the model should therefore not be faulted for failing to forecast them…”
Oh, ok. So the model works when investors are “rational” and fails when they aren’t. Does that make investors simply stupid some of the time for not behaving like the model predicted?
So what do the stats say? What is the correlation? The graph looks nice at first, but upon a closer look there are lots of gaps. It looks like there are several instances where the market fell before the predicted performance lines did. Even a stopped clock is right twice a day.
isn’t that just the equity risk premium v t-bills?
earnings growth + dividend yield = return (assuming no change in multiples) v risk free t-bill rate is risk premium..
the failure probably can be explained by the change in multiples which is not captured by those 3 variables.
i kind of think their is always some sort of irrational behavior somewhere in the market.. if there wasn’t there would be no opportunity
I don’t trust models that are only right on the curve-fitted portion that was available before the model was released. If you vary the parameters of the model a bit you can probably make it fit better. It also doesn’t look like the fit is very good.
I can believe that this is a model of where the market would be if everyone invested rationally. Every time there is a major divergence, the market eventually corrects. You can see the large unpredicted overpricing prior to the 87 crash, another one leading to what looks like a crash in 91, and the whole internet bubble.
An interesting indicator, but you would have lost money using it in the straight-forward fashion. It looks like it would have issued a buy signal in the spring of 2000 after suggesting that you sell from 1995 on. If you bought every time it crossed above 1.5 (which corresponds to 0% return on the graph), you would have managed to buy at the peak almost every time. Perhaps this should be used as a contrarian indicator. In which case it is suggesting that the market is going to continue in a bull market.
I don’t buy the “irrational” argument either. Instead, it looks to me like this model either misses an important variable that changed or accellerated its pace of change after the mid 1990s (being very generous), or it is simply a case of a [four? five?] parameter model selected out of a variety of potential models on the basis of its ability to fit a historical time series (i.e. “back testing” really means “adjusting the parameters to produce the best possible fit”), and failing badly to capture the behavior of the system being modelled from then on.
Because of the ability to select from a number of models and adjust parameters to fit a historical data series, you really need to look at the performance of the model from the time that the model and its parameters were fixed – i.e. when it was published – forward.
Published in 1993, not even close to accurate from 1995 onward. No offense to the modelers intended – it was a worthy effort – but that doesn’t inspire confidence.
“…misses an important variable that changed or accellerated its pace of change after the mid 1990s…”
The World Wide Web, for example.
This model Barry popped up is not a trading model. It’s an anecdotal model to add to your tool kit. And, as this model shows, this is a very dangerous market and I don’t see anything that leads me to believe a bottom is set. I keep waiting but it hasn’t happened for my work. It could set soon but………now it appears regardless of what sentiment says, every Tom Dick and Harry is expecting a rally.
The Transports and Small Caps are still at levels that are as out of whack as Big Tech was in 2000 by many measures. If we go to higher highs on the same themes, this is simply going to make the downside pain even more brutal IMO.
“There were no headlines last week announcing it, so most stock-market bulls are unaware of the good news. But a market timing model with a good long-term record has flashed the all-clear signal to re-enter the market.”
“The model is based on a statistic reported each week in the Value Line Investment Survey… After being cautious about the stock market for some time, the model based on this statistic is now forecasting healthy gains.”
– Hulbert talking about the same indicator, New York Times, November 5, 2000
http://www.nytimes.com/2000/11/05/business/05STRA.html
Put another way: it’s a data point, but I wouldn’t trade on it.
wcw, that is hilarious. I guess Hulbert is an NBA coach. ie, Recycled into new opportunities regardless of his long term effectiveness. There is some validity to the model but………………not to trade off of.
Hey, I guess Hulbert would have us in cash from 1997 to 2000 and buying then at 2000 for the next 80% dump.
Too cool!
So what’s the model or equation?
How do I, the average investor, apply it to the present day statistics in real time?
Edward Tufte at Junk Charts has a very interesting critique of Hulbert’s chart:
Hulbert “took the 13 years of poor performance, and mixed in 26 other years, of which 22 were used in constructing the original model. This is equivalent to averaging training accuracy and validation accuracy. (Training data is used to build a model; validation data is used to test its performance on previously unseen data.) Training accuracy is always high or else the modeler would have rejected the model; thus, no honest one would make claims based on training accuracy. Measured on validation data alone, this model is most definitely foul”
Read the rest: Illusion of success.
As GRL asked: what *is* the model? I thought it was kind of strange for the NYT to write about the inputs and the predictions, but to say nothing about the algorithm. Is it proprietary?
From the article nor the chart displayed above I can not deduce a trading model. When I first read this article I though the strategy depicted was poorly chosen as it seemed a classic case of curve-fitting. I couldn’t even understand from the article why the author thought it was successful in the past!
From the article nor the chart displayed above I can not deduce a trading model. When I first read this article I though the strategy depicted was poorly chosen as it seemed a classic case of curve-fitting. I couldn’t even understand from the article why the author thought it was successful in the past!
Models are blind.
This one’s 1990s failure has a lot to do with the inability to ‘see’ what some have termed “stock market Keynesianism” that kicked in with the 1995 reverse Plaza Accord and then the Asian Crisis. A strengthening dollar followed upon by a collapse in the most dynamic economic region, and liquidity flooded into the U.S. – assisted, of course, by AG’s 1995-99 interest rate policies.
Often it’s forgotten that profit – not earnings – of production capital in the U.S. began falling so early as 1997. The last leg of the bull was not driven by profit.
A strictly rational market would have fallen away rather than bubbled up, but ‘strictly rational’ is soley a province of theory.
A blink at the stock market
The Blog The Big Picture picked up on a fascinating research published in the NY times. It describes a very simple econometric model from Professor William Reichenstein, that predicts the stock market 6 month ahead.
Thats stuf that…