How (not) to Build a Market Indicator: The Hindenburg Omen
A recurring theme at "A Dash" has been the seductive power of charts and trading systems. Televised commercials from brokerage firms suggest that anyone with a computer can develop and test a system. One of my articles about this danger was the story of a rookie trader, who failed despite his intelligence and system testing. (Thanks to the Internet Wayback Machine for saving this story.)
In last week's market preview I suggested that people should know enough to dismiss stories like "The Hindenburg Omen." I got a number of emails from readers who asked me to elaborate. It is a topic that will not die as Barry Ritholtz shows with his discussion of the searches on Google Trends.
In this article I hope to provide a simple, non-technical explanation of why you should pay no attention to this "omen." I will do this by describing several important steps in building a system or indicator, and the perils of going wrong. I hope to demonstrate that it is nearly impossible to detect a flawed system, even for experts. As a result, even skeptics and fair-minded journalists have trouble showing what might be wrong with the Hindenburg Omen.
How (not) to Build a System
Here are the key steps to building a trading system or indicator that will look fantastic on paper, but be totally ineffective in practice.
- Start with the conclusion. Do not being by thinking about the problem or generating hypotheses, since those might not work. Instead, look directly at the dependent variable (what you are trying to predict).
- Take the result you seek on the dependent variable and run a correlation matrix with hundreds of possible "causes."This is easy to do with so many technical indicators available. Pure chance results in a correlation when you have so many possibilities.
- Find a plausible hypothesis. This is an easy task for smart people. In this article I showed how a group of graduate students at a leading university were tricked by a clever professor. People do not realize how easy it is to find "explanations" when given the answer unless they have had an experience like this.
- Start optimizing! Your original relationship will not be perfect. You can make it better by looking at the failing cases and getting rid of them. You can do this by adding new variables to rule out the "bad cases." You can get really precise with values, specifying a very narrow range that qualify. You can cleverly combine variables.
- Use as many variables as needed. This violates a statistical principle about "degrees of freedom" but who cares? The idea is that if you have very few cases and too many variables you are overfitting your model.
If you follow these steps, you have a system or indicator that looks excellent. Even an expert cannot prove it wrong from existing data. This is why the expert is interested in how the model was built, not just the final result.
Evaluating the Hindenburg Omen
After reading scores of articles on this subject, I think this is the best explanation of the development of the model and subsequent tweaking. If you look at the many parts of the "Omen", the carefully optimized parameters and the subsequent tweaks, you will see that all of the system development rules have been violated. The only one we cannot see for sure is the original "hypothesis" which referred to the divergence in new highs and new lows.
Even if you accept the system as stated, the key feature, the number of new highs and new lows, is flawed. Here is the take from Jeffrey Saut of Raymond James, referring to the current signal:
…(T)he vast majority of “stocks” making new highs were interest sensitive
closed-end funds, preferred stocks, or some other kind of fixed income
product, which by my pencil are not stocks. Therefore I'll say the same
thing I said two weeks ago, “I don't think a Hindenburg Omen has been
registered; and even if it has, its track record is spotty.”
CXO Advisory, one of our featured sites, does a typically professional job in analyzing this indicator. The technically minded should read the entire article and compare the analysis to the summary here.
Why This Story is Important
The hype around this topic is incredible. Conclusions range from those who see a near-certain market crash to those suggesting a 25% possibility. People are taking this very seriously. Art Cashin, who is the voice of the NYSE traders, talks about it on CNBC. The original Zero Hedge article, a typically uber-bearish scare piece, instantly went viral. I got multiple emails, including some from brokers at big-name firms. There is an eagerness to confirm the bearish bias in this summer of negativity.
Many observers who do not have specific training in research methods find the results to be very persuasive. Even if they sense a problem, they still see it as some kind of warning signal, since the correlation of the signal to the result is so clear. It is not a matter of intelligence or good intentions. Smart people are easily fooled on this one.
This problem is not obvious to people lacking specific training in research methods or system development. This can and does include mathematicians, engineers, and physicists.
The power tools of system development have outstripped the capabilities of the users, with a dangerous result.
It is now deceptively easy to create a trading indicator that is only useful for "predicting" the past.
Skeptical Observers, Great Opportunity
Concerning the Hindenburg Omen, several savvy market observers show skepticism — Barry Ritholtz (calling it "a backward-looking indicator that doesn't consider causation. He labels it 'recession porn,'" Josh Brown, Mark Hulbert, and Tobias Lefkovich.
I hope that this analysis has added some additional reasoning and explanation.
Meanwhile, for investors, I strongly urge a focus on the actual performance of corporations. While our official posture for trading accounts has been neutral or bearish for several months, our models have now started to identify good trading opportunities. For long-term investors the picture is even more decisive. There are many stocks trading at prices that value-oriented managers see as a generational opportunity.
It is important not to be sidetracked by pseudo-science.