The Inherent Problem in Stock Market Research

There is a dilemma in most stock market resesarch.

One the one hand, any theory is hungry for data.  It takes many, many cases to demonstrate a relationship with confidence.  Complicating this is the idea that one  should only look at part of the data, and use the rest of it for out-of-sample testing.  Those trying to develop trading systems have a basic awareness of this problem.  Those who operate as "market strategists" often do not.  They take all of the available data and draw a conclusion.

On the other hand, many problems have very few historical instances.  The study of market tops, market bottoms, Fed rate increases, and the like, have a relatively small number of instances even if one goes back decades for the analysis.

The problem?  The old data are simply not relevant.  Looking at stock market data from a hundred years ago is like looking at baseball statistics from the dead ball era.  Would you try to compare Ty Cobb to Barry Bonds or Albert Pujols?

When the world changes, the old findings may not matter.  Arguing that these data are relevant puts a major burden on those advancing the research, a burden that they typically ignore.

Here is a short list of things to think about:

• In the 70’s the options and futures markets developed.  Everything changed, as major players had large derivative positions.
• Decimilization and lower commissions have made it much easier for investors to switch positions.  The entire psychology chanced.
• Tax law changes have altered the incentives.
• ETF’s have people (investors, their advisors, and fund managers) trading in stocks where they know nothing about the underlying fundamentals.
• Program trading makes up much of the volume — over 70% in some reports.  What does this do to any technical analysis that uses volume as an indicator?
• Information has changed dramatically.  Even investors who do not know how to follow a particular stock have access to many opinions about it in a short period of time.  Do we really want to compare this with what people knew during the Taft Administration?

The conclusion is that many propositions cannot be effectively addressed through quantitative research.  There are not enough cases that are relevant.  Trying to do this anyway means that one is making a big decision based upon little information of dubious quality, another way of being Fooled by Randomness.