Can technical analysis beat the market? Part III, How CASTrader will use it

In the words of Neely in one of the Part I papers, "First, profit opportunities will generally exist in financial markets. Second, these profit opportunities will tend to erode over time in response to the forces of learning and competition. Finally, more complex strategies will persist for longer than simple ones."  Neely is describing relevant predictions of the Adaptive Market Hypothesis (AMH) I discussed in Part II.  CASTrader will be the ultimate participant in our adaptive markets, being composed of thousands or millions of adaptive agent traders.  Under the AMH, CASTrader will reflexively adapt along with the markets and hopefully continuously innovate profits.

Below are the principles for implementation of technical metodologies in CASTrader based on the above:

1. Methodologies should be unique.  In order to maximize the chance for success, CASTrader methodologies will favor unknown or unpublished methodologies that have less chance of having been previously exploited.
2. Methodologies should remain proprietary.  There is no reason to give away the methodology, because others can then exploit it.  It seems there is no end to academic studies of methodologies that stop working once they are published.
3. Methodologies should be constantly innovated.  Profits will disappear as methodologies get exploited.  CASTrader must be capable of finding new methodologies to exploit.  Genetic methods offer a nice way to accomplish this.
4. Methodologies must be capable of evolving to high levels of complexity.  More complex methods offer less chance of exploitation and according to AMH, will persist longer than simpler ones.  Again, genetic methods offer a nice way to accomplish this in combination with adding other classes of analysis such as fundamental, economic, demographic and other models.  Technical analysis will be added first, simply because it is the easiest.
5. No data snooping.  CASTrader must be developed using historical data, and then, if it appears to perform, go live using real money.  It's crucial that the apparent performance at that point is not data snooping fool's gold.  I believe the use of the Kelly Formula provides a form of natural protection against data snooping in that the Kelly Formula instantly adjusts capital allocations to methodology performance.  Phantom methodologies lose capital, real strategies are allocated capital in an optimum way.  Other means of avoiding data snooping are the subject of a later post.