[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[EquisMetaStock Group] Fundamental Design Flaws in v8.01 System Tester.



PureBytes Links

Trading Reference Links

Refer to "Bugs in v8.01".

Version 7.22 System Tester was designed to produce the algorithm 
parameter values that would maximize the return for a given stock 
over the backtest period.  If one wanted to ensure the algorithm 
would work for a portfolio of stocks, the process was repeated for 
two or three dozen stocks, the parameter values averaged, and then 
used in the final model as fixed values.  The algorithm was then 
tested against the set of stocks used in its development to validate 
algorithm performance.

The algorithm developed in this manner with its fixed coefficent 
values had two characteristics.  First, it could be relied upon to 
produce reliable results for an extended period of time, and second, 
it could be relied upon to give reasonable results for stocks other 
than those used in the development phase.  That, of course, assumes 
the fundamental algorithm was cleverly designed.  

When version 8.0 arrived on the scene, it was assumed by previous 
users that the new System Tester had been designed to perform the 
process of averaging the coefficent values automatically to achieve a 
general purpose algorithm.  This would relieve much of the drudgery 
associated with running the system tester on stock after stock and 
manually analyzing the coefficient values. Unfortunately, this was 
not to be.

Instead, the new System Tester simply adjusts the coefficient values 
to maximize the portfolio return for that particular point in time. 
However, if that same approach is applied next week, next month, or 
next year,it will generate significantly different parameter values 
each time. If a portfolio of different stocks is "optimized" the 
parameter values will be different than those for the original 
portfolio. In short, one cannot use the parameter values on a 
different portfolio or at future times and have confidence that they 
are producing reliable results. 

If one wishes to use v8.01 to produce a statistically valid set of 
parameter values, then that person must go through the same procedure 
used with v7.22 of running a collection of stocks, one stock at a 
time and then averaging the parameter values.

The v8.01 System Tester is designed to simulate real-time action by a 
broker handling orders.  This it may do, but the results are totally 
useless, because they don't apply to the future, nor to a different 
portfolio.  They give the trader no reliable guidance as to what to 
buy and what to sell.

If this is hard to understand, then let me explain it with a 
hypothetical example. Consider a portfolio with two stocks.  You have 
just run the the v8.01 System Tester and now have a set of indicator 
parameter values which you apply to your Indicator Builder functions 
and Expert Advisor.  The next day one stock triples in value and the 
other falls 80 percent in value. You run the System Tester and find 
that the parameter values are now significantly different.  Do you 
use the old values or the new for the future? You now decide to sell 
the loser and replace it with another stock. You rerun the System 
Tester and find a third set of parameter values.  Are these valid for 
future use in the Expert Advisor and the Indicator functions?  Should 
you now change your Expert Advisor and Indicator Builder functions?
The answers are no and no. If the indicator parameters had been 
obtained by averaging parameter values for two or three dozen stocks, 
then those parameter values would be the same in all cases in this 
example.  

In simple terms, the current System Tester is the equivalent of 
shooting an arrow at a wall and then painting a bulls-eye around the 
arrow stuck in the wall. 

There is a way to modify the Metastock 8.01 so that it simulates a 
broker.  This requires two things. First, the System Tester needs to 
be reconfigured to produce statistically valid coefficient values 
automatically.  In short, simply upgrade the v7.22 System Tester by 
providing a wrapper that manages multiple stocks and the statistical 
analysis.  Second, create a Broker model that simulates the functions 
of a broker using the statistically valid algorithm provided by the 
System Tester.  This would then allow the user to vary the 
instructions to the Broker in a way that provides the optimum 
performance.  The Broker model could be back-tested, just like the 
System Tester.

The parameter values for the algorithm could then be updated 
periodically and applied to the Broker, the Expert Advisor, and the  
Indicator Builder functions.

As an after-thought.  It would certainly be nice if there were a user 
definable data array, similar in principle to the Indicator Builder, 
where algorithm parameter values could be written one time and then 
used automatically by the Expert Advisor, Indicator Builder 
functions, and the Broker.


------------------------ Yahoo! Groups Sponsor ---------------------~-->
Buy Ink Cartridges or Refill Kits for your HP, Epson, Canon or Lexmark
Printer at MyInks.com. Free s/h on orders $50 or more to the US & Canada.
http://www.c1tracking.com/l.asp?cid=5511
http://us.click.yahoo.com/mOAaAA/3exGAA/qnsNAA/BefplB/TM
---------------------------------------------------------------------~->

To unsubscribe from this group, send an email to:
equismetastock-unsubscribe@xxxxxxxxxxxxxxx

 

Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/