|
PureBytes Links
Trading Reference Links
|
> Optimization should be banned, even though past trading
> patterns can be
> used as a barometer of the future, optimization trading
> systems will not
> and cannot repeat their past performance in real time.
Please, folks. A distinction needs to be made between two definitions.
Optimization: For trading system signal generators, a means of
adjusting (often mechanically or automatically) variable parameters such
that during execution on historical data, the greatest net result from
those data can be determined regardless of the behavioral changes that
occur within the signal generator.
Optimization: For systems of any kind, a process of evaluating
different configurations of systematic componenets to discover the best
possible behavioral result of the system. "Best possible behavioral
result" is defined as part of the system's objectives.
I know that I'm being picky, but this is a subject near and dear to my
heart. The first definition is generally what people think of when they
look at Tradestation's ability to re-test a signal generator using many
different moving average values in order to get the best summary report.
I would heartily agree... no, that's an understatement... I would swoon
with emphatic somethin' or other that this is silly, unproductive and
doomed to failure.
The second definition is ESSENTIAL in system development and testing.
Yes, I shouted. If you do not optimize a system in any way, then you
make either or both of the following assertions:
a) The first thing I come up with is the best possible thing.
b) Once a system is created, any attempt to improve it will result in
systematic failure.
There's also (c), "I don't have a clue how to create and evaluate
systems", which is okay and a much better answer than (a) or (b), both
of which are jest plane rong. Successful systems of every kind should
have feedback loops, where stuff goes into the system, gets changed,
comes out the other end, and the droppings are examined to see if the
system is behaving in a satisfactory way. If problems are evident, it
may or may not be of benefit to fix them, but it's always worth
evaluating.
Optimizing a trading system is a fine and necessary thing. I strongly
believe that most people do Definition #1 optimization, though, because
they believe that the "best possible behavioral result" (whether they
consciously think that way or not) of a trading system is to return the
greatest possible profit across a period of time. Using Tradestation to
crank through a bunch of oscillator values is evidence of this belief.
But I believe in different things. I say the best possible result of a
trading system is:
a) To behave as consistently as possible over time through a variety
of different types of market motion; and
b) To achieve a satisfactory level of profitability over time while
accomplishing (a).
"Satisfactory" is different for every person. In my reality, if a
system behaves in a consistent way in trends and chop and doldrums and
wild spikey psychosis, then it's probably going to behave consistently
into the future because that's what markets are likely to do in the
future (trend or chop or sit there or go schizoid). If it's profitable
while it's behaving consistently, then we're all set to give it a try.
Wrap it up, Sticks. They're all asleep, now.
So how to optimize, then, if testing different indicator values is bad?
Aha! It's not the testing of different values that is evil. What's
evil is looking at the summary report as you do it rather than the
behavior of the system. If Tradestation tells you that an 11-day moving
average returns the greatest profit over a ten year period, what do you
know about the system's behavior? Nothing. All you know is that your
data set worked best with an 11-day MA, but you don't know WHY it was
the best fit.
A "trading system" is not a canned signal generator, as most of the
folks on this list already know; consequently, optimizing a canned
signal generator in Tradestation does not necessarily (and isn't likely
to) mean that you have optimized your trading system.
Bong! Bong! Gee, look at the time. Must be getting back to being a
broker, now; I'm not a process analyst after the clock strikes twelve.
The "Reverse Cinderella Syndrome", or something. Once, I was a
princess. Now, I'm just "one of *them*"...
Sticks
--
Troutman, Defender of Sticks troutman@xxxxxxxxxxxxxxxxxxx
(aka) http://www.defendercapital.com/
Jonathan Matte, President No banner ads, no cookies,
Defender Capital Management, Inc. no sekrit sniffers!
Introducing Broker, Commodity Trading Advisor
|