PureBytes Links
Trading Reference Links
|
Chuck:
This is a very lucid post and I agree with your comments. How would you suggest
a person [that is relatively uneducated about trading] go about kicking the
tires on a trading system that they see advertised? Let's choose a system from
the webpage in your post.
Best,
Tim Morge
CRLeBeau@xxxxxxx wrote:
>
> I am a system vendor and would like to defend the idea that good systems can
> be purchased. This response is difficult to do without having it sound like a
> commercial message. I'll do my best.
>
> There are good systems for sale and there are bad systems for sale. It would
> be an error to assume that all are good or all are bad. It is also very
> difficult to tell the difference before you buy one. Remember that the real
> cost of a system is not the purchase price but the money that you might lose
> trading it.
>
> I have read on this list messages to the effect that you need to see x
> thousands of real time trades. This method borders on the rediculous if the
> system being evaluated is a long term trend following systems with five or six
> trades per market per year. You would have to start worrying about year 3000
> compliance if you took that approach. Even if you had thousands of actual
> trades you would have very little assurance that trades in the future would be
> profitable. The CFTC mandated warning that past performance is no assurance
> of future results is absolutely true and there is no way around it. It
> doesn't matter whether the trades are actual or hypothetical. To quote Yogi
> Berra: "The future ain't what it used to be."
>
> Some sophisticated system testers argue the case for complex out of sample
> testing methods. This is a good idea on the surface but it has its
> limitations. If the out of sample data is similar to the in sample data the
> test will produce good results. If the out of sample data is different than
> the in sample results then the test will show poor results. What is the data
> sample in the future going to look like? No one knows. Any attempts to infer
> too much from historical testing regardless of the methods used is dangerous
> and is similar to arguing about how many decimals you should round to on your
> WA guesses. (Ask Neal Weintraub for his views of backtesting if you want an
> earfull on this topic.)
>
> Using any system is a pure leap of faith in my opiniion. You may be better
> off relying on a system designed by someone with practical knowledge and
> experience than trying to do it yourself and learning by your mistakes. By
> the time you learn enough using that method you may not have any money left.
>
> For a sample of systems that I think are good, go to
> http://traderclub.com/
>
> Unfortunately there are no short term S&P systems listed there. Our best
> effort on that market to date turned out to be an accidental curve fit and
> will never see the light of day. Maybe I should post the code here and see if
> anyone can solve the curve fit problem.
>
> Chuck
|