|
PureBytes Links
Trading Reference Links
|
At 04:58 PM 4/28/98 +0100, Bill Eykyn wrote:
>
>Can you tell us any - repeat any - system you have so tested, in the manner
>you have so stated, and what the results have been.
>
>
>
I have never tested any commercial systems, only ones that I have created.
I would encourage traders to develop their own systems. When doing so, I
would offer several suggestions.
Keep it very simple. Do not use more than one or two indicators. Every
rule must be clearly defined in black and white. Someone else should be
able to test/trade the system. Focus on the long term. I would suggest
working with weekly charts, which poses another problem in itself. One
must get charts that are contract by contract in weekly format. You do not
want continuous; you must look at every contract individually and roll your
position just like you would in real time. I would try to roll quarterly
or with a March-July-December format as the contract allows. Also roll the
same day every time with SET rules for situations in which the new contract
would be flat or have the opposite position etc. My day has always been
the last trading day before the deliverable month. I.e. if I were in May
grains currently, I would roll on Thrusday, April 30, on the close. Test
in the walk forward mannor, because if you want to modify your rules
mid-stream, you need 'virgin' data on which to test it. The old data on
which you based your rules is thrown completely out. Run the various
statistical tests, z-score, standard dev of returns etc. at points through
the testing. As the data set becomes bigger these numbers for the entire
set should stabalize and the additional data should not deviate by more
than two stddev. I have read where 30 trades is enough, but I have found
that number MUCH too low. I usually look at 500 or more. When you test
write down: entry date, exit date, entry price, exit price, and initial
risk. Do not look at position size yet. After you have finished the
method testing, you must work your money management, which will determine
position size.
You will find, as you test a system, that you see the same trades over and
over again. Each market blends together and the process is very
repetitive, which is good. For my ego and weak human brain this helps
'prove' the validity of technical trading. I am a firm believer that this
intangible gives one more confidence in his/her creation; therefore, this
testing should be done by HAND. Hand testing will also keep one from
optimizing (read as curve fitting) the system, as every 500 trades will
take 30-35 hours to test and evaluate. Another 40 hours will be needed to
test and evaluate money management. Remember also not to curve fit the
money management.
Scot Billington
|