|
PureBytes Links
Trading Reference Links
|
Thanks so much Dave, I value advice, opinions and feedback from experienced
traders ... I am not one of you yet but am aspiring to become one :-) give
me a year or two and heh, I hope to be able to shake your hand with
confidence :-)
Actually your post is full of good stuff, more than I want to reply to
without having given it some thought. So, if its OK I'd love to respond to
some of your points later.
take care,
Herman.
At 10:04 AM 7/3/01 -0700, you wrote:
>Hi Herman,
>
>Your described approach is very methodical, and you are raising some good
points about the
>pitfalls.
>
>From the point of view of just another user, I think that all approaches
will show one shortcoming
>or another when viewed critically. However, a combination or summation of
many approaches gives
>me a certain comfort level that I find hard to trade without.
>
>I will use the same approach that you outline below, but I will start my
in sample and out of
>sample testing differently.
>
>I start by looking at a certain market, and trying to break that up into
pieces with different
>character, e.g. trending, trading range, reversing, etc. So, if I'm
trying to build a reversing
>system using an oscillator(s), I'll want to select those segments of the
market that are reversing
>or trading in a range. Assuming I have enough bars, I can break those
into segments for in and
>out of sample testing.
>
>Once I've found a successful approach to the specific character of that
market, I can expand my
>testing to a larger portion of the historical data, but I'll need to find
a way to turn the system
>on and off depending on what type of market behaviour is being exhibited.
Then I'll run this
>overall system against the larger data set, in much the same manner as you
describe below.
>
>Yes, it's somewhat redundant and misleading because I'm reusing the same
data, although on an
>expanded scale.
>
>If this works successfully on a market (or a specific security), I can
make life easier by testing
>this approach across a portfolio. Certain markets are more likely to
behave a certain way, more
>frequently, for a longer duration. In other words, my system will test
better on those markets
>which tended to have greater periods of reversing or or trading range
behaviour, for example.
>Taking the S&P futures, you'll see this market is very choppy and tends to
reverse a lot. It
>*WON'T* be a good choice for a trend following system, but may well be a
good choice for my
>example system approach.
>
>A personal belief of mine is that market behavoiurs are persistent. The
S&P is much more likely
>to continue it's reversing behaviour in the future. I'm fairly confident
that next month, it
>won't become a trendy market.
>
>So by choosing those markets that have worked well in the past, I feel
this improves my chances
>for a system to work well in the future.
>
>A sidenote here is that I'm making certain assumptions about basic system
design. I subscribe
>heavily to the signal / trigger approach as well as Chuck LeBeau's methods
for developing robust
>systems. (www.traderclub.com) I really like adaptive rather than fixed
approaches. Also, you'll
>notice that markets look very different depending on the timeframe:
comparing a 3 minute chart of
>the S&P with a daily chart is a good example.
>
>Just some somewhat random ideas to go with your thoughts....
>
>Dave Nadeau
>Fort Collins, CO
>
>
>--- Herman van den Bergen <psytek@xxxxxxxx> wrote:
>> My system development sequence has always been (and I believe many others
>> work this way too):
>>
>> 1) Performe my first System Tests on historical data preceeding 2001.
>> 2) If the results look promising, I would perform a system test on more
>> recent, 2001 data.
>> 3) If that looks good too I may assume my system to be valid and continue
>> work on it, if not I modify the system and start all over again.
>>
>> So, each testing phase includes extensive historical testing. Isn't this
>> like putting the cart before the horse? After all, if my system doesn't
>> make money during the most recent six months, why would I be interested to
>> learn whether it earned money a long time ago? The fact is that if it
>> didn't show profits in recent months I am not going to use it no matter how
>> good it appeared to be in the past. Would you?
>>
>> So, unless somebody can straighten me out, I'll be, from now on, testing my
>> systems on recent data first and then, only if it shows profit in recent
>> months, will I verify it's performance on historical data to get to know
>> the system better.
>>
>> Looks like this will save me a lot of time... right or wrong?
>>
>> Happy development!
>> Herman.
>>
>
>
>__________________________________________________
>Do You Yahoo!?
>Get personalized email addresses from Yahoo! Mail
>http://personal.mail.yahoo.com/
>
|