|
PureBytes Links
Trading Reference Links
|
You have to be very careful with optimized systems. They can look great on
paper but many tend to fail once real trading starts. Analyze each trade
closely. Is there only one or two trades with tremendous gains while there
are many trades with many small losses? You need to see consistent profits
thru out the entire testing periods. Another thing to look for is
robustness. For example, if you optimize a system which uses a moving
average and the best returns are for a 14 day moving average, check to see
if you still get decent returns with a 9, 10, 11, 12, 13, 15, 16, 17 day
moving average. If you still get decent returns that is a good sign. If
the system falls apart when you try a 13 day moving average for example,
then it is not very robust. You also need to look at the maximum number of
losses in a row. For example it the maximum losses in a row is five, could
you have stuck with the system had you started trading just as the first
loss happened? Same with maximum drawdowns. Even though the system
returned 100% on average, could you have stuck with the system if it had a
drawdown of 40%? Although this is mechanical investing, very few can
suppress their emotions when real money is involved. A lot of second
guessing goes on. Moral: optimized systems are not the holy grail you
think they will be.
Rick
-----Original Message-----
From: owner-metastock@xxxxxxxxxxxxx
[mailto:owner-metastock@xxxxxxxxxxxxx]On Behalf Of Herman van den Bergen
Sent: Monday, May 27, 2002 8:37 AM
To: metastock@xxxxxxxxxxxxx
Subject: Re: QQQ System Test
At 03:45 PM 5/25/01 -0700, you wrote:
>>... Just curious how this compares with other systems
>> trading QQQ.
>
>first you need to only include your out-of-sample
>performance. Good would be 100% / year. Above means
>nothing.
Hi Jack and other readers,
Sounds like you have some experience in this area - would you care to share
some?
When I run a System test on the QQQ, say for the period 11/3/99 - 11/24/00,
I get a long list of results. One for each opt combination. In my case I
get 164 results that give me >100%.
Judging by the percentage gain a large number of those "winners" work just
fine on out-of-sample data for the period 11/24/00 to today.
How do I select the best system from many "good" performers, I mean
something better than just looking at the linearity of the equity charts?
Are there other ways to rate/test a system for future performance?
Thanks for any comments you may have,
Herman.
|