|
PureBytes Links
Trading Reference Links
|
As a position trader, I would require satisfactory results across at least a
couple of 5 year periods including a number of major declines and a bear market
and I want to find comparable results across each of those periods. Using
sub-day charts you might very well accomplish the same validation by using a
couple of years broken up into samples which include all types of markets.
When evaluating systems across history which includes a large variation in price
range, e.g. 30 years of S&P, I place little or no emphasis on dollar winnings
because a few trades at the very high prices near the end of the series can skew
the whole thing. With any system one of the first things I look at is consective
losses because I know I'm unlikely to trade a system against a long losing run.
Another item I look at (and frequently omtipmize) is the average trade because
that, when multiplied by the position size tells me if it's worth my time to
trade the system. One other thing I do is to optimize the long and short sized
individually. Very few systems trade both sides well and I want to know if a
system is strong or weak on each side - some systems are suitable for one side
only. Final thing I look at are my system variables. I start with system
variables which "make sense" to me, then run various optimizations with
relatively large increments e.g. 9 to 39 by 5. I want a system which performs
well across a range of variables should perform which make some logical sense
e.g. if a system works best at 19, it should do ok with 14 and 24. In a
nutshell, I use optimization to learn about the sensitivites of the system
rather than to fine tune to highest profit.
And finally, when the variables have been chosen, there is no substitute for
going through the run trade by trade to learn more about how a system trades and
where its strengths and weakness' lie.
Earl
-----Original Message-----
From: Logan467@xxxxxxx <Logan467@xxxxxxx>
To: omega-list@xxxxxxxxxx <omega-list@xxxxxxxxxx>
Date: Tuesday, February 24, 1998 10:12 AM
Subject: System testing
>On another subject, when you test a system, what criteria do you use? Is it
>the percentage of winning trades, Ratio of average win/ average loss, profit
>factor, return on account or a combination of these and others? What should
>the numbers be for these criterias for the system to be considered a "good"
>system? And how long should one back test the system for the system to be
>consider "robust"?
|