Page 1 of 1

using ggo and gwfo

Posted: Wed Nov 22, 2006 9:20 pm
by kot
Hi,
I am a ggo and gwfo user. i got the license a few months before Wouter
sold the company.
I am still trying to figure out if the whole process of developing a trading system using the ggo and basing it on the methodology set by Pardo in his book "Design, testing and optimization of trading systems" is a reliable process.
From my experience - some of the strategies stopped working immediately after the Oos period ended and some continued working.
i was not able to define the reasons why. all my strategies had at the most 5 parameters that were optimized.
i would really be happy to learn from other peoples experience.
thanks
Kot

Posted: Thu Nov 23, 2006 9:13 am
by michal.kreslik
What was your OOS testing period size in relation to the IS period?

Michal

Posted: Fri Nov 24, 2006 7:13 am
by kot
i use the default settings of 70% in sample and 30% out sample.

Posted: Fri Nov 24, 2006 9:17 am
by michal.kreslik
Ok, and what was the number of trades generated during the in-sample period?

Michal

Posted: Fri Nov 24, 2006 8:42 pm
by kot
about 30.




the images show the graph chosen by me in the ggo and the graph with real trading in tradestation.
thanks
kot.

Posted: Fri Nov 24, 2006 10:42 pm
by michal.kreslik
30 trades isn't statistically significant quantity. The probability that the results were obtained due to chance alone decreases with the statistical significance.

Would you infer after three heads coin tosses that the coin is biased?

Michal

Posted: Sat Nov 25, 2006 8:25 am
by kot
sorry, my mistake. in the in sample period there were about 80 to 90 trades and in the out of sample period there were 30 trades.
kot.

Posted: Mon Nov 27, 2006 7:24 am
by kot
i would like to rephrase my question.
Lets say you got a terrific result in GGO - you have more then a 150 trades in the in sample period and 75 trades in the out of sample period, the graph shows that the two regression lines are converging and the unseen grad is between 70 percent to 130 percent. What do you think are the odds for the strategy to work real time? and When will you know if the strategy stopped working?
kot