anyone using ninjatrader's walk forward tool? i understand the basic concept and the importance of out of sample testing, but i am not grasping the functionality of the tool. i was hoping someone with working knowledge of it can translate.

Hi Caprica,
Yes I use it extensively. That is the only tool to use. Optimizer by it self is use less. What it does it actually runs optimizer and then back test on consecutive dates and then advances the optimizer by back test period and run it again and again from the end of optimization period runs back test.
Example: (dd/mm/yy)
From: 23/02/09
To: 16/08/09
Optimization Period: 21 (days)
Test Period: 14 (days)

Optimizer Back Test
23/02/09-15/03/09 16/03/09-29/03/09
09/03/09-29/03/09 30/03/09-12/04/09
23/03/09-12/04/09 13/04/09-26/04/09
06/04/09-26/04/09 27/04/09-10/05/09
20/04/09-10/05/09 11/05/09-24/05/09
04/05/09-24/05/09 25/05/09-07/06/09
18/05/09-07/06/09 08/06/09-21/06/09
01/06/09-21/06/09 22/06/09-05/07/09
15/06/09-05/07/09 06/07/09-19/07/09
29/06/09-19/07/09 20/07/09-02/08/09
13/07/09-02/08/09 03/08/09-16/08/09

The only drawback is that for each test (optimization or back test) it needs to collect data to plot the indicators. That means that if you use MA of length 80 on a 5 minute time frame on Pit hours, it won't trade the first day in each test. (Each trading day has 81 bars)!
Thats why I approached Piersh to try to correct it, but because he couldn't help me, I'm doing it manually. I run optimizer, then test, then advance the optimizer and run test again.

Baruch

baruchs thank you i would very much like your help with this.

so lets say i want to backtest 60 days of data say from 6/25/2009 to 8/25/2009 (today). and i want to use walk-forward to test out of sample data. can I do this?

or do i have to give up some historical data and not use all sixty days worth? in other words lets say the most historical data i have is going back to 6/25/2009 and nothing prior, and i want to use all available historical data, is this going to work or not?

Hi Caprica,
No, you don't understand it correctly. Walk Forward == Optimize + Back test.
The idea is to optimize your strategy on a given data and then test it on next period.
This way you will see if your strategy is really profitable. Optimizer by it self creates curve fitting, but if it is profitable on the next data then its great, if not then its not so great.
Optimizer: is making lots of back tests with different parameters values, and picking the best values. If you run back test with same values on the same period you will receive same performance, but its worthless.
The bottom line is that you need much more data for testing, and I don't understand why you don't have at least 1 year of data.

Baruch

baruchs 

Hi Caprica, No, you don't understand it correctly. Walk Forward == Optimize + Back test. The idea is to optimize your strategy on a given data and then test it on next period. This way you will see if your strategy is really profitable. Optimizer by it self creates curve fitting, but if it is profitable on the next data then its great, if not then its not so great. Optimizer: is making lots of back tests with different parameters values, and picking the best values. If you run back test with same values on the same period you will receive same performance, but its worthless. The bottom line is that you need much more data for testing, and I don't understand why you don't have at least 1 year of data. Baruch

hi baruch i think i understand it now. i dont have a 1 year tick data and my strat is tick data not minute data. i guess i could buy it but that just creates a bunch of new problems because of continuous contracts and etc with ninjatrader.

thanks for the info.