Periods for Walk Forward optimization

[khoga]

So I've spent a good part of the weekend watching youtube videos on walk forward analyzer. I get it, though I do think a good basic strategy should generalize well to an environment for more than a period or two

My opinions aside, I've seen some conflicting information on the WF opt on how it uses periods.

For example's sake, let's say we have two year's worth of end of day data for the optimization (e.g. 1/1/2009-12/31/2010). Let's put the optimization period at 2 months and the out-of-sample period at 1 month.

A couple of questions:

- if i started the optimization from 1/1/2010, not 2009, would the WF use the last two months of 2009 for the opt and the first month of 2010 as the out of sample OR would the WF use the first two months of 2010 for the opt and the 3rd month of out of sample?

- lets assume that we optimized on months 1 &2 and tested on month 3, for this next question. for the next optimization period, does WF use months 2&3 to test on month 4 OR does WF use months 4&5 to test on month 6 OR is it months 3&4 to test on month5?

The first method gives much more feedback on a "real life" scenario where you'd retune at the end of month 3, but the more I'm reading/watching the more I get the sense that it's one of the last two scenarios....which would be a shame as it wastes alot of data that could otherwise be used.

I really appreciate your help in clearing this up for me!

Thank you,
Tim

[NinjaTrader_RyanM1]

Hi Tim,

The walk forward start date is the start of the first optimization period. It won't look at bars prior to the start date, but you can adjust the start date back if that's what you're going for.

[khoga]

start date

so, in my example, the WF would look at months 1&2 for opt and then test on month 3, correct?

what about the second iteration of the WF? can you help me understand the second question i had asked?

Tim

[NinjaTrader_RyanM1]

It's a continuous process: Optimizing parameters > using those parameters in strategy (testing) > Optimizing parameters...

The length of time spent optimizing is the Optimization Period, and the length of time testing is the Test Period.

Sorry, I'm not sure what is your second question. Can you please clarify?

[khoga]

second question

- lets assume that we optimized on months 1 &2 and tested on month 3 (which we've just confirmed), for this next question.

for the NEXT optimization period,
1) does WF use months 2&3 to test on month 4 OR
2) does WF use months 4&5 to test on month 6 OR
3) is it months 3&4 to test on month5?

1 is optimal, but it's not really what I think i'm seeing...though that could be due to the periods i've used thus far.

[NinjaTrader_RyanM1]

I believe it would be 1. The test period uses the optimization parameters for the optimization period prior to its start. If it's still unclear, please post a screenshot of some tests/settings you're running on SampleMaCrossOver. Make sure to include both Optimization Period and Test Period visible so we're working from the same example for reference.