Understanding How to work with Walk Forward

[freewind]

Hi,

I'm trying to use the Walk Forward feature but I don't fully understand how it works.
The tutorial for this feature is very basic and I'll deeply appreciate if some one can guide me on how to use this feature.
For example I have a strategy that have 5 parameters which I optimized over the last 2.5 years. How can I check them with the Walk Forward feature? For what I need to set the Test Period and Optimized Period and what is the meaning of the results I'm seeing?

Tx in advance!

[NinjaTrader_Bertrand]

freewind, it's basially designed to offer insight into how your parameters optimizations are holding up as you continue to optimize and trade this strategy across the data set you have available. You would for example optimize on 3 months of data and then use those 'best' paramters found to trade the next month, after this step you would reoptimize the strategy again on the past 3 months of data and use the fresh found 'best' paramters set to trade the next month and so on until you're through the dataset. If you then evalulate the various test period reports you would see if the strategy is holding up on 'unseen' data or not.

[freewind]

Thanks Bertrand.
I don't fully understand the iteration of the Walk Forward. For example I optimized the data for 1/1/2010-1/6/2011, with optimization of 90 days and 90 days testing.
What will be the iterations of such setting?

[NinjaTrader_Bertrand]

It would then take the first 90 days, optimize on it and then trade the next 90 days, the test period and continue to do this until the end of your data set selected is reached.

[freewind]

but what are the steps size between each iteration?

[NinjaTrader_Bertrand]

It would use whatever you specify for optimization range and steps under Parameters, this would be the same as regular optimization.

[freewind]

I still don't get it.
Example:
I have 2 years I want to walk forward on, 90 days optimization 90 testing.
The first iteration the first 90 days of my data optimization and then the following 90 days for testing.Now I have left with 1.5 years, this is the next iteration of run forward half year later next 90 days for optimization and the 90 days for testing etc.?

I apologize for all those questions but I want to fully understand this feature which is not well documented.

[NinjaTrader_Bertrand]

Yes, correct as the optimization and test period is 90 days here as you have specified. It's going slice by slice to your dataset to simulate the 'optimize and trade fresh data' cycle you would normally have in real life as well with your strategy.

[freewind]

Thanks I got it.
Now I need to think on a way to use it.

[freewind]

Bertrand,
How I interpret the results? I see each result in the optimization tab and walk forward tab has use different parameters, so how can I compare what are the best parameters to trade with?

Tx in advance!

[NinjaTrader_Bertrand]

freewind, you would not get a single 'best' parameter set here - you rather see how your continuous optimization is holding up over time, which is essentially what you need to answer if employing any kind of optimization in a strategy. If you see a grouping of the same or close parameters in the best individual optimization results saved, this is then a more stable parameter overall as one that would appear only one time for an specific data set where it delivered it's best results.

[freewind]

Bertrand, I'm sorry but I still don't understand how to choose from a Walk Forward optimization the parameters, in the end my strategy, when running live, can run only with one set of parameters and the goal is to pick the most stable one.
If you can give a practical guide how to use the Walk forward it would be great.
One more thing is that I don't understand the connection between the Optimizer tab and Walk Forward tab if each result refer to a different parameter set!

Tx for your time.

[NinjaTrader_Bertrand]

freewind, I could unfortunately not give you an example, as picking parameters to use in your strategy on the unseen data is not an exact science - the WalkForward optimization will let you judge how good the continuous optimization in your set parameter ranges is holding up over time, if you're happy with the results you would apply the same optimization to your past data set to find the new best parameters to apply going forward.