Reply

Choosing the correct reoptimization method

1 replies

OceanOnline

Customer, bbp_participant, community, 9 replies.

Visit profile

10 years ago #111581

Hello,

 

still not sure what technique to use to periodically reoptimize your strategy?

 

As it is said in the articels, it is wise to use the WF Matrix to 1. Verify strategy robustness & 2. Find the optimal period for strategy reoptimization. 

 

So, for example, if we are discovering a good 3×3 cluster and the Matrix element tells us to reoptimize the strategy eg. every 260 days on 470 days of historical data, would I use now the “Simple Optimization” or better the “Walk – Forward Optimization”? Would I set the correct dates to only optimize the strategy to the mentioned 470 days or optimize it to all available historical backtest data? And if so WF Optimization method is also asking for OOS percentage. But I just would like to optimize the strategy – should I give it a only 1 day OOS data? Hmm, maybe some knows more about the last bit of procedure?

 

In addition, should we not safe the strategy settings for the next optimization or should the strategy in eg. 260 days again be tested against WF Matrix? Should any other bit of information be safed for proper future optimization?

 

Thank you in advance

Marcus

 

0

Mark Fric

Administrator, sq-ultimate, 2 replies.

Visit profile

10 years ago #122627

WF matrix test has two important outputs:
1. if the strategy passed (there is a 3×3 cluster)
2. what is the optimal reoptimizaiton period, for example it should be reoptimized every 260 days on 470 days history

The point 2. generally tells you that the strategy has the best results if you reoptimize it every 260 days.

So before you put strategy to MT4 demo/real account you should optimize it first and then reoptimize it again after 260 days, then again after another 260 days and so on.

To optimize the strategy you have to run just simple optimization – you don’t need any walk-forward test or matrix.
Just start simple optimization now on a data of 470 days (roughly 15-16 months back).
The best parameters found by optimization should be then used in live trading.

Then after 260 days from now you’ll repeat the optimization, again on history of 15-16 months back from the current date at that time.

The periods 260 and 470 don’t need to be 100% exact, 260 days is roughly 9 months, optimization should be on history of 15-16 months back.

I hope I explained it better now.

Mark
StrategyQuant architect

0

Viewing 1 replies (of 1 total)