Hi, I’m ready to go live on demo account with a new strategy. I’ve done WalkForward test 70/30. Results are good but maybe too good to don’t think that this strategy could be overfitted. What do you think about it? Is there any value on report that might lead you to think it’s overfitted?
Looks good!
No brainer … bang it on Demo WF and get on with another strategy?
Let us know how it goes?
Share it with us as our Christmas present!? 🙂
We have to assume that that WF result was done with a fixed dummy value and all other variable values fixed rather than with optimising values? We also have to assume that you did not develop and fit the strategy to all the available data and then do a WF on all the available data?
Yeah I didn’t assume above, but above is the more likely scenario!
Looks good!
No brainer … bang it on Demo WF and get on with another strategy?
Let us know how it goes?
Share it with us as our Christmas present!? 🙂
Christmas has passed GraHal 😀 perhaps for the next one 😉
We have to assume that that WF result was done with a fixed dummy value and all other variable values fixed rather than with optimising values? We also have to assume that you did not develop and fit the strategy to all the available data and then do a WF on all the available data?
That’s a problem for me. Having built the strategy with multiple timeframe I have many variables. So I’ve done WF with only a variable. I think that so is not correct? Right? How could I run a WF with more than 10 variables?
Hi – I think it is the idea to NOT using any variables (parameters which vary !), hence do not optimize (further).
Perhaps Vonasi can confirm that indeed this is his idea about it (it would be mine).
How could I run a WF with more than 10 variables?
It’s probably too late now as you have optimised your many variables over all the data available to you and so there is no TRUE OOS date to Walk Forward over??
Is above a correct statement?
PS I’ve crossed you off my Christmas card list! 🙂
How could I run a WF with more than 10 variables?
It’s probably too late now as you have optimised your many variables over all the data available to you and so there is no TRUE OOS date to Walk Forward over??
Is above a correct statement?
PS I’ve crossed you off my Christmas card list! 🙂
Yes, it’s correct. But when I build a strategy I think it’s normal found the best value that could perform. Obviously being many I could not optimize them all together but for blocks.
RE: PS: Nooo 🙁
Hi – I think it is the idea to NOT using any variables (parameters which vary !), hence do not optimize (further).
Perhaps Vonasi can confirm that indeed this is his idea about it (it would be mine).
Are you suggesting using a dummy variable?
Are you suggesting using a dummy variable?
Yes. Fix all the variables and then run the WF with a dummy variable that does nothing in the code. This then gives you an in sample and out of sample WF analysis of your finished strategy. Because it breaks it down into smaller sections of data to test and compare to other small sections of data there is a possibility for the test to see trades that perhaps were not seen in your standard backtest because perhaps you were already on the market. Personally I would develop the original strategy on only part of our back data then run it on the bit I didn’t develop it on. If it works the same or similar then a WF with a fixed dummy value would be the next test.
The WF done with variables optimising is a way for you to test and see if your strategy worked best with a single value of close range of values. Personally I rarely use it except perhaps just to check out one variable and see how import its value actually is.
Generally these days I don’t use the WF tool as I prefer running my strategies using my SEB testing idea which tells me a lot more about how curve fitted they are.
Here it is. It seems to be good again. Thank you all
