What’s the best way to decide what data to build on?
2 weeks ago #280947
I’m looking into how best to select the IS and OOS data…
In the StrategyLab lessons, the H1 strategy was built using data from 2003 – 2019 and all of it was effectively IS.
In the masterclass Quastic lessons – building a similar strategy – the data used is 2013-2016 and the IS and OOS is set as 50/50.
So… how do we best decide what data to use for the build? How do we decide whether it is best to divide that data into IS and OOS or just leave it all as IS and use a shorter, newer period as OOS?
There is not a single best way to set this. The idea behind OOS is to have decent amount of “unseen” data for performance verification. Some traders use tiny IS part and check on large amount of OOS data while others use various IS / OOS splits settings.
OOS should be ideally set to data that kind of differs from IS meaning the market regime changes between the data period to “stress test” a strategy using OOS data
1 week ago #281065
Thank you ever so much Tomas, that is very useful.
Is there any real difference between…
a) explicitly specifying a range as OOS within a larger range
b) not building on a particular range and then using that in robust tests as a kind of “manual OOS”
I’m thinking that there isn’t real difference (assuming the dates are the same in both scenarios)?
I’m guessing that the ability to specify IS and OOS periods allows us to select a variety of each over a range?
Viewing 3 posts - 1 through 3 (of 3 total)