Builder OOS is true or fake?

Builder OOS is truely out of SQXs vision or fake?

Lets say date range is 2015.01.30 to 2020.08.30

In sample 2015.01.30 to 2018.01.15

Out of sample 2018.01.16 to 2020.08.30

Is it the same reliablity as if builder In sample 2015.01.30 to 2018.01.15

Plus

Retester 2018.01.16 to 2020.08.30

?

OOS data period is used just for verification. The IS data is used to optimize and train the strategy on the historical data to best fit to ranking criteria set.

But logically you are always fitting the strategy on the market prices history no matter if IS or OOS data is used. A good practice is to use some “unseen” period of historical data and verify the strategy after it passes OOS tests

Is oos unseen for sqx?

Hello,

better check this article on data parts https://strategyquant.com/blog/training-validation-test-periods/ since there are also types of IS data. OOS data is “unseen” for the strategy training part but since you set “OOS filters” usually OOS data is considered during the build phase. For that reason you can keep another part of data unused as “OOS2” and use only in retester to verify strategies produces in the builder

if you dont set any filtering for OOS – these data will be “unseen”

and if OOS filter is set than SQX curve fits?

then what is the point of OOS?

Gin wrote:

and if OOS filter is set than SQX curve fits? then what is the point of OOS?

Because if you do not use filters the data is unseen.  If you use filters then it is no longer OOS.

hankeys wrote:

if you dont set any filtering for OOS – these data will be “unseen”

Sqx generates a strategy on IS

Then this strategy is just simply tested on oos without fitting

If oos filter has been passed successfully then strategy is added to databank

Else strategy discarded without any further curve fittings to satisfy oos filters?

Right?