Dear Support Team,
Since in the payout function, we can access the features data of "bblive" and "carry12", with range of 2007-2016.
Can we train our model on the whole date range (2007-2016)? And freeze this model for all the stressing testing? (like 2007-2008 ,2007-2013 ... so on)
Yes you can, you can do anything you want with the data that is given to you. Sub-sampling and cross validations are good methods to make your model more robust to forecast the future rather than explaining the past !