I am seeing this behaviour in the ML strategies my solution is to have a few differenent last training years.

[nikolas-joyce]

No description provided.

[nikolas-joyce]

i have a configuration like this:

and it seem to trade for a time and then lean not to trade.

when i advance the last training year is seem to overcome this inactivity.


and then advance the last training year again:

I am proposing to include all 3 in a potential campaign to reduce the chance of inactivity.

Do you have any objections or do you see and alternatives to this approach?

[dmveden]

Not enough reversals points in the training data can cause such situation. Also keep in mind that last date in training data looks like - 31-12-'last training year', i can disable in-sample predictions if you want