@Ugur MULUK

I have found your answers about this issue helpful when naively predicting time series. Now I understand I must do some clever work to the features to avoid the regressor's cheating issue.

I like your suggestion of switching to a binary (or multi-class) classification for predictions. I'm testing it right now. But my question is: How can I make further predictions with my binary/multi-class model?

Now I don't have prices as labels and I need to go beyond one single prediction.

Any suggestion would be very welcomed. Thank you