Hi,
In Chapter 11 (or notebook 5 There is an example of creating trading signals with Decision Trees for Japanese equities.
I have always learned to work with timeseries with a certain window, for example looking at all the data (incl features created by talib) the past 30 days and forecasting returns for the next day. In the book and notebook, we are creating this timeseries in new features with columns like ret_1, ret_5, etc. Are we not losing too much information this way if we make predictions on one timestamp instead of on a full window?