Right now, it seems that the Models, use the last date on the data to start counting the days into the future.
But they use the execution date to count for the FORE_30, FORE_90, etc...
We think it is better to have a selector to set if we want the prediction from the execution time, or the last date on the data.
Now it is not consistent.
Last date on data for metricA: 01-Aug-2020
Execution date: 01-Sep-2020
Model to forecast 90 days in the future from execution time (expected date 30-Nov-2020)
For the metricA is considered the last date of the data and then add 90 days: 30-Oct-2020 !!!
Statistic FORE_90 date, is considered from execution date: 30-Nov-2020 there is NO value for metricA, as the forecast ends on 30-Oct-2020
So, there is a mismatch.
If we want to see the data 90 days into the future, should be a selector, to set it from EXECUTION time (by default) or from Last date with data available, but with some consistency, not to confuse the dates.
In summary: Use execution time to start the prediction period into the future.