Dear sir,

As far as I know, both periods are worthful and needed. The overlap period of your observation and historical data of GCMs is needed for bias correction or downscaling ((start year)to the end of 2014). This is vital for having valid datasets (having downscaled or bias corrected of future datasets of different GCMs under different scenarios).

After downscaling or bias correction of GCMs data using overlap observation data, you will encounter time series of future data which is started from 1/1/2015. In this step, you will be able to compare data of any period of the future scenarios (near-mid-far future) with your observation data((...) to 2019) for extracting the variations of any parameters under different scenarios .

for more clarification, put the future GCM data of "2014 to current time" away and then your future data would be (2022)-2100. you can compare any period of that with your observation period (they should have same extent).