Norbert
TSLA will win
That procedure would yield noisier shifts in projections from day to day. As it is the last day observation is the most influential on the projection. Shifting the last 14th day would be influential enough to add. Going with a deeper history like this means the starting date is fixed and stabilizing the projection.
The curvature through the last 14 days is pretty slight. So fitting a line (on logN) would not be that different.
So where would a line point to? Closer to 100K, I guess?