I could see the argument that the weighting needs to taper off faster. The goal is to predict every game in the season, not just get the last games correct. The model does perform worse as the season wears on, but I believe that's largely because the games become less predictable (Alabama vs Michigan instead of Alabama vs Western Kentucky).

In insurance you look heavily at prior performance to predict the future. It would be not great to write the account at an overall average rate level and then just start making adjustments as you go.