How well did the projections perform?

Compare historical model errors with a baseline built only from earlier results.