As \(\ell\) gets large, it is most reasonable to simply forecast the sample mean, because of how they converge