In Appendix B, the distribution of the forecast errors of all 218 models for a few periods are shown, and we see that the individual forecasts aren’t uniformly spread around the true value, from which we conclude that the models have different variances.