ReadonlybaselineReadonlybeatsnull when no origin could be scored.
ReadonlydieboldDiebold–Mariano of the model's QLIKE terms against the baseline's (HAC
lags for the windows' overlap, ⌈horizon/step⌉ − 1): is the difference
between loss and baselineLoss more than noise? null with too few
scored origins or no variance in the differences.
ReadonlylossReadonlyoriginsReadonlyscoredEach scored origin's QLIKE term and the baseline's, in order: loss is their mean.
ReadonlyskippedOrigins dropped because a fit did not converge or a forecast was not usable.
A walk-forward's grade: the model's and the baseline's mean QLIKE, each scored origin's terms, and their Diebold–Mariano test.