ENH: out-of-sample moment conditions, score, GMM, MLE

[josef-pkt]

looks interesting for diagnostics

out of sample or forecast moment conditions, and it's asymptotic distribution

Hoffman, Dennis, and Adrian Pagan. “PRACTITIONERS CORNER: Post-Sample Prediction Tests for Generalized Method of Moments Estimators.” Oxford Bulletin of Economics and Statistics 51, no. 3 (1989): 333–43. https://doi.org/10.1111/j.1468-0084.1989.mp51003007.x.

We still have the problem that loglike and score can only be evaluated at the model data, endog, exog, i.e. no predictive, out-of-sample statistics. Similarly for GMM, momcond are only in-sample.
In GLM we could use freq_weights to only estimate on a subsample.