Objectives. To test whether product characteristics available in public Food and Drug Administration (FDA) data, measured before a fixed date, identify prescription drugs that are added to the FDA shortage list after that date.Study Design. Temporal validation at two cutoff dates, applying a previously developed cross-sectional risk screen.Methods. Drugs marketed as of each cutoff were identified from National Drug Code directory marketing dates. Drugs already on the shortage list at the cutoff were excluded. Features were computed only from products marketed at the cutoff; the outcome was first shortage listing after it. L2-penalized logistic regression, a random forest and a fixed-weight three-item rule were compared under repeated stratified cross-validation. Because the rule has no fitted parameters, its score is a fully out-of-time prediction; the two fitted models are not, and are interpreted accordingly.Results. At a January 2025 cutoff, 125 of 2,779 eligible drugs (4.5%) were newly listed. Area under the receiver operating characteristic curve (AUC) was 0.732 for the random forest, 0.692 for logistic regression and 0.655 for the rule, with average precision (AP) of 0.143, 0.117 and 0.116 against a chance value of 0.045. At a January 2023 cutoff, 137 of 2,577 drugs (5.3%) were newly listed, with AUCs of 0.705, 0.675 and 0.649 and AP of 0.119, 0.133 and 0.125 against chance of 0.053. Discrimination was higher among injectables on both measures, but the two measures ranked the approaches differently: the rule achieved the highest AUC at both cutoffs (0.806 and 0.742) while logistic regression achieved the highest average precision at both (0.266 and 0.224). All between-approach differences were smaller than the fold-to-fold standard deviations.Conclusions. Public product characteristics carry modest forward signal for shortage listing, concentrated in injectables. A three-item rule with no fitted parameters discriminates comparably to fitted models out of time, though which approach ranks highest depends on the measure used and no approach leads on both. Absolute precision remains low, so the ranking is a screening aid rather than a forecast.