1[2 {"id":"C1","text":"In real time, the AR(1) model had lower RMSE than the naive forecast at h=1 (1.48 vs 2.01 pp), a 26% improvement.","results_keys":["rmse_model_h1","rmse_naive_h1","rel_improvement_rmse_h1"],"status":"supported","caveat":"Only 8 origins; no significance test."},3 {"id":"C2","text":"At h=2 the AR(1) RMSE was 1.40 vs 2.27 pp for naive (38% improvement), and AR(1) had a smaller absolute error at all 8 origins.","results_keys":["rmse_model_h2","rmse_naive_h2","rel_improvement_rmse_h2","n_origins_model_abs_error_lower_h2"],"status":"supported","caveat":"Overlapping targets across origins; driven partly by the 2025Q1 swing."},4 {"id":"C3","text":"MAE ranks the models the same way at both horizons.","results_keys":["mae_model_h1","mae_naive_h1","mae_model_h2","mae_naive_h2"],"status":"supported","caveat":"Small sample."},5 {"id":"C4","text":"Mean absolute revision from first release to the 2026-10-30 vintage for the h=1 targets is 0.56 pp, about a third of forecast RMSE.","results_keys":["mean_abs_revision_h1_targets","rmse_model_h1"],"status":"supported","caveat":"Later revisions (e.g. future annual updates) may change this."},6 {"id":"C5","text":"The AR(1) coefficients were stable across origins (phi about 0.34).","results_keys":["ar1_coefficients_by_origin"],"status":"supported","caveat":"Depends on the 1985 start and the COVID exclusion."}7]8