1[2 {"id":"C1","text":"At h=1 the AR(2)-in-differences model does not beat the naive forecast: RMSE 0.118 vs 0.117 pp (relative improvement -0.010).","results_keys":["rmse_model_h1","rmse_naive_h1","rel_improvement_rmse_h1"],"status":"supported","caveat":"Difference not statistically distinguishable from zero (DM p=0.78); 35 evaluated pairs."},3 {"id":"C2","text":"At h=3 the AR(2)-in-differences model is worse than naive: RMSE 0.174 vs 0.159 pp (relative improvement -0.099).","results_keys":["rmse_model_h3","rmse_naive_h3","rel_improvement_rmse_h3"],"status":"supported","caveat":"DM p=0.036 but sensitive to COVID-exclusion choice (gap shrinks to -0.8% without exclusion); overlapping errors, short calm window."},4 {"id":"C3","text":"MAE ranking agrees with RMSE: naive has lower MAE at both horizons.","results_keys":["mae_naive_h1","mae_model_h1","mae_naive_h3","mae_model_h3"],"status":"supported","caveat":"Data rounded to 0.1 pp."},5 {"id":"C4","text":"35 of 36 (origin, horizon) pairs per horizon were evaluated; the missing 2025-10 observation excluded one target per horizon.","results_keys":["n_evaluated_h1","n_evaluated_h3","data_issues"],"status":"supported","caveat":""},6 {"id":"C5","text":"Results apply to the final-vintage series and do not establish real-time forecast performance.","results_keys":[],"status":"not_evaluated","caveat":"No real-time vintages were available."}7]8