1[2 {"id":"C1","text":"At h=1 the AR(p)-on-differences model (BIC chose p=1) has higher RMSE than the naive forecast (0.131 vs 0.117 pp; relative improvement -0.12).","results_keys":["rmse_model_h1","rmse_naive_h1","rel_improvement_rmse_h1"],"status":"supported","caveat":"35 evaluated pairs, final-vintage data, one model specification."},3 {"id":"C2","text":"At h=3 the model has higher RMSE than naive (0.181 vs 0.159 pp; relative improvement -0.14).","results_keys":["rmse_model_h3","rmse_naive_h3","rel_improvement_rmse_h3"],"status":"supported","caveat":"Overlapping errors; 35 evaluated pairs."},4 {"id":"C3","text":"MAE ranks the forecasts the same way: naive beats the model at both horizons.","results_keys":["mae_naive_h1","mae_model_h1","mae_naive_h3","mae_model_h3"],"status":"supported","caveat":""},5 {"id":"C4","text":"Diebold-Mariano tests indicate the model's loss is significantly larger than naive at the 5% level at both horizons (p=0.007, 0.019).","results_keys":[],"status":"supported","caveat":"Statistics are in out/supplementary.json, not results.json; small sample and approximate HAC make p-values indicative only."},6 {"id":"C5","text":"Over 2023-07 to 2026-06, the simple AR model did not beat the naive forecast at either horizon; without COVID dummies it is essentially tied with naive.","results_keys":["rel_improvement_rmse_h1","rel_improvement_rmse_h3"],"status":"supported","caveat":"The no-dummy sensitivity is in supplementary.json. This is not a general claim about all simple models or about real-time performance."}7]8