1[2 {"id":"C1","text":"At h=1 the AR2_diff model does not beat the naive forecast: its RMSE and MAE are higher (relative RMSE improvement -0.023).","results_keys":["rmse_model_h1","rmse_naive_h1","mae_model_h1","mae_naive_h1","rel_improvement_rmse_h1","n_evaluated_h1"],"status":"supported","caveat":"35 evaluated pairs; final-vintage data; single a-priori specification."},3 {"id":"C2","text":"At h=3 the AR2_diff model does not beat the naive forecast: its RMSE and MAE are slightly higher (relative RMSE improvement -0.008).","results_keys":["rmse_model_h3","rmse_naive_h3","mae_model_h3","mae_naive_h3","rel_improvement_rmse_h3","n_evaluated_h3"],"status":"supported","caveat":"35 overlapping 3-month-ahead errors; final-vintage data."},4 {"id":"C3","text":"The h=1 loss difference is statistically significant at 5% (DM p=0.017, model worse); the h=3 difference is not (p=0.18).","results_keys":["dm_stat_h1","dm_pvalue_h1","mean_loss_diff_h1","dm_stat_h3","dm_pvalue_h3","mean_loss_diff_h3"],"status":"supported","caveat":"Approximate small-sample DM test; the magnitude of the difference is economically negligible."},5 {"id":"C4","text":"One missing observation (2025-10-01) reduces evaluated pairs to 35 per horizon; 941 observations are in the last training window.","results_keys":["n_evaluated_h1","n_evaluated_h3","n_analysed","n_input_rows","data_issues"],"status":"supported","caveat":""}6]7