1[2 {"id":"C1","text":"Over 48 origins the employment-augmented direct ARDL model has lower RMSE and MAE than the naive forecast at h=1 and h=3.","results_keys":["rmse_naive_h1","rmse_model_h1","mae_naive_h1","mae_model_h1","rmse_naive_h3","rmse_model_h3","mae_naive_h3","mae_model_h3"],"status":"supported","caveat":"Current-vintage data; post-COVID evaluation window only."},3 {"id":"C2","text":"Relative RMSE improvement over naive is about 9% at h=1 and 17% at h=3.","results_keys":["rel_improvement_rmse_h1","rel_improvement_rmse_h3"],"status":"supported","caveat":"Point estimates; imprecise."},4 {"id":"C3","text":"The improvement is not statistically significant by a HLN-corrected Diebold-Mariano test (p about 0.19 and 0.22).","results_keys":["dm_stat_h1","dm_pvalue_h1","dm_stat_h3","dm_pvalue_h3"],"status":"supported","caveat":"Small sample of 48 overlapping errors; DM test has low power."},5 {"id":"C4","text":"Forecast errors increase with horizon for both methods.","results_keys":["rmse_naive_h1","rmse_naive_h3","rmse_model_h1","rmse_model_h3"],"status":"supported","caveat":""},6 {"id":"C5","text":"Under the publication-lag rule the last origin uses GDP through 2026-02 and employment through 2026-03, and the final estimation sample has 316 observations (h=1).","results_keys":["last_target_date_available","last_predictor_date_available","n_analysed"],"status":"supported","caveat":""},7 {"id":"C6","text":"Employment growth per se drives the improvement.","results_keys":["last_estimation_sample"],"status":"not_evaluated","caveat":"No ablation without employment was run; small positive coefficient suggests mean reversion may dominate."}8]9