1[2 {"id":"C1","text":"The naive forecast (GDP YoY of t-2) has RMSE 0.814 pp at h=1 and 1.140 pp at h=3 over the 48 origins.","results_keys":["rmse_naive_h1","rmse_naive_h3","mae_naive_h1","mae_naive_h3"],"status":"supported","caveat":"Current vintage data; publication timing simulated only."},3 {"id":"C2","text":"The direct ARDL model with LFS employment growth has lower RMSE and MAE than naive at both horizons (RMSE 0.666 vs 0.814 at h=1; 0.851 vs 1.140 at h=3).","results_keys":["rmse_model_h1","rmse_model_h3","mae_model_h1","mae_model_h3"],"status":"supported","caveat":"Point estimates on 48 origins only."},4 {"id":"C3","text":"Relative RMSE improvement is 18.1% at h=1 and 25.3% at h=3.","results_keys":["rel_improvement_rmse_h1","rel_improvement_rmse_h3"],"status":"supported","caveat":"Not statistically significant at 5% (see C4)."},5 {"id":"C4","text":"The improvement over naive is not statistically significant at 5% (DM p=0.126 at h=1, p=0.221 at h=3).","results_keys":["dm_stat_h1","dm_pvalue_two_sided_h1","dm_stat_h3","dm_pvalue_two_sided_h3"],"status":"supported","caveat":"DM test with 48 overlapping errors has low power; HAC choice (h+1 lags) is a judgement."},6 {"id":"C5","text":"All forecasts respect the information rules (GDP <= t-2, employment <= t-1, training targets <= t-2).","results_keys":["last_target_date_available","last_predictor_date_available"],"status":"supported","caveat":"Verified by ledger date checks and a perturbation test; vintage revisions not simulated."},7 {"id":"C6","text":"The gain is attributable specifically to employment information.","results_keys":[],"status":"not_evaluated","caveat":"No employment-free regression benchmark estimated; the report does not make this claim."}8]9