{"id":"W2521147242","doi":"10.1002/for.2439","title":"Benchmark Forecast and Error Modeling","year":2016,"lang":"en","type":"article","venue":"Journal of Forecasting","topic":"Monetary Policy and Economic Impact","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Statistics Canada","funders":"","keywords":"Benchmarking; Benchmark (surveying); Computer science; Autoregressive model; Process (computing); Regression; Regression analysis; Software; Data mining; Econometrics; Machine learning; Statistics; Mathematics; Economics","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01089495,0.00139805,0.001500974,0.002248612,0.0006085905,0.002822136,0.002420263,0.001830295,0.004871683],"category_scores_gemma":[0.05699069,0.0005217915,0.001320783,0.0034315,0.0007688122,0.002833421,0.001658476,0.002424,0.001502398],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001807312,"about_ca_system_score_gemma":0.001508923,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01451991,"about_ca_topic_score_gemma":0.006241895,"domain_scores_codex":[0.9937252,0.003038275,0.0003854526,0.0009650901,0.001479441,0.000406438],"domain_scores_gemma":[0.9858323,0.00692122,0.001939191,0.001931792,0.003144128,0.0002313587],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001559102,0.00006921471,0.007486093,0.0001744246,0.0001127409,0.00008713602,0.0002015451,0.761914,0.0004577326,0.1332185,0.009175289,0.08694743],"study_design_scores_gemma":[0.00001310932,0.00003348447,0.001516446,0.00005042992,0.00001811481,0.00002150714,0.00004663521,0.9566143,0.0008274913,0.03727923,0.003548353,0.0000308416],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02377605,0.0003735039,0.9654249,0.0006075391,0.000210806,0.0001799067,0.001492892,0.001260959,0.006673445],"genre_scores_gemma":[0.6275318,0.0007803474,0.3540612,0.000255038,0.0003057736,0.0008147642,0.005863852,0.0004853286,0.009901892],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01451991,"threshold_uncertainty_score":0.05761862,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1954053801729682,"score_gpt":0.2359680379271932,"score_spread":0.04056265775422502,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}