{
  "task_id": "F3-conf-03",
  "n_origins": 60,
  "first_origin": "2021-09-01",
  "last_origin": "2026-08-01",
  "lambda": 50.0,
  "normalisation_window": "expanding",
  "n_features": 8,
  "rmse_naive": 0.8830699166753823,
  "mae_naive": 0.6989594000000001,
  "rmse_mean": 0.5439355781141625,
  "mae_mean": 0.42929777256194207,
  "rmse_ridge": 0.5607989303521244,
  "mae_ridge": 0.45141289074453916,
  "rel_improvement_vs_naive": 0.3649439078805412,
  "rel_improvement_vs_mean": -0.031002480654837017,
  "feasibility": "complete",
  "feasibility_reason": null,
  "data_issues": [
    {
      "issue": "neworder_chg missing for 1992-01 and 1992-02 (NEWORDER starts 1992-02)",
      "handling": "no effect: training rows start at 1992-03 as specified, so these cells are never used"
    },
    {
      "issue": "data/README.md states the first month of every change series is empty, but ip_growth, oil_chg and permit_chg are populated for 1992-01",
      "handling": "values used as given (only ip_growth 1992-01/02 enters, as lags of the 1992-03 training row, as the task specification implies)"
    },
    {
      "issue": "extreme COVID-19 observations (e.g. ip_growth 2020-04 = -14.1, 2020-06 = +6.3) inside every training window",
      "handling": "retained; the ridge specification is fixed by the task and no outlier treatment is allowed for"
    },
    {
      "issue": "current-vintage (revised) data with publication lags ignored",
      "handling": "results describe pseudo-out-of-sample accuracy on revised data, not real-time performance"
    }
  ],
  "n_input_rows": {
    "ip_predictors.csv": 417
  },
  "n_analysed": 413,
  "random_seed": null,
  "dm_stat_ridge_vs_naive": -3.3649594921847754,
  "dm_pvalue_ridge_vs_naive": 0.0013506687241571726,
  "dm_stat_ridge_vs_mean": 0.5392198591267081,
  "dm_pvalue_ridge_vs_mean": 0.591764258342193
}