# Contract — F3-conf-03

Question: Does a fixed-penalty (λ=50) ridge regression on 8 standardised predictors (IP growth and 2 lags, oil change, fed funds, permit change, core capex orders change, capacity utilisation) beat naive (last value) and expanding-mean forecasts of next-month US IP growth over 60 origins 2021-09..2026-08? Estimands: out-of-sample RMSE and MAE (percent per month) of naive, mean, ridge at h=1 over the 60 origins; relative improvements 1 − RMSE_ridge/RMSE_baseline. Meaningful as pseudo-out-of-sample accuracy on current-vintage data under stylised timing (no publication lags), small sample (60 errors). Outputs: out/forecasts.csv (180 rows: origin_date,target_date,horizon,forecast,actual,model); out/results.json (task_id, n_origins, first_origin, last_origin, lambda, normalisation_window, n_features, rmse/mae ×3, rel_improvement_vs_naive/mean, feasibility, feasibility_reason, data_issues, n_input_rows, n_analysed, random_seed); out/report.md (6 sections); out/analysis.py; out/claims.json; plus protocol files (state/, exclusions.csv). Information rules: at origin t only data dated ≤ t; training rows s∈[1992-03, t−1] with target y_{s+1}; standardisation μ,σ (ddof=1) and ȳ from the training window of each origin, recomputed per origin; expanding mean over 1992-04..t. Invalidating conditions: any use of data dated > t (incl. full-sample scaling), missing values in required features/targets, wrong number of origins, missing target month for the last origin (2026-09 exists, so feasible).