# Contract (P0) — F3-conf-02

Question. How accurately could US quarterly real GDP growth (A191RL1Q225SBEA, % q/q SAAR) be forecast 1 and 2 quarters ahead in real time (8 origins 2024-01-31 … 2025-12-31), and does a simple time-series model beat the naive (last value) forecast? Estimands. Per horizon h∈{1,2}: RMSE and MAE over 8 origins of naive and of one model, vs FIRST-RELEASE actuals; rel. improvement 1−RMSE_model/RMSE_naive; mean |current − first release| over the 8 h=1 target quarters. Meaningful only as descriptive real-time accuracy over a tiny sample (n=8 per horizon); no inference of significance is warranted. Outputs. out/forecasts.csv (32 rows; origin_date,target_date,horizon,forecast,actual,model,vintage_date,train_last_date,train_last_value,n_train); out/results.json (keys listed in TASK.md); out/report.md (6 sections); out/analysis.py; out/claims.json. Information rules. At origin d only vintage file d is used (its values for all quarters). current_vintage.csv only for the revision statistic. h counted from last quarter in vintage file. Actuals = first_release.csv. Naive = last value of vintage file. Invalidity/impossibility. Missing vintage or first-release value for any target; use of current vintage in estimation; model tuned on out-of-sample results (model spec fixed a priori: AR(1) OLS, 1985Q2+ dependent sample, 2020Q2–2020Q4 dependent obs excluded so COVID outliers never enter as y_t or y_{t-1}; iterated 2-step forecast).