# Contract — F2-conf-04

Question. Okun's law in U.S. quarterly data: how does the quarterly change in the unemployment rate co-move with quarterly real GDP growth, with uncertainty robust to serial correlation?

Estimand. β in the linear projection Δu_t = α + β g_t + e_t, u_t = mean of the 3 monthly UNRATE values of quarter t (pp), Δu_t = u_t − u_{t−1}, g_t = 100·ln(GDPC1_t/GDPC1_{t−1}) (percent, not annualised). Sample 1960Q1–2019Q4 (240 quarters; 1959Q4 levels feed the first differences). β is a linear-projection coefficient (descriptive association, not causal); OLS is consistent for it under stationarity/weak dependence with finite moments. HAC (Newey–West, Bartlett) SE is valid under serially correlated, heteroskedastic e_t with bandwidth growing slowly with T.

Outputs. out/results.json with keys task_id, beta_hat, alpha_hat, r_squared, se_iid, se_hac, hac_lags, beta_ci_low, beta_ci_high, okun_negative_5pct, mean_growth, first_quarter, last_quarter, n_analysed, feasibility, feasibility_reason, data_issues, n_input_rows, random_seed; out/report.md (Question; Data and exclusions; Method; Results; Uncertainty; Limitations and open issues); out/analysis.py; out/claims.json; out/figure1.png (scatter + fitted line); plus protocol files (out/state/*, out/exclusions.csv).

Information rules. Only data/ files; current vintage taken as given; observations outside 1960Q1–2019Q4 only used for 1959Q4 levels; no window extension.

Invalidating conditions. Missing/incomplete months in a sample quarter (or 1959Q4), missing GDP quarters, wrong growth definition (annualised), sample ≠ 240 quarters, using data beyond 2019Q4.