# Contract — F6-conf-03 (Okun's law, US 1960Q1–2019Q4)
**Question.** How much does the quarterly change in the US unemployment rate respond to quarterly real GDP growth, and is that response the same before and after 1984?
**Estimand.** β in Δu_t = α + β g_t + ε_t, with u_t = mean of the 3 monthly UNRATE values in quarter t, Δu_t = u_t − u_{t−1} (pp), and g_t = 100·Δln GDPC1_t (percent, not annualised). Sample: 240 quarters, 1960Q1–2019Q4 (the 1959Q4 levels are used only for the first differences). β is a descriptive (reduced-form) association. It is meaningful as a stable linear projection coefficient if the relation is linear and roughly stationary. It is not a causal effect.
**Uncertainty.** Newey–West HAC (Bartlett) SE for β, with the lag length chosen and reported. 95% CI = β ± 1.96·SE.
**Stability.** Separate OLS on 1960Q1–1983Q4 (96 quarters) and 1984Q1–2019Q4 (144 quarters). Classical Chow SSR F-test of the joint equality of (α, β), with df (2, n−4) = (2, 236).
**Outputs.** results.json keys: sample_start, sample_end, n_sub1, n_sub2, baseline_metric, model_metric, effect_estimate, intercept, effect_se, hac_lags, effect_ci_low, effect_ci_high, beta_sub1, beta_sub2, chow_f_stat, chow_p_value, stability_conclusion, conclusion, task_id, feasibility, feasibility_reason, data_issues, n_input_rows, n_analysed, random_seed. Also figure1.png, table1.csv (period,n,intercept,slope,slope_se), report.md (6 sections), analysis.py, claims.json.
**Information rules.** No quarter after 2019Q4 may enter any estimate. Use only data/GDPC1.csv and data/UNRATE.csv.
**Invalidity / impossibility.** The task would be infeasible if data were missing inside 1959Q4–2019Q4 or a month was missing from a quarter's average. Any use of post-2019 data would invalidate the result. Note that the GDP vintage is from 2026, so revised data differ from real-time data. That does not matter here because the question is ex post.