step,rule,n_before,n_after,n_removed,rationale
1_lag,drop first quarter (no lagged value for AR(1)),314,313,1,AR(1) needs y_{t-1}
2_window,keep dependent quarters >= 1985-01-01,313,163,150,"post-1984 (Great Moderation) dynamics, fixed a priori"
3_covid,drop dependent quarters 2020Q2-2020Q4,163,160,3,COVID outliers would dominate OLS