step,rule,n_before,n_after,n_removed,rationale
h1_last_origin_0,candidate pseudo-origins s = every month in file,345,345,0,all monthly rows
h1_last_origin_1,target y(s+1) observed at origin (<= 2026-02-01),345,337,8,training targets must be published at origin
h1_last_origin_2,"lagged regressors y(s-2), e(s-1) exist",337,335,2,start of sample
h1_last_origin_3,drop rows with target or regressor month in 2020-03..2021-12,335,310,25,pandemic outliers would dominate OLS (pre-specified)
h3_last_origin_0,candidate pseudo-origins s = every month in file,345,345,0,all monthly rows
h3_last_origin_1,target y(s+3) observed at origin (<= 2026-02-01),345,335,10,training targets must be published at origin
h3_last_origin_2,"lagged regressors y(s-2), e(s-1) exist",335,333,2,start of sample
h3_last_origin_3,drop rows with target or regressor month in 2020-03..2021-12,333,306,27,pandemic outliers would dominate OLS (pre-specified)