sample_start_rule

Back to L1 | Browse all axes | Browse all options

Axis sample_start_rule on sub-layer l1_e (layer l1).

Sub-layer

l1_e

Axis metadata

  • Default: 'max_balanced'

  • Sweepable: False

  • Status: operational

Operational status summary

  • Operational: 3 option(s)

  • Future: 0 option(s)

Options

earliest_available – operational

Start at the panel’s earliest date; tolerates leading missing values.

Keeps every row; lets the L1.C raw_missing_policy and L2 imputation handle leading NaNs. Useful when the L2 EM-factor imputer can recover early observations and dropping them would lose informative history.

When to use

Studies that want maximum sample length and trust L2 imputation to handle leading NaNs.

References

  • macroforecast design Part 1, L1: ‘data definition is the recipe layer that pins source, target, geography, and horizon – everything downstream branches off these choices.’

Related options: max_balanced, fixed_date

Last reviewed 2026-05-04 by macroforecast author.

fixed_date – operational

Pin the start date in leaf_config (e.g., 1985-01-01).

Requires leaf_config.sample_start_date (ISO date). The L1 loader trims to that date verbatim. Useful for replication scripts that need an exact sample window matching a published paper.

When to use

Replication scripts; ablation studies over alternative start dates.

References

  • macroforecast design Part 1, L1: ‘data definition is the recipe layer that pins source, target, geography, and horizon – everything downstream branches off these choices.’

Related options: max_balanced, earliest_available

Last reviewed 2026-05-04 by macroforecast author.

max_balanced – operational

Start at the first date where every requested series is observed.

Computes the latest first-observation date across every column in the panel and trims earlier rows. Guarantees a balanced panel without imputing leading missing values.

Default for studies that mix series with different start dates (common on FRED-MD because some series only begin in the 1980s).

When to use

Default for FRED-MD/QD studies with mixed start dates.

When NOT to use

Custom panels where every series shares the same start date (use earliest_available to keep all rows).

References

  • macroforecast design Part 1, L1: ‘data definition is the recipe layer that pins source, target, geography, and horizon – everything downstream branches off these choices.’

Related options: earliest_available, fixed_date

Last reviewed 2026-05-04 by macroforecast author.