variable_universe

Back to L1 | Browse all axes | Browse all options

Axis variable_universe on sub-layer l1_c (layer l1).

Sub-layer

l1_c

Axis metadata

  • Default: 'all_variables'

  • Sweepable: False

  • Status: operational

Operational status summary

  • Operational: 5 option(s)

  • Future: 0 option(s)

Options

all_variables – operational

Use every series in the chosen dataset.

FRED-MD/QD ships ~130 / ~250 series respectively. all_variables uses every one of them as predictors (target excluded). Standard for high-dimensional forecasting comparisons (PCR, lasso, factor models).

When to use

Default. Any high-dimensional benchmark following McCracken-Ng.

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.’

  • McCracken & Ng (2016) ‘FRED-MD: A Monthly Database for Macroeconomic Research’, Journal of Business & Economic Statistics 34(4). (doi:10.1080/07350015.2015.1086655)

Related options: missing_availability, official_transform_policy

Last reviewed 2026-05-04 by macroforecast author.

core_variables – operational

Restrict to McCracken-Ng’s curated ‘core’ subset (~30 series).

Smaller predictor set covering output, prices, money/credit, interest rates, and labor. Useful when a study wants a low-dimensional benchmark or replicates a paper that used the core set explicitly.

When to use

Low-dimensional benchmark; comparison against published ‘core’ panel results.

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.’

  • McCracken & Ng (2016) ‘FRED-MD: A Monthly Database for Macroeconomic Research’, Journal of Business & Economic Statistics 34(4). (doi:10.1080/07350015.2015.1086655)

Related options: missing_availability, official_transform_policy

Last reviewed 2026-05-04 by macroforecast author.

category_variables – operational

Restrict to one McCracken-Ng category (e.g., ‘output_and_income’).

Uses one of the 8 (FRED-MD) / 14 (FRED-QD) category groupings as the predictor set. Requires leaf_config.variable_category naming the chosen category.

When to use

Within-category importance studies; testing whether one block alone is sufficient.

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.’

  • McCracken & Ng (2016) ‘FRED-MD: A Monthly Database for Macroeconomic Research’, Journal of Business & Economic Statistics 34(4). (doi:10.1080/07350015.2015.1086655)

Related options: missing_availability, official_transform_policy

Last reviewed 2026-05-04 by macroforecast author.

target_specific_variables – operational

Use a custom predictor list keyed to the target.

Requires leaf_config.target_specific_columns: {target: [predictors...]}. Different targets see different predictor sets. Useful when domain knowledge says only certain series are relevant for a given target (e.g., housing-target studies use housing-block predictors).

When to use

Domain-specific studies where each target has a known predictor block.

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.’

  • McCracken & Ng (2016) ‘FRED-MD: A Monthly Database for Macroeconomic Research’, Journal of Business & Economic Statistics 34(4). (doi:10.1080/07350015.2015.1086655)

Related options: missing_availability, official_transform_policy

Last reviewed 2026-05-04 by macroforecast author.

explicit_variable_list – operational

Use exactly the columns listed in leaf_config.variable_universe_columns.

Most flexible option. The recipe author supplies the full predictor column list in leaf_config; macroforecast filters the panel to that list verbatim. No grouping logic, no category lookup.

When to use

Replication scripts that need an exact predictor set; ablations.

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.’

  • McCracken & Ng (2016) ‘FRED-MD: A Monthly Database for Macroeconomic Research’, Journal of Business & Economic Statistics 34(4). (doi:10.1080/07350015.2015.1086655)

Related options: missing_availability, official_transform_policy

Last reviewed 2026-05-04 by macroforecast author.