transform_policy
Back to L2 | Browse all axes | Browse all options
Axis
transform_policyon sub-layerl2_b(layerl2).
Sub-layer
l2_b
Axis metadata
Default:
'apply_official_tcode'Sweepable: True
Status: operational
Operational status summary
Operational: 3 option(s)
Future: 0 option(s)
Options
apply_official_tcode – operational
Apply McCracken-Ng’s series-by-series stationarity transforms.
Each FRED-MD/QD series ships with a transformation code (1-7) mapping to a stationarity transform. apply_official_tcode runs the canonical mapping per series:
1 = level
2 = first difference
3 = second difference
4 = log
5 = first difference of log (≈ growth rate)
6 = second difference of log
7 = log diff of (1 + growth rate)
Applied per-origin within walk-forward to avoid look-ahead.
When to use
Default for FRED-based studies. Canonical replication path.
References
macroforecast design Part 2, L2: ‘preprocessing is the only layer with a strict A→B→C→D→E execution order; every cell follows the same pipeline.’
McCracken & Ng (2016) ‘FRED-MD: A Monthly Database for Macroeconomic Research’, JBES 34(4). (doi:10.1080/07350015.2015.1086655)
Related options: no_transform, custom_tcode, transform_scope
Last reviewed 2026-05-04 by macroforecast author.
no_transform – operational
Skip transforms; pass raw levels through.
Useful for tree-based / ranking models that don’t need stationarity, or for studies that apply alternative transforms in L3 (Hodrick-Prescott filter, Hamilton (2018) detrender, etc.).
When to use
Tree / forest models; alternative-transform studies; custom panels with already-transformed data.
References
macroforecast design Part 2, L2: ‘preprocessing is the only layer with a strict A→B→C→D→E execution order; every cell follows the same pipeline.’
Related options: apply_official_tcode, custom_tcode
Last reviewed 2026-05-04 by macroforecast author.
custom_tcode – operational
User-supplied per-series t-code map.
Requires leaf_config.custom_tcode_map: {series_name: int}. Macrocast applies the same 1-7 mapping as apply_official_tcode but reads codes from the user’s dict instead of the bundled FRED metadata. Useful for custom panels where the user wants the McCracken-Ng transform vocabulary.
When to use
Custom panels with user-defined stationarity codes.
References
macroforecast design Part 2, L2: ‘preprocessing is the only layer with a strict A→B→C→D→E execution order; every cell follows the same pipeline.’
Related options: apply_official_tcode, no_transform
Last reviewed 2026-05-04 by macroforecast author.