mixed_frequency_representation

Back to L2 | Browse all axes | Browse all options

Axis mixed_frequency_representation on sub-layer l2_a (layer l2).

Sub-layer

l2_a

Axis metadata

  • Default: 'calendar_aligned_frame'

  • Sweepable: True

  • Status: operational

Operational status summary

  • Operational: 5 option(s)

  • Future: 0 option(s)

Options

calendar_aligned_frame – operational

Default: keep selected mixed-frequency columns on the experiment calendar.

When a panel mixes monthly and quarterly columns (FRED-SD by default; any custom panel that declares per-column native frequency in metadata), the default representation flattens all columns to the experiment calendar via the L2.A quarterly_to_monthly_rule / monthly_to_quarterly_rule alignment rules. The panel emerges as a single rectangular frame; downstream layers see a uniform sampling grid.

When to use

Default for mixed-frequency studies; pairs with the canonical L2.A alignment rules.

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: drop_unknown_native_frequency, drop_non_target_native_frequency, native_frequency_block_payload, mixed_frequency_model_adapter

Last reviewed 2026-05-04 by macroforecast author.

drop_unknown_native_frequency – operational

Drop columns whose native frequency cannot be inferred.

Restricts the panel to columns whose native sampling rate is either declared in the L1 metadata or detectable from the FRED-SD workbook. Columns with unknown native frequency are dropped before any frequency-alignment rule fires.

When to use

Studies that demand strict provenance over per-column native frequency.

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: calendar_aligned_frame, drop_non_target_native_frequency

Last reviewed 2026-05-04 by macroforecast author.

drop_non_target_native_frequency – operational

Keep only columns whose native frequency matches the experiment frequency.

Restricts the panel to columns whose native sampling rate equals the L1 frequency. For a monthly experiment the quarterly columns are dropped (and vice versa). Useful when the user wants a strict single-frequency panel without any interpolation artifacts.

When to use

Strict monthly-only or quarterly-only panels; single-frequency benchmarks.

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: calendar_aligned_frame, drop_unknown_native_frequency

Last reviewed 2026-05-04 by macroforecast author.

native_frequency_block_payload – operational

Emit per-frequency block metadata for downstream models.

Keeps the panel intact (no alignment / drop) and instead publishes a fred_sd_native_frequency_block_payload.json manifest entry that lists each column’s native frequency. Models that consume mixed-frequency input directly (e.g. MIDAS, mixed-frequency factor models) can read this metadata from context['auxiliary_payloads'].

When to use

Researcher-owned MIDAS / mixed-frequency factor model studies.

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: mixed_frequency_model_adapter, calendar_aligned_frame

Last reviewed 2026-05-04 by macroforecast author.

mixed_frequency_model_adapter – operational

Block payload + a model-adapter contract for MIDAS-style fits.

Strictest option: emits the per-frequency block payload (see native_frequency_block_payload) plus a model-adapter contract that the L4 model_family must honour. The adapter validates that the registered model_family either declares MIDAS-style mixed-frequency support or registers via mf.custom_model with the appropriate auxiliary_payloads consumption. Runtime writes fred_sd_mixed_frequency_model_adapter.json with the adapter contract details.

When to use

Built-in MIDAS families (midas_almon, midasr) or registered custom mixed-frequency models.

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: native_frequency_block_payload, calendar_aligned_frame

Last reviewed 2026-05-04 by macroforecast author.