Target (y) And Predictor (x) Definitions

Parent: Layer 1: Data Source, Target y, Predictor x
Current group: Target (y) and predictor (x) definitions

This group names the forecasting target y and, for FRED-backed routes, the eligible raw predictor x columns. Layer 1 does not decide how y is transformed, how horizon targets are built, how x is lagged, or which representation reaches a model. Those decisions start in Layer 2.

FRED column dictionaries are not maintained in this Layer 1 page. Use 5. FRED-Dataset for the current FRED-MD, FRED-QD, and FRED-SD column definitions before writing explicit y/x lists.

Target (y) Definition

target_structure has two user-facing choices: Single Target and Multiple Targets. The canonical recipe values remain single_target and multi_target.

Value	Required payload	Meaning
`single_target`	`leaf_config.target`	Single Target: one y series. Required by one-target Study Scope values.
`multi_target`	`leaf_config.targets`	Multiple Targets: two or more y series. Required by multiple-target Study Scope values.

Target payload:

target: required for single_target.
targets: required for multi_target.
horizons: required forecast horizons.
sample_start_date / sample_end_date: optional sample-period bounds.

Layer 2 boundary:

horizon_target_construction decides whether y is level, difference, log-difference, direct average, path-average growth, or another supported horizon target representation.
target_transform, target_normalization, target missing policies, and target outlier policies are representation decisions, not source-frame identity.

FRED-MD/QD Predictor (x) Universe

variable_universe is a FRED-MD/QD metadata axis. It filters FRED-MD/FRED-QD source columns that are eligible as candidate predictors x before Layer 2 builds lags, factors, feature blocks, rotations, or custom representations. The current all-column dictionaries live in 5.1 FRED-MD and 5.2 FRED-QD.

For custom_source_policy: custom_panel_only or standalone dataset: fred_sd, this axis is hidden by default. Custom-only x columns are defined by the custom file itself. FRED-SD x columns are defined by state scope and series scope in 4.1.4 FRED-SD Predictor Scope, with the current generated column dictionary in 5.3 FRED-SD.

Value	Required payload	Meaning for FRED-MD/QD
`all_variables`	none	Use all eligible non-date source columns except target y. For FRED-MD/QD this means the whole selected FRED panel after source loading and official transform policy.
`core_variables`	none	Use the package core macro subset: `INDPRO`, `PAYEMS`, `CPIAUCSL`, `FEDFUNDS`, `GS10`, `M2SL`, `UNRATE`, when those columns exist in the selected panel.
`category_variables`	`leaf_config.variable_universe_category_columns` and `leaf_config.variable_universe_category`	Use a named category from a user-supplied category map. The map should be built from the FRED-MD or FRED-QD all-column table, or another documented study taxonomy.
`target_specific_variables`	`leaf_config.target_specific_columns`	Use a different x list for each y. Write these lists after inspecting the selected dataset’s all-column table in 5. FRED-Dataset.
`explicit_variable_list`	`leaf_config.variable_universe_columns`	Use one explicit x list for all target y series. Write the list after inspecting the selected dataset’s all-column table in 5. FRED-Dataset.

Current package behavior:

all_variables uses the loaded panel columns directly.
core_variables uses the fixed package subset listed above.
category_variables does not currently infer built-in FRED category maps at runtime. It requires leaf_config.variable_universe_category_columns.
target_specific_variables and explicit_variable_list are user-authored lists. They should be written against the column names visible in the selected FRED-MD/QD panel. Use 5. FRED-Dataset as the current column reference.

Category map example:

path:
  1_data_task:
    fixed_axes:
      variable_universe: category_variables
    leaf_config:
      variable_universe_category: labor
      variable_universe_category_columns:
        labor: [PAYEMS, UNRATE]
        prices: [CPIAUCSL]
        policy: [FEDFUNDS, GS10]

Explicit list example:

path:
  1_data_task:
    fixed_axes:
      target_structure: single_target
      variable_universe: explicit_variable_list
    leaf_config:
      target: INDPRO
      horizons: [1, 3, 6]
      sample_start_date: "1980-01"
      sample_end_date: "2019-12"
      variable_universe_columns: [RPI, UNRATE, CPIAUCSL, FEDFUNDS]

Target-specific example:

path:
  1_data_task:
    fixed_axes:
      target_structure: multi_target
      variable_universe: target_specific_variables
    leaf_config:
      targets: [INDPRO, UNRATE]
      horizons: [1, 3]
      target_specific_columns:
        INDPRO: [RPI, UNRATE, CPIAUCSL]
        UNRATE: [PAYEMS, CLAIMSx, INDPRO]

FRED-SD rule:

FRED-SD state and series filters are handled by 4.1.4 FRED-SD Predictor Scope.
Standalone fred_sd should use State Scope / State List and Series Scope / Series List, not variable_universe.
Composite fred_md+fred_sd or fred_qd+fred_sd routes use variable_universe for the FRED-MD/QD portion and the FRED-SD scope axes for the state-level portion.