regime_estimation_temporal_rule
Back to L1 | Browse all axes | Browse all options
Axis
regime_estimation_temporal_ruleon sub-layerl1_g(layerl1).
Sub-layer
l1_g (gated)
Axis metadata
Default:
'expanding_window_per_origin'Sweepable: False
Status: operational
Operational status summary
Operational: 3 option(s)
Future: 0 option(s)
Options
expanding_window_per_origin – operational
Re-estimate regimes on every expanding window.
Default for estimated_* regime methods. Avoids look-ahead by re-fitting the regime model on data through each origin date.
When to use
Default; OOS-safe regime estimation.
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: rolling_window_per_origin, block_recompute
Last reviewed 2026-05-05 by macroforecast author.
rolling_window_per_origin – operational
Re-estimate regimes on a fixed-length rolling window.
Uses the most-recent params.window observations only. Useful when regime structure drifts over time.
When to use
Drifting / non-stationary regime structure.
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: expanding_window_per_origin, block_recompute
Last reviewed 2026-05-05 by macroforecast author.
block_recompute – operational
Re-estimate every leaf_config.regime_recompute_interval origins.
Cheap approximation to per-origin re-fits. Caches the regime classification between recompute boundaries.
When to use
Long sweeps where per-origin regime re-fits are infeasible.
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: expanding_window_per_origin, rolling_window_per_origin
Last reviewed 2026-05-05 by macroforecast author.