quantdata.dysanos.grid.gendysanos.compare_surface_spot_drift#

Compare autonomous and log-spot-driven ML-SANOS surface drifts.

The comparison is deliberately based on chronological one-step forecasts. An in-sample comparison is not decisive because the log-spot-driven model nests the autonomous model and adds one coefficient to every surface-state equation.

class quantdata.dysanos.grid.gendysanos.compare_surface_spot_drift.SurfaceSpotDriftComparisonConfig(*, training_fraction=0.7, minimum_rolling_training=600, rolling_validation_size=200, hac_lag=20, covariance_eigenvalue_floor=1e-08)[source]#

Bases: object

Configuration for chronological surface-drift comparisons.

training_fraction: float = 0.7#
minimum_rolling_training: int = 600#
rolling_validation_size: int = 200#
hac_lag: int = 20#
covariance_eigenvalue_floor: float = 1e-08#
quantdata.dysanos.grid.gendysanos.compare_surface_spot_drift.compare_surface_spot_drifts(states, time_y, *, config=SurfaceSpotDriftComparisonConfig(training_fraction=0.7, minimum_rolling_training=600, rolling_validation_size=200, hac_lag=20, covariance_eigenvalue_floor=1e-08))[source]#

Compare surface-only, log-spot-driven, and time-trend drift forecasts.

quantdata.dysanos.grid.gendysanos.compare_surface_spot_drift.main()[source]#