`hamilton_filter` – Hamilton (2018) regression-based detrend (HP-filter alternative).#

Back to op axis | Back to L3 | Browse all options

Operational op under axis op, sub-layer L3_A_step_op, layer l3. Standalone callable: mf.functions.hamilton_filter_transform.

Function signature#

mf.functions.hamilton_filter_transform(
    panel: pd.DataFrame,
    h: int,
    p: int,
) -> pd.DataFrame

Parameters#

name	type	default	constraint	description
`panel`	`pd.DataFrame`	—	—	Input panel. Each column is a variable; rows are time periods. Series is promoted to a single-column DataFrame internally.
`h`	`int`	`8`	>= 1	Forecast horizon (periods ahead). Hamilton (2018) recommends h=8 for quarterly, h=24 for monthly.
`p`	`int`	`4`	>= 1	Number of lags used in the regression.

Returns#

pd.DataFrame — scalar result.

Behavior#

Regression-based two-sided alternative to the HP filter advocated by Hamilton (2018) for its better real-time properties. Default lookback h = 8 (quarterly) / 24 (monthly). Uses statsmodels hamilton_filter.

When to use

Real-time / one-sided detrending where HP’s two-sided smoothing is inappropriate.

In recipe context#

Set params.op = "hamilton_filter" in the relevant layer to activate this op within a recipe:

# Layer L3 recipe fragment
params:
  op: hamilton_filter

References#

macroforecast design Part 2, L3: ‘feature engineering is a DAG of typed transforms; cascade-depth bounds the longest chain at cascade_max_depth.’
Hamilton (2018) ‘Why You Should Never Use the Hodrick-Prescott Filter’, RES 100(5): 831-843.

hamilton_filter – Hamilton (2018) regression-based detrend (HP-filter alternative).#