DoubleMLForecaster

class DoubleMLForecaster(outcome_fcst, treatment_fcst, residual_fcst=None, exposure_vars=None, starting_window=0.1)[source]

Double Machine Learning forecaster for causal time-series forecasting.

Implements an adaptation of Double Machine Learning (DML) framework [1] for time-series, enabling deconfounded estimation of causal effects.

The forecaster uses a three-step residualization process to separate causal effects from confounding influences:

Fit procedure

Split X into exposure variables X_exposure and confounder variables X_confounder. X_exposure = X[exposure_vars], X_confounder = X.drop(columns=exposure_vars).
Fit the outcome forecaster to obtain residuals outcome_fcst.fit(y=y, X=X_confounder, fh=y.index), y_pred = outcome_fcst.predict(X=X_confounder), y_res = y - y_pred.
Fit the treatment forecaster to obtain exposure residuals treatment_fcst.fit(y=X_exposure, X=X_confounder, fh=y.index), X_exposure_pred = treatment_fcst.predict(X=X_confounder), X_exposure_res = X_exposure - X_exposure_pred.
Fit the residual forecaster to learn deconfounded causal relationship. residual_fcst.fit(y=y_res, X=X_exposure_res, fh=fh).
Refit the outcome and treatment forecasters for use during prediction. outcome_fcst.fit(y=y, X=X_confounder, fh=fh), treatment_fcst.fit(y=X_exposure, X=X_confounder, fh=fh).

Predict procedure

Split new X into exposure variables X_exposure and confounder variables X_confounder. X_exposure = X[exposure_vars], X_confounder = X.drop(columns=exposure_vars).
Compute the base (confounder-driven) forecast: y_pred_base = outcome_fcst.predict(X_confounder).
Compute the residualized exposures: X_exposure_pred = treatment_fcst.predict(X_confounder). X_exposure_res = X_exposure - X_exposure_pred.
Compute the causal (residual) forecast: y_pred_res = residual_fcst.predict(X=X_exposure_res).
Combine both components to obtain the final prediction: y_pred = y_pred_base + y_pred_res.

Parameters:

outcome_fcstsktime forecaster: Base forecaster modeling the outcome variable conditional on confounders.
treatment_fcstsktime forecaster: Forecaster modeling the exposure variables conditional on confounders.
residual_fcstsktime forecaster, optional (default=None): Forecaster modeling the residual (deconfounded) relationship between outcome and treatment. If not provided, a default forecaster is created using make_reduction(LinearRegression(), strategy="recursive"), a recursive reduction forecaster built from a linear regression model, providing a simple and interpretable baseline.
exposure_varslist of str, optional (default=None): Names of columns in X representing exposure (treatment) variables. The remaining columns are treated as confounders. If None, all features in X are treated as confounders and used only with the outcome forecaster. In this case, the treatment forecaster is not used, and None exposure residuals are passed to the residual forecaster, which therefore operates only on y.
starting_windowfloat, int or None, optional (default=0.1): If float, must be between 0.0 and 1.0, and is interpreted as the proportion of the starting dataset to ignore when doing in-sample predictions. Proportions are rounded to the next higher integer count of samples (ceil). If int, is interpreted as total number of samples to ignore.

Attributes:

outcome_fcst_sktime forecaster: Fitted clone of the outcome forecaster.
treatment_fcst_sktime forecaster: Fitted clone of the treatment forecaster.
residual_fcst_sktime forecaster: Fitted clone of the residual forecaster.

Notes

All provided component forecasters (outcome, treatment, residual) make proper use of exogenous data (X). In particular, the outcome and treatment forecasters must condition on both confounder and exposure variables. If a forecaster ignores exogenous inputs, the model reduces to a standard residual-based forecaster and loses its causal interpretation.
The outcome and treatment forecasters must support in-sample prediction, as residuals are computed from fitted values on the training data. If a forecaster does not natively support in-sample prediction, it can be wrapped using a utility such as OosForecaster to enable this functionality.
The residual forecaster should ideally be a simple interpretable model, such as a linear regression or reduced-form model, to preserve transparency of causal effect estimates.

References

[1]

Chernozhukov, V., Chetverikov, D., Demirer, M., Duflo, E., Hansen, C., Newey, W., & Robins, J. (2018). Double/debiased machine learning for treatment and structural parameters.

Examples

>>> from sktime.datasets import load_longley
>>> from sktime.forecasting.causal import DoubleMLForecaster
>>> from sktime.forecasting.naive import NaiveForecaster
>>> from sktime.split import temporal_train_test_split
>>>
>>> y, X = load_longley()
>>> y_train, y_test, X_train, X_test = temporal_train_test_split(y, X, test_size=4)
>>>
>>> # Assume 'GNP' is our exposure variable of interest
>>> exposure_vars = ['GNP']
>>>
>>> # Set up forecasters for nuisance functions
>>> outcome_fcst = NaiveForecaster()
>>> treatment_fcst = NaiveForecaster()
>>>
>>> # Create DoubleMLForecaster
>>> dml_forecaster = DoubleMLForecaster(
...     outcome_fcst=outcome_fcst,
...     treatment_fcst=treatment_fcst,
...     exposure_vars=exposure_vars
... )
>>>
>>> # Fit and predict
>>> fh = [1, 2, 3]
>>> dml_forecaster.fit(y_train, X=X_train, fh=fh)
DoubleMLForecaster(exposure_vars=['GNP'], outcome_fcst=NaiveForecaster(),
                   residual_fcst=RecursiveTabularRegressionForecaster(estimator=LinearRegression(),
                                                                      window_length=3),
                   treatment_fcst=NaiveForecaster())
>>> y_pred = dml_forecaster.predict(X=X_test)

Methods

`check_is_fitted`([method_name])	Check if the estimator has been fitted.
`clone`()	Obtain a clone of the object with same hyper-parameters and config.
`clone_tags`(estimator[, tag_names])	Clone tags from another object as dynamic override.
`create_test_instance`([parameter_set])	Construct an instance of the class, using first test parameter set.
`create_test_instances_and_names`([parameter_set])	Create list of all test instances and a list of names for them.
`fit`(y[, X, fh])	Fit forecaster to training data.
`fit_predict`(y[, X, fh, X_pred])	Fit and forecast time series at future horizon.
`get_class_tag`(tag_name[, tag_value_default])	Get class tag value from class, with tag level inheritance from parents.
`get_class_tags`()	Get class tags from class, with tag level inheritance from parent classes.
`get_config`()	Get config flags for self.
`get_fitted_params`([deep])	Get fitted parameters.
`get_param_defaults`()	Get object's parameter defaults.
`get_param_names`([sort])	Get object's parameter names.
`get_params`([deep])	Get a dict of parameters values for this object.
`get_pretrained_params`([deep])	Get pretrained parameters of this estimator.
`get_tag`(tag_name[, tag_value_default, ...])	Get tag value from instance, with tag level inheritance and overrides.
`get_tags`()	Get tags from instance, with tag level inheritance and overrides.
`get_test_params`([parameter_set])	Return testing parameter settings for the estimator.
`is_composite`()	Check if the object is composed of other BaseObjects.
`load_from_path`(serial)	Load object from file location.
`load_from_serial`(serial)	Load object from serialized memory container.
`predict`([fh, X])	Forecast time series at future horizon.
`predict_interval`([fh, X, coverage])	Compute/return prediction interval forecasts.
`predict_proba`([fh, X, marginal])	Compute/return fully probabilistic forecasts.
`predict_quantiles`([fh, X, alpha])	Compute/return quantile forecasts.
`predict_residuals`([y, X])	Return residuals of time series forecasts.
`predict_var`([fh, X, cov])	Compute/return variance forecasts.
`pretrain`(y[, X, fh])	Pre-train forecaster on panel (global) data.
`reset`()	Reset the object to a clean post-init state.
`save`([path, serialization_format])	Save serialized self to bytes-like object or to (.zip) file.
`score`(y[, X, fh])	Scores forecast against ground truth, using MAPE (non-symmetric).
`set_config`(**config_dict)	Set config flags to given values.
`set_params`(**params)	Set the parameters of this object.
`set_random_state`([random_state, deep, ...])	Set random_state pseudo-random seed parameters for self.
`set_tags`(**tag_dict)	Set instance level tag overrides to given values.
`update`(y[, X, update_params])	Update cutoff value and, optionally, fitted parameters.
`update_predict`(y[, cv, X, update_params, ...])	Make predictions and update model iteratively over the test set.
`update_predict_single`([y, fh, X, update_params])	Update model with new data and make forecasts.