Causal Inference

polars-ts provides causal inference methods for evaluating the impact of interventions on time series.

CausalImpact

Bayesian structural time series counterfactual analysis. Estimates what would have happened without an intervention.

import polars_ts as pts
from datetime import date

result = pts.causal_impact(
    df,
    intervention_date=date(2024, 3, 1),
    n_seasons=7,
)

print(result.summary())
# Pointwise and cumulative causal effects with credible intervals

With covariates

result = pts.causal_impact(
    df,
    intervention_date=date(2024, 3, 1),
    covariates=["weather", "demand"],
)

Synthetic Control

Donor-pool weighted counterfactual estimation for panel data with treated and control units.

sc = pts.SyntheticControl()
sc.fit(panel_df, treated_unit="unit_A", intervention_date=date(2024, 3, 1))

result = sc.estimate()
print(result.att)  # Average treatment effect on the treated

Placebo tests

# Run placebo tests to assess statistical significance
placebo = sc.placebo_test(n_permutations=100)
print(placebo.p_value)

When to use which

Method	Use case
CausalImpact	Single treated series, Bayesian counterfactual
Synthetic Control	Panel data with donor units, treatment effect estimation