Stationarity checks help decide whether a series should be modeled in levels, differences, or another transformed state. The lab exposes two tests with opposite null hypotheses so the result is less dependent on a single test convention.
Tests
- ADF tests the null hypothesis of a unit root. Rejection supports stationarity under the selected deterministic specification.
- KPSS tests the null hypothesis of stationarity. Rejection points toward a unit root or a missing deterministic component.
Deterministic Terms
ADF supports no deterministic term, a constant, or a constant plus trend. KPSS supports a constant or a constant plus trend. The choice should match the economic reading of the series. A price level, a ratio, and a growth rate should not be treated as interchangeable inputs.
Lag Choice
The implementation uses finite observations only. Default lag choices follow common sample-size rules, and users can override them. Critical values are asymptotic. Short samples should be read with caution.
Interpretation
ADF rejection and KPSS non-rejection point in the same direction: the transformed series is plausibly stationary. Disagreement is common in macro data. When tests disagree, inspect the chart, the transform, the deterministic term, and any known breaks before treating the result as decisive.
API
See the Stationarity API for request and response fields.
References
- Said and Dickey, 1984.
- MacKinnon, 1996.
- Kwiatkowski, Phillips, Schmidt, and Shin, 1992.
- Schwert, 1989.