Stationarity Testing in Financial Time Series — ADF vs KPSS and Null Hypothesis Asymmetry
Unit-root null vs stationarity null, size/power trade-offs, structural breaks & rolling windows for live models.
Higher education in Financial Engineering and Money & Capital Markets. SPK (Turkey CMB) licence. 16 years across institutional markets, research, and quant-driven analytics.
Competing null hypotheses
ADF tests unit root as null — rejection supports stationarity. KPSS flips the logic: null is trend stationarity; rejection suggests non-stationarity.Using both provides triangulation; discordant outcomes flag nuanced dynamics (e.g. near unit root processes).
Breaks & regime shifts
Macro shocks (COVID, sudden central bank regime flips) invalidate full-sample stationarity claims. Employ rolling ADF windows or Zivot–Andrews style break-aware tests when justified.
Link to strategies
Mean-reversion engines require short-horizon stationary spreads; momentum engines may explicitly exploit unit-root drifts in logs with cointegration overlays.
Finvestopia context
Our longer educational pieces encourage humble stationarity claims — markets rarely offer textbook constancy without maintenance.Related entries
Leverage asymmetry, fat tails, EWMA shortcuts vs likelihood-based GARCH, and online calibration for prod systems.
H≈0.5 random walk vs trending/mean-reverting mythology, estimation variance, and why Hurst ≠ free alpha.
Educational content authored by our team — informational only, not investment advice.