Definition
A strictly stationary process has a joint distribution unchanged by shifting time. Weak or covariance stationarity usually means a constant mean, constant finite variance and autocovariance depending only on lag. The chosen definition matters because many time-series tests and models rely on it.
How it is calculated
Weak stationarity requires constant mean mu, constant variance sigma squared, and covariance gamma_k that depends on lag k rather than calendar time t.
| Symbol | Meaning | Unit |
|---|---|---|
| E | expectation | X units |
| X | time-series value | series units |
| mu | constant mean | X units |
| Var | variance | squared X units |
| sigma | standard deviation | X units |
| Cov | covariance | squared X units |
| k | lag | periods |
| gamma | autocovariance at lag k | squared X units |
Interpretation
Stationarity is the idea that the probabilistic environment does not change merely because the calendar advances. A stationary return series can still be unpredictable and dependent. Prices commonly need transformations before a stationary model is plausible.
Assumptions and conventions
-
The stationarity definition is stated as strict or weak.
-
The sampling window is appropriate for the claimed stable moments.
Mean and sigma use X units; variances and covariances use squared X units.
Limitations and common misreadings
-
Stationary does not mean constant or independent.
-
A unit-root test result is not the definition of stationarity.
References
- Peter J. Brockwell, Richard A. Davis. (2016). Introduction to Time Series and Forecasting. Springer Texts in Statistics. Chapter 2, Stationary Processes. doi:10.1007/978-3-319-29854-2. Source

