Stationarity
Stationarity is the fundamental simplifying assumption in time series analysis: the statistical properties of the process do not change over time.
1. Strict Stationarity
A stochastic process $\{X_t, t \in T\}$ is strictly stationary if for any $n \geq 1$, any times $t_1, \dots, t_n$, and any lag $h$, the joint distribution of $(X_{t_1}, \dots, X_{t_n})$ is the same as that of $(X_{t_1+h}, \dots, X_{t_n+h})$.
That is:
$$\mathbb{P}(X_{t_1} \leq x_1, \dots, X_{t_n} \leq x_n) = \mathbb{P}(X_{t_1+h} \leq x_1, \dots, X_{t_n+h} \leq x_n)$$2. Weak (Second-Order) Stationarity
A stochastic process $\{X_t\}$ is (weakly) stationary if:
- $\mathbb{E}(X_t) = \mu$ is constant (independent of $t$)
- $\text{Var}(X_t) = \sigma^2$ is constant (independent of $t$)
- $\text{Cov}(X_t, X_s) = \text{Cov}(X_{t+h}, X_{s+h})$ for all $t, s, h$
Condition 3 means the covariance depends only on the lag $h = |t - s|$, not on the time index itself.
3. Relationship
- Strict stationarity $\Rightarrow$ weak stationarity (provided $\mathbb{E}(|X_t|) < \infty$)
- Weak stationarity $\not\Rightarrow$ strict stationarity in general
- Key exception: Weak stationarity + Gaussian process $\Rightarrow$ strict stationarity
This is because a Gaussian process is fully characterized by its mean and covariance structure.
4. Consequences of Weak Stationarity
- Constant mean function: $\mu_X(t) = \mathbb{E}(X_t) = \mu$
- Constant variance function: $\sigma_X^2(t) = \text{Var}(X_t) = \sigma^2$
- Autocovariance Function (ACVF) depends only on lag: $\gamma_X(t, s) = \gamma_X(h)$ where $h = |t-s|$
5. Examples
- A sequence of i.i.d. random variables is always strictly stationary
- White Noise Process $\text{WN}(0, \sigma^2)$ is always weakly stationary, but may not be strictly stationary
- A process with time-dependent mean $\mu(t) = t$ is not stationary