A covariance stationary time series must have:

1. Constant mean (stationary mean)
2. Constant variance (homoscedasticity)
3. Constant autocovariance structure (autocovariance depends only on lag, not time)

Option D (a nonconstant variance) is the exception because it violates the requirement for constant variance. A nonconstant variance indicates heteroscedasticity, which means the time series is not covariance stationary.

The other options are all characteristics of covariance stationarity:

• A: Stability of autocorrelation (autocorrelation depends only on lag)
• B: Stability of the mean (constant mean over time)
• C: Stability of the covariance structure (autocovariance depends only on lag)