An analyst is modeling a time series $x_t$ as an AR(1) process and performs the following regression on the errors of the model, $\varepsilon_t$:

$$\varepsilon_t^2 = a_0 + a_1 \varepsilon_{t-1}^2 + u_t$$

where $u_t$ is an error term. If the coefficient $a_1$ is found to be positive and statistically significant, the analyst can conclude that $x_t$: