I have a time series with a huge disruption in a pair of its values, which if I understand well makes squared residuals not independent. The ADF test suggests the process is stationary, and there seems to be no trend in the plot. However, I'm not sure if such outlier would invalidate stationarity in my process.
auto.arima, as well as the ACF and PACF, suggest an ARMA model, but I don't know if it's appropriate. Still, squared residuals are not independent so it isn't precisely a good model.
I don't know how to solve this if I should try GARCH or differentiating and I'm a bit at a loss.
Can someone shed some light on the issue? Thanks