If two time series have the same autocovariance function, are they the same process?
For me, this holds only if we suppose that the errors $\epsilon_t$ are normal, because in that case, the resulting $x_t$ are normally distributed and their joint distribution is fully characterized by their first two moments.