AbstractWe look at a statistical aspect of a recent case study. Although our discussion does not invalidate the general conclusions obtained there, it does suggest that the time series modelling may have been sub-optimal, and indicates how an alternative model specification could more probably have generated the sample statistics that were actually observed. The implication of our remarks is that discrimination between deterministic and stochastic non-stationarity, even if overlooked on preliminary inspection of the original data, will also be apparent from the raw sample autocorrelation plots.