“…Note that, as a consequence of the previous theorem, under the assumptions D(u n ) and D (u n ) the statement of Theorem 3.1.6 remains valid. Therefore, in a variety of applications where one needs to study extremes of time series featuring a sufficiently fast decay of correlations, GEV-and GPD-based statistical inference methods are used almost interchangeably, under the overall consensus that following the POT approach is more efficient when the time series are not exceptionally long [3,5,43,85,4,44]. Differences between the two methods emerge when extremes come in clusters; this is discussed below in Sections 3.2.2-3.4.…”