“…If the objective is a locally stationary extension of classical spectral analysis, only the autocovariances Cov(X t,T , X s,T ) have to be approximated. In the quantile-related context considered here, the joint distributions of X t,T and X s,T are the feature of interest, and traditional autocovariances are to be replaced with autocovariances of indicators, of the form Cov(I {X t,T ≤q t,T (τ 1 )} , I {X s,T ≤q s,T (τ 2 )} ), where q t,T (τ 1 ) and q s,T (τ 2 ) stand for the τ 1 -quantile of X t,T and the τ 2 -quantile of X s,T , respectively, with τ 1 , τ 2 ∈ (0, 1); see Li (2008Li ( , 2012, Hagemann (2013), or Dette et al (2015). Such covariances only depend on the bivariate copulas of X t,T and X s,T .…”