“…In the nonparametric case, we apply a kernel density method (see Nadaraya, 1964) which does not impose a specific theoretical distribution but instead derives the distribution of losses by smoothing the empirical distribution with an appropriate kernel function and bandwidth parameter (see Chen, 2008; Scaillet, 2004). For all six methods, we estimate the full set of parameters (time‐series model and distribution model) by the two‐step procedure of McNeil and Frey (2000) and Bhattacharyya et al (2008), which, in parametric approaches, can reduce misspecification error (in the parameters of the time‐series model) related to an incorrect distributional choice (see Ergen, 2015) and, in nonparametric ones, is the de facto standard (see Gao & Song, 2008). To characterize past investment risk in commodity futures markets and analyze its behavior in turbulent phases, like, recessions and stock market downturns, we apply the estimators to our entire range of return data.…”