Discussion of a paper by D. M. Hershfield, ‘An empirical comparison of predictive values of three extreme value procedures’