“…If we let F τ , h = μ τ , h + x τ , h , then under some assumptions , where ϕ h is a constant which depends on the distribution function of the data and the loss function, and V τ − h ( y τ ) is the conditional variance: see Patton and Timmermann (, Proposition 2). Clements () investigate whether asymmetric loss accounts for the inconsistencies reported in Table , but find little support for the contention. Our interest here is whether (anti‐)herding explains the inconsistencies, and whether we can use the above testing approach (‘Testing procedure’) to test this hypothesis in the presence of asymmetric loss.…”