Recent results have shown backtests of expected shortfall (ES) are necessarily approximated, in the sense that they are unavoidably sensitive to possible errors in the prediction of value-at-risk. Carlo Acerbi and Balazs Szekely introduce a backtest for ES that minimises such sensitivity. The bias is small: the effect is generally negligible for small VAR discrepancies. Moreover, the bias is prudential, in the sense that any imperfect VAR prediction results in a more