Measure 1

 It is interesting to note that when viewed in terms of measure 1, historical simulation does appears to be a reasonable model which neither significantly overestimates or underestimates risk over the full range of confidence levels. In fact, on the basis of this measure, one could perhaps reject model 5 (tail emphasized GARCH(1,1)) as being too conservative - in the sense of always over-estimating risk. In general the message from figures 1u and 1m is that models 1 and 3 have similar behaviours at low confidence level but in the regime beyond the $95\%$ level - tables are turned and - model 1 becomes far superior than all models except model 5. We also find that there is not much too choose between models 2 and 4.