There are many ways to define risk. It all depends on preferences and specific circumstances.
One way, and that used in the risk summaries on these pages, is simply to take 99% daily VaR and estimate it with a number of different methods, and take the average. This is a short-term risk measure, and the model averaging sort of smooths out some of the model risk.
A different way is to use a fat tailed method to unconditionally estimate the daily probability of a large outcome over some long time period, perhaps the expected worst day in 25 years. Then, it is best to use extreme value theory (EVT).
Neither approach is very informative about long-term risk, weekly, annual or decade. A different methodology is needed.
© All rights reserved, Jon Danielsson, 2018