Vide subject,

Anyone?

Technically there are 252 trading days per year, 21 per month in the US. I think we use 20 because it makes the math much easier to work with @ 5% VAR.

agree with chuck. when you compute annual VAR you use 250 days, so dividing that by 12 would give you 20.83. looking at that question, however, that is ultra ultra minutae so i wouldnt get bogged down with that.

CFAI volume 5 - Reading 26 - Section 5.3 “The Advantages and Limitations of VAR”:

“Users of VAR should routinely test their system to determine whether their VAR estimates prove accurate in predicting the results experienced over time […] It is extremely important to go through this exercise, ideally across multiple time intervals, to ensure that the VAR estimation method adopted is reasonably accurate. For example, if the VAR estimate is based on daily observations and targets a 0.05 probability, then in addition to ensuring that approximately a dozen threshold violations occur during a given year, it is also useful to check other, shorter time intervals, including **the most recent quarter (for which, given 60-odd trading days**, we would expect approximately three VAR exceptions—i.e., losses greater than the calculated VAR), and **the most recent month (20 observations,** implying a single VAR exception)”.

In other words the aim is to test the accuracy of VAR estimates and to verify that the **approximate** 12 annual instances (250 * 0.05 = 12.5) where loss predicted by VAR are exceeded are relatively evenly distributed throughout the time horizon under consideration.

So, wanting to be precise:

Quarterly we would expect 250 / 4 = 62.5 days * 0.05 = 3.125 exceptions

Monthly we would expect 250 / 12 = 20.84 days * 0.05 = 1.041 exceptions.

Bottom line 60 and 20 are good enough to figure out the number of exceptions that are consistent with our var calculation.

Good luck, Carlo