What's new

# Return Aggregation VaR.

#### shivanin

##### Member

Hi David,

This is screen shot from notes on VaR.
Can you please explain how do we get answer 1.3% in " Historical Simulation VaR" methodology??

Thanks.

#### David Harper CFA FRM

##### David Harper CFA FRM
Staff member
Subscriber
Hi @shivanin Those XLS are here https://www.bionicturtle.com/topic/learning-spreadsheet-allen-chapters-2-3/ in this case, I just did a lazy: =ABS(PERCENTILE(E19:E23,1-confidence)); i.e., ABS of -1.30%. You will note this is not the correct 95% HS: there are only 5 losses (as the point is not HS per se but aggregation), such that the correct 95% loss is here the worst loss, but I just used the percentile to introduce some smoothing into a very jagged discrete distribution. Thanks,

#### shivanin

##### Member

But... I kinda get what you are saying.

Also... On the exam if we have to find 95% VaR for let's say 100 observations... Do we take 5th worse or 6th worse value??

#### David Harper CFA FRM

##### David Harper CFA FRM
Staff member
Subscriber
@shivanin technically the 6th worst (but on the exam the 5th is acceptable, as is the average of 5th and 6th; GARP is well aware of the fact that authors differ here). Thanks,

#### shivanin

##### Member
Hi David,

On the same lines... For the purpose of exam if not mentioned
What do we assume
1) one tail or two tail in case of confidence interval
2) annual or semiannual compounding?

#### David Harper CFA FRM

##### David Harper CFA FRM
Staff member
Subscriber
@shivanin
1. A hypothesis test (i.e., CI or significance) should be specified, but if not specified, assume two-tailed. (VaR is always one-specific)
2. Compound frequency must be specified, or can be inferred from the question. Thanks,