Random questions about videos

[shanlane]

Hello,

I had a couple of very random (and hopefully easy) questions about videos 7a and 7b from last year I was hoping someone could answer for me.

1. When talking about the spread for LVaR, how exactly is that computed? What is a 2% spread? Does that just mean, for example, if a security has a mid marlet price of $100 and its spread is $2 ($99 bid and $101 offer)then it has a spread of 2%?

2. In slide 47 of 7b, it says that a heavy tailed distribution is one in which its tails are not exponentially bounded. Could you please briefly explain what is meant by this? The reason I ask is that in the source material, it looks like this definition is reversed. It says that it consideres a distribution light tailed if it has "finite moments of all order". The exponential, weibull, gamma and Lognormal have "infinite" maximal moments and the other distributions have finite moments. Yet the four that have "infinite" moments are the ones that are considered light tailed. Is something backwards?

3. In slide 53, you say that LDA assumes losses are iid and distributed according to some distribution F . Are you actually referring to the F distribution or are you just following some convention from the reading that names the severity distribution "distribution F"?

4. Finally, you use the term "unit of measure" quite frequently and I am not really sure what this is referring to.

Sorry for all of the random questions

Shannon

 

[David Harper CFA FRM]

Hi Shannon,

1. Yes, exactly as you calculate: the spread is always (Jorion, Dowd, Alexander) the difference in bid/ask divided by the MIDPOINT
2. I need to bookmark and come back. First, I agree the reading here is confusing (it leans on an alternative definition of heavy tail: that the tails are not exponentially bounded), it took me a long while to reconcile their language; Second, FWIW, the exam won't use this definition, as you may know, the exam will refer to heavy tail as kurtosis > 3.0. I will try to revert on this Sat/Sunday, after we get the Study Notes due out.
3. The latter, not an F-distribution, but rather a "generic" loss distribution denoted with F(.); btw, in the Deutsch Bank LDA case study (the source), they ended up grafting "piecewise" a POT EVT (parametric; heavy-tailed) onto an empirical distribution for the body. So, the F(.) distribution referred to this, which is nothing like the sampling F-distribution.
4. It's the "cell" sufficiently large to aggregate severity + frequency, but small enough to capture the business unit/event type.
   See this practice question (do you like how how i try to use questions to impart terms?! ): "11.5. A unit of measure (UOM) is the disaggregated level at which a bank starts distinguishing, specifying and then estimating the frequency and severity distributions. If a bank holding company maps all internal loss data exactly to the Basel II/III UOMs under the Advanced Measurement Approach (AMA) to operational risk, with 1:1 correspondence against all business types and events, how many UOMs will the bank populate before any aggregation?"

Thanks, David

 

[shanlane]

Thank you so much for clearing that up!

I also like (please excuse the sarcasm) how this author classifies lognormal as a light tailed dist when in part 1 it was usually referred to as a heavy tailed dist. As you mention, it is the excess kurtosis that gave it that distinction, but this certainly seems like they are trying to confuse me . Back to studying!

Thanks again!

Shannon

 

[David Harper CFA FRM]

Ha, if the sarcasm is not pointed in my direction, I say "bring it on!"

 

[shanlane]

No way! You are awesome! Some of these authors, on the other hand...

Thanks again,

Shannon