An interesting request came through from one of our customers today and is worth sharing on the Causal Capital blog because a lot of risk practitioners struggle with some of the more complex areas of modelling uncertainty. This is especially the case when risk managers are attempting to assess the size of impacts from catastrophes.
Martin, today our risk team ran a scenario analysis exercise with a business unit and it ended up turning into a bit of a shambles when we came to assessing how large a specific catastrophe or scenario could become. What's worse is that one of the senior managers said and I quote "this scenario analysis process you guys put to us is not based on evidential data and is complete codswollop, why don't you use proper techniques for measuring extreme losses such as Extreme Value Theory". We left the meeting a little embarrassed and Googled EVT to be presented with some mathematically complex and nasty papers which I will send to you.
Please can you take a look at these papers and explain what Extreme Value Theory is, where it is used, how it works, please decrypt the language of math in these papers into commercial English. If you can show us an example of this EVT technique being applied to a business problem that would be awesome.
Response
In response to this request, a presentation has been created which can be downloaded from the following [LINK] but to make it easier to appreciate, I have applied some notes to the presentation in the format of a mini video. All this done quite quickly might I add.
Please can you take a look at these papers and explain what Extreme Value Theory is, where it is used, how it works, please decrypt the language of math in these papers into commercial English. If you can show us an example of this EVT technique being applied to a business problem that would be awesome.
Response
In response to this request, a presentation has been created which can be downloaded from the following [LINK] but to make it easier to appreciate, I have applied some notes to the presentation in the format of a mini video. All this done quite quickly might I add.
A quick run through of the slides to add context | Martin Davies
The paper from Eric Gilleland and Richard Katz [LINK] in the journal of statistics is phenomenal, well written up, straight forward to grasp and these guys are fantastic. Importantly the journal includes a working model of Extreme Value Theory that can be run up in R-Project and we feature this working model in our presentation. Being able to actually run EVT is invaluable because it's great to talk models all day long but being able to operate them is walking the talk I suppose.
No comments:
Post a Comment