Margin Call the movie. An explanation with more details than those provided in standard media

[Aber]

I know this is not the first topic on this movie but I am going to post any way.

The movie is amazing! No doubt. This along with The big Short are probably the only real good finance movies out there. There a lot of articles explanaining the movie but I would love for someone to explain it in more detail to someone who actually does have industry knowledge (like me). I am very interested in the conversation at the beginning of the movie. Here is a brief summary:

"This is basically everything that we have on our books at any given time. But what Eric was trying to do here is work it for levels of volatility that fall outside the limits of our standard VAR model (...) The volatility boundaries are basically set using historic patterns then stretching them out another 10-15%. roughly. "
" We are starting to test those historic patterns (...). Monday, last Friday, last Wednesday and Monday. "

What is going here? The way I see it is that their VAR model should look differently because they have recently observed higher volatility. So their VAR numbers are higher today than the were last month. Why is this such a big deal? I don't understand the phrase "levels of volatility that fall outside the limits of our standard VAR model". Please explain! Feel free to comment on the whole movie as well

 

[mhy]

The way I interpreted was:

-They have a Monte Carlo simulation that runs on their portfolio to generate their VaR. That simulation has volatility parameters which are periodically re-evaluated based on historical levels.

-Their model hasn't been re-calibrated in a while so their volatility parameter estimates are way off recent market levels

-As a result during the past few weeks their realized VaR has been far beyond what their model would have expected

-Eric Dale was trying to do simulations at the far tails (looking for 1%, black swan events, in terms of vol) thus trying to see what happens if their model estimate of volatility was totally wrong.

 

[vertigo]

it is probably my favorite movie on wall street because it is surprisingly accurate on many issues. there are two lines from the Gorilla that are unforgettable
"speak to me as you would a child or a golden retriever"
and
"three ways to win: be first, be smarter or cheat. and i dont cheat"
they are unforgettable because they are practical and useful.

The way I interpreted was:

-They have a Monte Carlo simulation that runs on their portfolio to generate their VaR. That simulation has volatility parameters which are periodically re-evaluated based on historical levels.

-Their model hasn't been re-calibrated in a while so their volatility parameter estimates are way off recent market levels

-As a result during the past few weeks their realized VaR has been far beyond what their model would have expected

-Eric Dale was trying to do simulations at the far tails (looking for 1%, black swan events, in terms of vol) thus trying to see what happens if their model estimate of volatility was totally wrong.

this is somewhat accurate but it is giving the quants and risk people too much credit. "periodically re-evluated based on historical levels" = nope. VaR is a terrible measure - saying "VaR has been far beyond what their model would have expected" is idiotic on three counts
1. believing that VaR being "far beyond" means anything useful - it does not - because VaR is garbage
2. believing what their model would have expected means anything useful - it does not - risk models are not used for prediction
3. this is the real killer: VaR hides the real risk. it builds a false confidence. a good metaphor for VaR: it will tell you if you do or do not have a bacterial infection, but when you get cancer (tail event), there is nothing it can do.

where Eric Dale goes wrong is "trying to do simulations at the fat tails". that is such a stupid thing to do - you cannot simulate fat tails, or rather, if you do so, then you are being very naive if you believe that is a true fat tail event. you would need billions of simulations to artifically create "1% black swan events". and they would not be characterised by volatility - no, market variables do not cause or characterise black swan events. it is the opposite - volatility is priced by how likely people think events will happen, e.g., Brexit or Greece leaving the EU.

 

[Mikey V]

What’s a good risk measure nowadays? Or are you hinting at the idea that there is no “good” measure?

 

[vertigo]

What’s a good risk measure nowadays? Or are you hinting at the idea that there is no “good” measure?

theoretically speaking it is possible for a risk measure to be "good" - it merely needs to be convex, ie. a convex risk masure. for example, expected shortfall is a convex risk measure.

there are major criticisms of expected shortfall, which (potentialy) make it a bad risk measure in practice. it's not intuitive (traders may dislike VaR but at least they know what it is) and if you have bad market data (which you will do), then the extreme quantiles are garbage and ES will be even worse than VaR!

my point is that from a theoretical perspective (mathematical theory, we dont care about implementation , the culture of a firm, how the risk model is implemented, etc) there are "good" risk measures. from a practitioner's perspective, they are all flawed - sticking to simple methods in these situations works best - historical simulation VaR with a long time horizon (at least five years) is probably the best we can do.

a major difference between an academic and a practitioner is that a practitioner can know something is garbage and still find a way to use it. you could do that with VaR. i am aware that i criticised VaR, but it will never go - the regulators need to justify their existence somehow and forcing banks to have an army of "risk management" teams is one way they justify themselves. "look, we make the banks secure!"