Forums  > Risk Management  > same risk models  
     
Page 1 of 1
Display using:  

nikol


Total Posts: 1231
Joined: Jun 2005
 
Posted: 2020-08-05 15:12
Philosophical:

If risk models are the same, because regulators tend to approve same models and because risk quants absorb same intellectual stuff (courses, articles, lectures etc.) can it be the cause of systemic problems?

Same trading strategies lead to swarm effects, it is clear, but can I say that risk models filter same risk profile (= trading strategies) and, therefore, can be the cause of the systemic problem too.

Its Grisha


Total Posts: 61
Joined: Nov 2019
 
Posted: 2020-08-05 15:41
Good thread - I think this is especially a problem in more long-only businesses where allocators generally have greater transparency on the portfolio construction. As a result, when allocators are looking at Vendor B risk model, the asset managers they hire are incentivized to also use Vendor B for portfolio construction. This way, client and manager see tracking error through the same lens.

Where the problem arises is if one of Vendor B's risk components is a misrepresentation of the actual vol associated with that exposure (as tends to be the case). Now an entire complex of allocators and managers is constructing portfolios based on that.

Outside of the sleepy world of long only, I think the quant quake of 2007 can be viewed as risk model crowding:

https://web.mit.edu/Alo/www/Papers/august07.pdf

If you look at classical stat arb strategies (a la Avellaneda and Lee 2008), these can be roughly described as the mean reversion of returns that differ from those returns implied by covariance. Since this is essentially using a risk model to try to generate alpha, crowding in vanilla stat arb is directly tied to crowding of risk models.

nikol


Total Posts: 1231
Joined: Jun 2005
 
Posted: 2020-08-05 17:38
@Its_Grisha

"Vendor B risk model, the asset managers they hire are incentivized to also use Vendor B"

Yes, I kept in mind this specific dynamics. It's a classic game under competition: the less margin is left, the more people are looking at the neighbor to copy "best methods, practice, knowledge" etc. Then the "Best everything" is becoming obsession.

I am not sure if Avellaneda-Lee paper tells about same models. Of course, if market goes X-steps down which causes Y amount of losses, the risk model should capture that. And of course, stop-losses are executed through "Management Intervention".

But what I mean is that the similarity of these models is even larger - it's just a copy, because they all use same Vendor. Than no surprise, these machines will issue same red flags across the industry.

Still, is there any solution?

Its Grisha


Total Posts: 61
Joined: Nov 2019
 
Posted: 2020-08-05 18:09
Yes - Avellenada and Lee boils down to PCA on returns, but there are also statistical risk model vendors offering essentially this solution with several refinements for robustness.

Generally what you can buy falls into two classes: PCA-style and Fundamental Factor-style, with a few hybrid approaches.

Not sure on the solution - In my experience some of the smarter firms are building it in-house, or at the very least customizing the vendor solution. So it may be a matter of natural selection here.


tabris


Total Posts: 1276
Joined: Feb 2005
 
Posted: 2020-08-05 23:49
I think something along similar line has been written about on the risk of clearing houses.

Example here

Basically systematic risk only can happen when non isolated events lead to chain reactions. So you can argue that our backward looking initial margin/variation margin risk model might not work for system wide shocks that might cause a CCP to run low on cash based on historical margins.

Dilbert: Why does it seem as though I am the only honest guy on earth? Dogbert: Your type tends not to reproduce.

ronin


Total Posts: 608
Joined: May 2006
 
Posted: 2020-08-08 14:29
> Still, is there any solution?

Sure there is, but you won't like it.

The solution is diversification of risk models. So that any specific trigger pulls only a small fraction of institutions, and others are reacting differently.

The reason you won't like it is that the current market is pretty mature, and it is the investors who are pushing the standard risk models on everybody. The solution implies that there would be some cataclysmic wipeout of a significant nuber of investors, so that the slate is wiped clean.

And it's not just theoretical. Everybody who got hammered in the quant space this year seems to have been using some variant of Barra, at least based on my personal anecdotal experience.

"There is a SIX am?" -- Arthur

nikol


Total Posts: 1231
Joined: Jun 2005
 
Posted: 2020-08-08 14:59
@ronin

> Sure there is, but you won't like it.

Later on, after spelling the question, I tended to get the same answer, but thinking a bit further, I conclude that 10 different money managers betting on same thing should (at least theoretically) react to the disaster in the same manner. The question is why did they get into this overcrowded market.

My solution, perhaps, you will not like:

Limit allocations to specific betting (=strategies) by "upper hand", such that there is no over-crowding effect - the one which induces systemic downfall/waterfall. Of course, that is possible only for regulated entities. Wealthy privates can still do what they want up to the point where they can destabilize the market (SEC intervenes).

Still, if the risk model is able to see how far the market is populated (there are more predators then preys), then the direction of bet should be reversed.

> variant of Barra

It is like an ancient Roman aqueduct:
in short run, you cannot survive without the water, but you might get health problem due to poisoning from lead (Pb) in the longer run.

zee4


Total Posts: 79
Joined: May 2010
 
Posted: 2020-08-08 18:19
I think it is true that lots of asset management companies and pension funds use Barra/Northfield for modelling and measuring risk. This and other related stuff can be sniffed out, to some extent, from their job postings too.

However, even if you are using the same risk model, isn't this just a small piece of the puzzle? I mean you can see all you factor and country exposure using these risk models, but its still up to you to have an exposure or not. You can also go long or short depending other signals perhaps.

Here is another observation, the value guys have been getting crushed in recent years, but some of the prominent ones have been increasing allocation to it. I feel like this is not a risk modelling problem.

Бухарский

ronin


Total Posts: 608
Joined: May 2006
 
Posted: 2020-08-13 10:44
A risk model is just a language to translate risk limits into risk capital. It isn't 'right' or 'wrong', in the same way that Black Scholes can't be 'right' or 'wrong'. It's just a translation tool.

The problem is stress propagation in a market where each participant is using the same proportional sizing and the same model telling them when to cut, vs a market where sizes and cuts are distributed.

One is brittle, the other is viscous. No prizes for telling which one we have.

"There is a SIX am?" -- Arthur

gamerx


Total Posts: 19
Joined: Sep 2012
 
Posted: 2020-08-24 15:02
>Same trading strategies lead to swarm effects, it is clear, but can I say that risk models filter same risk profile (= trading strategies) and, therefore, can be the cause of the systemic problem too.

Interesting idea, but I doubt risk models really funnel everyone into the same strategies. Risk models cover such a finite number of dimensions that they are simply a drop in the ocean.

> And it's not just theoretical. Everybody who got hammered in the quant space this year seems to have been using some variant of Barra, at least based on my personal anecdotal experience.

Well....it is one of the industry standard. Most quants are using it, and most quants are probably not that different from one another especially during a quant crash. Doubt we can infer too much from that. If anything, the hammering this year really suggests that existing risk models are lacking and that many of their "alphas" are now betas or risk premia.
Previous Thread :: Next Thread 
Page 1 of 1