Credit models get even more complicated

By Felix Salmon
August 10, 2009
This is the kind of thing which makes Nassim Taleb tear his hair out. The 32-page, equation-filled IMF paper on new developments in credit risk modelling (which, yes, spends a lot of time on Gaussian copulas) seems to accept as an article of faith that the problem with credit risk models was that they weren't sophisticated and complicated enough.

" data-share-img="" data-share="twitter,facebook,linkedin,reddit,google" data-share-count="true">

This is the kind of thing which makes Nassim Taleb tear his hair out. The 32-page, equation-filled IMF paper on new developments in credit risk modelling (which, yes, spends a lot of time on Gaussian copulas) seems to accept as an article of faith that the problem with credit risk models was that they weren’t sophisticated and complicated enough.

One thing I’m quite sure of: the kind of equations being flung around in this paper are not going to be intuitively understood by underpaid regulators. This could then constitute a good test of any new regulatory regime: will the regulators roll over and say “oh well you’re very clever we’re sure you know what you’re doing”, or will they slap down any attempt to use these newfangled models to persuade regulators that everything’s perfectly safe? I do hope it’s the latter.

Update: A reader finds a classic quote on page 19 of the paper:

With sufficient detailed information, it is possible to design an optimal algorithm that generates contagion in the network and, therefore, correlation and multiple defaults, following the failure of one or more firms to honor its liabilities.

Yes, people still believe in “optimal algorithms”, even now, even after all we’ve been through.


We welcome comments that advance the story through relevant opinion, anecdotes, links and data. If you see a comment that you believe is irrelevant or inappropriate, you can flag it to our editors by using the report abuse links. Views expressed in the comments do not represent those of Reuters. For more information on our comment policy, see

actually, this paper is mild on the math. anyone working in credit needs to already know this (e.g., Merton)…math isn’t a bad thing…

it is pretty advanced for the basel II gnomes, remember they think that correlations are STATIC…

Here is a reality check on precision implied in credit risk modeling. implied-precision-in-pd.html

David, I think anyone working in credit needs to actually try to understand the company whose risk they are taking.

When I look at these papers, they seem very clever and interesting. As ways of looking at these investments and risks, they could be useful in a heuristic sense. But I’m amazed that people invest using these models. Although using math, they don’t seem to generate precision. On the other hand, as I said, they are interesting ways of viewing these investments. But I’m no expert. I do think that everyone would get something out of reading the paper.

However, I’m sure that I could find people who could help me more clearly understand the points being made and how dicey they are for real world use. I would certainly do this before allowing someone to spend my money, especially if I had a lot.

Some of these equations are the foundations of Basel II, so the regulators would know them.

Over-reliance on model in socials sciences is bad, but turning regulators and bankers into Luddites is not much better either

Posted by MrM | Report as abusive

Felix, the FT had a good editorial on the prevalence of deferring to numbers in executive decision-making today. Makes some points that are germane here…

Posted by Eric Dewey | Report as abusive

Just remember ladies & gentlemen: GiGo. You could have David Shaw and James Simons working at the SEC or any other regulator but the complexity of the models/math is irrelevant when the assumptions are rooted-in nonsense. I remember doing refis and acquisitions where debt/EBITDA (sometimes just senior debt!) was modeled to ramp-up to 12x post-deal, and these weren’t LBO deals, they were corporate! A few tweaks in the underlying assumptions though and the deal got down to 6x with 4 strokes of the keyboard.

I don’t have the answer (yet?) but I’m quite certain there is, and will continue to be an inverse correlation between complexity of teh maths/models and regulators’ ability to actually do their job.

Who needs these fancy credit models when internet aggregators can provide you with a free credit score

Oops – sorry wrong thread.

Posted by Doug | Report as abusive

This is another alarming sign that the Panglossian paradigm of efficient markets and the magic equations that describe them have survived the Great Recession. Indeed, the hydra has grown additional heads, or at least terms…

Posted by The Scientician | Report as abusive

William: tell me more about this “understanding the company you are lending to” idea of yours; I believe that it may have applications to equity investment.

Posted by dsquared | Report as abusive

right. on the other hand, modeling risk isn’t a complete waste of time, and I’d rather have models that incorporate “contagion in the network” than not. What you make of these models, is another matter.

Posted by Luis Enrique | Report as abusive

The math allows us to deceive ourselves and pretend to know much more than we do.

Posted by Dan | Report as abusive

Felix, an interesting perspective on this dilemma may come from the hard sciences. Regulators were designed and commissioned to regulate the activities of relatively benign financial institutions, with tools more akin to Newtonian physics than the Quantum Calculus that now seems to be required.

Management must seek to understand the implications of their decisions. When a portfolio consists of 6,000-10,000 loans, that is one thing – mundane, predictable, and “modelable.” However, when the portfolio and securitizations and hedges value at 1,000 or 10,000 times the size of that “Newtonian” portfolio, something changes in the game. A model can predict behavior, but it will, under no circumstances, provide any more clarity than Newtonian calculus in an Einsteinian context.

Just a thought, but perhaps we had best prepare ourselves for the contingency that behemoth banking is simply beyond prudent capability – I am not saying we can’t do it, we clearly are doing it – rather I am saying that it isn’t a good idea, systemically.

Limitations. We have them.

Posted by Ken | Report as abusive

I keep reading that quote in the update, and failing to understand. It seems to be saying “We can do better at causing widespread disaster,” but that makes no sense at all. Are they perhaps pointing out that the downside effects of the algorithms also have to be taken into account?

Posted by Ken | Report as abusive

By the way, when I say that these models can be interesting, I’m referring to this paper by Li especially: Default%20Correlation-%20A%20Copula%20Fu nction%20Approach.pdf

Here’s the point:

“There exist three methods to obtain the term structure of default rates:
(i) Obtaining historical default information from rating agencies;
(ii) Taking the Merton option theoretical approach;
(iii) Taking the implied approach using market prices of defaultable bonds or asset swap spreads.”

It strikes me that looking at these products and risks through these various methods can be quite interesting, even if dubious for actual investing.

Read it last night. Turns out to be a nicely-written & succinct primer on the state of credit risk models; including highlighting drawbacks (e.g., Gaussian copula).

highlights @ e/imf_on_recent_advances_in_credit_risk_ modeling_credit/

“seems to accept as an article of faith that the problem with credit risk models was that they weren’t sophisticated and complicated enough.”

This freakes me out. Looks like people do not understand that complexity generates chaos. Conversely, to fight chaotic behaviour one needs regulation and simplicity…