How AIG FP Brought Down the World

Michael Lewis is back in Vanity Fair, with a really good article about Joseph Cassano and AIG Financial Products. Or half of a really good article, at any rate — the second half of the story is fantastic, and gives by far the best English-language account of what happened at FP and how Cassano, by commission and omission, enabled it. (The piece isn’t online, just a précis — I do hope VF gives up this annoying habit soon.)

There’s a bit too much irrelevant throat-clearing at the beginning of the piece, though, and there’s also this very peculiar passage:

The public explanation of AIG’s failure focused on the credit-default swaps sold by traders at AIG FP, when AIG’s problems were clearly broader. There was the mortgage-insurance unit in North Carolina, United Guaranty, that had taken on all sorts of silly risks in the past two years, lost several billion dollars, and replaced their CEO. There were the fund managers at AIG, the parent company, who had blown nearly $50 billion in trades in subprime mortgages — that is, they had lost more than AIG FP, whose losses stood at around $45 billion.

Later on in the piece, after it starts getting good, Lewis explains how “if it hadn’t been for AIG FP the subprime-mortgage machine might never have been built, and the financial crisis might never have happened”. So doesn’t it make perfect sense for the public explanation of AIG’s failure to focus on FP? After all, as Lewis shows, FP literally had no idea what it was doing. He tells the story of Gene Park, who examined FP’s business at the end of 2005 and found that it was insuring deals which were 95% subprime:

Park then conducted a little survey, asking the people around AIG FP most directly involved in insuring them how much subprime was in them. He asked Gary Gorton, a Yale professor who had helped build the model Cassano used to price the credit-default swaps. Gorton guessed the piles were no more than 10 percent subprime. He asked a risk analyst in London, who guessed 20 percent. He asked Al Frost [FP's main liaison to Wall Street], who had no clue.

Cassano simply trusted the models (which were built with a lot of Gorton’s help), even when the models weren’t designed for subprime in particular, or even really for mortgages.

I guess the message of Lewis’s piece is that FP caused the global financial crisis, even if it didn’t necessarily cause the complete downfall of AIG — that AIG ended up buying in to the bubble created by FP, just companies like Citigroup and Bear Stearns did. Or, to put it another way, FP brought down the financial markets, and the crashing financial markets brought down AIG. You can blame the end of the world on Cassano, but there were a lot of people inside AIG but outside Cassano’s little group who ended up buying into the markets he helped to create and inflate.

Comments

[Abe Mishima:]

The financial history of the late 20th and early 21st centuries will be analyzed thusly. We trusted the stability of our financial system to complex pieces of software, written by PhDs and Masters holders who had not one bit of common sense. They gave these programs to nitwits who had no idea of how to use them. The kind of people who brush off technicalities with statements like, "Don't bother me with the details." Imagine Homer Simpson at the helm of a control board in a nuclear power plant. Makes you shudder, doesn't it? It shouldn't. The software for controlling a nuclear power plant is idiot proof. CDO and CDS modeling software is not. To the extent the NRC regulates nuclear power, there should be a Department of Homeland Financial Security that controls the investment industry. What a shame the disaster had to happen!

Jul 02 05:47 PM

[nym:]

Available online www.scribd.com/doc/170...

Jul 02 08:53 PM

[derryl:]

Madoff's investors did not carefully scrutinize his model because he gave them 12% returns. Why would anyone care what's in the financial products they are selling when they are getting millions of dollars in bonuses for selling them? Even if somebody sees that the models can't work, the bonuses are real. You get what you pay for.

Jul 03 01:02 AM

[Rhunzzz:]

How about sewing Cassano into a sack and dumping him into the Thames?

Jul 03 02:14 AM

[nobby73:]

Most risk systems work in a similar way as if using historic weather patterns in July to predict probability of snow in January. Every time the financial markets go through a trauma, whether it be LTCM the S&L disaster, Russia, the dotcom crash or any other going back centuries, there is a collective wiping of memory, a clearing of the slate and a belief that we are entering a new future and we will be smarter this time. Those who seem too affected by it are cast out, considered burnt out or too bearish.

The problem is, nothing makes people more stupid than a trade that moves in the right direction. They lose all doubt and they think that if the models are working, i.e. low variance and regular profit, they have cracked it. What is the alternative? They are making money for the firm and for themselves and if they have little reason to question things. Unfortunately, it does not seem the main criteria for senior management in US financial companies is for them to be smart enough to understand the flaws in risk methodology. In my experience, good and brave management in finance is when they say we need to cut this lucrative business, because it is suspiciously good and the markets are in stampede mode, so if this turns bad, we will be stuck when liquidity vanishes.

In fact, very few risk models I have seen accurately assess this fundamental feature of trading positions - when the markets turn, it's not the falling prices that get you, it's the illiquidity gaps. The scientific term for the appearance of stability and order prior to catastrophic collapse is "critical steady state", which is a bit like dense traffic on the freeway all travelling at high speed. The traffic flow appears smooth and orderly, but it only takes one vehicle to break to set of a chain reaction which will bring everything behind to a halt.

In my view, part of the problem has been that finance has taken certain concepts of dynamic systems from science, but maybe it did not use the right ones, i.e. ones that involve human behaviour rather than random particles.

AIG for example took great comfort in the fact the market appeared to be highly liquid, without realizing it was themselves who made a key component of the liquidity.

Jul 03 05:09 AM

[john s. gordon:]

computer models can replicate the data that were fed into them when they were built.
the really interesting phenomena occur outside the boundaries of the initial data base.
interesting parallel here - the computer software on the crashed air france airbus couldn't handle all the conflicting inputs that were being thrown at it.
> jack

Jul 03 08:46 AM

[John Lounsbury:]

Felix - - - Excellent article.
nym - - - Thanks for the link to the full article.
nobby73 - - - Great tutorial. The information you provide explains why models need to be tested extensively with "out of sample" data. The truly wise also use "stress tests", i.e., out of sample data with hypothetical distortions. The result of following this procedure is that many seemingly good models would be discarded. Perhaps that is why so many models are implemented without exhaustive testing. And why we keep repeating disasters. I like to use models, but always with a close monitor for when reality shows relatively minor divergence from the model prediction. Yes, I get stopped out prematurely from a modeled trade at times, but I avoid going down in flames. Finally, I don't run the liquidity risk that a major player like AIG does.
John Gordon - - - You understand everything I just said. I hope you don't object to my expanding on the subject.

Jul 03 11:49 AM

[paulsjj:]

Wall Street built a machine.
Alan G. turned the crank.
AIG & GS gamed the mart,
and blew the whole d... thing apart.

Jul 03 11:57 AM

[BerkeleyBob:]

If you check the past few issues of Wired, there was an interesting article which I did not fully grasp about the risk prediction algorithm. Think the article was titled "The Equation That Failed to Predict". The central point was that the algorithm was mis-applied.

Jul 03 11:59 AM

[LIGUY73:]

AIG FP did not bring down the World. The Sub prime mortgage did. Saying otherwise is a miscarriage of journalistic standards.

What is wrong with you people!!

Jul 06 10:03 AM

[csyy:]

sorry, here is a dull question, do you think AIG will back to $30?

Jul 06 01:52 PM

[caspar:]

Does anybody know what's happening in Europe with AIG Europe S.A. ? I believe they are trying to get seperated from AIG (US) ?! Currently, I am trying to get a clear picture of the whole situation at AIG Europe S.A., you know the whole spiderweb with little daughter holdings here and there, a small money transfer here and there... anybody knows anything about this?? gr. Caspar

Jul 07 03:32 AM