kenberg, on Mar 25 2010, 07:56 AM, said:
For the collaboration to work effectively, the mathematician has to ask the RWP what it is that he really wants to do, the RWP has to explain it, and the M has to take the time and make the effort to understand the explanation.
...
The people that work in the mathematics of finance are very smart people and they know all of this stuff that I just said. Still, it seems possible that at least some of the current difficulty springs from a very sophisticated version of the same basic issue.
I just re-read an excerpt from Gillian Tett's
book on this topic. It sounds like the "very smart people" were indeed onto the problem and trying to explain it and that the RWP's were not listening.
Quote
JP Morgan statisticians knew that company debt defaults are connected. If a car company goes into default, its suppliers may go bust, too. Conversely, if a big retailer collapses, other retail groups may benefit. Correlations could go both ways, and working out how they might develop among any basket of companies is fiendishly complex. So what the statisticians did, essentially, was to study past correlations in corporate default and equity prices and program their models to assume the same pattern in the present. This assumption wasn’t deemed particularly risky, as corporate defaults were rare, at least in the pool of companies that JP Morgan was dealing with. When Moody’s had done its own modelling of the basket of companies in the first Bistro deal, for example, it had predicted that just 0.82 per cent of the companies would default each year. If those defaults were uncorrelated, or just slightly correlated, then the chance of defaults occurring on 10 per cent of the pool – the amount that might eat up the $700m of capital raised to cover losses – was tiny. That was why JP Morgan could declare super-senior risk so safe, and why Moody’s had rated so many of these securities triple-A.
The fact was, however, that the assumption about correlation was just that: guesswork. And Demchak and his colleagues knew perfectly well that if the correlation rate ever turned out to be appreciably higher than the statisticians had assumed, serious losses might result. What if a situation transpired in which, when a few companies defaulted, numerous others followed? The number of defaults required to set off such a chain reaction was a vexing unknown. Demchak had never seen it happen, and the odds seemed extremely long, but even if there was just a minute chance of such a scenario, he didn’t want to find himself sitting on $100bn of assets that could conceivably go bust. So he decided to play it safe, and told his team to look for ways to cut their super-senior liabilities again, irrespective of what the regulators were saying.
That stance cost JP Morgan a fair amount of money, because it had to pay AIG and others to insure the super-senior risk, and those fees rose steadily as the decade wore on. In the first such deals with AIG, the fee had been just 0.02 cents for every dollar of risk insured each year. By 1999, the price was nearer 0.11 cents per dollar. But Demchak was determined that the team must be prudent.
Around the same time, the JP Morgan team stumbled on a second, potentially bigger problem. As the innovation cycle turned and earnings declined from the early Bistro deals based on pools of corporate loans, Demchak asked his team to explore new uses for Bistro-style deals, either by modifying the structure or by putting new kinds of loans or other assets into the mix. They decided to experiment with mortgages. Terri Duhon was at the heart of the endeavour. Only 10 years earlier, Duhon had been a high-school student in Louisiana. When she told her relatives she was going to work in a bank, they had assumed she was going to be a teller. Now she was managing tens of billions of dollars. She was trained as a mathematician, and she thrived on adrenaline, riding motorbikes in her spare time. Even so, she found the thought of being in charge of all those zeros awe-inspiring. “It was just an extraordinary, intense experience,” she later recalled.
A year after Duhon took on the post, she got word that Bayerische Landesbank, a large German bank, wanted to use the credit derivatives structure to remove the risk from $14bn of US mortgage loans it had extended. She debated with her team whether to accept the assignment; working with mortgage debt wasn’t a natural move for JP Morgan. But Duhon knew that some of the bank’s rivals were starting to conduct credit derivatives deals with mortgage risk, so the team decided to take it on.
As soon as Duhon talked to the quantitative analysts, she encountered a problem. When JP Morgan had offered the first Bistro deals in late 1997, it had access to extensive data about all the loans it had pooled together. So did the investors who bought the resulting credit derivatives, since the bank had deliberately named all of the 307 companies whose loans were included. In addition, many of these companies had been in business for decades, so extensive data were available on how they had performed over many business cycles. That gave JP Morgan’s statisticians, and investors, great confidence in predicting the likelihood of defaults. But the mortgage world was very different. For one thing, when banks sold bundles of mortgage loans to outside investors, they almost never revealed the names and credit histories of the individual borrowers. Worse, when Duhon went looking for data to track mortgage defaults over several business cycles, she discovered it was in short supply.
While America’s corporate world had suffered several booms and recessions in the later 20th century, the housing market had followed a steady path of growth. Some specific regions had suffered downturns: prices in Texas, for example, fell during the Savings and Loans debacle of the late 1980s. But since the second world war, there had never been a nationwide house-price slump. The last time house prices had fallen significantly en masse, in fact, was way back in the 1930s, during the Great Depression. The lack of data made Duhon nervous. When bankers assembled models to predict defaults, they wanted data on what normally happened in both booms and busts. Without that, it was impossible to know whether defaults tended to be correlated or not, in what circumstances they were isolated to particular urban centres or regions, and when they might go national.
Duhon could see no way to obtain such information for mortgages. That meant she would either have to rely on data from just one region and extrapolate it across the US, or make even more assumptions than normal about how defaults were correlated. She discussed what to do with Krishna Varikooty and the other quantitative experts. Varikooty was renowned on the team for taking a sober approach to risk. He was a stickler for detail and that scrupulousness sometimes infuriated colleagues who were itching to make deals. But Demchak always defended Varikooty. His judgment on the mortgage debt was clear: he could not see a way to track the potential correlation of defaults with any confidence. Without that, he declared, no precise estimate could be made of the risks of default in a pool of mortgages. If defaults on mortgages were uncorrelated, then the Bistro structure should be safe for mortgage risk, but if they were highly correlated, it might be catastrophically dangerous. Nobody could know.
Duhon and her colleagues were reluctant simply to turn down Bayerische Landesbank’s request. The German bank was keen to go ahead, even after the uncertainty in the modelling was explained, and so Duhon came up with the best estimates she could to structure the deal. To cope with the uncertainties the team stipulated that a bigger-than-normal funding cushion be raised, which made the deal less lucrative for JP Morgan. The bank also hedged its risk. That was the only prudent thing to do, and Duhon couldn’t see herself doing many more such deals. Mortgage risk was just too uncharted. “We just could not get comfortable,” Masters later said.
In subsequent months, Duhon heard through the grapevine that other banks were starting to do credit derivatives deals with mortgage debt, and she wondered how they had coped with the lack of data that so worried her and Varikooty. Had they found a better way to track the correlation issue? Did they have more experience of dealing with mortgages? She had no way of finding out. Because the credit derivatives market was unregulated, details of the deals weren’t available.
The team at JP Morgan did only one more Bistro deal with mortgage debt, a few months later, worth $10bn. Then, as other banks ramped up their mortgage-backed business, JP Morgan largely dropped out. Eight years later, the unquantified mortgage risk that had frightened off Duhon, Varikooty and the JP Morgan team had reached vast proportions. And it was spread throughout the western world’s financial system.
My wife and I have a saying: "just because I'm not listening doesn't mean I don't care". Fortunately, she is not a quant and I am not an investment banker, or vice versa.
If you lose all hope, you can always find it again -- Richard Ford in The Sportswriter
Help
