Notes for a speech to the Regional Bond Dealers Association

By Felix Salmon
April 23, 2009

Blogging will be light today and tomorrow, since I’m flying to and from the annual meeting of the Regional Bond Dealers Association in Dallas. They’ve asked me to give a speech: here are my notes.

Like everybody else at this conference, I’m sure, I’m here to talk about risk – the main part of what you all do for a living. You buy it, you sell it, you measure it, you underwrite it. But like most of us, I’ve changed my view on risk considerably over the past couple of years. And it seems to me that one of the biggest mistakes that we all made during the credit boom was that everybody overestimated the demand for risk, when in reality there was much more demand for safety.

I believed along with Alan Greenspan that when it comes to debt instruments in general, and credit derivatives in particular, “These instruments enhance the ability to differentiate risk and allocate it to those investor most able and willing to take it.”

But if you look at what happened in practice, the art of securitization always seemed designed to create ever-increasing quantities of risk-free debt. Banks thought they were selling loans and mortgages to people who wanted the risk, but they weren’t: they were carefully packaging those loans and mortgages into bonds carrying a triple-A credit rating. And people buying triple-A risk don’t want any risk at all.

So what happened to the risk? The answer is that it was essentially modelled away. And one of the most important enablers of that modelling is the subject of my Wired cover story, the Gaussian copula function.

But the story is bigger than one formula.

Consider the very first synthetic CDO: it was called Bistro, and it was issued by JP Morgan in the late 1990s. The bank sold off the risk associated with $10 billion in loans on its balance sheet – this was credit risk which the bank didn’t want, even though it valued its lending relationshipsvery highly. So it kept the loans on its book, and kept the relationships, but sold off the risk by using credit default swaps to structure a synthetic CDO.

But here’s the astonishing thing. The credit risk on that $10 billion in loans managed to somehow get squeezed into a CDO of just $700 million: there was no chance that all the loans would sour at the same time, and JP Morgan managed to persuade Moody’s that the CDO could be just 7% of the size of the underlying loan pool, while hedging all of the credit risk.\

And it gets better: of that $700 million, fully two-thirds carried a triple-A rating. And triple-A, for those of you who remember as far back as 2006, means “no credit risk at all” – it means “risk free” – it means “you’re only taking interest-rate risk”. Which makes no sense when the whole point of credit default swaps is to separate out credit risk from interest-rate risk.

Add in the double-A rated tranches of Bistro, and you have holders of less than $200 million of risky paper taking substantially all the credit risk on $10 billion of corporate loans. That’s less than 2%. And Moody’s happily signed off on this, for two reasons. One was that the business of rating structured-finance vehicles was highly profitable; the second was that their entire business and reputation was based on the idea that they could model credit risk. If they couldn’tmodel credit risk, then they couldn’t rate credit. And so they were backed into a corner, and forced to apply their storied credit ratings to structured products which were simply the logical conclusion of their own models, rather than the result of a fundamentals-based look at a certain credit.

If you take a step back, you can see what’s going on here. You and I and Alan Greenspan all thought that credit derivatives were wonderful things because they moved credit risk out of the hands of people who didn’t want it, like banks, and into the hands of people who did want it.

In reality, however, the appetite for risk was never nearly as great as we all thought. $10 billion of loans becomes less than $200 million of credit-risk instruments, and everybody else reassures themselves that they’ve managed to reduce their credit risk to zero, even as the people holding that $200 million in synthetic CDO tranches are reassured by their own single-A or triple-B credit ratings that theyaren’t taking a particularly large amount of risk either.

And of course you know what happens next: some bright spark invents the CDO-squared, which seems to reduce the total amount of risk even further. You take the mezzanine debt, the triple-B stuff, and you do all manner of securitization magic to it, and it turns out that you can turn most of that into triple-A paper, too!

Because it was all triple-A, no one felt much in the way of need to do any analysis of their own: it’s almost impossible to overstate the power of the laziness of the bond investor. You know this from your own work with municipal issuers: the reason for those monoline wraps is not because the issuers have a lot of credit risk, but because the investors are lazy, and don’t want to do their homework, and reckon they can get out of doing their homework so long as there’s a monoline guarantee. Essentially, they’re outsourcing their own job to the monolines. Which might be reasonable for a small retail investor, but is not a good idea if your job is to invest in fixed-income instruments which carry a higher yield than Treasury bonds.

Of course, we all know how reliable those monoline guarantees turned out to be – and that’s a related story. The monolines, just like the ratings agencies, believed far too much in the power of models.

But things didn’t go completely insane until the technology behind Bistro started being used on asset-backed bonds in general, and mortgage-backed securities in particular.

Once again, we have a situation where everybody is trying to farm off risk to everybody else, to the point at which everybody thinks that someone else has it. For one thing, virtually nobody ever even stopped to worry about credit risk in the MBS market – I know that I didn’t, until it was far too late. I believed what the professionals told me, which was that the only thing a mortgage-bond investor needs to worry about is prepayment risk, and that credit risk is a non-issue.

But even those people who did stop to worry about credit risk were rapidly reassured. Most mortgages were always sold to Fannie and Freddie – and, presto, all that risk magically disappeared. These were hugely profitable corporations, what could possibly go wrong?

Then of course there were the non-conforming mortgages, mostly subprime, which couldn’t get sold to Frannie. How could you securitize those? They all looked very similar, with the same originators and underwriters and loan-to-value ratios and underlying FICO scores and so on and so forth. And so the key aspect of securitization which allowed the ratings agencies to dole out triple-A ratings like so much confetti – diversification – would seem to have been missing.

At least in the Bistro deal, the underlying loans came from a broad and healthy group of companies. When people started securitizing subprime mortgages, the underlying assets were neither broad nor healthy. And so were the seeds of disaster sown.

Common sense says that you can’t start lending money to very risky borrowers without taking on lots of credit risk – but somehow, by the time the loans made their way through the system, almost nobody thought that they were taking on credit risk. Most of the participants in the market thought they had triple-A-or-better debt, and they all believed, without ever really stopping to check, that the enormous amounts of credit risk being produced were being willingly held elsewhere.

And so we come to David Li’s Gaussian copula function. What the copula did was, in effect, nullify the effects of common sense: it blinded bankers and traders and bond investors with quant science. The formula decided that you could measure the degree of diversification in a mortgage pool scientifically, by looking at a single number known as correlation. Correlation was treated as a constant – which was ridiculous on its face – and then the ratings could be derived from it. By plugging in a suitably low correlation number at one end, you could churn out triple-A ratings and healthy bond valuations at the other. And since the trade was so incredibly profitable for anybody who entered into it, no one asked too many questions, and everybody piled in.

Of course, on Wall Street, if everybody is making the same trade, that’s a tried-and-true recipe for bubbles and crashes, which is exactly what we got.

It turns out that while everybody was concentrating on credit ratings, no one spent nearly enough time worrying about model risk. All those models, including the Gaussian copula function, which were used to generate the ratings, turned out to have enormous flaws. For one thing, models generally try to describe some external reality – but in this case, the models were driving the reality, creating feedback loops which were entirely outside the ability of the modelers to comprehend or hedge.

More generally, the models were based on data from a period of time when nothing ever blew up, and as a result they had a tendency to produce results saying that nothing was ever going to blow up. Eventually, you got to the ridiculous situation where the chief financial officer of Goldman Sachs could get on a quarterly earnings call and talk with a straight face about 25-sigma events, as though such concepts had real meaning.

At this point, there might be a couple of you in this audience feeling just a tiny bit smug about all this. Sure, you’ve been hit by the financial crisis – we all have. But you never got into the mortgage securitization space, you never traded correlation, you never got blindsided by a multi-billion-dollar “liquidity put” you never even knew that you had written. In other words, it wasn’t your fault.

But it’s worth asking why the regional bond dealers managed to dodge the bullet. And the answer, I’m afraid, is basically that you got lucky.

For one thing, you don’t have massive balance sheets. JP Morgan, when it did its Bistro deal, was perfectly happy keeping $10 billion in assets on its balance sheet. In New York, at the time, big balance sheets were considered a good thing: they were ways of making lots of money, and even investment banks without a commercial bank attached – Goldman Sachs is the prime example here, but you could look just as easily at Morgan Stanley or Lehman Brothers or Merrill Lynch – had as much as $1 trillion of assets. Why did they need such an enormous balance sheet? No one really asked. But it did make it easy to hide things like super-senior risk.

Remember that the Bistro deal was only for $700 million, which meant that JP Morgan kept $9.3 billion of so-called “super-senior” risk on its own books – unless and until it managed to hive that risk off to AIG. Later entrants to the game, like Citi and Merrill, never bothered to sell much if any of their super-senior exposure, and when suddenly correlations spiked and mortgages across the country started defaulting at the same time, they realized that their models had been flawed, that they hadn’t sold off all their credit risk after all, and that they had hundreds of billions of dollars in risk so well buried in these trillion-dollar balance sheets that no one really knew it was there.

So, congratulations on not being huge. And congratulations too on largely avoiding the securitization/ABS space, which was mainly the province of the big banks with lots of warehousing capacity.

But if the next shoe does drop, it’s likely to be munis, and that’s bread and butter for a lot of you guys. Correlations can go to 1 in any market, not just ABS. And although the locus of the crisis was ABS, there’s no particular reason that it couldn’t have been munis instead.

A lot of people think that municipal bonds are just inherently very safe things, but we just don’t live in a world of “inherently very safe”. I’d highly encourage you all to get out your copy of the last Berkshire Hathaway annual report, where Warren Buffett talks about the risks in the muni market:

The rationale behind very low premium rates for insuring tax-exempts has been that defaults have historically been few. But that record largely reflects the experience of entities that issued uninsured bonds…

A universe of tax-exempts fully covered by insurance would be certain to have a somewhat different loss experience from a group of uninsured, but otherwise similar bonds, the only question being how different. To understand why, let’s go back to 1975 when New York City was on the edge of bankruptcy. At the time its bonds – virtually all uninsured – were heavily held by the city’s wealthier residents as well as by New York banks and other institutions. These local bondholders deeply desired to solve the city’s fiscal problems. So before long, concessions and cooperation from a host of involved constituencies produced a solution. Without one, it was apparent to all that New York’s citizens and businesses would have experienced widespread and severe financial losses from their bond holdings.

Now, imagine that all of the city’s bonds had instead been insured by Berkshire. Would similar belt- tightening, tax increases, labor concessions, etc. have been forthcoming? Of course not. At a minimum, Berkshire would have been asked to “share” in the required sacrifices. And, considering our deep pockets, the required contribution would most certainly have been substantial.

Local governments are going to face far tougher fiscal problems in the future than they have to date…

When faced with large revenue shortfalls, communities that have all of their bonds insured will be more prone to develop “solutions” less favorable to bondholders than those communities that have uninsured bonds held by local banks and residents. Losses in the tax-exempt arena, when they come, are also likely to be highly correlated among issuers. If a few communities stiff their creditors and get away with it, the chance that others will follow in their footsteps will grow. What mayor or city council is going to choose pain to local citizens in the form of major tax increases over pain to a far-away bond insurer?

To put it simply: if one muni defaults, that’s nasty for its creditors, including the monolines. And default is much more likely now than it was when most munis were unwrapped – insurance, as any insurer will tell you, is rife with moral hazard.

But if five or six munis default, things get much, much worse. At that point, the cost of default for a wrapped muni issuer plunges, and possibly even goes negative. Once a few munis default, no one’s going to lend to any muni, even the ones which are current on their debt. So why bother staying current? Why not just default and let the insurer, rather than your local taxpayers, take most of the pain?

In other words, there’s a very serious, and pretty much impossible to hedge, risk of snowballing muni defaults.

The fact is that the muni market is still heavily reliant on monoline wraps, which are at heart an artifact of the credit bubble, and of the fact that no one wanted to do homework or admit that they were taking risk. Those days are over now, and the new financial world which emerges from the current rubble is going to be one where investors are forced to face up to the fact that risk is endemic and can’t simply be modelled away.

What’s that going to mean for your business? I fear the news isn’t good. No fixed-income investors have the time to do detailed credit analysis on a regional hospital. That’s something banks do: these things should often by rights be loans rather than bonds. Which implies that we’re going to move back to a world of reintermediation, with less of a role for bond dealers and more of a role for boring bankers who know their clients and do their homework. And more generally, the financial sector is going to be a much smaller part of the economy than it has been over the past couple of decades.

So even if and when the economy rebounds, I wouldn’t expect your business to necessarily rebound with it. Ask yourself how many of your buy-side clients really want to analyze and buy substantial amounts of credit risk, which is the main product that you’re selling. Remember that they can’t be lazy any more, and rely on copulas and credit ratings and monline wraps, they have to do it all themselves. Do you see a business selling to these people? I hope so, because that’s going to be a large part of your job from now on, and I wish you all good luck.


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

betcha the MC uses the phrase “influential blogger” somewhere in his intro.

Posted by otto | Report as abusive

Felix: I’ve been following your blog for the past six months and I have to say that this is a really, really excellent summary of what happened. It is nicely tailored to the audience, too. The regional bond dealers are getting their money’s worth.

They might prefer that you wait until after the talk to post the notes, however.

Posted by Tentakles | Report as abusive

Might want to put some chicken wire up before telling them their business is not going to be picking up anytime soon ;-)

Posted by Chris | Report as abusive

Felix, you exactly true.
“One of the biggest mistakes that we all made during the credit boom was that everybody overestimated the demand for risk, when in reality there was much more demand for safety.”

I think, safety for our next burden.

Wow, I give you credit for be willing to deliver bad news. The temptation with these things is always to find a crowd-pleasing silver-lining. Kudos for sticking to your guns.

I think there’s an interesting question here about the magnitude of AIG’s role in all this. AIG became the toxic waste dump where all the unwanted credit risk was buried. Clearly some of the insiders knew that they had real risk, which is why they wanted CDS on AAA MBS in the first place. If AIG hadn’t been there to take all the risk, wouldn’t the system have backed up to the point the bubble would have been smaller?

While I (hopefully) have your attention, I think there’s an interesting parallel here to the “Winner’s Curse” phenomenon with auctions where the bidder who most overestimates the value of an item places the highest bid and thus wins the auction. When risk becomes tradeable, the bidder who most underestimates the risk involved will be the one who winds up holding the risk. Coupled with policies that allowed firms to essentially determine their own reserve requirements — this seems like a recipe for creating blow-ups.

Posted by DCreader | Report as abusive

“Of course, on Wall Street, if everybody is making the same trade, that’s a tried-and-true recipe for bubbles and crashes, which is exactly what we got.”

At the risk of constantly being the one kid in class who never gets it, as regards David Li’s Gaussian copula function, is it:

1) The 1st user thought that he was making money, which led people to follow him, but he was actually losing money. Eventually, the losses added up to a figure that was a very substantial loss to society.

2) The users were making money until a certain money-losing number of users was reached.

As for stocks, I’m having a problem understanding how everybody can be doing the same thing. For every buyer, doesn’t there have to be a seller?

I sold my house last year. I made money. Even though that house has probably gone down in value, the people who bought it are still in it because they can afford the payments, taxes, etc. When I bought that house, the price fell 10% to 15%, but I could afford the upkeep. I even petitioned for a decrease in my property taxes.

The problem has to be that people took out loans that they can’t afford to pay, and lenders gave out loans to people who can’t afford to pay. In other words, the problem was the amount of debt as against the resources that people had.

In the case of CDOs, surely the bottom line has to do with the situation that I just described. The problem is that people don’t have enough resources to make good on the claims against them. Again, it’s not really the model, it’s the lack of resources. People were trying to get in on a rising market on the cheap. That is, they were investing more and more with less and less backing, and lending to people with less and less backing. I’m sorry, that’s absolutely clear, whatever your models tell you about the future.

Suppose the investors had used an eight ball, a ouija board, a medium, etc., claiming that they believed that it worked. Would we be as forgiving of their mistakes? What is it about a math model that leads people to conclude that it mirrors or predicts reality? In that sense, what differentiates the people who invested in Madoff as opposed to AIG? They both simply trusted people and the methods that they claimed that they were using.

If the mis-interpreters of Li were followed because they believed that they were right, what possible law can stop people believing in other people.Laws could only help if the managers invested against sensible criteria or were committing fraud. There’s no law capable of stopping people from making bad investments or loans with people that they trust.

Bottom line, isn’t all that bond dealers are marketing is trust?

I have to quibble with the Bistro description. They didn’t persuade the rating agencies that they had hedged all the credit risk. JP Morgan retained the first loss of the pool as well as the super senior risk. What they did was hedge the risk above the first loss position to some level well above the expected loss, gaining capital relief on the difference between the threshold percentage and 8%. If they persuaded anyone, it was the regulators.

Also, anyone who thinks they aren’t taking much credit risk on a BBB bond is insane.

Posted by Ginger Yellow | Report as abusive

Incidentally, I never heard the notion that AAA was “risk free” until about two years ago, even in the sense in which treasuries are “risk free”. (I may be crazy wrong here, but I believe GM was rated AAA at the beginning of this decade, and it was certainly clear by 2005 that long-dated debt held four years earlier was not “risk free”, even in the approximate sovereign sense.) I also had never heard until a year or two ago that U.S. Treasuries were even rated, though I have since read on occasion that it’s AAA, and have read the complaint that therefore other AAA rated bonds are rated as “as safe as treasuries”. (This is like saying that, because Einstein had a high school diploma, anyone with a high school diploma has been rated “as smart as Einstein”. The mechanism simply doesn’t make the distinction; it’s a bit of a leap to then say that the mechanism is asserting that the distinction doesn’t exist.)

I very much doubt that, even when GM was AAA, its bonds ever traded at as low a yield as comparable treasuries.

I can hear the standing ovation now…

Felix, I’m supporting dWj over here. Sadly you’re creating wide-sweeping realisations without truly understanding what you’re talking about. Just about every paragraph is written convincingly, but the data that supports your argument is mostly false. For example: first CDO late 1990s. Um, no. AAA means only interest rate risk? Um, no. Also this had very little to do with Gaussian copula back in those days. I could go on and on but does it really matter. Anybody who can criticize somehow has an equally unsophisticated, applauding audience.

Posted by educated investor | Report as abusive

I can see that you are adding a new many work directly into your blog. Maintain publishing the great work. Some actually helpful tips in that room. Bookmarked. Pleasant to discover your site. Thank you!

Enormous breakfast with ready to vist web page while carries types of information which got seacrching