The 2007-08 Subprime Mortgage Meltdown, Personally

Life on a trading floor losing $45 billion, physicists modeling mortgage loan defaults, telling John Paulson to short mortgages, flak for revealing Ambac’s losses, and the best subprime analyst.

When Laurie Goodman hired me into UBS’ structured products research team in 2001, I became a colleague of Tom Zimmerman. Tom is an unsung hero of the subprime mortgage meltdown. He started raising alarms about residential mortgage credit quality in 2005. In February 2007, he published a research report naming six triple-B minus subprime bonds he said were going to default.[1] For the next two years, as the mortgage crisis deepened, Tom and his team (David Liu, Shumin Li, and Rei Shinozuka) updated their published list of predicted defaults. For example, in April 2008, Tom predicted the defaults of 35 subprime bonds at that time rated triple-A, 45 rated double-A, 28 rated single-A, and 57 rated triple-B.[2] No other sell-side analyst published such forecasts or was so publicly pessimistic on subprime mortgages as Tom.[3] But Tom was prescient in all his default predictions.

For more on Tom’s predictions versus the credit rating agencies, see Getting Rid of Issuer Pay Will Not Improve Credit Ratings, But Maybe Investor Revenue Will.

Tom’s model was simple; akin to predicting how tall a child will grow from how tall he is now. For example, an 18-month girl will about double her height by the time she is full grown. Historically, a mortgage portfolio’s lifetime defaults are five times the one-year statistic. Tom took the defaults a mortgage portfolio had already experienced and added his calculation of coming defaults from mortgages currently in arrears. He then multiplied these defaults by a factor that depended on how seasoned the portfolio was. For example, he multiplied by five if the portfolio was one-year old. Later, Tom adjusted his model as mortgage lenders stopped making subprime loans and it became impossible to refinance floating-rate loans when teaser rates expired and required payments increased.

In 2008, UBS’ senior management had lost faith in UBS risk management’s Pollyannish view of mortgage losses, but they still didn’t accept Tom’s pessimistic view. Management hired a half dozen statisticians and physicists to model mortgage losses. These guys’ mother tongues were Chinese, Russian, and stochastic calculus; and they rented a supercomputer to crunch the millions of datapoints they were using to predict defaults.

After a month of work, they made a presentation to UBS’ mortgage traders, salesmen, and analysts. The first objection came from the UBS executive in charge of buying mortgage loans who said their model had the wrong sign on borrower FICO scores. It should be positive, he said. The Russian explained that the negative coefficient meant that as FICO increased, indicating better borrower credit quality, mortgage defaults declined. The mortgage buyer said he understood what the sign meant, but in his firsthand experience, higher FICO scores were associated with greater default incidence.

The physicist’s model and the mortgage loan buyer were both right. Mortgage originators offered loans with less conservative underwriting requirements, such as higher debt ratios and no income verification, to people with higher FICO scores. It turned out in 2007 that less conservative underwriting trumped higher FICO scores, and these loans defaulted more frequently than more conservatively underwritten loans with lower FICO scores. But the physicists had taken a kitchen sink approach to independent variables and searched loan tapes for every possible explanatory variable. They incorporated borrower debt ratio, income verification, and other underwriting characteristics so, after accounting for those variables, the sign on FICO score should be negative.

But Tom’s team noted a genuine problem with the physicist’s model: much of the data they relied on was fraudulent. On their mortgage applications, people frequently lied about their income and whether they were buying the house to live in or as an investment. Such fraud was particularly prevalent among loans that defaulted, exactly the loans the physicists were trying to identify statistically. If they had talked to Tom, the physicists would have learned this before they started regressing variables.

Besides the bad data, the physicists’ efforts were hampered by the lack of historical experience necessary to model the new types of loans being made. They had data on the loss experience of subprime mortgage loans with low FICO scores or low downpayments or adjustable-rate resets or no income documentation. But rather than incorporating one of these risks, newer mortgage loans incorporated two or more of these risks simultaneously. The physicists had no historical data to tell them whether these risks were additive in terms of producing losses or multiplicative!

Moreover, the physicists had no data on how these risks produced losses in the environment of drastically falling home prices that was occurring in 2007-09. Nor did they understand how people’s willingness to pay their debts had changed. Falling house prices meant that some borrowers had negative equity in their homes, the value of their home was less than the mortgage balance, and these people had an economic incentive to hand the house keys to the bank and walk away. Or stop making mortgage payments and live in the house rent free until evicted.

Meanwhile, Tom’s approach focused on the actual performance of the loans rather than reported loan characteristics and the status of borrowers at the time of loan origination. In the end, Tom’s predictions proved much more accurate than the physicists’ predictions.

As UBS’ subprime-bond-backed CDO (collateralized debt obligations) analyst, I was the beneficiary of Tom’s analysis. It was logical that I use Tom’s subprime bond predictions in my CDO analysis as the relationship between the two is mathematical. When the subprime bonds that the CDO owns default, the CDO can’t pay its own obligations. Because Tom was right and early about subprime bonds, I was right and early about CDOs. I owe my success in analyzing CDOs to Tom.

UBS eventually took a $45 billion loss from subprime-mortgage-related securities. An outside law firm, hired in 2008 to analyze what went wrong, opined “there appears not to have been sufficient discussion of, or actions upon, concerns surrounding subprime … even though UBS’s research team issued research reports on this area.”

The Time I Upset People Over Ambac’s CDO Losses

In November 2007, UBS’ equity analysts hosted a conference on bond insurance companies. MBIA, FGIC, and Ambac all insured subprime-backed CDOs, so the equity analysts wanted me to speak on the risk to the insurers. I took a quick look and didn’t find much disclosure by MBIA and FGIC. But Ambac published the names and amounts of most of the CDOs they insured.

Looking at the CDOs Ambac insured, I knew losses would be in the billions, so I told the equity analysts that that was what I was going to say if their invitation still stood. That was fine with them, and I did a careful analysis of Ambac’s CDO portfolio using Tom’s subprime bond by subprime bond loss predictions. I came up with $3.75 billion of CDO losses for Ambac.[4] Mind you, Ambac only had $284 million of loss reserve, $5.6 billion of equity, and other exposures to subprime mortgages. Before the conference I met with Ambac executives at their office and didn’t hear anything to change my view.

The first sign of the furor I caused was an angry call from an Ambac equity investor I fielded right after I gave my presentation. I, and my analysis, were “reprehensible.” I must be part of an illegal effort to short the stock and talk it down. Later, at an investor conference in December 2007, Ambac’s CEO expressed his view of my analysis. He accused me of looking at their CDO portfolio as if the CDOs held only 2006 vintage mortgage loans instead of looking at their actual portfolios, “…it was really outrageous. They were in our office like a day or two before [my presentation]. They knew exactly what was in that portfolio.”

But potentially the worse feedback for me professionally came from UBS’ municipal bond department. They cared about what I said about Ambac and other bond insurers because many of the municipal bonds they traded and held in inventory were insured by these companies. I hadn’t given them a heads-up about my talk. I should have, but frankly, I didn’t think about municipal bonds!

What made the situation worse was that amid the backlash, Moody’s publicly affirmed Ambac’s Aaa rating four weeks after my talk and written report. S&P also stuck to their AAA Ambac rating. My critics argued that surely these two teams of credit analysts, whose sole job was to analyze bond insurers, had a better grasp of Ambac’s credit quality than a CDO analyst who admitted to only spending a couple day’s effort. But Moody’s base-case loss assumption for 2006-vintage subprime mortgage loans was 11% and its stress-case loss assumption for that same vintage was 19%. I guess S&P’s loss assumptions and AAA stress-case losses were similar.

These base and stress case losses fed into Moody’s and S&P’s subprime bond and ABS CDO analysis. The practical effect of the assumptions was that to retain a AAA rating, a bond insurer had to be able to withstand losses in its insured portfolio arising from 19% losses on 2006 subprime mortgage loans.[5] Meanwhile, Tom base-case loss assumption for 2006-vintage mortgage loans was 19.7%. Tom’s base case prediction was higher than the stress test Moody’s (and I guess S&P) used to award triple-A ratings! The difference between Tom’s expected losses on 2007-vintage loans over the rating agencies’ triple-A stress case must have been even greater.

But my boss, Laurie Goodman, had my back against my critics. She took a call from a managing director in UBS’ municipal bond business, and I could hear her side of the conversation across our trading floor desk. Several times, “Well, that’s his opinion.” And then finally, “No I don’t think we need a meeting, there’s nothing to discuss.”

For more on Laurie’s management of UBS’ securitized products research group during the subprime mortgage credit crisis, see Sell-Side Research Independence in Theory and Practice: Analysts and Mortgage Traders Clash in 2007.

In January 2008, a couple months after my talk, Ambac announced a $3.5 billion write down, plans to raise $1 billion in equity, and the departure of its CEO, the guy who called my analysis “outrageous.” In March, the Ambac executives who back in November told me everything was great with their CDO portfolio called me up. Could they publish my November article detailing my $3.75 billion CDO loss prediction on Ambac’s website? I told them to wait and re-ran my analysis using Tom’s latest loss predictions. Now my prediction of Ambac’s CDO losses was $4.7 billion. I told the Ambac guys they couldn’t post my earlier out-of-date prediction. It eventually became clear, even to the rating agencies, that Ambac was not able to pay all its insured claims. In July 2009 (20 months after my speech!), Moody’s and S&P downgraded Ambac from investment grade to Caa2 and CC, respectively.

The Time I Told John Paulson to Short Mortgage Credit

By December 2007, Tom’s predictions of subprime bond defaults were astronomical. For example, he predicted 2006 triple-B subprime bonds were going to experience 89% principal loss. Using Tom’s subprime bond loss model and applying it to 208 CDOs comprised of triple-B and single-A subprime bonds, I found that 86% of senior triple-A CDO tranches would default with losses averaging 51%. Across defaulting and non-defaulting CDOs, expected loss was 43%.[6] These soon-to-default CDO tranches were still rated triple-A by the credit rating agencies.

No other CDO analyst was presenting figures like mine, and I got a lot of negative feedback. It wasn’t appreciated that my report came out as investors and banks were marking their portfolios to market for year-end financial reports that ultimately affected bonuses. But one person who liked my article and asked for a meeting was a very significant investor, Paolo Pellegrini, a partner at Paulson & Co.

It was well-known that Pellegrini and John Paulson had made $15 billion for their clients in 2007 by shorting mortgage credit. They had also made billions of dollars for themselves. So I was excited to speak to Pellegrini. I was even more excited when Paulson joined our meeting.

I gave them an overview of how I had applied Tom’s subprime bond analysis to CDOs and went through my results. But Paulson was having none of it. He said my results were ridiculously optimistic. Every senior triple-A CDO was going to default, not 86%, as I had determined, and their losses would be far greater than the 51% I predicted. I was sort of happy to hear this critique since up to then I had only heard how overly pessimistic I was.

But Paulson’s critique of my work continued, and he was repeating himself. It became a tirade only interrupted when he turned on Pellegrini for inviting me to their office and wasting their time. He sarcastically asked Pellegrini if he wanted to buy CDOs. I defended myself as best I could. I suggested that even if my results were as wildly optimistic as Paulson said, they still ranked CDOs by how bad they were. That might be useful.

Still Paulson was not appeased. I was a Pollyannish shill for subprime mortgages and subprime-mortgage-backed CDOs. I was ignorant to the true dimension of the destruction. He kept going until I had enough and said, “If you think it’s going to be so bad, maybe you should short mortgages.” I guess I said it with a perfectly straight face because Paulson, Pellegrini, the two Paulson analysts in the meeting, and the UBS salesman who had brought me all looked at me dumbfounded. After several seconds of silence, Paulson practically screamed “I am short mortgages.” I quietly said “Yes, I know that,” and the meeting ended.

When I got back to my office, Laurie asked me how the meeting went with Pellegrini. She was impressed Paulson had joined the meeting. Then I told her that Paulson had explained to me how stupid I was. Laurie said that Paulson had explained that to her, too, and Tom was the only analyst he thought had any intelligence.

Besides how dumb I was, I learned something else from Paulson. Security firms had approached Paulson & Co about buying super senior CDOs at 70 cents on the dollar. The incentive was that the security firms would finance the CDOs at 60 cents on the dollar on a non-recourse basis. This meant that Paulson could return the CDO to the security firm and Paulson’s debt would be extinguished. Paulson’s hedge fund would essentially be paying 10 cents for an option on the CDO struck at 60 cents. It would make money if it could get 70 cents of value from holding or selling the CDO. Paulson of course thought the CDOs weren’t worth anywhere near 70 cents and turned them down.

Paulson said the security firms were doing this to get the CDOs off their balance sheets. Which struck me as an odd accounting result. When it was clear the CDO was not worth 60 cents, the CDO would be put back to the security firm and wind up on its balance sheet again.

But was Paulson right? Were my CDO loss predictions outrageously low? Yes, it turns out. Over 2008, Tom just about doubled his loss estimates on mortgage loans. In 2012, the Philadelphia Fed did what is maybe the best autopsy of senior triple-A CDO losses. In the best apples-to-apples comparison to my December 2007 43% expected loss figure, the Fed estimated 67% - 76% losses.[7]

Working Through the Crisis

As you might guess, the mood on UBS’ mortgage trading floor at 6th and 51st in Manhattan got gloomier and gloomier over 2007-08. Monthly trustee reports noted more and more delinquencies, defaults, and losses. Salesmen, traders, and structurers had time on their hands as business ground to a halt. There were successive rounds of layouts, with each round producing “the team we are going forward with.” UBS’ stock price fell from $65 to $7, making the deferred compensation we had earned in previous years worthless.

Excess time and Wall Street’s morbid sense of humor combined with the desire to put our losses into relatable context. When we were halfway to the $45 billion loss the mortgage trading floor eventually recognized, a salesman framed the amount in terms of gold. He said it was equivalent to a stack of Krugerrands 55 miles high or a block of gold measuring 16 feet by 16 feet by 16 feet.

A trader thought about how much work it would take to throw away that much money. If we busted out our windows on the 11th floor, it would take five guys eight hours to pitchfork enough $100 bills onto the street below to get rid of that much money. An economist wondered what midtown inflation would be if we did throw that much money out the windows. He thought a falafel sandwich at the stand on 6th Avenue would cost $1 million. Morbid calculations, gallows humors, and watching a couple in the Time-Life office building across 51st Street have afternoon sex broke the monotony.[8]

Meanwhile, my colleagues and I in research had a lot to do. The relentless stream of bad news brought opportunities to say something important to investors. I got some coverage in the popular press for my August 2007 summary of the meltdown: “The greatest ratings and credit risk-management failure ever.” These were the most exciting and most productive two years of my career, even if they weren’t good remuneratively. It was also like being in a 24-month disaster movie. I was finally laid off in November 2008 when UBS exited the CDO business.

Looking Back on It

When I think about the Subprime Mortgage Meltdown, or the Global Financial Crisis, or the Great Recession, whatever you want to call it, I first picture a mortgage broker and a couple sitting around a kitchen table in a small apartment. “I can get you into your own home for about what you are spending on rent. I’ll show you how to fill out the loan application” says the broker. Right there, around that kitchen table, at that specific time, is where the lies began.

And then I think of all the lies and liars downstream of that kitchen table: originating banks, security firms, asset managers, rating agencies, insurers, investors, and my former profession, sell-side securities analysts. Home ownership rose to 70% of American families from 65%; a five-percentage-point increase in American dream realization! And trillions of dollars in losses and uncountable human suffering resulted. One of the worse things humans have ever done to one another without using weapons of war.

Besides publishing the stories of finance professionals on this website, Doug fund-raises to stage George Balanchine’s seldom-performed ballets at BalanchinePatrons.org, and undertakes consulting engagements. Some of his articles on CLOs, CDOs, other structured finance products, default correlation, credit analysis, rating agencies, and George Balanchine can be found on Academia and Social Science Research Network.

Comment from Joe Pimbley: “Excellent story! I love the history! I congratulate you on the quality of your work and your professional steadfastness (if that's the right word). The anecdotes that stand out for me are (i) the Ambac loss projection with the flak from the UBS muni group and the ABK CEO and (ii) the meeting with Paulson. A funny thing is I've never met Paulson and wouldn't be able to pick him out of a lineup. For no good reason, apparently, I pictured him as a highly clever, under-the-radar investor. But why was he such a butt in your meeting? He would have gotten far more value from his conversation with you by asking polite questions about your analysis such as “what scenarios could make subprime losses even worse than you project?’”

- - - - -
[1] “A Simple ABX Loss Projection Model,” in UBS Mortgage Strategist, 27 February 2007, page 32.

[2] “Rating the Rating Agencies on Subprime” in UBS Mortgage Strategist, 1 April 2008, page 23.

[3] Chris Flanagan at JPMorgan named 11 predicted bond defaults in February 2007, but was never again so publicly specific. I suspect he was more forthcoming with subprime investors in private.

[4] “Monoline CDO Losses” in Monolines and Mortgage Insurers: Implications from the Subprime Crisis, UBS Lunch & Learn Series, 19 November 2007.

[5] “Moody’s Rating Actions on Monoline Financial Guarantors” in UBS CDO Insight, 16 December 2007, page 2.

[6] “Mortgage and ABS CDO Losses” in UBS CDO Insight, 13 December 2007, page 8.

[7] Larry Cordell, Yilin Huang, and Meredith Williams, Working Paper No. 11-30/R Collateral Damage: Sizing and Assessing the Subprime CDO Crisis, Federal Reserve Bank of Philadelphia, May 2012. Mr. Cordell updated his paper in Larry Cordell, Greg Feldberg, and Danielle Sass, “The Role of ABS CDOs in the Financial Crisis” The Journal of Structured Finance, Spring 2019. I use statistics from the 2012 paper because it breaks out senior AAA CDO losses.

[8] The afternoon sun reflected off our south-facing windows turning them into mirrors. Maybe the couple thought if they couldn’t see us, we couldn’t see them. Or maybe they volunteered us into their exhibitionist kink. Not that anyone across the street was affronted. Some of the salesmen brought binoculars to the office.

To comment on a story or offer a story of your own, email Doug.Lucas@Stories.Finance

Copyright © 2024 Douglas J Lucas. All rights reserved. Used here by permission. Short excerpts may be republished if Stories.Finance is credited or linked.