LemonsAndCDOsWhyDidSoManyLenders Preview

Preliminary Comments appreciated!
Lemons and CDOs

Why Did So Many Lenders Issue Poorly Performing CDOs?

Oliver Faltin-Traeger
Blackrock, Inc.

Christopher Mayer
Columbia Business School & NBER
Visiting Scholar, Federal Reserve Bank of New York
[email protected]

December 28, 2011

The authors wish to thank Patrick Bolton, Edward Morrison, Doron Nissim, Tomasz Piskorski,
Edward Vytlacil, Daniel Wolfenzon, and seminar participants at Columbia Business School,
Columbia Law School, and Harvard Business School. The opinions in this paper reflect our own
views and neither our colleagues nor the Federal Reserve Bank of New York are responsible for
any flaws or omissions in the analysis. The Paul Milstein Center for Real Estate at Columbia
Business School provided critical funding to support this research.

Lemons and CDOs
Why Did So Many Lenders Issue Poorly Performing CDOs?

Abstract
Collateralized Debt Obligations (CDO) played a key role in the growth of Asset-Backed
Security (ABS) issuance between 2004 and 2007 by providing a mechanism for lower-rated ABS
to be used as collateral for the creation of AAA securities. Using a database published by
Pershing Square Capital Management covering all of the assets underlying 528 CDOs and CDO-
Squareds issued from 2005 through 2007 and using rating history and other information from the
ABSNet database, we compare the characteristics and performance of ABS observed in a CDO
with other ABS not observed in a CDO. We find that CDO assets tend to be lower rated
securities from the lowest quality asset classes and vintages, and with a higher spread at
issuance. CDO assets performed much worse than comparable securities that were not included
in a CDO. When we control for the initial rating, CDO assets have a downgrade severity that is
at least twice as bad as comparable ABS not included in a CDO. Synthetic CDOs assets perform
worse than cash CDO assets, but assets included in both cash and synthetic CDOs perform worst
of all (with a downgrade severity about two and one-half times worse than the average
downgrade severity). Even when we include controls for a wide variety of observable
characteristics, including initial yield, CDO assets still underperform comparable ABS by
between 50 and 100 percent. These results suggest that CDO originators successfully sold
securities and insurance against the worst performing ABS assets, but also that buyers of CDOs
would have had a hard time analyzing these securities based on observable characteristics alone.
1
1. !"#$%&'(#)%"
They structured like mad and travelled the world, and worked their tails to make some
lemonade from some big old lemons.
- Former head of Goldman Sachs mortgage department in an internal email released
during the U.S. Senate Permanent Subcommittee on Investigations hearings of April 27,
2010.

Recent headlines regarding the exceptionally poor performance of collateralized debt
obligations (CDOs) were highlighted by SEC charges against Goldman Sachs for defrauding
investors by misstating and omitting key facts about a CDO transaction, Abacus 2007 AC-1.
1

The litigation served to highlight the exceptionally poor performance of a group of assets whose
issuance ballooned with the growth of private asset-backed securities (ABS) between 2004 and
2007 (see Figure 1). According to the Securities Industry and Financial Markets Association
(SIFMA), issuance of US dollar denominated CDOs more than doubled in 2006 reaching over
$421 billion before collapsing in the second half of 2007.
2

In our study, below, we examine the performance of ABS that serve as underlying
collateral or reference securities in 528 cash, hybrid, and synthetic CDOs posted online by
Pershing Square Capital Management and insured by Ambac or MBIA from 2005 to 2007. We
match these underlying securities into a database of more than 84,000 ABS from ABSNet

1
SEC Press release 2010-59, April 16, 2010. On July 15, Goldman Sachs agreed to pay a fine of $550 million and
reform its business practices to settle the SEC charges (SEC Press release 2010-123, July 15, 2010).
2
Aggregate issuance data based on SIFMA reports available at: http://www.sifma.org.
2
containing rating history, initial yield, collateral type, sponsor, and a variety of other important
variables.
Our data suggest that CDO assets were of relatively poor quality based on readily
observable attributes. CDO assets tended to be lower rated securities from the lowest quality
asset classes and vintages, and with a higher spread at issuance. Despite the fact that CDOs came
from low quality ABS, a higher percentage of CDO assets were rated by all three credit rating
agencies than the typical ABS security not in a CDO, which might have given buyers excessive
confidence in the quality of the assets.
Nonetheless, the empirical results confirm that assets included in cash and synthetic
CDOs performed extraordinarily poorly relative to seemingly comparable securities that were
not included in a CDO. When we control for the initial credit rating, CDO assets have a
downgrade severity that is at least twice as bad as comparable ABS not included in a CDO.
Synthetic CDOs assets perform worse than cash CDO asset, but assets included in both cash and
synthetic CDOs perform worst of all (with a downgrade severity about two and one-half times
worse than the average downgrade severity). Even when we include controls for a wide variety
of observable characteristics, including initial bond yield, CDO assets still are downgraded more
than comparable ABS by between 50 and 90 percent.
These results suggest that CDO originators successfully sold securities and insurance
against the worst performing ABS assets, but also that buyers of CDOs would have had a hard
time analyzing these securities based on observable characteristics alone. The poor performance
of cash CDOs is consistent with the ratings arbitrage hypothesis, as is the fact that securities in
more highly structured deals (CDO squared) perform even worse than those in straight CDOs.
3
Reference securities that show up in multiple synthetic CDOs also perform worse than ABS that
appears in only one CDO in our sample.
Our finding that synthetic CDOs performed even worse than cash CDOs seems to suggest
that issuers of CDOs had a strong motive to bet against the performance of the underlying
collateral inside a CDO. It would have been very hard to randomly choose securities with such
poor ex-post performance. This hypothesis is reinforced by the fact that synthetic CDOs
issuance hit its peak just at the time that the housing bubble started to burst, and after the peak
issuance of cash CDOs.
The data also allow us to examine the role of the sponsor in subsequent performance of
the CDOs. Many critics have argued that the proliferation of smaller, less reputable and poorly
regulated issuers was a factor in the crisis. Counter to findings in Faltin-Traeger, Johnson, and
Mayer (2010a) that high quality ABS sponsors issue better performing securities, our results
show that more highly regulated sponsors chose especially poor quality securities for their
CDOs, securities that performed even worse than the average security inside a CDO (a very low
standard). ABS included in CDOs issued by both foreign and domestic banks suffered especially
severe downgrades, a finding that persists even when we control for all observable bond
attributes. ABS in CDOs issued by major investment banks performed better-than-average based
on rating alone, although those results becomes insignificant when we control for bond yield and
other observable characteristics. Finally, we examine14 ABACUS deals issued by Goldman
Sachs. Conditional on rating and initial security yield, bonds included in the ABACUS deals
suffer less severe downgrade severity than ABS inside the average CDO deal (although the
underlying still perform considerably worse than equivalent ABS not included in a CDO).
4
However, once we control for the various observable characteristics, the underlying ABS in
ABACUS deals perform worse than average.
Finally, we show that ABS that appear in both cash and synthetic CDOs have a 50
percent worse downgrade severity than ABS in either cash or synthetic CDOs on their own.
These results are consistent with the hypothesis that equity buyers in cash CDOs might have
tried to offload the risk from the worst performing underlying ABS by shorting the cash
positions in subsequent synthetic CDOs. Consistent with this hypothesis, we find that reference
securities appearing in both cash and synthetic CDOs show up first in the cash CDO about 75
percent of the time. Particularly striking, though, is that assets in both cash and synthetic CDOs
received an average of 2.5 out of a possible three ratings, compared to an average of 2.1 ratings
for the typical ABS outside a CDO, underscoring the inability or unwillingness of rating
agencies to discipline adverse selection of the worst assets into CDOs.
The next section summarizes the relevant literature on ratings and structured finance.
Section 3 discusses the data, while Section 4 presents the basic empirical results. Section 5
presents a brief conclusion and policy discussion.
2. +$,-)%'. /)#,$0#'$, %" 123. 0"& 1$,&)# 40#)"5.
Many commentators argue that the ABS market suffers from information asymmetries
due to the opacity that securitization creates. While investors may not have fully appreciated the
asymmetries present in structured finance markets, their existence in other markets has been
thoroughly examined. A long and established literature explores the potential effects of adverse
selection on market allocation including Akerlof (1970) and Rothschild and Stiglitz (1976), the
5
implications for debt markets including Jaffee and Russell (1976) and Stiglitz and Weiss (1981),
and the implications for equity markets including Myers and Majluf (1984).
Empirical research confirms the importance of information asymmetries, but also shows
that such asymmetries do not necessarily lead to financial crises and market failures the way
some commentators have claimed about the CDO market. When facing adverse selection, buyers
can typically respond by paying a lower price for an asset. For example, Genesove (1993) shows
that buyers at auction pay less from used car dealers than for cars that come from new car
dealers, where some sellers will trade-in a used car every three years to buy a new car whether or
not anything is wrong with their existing car. In the ABS market, Downing, Jaffee, and Wallace
(2009) show that buyers of Freddie Mac SPVs recognized risks involved and demanded a
lemons premium of 13 to 45 percent of the overall prepayment spread. Only in rare cases
should adverse selection cause a complete market failure.
Of course, information asymmetries and adverse selection can still result in inefficient
transactions taking place. Several authors have examined the likelihood that some lenders
originated mortgages with greater risk due to their ability to sell the loans in the securitized
market. Keys, et. al. (2008), Mian and Sufi (2008), and Berndt and Gupta (2008) show that
originators made riskier loans when they were able to securitize these loans, although Bubb and
Kaufman (2009) disagree. Jiang, Nelson, and Vytlacil (2010) provide evidence of specific
agency conflicts associated with the mortgage origination process.
In the case of rated securities, the market has relied on third-parties (rating agencies) to
certify the quality of underlying collateral. However, this system appears to have failed at a
crucial time in the development of the market for rated securities. Theoretical models point to a
number of possible reasons that a third party ratings system might fail. Bolton, Freixas, and
6
Shapiro (2009) suggest that ratings become less informative at the peak of a market when there
are more nave investors in the market. As well, the authors argue that competition between
ratings agencies for business also leads to lower quality of ratings. Skreta and Veldkamp (2009)
point instead to increasing complexity, suggesting that as asset complexity increases, rating
agencies are more likely to offer a wider range of ratings, increasing the scope for ratings
shopping even if rating agencies issue purely unbiased ratings.
A growing empirical literature documents the extremely poor performance of structured
finance credit ratings in general and CDO ratings in particular. Benmelech and Dlugosz (2009)
find that CDO tranches rated by only one agency, especially S&P, were more likely to be
downgraded and, conditional on being downgraded, to suffer more severe downgrades. Coval,
Jurek, and Stafford (2009a) point out that ratings of CDOs were highly unreliable due to models
that were highly sensitive to even small errors in economic projections or losses and that
underestimated the correlation of risks across various debt securities. Aschcraft, Goldsmith-
Pinkham, and Vickery (2009) find that projected mortgage delinquency rates on subprime and
Alt-A ABS from a loan-level econometric model are strongly correlated with ex-post default,
suggesting ratings did not fully reflect information on mortgage risk available at deal origination.
As well, they show that, conditional on fundamentals, subordination levels (the buffer that
protects the highest rated securities from losses) declined by about 20% between mid-2005 and
mid-2007.
Other authors investigate the source of such ratings failures. Becker and Milbourne
(2008) show that competition between S&P and Moodys leads to lower ratings. Faltin-Traeger
(2010) presents empirical evidence that sponsors chose to obtain ratings from rating agencies
7
that tended to rate its deals more favorably or with lower levels of subordination, implying that
the issuer pays model may have given sponsors too much influence over the rating process.
Authors come to different conclusions about the extent that CDOs investors understood
the risks they carried. Coval, Jurek, and Stafford (2009b) argue that many structured finance
instruments could be characterized as economic catastrophe bonds, but that they offered far less
compensation than alternatives with comparable payoff profiles. The authors suggest that buyers
focused on expected payoffs as measured by ratings, while ignoring the state of the economy in
which defaults occur. However, Collin-Dufresne, Goldstein, and Yang (2010) price long-dated
S&P 500 options and tranche spreads on the five-year CDX index, calibrating the model using
the entire term structure of CDX spreads. The authors point out that their model matches the time
series of tranche spreads, offering a resolution to the seeming puzzle reported by Coval, Jurek,
and Stafford (2009b).
Most authors find that market prices did not sufficiently differentiate between the risks
posed by different securities. Adelino (2009) and Faltin-Traeger, Johnson and, Mayer (2010b)
demonstrate the spreads paid by securities buyers often help predict subsequent downgrades,
controlling for ratings, and that less complicated structures obtained slight pricing premiums.
However, the predictive power of spreads was far from perfect. Adelino shows that spreads do
not predict the likelihood of downgrade for AAA securities, which represented the bulk of MBS
securities issued in the crisis. Faltin-Traeger, Johnson, and Mayer show that securities issued by
the highest rated sponsors, if anything, required a spread premium, despite the fact that that these
high quality sponsors issued ABS that had the smallest likelihood of downgrade.
8
3. 123 6#$'(#'$, 0"& 40#)"5.
This paper attempts to disentangle a number of alternative hypotheses about the failure of
the market for rated securities. We examine a market, CDOs, that has received much attention as
a potential cause of the crisis. Given that CDOs are primarily composed of other rated securities,
there may be greater potential for information and ratings failures for CDOs than for other asset-
backed securities.
7898 10.: ;<%= -,$.'. 6>"#:,#)( 123.
Non-synthetic or cash flow CDOs appear to have played a key role in the growth of the
ABS market because they produced AAA-rated securities using the cash flows from a pool of
underlying ABS that were mostly rated below AAA. According to the Pershing Square dataset
used in this paper and described below, about 30% of the face value of the securities underlying
CDOs active in the second half of 2007 were rated AAA by S&P, but over 80% of the face value
of those CDOs obligations were rated AAA (see Figure 2). While this analysis does not take
into account the value of bond insurance and other forms of credit support embedded in the
CDOs, the rating transformation that occurred during the structuring process greatly expanded
the volume of capital available to ABS issuers. Some have referred to the process of pooling
low-rated ABS to collateralize the issuance of CDO liabilities that are predominantly rated AAA
as ratings arbitrage because the process allegedly takes advantage of arbitrage opportunities
in the credit rating agencies models. This transformation created additional AAA rated
securities that could be purchased directly by investors who had a preference for apparently low-
risk securities with high ratings.
A second function of CDOs was to allow the creation of custom securities that permitted
large and presumably sophisticated investors to express opposing views about the expected
9
performance of a particular pool of ABS. So-called synthetic CDOs did not involve the
purchase of an actual pool of underlying assets but instead were created to allow investors to
receive cash flows based on the performance of a pool of ABS specified by the deal documents
(reference securities) but not actually owned by the CDO trust. The CDO trust might purchase
credit default swaps (CDS) on the reference securities. The pool of CDS was designed to mimic
the cash flows that actual ownership of the securities referenced by the CDS would provide.
Since the CDO manager was not constrained by the need to purchase the underlying ABS,
synthetic CDOs could be arranged relatively quickly and sponsors could choose from a large
variety of reference assets.
3
In some cases, issuers created hybrid CDOs that contained a mix
of actual ABS and credit default swaps referencing ABS.
Motives for participation in synthetic CDOs varied widely. Some investors used CDOs to
hedge an existing investment position (purchase protection) or to receive regular investment
income. Others appear to have been speculating on the demise or success of the housing and/or
mortgage markets through the performance of reference securities. Whatever the motives,
issuance of synthetic CDOs grew rapidly at the tail end of the housing and credit boom.
Although pure synthetic CDOs represented less than one-third of overall CDO issuance, the
aggregate face value of synthetic CDOs rose substantially in 2007, while the issuance of cash
flow CDOs fell (see Figure 1), possibly suggesting the demand by some investors to speculate on

3
Synthetic CDOs also have more flexible structures than cash CDOs. Funded synthetic CDOs closely mimic cash
CDOs in that a buyer pays cash to purchase CDO tranches and receives regular payments based on the performance
of the underlying collateral. More common were unfunded synthetic CDOs in which an investor receives regular
spread payments (premiums) in return for cash payments in the event of a default.
10
the future performance of the housing market at a time when the housing bubble had started to
collapse.
4

78?8 6@%".%$ !"(,"#)-,.
The incentives for placing different types of securities in a CDO depend, in part, on
whether a deal is cash flow or synthetic.
5
In cash flow CDOs, sponsors have relatively clear
incentives to choose poorly performing underlying. Rating agencies were transparent in stating
that they evaluated the expected performance of a previously rated security based only on the
rating of the security. Therefore consider an issuer who must decide which of two AA-rated
subprime mortgage-backed securities to include in a CDO. One security sells at par, while the
other (riskier) security sells at 95 percent of par. Since the rating agency treats both securities as
equivalent in terms of the impact on the CDO rating, the issuer is likely to purchase the cheaper
(riskier) security. Of course, if investors fully understood the process and the rating had no
independent impact on valuation, such ratings arbitrage would not create financial value.
However, regulators provide preferential capital treatment for highly rated securities so CDOs
provided an opportunity for purchasers to acquire these riskier (and higher yielding) securities
while still apparently meeting strict risk-based capital requirements.

4
Data provided by S&Ps RatingsXpress database of structured finance ratings. Some now high-profile investors
such as John Paulson appear to have used CDOs to place large bets on the demise of subprime securities in 2007.
5
In some CDOs, the sponsor of the CDO might be a different party than the collateral manager, who was
responsible for choosing securities to put into a CDO. Nonetheless, the sponsor of a CDO can influence the quality
of the underlying ABS pool through its choice of collateral manager and through communication to that collateral
manager. As well, purchasers of different CDO tranches also had to approve the types of securities placed in the
CDO, so these purchasers might have also been in favor of taking advantage of as much ratings arbitrage as
possible.
11
Since CDOs involved a second layer of structuring beyond the original ABS, the
potential for ratings arbitrage was greater than what would have otherwise been possible by
pooling riskier-than-average mortgages into subprime MBS. In addition, CDO-squareds
involved using CDO securities as collateral for the creation of a higher-level CDO and provided
still another layer of securitization, multiplying once again the potential degree of ratings
arbitrage.
In the case of a synthetic CDO, however, it is not clear why the sponsor would seek out
risky securities in the same way since no ABS are actually purchased. Because a synthetic CDO
is composed of a pool of CDS, the positions of the buyer and seller are completely symmetric.
Thus, sponsors cannot buy low and sell high, as in the case of ratings arbitrage with cash CDOs.
If relatively risky reference securities are chosen, one side is implicitly choosing to be long the
risky securities while another side is short those same risky securities. While ratings arbitrage
still makes sense based on observables (i.e. it still makes sense to put A and AA rated securities
together to get a lot of AAA rated securities), it is unclear why either side might prefer to choose
higher or lower yielding securities conditional on rating. After all, if one side is long the AAA
rated tranche to take advantage of lower capital requirements, the other side is short that same
AAA rated tranche.
While ex-ante incentives do not appear to favor a choice of either unduly risky or safe
securities (conditional on rating) to include in a CDO, parties on one side or the other of the
CDO might want to manipulate reference securities their favor. The long side of a CDO (the
party who receives the regular yield) would prefer relatively safe reference ABS, while the short
side would prefer the inclusion of risky securities with large expected losses.
12
In the SECs complaint against Goldman Sachs, the government alleges that the party on
the short side of the transaction, Paulson and Co., was able to skew the reference asset pool
towards assets that were disproportionately likely to take losses and that Paulsons position in the
transaction was not adequately disclosed to investors on the long side of the transaction.
6

Paulsons participation in a large number of CDO transactions has led some to speculate that
such behavior was the norm rather than the exception.
Similarly, a recent book on the financial crisis, Smith (2010), suggests that [t]he really
smart guys were the ones who used the bottom tranches to fund a short subprime bet, and
that another hedge fund, Magnetar, went into the business of creating subprime CDOs on an
unheard-of scale. According to the books sources, Magnetar was involved in the sale of
approximately $30 billion in CDO securities, and it ended up driving between 35% and 60% of
subprime issuance in 2006.
7
Allegedly, most of the CDOs associated with Magnetar have turned
out to be nearly worthless.
8

7878 123 1%A@<,B)#> 0"& 3@0()#>
Even if CDO sponsors had incentives to choose low-quality collateral assets, it is not
immediately clear why asymmetric information should be an issue since the composition of the
ABS pool was available to all investors. The securities in the pool were clearly listed in deal

6
SEC Press release 2010-59, April 16, 2010.
7
See pp. 259-262 and Appendix II of Smith (2010) for a more detailed discussion of the Magnetar trade.
8
See The Magnetar Trade: How One Hedge Fund Helped Keep the Bubble Going, by Jesse Eisinger and Jake
Bernstein, ProPublica - April 9, 2010. The full article reports that An independent analysis commissioned by
ProPublica shows that these deals defaulted faster and at a higher rate compared to other similar CDOs. According
to the analysis, 96 percent of the Magnetar deals were in default by the end of 2008, compared with 68 percent for
comparable CDOs. Magnetar disputes these findings in a letter posted in Propublica.
13
documents and an interested investor could have performed an independent evaluation of the
underlying ABS.
Instead the issue may be better characterized as costly information acquisition. As a
consequence of the size and complexity of the collateral pool, buyers would have faced a
difficult task in pricing each of the many underlying assets. Most CDO deals in the dataset used
in this paper had at least one hundred underlying securities, each of which was tied to thousands
of mortgages or other types of debt (see Figure 3). Because of this complexity, CDO ratings
depend critically on a variety of modeling assumptions about the overall performance and the
correlation in performance of underlying assets and the extremely complicated division of cash
flows between more than a dozen securities that make up a typical cash CDO. Synthetic CDOs
often had fewer tranches, but equally complex rules.
Evaluating CDO performance is complicated by several additional factors. Liabilities
from one CDO can be repackaged within a second CDO (referred to as a CDO-squared). The
performance of one CDO tranche may therefore depend upon not only the cash flows from the
ABS in that CDO but also the ABS underlying any CDO tranches that the CDO of interest owns
as collateral. Each CDO in the dataset used in this paper invests in an average of 119 securities
and about 5% of those are themselves CDO liabilities. Going only one level down in a set of
120 securities, an investor must evaluate the 114 ABS directly underlying the CDO and 6
additional sets of 114 securities underlying the 6 CDO tranches in the pool for a total of nearly
800 ABS.
Evaluation is further complicated by the fact that synthetic CDOs did not involve the
purchase of actual ABS collateral but instead amounted to agreements between two parties to
exchange cash flows. However, the fact that synthetics did not involve the creation of new loans
14
meant that the same aggregate pool of ABS was supporting a larger volume of CDO liabilities,
potentially increasing the correlations among CDO tranches. In 2007, Bloomberg reported that
Moody's also said it's concerned that the growth of synthetics, or credit swaps, may leave
more CDOs invested in other CDOs exposed to the same bonds as they are. The company said its
models were developed using the data that was available at the time, such as transactions
backed by cash collateral. Moody's is now working on a research project to reassess the
correlation between CDOs at time when exposures can be infinitely replicated, it said.
Third, many CDO deals were backed by actively managed pools of assets, creating
significant flexibility with respect to the choice of securities in the collateral pool but also
making them much more difficult to evaluate. According to a report by Fitch Ratings (2006)
describing the popularity of one type of actively managed CDO, Market Value CDOs are
enjoying a revival with issuance more than doubling in 2006 from the year before. MV CDOs
appeal to managers because they offer greater trading flexibility and can invest in a wide range
of assets, including high yield bonds and bank loans and [structured finance] securities. These
CDOs effectively enabled institutional investors to invest in actively managed funds that
invested in a variety of financial products and therefore provided a higher return than other
AAA-rated securities.
The difficulties involved in accurately estimating the value of CDO liabilities led former
Fed Chairman Greenspan to warn investors that the credit risk profile of CDO tranches poses
challenges to even the most sophisticated market participants [and investors should not] rely
solely on rating-agency assessments of credit risk. However, many investors nonetheless did not
have access to the considerable resources needed to perform their own due diligence and
continued to rely on credit ratings. According to Mason and Rosner (2007):
15

The ability to repackage financial securities and call them something else, with
no fundamental change to their risk characteristics, in order to achieve an
improved bond rating is the fundamental source of ratings arbitrage. As long as
ratings agencies mean different things when referring to CDOs, ABS, and
Corporate debt, incentives will continue to be skewed by risk arbitrage.
Furthermore, embedding ratings into regulation through ERISA and Basel II only
worsens the incentives to use opacity to the issuers benefit (and the investors
loss).
78C8 D>@%#:,.,.
Ratings on the securities issued by a CDO are primarily based upon the ratings of the
underlying assets and assumptions about the correlation of the performance of those assets. As
noted above, this process gives perverse incentives for CDO sponsors. From a pure ratings
perspective, a sponsor might be indifferent between a relatively cheap AAA-rated security (that
investors judge to have barely made the AAA standards) and a much more expensive, high
quality AAA security (that investors believe faces virtually no chance of default). A sponsor
interested in creating the highest quality CDO at the lowest cost would choose the most highly-
rated underlying ABS that the market views as being riskiest and thus the least expensive,
conditional on rating.
We examine several hypotheses about the quality of ABS were placed into CDOs:
1) Random Selection: Given that investors often knew the precise securities that were
placed in CDOs, they should have seen through any attempt by sponsors to select low
16
quality ABS as collateral. In this case, CDOs would contain a random selection of
ABS as collateral.
2) Ratings Arbitrage: Investors might be willing to accept (or even prefer) lower quality
ABS in CDOs in an attempt to arbitrage capital requirements. Such a hypothesis
might explain why cash CDOs contain the lowest quality collateral, as the sponsors of
these CDOs would have the highest willingness to pay for the worst securities, as
they could place the low quality securities into CDOs, taking advantage of ratings
arbitrage. Of course, the incentives for synthetic CDOs would not be the same as cash
CDOs, since buyers and sellers held completely symmetric positions. Any gain on
one side of the transaction must be made up by a loss on the other side.
3) Investors were misled: Some less sophisticated investors might have been misled
when they purchased highly complex securities. Instead, nave investors might have
relied on purchasing securities rated by multiple rating agencies and sponsored by
some of the most reputable financial institutions in the US and abroad. If this trust
were successful, we should expect securities rated by multiple rating agencies or
sponsored by regulated or high quality issuers to perform better conditional on
observable attributes.
We will provide evidence for these incentives using a database of CDO holdings to show
that securities purchased in CDO deals have higher yields and more ratings, but also poorer ex-
post rating performance in terms of downgrade severity.
17
4. 20#0
Under normal circumstances, the collateral underlying CDOs is not available to academic
researchers as most CDOs were issued as private securities under SEC rules that required only
limited disclosure (so-called 144A offerings). In return, sponsors agreed to limit the types of
investors who could purchase CDOs to the most sophisticated investors. In contravention of
these rules, on January 30
th
, 2008 Pershing Square Capital Management distributed a database
covering all of the assets underlying CDOs and CDO-Squareds insured by Ambac or MBIA from
2005-2007 in what Pershing Square claimed was an attempt to bring attention to the large
liabilities of bond insurers Ambac and MBIA.
9
We use this database to examine the performance
of assets inside these CDOs.
The information reported by Pershing Square was collected from trustee reports dated
between 2006Q3 and 2008M1, although virtually all of the deals in the dataset (99%) are
described by a trustee report dated in 2007 or January 2008. We therefore restrict our definition
of CDO assets to securities observed in a CDO trustee report dated in this period. The resulting
dataset covers 528 deals and approximately 30,000 unique underlying securities identified by
CUSIP.
Documentation accompanying the Pershing Square dataset indicates that it covers all
CDO deals closed between 2005 and 2007, but possibly omits some synthetic deals. In order to
examine how many deals may be missing, we merge the CDO liabilities reported by Pershing
Square with S&Ps RatingsXPress dataset, which covers all of S&Ps public structured finance
ratings. For comparability, we drop deals that would be unlikely to contain only ABS, including
CLOs, TruPS, and CDOs backed by corporate debt. Of the remaining CDO deals in the S&P

9
Pershing Square was reputed to hold an appreciable short position in the stock of Ambac and MBIA at that time.
18
dataset, 76% are also in the Pershing Square dataset. Using downgrade severity regressions
similar those described below for underlying ABS, we find no statistically significant difference
in the downgrade performance of CDO tranches for S&P rated deals that were in the Pershing
Square database and those S&P rated CDO tranches that did not merge.
To provide additional information about the assets within the CDOs, we use data on
ratings history, coupon, and sponsors of ABS from Lewtan Technologies ABSNet securitization
database. The database provides the complete history of rating actions by Standard and Poors
(S&P), Moodys Investors Service (Moodys), and Fitch Ratings (Fitch) and identifies the issuer
for 150,000 securities issued between 1995 and 2008. We restrict the sample to those ABS
securities issued after 2001 where we observe the underlying collateral type, coupon history,
sponsor and servicer, resulting in a sample of 86,294 securities.
To measure performance, we calculate the number of fine rating notches each security
was downgraded. For a security rated AAA, a downgrade of one fine notch would result in a
rating of AA+; two fine notches would be a AA; three fine notches would be AA-. Each ratings
category has three fine notches between AA and CCC. We treat as censored any ratings reported
to be at or below CCC-, as downgrades below this level are not consistently reported. This
downgrade severity measure takes a value of zero if a securitys rating remained unchanged or
improved.
10
We use S&P ratings because S&P rated the largest fraction of securities in the
sample. Results are comparable regardless of which rating agencys ratings are used.
For our baseline analysis, we focus on downgrades during the two-year period at the
heart of the crisis between July 1, 2007 and July 1, 2009, although we consider other time
periods to control for possible measurement error in when we observe ABS underlying each

10
Very few securities were upgraded during our sample period beginning in Jluy, 2007.
19
CDO. On specific concern is that turnover in the collateral pool might undermine the clarity of
our results because we only observe a snapshot of the collateral pool at the time of the trustee
report. In some cases, the CDO manager may choose to subsequently change the composition of
the pool. However, for a subset of the CDO deals in our sample, we have access to servicer
reports that show monthly changes in each collateral pool. In these pools, turnover appears to be
low a few percent per month. We also address this issue in other specification tests by limiting
our analysis to data taken from trustee reports dated July, 2007 through January, 2008 and
measuring performance beginning in February, 2008. As well, we consider various endpoints for
our analysis, including July, 2008, July, 2009, and June, 2010.
For each security, we proxy for the date of issuance using the date each security is first
rated. Table 1 shows the mean downgrade severity by vintage and the distribution of securities in
the ABSNet database by whether they are observed in a CDO. Securities observed in CDOs tend
to be of more recent vintages, which also have higher downgrade severities. For example, the
2006:h2 vintage represents about 9 percent of all outstanding securities issued since 2001, but
between 16 and 22 percent of all ABS used in CDOs. That vintage had an average downgrade
severity of more than 9 fine notches; so a security rated AAA at issuance would on average be
rated BBB- by the middle of 2009. Despite being a relatively late vintage and thus not available
to be placed into CDOs for a very long period of time, the 2006h2 vintage was especially likely
to placed in synthetic CDOs (22 percent of all ABS in synthetic CDOs) and even more likely to
be used as a reference asset more than once in synthetic CDOs (24 percent) as shown in the
column on the bottom right hand side of the table.
Table 2 lists the same relative distribution of securities in and out of CDOs, but by
collateral type instead of vintage. There are 22 types of collateral backing securities in our
20
database, of which the smallest 13 types have been consolidated into a category labeled Other.
Once again, the worst performing collateral typehome equityis much more highly
represented within securities included in CDOs than those never consolidated into CDOs.
Next we consider the rating agency that rated the CDOs compared to those that rated the
underlying securities in Table 3. Not surprisingly, the ABS included in CDOs were much more
likely to have been rated by the particular agency that also rated the CDO. This is because most
agencies treated their own rated ABS in an advantageous fashion when rating a CDO that
included collateral rated by competing agencies. Even more striking is the extent to which the
CDOs themselves were likely to be rated by at least two agencies, and often by all three
agencies. S&P and Moodys were involved in rating the bulk of the CDOs in the database.
Clearly buyers of CDOs might have been nervous about ratings shopping and required rating by
multiple issuers, a requirement that in the ABS market has been shown to have provided more
reliable ratings performance. However, obtaining multiple ratings did not provide stronger
protection for buyers of CDOs. Table 4 shows that the CDOs rated by all three rating agencies
were, if anything, more likely to be downgraded relative to CDOs that were rated by only two
agencies. As well, the previously documented problems with CDOs rated by S&P are apparent in
this table.
Table 5 reports summary statistics for a number of other important variables for our
analysis. One potentially valuable control variable is the yield on the ABS security. Many
analysts claimed that issuers of CDOs searched for high yielding securities conditional on rating,
with the high initial yield serving as a proxy for ABS securities that were especially low quality.
Alternatively, higher yielding securities were relatively cheap compared to the par value of their
collateral and thus especially attractive for inclusion in a CDO if rating agencies did not
21
differentiate between ABS of a given rating. Such a yield might have also served as a signal to
potential purchasers of CDOs or rating agencies about the quality of underlying collateral. While
we would like to observe the current yield (or price) of the security at the time it was included in
the CDO, such data is typically not available as most of the ABS were traded quite infrequently
and the prices of the trades were not publically recorded. Instead, we use data on the coupon at
the time of issuance and convert that coupon yield into a spread at issuance assuming the security
was issued at par. To validate the assumption that most securities trade at par upon issuance, we
examine price data from Bloomberg, which is available for a portion of our sample. The median
issuance price in the 37% of the sample that merged was par and 95 percent of the sample had an
issuance price greater than 99.8% of the par value. We therefore use the coupon spread to proxy
for the yield spread at issuance.
11

Table 5 shows the mean coupon spread in percentage points and the mean floating-rate
fraction for securities observed and not observed in a CDO. It also shows the mean number of
ratings and mean rating for each subsample of securities. CDO assets tend to have more ratings
and lower ratings. The numerical rating mean is created by assigning a numerical value to each
fine rating notch beginning by assigning 1 to AAA, 2 to AA+, 3 to AA, etc. Therefore a higher
mean rating numerical value corresponds to a worse mean rating. CDO assets also tend to have
slightly longer expected maturities. This may reflect the skew in distribution towards home
equity MBS relative to credit card and auto loan ABS, which tended to have shorter maturities.
Finally, for AAA securities, we calculate subordination below the AAA tranche as in Ashcraft,

11
For each fixed-rate security, ABSNet provides an expected maturity. We create the spread by subtracting the yield
on the coincident Treasury with the closest maturity to the securitys expected maturity. For floating-rate securities,
the benchmark interest rate is provided (most often 1-month LIBOR). This rate is subtracted from the initial coupon
to create the spread for floaters.
22
Goldsmith-Pinkham, and Vickery (2009), defined as 1 [face value of AAA
securities]/[aggregate face value of underlying loans at origination]. This subordination value is
calculated using the underlying loan balance at origination so it may not be identical to
subordination at the beginning of 2007H2 depending upon loan balance changes that have
occurred after origination. However, this variable should still be informative about the
subordination level at 2007H2, even if it may be somewhat stale.
Next we examine the distribution of ABS ratings for securities within CDOs in Table 6
based on their initial rating on July 1, 2007. Securities observed in a CDO tended to have ratings
below AAA, which is consistent with the idea that issuers were using CDOs to improve the
ratings of otherwise lower quality collateral. Once again, synthetic CDOs appear to be using
relatively lower quality collateral, measured by initial rating, relative to cash CDOs.
Table 7 serves to preview our regression results by examining the subsequent downgrade
severity of ABS inside and outside CDOs based on the initial rating of the ABS. Securities
observed in CDOs have strikingly higher downgrade severities regardless of the initial rating.
We consider three time periods: securities downgrades about one year after our initial
observation (July 1, 2008 versus July 1, 2007), two years afterwards (July 1, 2009), and three
years afterwards (June 1, 2010). For the first two years of the sample, ABSs suffered more than
twice the downgrade severity when they were included in a CDO. And ABS included in
synthetic CDOs or in both non-synthetic (cash) and synthetic CDOs appeared to perform even
worse than ABS included only in cash CDOs. The difference in downgrade severity between the
columns appears to moderate after three years, but that might be due to censoring as CDOs can
only hit a lower bound of CCC- in our database. We control for such censoring, as well as
differences in underlying collateral and other observable attributes in the regressions that follow.
23
Finally, Table 8 considers CDO squared securities, by examining the performance of
CDOs that were included as underlying collateral in other CDOs. Greater levels of complexity
appear to offer the opportunity for even more cherry picking of securities by expected
downgrade severity. The worst downgrades are suffered by AAA and AA CDOs included in
other CDOs, with a mean downgrade severity of almost 10 fine notches in less than two year; the
equivalent of going from the highest possible rating to junk status in a relatively short period of
time.
5. 6,('$)#> +,$;%$A0"(,
E898 6,('$)#> 1:0$0(#,$).#)(. 0"& 2%="5$0&, 6,-,$)#>
To begin, we take all securities issued between 2002 and 2007H2 and compare the
downgrade severities of those observed in a CDO with the rest of the sample of ABS not
included in a CDO. To begin, we run a Tobit regression of downgrade severity on fixed effects
indicating whether each security was observed in at least one non-synthetic CDO, at least one
synthetic CDO, or a combination of both synthetic and non-synthetic deals.
The results, shown in Table 9, are striking. When we include only includes dummy
variables for the initial rating (as of July 1, 2007), column (1) shows that securities observed in a
CDO are downgraded at least 3 more fine-rating notches more than those not included in a CDO,
an effect that is twice as large as the mean downgrade rate of 3.0 fine ratings over the same two
year time period for ABS not used as collateral. Securities inside synthetic CDOs performed
even worse, with a downgrade severity of about 4.1 fine notches. Securities inside both synthetic
and cash CDOs performed worst of all, with a downgrade severity of 4.6 fine notches, about two
24
and half times worse than average. These effects are little changed in column (2) when we
include fixed effects for the rating agency.
The inclusion of additional security characteristics reduces the estimated coefficients by
about half, although the effects remain economically and statistically significant. In column (3)
we include a fixed-effect identifying if a securitys coupon is floating-rate, the mean spread
calculated by vintage half-year separately for fixed and floating securities, and the coupon spread
difference from the mean. A higher coupon spread difference from the mean is predictive of
worse performance, indicating that the market was capturing some of the performance risk in
these securities. As well, floating-rate securities tended to perform worse than fixed-rate
securities, consistent with other research evaluating ABS performance between 2000 and 2008
(Faltin-Trager, Johnson, and Mayer, 2010a).
12

In column (4) we include a number of additional controls, including fixed effects for
each securitys collateral type and half-year of initial rating. We also include all variables from
the base specification of Faltin-Traeger, Johnson, and Mayer (2010a) to further control for any
remaining potentially relevant information. These include fixed effects identifying the S&P
issuer credit rating of the sponsor at the time of issuance of the security; a fixed effect identifying
whether the sponsor issued ABS in more than four collateral type categories in the three years
prior to issuance of the security; and fixed effects indicating whether the sponsor also services
the securitys collateral or the servicer is unidentified.
The coefficients on these controls are shown in Table 18. The results are consistent with
those from Faltin-Traeger, Johnson, and Mayer (2010a). Securities issued by sponsors with an

12
See Faltin-Trager, Johnson, and Mayer (2010a) and Standard & Poors Press Release. Lower Property
Valuations Drove 2009 Floating-Rate CMBS Downgrades. February 18, 2010.
25
investment-grade credit rating and more lines of structured finance business perform better than
average.
13
As well, deals in which the sponsor also performs servicing have lower downgrade
severities. Regressions in columns (4) and (5) also include fixed effects indicating if each
security was upgraded or downgraded before 2007H2. Consistent with other work on the serial
correlation of rating transitions, we find that securities downgraded before 2007H2 also have
significantly higher subsequent downgrade severities. As well, securities rated AAA with higher
levels of subordination perform better ex post.
Finally, in column (5) the floating-rate fixed effect and difference from mean spread are
included. The mean spread is not included because vintage half-year fixed effects are included in
this specification. The coefficient on the difference from the mean spread is still positive,
however it is significantly smaller, indicating that the other controls are capturing some of the
information imbedded in spreads. And as before, the coefficients on the CDO fixed effects
decrease but remain large and significant.
E8?8 6'FG.0A@<, 6@,();)(0#)%".
We now examine the regressions presented in the previous section for several subsamples
to better understand the results. In Table 11, we re-run the specification that includes a complete
set of controls (column 5 of Table 10) separately for securities sorted by initial rating to pick up
any non-linearities that might exist in the downgrade process. For all ABS with initial ratings of
investment grade (BBB) or higher, the coefficients on the CDO fixed effects are positive and in
most cases they are significant. The results provide further support for our conclusions that cas

13
The omitted sponsor rating category corresponds to any rating below BBB.
26
CDOs appear to perform better than synthetic CDOs. However, for securities rated BB and
below (representing 7 percent or less of the sample), the CDO fixed effects are small and not
statistically different from zero. Table 12 provides a similar set of regressions corresponding to
the specification from column (1) of Table 10, which is the specification that only includes
controls for the initial rating of the ABS security. Once again, the coefficients on the CDO fixed
effects are positive and significant in each case, including for securities rated BB and below.
In order to examine the extent to which stale spread and subordination information is
affecting the results, we report the results of running the same specifications as in Table 10, but
restricting the sample to securities originated in 2007H1 only. Results are shown in Table 13.
The coefficients on the inclusion in a CDO are much smaller, although still highly statistically
significant. These results suggest that including more recently rated securities might moderate
the estimated impact, possibly suggesting that ratings arbitrage is best accomplished for
securities that are in the market longer. Sponsors might have an easier time predicting the
downgrade performance of older securities, where the market might have more timely
information than the rating agencies.
Next we restrict the sample to the performance of CDO in Table 8. These regressions
compare the performance of CDO liabilities that are observed as underlying in CDO-squareds
with the remaining CDO liabilities in the ABSNet sample. The results indicate that CDOs that
are re-securitized tend to perform even worse than other repackaged ABS, with a downgrade
severity of 3.3 fine notches for cash CDOs and an incredible 7.8 fine notches for CDOs in both
cash and synthetic CDOs. Once again we also find that there is strong serial correlation in rating
transitions for CDO liabilities and that higher subordination levels help protect investors.
27
E878 H'AF,$ 0"& I>@, %; 123.
In order to examine whether securities observed in more CDOs perform worse, we run
the same specification as in Table 10, but include fixed effects indicating whether a security was
observed in 2 CDOs and more than 2 CDOs along with the usual set of control variables. Results
shown in Table 15 indicate that securities observed in two or more CDOs perform significantly
worse than securities observed in only one synthetic or non-synthetic deal. These results are also
consistent with the hypothesis that CDOs were chosen, in part, for their relative likelihood of
performing poorly (and/or being relatively cheap).
One hypothesis for the especially poor performance of securities in both cash and
synthetic CDOs is that buyers of CDOs with ABS might have used synthetic CDOs to hedge the
performance of the worst ABS. Consistent with that hypothesis, securities are issued first in the
cash CDOs about 75 percent of the time prior to their inclusion in a synthetic CDO. The data do
not suggest appreciable support for this hypothesis. For securities in both a synthetic and non-
synthetic CDO, we examine which type of CDO had the earliest closing date. Again, we run the
same specification as in Table 10 but include a fixed effect that equals one for those securities in
both a synthetic and non-synthetic CDO where the non-synthetic deal closed first. Results in
Table 16 illustrate that in the simpler specifications, these securities perform relatively better,
however when all the available controls are included, their performance is similar to other CDOs
included in both cash and synthetic CDOs.
E8C8 +,$;%$A0"(, %; J0$)%'. 6@%".%$.
Finally, we examine the role of various sponsors in the performance of CDOs. We divide
CDOs into four categories: those sponsored by a domestic bank, those sponsored by a foreign
28
bank, those sponsored by an investment bank (as of July, 2007), and all other sponsors. The list
of sponsors is given in Appendix Table 8. The results suggest an important role of sponsor in
CDO performance, but that the CDOs sponsored by the most highly regulated sponsors (both
domestic and foreign banks) performed the worst. Ironically, investment banks that have
generated much attention for having contributed to the crisis appear to have sponsored CDOs
that performed, if anything, better than the average CDO. Appendix Table 2 shows the
performance of the Abacus deals that the SEC referenced in its complaint against Goldman
Sachs. The Abacus deals seemed to suffer even slightly fewer downgrades relative to deals
sponsored by other investment banks.
6. 1%"(<'.)%"
Collateralized Debt Obligations (CDO) played a key role in the growth of Asset-Backed
Security (ABS) issuance between 2004 and 2007 by providing a mechanism for lower-rated ABS
to be used as collateral for the creation of AAA securities. Using a database published by
Pershing Square Capital Management covering all of the assets underlying 528 CDOs and CDO-
Squareds insured by Ambac or MBIA from 2005 to 2007 and using rating history and other
information from the ABSNet database, we compare the characteristics and performance of ABS
observed in a CDO with other ABS not observed in a CDO. We find that CDO assets tend to be
lower rated securities from the lowest quality asset classes and vintages, and with a higher spread
at issuance, although also more likely to be rated by all three rating agencies.
CDO assets performed much worse than comparable securities that were not included in
a CDO. When we control for the initial rating, CDO assets have a downgrade severity that is at
least twice as bad as comparable ABS not included in a CDO. Synthetic CDOs assets perform
29
worse than cash CDO asset, but assets included in both cash and synthetic CDOs perform worst
of all (with a downgrade severity about two and one-half times worse than the average
downgrade severity). Even when we include controls for a wide variety of observable
characteristics, including initial bond yield, CDO assets still underperform comparable ABS by
between 50 and 100 percent. These results suggest that CDO originators successfully sold
securities and insurance against the worst performing ABS assets, but also that buyers of CDOs
would have had a hard time analyzing these securities based on observable characteristics alone.
These results suggest an appreciable failure of the regulatory process that relies on
heavily on credit ratings to discipline the investment behavior of regulated entities through
giving preferential capital treatment for low-rated securities. Issuers were able to game the
system and obtain high ratings on low quality assets that performed extremely poorly in the
crisis. This fact is underscored by the failures of AIG, several investment banks, and many
regulated banks, especially in Europe, and who clearly sought out high yielding securities with
the seeming safety of top ratings.
Our findings also suggest that reforming the ratings process will not be easy. Bolton,
Freixas, and Shapiro (2009) have suggest requiring issues be rated by more than one rating
agency. While this rule would have eliminated problems with single-rated ABS that performed
poorly relative to ABS with multiple ratings, the assets underlying the CDOs in our sample were
more likely to receive two or three ratings relative to ABS outside CDOs, yet still performed
considerably worse. A second possible reform that some have raised is to require rated ABS to
have a separate ratings category (presumably with higher required capital) relative to corporate
rated securities. Nonetheless, this might not be sufficient to have prevented the problems in
CDOs. CDO quality was much worse than comparably rated ABS. Our results suggest that the
30
more highly structured the product (e.g., CDOs versus ABS or CDO squared versus plain
CDOs), the less informative ratings were in predicting performance. The experience of CDOs
suggests we have a long way to go to finding a process of using third-party credit rating agencies
to discipline regulated entities from taking on excess risk.

31
7. 4,;,$,"(,.
Adelino, Manuel. 2009. Do Investors Rely Only on Ratings? The Case of Mortgage-Backed
Securities. Working paper.
Akerlof, George. 1970. The Market for Lemons: Quality Uncertainty and the Market
Mechanism. The Quarterly Journal of Economics, 84(3), 488500.
Ashcraft, Adam, Paul Goldsmith-Pinkham, and James Vickery. 2009. "MBS Ratings and the
Mortgage Credit Boom. Working paper.
Ashcraft, Adam B. and Til Schuermann. 2008. "Understanding the Securitization of Subprime
Mortgage Credit," Foundations and Trends in Finance 2, No. 3 (July): 191-309.
Becker, Bo and Todd Milbourn. 2008. Reputation and Competition: Evidence from the Credit
Rating Industry," HBS Finance working paper 09-051.
Benmelech, Efraim and Jennifer Dlugosz. 2009. The Alchemy of CDO Credit Ratings. Journal
of Monetary Economics. Vol. 56, No. 5, pp. 617634.
Bergstrom, Rupini. 2007. Bond Raters Get Subpoenas. The Wall Street Journal. October 26,
2007.
Bolton, Patrick, Xavier Freixas, and Joel Shapiro. 2009. The Credit Ratings Game.
Unpublished manuscript from Columbia Business School.
http://ssrn.com/abstract=1428288.
Bubb, R. and A. Kaufman. 2009. Securitization and Moral Hazard: Evidence from a Lender
Cutoff Rule. Unpublished manuscript from Harvard University.
http://ssrn.com/abstract=1477891.
Collin-Dufresne, Pierre, Robert S. Goldstein, and Fan Yang. 2010. On the Relative Pricing of
Long Maturity S&P 500 Index Options and CDX Tranches. Working paper.
Coval, Joshua D., Jakub W. Jurek, and Erik Stafford. 2009a. "The Economics of Structured
Finance." Journal of Economic Perspectives, 23(1), 3-25.

32
Coval, Joshua D., Jakub W. Jurek, and Erik Stafford. 2009b. "Economic Catastrophe Bonds.
The American Economic Review, 99(3), 28-66.
Downing, Chris, Dwight Jaffee, and Nancy Wallace. 2009. "Is the Market for Mortgage Backed
Securities a Market for Lemons?" Review of Financial Studies, 22(7), 2457-2494.
Faltin-Traeger, Oliver. 2010. Picking the Right Rating Agency: Sponsor Choice in the ABS
Market, Columbia Business School mimeo.
Faltin-Traeger, Oliver, Kathleen Johnson, and Christopher J. Mayer. 2010a. "Rating Asset-
Backed Securities." Unpublished manuscript from Columbia Business School, February.
Faltin-Traeger, Oliver, Kathleen Johnson, and Christopher J. Mayer. 2010b. "Issuer Credit
Quality and the Price of Asset-Backed Securities." The American Economic Review
Papers and Proceedings, 100(2), forthcoming.
Farhi, Emmanuel and Jean Tirole. 2009. Collective Moral Hazard, Maturity Mismatch and
Systemic Bailouts, Harvard University working paper, June 29.
Fitch Ratings. 2002. Survey Shows Majority of Structured Finance Executives Oppose
Notching as Practiced by Moodys and S&P. Press release, March 27, 2002.
Fitch Ratings. 2006. Credit Policy Special Report 2007 Global Structured Finance Outlook:
Economic and Sector-by-Sector Analysis December 11 2006.
Genesove, David, 1993. "Adverse Selection in the Wholesale Used Car Market. Journal of
Political Economy, 101(4), August, 644-665.
Jaffee, Dwight M. and Thomas Russell. "Imperfect Information, Uncertainty, and Credit
Rationing," Quarterly Journal of Economics. Nov. 1976, 90(4), pp. 651-66.
Jiang, Wei, Ashlyn Nelson, and Edward Vytlacil. 2009. Liars Loan? Effects of Loan
Origination Channel and Loan Sale on Delinquency. Unpublished manuscript from
Columbia Business School. http://ssrn.com/abstract=1421462

33
Keys, B., T. Mukherjee, A. Seru, V. Vig. 2008. Did Securitization Lead to Lax Screening?
Evidence From Subprime Loans, 2001-2006. Unpublished manuscript from SSRN.
http://ssrn.com/abstract=1093137
Krozner, Randall S. 2007. Remarks to the 2007 Credit Markets Symposium at the Charlotte
Branch of the Federal Reserve: Recent Innovations in Credit Markets (Mar. 22, 2007).
Available at
http://www.federalreserve.gov/BoardDocs/Speeches/2007/20070322/default.htm
Mason, Joseph R. and Josh Rosner, Where Did the Risk Go? How Misapplied Bond Ratings
Cause Mortgage Backed Securities and Collateralized Debt Obligation Market
Disruptions," 2007. SSRN Working Paper #1027475.
Mian, Atif, and Amir Sufi. 2008. The Consequences of Mortgage Credit Expansion: Evidence
from the 2007 Mortgage Default Crisis. NBER Working Paper No. W13936, April.
Myers, Stewart C. and Majluf, Nicholas. "Corporate Financing and Investment Decisions When
Firms Have Information That Investors Do Not Have," Journal of Financial Economics.
June 1984, 13(2), pp. 187-221.
NERA Economic Consulting. 2003. Credit Ratings for Structured Products: A Review of
Analytical Methodologies, Credit Assessment Accuracy, and Issuer Selectivity among
the Credit Rating Agencies.
Nomura Securities International. 2006. Rating Shopping Now the Consequences.
Poole, William. 2009. Moral Hazard: The Long-Lasting Legacy of Bailouts, Cato Institute
working paper, April 30.
Rothschild, Michael and Joseph E. Stiglitz. "Equilibrium in Competitive Insurance Markets: An
Essay on the Economics of Imperfect Information," Quarterly Journal of Economics,
Nov. 1976, 90(4), pp. 630-49.
Skreta, Vasiliki, and Laura Veldkamp. 2009. Ratings Shopping and Asset Complexity: A
Theory of Ratings Inflation. Journal of Monetary Economics. Vol. 56, No. 5, pp. 678
695.

34
Smith, Yves. 2010. Econned: How Unenlightened Self Interest Undermined Democracy and
Corrupted Capitalism. New York, NY: Palgrave Macmillan.
Stiglitz, Joseph E. and Weiss, Andrew. "Credit Rationing in Markets with Imperfect
Information," Amer. Econ. Rev., June 1981, 71(3), pp. 393-410.


35

Figure 1: Aggregate Dollar Volume of CDOs Rated by S&P by Quarter
This figure shows the aggregate face value of CDO securities rated by S&P in each year 2001
through 2008. Values are calculated by quarter. Data is provided by the S&P RatingsXpress
issue/maturity database.

0
20
40
60
80
100
120
2001q1 2002q1 2003q1 2004q1 2003q1 2006q1 2007q1 2008q1
"
#
#
$
%
#
&
'
%

(
&
)
%

*
&
+
,
%

-
.

/
0
+
1

Cashow and Pybrld SynLheuc

36
Figure 2: Distribution of Securities by Rating
This figure shows the percentage distribution of securities in the ABSNet database observed in a
CDO in 2007H2 and the same type of distribution for the corresponding CDO liabilities in the
Pershing Square database.

Distribution calculated using security face value at origination

0
10
20
30
40
30
60
70
80
90
AAA AA A 888 < 888
203'$04,567 68 9%),$05%3 4: ;&57#
underlylng A8S CuC Llablllues

37
Figure 3: Histogram of Number of Underlying Securities per Deal

0
10
20
30
40
30
60
70
80
90
100
20 40 60 80 100 120 140 160 180 200 220 240 260 280 300 More
<
,
/
4
%
$

6
8

2
%
&
+
3

<,/4%$ 68 =7>%$+:07# 9%),$05%3 ?%$ 2%&+
<,/4%$ 68 =7>%$+:07# 9%),$05%3 ?%$ 2%&+

38
Figure 4: AAA Subordination by Vintage

Subordination is calculated as 1 [face value of AAA securities] / [face value of underlying
loans]. It is calculated by deal and then averaged by year of origination.

0.23
0.33
0.43
0.33
0.63
0.73
0.83
1999 2000 2001 2002 2003 2004 2003 2006 2007 2008
9
,
4
6
$
>
0
7
&
5
6
7

4
%
+
6
@

9
%
7
0
6
$

"
"
"

""" 9,46$>07&567 4: *07'&#%
CuCs CLher AsseL Classes

39
Figure 5: Distribution of S&P Rating Transitions after 2007H2 by Subsample
This figure shows the distribution of S&P rating transitions for two subsamples of securities in
the ABSNet dataset, those observed in a CDO in the Pershing Square dataset in 2007H2 and the
rest of the securities in the dataset that are active at the end of 2007. A transition of -1
corresponds to a downgrade of one fine rating notch, a downgrade from AA+ to AA, for
example.

0
10
20
30
40
30
60
70
80
90
-20 -18 -16 -14 -12 -10 -8 -6 -4 -2 0 2 4 6 8 10 More
203'$04,567 68 9AB ;&57# C$&7305673 &D%$ EFFG
CLher A8S Cbserved ln a CuC

40
Table 1: S&P Performance and Distribution of Underlying Securities by Vintage
This table shows the mean number of fine rating notches that a security was downgraded by S&P
from end of 2007H1 to end of 2009H1. It also shows the number and distribution of securities
not observed in a CDO, securities observed as collateral in non-synthetic deals and securities
observed as reference assets in synthetic deals.

Mean Collateral in a Reference asset Both collateral and
Halfyear of downgrade Not in a CDO non-synth CDO in a synth CDO a reference asset
initial rating severity N Dist. N Dist. N Dist. N Dist.
2002h1 0.5 3,154 5% 60 0% 9 0% 2 0%
2002h2 0.6 4,016 6% 106 1% 9 0% 0 0%
2003h1 0.5 5,989 9% 225 2% 31 1% 10 0%
2003h2 0.7 6,418 10% 339 2% 73 4% 45 1%
2004h1 1.0 5,979 9% 563 4% 162 8% 178 4%
2004h2 1.7 6,001 9% 1,600 11% 163 8% 469 9%
2005h1 2.1 5,764 9% 2,262 16% 155 7% 825 16%
2005h2 4.4 7,681 12% 2,985 21% 274 13% 1,101 22%
2006h1 7.9 6,080 9% 2,530 17% 314 15% 1,166 23%
2006h2 9.1 5,876 9% 2,280 16% 457 22% 876 17%
2007h1 9.1 7,746 12% 1,514 10% 423 20% 384 8%
Total 64,704 100% 14,464 100% 2,070 100% 5,056 100%

Collateral in 2 Collateral in >2 Reference asset Reference asset
Halfyear of non-synth CDO non-synth CDO in 2 synth CDO in >2 synth CDO
initial rating N Dist. N Dist. N Dist. N Dist.
2002h1 8 0% 1 0% 0 0% 0 0%
2002h2 15 0% 2 0% 0 0% 0 0%
2003h1 31 1% 6 0% 0 0% 0 0%
2003h2 43 1% 15 1% 9 3% 1 1%
2004h1 70 2% 29 1% 31 10% 13 7%
2004h2 322 10% 205 8% 33 10% 6 3%
2005h1 507 16% 611 23% 17 5% 10 5%
2005h2 688 22% 639 24% 28 9% 18 9%
2006h1 623 20% 571 21% 53 17% 38 20%
2006h2 494 16% 344 13% 66 21% 46 24%
2007h1 346 11% 255 10% 78 25% 61 32%
Total 3,147 100% 2,678 100% 315 100% 193 100%


41
Table 2: Collateral Type Distribution of Underlying Securities
This table shows the mean number of notches each security is downgraded by S&P from end of
2007H1 to end of 2009H1 by collateral type. It also shows the number of securities and
distribution for securities not observed in a CDO, securities observed as collateral in non-
synthetic deals and securities observed as reference assets in synthetic deals.

Mean Collateral in a Reference asset Both collateral and
downgrade Not in a CDO non-synth CDO in a synth CDO a reference asset
Collateral type severity N Dist. N Dist. N Dist. N Dist.
Auto loans 0.3 1,335 2% 11 0% 5 0% 2 0%
CDOs 4.8 1,699 3% 434 3% 105 5% 285 6%
CMBS 0.3 2,406 4% 164 1% 106 5% 78 2%
Credit cards 0.2 734 1% 26 0% 17 1% 8 0%
Home equity 5.0 33,363 52% 10,224 71% 1,390 67% 4,171 82%
RMBS 3.7 23,847 37% 3,572 25% 428 21% 489 10%
Student loans 0.0 1,108 2% 24 0% 15 1% 21 0%
Other 0.0 212 0% 9 0% 4 0% 2 0%
Total 64,704 100% 14,464 100% 2,070 100% 5,056 100%

non-synth CDO non-synth CDO in 2 synth CDO in >2 synth CDO
Collateral type N Dist. N Dist. N Dist. N Dist.
Auto loans 1 0% 0 0% 0 0% 0 0%
CDOs 99 3% 121 5% 15 5% 6 3%
CMBS 35 1% 18 1% 18 6% 14 7%
Credit cards 6 0% 2 0% 3 1% 0 0%
Home equity 2,478 79% 2,211 83% 219 70% 161 83%
RMBS 5 0% 1 0% 0 0% 0 0%
Student loans 517 16% 318 12% 58 18% 12 6%
Other 6 0% 7 0% 2 1% 0 0%
Total 3,147 100% 2,678 100% 315 100% 193 100%


42
Table 3: Number of Securities Rated by Each Combination of CRAs Underlying CDOs
Rated by Each Combination of CRAs
Each cell reports the number of underlying securities that were rated by a particular combination
of rating agencies (by row) and observed as collateral in a CDO rated by each combination of
rating agencies (by column). Ratings are taken from the end of 2007H1. The letters S, M, and F
correspond to S&P, Moodys, and Fitch respectively. A security will only occupy one row but
may be represented in more than one column if it was included in multiple CDO deals rated by
different combinations of rating agencies.

Observations in a CDO rated by Not in
Underlying rated by F M S MF SF SM SMF a CDO
F 49 0 0 0 271 61 38 6,786
M 0 0 0 0 16 130 46 1,763
S 18 0 0 1 847 533 293 8,903
MF 0 0 1 7 31 532 175 6,753
SF 3 0 2 9 249 701 304 14,447
SM 2 0 47 111 666 10,749 3,350 28,315
SMF 2 0 51 135 482 6,306 2505 13,013
Total 74 0 101 263 2,562 19,012 6,711 79,980


43
Table 4: Fraction Downgraded by Each Combination of CRAs Underlying CDOs Rated by
Each Combination of CRAs
Each cell reports the percent of underlying securities that were downgraded by any rating agency
from end of 2007H1 to end of 2009H1. Each cell corresponds to securities rated by a particular
combination of rating agencies (by row) and observed as collateral in a CDO rated by each
combination of rating agencies (by column). Ratings are taken from the end of 2007H1. The
letters S, M, and F correspond to S&P, Moodys, and Fitch respectively. A security will only
occupy one row but may be represented in more than one column if it was included in multiple
CDO deals rated by different combinations of rating agencies.

Observations in a CDO rated by Not in
Underlying rated by Fitch Moodys S&P M and F S and F S and M SMF a CDO
Fitch 98% 87% 90% 87% 48%
Moodys 25% 70% 67% 35%
S&P 78% 0% 55% 66% 65% 39%
Moodys and Fitch 100% 14% 65% 67% 48% 31%
S&P and Fitch 100% 50% 11% 78% 82% 67% 28%
S&P and Moodys 100% 91% 72% 86% 92% 92% 57%
All three agencies (SMF) 50% 96% 65% 78% 84% 87% 37%

44
Table 5: Summary Statistics for ABSNet Securities by Observation in a CDO
This table shows summary statistics by subsample. In order to calculate mean ratings, a numerical value was assigned to each rating,
assigning 1 to AAA, 2 to AA+, 3 to AA, etc. Therefore, a higher mean rating numerical value corresponds to a worse mean rating.

Collateral in a Reference asset Both collateral and
Not in a CDO non-synth CDO in a synth CDO a reference asset
Number of securities 80,007 15,517 2,173 5,134
Mean number of ratings 2.1 2.3 2.3 2.5
Mean Fitch rating 2.9 4.7 6.5 7.0
Mean Moody's rating 2.5 4.8 6.9 7.3
Mean S&P rating 3.0 4.9 6.6 6.9
Mean expected maturity (years) 29.7 31.1 31.1 31.2
Mean coupon spread 0.58 0.90 1.20 1.30
Floating-rate fraction 56% 80% 87% 96%
Mean subord. of AAA securities 13% 17% 18% 20%
Mean S&P downgrade severity
- as of Jul 1 2008 0.7 3.0 4.5 4.9
- as of Jul 1 2009 3.0 7.2 8.2 9.0
- as of Jun 1 2010 6.4 12.0 11.3 12.1


45
Table 6: Distribution of Securities by S&P Rating at the Start of the Performance Observation Period
This table shows the number and distribution of securities in the ABSNet dataset by coarse rating where rating is observed at the end
of 2007H1.

Collateral in a Reference asset Both collateral and
Rating at end Not in a CDO non-synth CDO in a synth CDO a reference asset
of 2007H1 N Dist. N Dist. N Dist. N Dist.
AAA 46,148 71% 2843 20% 295 14% 202 4%
AA 4,673 7% 4752 33% 290 14% 939 19%
A 4,333 7% 3,764 26% 416 20% 1,252 25%
BBB 4,245 7% 2,125 15% 924 45% 2,528 50%
BB and below 5,305 8% 980 7% 145 7% 135 3%
Total 64,704 100% 14,464 100% 2,070 100% 5,056 100%

Rating at end non-synth CDO non-synth CDO in 2 synth CDO in >2 synth CDO
of 2007H1 N Dist. N Dist. N Dist. N Dist.
AAA 474 15% 203 8% 15 5% 4 2%
AA 1272 40% 1,257 47% 21 7% 1 1%
A 906 29% 957 36% 66 21% 28 15%
BBB 400 13% 230 9% 185 59% 151 78%
BB and below 95 3% 31 1% 28 9% 9 5%
Total 3,147 100% 2,678 100% 315 100% 193 100%


46
Table 2: S&P Downgrade Severity by Rating at the Start of the Performance Observation Period
This table shows the mean number of fine rating notches that a security was downgraded by S&P. Each row corresponds to the
security rating at the end of 207H1.

Downgrade as of Jul 1 2008
Rating at end Collateral in a Reference asset Both collateral and
of 2007H1 Not in a CDO non-synth CDO in a synth CDO a reference asset
AAA 0.16 0.28 0.39 0.52
AA 1.42 3.03 3.11 3.86
A 2.10 4.79 5.40 5.83
BBB 2.94 3.75 5.74 5.13
BB and below 2.30 2.88 5.80 5.14
AAA 2.41 5.36 4.21 5.36
AA 4.08 7.62 8.85 8.90
A 4.37 8.47 9.44 9.86
BBB 5.13 7.14 8.86 8.99
BB and below 4.64 5.75 7.36 7.30
Downgrade as of Jun 1 2010
AAA 6.00 12.36 8.88 9.89
AA 8.07 12.65 14.01 13.67
A 8.07 12.49 12.45 13.06
BBB 7.85 10.70 11.14 11.44
BB and below 6.31 8.13 8.97 8.76


47
Table 3: S&P Downgrade Severity by Rating and Asset Type
This table shows the mean number of fine rating notches that a security was downgraded by S&P. Each row corresponds to the
security rating at the end of 2007H1. The first two columns correspond to non-CDO ABS securities only and the second set of two
columns correspond to CDO securities.

Rating at end Non-CDO ABS CDOs
of 2007H1 Not in a CDO In a CDO Not in a CDO In a CDO
AAA 0.16 0.25 0.36 1.32
AA 1.44 3.19 0.93 2.52
A 2.20 5.17 0.59 2.00
BBB 3.04 4.82 1.39 2.09
BB and below 2.36 3.56 0.75 0.60

AAA 2.38 4.99 4.32 9.90
AA 4.11 7.82 3.42 9.72
A 4.49 8.94 2.63 6.59
BBB 5.24 8.40 3.56 5.08
BB and below 4.75 6.26 1.98 1.80
Downgrade as of Jun 1 2010
AAA 5.99 11.84 6.80 13.03
AA 8.18 12.91 5.51 11.91
A 8.33 12.74 4.44 8.59
BBB 8.02 11.26 5.50 7.71
BB and below 6.42 8.41 3.65 5.33

48
Table 4: Tobit Regressions of S&P Downgrade Severity on ABS Characteristics
This table reports coefficients from Tobit regressions run over all securities in the ABSNet
database active at the end of 2007H1 where the dependent variable indicates how many fine
rating notches a security was downgraded by S&P from end of 2007H1 to end of 2009H1. The
dependent variable takes the value zero if no transition occurred or if an upgrade occurred. All
ratings below CCC- are treated as censored. Subordination is calculated as 1 [face value of
AAA securities]/[total balance of underlying loans]. Subordination is calculated for AAA
securities only and interacted with a fixed effect indicating AAA securities. Coupon spreads are
calculated using the coupon at origination, not during 2007H1. Standard errors are clustered by
deal.

(1) (2) (3) (4) (5)
Variables All All All All All
Security in non-synthetic CDO 3.021*** 2.891*** 2.653*** 1.466*** 1.455***
(0.119) (0.120) (0.120) (0.083) (0.084)
Security in synthetic CDO 4.112*** 3.894*** 3.802*** 1.723*** 1.738***
(0.210) (0.208) (0.206) (0.154) (0.158)
Security in both synth & non-synth CDO 4.642*** 4.457*** 4.053*** 2.714*** 2.734***
(0.181) (0.185) (0.184) (0.140) (0.141)
Security downgraded before 2007H2 3.014*** 3.282***
(0.360) (0.374)
Security upgraded before 2007H2 -0.637*** -0.756***
(0.114) (0.120)
(AAA)x(Subordination) -2.393*** -2.470***
(0.423) (0.444)
Floating-rate security 0.995*** -0.002
(0.140) (0.110)
Mean spread 1.864***
(0.190)
Difference from mean spread -0.237*** 0.021
(0.024) (0.019)
Constant 2.389*** 2.213*** 1.177*** -1.487*** -1.538***
(0.098) (0.216) (0.254) (0.302) (0.315)
Observations 86294 86294 86294 86294 86294
R-squared 0.029 0.030 0.033 0.100 0.101
Initial rating FE included Yes Yes Yes Yes Yes
Rating agency FE included No Yes Yes Yes Yes
Vintage halfyear FE included No No No Yes Yes
Collateral type FE included No No No Yes Yes
Sponsor rating FE included No No No Yes Yes
Sponsor diversification FE included No No No Yes Yes
Servicer FE included No No No Yes Yes
*** p<0.01, ** p<0.05, * p<0.1

49

Table 5: Tobit Regressions by Initial Rating
Continued from Table 4, these regressions are run by security rating at the end of 2007H1. This
table reports coefficients from Tobit regressions run over securities in the ABSNet database
active at the end of 2007H1 where the dependent variable indicates how many rating notches a
security was downgraded by S&P from end of 2007H1 to end of 2009H1. The dependent
variable takes the value zero if no transition occurred or if an upgrade occurred. All ratings
below CCC- are treated as censored. Standard errors are clustered by deal.

(1) (2) (3) (4) (5)
Variables AAA AA A BBB BB and below
Security in non-synthetic CDO 1.306*** 0.614*** 0.798*** 0.707*** 0.016
(0.138) (0.163) (0.174) (0.182) (0.190)
Security in synthetic CDO 0.803** 0.558 1.668*** 1.178*** -0.287
(0.335) (0.285) (0.329) (0.255) (0.444)
Security in both synth & non-synth CDO 2.143*** 0.578** 1.784*** 1.608*** -0.207
(0.382) (0.285) (0.255) (0.217) (0.379)
Security downgraded before 2007H2 3.390** -0.061 2.681*** 3.341*** 3.070***
(1.720) (0.182) (0.994) (0.891) (0.368)
Security upgraded before 2007H2 0.745*** 0.138 -0.375 -0.223 -0.481*
(0.116) (0.182) (0.259) (0.318) (0.259)
(AAA)x(Subordination) -4.277***
(0.508)
Floating-rate security -0.408*** 0.911*** 1.223*** 0.815*** 0.625***
(0.122) (0.063) (0.236) (0.242) (0.225)
Difference from mean spread 0.058*** -0.191*** -0.128** 0.189*** -0.066
(0.020) (0.063) (0.063) (0.064) (0.040)
Constant -0.609** -2.771*** -3.691*** -5.208*** -3.863***
(0.294) (0.163) (0.764) (0.802) (0.905)
Observations 49488 10654 9765 9822 6565
R-squared 0.068 0.109 0.128 0.126 0.167
Rating agency FE included Yes Yes Yes Yes Yes
Vintage halfyear FE included Yes Yes Yes Yes Yes
Collateral type FE included Yes Yes Yes Yes Yes
Sponsor rating FE included Yes Yes Yes Yes Yes
Sponsor diversification FE included Yes Yes Yes Yes Yes
Servicer FE included Yes Yes Yes Yes Yes
*** p<0.01, ** p<0.05, * p<0.1


50
Table 6: Tobit Regressions by Initial Rating Continued
Continued from Table 4, these regressions are run by security rating at the end of 2007H1. This
table reports coefficients from Tobit regressions run over securities in the ABSNet database
security was downgraded by S&P from end of 2007H1 to end of 2009H1. The dependent
variable takes the value zero if no transition occurred or if an upgrade occurred. All ratings
below CCC- are treated as censored. Standard errors are clustered by deal.

(1) (2) (3) (4) (5)
Variables AAA AA A BBB BB and below
(0.184) (0.204) (0.213) (0.227) (0.262)
(0.407) (0.523) (0.448) (0.321) (0.613)
(0.509) (0.337) (0.301) (0.237) (0.579)
Constant 2.410*** 4.179*** 4.588*** 5.728*** 5.569***
(0.098) (0.148) (0.149) (0.157) (0.154)
Observations 49488 10654 9765 9822 6565
R-squared 0.003 0.011 0.021 0.018 0.003


51
Table 7: Tobit Regressions Restricted to 2007H1 Vintage
These regressions are restricted to securities issued during 2007H1 only. This table reports
coefficients from Tobit regressions run over securities in the ABSNet database active at the end
of 2007H1 where the dependent variable indicates how many rating notches a security was
downgraded by S&P from end of 2007H1 to end of 2009H1. The dependent variable takes the
value zero if no transition occurred or if an upgrade occurred. All ratings below CCC- are treated
as censored. Standard errors are clustered by deal.

(1) (2) (3) (4) (5)
Variables 2007H1 2007H1 2007H1 2007H1 2007H1
Security in non-synthetic CDO 0.774** 0.968*** 1.035*** 1.048*** 1.043***
(0.322) (0.337) (0.341) (0.295) (0.296)
(0.472) (0.479) (0.481) (0.455) (0.452)
Security in both synth & non-synth CDO 0.815 0.903* 1.091** 1.176** 1.287***
(0.499) (0.518) (0.516) (0.461) (0.459)
(1.855) (1.872)
Floating-rate security -0.451 -0.277
(0.539) (0.442)
Difference from mean spread 0.127 0.288***
(0.123) (0.088)
Constant 7.468*** 6.512*** 6.722*** 1.466 1.489
(0.398) (0.904) (0.990) (1.371) (1.380)
Observations 10067 10067 10067 10067 10067
R-squared 0.027 0.031 0.030 0.061 0.062
*** p<0.01, ** p<0.05, * p<0.1


52
Table 8: Tobit Regressions Restricted to CDO Underlying
These regressions are restricted to CDO securities only. This table reports coefficients from
Tobit regressions run over securities in the ABSNet database active at the end of 2007H1 where
the dependent variable indicates how many rating notches a security was downgraded by S&P
from end of 2007H1 to end of 2009H1. The dependent variable takes the value zero if no
transition occurred or if an upgrade occurred. All ratings below CCC- are treated as censored.
Coupon spreads are calculated using the coupon at origination, not during 2007H2. Standard
errors are clustered by deal.

(1) (2) (3) (4) (5)
Variables CDOs CDOs CDOs CDOs CDOs
(0.513) (0.509) (0.506) (0.433) (0.436)
(0.903) (0.884) (0.876) (0.779) (0.787)
(0.773) (0.760) (0.759) (0.629) (0.645)
(1.333) (1.348)
(0.812) (0.813)
(AAA)x(Subordination) 5.026*** 5.025***
(1.810) (1.804)
Floating-rate security 0.238 -0.415
(0.424) (0.394)
Mean spread 0.160
(0.577)
Difference from mean spread -0.138* 0.018
(0.079) (0.093)
Constant 4.832*** 6.985*** 6.783*** 4.946*** 5.378***
(0.367) (0.637) (0.778) (1.501) (1.533)
Observations 2523 2523 2523 2523 2523
R-squared 0.025 0.034 0.035 0.057 0.058
*** p<0.01, ** p<0.05, * p<0.1


53
Table 9: Number of CDO Deals and S&P Downgrade Severity
These regressions include additional fixed effects for the number of CDO deals in which a
security was observed.

(1) (2) (3) (4) (5)
(0.142) (0.134) (0.131) (0.090) (0.091)
(0.773) (0.212) (0.210) (0.156) (0.160)
(0.186) (0.186) (0.185) (0.144) (0.144)
Security in 2 CDOs 0.412** 0.633*** 0.409** 0.422*** 0.441***
(0.178) (0.173) (0.172) (0.133) (0.133)
Security in > 2 CDOs 0.233 0.768*** 0.366* 0.539*** 0.560***
(0.227) (0.218) (0.218) (0.169) (0.169)
(0.369) (0.382)
(0.114) (0.120)
(0.408) (0.428)
Floating-rate security 1.481*** 0.000
(0.100) (0.110)
(0.234)
(0.023) (0.019)
Observations 86294 86294 86294 86294 86294
*** p<0.01, ** p<0.05, * p<0.1


54
Table 10: Type of CDO First Closed
This table includes a fixed effect that indentifies whether a security was observed first in a non-
synthetic or synthetic CDO deal for those securities that were observed in both.

(1) (2) (3) (4) (5)
Non-synthetic deal before synthetic -3.722*** -3.415*** -3.506*** -0.155 -0.120
(0.259) (0.258) (0.256) (0.209) (0.211)
(0.128) (0.124) (0.121) (0.083) (0.084)
(0.217) (0.208) (0.206) (0.154) (0.158)
(0.209) (0.209) (0.206) (0.166) (0.167)
(0.368) (0.382)
(0.114) (0.120)
(0.408) (0.428)
(0.100) (0.110)
(0.234)
(0.023) (0.019)
Observations 86294 86294 86294 86294 86294
*** p<0.01, ** p<0.05, * p<0.1


55
Table 11: Alternate Trustee Report and Performance Period
This table reports coefficients from Tobit regressions run over all securities in the ABSNet
database active at the end of 2008M1 where the dependent variable indicates how many rating
notches a security was downgraded by S&P from end of 2008M1 to end of 2009H1. The
dependent variable takes the value zero if no transition occurred or if an upgrade occurred. All
ratings below CCC- are treated as censored. Standard errors are clustered by deal.

(1) (2) (3) (4) (5)
(0.110) (0.111) (0.111) (0.079) (0.080)
(0.201) (0.199) (0.199) (0.153) (0.157)
(0.163) (0.169) (0.168) (0.141) (0.141)
Security downgraded before 2008M1 2.258*** 2.310***
(0.159) (0.161)
Security upgraded before 2008M1 -0.852*** -0.978***
(0.106) (0.112)
(0.426) (0.444)
(0.137) (0.108)
(0.200)
Difference from mean spread -0.212*** 0.044**
(0.024) (0.019)
Constant 2.387 2.160*** 1.265*** -1.149*** -1.163***
(0.098) (0.211) (0.246) (0.286) (0.299)
Observations 86294 86294 86294 86294 86294
R-squared 0.016 0.016 0.019 0.081 0.081
*** p<0.01, ** p<0.05, * p<0.1

56
Table 18:
These regressions include additional fixed effects for whether a security was in CDO sponsored
by Domestic Bank, Foreign Bank, Major Investment Bank.
(1) (2) (3) (4) (5)
(0.122) (0.122) (0.122) (0.086) (0.087)
(0.212) (0.210) (0.209) (0.158) (0.162)
(0.194) (0.196) (0.196) (0.150) (0.151)
Security in CDO sponsored by Domestic Bank 1.271*** 1.257*** 1.371*** 1.021*** 1.069***
(0.248) (0.248) (0.242) (0.194) (0.196)
Security in CDO sponsored by Foreign Bank 0.715*** 0.672*** 0.511*** 0.281** 0.273*
(0.179) (0.179) (0.178) (0.139) (0.139)
Security in CDO sponsored by Major Investment Bank -1.075*** -1.061*** -0.971*** 0.016 0.063
(0.179) (0.180) (0.177) (0.144) (0.146)
(0.360) (0.374)
(0.114) (0.120)
(0.422) (0.444)
(0.139) (0.109)
(0.190)
(0.024) (0.019)
Constant 2.394*** 2.220*** 1.186*** -1.486*** -1.537***
(0.098) (0.216) (0.254) (0.302) (0.315)
Observations 86294 86294 86294 86294 86294
R-squared 0.029 0.030 0.034 0.100 0.101
*** p<0.01, ** p<0.05, * p<0.1


57
Table 19:
These regressions include additional fixed effects for whether a security was in CDO with the
same sponsor.

(1) (2) (3) (4) (5)
(0.120) (0.120) (0.120) (0.083) (0.084)
(0.210) (0.208) (0.206) (0.154) (0.158)
(0.181) (0.185) (0.183) (0.141) (0.141)
Security in CDO with the same Sponsor -0.789*** -0.812*** -0.730** 0.965*** 1.002***
(0.288) (0.287) (0.285) (0.271) (0.279)
(0.358) (0.372)
(0.114) (0.120)
(0.422) (0.443)
(0.140) (0.109)
(0.189)
(0.024) (0.019)
Constant 2.399*** 2.223*** 1.187*** -1.487*** -1.540***
(0.098) (0.216) (0.254) (0.301) (0.315)
Observations 86294 86294 86294 86294 86294
R-squared 0.029 0.030 0.033 0.100 0.101
*** p<0.01, ** p<0.05, * p<0.1


58
Appendix 1: Control Coefficients from Base Specification Regressions
This table reports coefficients from control variables not reported in Table 4. The columns
correspond to the regressions indicated in Table 4.

Variables (1) (2) (3) (4) (5)
Vintage (2002H2) -0.003 0.048
(0.119) (0.126)
Vintage (2003H1) -0.388*** -0.335***
(0.113) (0.120)
Vintage (2003H2) -0.544*** -0.542***
(0.113) (0.119)
Vintage (2004H1) -0.554*** -0.503***
(0.116) (0.124)
Vintage (2004H2) -0.225* -0.158
(0.126) (0.136)
Vintage (2005H1) -0.153 -0.095
(0.147) (0.155)
Vintage (2005H2) 2.298*** 2.378***
(0.194) (0.201)
Vintage (2006H1) 5.893*** 6.029***
(0.235) (0.240)
Vintage (2006H2) 7.340*** 7.485***
(0.262) (0.270)
Vintage (2007H1) 7.714*** 7.879***
(0.284) (0.296)
Collateral type (Others) -0.876** -1.291***
(0.400) (0.437)
Collateral type (CDOs) 3.847*** 3.938***
(0.426) (0.446)
Collateral type (CMBS) -1.252*** -1.347***
(0.343) (0.405)
Collateral type (Credit cards) -2.265*** -2.208***
(0.397) (0.419)
Collateral type (Home equity) 1.568*** 1.612***
(0.231) (0.255)
Collateral type (RMBS) 1.613*** 1.639***
(0.258) (0.282)
Collateral type (Student loans) -1.043*** -0.994**
(0.378) (0.394)
Initial rating (AA+) 1.309*** 1.325*** 0.506*** 1.601*** 1.559***
(0.168) (0.168) (0.183) (0.134) (0.132)
Initial rating (AA) 1.800*** 1.917*** 1.090*** 2.458*** 2.399***
(0.122) (0.124) (0.144) (0.095) (0.093)
Initial rating (AA-) 3.866*** 3.896*** 3.209*** 3.509*** 3.485***
(0.189) (0.189) (0.205) (0.142) (0.142)


59
Appendix 1 Continued

Variables (1) (2) (3) (4) (5)
Initial rating (A+) 3.846*** 3.892*** 2.535*** 3.939*** 3.927***
(0.183) (0.184) (0.234) (0.140) (0.140)
Initial rating (A) 1.961*** 2.108*** 0.740*** 2.952*** 2.895***
(0.123) (0.127) (0.192) (0.097) (0.097)
Initial rating (A-) 3.514*** 3.590*** 2.312*** 3.892*** 3.878***
(0.162) (0.165) (0.219) (0.131) (0.133)
Initial rating (BBB+) 3.409*** 3.484*** 0.870*** 4.176*** 4.178***
(0.167) (0.169) (0.330) (0.131) (0.134)
Initial rating (BBB) 2.101*** 2.283*** -0.423 3.192*** 3.136***
(0.130) (0.134) (0.300) (0.104) (0.105)
Initial rating (BBB-) 4.407*** 4.558*** 2.156*** 4.784*** 4.812***
(0.178) (0.182) (0.327) (0.147) (0.151)
Initial rating (BB+) 4.994*** 5.117*** 2.454*** 5.228*** 5.420***
(0.246) (0.250) (0.403) (0.212) (0.226)
Initial rating (BB) 3.127*** 3.552*** 0.181 3.749*** 3.635***
(0.144) (0.155) (0.332) (0.130) (0.132)
Initial rating (BB-) 1.128*** 1.285*** -0.822** 2.826*** 2.759***
(0.289) (0.296) (0.407) (0.216) (0.243)
Initial rating (B+) -0.829*** -0.651*** -1.761*** 2.170*** 1.989***
(0.237) (0.246) (0.352) (0.220) (0.258)
Initial rating (B) 3.421*** 3.991*** 1.079*** 4.095*** 3.990***
(0.152) (0.170) (0.316) (0.141) (0.145)
Initial rating (B-) -0.311 -0.147 -0.621* 2.915*** 2.798***
(0.272) (0.285) (0.354) (0.226) (0.282)
Initial rating (Below B) 6.990*** 7.143*** 4.804*** 5.532*** 5.502***
(0.721) (0.729) (0.781) (0.604) (0.673)
Rated by S&P only -0.583** -0.373 -0.048 0.214
(0.226) (0.245) (0.175) (0.182)
Rated by S&P and Fitch -0.066 0.164 0.339* 0.355*
(0.269) (0.290) (0.203) (0.211)
Rated by S&P and Moody's 0.488** 0.523** -0.002 0.030
(0.217) (0.222) (0.161) (0.161)
Parent rating FE (AAA) 1.141*** 1.096***
(0.235) (0.239)
Parent rating FE (AA) 0.817* 0.817*
(0.419) (0.443)
Parent rating FE (A) 0.035 -0.009
(0.170) (0.174)
Parent rating FE (BBB) 0.739*** 0.726***
(0.136) (0.140)
Parent rating FE (NR) -0.158 -0.173
(0.168) (0.172)
I(collateral types issued by seller > 4) -0.905*** -0.944***
(0.136) (0.139)
I(seller = servicer) -0.128 -0.182
(0.127) (0.128)
I(servicer unidentified) 0.143 0.098
(0.223) (0.228)

60

Appendix 2: ABACUS Performance
This table reports coefficients from Tobit regressions run over securities in the ABSNet database
security was downgraded by S&P from end of 2007H1 to end of 209H1. The dependent variable
takes the value zero if no transition occurred or if an upgrade occurred. All ratings below CCC-
are treated as censored. Coupon spreads are calculated using the coupon at origination, not
during 2007H2. Standard errors are clustered by deal.

(1) (2) (3) (4) (5)
Security in an ABACUS deal -2.381*** -2.136*** -1.689*** 0.584** 0.801***
(0.325) (0.325) (0.315) (0.242) (0.254)
(0.128) (0.124) (0.121) (0.083) (0.084)
(0.216) (0.207) (0.206) (0.157) (0.160)
(0.187) (0.187) (0.185) (0.143) (0.144)
(0.368) (0.382)
(0.114) (0.120)
(0.408) (0.428)
(0.100) (0.110)
(0.234)
(0.023) (0.019)
Observations 86294 86294 86294 86294 86294
*** p<0.01, ** p<0.05, * p<0.1


61
Appendix 3: Time Series
This table reports coefficients from Tobit regressions run over securities in the ABSNet database active at the end of 2007H1 where
the dependent variable indicates how many rating notches a security was downgraded by S&P 6-, 12-, 18-, 24-, 30-, and 35-month
after end of 2007H1. The dependent variable takes the value zero if no transition occurred or if an upgrade occurred. All ratings below
CCC- are treated as censored. Coupon spreads are calculated using the coupon at origination, not during 2007H2. Standard errors are
clustered by deal.

35 months 30 months 24 months 18 months 12 months 6 months
Jun 1 2010 Dec 31 2009 Jun 30 2009 Dec 31 2008 Jun 30 2008 Dec 31 2007
(1) (5) (1) (5) (1) (5) (1) (5) (1) (5) (1) (5)
Variables All All All All All All All All All All All All
Security in non-synthetic CDO 6.108*** 2.214*** 5.034*** 1.894*** 3.021*** 1.455*** 2.299*** 1.213*** 1.031*** 0.519*** 0.276*** 0.090***
(0.164) (0.105) (0.147) (0.097) (0.119) (0.084) (0.099) (0.071) (0.064) (0.054) (0.031) (0.027)
Security in synthetic CDO 6.231*** 2.021*** 5.581*** 1.880*** 4.112*** 1.738*** 3.449*** 1.574*** 2.125*** 1.218*** 0.563*** 0.303***
(0.324) (0.227) (0.287) (0.205) (0.210) (0.158) (0.177) (0.138) (0.131) (0.112) (0.071) (0.069)
Security in both synth & non-synth CDO 7.226*** 2.562*** 6.135*** 2.358*** 4.642*** 2.734*** 4.088*** 2.648*** 2.209*** 1.466*** 0.255*** -0.024
(0.255) (0.187) (0.234) (0.174) (0.181) (0.141) (0.153) (0.124) (0.128) (0.106) (0.053) (0.053)
Security downgraded before 2007H2 2.948*** 3.179*** 3.282*** 2.998*** 2.209*** 1.814***
(0.545) (0.483) (0.374) (0.311) (0.226) (0.151)
Security upgraded before 2007H2 0.629*** 0.228 -0.756*** -1.022*** -0.901*** -0.195***
(0.205) (0.188) (0.120) (0.096) (0.065) (0.036)
(AAA)x(Subordination) -6.812*** -5.366*** -2.470*** -0.595* 1.617*** 1.245***
(0.577) (0.529) (0.444) (0.347) (0.230) (0.151)
Floating-rate security 0.413*** 0.625*** -0.002 0.196** 0.227*** 0.012
(0.123) (0.115) (0.110) (0.084) (0.044) (0.022)
Mean spread

Difference from mean spread 0.001 -0.024 0.021 0.014 -0.014** 0.009***
(0.022) (0.020) (0.019) (0.016) (0.007) (0.003)
Constant 6.280*** -4.675*** 5.089*** -3.738*** 2.389*** -1.538*** 1.333*** -0.561** 0.090*** -0.420*** 0.003 -0.295***
(0.156) (0.491) (0.129) (0.434) (0.098) (0.315) (0.068) (0.245) (0.010) (0.155) (0.004) (0.072)
Observations 86294 86294 86294 86294 86294 86294 86294 86294 86294 86294 86294 86294
R-squared 0.031 0.146 0.031 0.134 0.029 0.101 0.037 0.100 0.055 0.092 0.036 0.054
Initial rating FE included Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Rating agency FE included No Yes No Yes No Yes No Yes No Yes No Yes
Vintage halfyear FE included No Yes No Yes No Yes No Yes No Yes No Yes
Collateral type FE included No Yes No Yes No Yes No Yes No Yes No Yes
Sponsor rating FE included No Yes No Yes No Yes No Yes No Yes No Yes
Sponsor diversification FE included No Yes No Yes No Yes No Yes No Yes No Yes
Servicer FE included No Yes No Yes No Yes No Yes No Yes No Yes
*** p<0.01, ** p<0.05, * p<0.1

62
Appendix 4: Marginal Effects of Logit Regressions
This table reports marginal effects coefficients from Logit regressions run over securities in the ABSNet database active at the end of
2007H1 where the dependent variable indicates the probability a security was downgraded by 6 fine rating notches or more by S&P 6-
, 12-, 18-, 24-, 30-, and 35-month after end of 2007H1. Coupon spreads are calculated using the coupon at origination, not during
2007H2. Standard errors are clustered by deal.

(1) (5) (1) (5) (1) (5) (1) (5) (1) (5) (1) (5)
Security in non-synthetic CDO 0.294*** 0.123*** 0.283*** 0.116*** 0.213*** 0.076*** 0.170*** 0.050*** 0.040*** 0.005*** 0.005*** 0.000
(0.007) (0.009) (0.007) (0.010) (0.008) (0.007) (0.008) (0.005) (0.003) (0.001) (0.001) (0.000)
Security in synthetic CDO 0.250*** 0.127*** 0.269*** 0.139*** 0.288*** 0.098*** 0.267*** 0.056*** 0.095*** 0.008*** 0.006*** -0.001**
(0.012) (0.021) (0.013) (0.023) (0.015) (0.017) (0.015) (0.011) (0.008) (0.002) (0.001) (0.000)
Security in both synth & non-synth CDO 0.333*** 0.155*** 0.334*** 0.156*** 0.336*** 0.160*** 0.320*** 0.123*** 0.084*** 0.010*** 0.003*** -0.002***
(0.009) (0.016) (0.010) (0.018) (0.013) (0.014) (0.013) (0.012) (0.007) (0.002) (0.001) (0.000)
Security downgraded before 2007H2 0.316*** 0.389*** 0.460*** 0.539*** 0.167*** 0.025***
(0.021) (0.023) (0.036) (0.041) (0.025) (0.005)
Security upgraded before 2007H2 0.128*** 0.098*** -0.035* -0.051*** -0.013*** -0.001
(0.024) (0.028) (0.020) (0.010) (0.003) (0.001)
(AAA)x(Subordination) -0.697*** -0.705*** -0.292*** -0.100*** 0.021*** 0.009***
(0.066) (0.070) (0.045) (0.026) (0.006) (0.001)
Floating-rate security 0.047*** 0.073*** 0.009 0.018*** 0.008*** 0.000
(0.012) (0.013) (0.009) (0.006) (0.002) (0.000)
Mean spread

Difference from mean spread 0.012*** 0.010*** 0.009*** 0.008*** 0.002*** 0.001***
(0.002) (0.003) (0.002) (0.001) (0.000) (0.000)
Mean 0.559 0.567 0.500 0.462 0.298 0.196 0.205 0.107 0.051 0.023 0.008 0.003
Observations 86074 86074 86074 86074 86074 86074 86074 86074 86074 86074 86074 86074
*** p<0.01, ** p<0.05, * p<0.1


63
2007H1 where the dependent variable indicates the probability a security was downgraded by 9 fine rating notches or more by S&P 6-
, 12-, 18-, 24-, 30-, and 35-month after end of 2007H1. Coupon spreads are calculated using the coupon at origination, not during

(1) (5) (1) (5) (1) (5) (1) (5) (1) (5) (1) (5)
Security in non-synthetic CDO 0.295*** 0.119*** 0.278*** 0.105*** 0.193*** 0.055*** 0.146*** 0.034*** 0.028*** 0.003*** 0.004*** 0.000
(0.007) (0.010) (0.008) (0.009) (0.008) (0.006) (0.007) (0.004) (0.003) (0.001) (0.001) (0.000)
Security in synthetic CDO 0.263*** 0.128*** 0.282*** 0.136*** 0.270*** 0.061*** 0.238*** 0.030*** 0.067*** 0.003*** 0.005*** 0.000
(0.012) (0.023) (0.013) (0.023) (0.014) (0.012) (0.014) (0.007) (0.006) (0.001) (0.001) (0.000)
Security in both synth & non-synth CDO 0.342*** 0.153*** 0.326*** 0.125*** 0.289*** 0.096*** 0.254*** 0.066*** 0.053*** 0.004*** 0.001 -0.001***
(0.010) (0.018) (0.011) (0.018) (0.013) (0.011) (0.013) (0.008) (0.005) (0.001) (0.001) (0.000)
Security downgraded before 2007H2 0.386*** 0.420*** 0.469*** 0.503*** 0.123*** 0.004**
(0.028) (0.033) (0.048) (0.057) (0.028) (0.002)
Security upgraded before 2007H2 0.120*** 0.088*** -0.034** -0.034*** -0.008*** 0.000
(0.028) (0.029) (0.015) (0.007) (0.002) (0.001)
(0.068) (0.067) (0.034) (0.019) (0.004) (0.001)
Floating-rate security 0.044*** 0.062*** 0.006 0.013*** 0.004*** 0.000
(0.013) (0.012) (0.007) (0.004) (0.001) (0.000)
Mean spread

(0.003) (0.002) (0.001) (0.001) (0.000) (0.000)
Mean 0.506 0.464 0.450 0.370 0.250 0.132 0.160 0.068 0.032 0.012 0.004 0.001
Observations 83679 83679 83679 83679 83679 83679 83679 83679 83679 83679 83679 83679
*** p<0.01, ** p<0.05, * p<0.1


64
2007H1 where the dependent variable indicates the probability a security was downgraded by 12 fine rating notches or more by S&P
6-, 12-, 18-, 24-, 30-, and 35-month after end of 2007H1. Coupon spreads are calculated using the coupon at origination, not during

(1) (5) (1) (5) (1) (5) (1) (5) (1) (5) (1) (5)
Security in non-synthetic CDO 0.288*** 0.106*** 0.264*** 0.085*** 0.168*** 0.040*** 0.106*** 0.017*** 0.009*** 0.000** 0.002*** 0.000
(0.008) (0.010) (0.008) (0.008) (0.008) (0.004) (0.007) (0.002) (0.001) (0.000) (0.001) (0.000)
Security in synthetic CDO 0.293*** 0.106*** 0.295*** 0.082*** 0.241*** 0.034*** 0.153*** 0.007** 0.011*** 0.000 0.001 0.000
(0.013) (0.023) (0.014) (0.020) (0.015) (0.008) (0.013) (0.003) (0.002) (0.000) (0.001) (0.000)
Security in both synth & non-synth CDO 0.306*** 0.103*** 0.299*** 0.095*** 0.238*** 0.064*** 0.146*** 0.018*** 0.011*** 0.000 -0.001* 0.000
(0.011) (0.017) (0.012) (0.015) (0.013) (0.008) (0.011) (0.004) (0.002) (0.000) (0.000) (0.000)
Security downgraded before 2007H2 0.198** 0.156* 0.104* 0.056** 0.003 0.000
(0.088) (0.088) (0.055) (0.028) (0.002) (0.001)
Security upgraded before 2007H2 0.132*** 0.096*** -0.031*** -0.015*** -0.001** 0.000
(0.029) (0.028) (0.011) (0.005) (0.000) (0.000)
(AAA)x(Subordination) -0.671*** -0.550*** -0.139*** -0.036*** 0.004*** 0.001
(0.066) (0.059) (0.026) (0.011) (0.001) (0.001)
Floating-rate security 0.037*** 0.049*** 0.001 0.006** 0.001*** 0.000
(0.011) (0.010) (0.005) (0.002) (0.000) (0.000)
Mean spread

Difference from mean spread 0.012*** 0.009*** 0.005*** 0.002*** 0.000*** 0.000
(0.002) (0.002) (0.001) (0.001) (0.000) (0.000)
Mean 0.428 0.323 0.376 0.249 0.188 0.086 0.098 0.033 0.007 0.002 0.002 0.000
Observations 79729 79729 79729 79729 79729 79729 79729 79729 79729 79729 79729 79729
*** p<0.01, ** p<0.05, * p<0.1


65
2007H1 where the dependent variable indicates the probability a security was downgraded from investment grade to non-investment
grade by S&P 6-, 12-, 18-, 24-, 30-, and 35-month after end of 2007H1. Coupon spreads are calculated using the coupon at
origination, not during 2007H2. Standard errors are clustered by deal.

(1) (5) (1) (5) (1) (5) (1) (5) (1) (5) (1) (5)
Security in non-synthetic CDO 0.315*** 0.145*** 0.300*** 0.129*** 0.217*** 0.070*** 0.167*** 0.042*** 0.030*** 0.004*** 0.004*** 0.001**
(0.007) (0.010) (0.008) (0.010) (0.009) (0.006) (0.008) (0.004) (0.003) (0.001) (0.001) (0.000)
Security in synthetic CDO 0.275*** 0.172*** 0.291*** 0.166*** 0.293*** 0.090*** 0.257*** 0.046*** 0.064*** 0.005*** 0.008*** 0.001*
(0.013) (0.023) (0.013) (0.025) (0.015) (0.015) (0.015) (0.009) (0.007) (0.002) (0.002) (0.000)
Security in both synth & non-synth CDO 0.366*** 0.190*** 0.364*** 0.183*** 0.349*** 0.150*** 0.322*** 0.110*** 0.060*** 0.008*** 0.005*** 0.000
(0.009) (0.018) (0.010) (0.019) (0.013) (0.013) (0.013) (0.011) (0.006) (0.002) (0.001) (0.000)
Security downgraded before 2007H2 0.117 0.126* 0.218*** 0.157*** 0.034** 0.009**
(0.076) (0.073) (0.071) (0.055) (0.013) (0.004)
Security upgraded before 2007H2 0.162*** 0.118*** -0.042*** -0.038*** -0.007*** -0.001**
(0.027) (0.029) (0.014) (0.007) (0.001) (0.001)
(0.070) (0.069) (0.037) (0.020) (0.003) (0.002)
Floating-rate security 0.039*** 0.065*** 0.012 0.020*** 0.006*** 0.001**
(0.013) (0.013) (0.008) (0.005) (0.001) (0.000)
Mean spread

(0.003) (0.003) (0.002) (0.001) (0.000) (0.000)
Mean 0.527 0.495 0.467 0.392 0.257 0.147 0.170 0.076 0.030 0.012 0.006 0.003
Observations 79729 79729 79729 79729 79729 79729 79729 79729 79729 79729 79729 79729
*** p<0.01, ** p<0.05, * p<0.1


66

Appendix 8:
This is a list of Domestic Bank, Foreign Bank, Major Investment Bank included in the
regressions of Table 18, and the number of CDOs in our sample that they have sponsored.

Domestic Bank # of CDO sponsored
Wachovia 5
J.P. Morgan Chase 4
PNC 4
E*Trade 3
Citigroup 2
Bank of America 1
SunTrust 1

Foreign Bank # of CDO sponsored
Societe Generale 26
Deutsche Bank 11
Credit Suisse 8
UBS 7
Fortis Bank 5
Rabobank 3
Royal Bank of Scotland 3
Royal Bank of Canada 2
Hypo Real Estate Group 2
Gulf International Bank 1
ING 1
KBC Bank 1

Major Investment Bank # of CDO sponsored
Goldman Sachs 16
Merrill Lynch 13
Lehman Brothers 12
Bear Stearns 9

LemonsAndCDOsWhyDidSoManyLenders Preview

Uploaded by

Copyright:

Available Formats

LemonsAndCDOsWhyDidSoManyLenders Preview

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

LemonsAndCDOsWhyDidSoManyLenders Preview

Uploaded by

Copyright:

Available Formats

Preliminary Comments appreciated!

Lemons and CDOs

You might also like