Вы находитесь на странице: 1из 12

Default Rates on

Structured Finance Securities


DOUGLAS J. LUCAS, LAURIE S. GOODMAN, AND FRANK J. FABOZZI

eceiu rating agency reports attack "material impairment" (either an uncured pay-

R
DOUGLAS J. LUCAS
is dirL'ctor <it t^DO :\ iong-held assumption ot partic- ment detanit or a dtjwngrading to C'a or C")
Research at UBS m ipants in the structured finance over tive years.
New York Cit\-,
market: that asset-backed securities Almost all market participants think tiie
d«uglas.lucas@ubs.com
(ABS), connnercial mortgage-backed securi- default rate tor BBB Resi B^"(~ has been much
LAURIE S . G O O D M A N ties (CMBS), .md residential mortgage- backed lower than 4.0%, let alone 13.4%. In fact, it is
IS a managing diRTtor and securities (RMBS) have lower historical default generally taken for granted that these assets
co-head of Global Fixed rates than equivalently rated corporate debt. default less frequendy than corporate bonds,
Intoinc Research at UBS
The implication ot this is important, not only which have 3.2% and 2.3% five-year default
ill Ncu- York City.
laurie.go(tdman@ubs.t'oni
for investors in these assets, but also tor investors rates according to SikP and Moody's. It is
in structured finance collateralized debt oblig- common, tor example, for SF C D O structurers
F R A N K J. FABOZZI ations (SF CDOs) backed by these assets. to use a base case of 0.36% defaults per year,
is the Frederick Fniiik For example, Cloodm.m and F.ibozzi or 1.80% defaults over five years, in their cash
adjunct professor of flow models of BBIi Resi B&C collateral.
[2002] argue that because the rating agencies
finance at the School of
Management of Yale Uni- ignore the historically low default rate ot struc- Investors care more about/iifKfc default
versity' in New Haven, C~T. tured finance collateral, they are more con- rates than pdst def'ault rates in SF CDOs and
fab«zzi32l@a()l.Ci)m servative in their ratings of SF CDOs than we are constantly reminded of the variability
CDOs backed by corporate bonds, and SF in credit quality of assets with the same rating.
CDOs thus otl^er investors relative value. It the But still. It would be nice if we could be a lot
rating agency reports are accepted at tace value, more certain of the past than is implied by a
it would appear that the Goodinan-Fabozzi range of five-year detault rates ranging from
conclusion lacks empirical support. 1.8% to 13.4%i.
Let s consider one important example tor We explore the discrepancy between the
investors in the CT^O market—BBB tranches default rates calculated by the rating agencies
from B and C credit qtiality first lien residen- and market intuition. We get into the nitty-
tial mortgages (Resi B&C, also known as sub- gritty of bow rating agency statistics are
prime home equities or home equities). SikP constructed betore we come to our own con-
published a statistic recently suggesting that an clusions. We find, to take one example, that a
average of 4.0% of BBB Resi B&C tranches more accurate historical default rate for BBB
are downgraded to D {S&l's default rating cat- Resi B ^ C over tive years is ].9% rather than
egory) over five years (see Hu, Pollsen, and the 4.0% to 13.4%. of the SixP and Moody's
Elengical |2OO3]). Hu, Cantor, Silver, Phillip, reports. Our estimate ot past pertormance is m
and Snailer [2()03| report a statistic suggesting line with the 1.8%, detault rate that SF C D O
that an average of 13.4% of these tranches suffer structurers commonly use to model the future.

44 111 1 Mil 1 R A i r s UN SiRtjrTURi":n FINANCE SECURITIES 2(1114


We discuss six rating agency reports. We first focus EXHIBIT 1
on three S&F reports from January and February 2003 Moody's Material Impairment Rates in ABS
tliat show the rate at which different types ot structured
hnance tranches have been downgraded to D {tive-year Health Care Receivables 40.0%
Franchise Loans 22.8%
rating transitions). In discussing S&P's results, we explain Manufactured Housing 12-1%
the advantages of an alternative matrix multiplying approach Autos 1.4%
where we extrapolate five-year downgrade rates from HEL 1.3%
short-term average rating changes. We then apply this Leases 0.8%
matrix mtiltiplying technique to two other rating transi- Credif Cards 0.3%
Equipment 0.0%
tion studies—a July 2003 S&P rating transition study, and
Fioor Pians 0.0%
a January 2003 Moody's rating transition study—and we Smaii Business Loans 0.0%
examine the Moody's study that presents material impair- Student Loans 0,0%
ment rates tor structured fmance tranches. Other Receivables 0.0%
Other ABS 1.0%
All ABS 2.7%
I. PRELIMINARY CAUTIONS
Soma". Hii. Ciiiior, Silver. Pliillip. dml SiuiUcr j2()0Jj.
How appropriate are historical results in predicting
future credit performance? While rating transition and defaults are driven to a large extent by the idiosyncratic
detault studies are necessarily pictures taken of a rear-view problems of these corporate sponsors. For example, 17%
mirror, the nature ot structured finance makes it hard to of all ABS def^aults are traceable to problems at Con-
get an accurate picture of even the past. In a corporate seco/Greentree. Even more dramatic, 62% ot all RMBS
bond rating transition or default study, the unit of study defaults are traceable to Quality Mortgage. With past
is the corporate entity. But for structured fmance, the defatilts so much a tlinction of individtial corporate prob-
object of study is the specific tranche issued by a specit'ic lems, predicting future default rates is problematical. Are
special-purpose entity. we going to have more or fewer CA)nseco/()reentrees
The sheer number of these tranches, each with its and Quality Mortgages?
own unique credit characteristics, makes the creation of Finally, dcfatilt rates are only half of the credit loss
databases difficult. The fact of a default is often ambiguous; story. Default severity, or loss in the event of a default, is
a missed coupon may occur unseen, and the lapse may be the second part of the credit loss formula. l!)efauk severity
rectified later. It may also be certain /fi'ir, judging from the among structured finance tranches seems to vary by under-
state of the special-purpose entity's collateral portfolio, lying assets and by the seniority and size ot the tranche.
that a tranche will eventually default later m its lite. We do not touch on this important credit factor here.
The corporate entity moreover continues on as debt We thus approach the historical reports of S&P and
is issued and retired. Each structured finance tranche, Moody's with a healthy degree of skepticism about what
however, has a limiteci life, and in a database of structiu^d they can tell us about the future. But maybe they can at
fuiance defaults, withdrawn ratings abound. We shall sec least give us a clearer picture of the rear-view mirror.
that the treatment of withdrawn ratings is an important
consideration in assessing structured finance default studies. II. S&P'S FIVE-YEAR TRANSITION TO D
The heterogeneity of structured finance assets means
that broad categories are made up of assets with disparate In three reports issued in January and February 2003,
performance. For example, in Moody's ABS category, SikV published five-year transitions to D rates, or the his-
health care, tranchise loan, and manufactured housing toric rate at which structured finance tranches have been
securitizations have had the highest default rates, as shown downgraded to P. We take the 1) rates as a close proxy
ui Exhibit 1. When the individual types of deals are aggre- for default.
gated into a broad category we get a distorted picture of We show in Exhibit 2 S&P's results for ABS (Ertuk,
the performance of the whole and of the parts. Coyne, and Elengical |2()03|): CMBS (Hu and C:hun
The defaults ot structured finance tranches are often |2OO3|); and RMBS (Hu, Pollsen, and Elengical 12003]).
directly linked to the originators and servicers of the We also show S&:P's five-year corporate credit default rate
underlying assets. In fact, overall structured finance from Brady, Vazza, and Bos |2OO3j). Boxed in the exhibit

SEI'TLMBLK 21)04 THIIJOUKNAI oi^ Fixci) INCOMT, 45


is the L'litry tor lUJB Resi B ^ C tranches. EXHIBIT 2
Keep in mind tliat Si^P's RMBS category iiicltides S&P Five-Year Transition to D Rates
transactions backed by prime and subprime first and
second residential liens, including home equity lines ot Corporate
Default
credit, home improvement loans, reverse mortgages, and ABS^ CMBS" RMBS'' Rate''
tax liens, so the nimibers in the exhibit merely suggest AAA 0.00% 0.00% 0.00% 0.10%
the actual downgrade rate of any specitic type of RMBS. AA 0.20% 1.93% 0.80% 0.27%
The same point holds tor the ditterent types of securiti- A 1.47% 1.97% 2.40% 0.62%
BBB 1.23% 0.93% 4.00% 3.20%
zations in the ABS and CMBS categories.
BB 4.62% 5.55% 9.90% 12.34%
Exhibit 2 contradicts two beliefs widely held by B 0.00% 13.68% 13.70% 26.59%
structured finance market participants. The first, as we
noted at the outset, is that detault rates tor structin"ed 'lliluk. Coyne, ami l-lci!i;iui! I2
hnance tiMns.tctions are lower than for corporate debt. '•Ilumii! CJuiu 1200}f.
The second is that CMBS and RMBS have lower default Ihi. l\'ll<n,. ,ui,! liU-Hiiical 120031.
r.ites than ABS. 'Bnuly, [;;;;.!, ami Bo^ 1200}j.
We see that many of the investment-grade structured
finance transitions to I) rates .ire higher than the corporate rating at a particular time) that cm enter the tive-year tran-
default rates for credits of the same rating. At the extreme, sition calculation is the one running from January I. I9'J8,
the AA CMBS rate is seven times the corporate AA rate. to January 1, 2(>()3. This means that any structured tinance
And the BBB RMBS, is 4.0% versus the corporate 3.2% securitization rated after January 1, 1998, is not part ot
rate. We also see that tor many rating categories, ABS tran- the five-year rating transition calculation.
sition rates ,ire lower than those for C^MBS and RMBS.
S&P's methodology greatly reduces the number ot
We believe that both of these ettects are a result ot structured tmance transactions incorporated in the tive-
Si^'P's transition study niethodt>logy rather than a reflec- year transition calculation. For example, there are 282
tion ot true credit experience. BBB RMBS with five or more years of rating history
included in the tive-year transition rates, but there are 530
S&P's Transition to D Methodology BBB RMBS with tour or tewer years of rating history
that are cxdiidcd from the five-year transition rates. Fur-
S&P calculates tive-year transition to 1) rates in a thermore, it appears that a disproportionate number ot
way that excludes recent rating experience. This can some- pre-1998 RMBS transitioned to D.
times lead to strange results, such as when the transition We demonstrate the tact of declining transition rates
to 1) rate is higher over a shorter time interval th.m it is by comparing the 1978-2002 RMBS average one-year
over a longer time interval. For example, alnu)st all ABS transititm matrix with the earliest available RMBS transi-
transitions to 1.^ rates over ilircc years are higher than ABS tion matrix SikV published, the 1978-2000 average one-
transition rates over /Jre years. year transition matrix. In Exhibit 3 we show the 1978-2000
S&Ps transition methodology requires that, to enter transition matrix iiii)iiis the 1978-2002 transition matrix.
the five-year transition matrix, the rating must be tive years The positive numbers tor BBB transitions to BB, B, CCC!!,
old. So the l.ist rdtiiii^ coliort {group of credits with the same

EXHIBIT 3
S&P Average 1978-2000 RMBS Transition Matrix Minus 1978-2002 Transition Matrix
Rating at End of One Year
ing

AAA AA A BBB BB B ccc cc c D


c AAA -0.04% 0.03% 0.00% 0.01 % 0.00% 0.00% 0.00% 0.00% 0.00% 0.01%
en != AA -2.84% 2.48% 0.21 % 0.04% -0.01% -0.03% 0.05% 0.00% 0.00% 0.00%
01 (0
A -0.82% -2.73% 3.07% 0.25% 0.01% 0.03% 0.04% 0.00% 0.00% 0.03%
BBB -0.08% -1.59% -0.84% 1.33% 0.35% 0.62% 0.10% 0.00% 0.00% 0.01%
tin

BB 0.00% 0.00% -1.13% -4.13% 4.36% 0.21% 0.40% 0.13% 0.00% 0.06%
EC B 0.00% 0.00% 0.00% 0.04% -2.30% 3.86% -1.81% 0.23% 0.00% -0.12%

46 RATES ON STULHIURI'D TiNANcr 211114


and I) mean that there were more BBB downgrades in Multiplying Transition Matrices
1978-20(11) tlian in 197S-2OO2. In other words, BBB down-
grades declined in 2(H)1 and 2002. We attempt to sidestep the methodological issues
This is consistent with our hypothesis that the small raised here by multiplyiini trcimitioti iiiatria's (which admit-
number of BBB credits before 1998 experienced more tedly stirs up its own methodological issues). Our pre-
downgrades than later BBB RMBS, and bias the five-year ferred way tt) calculate the five-year transition to D rates
transition to D statistic upward. The same result, declining from S&P data is to "multiply" transition matrices. By this
transition rates in recent ye.irs, holds true for the other we tiieati, for example, to look at a six-month average
RMBS rating categories too. transition matrix and see where BBB RMBS transition to
Another aspect ofS&l^'s methodology that affects its after six months. Most ratings will remain the same, but
five-year results is the treatment of ii'ithdnni'ii ratings. In some percentage of them will be upgraded or downgraded.
calculating five-year rating transitions, S&;I* excludes struc- Then, we put the vector of ratings and percentages back
tured finance credits whose ratings were withdrawn some- into the transition matrix to see where the originally rated
time over the five-year period (unless the credit was rated BBB RMBS migrate in one year. We continue this process
I) before its rating was withdrawn). An example shows until we have transitioned BBB RMBS ten times to arrive
the implication of this approach. at cumulative five-year transitions.
SLippose there are HH) BBB ratings at the beginning The advantage of this approach is that we use all
of a five-year period, and ISO of the ratings are withdrau'ii available data because the six-month transition matrix
sometime over the period. Further, suppose that one BBB incorporates data from recent periods, even January 1,
credit was downgraded to O, The five-year transiti(.)n to 2003. tojuly 1, 2003. It also prevents average transition
D rate would be 1 divided by 50, or 2%, because the 50 to I) rates from being higher over shorter periods than the\'
withdrawn ratings are excluded from the calculation of are over longer periods.
transition rates. The alternative treatment would be to Another advantage of this approach is that it uses
tre.it withdrawn ratings as stable ratings and calculate the S&P"s latest structured finance rating transition study
transition to I) rate as 1 over 100 or 1%. (Ertuk, Elengical, and Gillis |2003]). This study makes
I'liere are arguments for both approaches. Treating some improvements in S&Ps rating database and method-
a withdrawn rating as a stable rating makes sense in that ology in creating average six-month rating transition
.1 withdrawal is usually not a negative credit event. In fact, matrices. Eirst, structured finance credits are more care-
it is an unamhiguoLisly good thing that a credit pays off fully categorized by type of asset and domicile. Second,
its debt and has its rating witlidr.nvn. On the other hand, for ratings that are withdrawn at the end of the six-month
if a credit is not outstanding, it does not have an oppor- perie)d, but that transitioned to some intermediate rating
tunity to default. Why keep congratulating a one-year during the six-month period, the last rating before with-
bond for not defiUilting four years after its maturity? drawn is taken as tbe ending rating. Thus, if a credit started
One thing is certain. Compared to including with- tlie six-month period as a BBB, transitioned to BB over
drawn tranches in its calculations, SikVs exclusion of with- the six months, but was witbdrawn by the end of six
drawn ratings iiicrciiscs the percentage of credits deemed nuMiths. this wxitild be taken as a transition to BB in S&P's
downgraded to D. And eventually, over a long enough stiidy.-
period, all credits in a cohort will either mature, thereby
having then- ratings withdrawn (and be excltided from Results of Multiplying Transition Matrices
the cohort), or defatilt. When this happens, the transi-
tion to 1) rates must be 100%. As we noted earlier, with- In Exhibit 4, wv shtjw tlie results of multiplying the
drawn ratings are a bigger problem in studies of structured average six-month matrices and compare those results to
fmance defaults where the short maturities of specific S&P's previous calculated five-year transitions and to ftve-
issues, rather than the ongoing existence of a corporate year corporate defaults.
entity, are the object of study. Note that for RMBS and (^MBS. multiplying the
But the SikV methodology used in the three studies six-month matrix ten times almost always gives much
published in 2003 is not tbe only way to calculate transi- lower five-year transition to I) rates than S&P's five-year
tion statistics, and is probably not the best way. It is also not transition matrix number. F-or example, the multiplying
the way S&'P and Moody's calculate their default statistics.' method vields a BBB RMBS transition rate of 1.62%

Si I ' l l MHI.K 2i)U4 7'Hr JouRNAi or FIXKD INCOMH 47


EXHIBIT 4
SȤcP Five-Year Transition to D Rates
Corporate
ABS CMBS RMBS Default
Multiply 5-Year Multiply 5-Year Multiply 5-Year Rate
AAA 0.01% 0.00% 0.00% 0.00% 0.00% 0.00% 0.10%
AA 1.16% 0.20% 0.00% 1.93% 0.02% 0.80% 0.27%
A 2.09% 1.47% 0.08% 1.97% 0.47% 2.40% 0.62%
BBB 13.80% 1.23% 0.65% 0.93% 1.62% 4.00% 3.20%
BB 42.56% 4.62% 4.36% 5.55% 4.50% 9.90% 12.34%
B 82.53% 0.00% 12.28% 13.68% 14.14% 13.70% 26.59%

EXHIBIT 5
S&P ABS Defaults
1998 1999 2000 2001 2002 Total
Auto 1 - - - 1 2
Credit Card .- - 2 2 4
Franchise Loan - - - 9 22 31
Manufactured Housing - - - 1 32 33
Other - 9 - - 1 10
Total 1 9 0 12 58 80
Soiinv: Brink, j20()Jj.

while SikP\ five-year matrix shows 4.00%. Tbis, inci- finance transitioTi rates are similar, but usually a little lower
dentally, is consistent with the SF C D O structurer assump- than for the U.S. ABS. It would seem that S&P faced
tions of a 1.8% default rate over five years for BBB Resi more surprises from strange U.S. ABS asset categories
B&C. We also see that CMBS gain what we thmk is their than from international ABS asset categories. On average,
rightful place at the top of structured finance credit quality U.S. C O O transitions are worse than U.S. CMI5S and
with the lowest transition rates. RMBS. but not as bad as U.S. ABS. European C D O tran-
Yet for ABS, the effect is the opposite, as transition sitions win the booby prize, being worse than U.S. ABS.
rates calculated by the multiplying method are higher than Also interesting are the international categories we
the five-year rate. This is because ABS credit performance could not calculate transition to D rates for: European
since 199S has been much poorer than prior to I99H. C'MBS, European RMBS. Asian structured finance, and
Tbis recent poorer credit performance is captured only by Australia/New Zealand strnctured finance. These cate-
the multiplying method. As we mentioned before, the gories have never had a transition to D. Of course, there
poor performance of ABS in recent years has mainly come are not as many tranches making up their statistics, par-
from new and untested asset classes. ticularly low-rated tranches, as in the U.S. structured
Exhibit 5 shows S&Ps classification of recent ABS finance categories.
defiuilts.
S&P Study Conclusion
Five-Year Transitions of International
Structured Finance and CDOs We think our method ot multiplying six-month
transition matrices from S&P's report by Ertuk, Eleng-
S&P has also carefully calculated six-month transi- ical, and Gillis |2(K)3| is better at assessing tlie long-term
tion rates tor CDOs and structured finance transactions credit quality of structured finance tranches than the five-
backed by assets originated outside the United States. year transition matrices in S&P's three studies published
Exhibit 6 shows transition to D rates tor the geographies in January and February 2003. The multiplication method
and structured fmance categories that could be calculated. includes data h-om all years in arriving at long-term results
European ABS and emerging market structured and eliminates the problem that short-term transition rates

48 R A I L S t'N SiRUciiURtn FINANC [. Stc SF.I'TEMlirR 2004


may be higher than loiig-rerm tiMiisition rates. It also EXHIBIT 6
makes use of S&.P's improved d.itab.ise and methods. S&P Other Five-Year Transition to D Rates
S&P's five-year RMBS transition rates are bigb
because S&P did not rate very many RMBS below AA Euro US Euro Corporate
ABS EMSF CDO CDO Default
until the last five years, and older RMBS have performed Multiply Multiply Multiply Multiply Rate
more poorly than more rccc-ntly issued RMBS. The mul- AAA 0,00% 0.00% 0,07% 0,54% 0,10%
tiplication methodolog\' leads to lower transition results AA 0,20% 0.00% 0,33% 1,32% 0,27%
for RMBS and CMBS that are in hne with our under- A 1,30% 1.26% 1,56% 4,02% 0,62%
standing of these products. It also calculates higher ABS BBB 7,22% 12.02% 2,20% 16,04% 3,20%
BB 39.35% 55.88% 4,41 % 41,18% 12,34%
transition rates by weighing in the recent poor perfor-
B 80.69% 80.03% 9.26% 90.24% 26.59%
mance of those tranches.
i from l-r G"i7fo IJ

III. MOODY'S RATING TRANSITION STUDY EXHIBIT 7


Moody's Five-Year Transition to Ca and C
Moody's structured finance transition study calcu-
lates average one-year transition matrices by modified rating Corporate
categories, adding the 1 s. 2s, and 3s to the letter rating des- Default
ABS CMBS RMBS CDOs Rate
ignation (see Hu and Cantor [2(103]). Moody's does not Aaa 0.01% 0,17%
0,00% 0,00% 0,30%
calculate rating transitions over multiple years. Aal 0,27% 0,00% 0,01% 1,87% 0,17%
We follow the same multiplying technique we used Aa2 0,49% 0,09% 0,03% 1,92% 0,33%
on the S&P average transition matrices to create Exhibit Aa3 2,75% 0,01% 0.09% 4.99% 0.29%
7. It shows the rate at which Moody's structured finance Al 0,80% 0,02% 0,49% 6.00% 0,47%
A2 0.40% 0,01% 0,12% 2,95% 0,68%
ratings have migrated to Ca and C^ over five years. Bear
A3 1.88% 0,01% 1,12% 7,89% 0,62%
in mind that Moody s does not have a I) rating category, Baa1 6.51% 0,56% 2,41% 9,20% 1,80%
and Ca and C rating,s usually indicate default. Boxed in Baa2 5.47% 0,77% 1,31% 19,70% 2,24%
Exhibit 7 are the Moody's categories that include Baa Baa3 9.65% 0,07% 3,47% 22,09% 4.23%
Resi BikC tranches. Ba1 24,74% 0,62% 4,95% 40,27% 7.61%
Ba2 28,60% 0.06% 4.57% 47.49% 9.42%
The tmcness of the modified rating categories leads 47,58%
Ba3 0,60% 10.46% 38.24% 20,70%
to the unexpected result that in some cases higher ratings B1 53,28% 0,83% 4,85% 83,87% 27,56%
have transitioned to Ca and C more often than lower rat- B2 47.22% 1,07% 11,57% 56,19% 34,49%
ings have. But it one regroups ratings into letter rating B3 83.90% 3,23% 22,56% 82,09% 44,40%
categories, we can compare them with the S&'P multiplied Ciihihued tro)ii Hu and Cantor 1200^1.
fivc-ycar traii'^ition to D matrices, CMBS and RMBS
rating transitions, espc-cially in the mvt'Stment grades, are hve-year Baa ABS transition rates range from .S,47% to
roughly similar. For CDOs, Moody's transitions are a lot 9.65% and average 7.21%. This is four times the 1.62% rate
more frequent, which we attribute to the significantly we calculated for S&P's BBB RMBS transitions.
L^reater C D O market share Moody's had over S&P for But as we saw in Exhibit 1, Moody's reports that
much of the study period, especially of C D O tranches home equity loan defaults (including Resi B&'C and other
rated below Aa. home equities) are a little less than half of those of A_BS
For ABS, especially tranches rated helow Aa, Moody's in total. Using this percentage (arbitrarily, because we
transitions are much less frequent than S&P's multiplied don't really know how to apply it to specific ratings), we
transition results in Exhibit 4. We think this is partly arrive at average five-year Baa Resi B&C transition rates
because of a different mix of transactions designated as ranging from 2.63% to 4.65% and averaging 3.47%. We
ABS by tiie two rating agencies. For example, Moody's show these calculations in the last three rows of Exhibit
classifies Resi B&C and other home equities as ABS, while 8. These Moody s Resi B&C] transition rates are twice as
S&P classihes them as RMBS. Unfortunately, this does not high as the 1.62% S&P BBB RMBS transition rate, but
provide much clarity as we try to figure out the histor- lower and we teel much more accurate than Moody's all-
ical default rate of Resi B&C. ABS Baa transition rates.
As we show in the first two rows of Exhibit 8, Moody's

Sl^'TTMEtFR 2004 OF FlXFll INPOML 49


EXHIBIT 8 expect that the treatment of withdrawn ratings makes a
Moody's Baa Resi B&C Five-Year Transitions to D bigger diflerence in the cumulative rating transitions like
tln>se ot S&P sliown in Exhibit 2. This is because over
Baa1 Baa2 Baa3
time, as cumulative cohorts age, withdrawn ratings make
ABS Transitions 6,51% 5,47% 9.65%
Averaqe ABS Transition 7,21%
up an increasingly greater proportion.
Relative Default Frequency
48,15%
of Resi B&C versus All ABS Moody's Material Impairment
Resi B&C Transitions 3.13% 2.63% 4,65% and Downgrade Study
Average Resi B&C 3.47%
Hu, Cantor, Silver. Phillip, and Snailer [2003]
cd fivin daiii in l:xhihiti t iiiitl 7.
examine "material impairments" of structured finance
securities. Moody's defhies a material impairment as a
EXHIBIT 9
payment default that has gone uncured or a ciowngrading
Moody's Five-Year Transition to Ca and C to the Ca or C rating categories. A rating of Ca or C in
Eliminating Withdrawn Ratings the absence of a payment det^ault may indicate that the
1

Corporate structured finance tranche is still paying its coupon, but


Default the condition of the underlyiiitf collateral augurs an
ABS CMBS RMBS CDOs Rate
almost-certain eventual default on interest or principal.
Aaa 0,02% 0.00% 0,00% 0,35% 0,17%
Aa1 0,32% 0.00% 0,01% 2,14% 0,17%
Alternatively, the presence of an uncured payment default
Aa2 0,53% 0,12% 0,03% 2,14% 0,33% in the absence of a Ca or C' rating may indicate that the
Aa3 2,93% 0,01% 0,09% 5.38% 0,29% payment default is slight or expected to be cured.
Al 0,93% 0,02% 0,52% 7.01% 0,47% About half of all material impairments so defined are
A2 0,47% 0,01% 0,13% 3,45% 0,68%
payment defaults of tranches rated above Ca or C".
A3 2,19% 0,02% 1,16% 8,48% 0.62%
Baai 7.15% 0,75% 2.47% 10,58% 1.80%
Moody's has not taken a public stand that the payment
Baa2 5.92% 0,79% 1.36% 20,95% 2,24% default IS severe or is going to continue. In fact, a high
Baa3 10.83% 0,08% 3.59% 23,38% 4,23% percentage of structured finance payment defaults later
Ba1 26.02% 0.78% 5,14% 41,67% 7,61% become cured, maybe about 20% to 3O'X). Thus, we view
Ba2 30,51% 0.08% 4,72% 48,78% 9,42% Moody's material impairment category as an expansive
Ba3 52,90% 0.72% 10,80% 39.76% 20,70%
definition of default.
B1 57,60% 1,06% 5,04% 85.25% 27,56%
B2 49,66% 1,39% 12,14% 58,00% 34,49% Another step in Moody's methodology" is to deduct
B3
UO 86,32%
OU,Oil /o 3,57%
0.01 lO 23,54%
jCJ,iJt 84,37% 44,40% halt of all withdrawn ratings in the calculation of default
Calailalcd from Hii iiiid (.'.iiiilor j200.ij. rates. So, if two ratings in a cohort ot 100 ratings were
withdrawn over the year, those defaulting over the year
would be compared to a denominator of 99 rather than
Effect of Withdrawn Ratings 100 or 98, But if next year two more ratings are with-
drawn, defaults are compared to a denominator of 97,
The Moodys data also allow us to explore the meth-
This splits the difference between counting withdrawn
odological question mvolving the treatment o{ with drawn
ratings as non-defaults and eliminatinu; them completely
ratings. In arriving M the downgrade rates m Exhibit 7.
from the defluilt statistics and tliereby biasing the default
we multiplied Moody's one-year transition matrix with
statistics. But again, one has to e.xpect that many struc-
withdrawn ratings treated as stable ratings. In Exhibit 9,
tured finance tranches mature every year and have with-
we sliow the result ot rcnuiving withdrawn ratings.
drawn ratings. To the extent this is so. this treatment will
As we expected, eliminating \\ ithdrawn ratings raises exaggerate the calculated structured finance def'aults. and
the transition rates, particularly tor ABS and C^DOs rat- the efTect is componnded as tranches season.
ings and particularly for speculative-grade ratings, but the
Exhibit 10 shows Moody's material impairment rates
difference is not great. We think the multiplying method
for structured finance tranches along with the agency's
eliminates a lot of the difference between the two with-
calculation of five-year corporate defiiult rates. Moody's
drawn rating methodologies, because over six months or
BBB Resi B&C are categorized as ABS (the hoxed
one year there are not very many withdrawn ratings. We
number).

50 DiTAUI.T R A I K S O N STRUrTURRI) FiNANCF. SFCURITIFS 2(H14


We see in Exhibit 10 tli.it CMliS inip.urnieiits are EXHIBIT 10
about ill line witii corporates. but that ABS and l^MHS Moody's Five-Year Material Impairment Rates—
arc liigher than corporates. Moody's structured finance Rolling Cohorts
results are generally a lot higher than those we achieved
from multiplying S&Fs six-month transitit^i matrix. P:irt Corporate
ABS CMBS RMBS Default Rate
of the relatively high structured fuiance default rate is 0.97% 0.12%
Aaa 0.06% 0.00%
attributable to the expansive definition ot tnaterial impair- Aa 3.13% 0.00% 0.61 % 0.26%
ments versus default and the elimination of withdrawn A 1.61 % 1.03% 1.42% 0,51%
ratings. Baa 13.44% I 2.23% 7.25% 2.25%
Ba 53.08% 5.17% 9,65% 11.36%
B 49.38% 19.43% 17.17% 32.31%
Rolling Cohort versus Original Issue Cohorts
Soiinr: Hu, Cuiiior, Sihvr, Phillip, and Siiaiter 120031.

Moody"s also provides a second set of five-year


impairment rates, calctilated in a slightly different manner, tranches tend to increase over time after issuance. Exhibit
that sheds further light on these default rates. Instead of 12 illustrates tbe pattern of marginal defaults yc-ar-by-year
the rolling cohort methodology, where a BBB Resi B&C- after initial issuance.
tranche issued on |aiiLiary 1. 2000, its part of tiie 2000, It shows that marginal detanks increase until three
2001, and 2002 cohorts, Moodys alternative original issue years after original issuance and then decline. This pat-
cohort methodology forms cohorts only from imi'ly Issued tern of defaults insures that the rolling cohort method
tranches. The tranche issued on January 1, 2000 would will produce higher default rates than the original issue
be part of only that one single cohort, and would coLint cohort method. This is because under the rolling coliort
only as a three-year default at its original issue rating. method, defaults in later years in tbe life of a tranche are
Structured finance defaults calculated by both weighted into the default rate of earlier years. Under these
riK'thodologies differ greatly, as shown in Exhibit I I. With circumstances, we feel the original cohort method pro-
the same exact data, detank rates for structured finance vides a better estimate of future structured finance default
tranches are about twice as high under tbe rolling cohort rates.'
niL'tbod as under tbe original issue metbod. In Exhibit 13. we show Moodys material impair-
Moody's points out that the reason for the different ment rates tor structured hnance tranches, using the orig-
restilts is that marginal defaults among structured finance inal cohort methodology, along with the rating agency's

EXHIBIT 11
All Structured Finance Cumulative Impairment Rates
5%

• Original Issue Method


B Rolling Cohort Method

Years

2(1114 TuiiJOURNA! Oh FlXFH KcdMh 51


EXHIBIT 12
Marginal Material Impairments
0.8%

0.6%

0.4%

0.2%

0.0%

Years Since Origination

EXHIBIT 13
those we achieve by multiplying S&Ps six-month tran-
Moody's Five-Year Material Impairment Rates-
Original Issue Cohorts sition matri.x shows Moody's defiult results are still gen-
erally higher. Again, part ot the relatively high structured
Corporate fmance deflmlt rate is attributable to the expansive defi-
Default
ABS CMBS RMBS Rate nition ot material impairment versus default, and part is
Aaa 0.05% 0,00% 0,58% 0,12% due to the partial elimination of withdrawn ratings.
Aa 2.83% 0,00% 1,01% 0.26% Using the ratio of HEL defaults to total ABS defaults
A 1,10% 0,23% 0,63% 0.51 % from Exhibit 1 {again blindly, since we still don't know
Baa 1 4,38% 0,83% 6,00% 2,25%
how to apply it to specific ratings), we arrive at an average
Ba 19,70% 1,77% 4,98% 11,36%
B 39,53% 5,66% 13,66% 32,31%
five-year Baa Resi B&:C. transition rate of 2.11%. We show
these calculations in Exhibit 14. This Moody's Resi B&C
Soitra-: Hii. Cmnor, Silver, Phillip, tiiid Siiailcr [2
transition rate is higher than the 1.62% S&P BBB RMBS
EXHIBIT 14 transition rate, but it is lower, and we feel much more
Moody's Baa Resi B&C Five-Year Material Impairments accurate, than Moody's all-ABS Baa default rate.

Baa
All ABS Material IV. CONCLUSION
Impairments 4,38%
Relative Frequency of Resi
In determining default rates for structured finance
B&C and All ABS Material 48.15% tranches, we think the most reliable sources are the S&P
Impairments multiplied five-year transition to ][) rates presented in
Resi B&C Transitions 2.11% Exhibit 4 and the Moody s original issue material impair-
Bxhihits I tvid
ment rates presented in Exhibit 13. We thus average the
two to produce the five-year default rates shown in Exhibit
calculation of five-year corporate default rates. Boxed in 15, We also take Moody's ABS rate and multiply it by
the exhibit is the Moody's category that includes BBB 0.4K to arrive specifically at Resi B&C default rates.
Resi B&C. The origin:il cohort methodology produces Obviously, there is a lot ot Kentucky windage in
lower material impairment rates than the rolling cohort our historical default estimates. Yet historical results, what-
methodology; especially for ABS and especially for the ever their exact number, are pictures of a rear-view mirror.
Baa category including Resi B&C. CMBS defaults are And in this case, they reflect the difficulty of conducting
now lower than corporate defaults, hut ABS and RMBS an investigation of structured finance det'aults and the
defaults are still higher than corporates. almost random etYect of corporate credit events on struc-
Comparison of these original issue results against tured finance.

52 DEFAUIT RAT^:s ON STRUCTURFD FIIMANCL SECUKITIRS SET'TEMHER 2004


EXHIBIT 15 REFERENCES
Estimated Historical Five-Year Default Rates for
Brady, Brook, Diane Vazza, and Roger J. Bos. "Corporate Detanks
Structured Finance Tranches and Corporate Bonds
Peak in 20(12 Amid Record Amounts of Detaults and DecliniTig
AB5 CMBS RMBS Rest B&C Corporates Credit Quality." l^atiiii><: Perforiiuvicc 2002: DeJauU Transition.
AAA 0.0% 0.0% 0.3% 0.0% 0.1% Recovery, and Spreads. Standard ik Poor's. February 2003.
AA 2.0% 0,0% 0.5% 0.7% 0,3%
A 1.6% 0.2% 0.6% 0.5% 0.6% Ertuk, Erkan, Patrick Coyne, and Jay Elengical. "Ratings Tran-
BBB 9.1% 0.7% 3.8% 1.9% 2.7% sitions 2002: U.S. ABS Weather a Turbulent Year." Standard &
BB 31.1% 3.1% 4.7% 7.0% 11.9% Poor's, January 31,2003.
B 61.0% 9.0% 13.9% 16.6% 29.5%
Ertuk, Erkan, Jay Elengical, and Thomas G. Gillis. "Rating
Transitions in Eirst-Half 2003: Global Structured Securities
ENDNOTES
Semiannual Review. " Standard Ik Poor's. July 29, 2003.
This article draws on material to be published in Lucas.
Goodman, Laurie S,. and Frank J. Eabozzi. CoUaterahrrcd Debt
Goodman, and Fabozzi, CoUatcraHzcd Debt C)bl{i;iitioiis: Swicfurcs
Obli_^alions: Stniaiircs and Analysis. Hoboken, NJ: John Wiley
and Analysis, 2nd ed. (Hobokcii. NJ:John Wiley & Sons. 2005).
& Sons, 2002,
'In short, marginal default rates (defaults within the tlrst
year, within the second year, and so on) are calculated tor each
rating cohort. Yearly marginal default rates are then averaged Hii, Jian, and Richard Cantor. "Structured Einance Rating
across all rating cohorts with the requisite iiunihcr of years oi Transitions: 19S3-2OO2: CoinparisoTis with Corporate l^atings
history. The one-year cuniularivc default rate is merely the average and Across Sectors." Moody's Investors Service, January 2003.
of every rating cohort's tlrst-year marginal default rate. Buc the
two-year cumulative default rate is the sum of the one-year cumu- Hu, Jian, Richard Cantor. Andrew Silver. Tad Phillip, and Joe
lative default rate and the average sccoiul-ycar marginal detault Snailer, "Payment Defaults and Material Impairments of U.S.
rate. The three-year cumulative dehiult rate is the sum of die Structured Einance Securities: 1993-2002," Moody's Investors
two-year cumulative detault rate and the aveniiie f/dnV-year mar- Service. December 2003.
ginal delault rate. And so it goes, adding average iiuiri;iiial default
rates to previously calculated ttiiiiuLilii'c default rates. I hi, losepb, and Roy Cbun. "Rating Transitions 2002: Respect-
able Credit Performance of U,S. CMBS." Standard & Poors,
This approach uses as much ot the availahle data as pos-
January 16, 2003.
sible. Last year's rating cohort, with only one year of data, con-
tributes something to the ten-year average cumulative default rate.
Hu,Joseph, Robert B, Pollsen, and Jay Elengical. "RatingTran-
The other suboptimal way to calculate a five-year default
sitions 2002: A Quarter Century of Outstanding Credit Per-
rate would be to average the cumulative default rates of every
ct)hort with five years of history. (This is what S&:V did with torniance of U.S. RMBS." Standard & Poor's, February 6, 2003.
structured finance rating transitions in its three studies pub-
lished in 2(103.) In this case, the last four annual cohorts would
To order reprints oj this artidc, plduc Lontiict Ajcuii MaHk at
contribute nothing to the average five-year cumulative detault
statistic. The marginal method also makes sure that cumulative imiaiik@iijomniili.coni or 212-224-3205.
default rates never decline over time.
-S&.P and Moody's point out that the multiplication
method implicitly assumes that rating downgrades are not seri-
ally correlated, i.e., that a tranche that has been downgraded is
not more likely to be downgraded tigaiii. relative to other
tranches that have not been downgraded. We know tbat down-
gracies ot corporate bonds are serially correlated, but we are
not sure bow serially correlated downgrades of structured
finance trancbes would atTect the overall results of tbe multi-
plication metbod.
^The original cohort method stilJ takes advantage of recent
detauit history troni tranches issued within the last tive years.
These detaults go into the calculation of average marginal def.uilr
in the particular year after issuance.

Si.l'U.Mltt-R 2(11)4 THE JOURNAL OI- FIXED 53