No. 2008/15
Risk Transfer with CDOs
Jan Pieter Krahnen and Christian Wilde
Center for Financial Studies
The Center for Financial Studies is a nonprofit research organization, supported by an
association of more than 120 banks, insurance companies, industrial corporations and
public institutions. Established in 1968 and closely affiliated with the University of
Frankfurt, it provides a strong link between the financial community and academia.
The CFS Working Paper Series presents the result of scientific research on selected
topics in the field of money, banking and finance. The authors were either participants
in the Center´s Research Fellow Program or members of one of the Center´s Research
Projects.
If you would like to know more about the Center for Financial Studies, please let us
know of your interest.
Prof. Dr. Jan Pieter Krahnen
Prof. Volker Wieland, Ph.D.
CFS Working Paper No. 2008/15
Risk Transfer with CDOs*
Jan Pieter Krahnen1 and Christian Wilde2
April 28, 2008
Abstract:
Modern bank management comprises both classical lending business and transfer of asset risk
to capital markets through securitization. Sound knowledge of the risks involved in
securitization transactions is a prerequisite for solid risk management. This paper aims to
resolve a part of the opaqueness surrounding credit-risk allocation to tranches that represent
claims of different seniority on a reference portfolio. In particular, this paper analyzes the
allocation of credit risk to different tranches of a CDO transaction when the underlying asset
returns are driven by a common macro factor and an idiosyncratic component. Junior and
senior tranches are found to be nearly orthogonal, motivating a search for the whereabout of
systematic risk in CDO transactions. We propose a metric for capturing the allocation of
systematic risk to tranches. First, in contrast to a widely-held claim, we show that (extreme)
tail risk in standard CDO transactions is held by all tranches. While junior tranches take on all
types of systematic risk, senior tranches take on almost no non-tail risk. This is in stark
contrast to an untranched bond portfolio of the same rating quality, which on average suffers
substantial losses for all realizations of the macro factor. Second, given tranching, a shock to
the risk of the underlying asset portfolio (e.g. a rise in asset correlation or in mean portfolio
loss) has the strongest impact, in relative terms, on the exposure of senior tranche CDOinvestors. Our findings can be used to explain major stylized facts observed in credit markets.
JEL Classification: G21, G28
Keywords: Credit Risk, Risk Transfer, Systematic Risk
* We are grateful for Financial support by Deutsche Forschungsgemeinschaft (DFG) and by Frankfurt University's Center for Financial
Studies (CFS). This paper is part of CFS' project on the Economics of Credit Risk Transfer. In developing the basic question of this paper,
we owe a lot to mini workshops with Günter Franke as well as Dennis Hänsel and Thomas Weber. Furthermore, we thank Falko Fecht,
Ian Marsh, Loriana Pelizzon, Isabel Schnabel, and two anonymous referees for helpful comments. We also thank seminar participants at
the 2005 Meeting of the German Finance Association in Augsburg, the 2005 Annual meeting of the Verein für Socialpolitik in Bonn, the
2005 Workshop on Credit Risk Management in Obergurgl, Max Planck Institut in Bonn, WHU Koblenz, the 30th Anniversary of the
Journal of Banking and Finance Conference in Beijing, the 2006 annual conference of CERF in Cambridge, the 2006 VHB conference in
Dresden, the 2006 BIS Workshop on Risk Management and Regulation in Banking in Basel, and the 2007 EFA conference in Ljubljana.
1 Goethe University Frankfurt and Center for Financial Studies, Correspondence: Goethe University Frankfurt, Finance Department,
Mertonstrasse 17-21 (PF 88), 60054 Frankfurt am Main, Germany, E-mail: krahnen@finance.uni-frankfurt.de, Phone: +49 69 798 22568,
Fax: +49 69 798 28951.
2 Goethe University Frankfurt, Finance Department, Mertonstrasse 17-21 (PF 88), 60054 Frankfurt am Main, Germany,
E-mail: wilde@finance.uni-frankfurt.de, Phone: +49 69 798 28381, Fax: +49 69 798 28951.
1
Introduction
Securitization of loan assets has become a common instrument of bank risk management in
recent years, prior to the credit crisis. According to a survey published by the European
Central Bank in 2004 (ECB 2004), about 8-15% of overall assets of large, international banks
were subject to a securitization transaction. Many observers, such as J.P.Morgan (2004),
believed the market for asset backed securities (ABS) to grow rapidly in the following couple
of years.
The securitization crisis started in July 2007 in the market for subprime mortgages. It has
drastically altered the growth prospects of structured finance markets in general, and of CDO
markets (Collateralized Debt Obligations) in particular. The dramatic decrease in liquidity
on secondary markets for tranche notes has aroused a large public interest in risk properties
of these instruments, where assessments vary greatly.
Most related papers treat the issue of risk allocation to tranches in the context of pricing
studies. Coval et al. (2007), for example, find that investors in senior tranche notes are greatly
underpaid for the (high) level of systematic risk inherent in these securities. They also consider
junior tranches to be exposed primarily to diversifiable risk which, according to the authors,
renders the common characterization of equity tranches as ”toxic waste” obsolete. Duffie
(2007), on the other hand, claims that junior-tranche prices are vulnerable to macroeconomic
performance, and Eckner (2007) finds the compensation per unit of expected loss for senior
tranche investors to be much higher than that of junior tranche investors.
Also, on the side of policy makers, views on the whereabout of systematic risk diverge
widely. Alan Greenspan, one of the forceful proponents of securitization, in a now-famous
quote, said that credit derivatives and other complex financial instruments have contributed
”to the development of a far more flexible, efficient, and hence resilient financial system than
existed just a quarter-century ago.” (Greenspan 2004). The main argument was that risk
transfer through credit securitization allows banks to hold less asset risk, and to be more
diversified. As Duffie (2007) points out, a more efficient distribution of risk in the economy
will also lead to a lowering of the cost of capital, and to further macroeconomic benefits.
However, there is practically no data available on the actual allocation of tranche notes,
and their respective risk profiles, among issuers and investors. Even supervisors apparently
have only limited reliable information as to which risk is being held inside or outside the
financial system (ECB 2007). On top of not knowing where tranches are being held, there is
even no consensus as to the content of systematic risk in the different layers of securitization
transactions, the junior and the senior tranche notes.
Our paper contributes to this debate by showing how, through the pooling and tranching
of underlying financial assets, risks are allocated to a set of junior, mezzanine, and senior
tranche notes, and how changes in risk will affect the risk properties of tranches.
The tranching process makes a measurement of the systematic component particularly
difficult. We use a one-factor model with a macro factor driving the performance of the
2
underlying loan portfolio and consequently also the associated tranches. This allows us to trace
the allocation of systematic risk to tranches. We propose a measure to estimate macro-factor
dependency of the individual tranches. In particular, total portfolio losses are decomposed
in a multidimensional way to states and tranches, where states are defined as the macro
factor taking values in certain ranges. Thus, this metric yields state- and tranche-dependent
loss statistics obtained from Monte Carlo simulations. The risk properties of tranches with
different seniority will be compared to those of straight bonds, which is the traditional, nonstructured finance, funding mode. The presented method also allows to introduce a shock
working on the risk properties of the underlying asset portfolio, and delineate the effect on
the risk properties of the bonds.
The rest of the paper is structured as follows: Section 2 describes how CDOs can be modeled. The model setup is presented and the implementation based on Monte Carlo simulation
is discussed. Section 3 covers the risk allocation to the individual tranches, both with respect
to the overall risk profile, tranche interdependencies, and systematic risk of tranches. In this
section, we establish a base case, describe individual tranche characteristics, provide robustness checks, and compare the results for the tranches with the properties of straight bonds.
Section 4 looks at various types of shocks to the reference portfolio and explores, how they
impact the risk characteristics of tranches. Section 5 concludes and discusses implications of
the findings for investors and regulators.
2
Modeling CDOs
This section focusses on analyzing the risk characteristics of CDOs and related financial
instruments1 . We construct a simple tool that allows us to portray the loss distribution of
asset portfolios, and of any tranche that is derived from the same underlying portfolio.
2.1
Model Setup
We apply a firm-value model to capture the occurrence of obligor default. More precisely,
we apply a structural one-factor correlated default model. The driving factor is a market
factor, and company value is modeled as the interplay of the market factor and a company
specific, idiosyncratic risk factor. This market model approach is the model of choice in most
corporate finance applications. We model company value Vn,t of each obligor n ∈ 1, 2, ..., N at
any time t before maturity as being driven by a generalized macroeconomic factor YtM that
is common to all securities, and an idiosyncratic component ǫn,t :
Vn,t =
q
q
M
ρM
Y
+
(1 − ρM
n t
n )ǫn,t
1
(1)
Collateralized Debt Obligations (CDOs) are structured financial instruments that exhibit two basic features: the pooling of underlying financial claims, and their tranching into a set of bonds, differentiated by the
degree of subordination.
3
with YtM ∼ Φ(0, 1) and ǫn,t ∼ Φ(0, 1), where Φ(0, 1) denotes the standard normal distribution. Thereby, we obtain correlated asset values of obligors. In case the sensitivities
p
M
ρM
n of firm values to the macroeconomic risk factor are the same for all obligors n, then ρn
corresponds to the mutual correlation coefficient for all assets.
The chosen one-factor approach is the model choice in many studies, such as Gibson
(2004), among others.
Obligor n is assumed to default if at any time t the value Vn,t of its assets lies below
the exogenously given default boundary Dn , i.e. Vn,t < Dn . Vn,t is assumed to be normally
distributed and is standardized such that Vn,t ∼ Φ(0, 1).
There is a simple relation linking every default boundary Dn to a particular default probability pn :
Dn = Φ−1 ∼ (pn ).
(2)
Usually, a fraction of the notional amount can be recovered in case of default. Let ψn
denote the recovery rate and θn the exposure size of security n. Portfolio loss is given as the
sum of individual loan losses. We define the portfolio loss rate P LR as the final value at
maturity of portfolio loss divided by the final value of all promised payments until maturity:
P LR =
PN
n=1 1{Tn >τn }
· θn · (Fn · (1 − ψn ) · expr(Tn −τn ) +Cn,τn ,Tn
,
PN
n=1 θn · (Fn + Cn,0,Tn )
(3)
where 1{Tn >τn } is an indicator function taking the value one if security n defaults during
its lifetime and zero otherwise. Tn represents maturity of security n, and tn is the time of
default. Fn denotes the redemption value and Cn,sn ,tn represents the present value at time
tn of all coupon payments for security n paid in the time interval [sn , tn ]. All payoffs are
discounted with interest rate r.
The applied firm value model (Eq. 1) is suitable for a simulation exercise.
2.2
Model Implementation
In the implementation, we do not need to apply simplifying assumptions to determine the loss
distribution of the underlying portfolio. Instead, we are able to fully profit from the Monte
Carlo Simulation procedure. Analytical approaches often rely on limiting assumptions, e.g.
that the portfolio is composed of an infinite number of securities with identical characteristics.
Thus, analytical models to some extent may be suitable for sensitivity analyses, but Monte
Carlo Simulation is more appropriate for real-world applications. All individual securities
in the portfolio can be accounted for by their specific exposure size, recovery rate, default
probability, and maturity. Furthermore, Monte Carlo Simulation allows to differentiate between obligors and individual securities. The occurrence of joint obligor defaults is modeled
by accounting for the sensitivity of each individual obligor to the common factor.
The loss distribution is simulated in 5 steps: First, a realization of the macro factor is
4
simulated until maturity. Subsequently, default scenarios are generated for all individual
obligors in the portfolio. Default occurs, if the simulated firm value of an obligor, based on
realizations of the macro factor and an idiosyncratic term, falls below the default boundary
which is determined by the default probability of the obligor. In the third step, individual
loan losses are obtained by applying a recovery rate to loan default. Fourth, portfolio loss is
given as the sum of realized individual loan losses. This corresponds to one realization in the
simulation. Fifth, many simulation runs yield the loss distribution of the entire portfolio.
The loss distribution depends on various input factors that may be grouped into three
categories: Individual loan components, portfolio components, and additional CDO features.
Individual loan components comprise maturity, credit quality, and credit migration probability, and expected recovery rate at default. Portfolio components comprise the sensitivities
of the individual loans to the common factor, portfolio diversification, and individual obligor
concentration. Furthermore, in practice, CDO loan portfolios present additional complications as they are dynamic portfolios with various restrictions concerning asset replenishment
over the lifetime of the issue. The implementation applied in this paper accounts for single
issuer default as well as portfolio characteristics, which are the focus of the investigation.
3
3.1
Risk Allocation to Tranches
Individual Tranche Characteristics
We now investigate the nature of risk transfer from the underlying portfolio to tranches.
This is at the heart of structured finance transactions, i.e. the pooling and reallocation of
individual risks to investors. The transfer of risks is non-proportional, due to the principle of subordination of tranches. The resulting risk allocation is estimated by Monte Carlo
simulation.
Let us consider as base case a reference portfolio with 10’000 loans. All securities have the
same characteristics: They are zero bonds with identical nominal value, 10 years to maturity,
7.63% default probability, 24.15% recovery rate, and identical exposure to the macro factor,
corresponding to a correlation of ρM
n = 0.15 between all securities. The applied default
probability corresponds to a Baa-rating for the bonds, according to Moody’s (2005), Exhibit
17. Overall, the base case represents a realistic setting for a typical CDO transaction. The
high number of loans is chosen intentionally to eliminate diversifiable risk to a large extent,
giving a clear picture of systematic risk in the analysis as shown later. The evolution of
individual-loan credit quality over time is simulated with 500’000 simulation runs.
Panel A of Figure 1 shows the obtained loss distribution for the reference portfolio. The
loss rate distribution has a typical shape for portfolios subject to credit risk, displaying a
substantial positive skewness. The two main parameters determining the shape of the loss
rate distribution are the default probability net of the recovery rate and the sensitivity of
the individual loans to the macro factor. The higher the macro factor sensitivity, the more
5
probability mass is shifted from the middle of the distribution to the tails, and vice versa.
Subsequently, following industry practice in the securitization market, the portfolio is
split into several tranches of strict subordination. In the subsequent analysis, the number
of tranches is assumed to be seven. Note that all results reported below remain essentially
unchanged if the number of tranches is smaller or larger, say 5 or 9. In practice, the tranches
are associated with different ratings by rating agencies. For given maturities of the tranches,
the ratings in turn correspond to specific default probabilities. We define the tranches by a
maximum default probability, which is fixed at the 1.01%, 2.57%, 3.22%, 7.63%, 19%, and
36.51% quantile of the loss rate distribution. These numbers correspond to the average issuerweighted cumulative default rates by whole letter rating for the period from 1920 to 2004 as
reported by Moody’s (2005). Thus, the six most senior tranches are rated Aaa, Aa, A, Baa,
Ba, and B according to Moody’s rating scheme, while the remaining first loss piece is not
rated. We number the tranches from 1 to 7, with the seventh tranche being the first loss
piece, or equity piece, which covers the residual loss. Tranche no.1, at the other end of the
spectrum, refers to the most senior tranche. All remaining tranches, nos. 2-6, are mezzanine
tranches.
Tranching is done with the intention of maximizing the size of each tranche except the
first loss piece, subject to the restrictions that the sizes of all more senior tranches are maximized and the default probability of that tranche is not greater than that required to obtain
a particular credit rating. Thus, a portfolio is tranched by first determining the lower attachment points of all tranches. The lower attachment point of each tranche is given as the
portfolio loss rate that is exceeded only with the default probability allowed for that tranche.
Since the upper attachment point of a tranche is identical to the lower attachment point of
the next senior tranche, the size of each tranche is given as the difference between the two
attachment points of that tranche. Thus, the number of different layers (tranches) and their
required maximum default probabilities, determined by the rating a tranche is supposed to
have, determine the attachment points and, correspondingly, the sizes of the tranches.
Applying the presented loss distribution of the total portfolio leads to the following tranche
sizes, represented as fraction of the total portfolio, starting from the most senior tranche:
0.7853, 0.0385, 0.0092, 0.0371, 0.0397, 0.0295, and 0.0607 for the equity piece. The detailed
summary statistics for the tranches are provided in Panel A of Table 1. Graphical representations of the loss distributions for different tranches (senior tranche, mezzanine tranche, and
first loss piece) are given in Figure 1.
As can be seen in Panel A of Table 1, the senior tranche is by far the largest part of the
entire transaction, making up 78.53% of the transaction. The expected loss rate is only 5
basis points, while expected loss given a default event is 495 basis points. The mean loss rate
is monotonic increasing in the degree of subordination. Its maximum value is 69.01% for the
equity piece. The default probability of the equity piece is 100%, as none out of 500’000 runs
in the simulation came out with a zero loss rate for the entire portfolio.
The senior tranche has the highest quality in all categories. The probability of default
6
is lowest, with no loss in 98.99% of all cases in this example. In addition, mean loss, loss
standard deviation, and loss given default are lowest among all tranches. Furthermore, the
senior tranche is by far the largest of all tranches, with a claim on 78.53% of the volume of
the underlying portfolio in the base case. In contrast to the senior tranche, the first loss piece
suffers a loss rate of 100% with a large probability of 36.51%. Furthermore, while low losses
occur at low frequency, higher losses occur with an increasing likelihood, peaking at a loss of
36% in the base case. Overall, the first loss piece (FLP) has the highest expected loss of all
tranches.
The numbers for the senior tranche are particularly striking, as they show a very low mean
loss given default, despite its large size. Panel D in Figure 1 explains why this is the case.
Realized portfolio losses that surpass the capacities of the more subordinate tranches cluster
at the low end of possible loss rates, without any observation exceeding a 40% loss rate in
the simulation runs. Clearly, the shape of the loss distribution for this tranche (as for the
other tranches as well) depends on the shape of the total portfolio loss distribution and the
applied tranching scheme. However, the differences are typically not very pronounced and
affect characteristics such as the steepness of the distribution.
As can be seen from Figure 1, the most senior mezzanine tranche, no. 2, displays a broad
tendency of a downward sloping distribution function throughout its domain. The loss rate
distributions of all mezzanine tranches have a similar shape. They are slightly downward
sloping as long as the lower cut-off value is larger than the mode of the loss rate distribution
of the total portfolio. This is the case in essentially all practically relevant cases since the
mode of the loss rate distribution typically lies in the domain of the first loss piece.
The distribution of the first loss piece, depicted in Figure 1, is single peaked in the interior
of its domain, abstracting from the spike at its upper boundary. This follows from the fact
that the lowest tranche comprises about two thirds of the cumulative loss rate distribution,
comprising the peak of the aggregate loss rate distribution.
Overall, the obtained results for the base case are typical for real-world securitization
transactions. To check the stability of the results, the base case is altered with respect to
three selected key characteristics of the reference portfolio. In particular, the correlation
coefficient is increased to 0.3, the default probability is increased to 0.19, and the number
of loans in the portfolios is reduced to 100. The resulting tranche statistics obtained when
tranching these portfolios with the same tranching scheme as applied for the base case are
given in Panels B, C, and D of Table 1.
Higher correlation (Panel B) than in the base case leads to a more fat-tailed portfolio loss
distribution while mean loss is not affected. More extreme loss realizations lead to smaller sizes
of extreme tranches (tranche no.1 and tranche no.7) and larger sizes of mezzanine tranches.
This can be seen in Panel B. In Panel C, the default probability is increased from 0.763 to
0.19, corresponding to a one notch downgrade, from Baa to Ba, according to Moody’s tables.
Increased default probability directly affects the mean loss of the reference portfolio and it
also leads to increased mean loss of all the tranches compared to the base case. Decreasing
7
the number of loans in the reference portfolio, as shown in Panel D, does not systematically
change the results compared to the base case.
Overall, we find the results in the base case to be robust with respect to changes of the
portfolio characteristics. While, of course, the numbers do change, the qualitative findings
are unchanged.
From the simulation exercise, we obtain a couple of insights. By tranching, the risks of
the underlying portfolio are allocated in a non-proportional way to the tranches. The loan
portfolio is transformed into several securities with entirely different risk characteristics. The
presented statistics illustrate that reference portfolios of average quality (a 7.63% default
probability over 10 years for all loans, conforming to a Baa rating, according to Moody’s) can
be divided into one large tranche of the highest quality, a couple of mezzanine tranches, and
a relatively small first loss piece in which the major proportion of credit risk is concentrated.
The tranches or only a selection of them, as is often intended, can subsequently be sold to
investors.
3.2
Tranche Interdependencies
In this section, we use the data generated in the simulation exercise in order to investigate the
correlation between tranche cash flows. The correlation between tranches of different issues is
analyzed, e.g. the correlation between two first loss pieces, or two senior tranches with distinct
underlying asset portfolios. Since we control the data generating process, we can trace the
effect of changes in the underlying asset correlations to the resulting tranche correlations.
Table 2 displays the bilateral correlations of all tranches (ranging from senior tranche to
the first loss piece) from two different CDOs with identical characteristics. Note that for large
portfolios, the correlation pattern converges in the limit to that of same-issue tranches. The
results indicate that tranches of similar credit quality, or seniority, have higher correlation
values than tranches with different credit quality. Tranche correlations decrease monotonically with increasing distance of quality between two tranches. The highest correlation values
are obtained for tranches with the same credit quality. These values are close to one. Correspondingly, the lowest bilateral correlation value (0.0729) is obtained for tranche 1, the most
senior tranche, and tranche 7, the most junior tranche. This shows that senior tranches are
almost orthogonal to junior tranches, in particular to the equity piece. Note that even lower
correlations can be attained by increasing the distance of tranches, e.g. by decreasing the
maximum default probability allowed for the senior tranche.
To determine the robustness of the obtained correlation pattern, the input parameters of
the base case are altered. Panel B of Table 2 displays the bilateral tranche correlations of two
different CDOs where the correlation between individual obligors in the reference portfolios
is increased to 0.3. The values of all individual correlations are larger than in the base case.
However, the correlation pattern is very similar to that of Panel A.
In Panel C of Table 2, the default probability is increased from 7.63% to 19%, correspond-
8
ing to a one-notch downgrade. While individual correlations rise for tranches with similar
credit quality, they decrease for tranches with different credit quality. In addition, lower
rated tranches have a higher correlation with the total reference portfolio and vice versa.
However, the overall correlation pattern is similar to that in Panel A and Panel B.
In an additional robustness check, the number of individual securities in the reference
portfolio is altered. In particular, the reference portfolio is assumed to consist of 100 loans
from different obligors, and the correlation between them is 0.15 as in the base case. The
results are presented in Panel D of Table 2. While the individual correlations are lower than
those in Panel A, again, the correlation pattern is very similar to that of the base case.
According to these simulations, the results confirm that portfolio risk is transferred to
tranches in a non-linear way. In particular, the risk associated with senior tranches is only
to a minor extent correlated with the risk of the first loss piece. Thus, a bank not selling all
tranches but retaining certain tranches will consequently be exposed to certain types of risk,
different from the original risk exposure.
3.3
Estimating the systematic risk of tranches
The objective of the analysis is to trace the macroeconomic-risk exposure of individual
tranches that are structured according to the principle of subordination.
With tranches given by the original tranching process and with simulated realizations
of the macroeconomic risk factor, the dependency of both the underlying portfolio and the
individual tranches to the macro factor can be extracted from the simulation results. The
simulation results allow us to estimate directly the relationship between the macro risk factor
and the realizations of particular tranches of an underlying loan portfolio. The aggregated
numbers are presented in Table 3.
Systematic risk, or macroeconomic-factor-dependent risk, of a security can be represented
by the average loss the security suffers in a particular state, determined by the macroeconomic
risk factor taking a certain value. Since the macro factor can take any value, there is an
unlimited number of states. For the exposition of the results obtained, all possible states
are aggregated to several broader states, determined by the macro factor taking values in
certain ranges. In order to provide a clear picture, the ranges are determined by the default
probabilities of the tranches. Thus, e.g. range no.1, covering the bad states, represents all
cases with the 1.01% lowest macro factor realizations, range no. 2 represents all cases with
macro factor realizations in the 1.01%-2.57% range, and range no. 7 stands for the highest
63.49% macro factor realizations, corresponding to the good states.
This structural-form representation of systematic risk avoids the linearity assumption
implicit in the more standard covariance, or beta, statistic of systematic risk. The measure
applied here to capture systematic risk is essentially a two-dimensional decomposition of total
reference-portfolio losses to states and tranches.
Note that the ranges, or states, are non-overlapping and add to the total range, consisting
9
of all macro factor realizations. With seven tranches given, we obtain by definition seven
states of the economy. This particular choice of seven states provides the clearest picture.
This is due to the fact that for each tranche, losses generally only tend to occur if the macro
factor is in a poor state, relative to the attachment point. Thus, the values in the bottom left
of each Panel are zero. This assumes, however, a minor role for idiosyncratic risk involved,
as is the case for large asset portfolios, such as the one in the base case with 10’000 loans.
With more idiosyncratic risk present, as is the case in Panel D with a portfolio of 100 loans,
the picture becomes more blurry and losses increasingly occur if the macro factor is in a good
state, relative to the attachment point.
In Table 3, the average loss for each tranche and the reference portfolio are shown for
the individual states. The average losses are shown both for all states (all realizations of the
macro factor) and for the seven states as described earlier. The entries in Table 3, therefore,
indicate the mean loss of a particular tranche in a particular state. Note that the numbers add
up column-wise for each tranche, line-wise for each state, and altogether to average total loss
of the portfolio (bottom right in each Panel). Thus, Table 3 shows the allocation of portfolio
risk to the tranches, indicating in which state of the macro factor these losses are occurring.
Panel A of Table 3 presents the allocation of macro-factor-dependent risk for the base case,
introduced earlier. The last row of Panel A shows that the average total portfolio loss of 5.79%
is mainly included in the first loss piece (4.19%), while the senior tranche only accounts for
0.04%, amounting to 0.68% of total portfolio risk. The last column shows how total portfolio
losses relate to macro-factor realizations.
In particular, the total portfolio (5.79% average loss) consists of 0.26% state-1 loss and
1.97% state-7 loss. As can be seen in the second last row of Panel A, all portfolio losses in
state 7 are borne by the first loss piece. State-1 losses, on the other hand, are shared almost
equally across all tranches, as tranche no. 1 bears 0.0004 and tranche no. 7 bears 0.0006 of
the total state-1 losses (0.0026). Thus, while the first loss piece bears losses attributable to
all states, tranche 1 only bears state-1 losses.
Panels B-D of Table 3 provide robustness checks of the results. Again, the portfolio
characteristics are altered with respect to correlation (Panel B), default probability (Panel
C), and number of loans in the portfolio (Panel D). Note that the states are the same as
in the base case. While the process of tranching leads to different tranche sizes than in the
base case, the applied tranching scheme and the default probabilities of the tranches are the
same. Overall, the numbers demonstrate that the basic pattern as discussed for the base case
remains unchanged.
In particular, we find that, for all variations of the base case, state-7 losses are entirely
covered by the first loss piece, and the most senior tranche only suffers in the bad state 1.
Or, to put it differently: the senior tranche covers only a certain share of the overall macro
tail risk, and at the same time it does not cover much else besides macro tail risk.
This section has an important result: under quite general assumptions about tranching,
the most subordinate tranche bears most macrofactor-related losses. Furthermore, the results
10
show that the impact of macro risk on the default rates of tranches varies systematically with
the rating quality of the tranche. According to the results, the more senior a tranche is, the
more likely is its default accompanied by a negative realization of the macro risk factor.
3.4
Comparing systematic tranche risk to the systematic risk of straight
bonds
The last section has analyzed how total systematic portfolio risk is allocated to individual
tranches in a standard stuctured finance transaction. For comparison, we now look at the
macro-factor dependence of non-tranched portfolios consisting of straight bonds with identical
default probabilities and ratings to that of the tranches discussed in the last section.
Panel A in Table 4 relies on Panel A of Table 3. The difference is that now, the numbers
are presented as fraction of total security loss over all states, i.e. column-wise they add up to
1, while row-wise they are weighted by number of observations. Panel B has seven portfolios,
each consisting of 10’000 bonds with a given rating, ranging from Aaa to B. Note that the
underlying portfolio in Panel A corresponds to the Baa-rated portfolio in Panel B.
The results indicate that while the Aaa-rated senior tranche no. 1 suffers almost all of its
losses in state 1, the untranched Aaa-rated straight-bond portfolio only suffers 8.25% of its
losses in state 1. In fact, this bond portfolio suffers losses in all states, even when the macro
factor has intermediate or good realizations. Moreover, all tranches exhibit higher state-1
losses than the bond portfolios with the same rating and lower state-7 losses. Note that the
results for the bonds portfolios are independent of the number of bonds in the portfolio. While
the results just presented are obtained for bond portfolios with 10’000 individual loans each,
the results do not change in the case of portfolios consisting of only one bond each.
These results demonstrate that tranches have completely different risk characteristics compared to straight-bond portfolios with the same default probability or rating. Tranches, especially the senior ones, are much more exposed to tail risks.
4
4.1
Shocks to portfolio and tranche risk
Risk characteristics of tranches
Having discussed the economics of tranching and the risk properties of tranches, we now
examine how changes in the original setting impact the risks of existing tranches, both in
absolute terms and relative to each other. In particular, we take tranching as given, i.e.
the attachment points of the tranches are fixed and the tranches are established as defined
securities, and investigate subsequent shocks to the quality of the reference portfolio. In the
model framework for CDOs applied in this paper, shocks affecting the loss rate distribution
of the underlying portfolio can be represented by changes of the default probability and by
changes of correlation.
11
These shocks may stem from several sources. Default probability may differ from the time
of tranching because of adverse realizations of the macroeconomic factor, but also simply
because it was misspecified, e.g. the original rating of the portfolio was false. Furthermore,
the quality of the portfolio, represented by the default probability, may suffer if the loans
are suddenly screened and monitored less well than before. This may happen if the first
loss piece is sold and the institution responsible for monitoring loses its incentives to ensure
timely repayment of loans. Correlation may differ from the time of tranching because the
commonality of obligors’ asset values increases, possibly due to business reasons.
Table 5 presents the summary statistics of existing tranches for the base case without
shock (Panel A), the case of a correlation shock (Panel B), and the case of a sudden increase
of the default probability (Panel C). Since tranching is taken as given, tranche size remains
constant by definition. A correlation increase has a different effect on mean tranche losses.
While mean loss increases for the senior tranches, it decreases for the junior tranches. The
same applies to the default probability and to mean loss given default. In Table 5, the
reversal occurs from tranche no.5 to tranche no.6 for mean loss and default probability, but
note that the exact turning point depends on the applied tranching scheme. The new default
probabilities in Panel B can be applied to determine the new tranche ratings by applying the
Table provided by Moody’s (2005), Exhibit 17. The new ratings are Baa (tranche nos.1-3),
Ba (tranche no.4), B (tranche no.5 and no.6), and tranche no.7 is not rated. The results also
show that the loss standard deviation of all tranches rises. This comes from the fact that
increasing correlation moves probability mass from the middle to the tails of the portfolio
loss rate distribution which naturally also affects the individual tranches. Panel C reports the
tranche statistics for increased default probability. The numbers demonstrate that not only
mean loss, default probability, and mean loss given default of the reference portfolio, but also
of all tranches increase. Loss standard deviation, in contrast, increases for senior tranches,
but decreases for junior tranches. Again, the new tranche ratings can be determined as Ba
(tranche no.1), B (tranche no.2 and no.3), while tranche nos.4-7 do not receive a rating.
4.2
Systematic risk of tranches
Having examined the general risk characteristics of tranches after shocks, we now discuss how
the allocation of systematic risk is affected by these shocks. Again, we look at a correlation
increase from ρ = 0.15 to ρ = 0.3 and an increase of the default probability to from 0.0763 to
0.19. The results are given in Table 6. Panel A in this Table is identical to Panel A of Table
3, to facilitate comparison.
We discuss the shock to asset correlation first. As can be seen from the last column in
Panel B, an increase in correlation boosts the losses of the reference portfolio in states 1 to 5,
and it lowers them in states 6 and 7. Since the overall mean-loss change is zero, the discussed
correlation shock leads to a redistribution of risk. The change of state-1 losses (losses in the
bad state, i.e. macro tail risk) is entirely borne by the senior tranche (tranche no.1), thereby
12
substantially increasing its risk position (from 0.0004 to 0.0018 for state 1).
Furthermore, in contrast to the base case (Panel A of Table 3), tranche 1 now also suffers
losses in the states 2, 3, and 4. As a result, total tranche-1 losses for all states rise from
0,0004 to 0.0031 in absolute numbers. The last row in Panel B shows that, with constant
total losses for the reference portfolio over all states, senior tranches (tranches 1 to 5 in this
case) bear more losses than in the base case, while the remaining more junior tranches have
less losses than before. Junior tranches apparently benefit from a decrease in high-state losses
(high realizations of the macro factor) while the senior tranches are hit disproportionately
by increased low-state losses (bad macroeconomic environment). Also, senior tranches are
generally more affected in good states than before.
Panel C of Table 6 presents the results for increased default probability. The numbers
are all equal or higher than in Panel A and show that all tranches are negatively affected.
However, the losses are split disproportionately to the tranches. In general, the higher the
seniority of the trance, the higher is the relative increase of losses. While tranche-7 losses
increase by 39% for all states, tranche-1 losses increase by 2850%. The senior tranches bear
a higher share of portfolio losses. This applies both to all states combined and individually.
In lower states, all tranches suffer higher losses than before.
The general picture that emerges from these results is as follows: if the initial loss distribution of the underlying portfolio is changing then, holding the original transaction fixed, there
is a strong incremental effect on the systematic risk of senior tranches in particular. For both
cases, increases in default probability and increases in correlation of the reference portfolio,
senior tranches experience by far the largest relative change in default expectation. For junior
tranches, in contrast, the relative change in mean loss due to an increase of reference portfolio
mean loss is relatively small, and they can actually benefit from an increase in correlation.
Naturally, the results are reversed in the case of a reduction in default probability or
correlation. Then, junior tranches will only have a slight benefit, or they will even suffer (if
correlation is falling), while senior tranches will quite generally benefit greatly.
5
Conclusion
In this paper, a methodology is suggested to capture the allocation of asset risk to tranches
of a multi-layer CDO transaction. The two-dimensional decomposition of losses to tranches
and states also allows to trace the effect of changes in asset risk on tranches. The obtained
results apply more generally to all structured products in financial markets.
The analysis suggests answers to three pertinent questions in today’s attempt to understand credit markets.
First, on the systematic risk of tranches: to what extent are junior and senior tranches
exposed to asset risk in general and to macro factor risk in particular? We show that the
common presumption about macro factor tail risk (extreme systematic risk) being largely
held by senior tranches is false. While senior tranches are in fact primarily exposed to macro
13
tail risk, the reverse is not true, as the share of macro tail risk borne by senior tranches is
quite limited.
Second, on the comparability of corporate-bond and CDO-tranche ratings: how does the
distribution of asset risk, in particular macro tail risk compare between these two financial
assets, given that they have the same letter rating? Applying the decomposition to a portfolio
of bonds, and comparing the results to the mapping obtained for the senior tranche of a
CDO, where the bond and the tranche both are rated Aaa, we find strikingly different tail
risk allocations. In particular, the senior tranche is basically a purebred tail-risk bond, while
the bond portfolio to a major part is exposed to non-tail risk, in addition to tail risk.
Third, on the sensitivity of tranches to changes in the underlying risk: how does the
allocation of risk respond to an unforeseen increase in asset correlation or overall default
probability, e.g. the downgrade of the underlying asset pool? To gauge the effect of a shock
to default probability, we compare the decomposition of portfolio losses before and after a
quality decrease commensurate to a rating downgrade, holding the original tranching structure
fixed, i.e. not allowing investors to renegotiate their contract.
Our findings show that the ultimate effect on tranche risk depends on the source of the
shock, i.e. whether asset correlations or default probabilities are changing. The most striking
result is that, in relative terms, the strongest increases in systematic risk is experienced by the
most senior tranche. This statement is true for correlation and default probability changes
alike.
What lessons can we draw from these results?
By applying the method of two-dimensional loss decomposition, one can trace the allocation of systematic risk to tranches quite generally for all structured products. This is essential
information for asset pricing and, perhaps even more importantly, for regulators who aim to
supervise systemic risk in banking. Thus, our findings are of particular relevance to issuers,
investors and regulators in the banking industry. Let us discuss these issues in turn.
First, while we do not intend to augment our analysis by a pricing model, we can make a
prediction concerning CDO-tranche price changes. This is also related to movements in credit
spreads, observed on housing markets during real estate recessions and other occasions of
deteriorating collateral quality. A shift of the underlying default expectation offers a possible
explanation for spread changes of bonds and tranche notes.
Second, the analysis of risk shifting can be linked to the literature on lending standards
in credit markets. For example, there is an extensive literature on the role of first loss piece
retention for maintaining proper lending standards (deMarzo 2005; Franke and Krahnen 2007).
Without equity piece retention, credit quality tends to decrease, provoking a shift of the loss
rate distribution. Thus, if an originator sells the equity piece of a given transaction, he
essentially provokes the risk distribution to shift, with consequences for tranche spreads, in
particular for the senior tranche.
Finally, the findings are relevant for bank risk management and for regulators aiming at
monitoring risk transfer in capital markets. The performed risk comparison of tranches and
14
straight bonds demonstrates that risk management based on ratings alone is not sufficient.
Thus, more generally, a transparent view of risk allocation to tranches and bonds will not
only improve risk management, it will also help to understand risk allocation in the economy.
15
References
[1] Andersen, T., T. Bollerslev, F. Diebold, and P. Christoffersen (2004). Practical volatility
and correlation modeling for financial market risk management, working paper, October.
[2] Coval, H. D., J. W. Jurek and E. Stafford (2007). Economic Catastrophe Bonds, working
paper, July.
[3] DeMarzo, P. (2005). The pooling and tranching of securities: a model of informed
intermediation, Review of Financial Studies 18, 1-35.
[4] Duffie, D. (2007). Implications in Credit Risk Transfer: Implications for Financial Stability, working paper, Stanford University.
[5] Eckner, A. (2007). Computational techniques for basic affine models of portfolio credit
risk, working paper, Stanford University.
[6] European Central Bank (2004). Credit risk transfer by EU banks: activities, risks, and
risk management. Report, May.
[7] European Central Bank (2007). Financial statistics for a global economy, Third European Central Bank conference on statistics, May.
[8] Franke, G., J. P. Krahnen (2007). Default risk sharing between banks and markets: the
contribution of collateralized debt obligations, NBER working paper, forthcoming: The
Risks of Financial Institutions, edited by M. Carey and R. Stulz, University of Chicago
Press.
[9] Gibson, M.S. (2004). Understanding the Risk of Synthetic CDOs, working paper.
[10] Greenspan, Alan (2004). Economic flexibility, Speech to HM Treasury Enterprise Conference, London, UK.
[11] Moody’s (2005). Default and recovery rates of corporate bond issuers, 1920-2004. Global
Credit Research. Moody’s Investors Service, January 2005.
[12] J.P. Morgan (2004). Global ABS/CDO Weekly Market Snapshot, Global Structured
Finance Research, JP Morgan New York, June 10.
16
Table 1: Summary statistics for tranches
This table presents summary statistics for the seven tranches, representing claims of strict subordination
on the underlying portfolio. The statistics indicate the allocation of losses of the underlying portfolio to
the individual tranches. The cut-off values for a particular tranche is determined by the default probability
allowed for that tranche as indicated in the fifth column and the default probability allowed for the next
senior tranche. The most junior tranche (tranche number 7) corresponds to the first loss piece. It bears the
first losses occurring in the portfolio, and only when its capacity is exhausted does the next senior tranche
take on losses. The columns present, from left to right, tranche number, tranche size, mean loss, loss standard
deviation, default probability, and mean loss given default (LGD). The last row of each panel displays the
statistics for the underlying portfolio. In Panel A (base case), the reference portfolio consists of 10’000 zero
bonds, and all of them are assumed to have a default probability of 7.63%, 10 years maturity, 24.15% recovery
rate, and a default correlation of 0.15. The loss distribution is calculated with 500’000 simulations. In Panel
B, the base case is altered and the default correlation is increased to 0.3. Panel C applies the portfolio
characteristics of the base case, except the default probability, which is increased to 19%. In Panel D, the
settings of the base case are applied, with the exception that the number of loans in the reference portfolio is 100.
Tranche
1
2
3
4
5
6
7
Total PF
Tranche
1
2
3
4
5
6
7
Total PF
Tranche
1
2
3
4
5
6
7
Total PF
Tranche
1
2
3
4
5
6
7
Total PF
Panel A: Base case
mean loss
loss std
default prob
0.05%
0.69%
1.01%
1.68%
11.82%
2.57%
2.88%
16.41%
3.22%
5.12%
20.32%
7.63%
12.47%
30.07%
19.00%
26.79%
40.89%
36.51%
69.01%
31.01%
100.00%
5.79%
4.55%
100.00%
Panel B: Different correlation (ρ = 0.3)
size
mean loss
loss std
default prob
67.72%
0.10%
1.31%
1.01%
7.02%
1.69%
11.84%
2.57%
1.71%
2.88%
16.40%
3.22%
6.85%
5.11%
20.30%
7.63%
6.93%
12.39%
29.99%
19.00%
4.52%
26.57%
40.79%
36.51%
5.25%
60.11%
37.51%
99.87%
100.00%
5.80%
6.89%
99.87%
Panel C: Different default probability (p=0.19)
size
mean loss
loss std
default prob
61.32%
0.07%
0.93%
1.01%
4.72%
1.69%
11.86%
2.57%
1.23%
2.88%
16.40%
3.22%
5.18%
5.15%
20.39%
7.63%
6.28%
12.60%
30.22%
19.00%
5.34%
27.02%
41.00%
36.51%
15.93%
73.83%
27.27%
100.00%
100.00%
14.42%
8.22%
100.00%
Panel D: Different number of loans (100 loans)
size
mean loss
loss std
default prob
77.24%
0.05%
0.72%
1.01%
4.55%
1.73%
11.88%
2.57%
0.76%
3.02%
17.12%
3.22%
3.79%
5.16%
20.54%
7.63%
4.55%
12.77%
30.28%
19.00%
3.04%
27.66%
41.64%
36.51%
6.07%
66.45%
34.21%
95.18%
100.00%
5.79%
4.95%
95.18%
size
78.53%
3.85%
0.92%
3.71%
3.97%
2.95%
6.07%
100.00%
17
mean LGD
4.95%
65.47%
89.54%
67.04%
65.61%
73.37%
69.01%
5.79%
mean LGD
9.80%
65.66%
89.45%
66.93%
65.19%
72.76%
60.19%
5.81%
mean LGD
6.93%
65.86%
89.48%
67.53%
66.33%
74.01%
73.83%
14.42%
mean LGD
5.12%
67.22%
93.92%
67.60%
67.22%
75.75%
69.81%
6.09%
Table 2: Bilateral correlations of tranches from two different CDO issues
This table displays the bilateral correlations of loss rates for all tranches from two different CDO issues
ranging from tranche number 1 (most senior tranche) to tranche number 7 (first loss piece). Both CDOs
have similar characteristics. In Panel A (base case), the reference portfolio consists of 10’000 zero bonds,
and all of them are assumed to have a default probability of 7.63%, 10 years maturity, 24.15% recovery rate,
and a default correlation of 0.15. The loss distribution is calculated with 500’000 simulations. In Panel
B, the base case is altered and the default correlation is increased to 0.3. Panel C applies the portfolio
characteristics of the base case, except the default probability, which is increased to 19%. In Panel D, the
settings of the base case are applied, with the exception that the number of loans in the reference portfolio is 100.
Panel A: Base case
CDO 1
1
2
3
4
5
6
7
CDO 2
1
0.9957
2
0.6061
0.9925
3
0.4315
0.8413
0.9785
4
0.3400
0.6637
0.8187
0.9935
5
0.2121
0.4141
0.5113
0.7324
0.9948
6
0.1305
0.2548
0.3146
0.4508
0.7421
0.9943
7
0.0729
0.1423
0.1757
0.2517
0.4144
0.6549
0.9961
Total PF
0.3697
0.5440
0.5936
0.7143
0.8341
0.8552
0.7647
7
0.0801
0.1515
0.1868
0.2676
0.4392
0.6927
0.9977
Total PF
0.4194
0.6077
0.6584
0.7802
0.8737
0.8422
0.6934
7
0.0723
0.1369
0.1686
0.2425
0.4002
0.6324
0.9980
Total PF
0.2915
0.4520
0.5068
0.6336
0.7883
0.8626
0.8375
7
0.0706
0.1426
0.1733
0.2458
0.4037
0.5745
0.7294
Total PF
0.3153
0.4723
0.5041
0.6077
0.7128
0.7161
0.6321
Panel B: Different correlation (ρ = 0.3)
CDO 1
1
2
3
4
5
6
7
CDO 2
1
0.9982
2
0.6262
0.9971
3
0.4460
0.8432
0.9910
4
0.3515
0.6648
0.8193
0.9976
5
0.2200
0.4161
0.5130
0.7347
0.9980
6
0.1356
0.2564
0.3160
0.4526
0.7430
0.9976
Panel C: Different default probability (p=0.19)
CDO 1
1
2
3
4
5
6
7
CDO 2
1
0.9957
2
0.6243
0.9935
3
0.4461
0.8442
0.9817
4
0.3500
0.6630
0.8157
0.9949
5
0.2179
0.4127
0.5081
0.7306
0.9963
6
0.1341
0.2540
0.3127
0.4497
0.7422
0.9963
Panel D: Different number of loans (100 loans)
CDO 1
1
2
3
4
5
6
7
CDO 2
1
0.7120
2
0.5033
0.6633
3
0.3803
0.5973
0.5925
4
0.3227
0.5689
0.6092
0.6883
18
5
0.2066
0.4062
0.4757
0.6169
0.7327
6
0.1252
0.2523
0.3053
0.4259
0.6317
0.7137
Table 3: Tranche loss in different states
This table displays the loss for each tranche and the underlying portfolio in different states, represented as
ranges of the macroeconomic factor. The losses are shown both for all realizations of the macro factor and
for the macro factor taking values in certain ranges that are given by the attachment points of the tranches.
The tranches range from tranche number 1 (most senior tranche) to tranche number 7 (first loss piece). In
Panel A (base case), the reference portfolio consists of 10’000 zero bonds, and all of them are assumed to
have a default probability of 7.63%, 10 years maturity, 24.15% recovery rate, and a default correlation of
0.15. The loss distribution is calculated with 500’000 simulations. In Panel B, the base case is altered and the
default correlation is increased to 0.3. Panel C applies the portfolio characteristics of the base case, except
the default probability, which is increased to 19%. In Panel D, the settings of the base case are applied, with
the exception that the number of loans in the reference portfolio is 100.
Panel A: Base case
State
1
2
3
4
5
6
7
All states
Tranche
1
0.0004
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0004
State
1
2
3
4
5
6
7
All states
Tranche
1
0.0007
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0007
State
1
2
3
4
5
6
7
All states
Tranche
1
0.0004
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0004
State
1
2
3
4
5
6
7
All states
Tranche
1
0.0003
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0004
2
0.0004
0.0003
0.0000
0.0000
0.0000
0.0000
0.0000
0.0006
Panel
3
4
5
6
0.0001
0.0004
0.0004
0.0003
0.0001
0.0006
0.0006
0.0005
0.0000
0.0002
0.0003
0.0002
0.0000
0.0007
0.0017
0.0013
0.0000
0.0000
0.0019
0.0033
0.0000
0.0000
0.0000
0.0023
0.0000
0.0000
0.0000
0.0000
0.0003
0.0019
0.0049
0.0079
B: Different correlation (ρ = 0.3)
7
0.0006
0.0009
0.0004
0.0027
0.0069
0.0106
0.0197
0.0419
Total PF
0.0026
0.0030
0.0011
0.0064
0.0122
0.0129
0.0197
0.0579
2
3
4
5
6
0.0007
0.0002
0.0007
0.0007
0.0005
0.0005
0.0003
0.0011
0.0011
0.0007
0.0000
0.0001
0.0004
0.0005
0.0003
0.0000
0.0000
0.0013
0.0031
0.0020
0.0000
0.0000
0.0000
0.0033
0.0051
0.0000
0.0000
0.0000
0.0000
0.0034
0.0000
0.0000
0.0000
0.0000
0.0000
0.0012
0.0005
0.0035
0.0086
0.0120
Panel C: Different default probability (p=0.19)
7
0.0005
0.0008
0.0003
0.0023
0.0060
0.0092
0.0124
0.0316
Total PF
0.0039
0.0044
0.0016
0.0087
0.0144
0.0126
0.0124
0.0580
2
3
4
0.0005
0.0001
0.0005
0.0003
0.0002
0.0008
0.0000
0.0000
0.0003
0.0000
0.0000
0.0010
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0008
0.0004
0.0027
Panel D: Different number of
6
0.0005
0.0008
0.0003
0.0024
0.0061
0.0043
0.0000
0.0144
loans)
7
0.0016
0.0025
0.0010
0.0070
0.0181
0.0279
0.0595
0.1176
Total PF
0.0043
0.0056
0.0022
0.0131
0.0273
0.0322
0.0595
0.1442
6
0.0003
0.0005
0.0002
0.0013
0.0029
0.0024
0.0008
0.0084
7
0.0006
0.0009
0.0004
0.0027
0.0069
0.0099
0.0189
0.0403
Total PF
0.0026
0.0030
0.0011
0.0064
0.0122
0.0129
0.0198
0.0579
2
0.0004
0.0003
0.0001
0.0001
0.0000
0.0000
0.0000
0.0008
3
0.0001
0.0001
0.0000
0.0000
0.0000
0.0000
0.0000
0.0002
4
0.0004
0.0005
0.0002
0.0006
0.0003
0.0000
0.0000
0.0020
19
5
0.0006
0.0010
0.0004
0.0028
0.0031
0.0000
0.0000
0.0079
loans (100
5
0.0005
0.0007
0.0003
0.0017
0.0021
0.0006
0.0000
0.0058
Table 4: Macro factor dependency of tranches versus untranched bond portfolios
This table displays the macro factor dependency of tranches with different ratings compared to macro factor
dependencies of untranched bond portfolios with the same ratings. Macro-factor dependency is represented
as losses occurring in different states, represented as ranges of the macroeconomic factor. The losses are
shown both for all realizations of the macro factor and for the macro factor taking values in certain ranges
that are given by the attachment points of the tranches. The tranches range from tranche number 1 (most
senior tranche) to tranche number 7 (first loss piece). In Panel A (base case), the reference portfolio consists
of 10’000 zero bonds, and all of them are assumed to have a default probability of 7.63%, 10 years maturity,
24.15% recovery rate, and a default correlation of 0.15. The loss distribution is calculated with 500’000
simulations. In Panel B, various portfolios are considered that solely consist of 10’000 bonds with a specific
rating, ranging from Aaa to B and determining the default probability. The applied default probabilities are
1.01% (Aaa), 2.57% (Aa), 3.22% (A), 7.63% (Baa), 19.00% (Ba), and 36.51% (B). All losses are given as
fraction of total losses of the examined tranche/bond portfolio attributable to a particular state. Securities
are given in columns, states are given in rows.
Panel A: Base case
State
1
2
3
4
5
6
7
All states
Tranche
1
0.9985
0.0017
0.0000
0.0000
0.0000
0.0000
0.0000
1.0000
2
0.5992
0.3985
0.0025
0.0000
0.0000
0.0000
0.0000
1.0000
State
1
2
3
4
5
6
7
All states
PF rating
Aaa
0.0825
0.0797
0.0274
0.1445
0.2316
0.2045
0.2298
1.0000
Aa
0.0636
0.0671
0.0238
0.1306
0.2250
0.2151
0.2747
1.0000
3
4
5
0.3503
0.1975
0.0810
0.5350
0.3050
0.1251
0.1079
0.1260
0.0521
0.0069
0.3698
0.3530
0.0000
0.0018
0.3875
0.0000
0.0000
0.0012
0.0000
0.0000
0.0000
1.0000
1.0000
1.0000
Panel B: Bond portfolios
A
0.0594
0.0640
0.0229
0.1269
0.2226
0.2172
0.2870
1.0000
Baa
0.0442
0.0519
0.0192
0.1109
0.2100
0.2230
0.3407
1.0000
20
Ba
0.0301
0.0390
0.0150
0.0912
0.1893
0.2231
0.4124
1.0000
6
0.0377
0.0582
0.0243
0.1646
0.4236
0.2906
0.0010
1.0000
B
0.0213
0.0298
0.0118
0.0749
0.1680
0.2164
0.4778
1.0000
7
0.0146
0.0226
0.0094
0.0639
0.1648
0.2535
0.4711
1.0000
Total PF
0.0442
0.0519
0.0192
0.1109
0.2100
0.2230
0.3407
1.0000
Table 5: Summary statistics for existing tranches
This table presents summary statistics for seven existing tranches for the case that characteristics of the
reference portfolio change at any time after the reference portfolio has been split into tranches. The statistics
indicate the allocation of losses of the underlying portfolio to the individual tranches. The most junior tranche
(tranche number 7) corresponds to the first loss piece. It bears all losses not covered by the other, more
senior, tranches. The columns present, from left to right, the tranche number, the tranche size, mean loss,
loss standard deviation, default probability, and mean loss given default (LGD). The last row of each panel
displays the statistics for the underlying portfolio. Panel A represents the base case as presented in Table 1.
The cut-off values for a particular tranche is determined by the default probability allowed for that tranche
as indicated in the fifth column. Panel B displays the tranche characteristics, taking tranche size as given
according to the base case and subsequently increasing the default correlation to 0.3. Panel C takes tranching
as given and displays the tranche characteristics after default probability rises to 19%.
Tranche
1
2
3
4
5
6
7
Total PF
Tranche
1
2
3
4
5
6
7
Total PF
Tranche
1
2
3
4
5
6
7
Total PF
Panel A: Base case
mean loss
loss std
default prob
0.05%
0.69%
1.01%
1.68%
11.82%
2.57%
2.88%
16.41%
3.22%
5.12%
20.32%
7.63%
12.47%
30.07%
19.00%
26.79%
40.89%
36.51%
69.01%
31.01%
100.00%
5.79%
4.55%
100.00%
Panel B: Increased correlation (ρ = 0.3)
size
mean loss
loss std
default prob
78.53%
0.40%
2.61%
4.19%
3.85%
5.38%
21.58%
6.79%
0.92%
7.20%
25.59%
7.62%
3.71%
9.78%
28.37%
12.30%
3.97%
16.26%
34.89%
21.03%
2.95%
26.10%
41.78%
32.13%
6.07%
56.62%
37.57%
99.87%
100.00%
5.80%
6.89%
99.87%
Panel C: Increased default probability (p=0.19)
size
mean loss
loss std
default prob
78.53%
1.50%
4.40%
18.47%
3.85%
23.89%
40.34%
30.00%
0.92%
31.69%
45.91%
33.44%
3.71%
41.27%
46.39%
49.74%
3.97%
59.96%
45.32%
70.54%
2.95%
78.40%
37.91%
85.90%
6.07%
95.65%
13.40%
100.00%
100.00%
14.42%
8.22%
100.00%
size
78.53%
3.85%
0.92%
3.71%
3.97%
2.95%
6.07%
100.00%
21
mean LGD
4.95%
65.47%
89.54%
67.04%
65.61%
73.37%
69.01%
5.79%
mean LGD
9.46%
79.16%
94.45%
79.48%
77.31%
81.25%
56.69%
5.81%
mean LGD
8.13%
79.64%
94.76%
82.98%
84.99%
91.27%
95.65%
14.42%
Table 6: Tranche loss in different states for existing tranches
This table presents the losses of the reference portfolio and the individual tranches for the case that the
characteristics of the reference portfolio change at any time after the reference portfolio has been split into
tranches. The loss for each tranche and the underlying portfolio is shown in different states, represented as
ranges of the macroeconomic factor. The losses are shown both for all realizations of the macro factor and for
the macro factor taking values in certain ranges that are given by the attachment points of the tranches. The
tranches range from tranche number 1 (most senior tranche) to tranche number 7 (first loss piece). In Panel
A (base case), the reference portfolio consists of 10’000 zero bonds, and all of them are assumed to have a
default probability of 7.63%, 10 years maturity, 24.15% recovery rate, and a default correlation of 0.15. The
loss distribution is calculated with 500’000 simulations. In Panel B, the base case is altered and the default
correlation is increased to 0.3. Panel C presents the losses for a tranched portfolio as in the base case that
experiences a default probability increase to 19%.
Panel A: Base case
State
1
2
3
4
5
6
7
All states
Tranche
1
0.0004
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0004
State
1
2
3
4
5
6
7
All states
Tranche
1
0.0018
0.0011
0.0002
0.0001
0.0000
0.0000
0.0000
0.0031
State
1
2
3
4
5
6
7
All states
Tranche
1
0.0022
0.0023
0.0008
0.0037
0.0029
0.0000
0.0000
0.0118
2
0.0004
0.0003
0.0000
0.0000
0.0000
0.0000
0.0000
0.0006
Panel
3
4
5
6
0.0001
0.0004
0.0004
0.0003
0.0001
0.0006
0.0006
0.0005
0.0000
0.0002
0.0003
0.0002
0.0000
0.0007
0.0017
0.0013
0.0000
0.0000
0.0019
0.0033
0.0000
0.0000
0.0000
0.0023
0.0000
0.0000
0.0000
0.0000
0.0003
0.0019
0.0049
0.0079
B: Increased correlation (ρ = 0.3)
7
0.0006
0.0009
0.0004
0.0027
0.0069
0.0106
0.0197
0.0419
Total PF
0.0026
0.0030
0.0011
0.0064
0.0122
0.0129
0.0197
0.0579
2
3
4
5
6
0.0004
0.0001
0.0004
0.0004
0.0003
0.0006
0.0001
0.0006
0.0006
0.0005
0.0003
0.0001
0.0002
0.0003
0.0002
0.0008
0.0004
0.0016
0.0018
0.0013
0.0000
0.0000
0.0008
0.0033
0.0034
0.0000
0.0000
0.0000
0.0001
0.0021
0.0000
0.0000
0.0000
0.0000
0.0000
0.0021
0.0007
0.0036
0.0065
0.0077
Panel C: Increased default probability (p=0.19)
7
0.0006
0.0009
0.0004
0.0027
0.0069
0.0104
0.0124
0.0344
Total PF
0.0039
0.0044
0.0016
0.0087
0.0144
0.0126
0.0124
0.0580
7
0.0006
0.0009
0.0004
0.0027
0.0069
0.0106
0.0359
0.0581
Total PF
0.0043
0.0056
0.0022
0.0131
0.0273
0.0322
0.0595
0.1442
2
0.0004
0.0006
0.0003
0.0017
0.0044
0.0019
0.0000
0.0092
3
0.0001
0.0001
0.0001
0.0004
0.0010
0.0012
0.0000
0.0029
4
0.0004
0.0006
0.0002
0.0016
0.0042
0.0064
0.0019
0.0153
22
5
0.0004
0.0006
0.0003
0.0018
0.0045
0.0070
0.0093
0.0238
6
0.0003
0.0005
0.0002
0.0013
0.0034
0.0052
0.0124
0.0231
Figure 1: Loss distribution of tranches
This diagram presents the loss distribution of a loan portfolio (Panel A) and three tranches (Panel B-D) at
maturity. The three tranches depicted are the first loss piece (Panel B), the most senior mezzanine tranche
(Panel C), and the most senior tranche overall (Panel D). The underlying portfolio consists of 10’000 securities
from different obligors. All securities have the same characteristics: They are zero bonds with 10 years
to maturity, 7.63% default probability, 24.15 % recovery rate, and identical exposure to the macro factor
(ρM
n = 0.15). The evolution of individual-loan credit quality over time is simulated with 500’000 simulation
runs. The horizontal axis shows the loss rate; the vertical axis shows the observed frequency, truncated at 5%,
0.3%, and 0.5% for the three tranches, respectively. There are several values surpassing these thresholds: For
the first loss piece, 100% loss occurs at a frequency of 36.51%. For the depicted mezzanine tranche, zero loss
occurs at a frequency of 87.43%, and 100% loss occurs at a frequency of 1.01%. For the senior tranche, zero
loss occurs at a frequency of 98.99%.
Panel A: Total portfolio
Panel B: First loss piece (tranche 7)
0.035
0.05
0.045
0.03
0.04
0.025
0.035
frequency
frequency
0.03
0.02
0.015
0.025
0.02
0.015
0.01
0.01
0.005
0.005
0
0
0.1
0.2
0.3
0.4
0.5
loss rate
0.6
0.7
0.8
0.9
0
1
0
Panel C: Mezzanine tranche (tranche 2)
0.1
0.2
0.3
0.4
0.5
loss rate
0.6
0.7
0.8
0.9
1
0.9
1
Panel D: Senior tranche (tranche 1)
0.003
0.005
0.004
0.002
frequency
frequency
0.003
0.002
0.001
0.001
0
0
0.1
0.2
0.3
0.4
0.5
loss rate
0.6
0.7
0.8
0.9
0
1
23
0
0.1
0.2
0.3
0.4
0.5
loss rate
0.6
0.7
0.8
CFS Working Paper Series:
No.
Author(s)
Title
2008/14
Stefan Mittnik
Tina Yener
Value-at-Risk and Expected Shortfall for Rare
Events
2008/13
Serena Lamartina
Andrea Zaghini
Increasing Public Expenditures: Wagner’s Law in
OECD Countries
2008/12
Jürgen Gaul
Erik Theissen
A Partially Linear Approach to Modelling the
Dynamics of Spot and Futures Prices
2008/11
Roman Kräussl
Niels van Elsland
Constructing the True Art Market Index - A Novel
2-Step Hedonic Approach and its Application to
the German Art Market
2008/10
Alan Muller
Roman Kräussl
Do Markets Love Misery? Stock Prices and
Corporate Philanthropic Disaster Response
2008/09
Christopher D.Carroll
Jirka Slacalek
Martin Sommer
International Evidence on Sticky Consumption
Growth
2008/08
Markus Haas
Stefan Mittnik
Multivariate Regime–Switching GARCH with an
Application to International Stock Markets
2008/07
Markus Haas
Stefan Mittnik
Mark S. Paolella
Asymmetric Multivariate Normal Mixture GARCH
2008/06
Charles Grant
Christos Koulovatianos
Alexander Michaelides
Mario Padula
Evidence on the Insurance Effect of Marginal
Income Taxes
2008/05
Dimitris Christelis
Dimitris Georgarakos
Michael Haliassos
Economic Integration and Mature Portfolios
Copies of working papers can be downloaded at http://www.ifk-cfs.de