Financial Engineering & Risk Management

Structured Credit: CDO’s and Beyond

M. Haugh G. Iyengar
Department of Industrial Engineering and Operations Research
Columbia University
 Securitization
Recall that securitization is the name given to the process of constructing new
securities from the cash-flows generated by a pool of underlying securities
– have already seen examples of securitization in the mortgage market.

The economic rationale behind securitization is that it enables the construction of

new securities with a broad range of risk profiles.
A broad range of investors may therefore be interested in these new securities
– even if they had no interest in the underlying securities.
This results in an increased demand for the underlying cash-flows
– and so the cost-of-capital is reduced for the issuers of the underlying
securities.

Credit default obligations (CDOs) are securities that are constructed from an
underlying pool of fixed-income securities
– first issued by banks in the mid-1990’s
– originally motivated by regulatory arbitrage considerations to supply CDO’s
to the market.
2
 Securitization
Collateralized loan obligations (CLOs) are securities that are constructed from an
underlying pool of loans.

The structured credit market was at the heart of the financial crisis
– terms like ABS, MBS, CDO and ABS-CDO’s became standard in the
financial (and mainstream) media.

Our introduction to structured credit and CDOs will:

1. Provide further examples of the securitization process
2. Allow us to introduce the (in)famous Gaussian copula model
3. Discuss the difficulties in risk managing structured credit portfolios
4. Emphasize just how crazy some parts of the financial markets had become in
the lead-up to the financial crisis.
The mechanics of how CDOs work have much in common with the mechanics of
credit default swaps (CDSs)
– so worthwhile to review and understand CDSs before studying CDO’s.
4
Financial Engineering & Risk Management
The Gaussian Copula Model

M. Haugh G. Iyengar
Department of Industrial Engineering and Operations Research
Columbia University
 Assumptions
We assume there are N bonds or credits in the reference portfolio.
Each credit has a notional amount of Ai .

If the i th credit defaults, then the portfolio incurs a loss of Ai × (1 − Ri )

– Ri is the recovery rate, i.e., the percentage of the notional amount that is
recovered upon default.
From a modeling perspective Ri is often assumed to be fixed and known
– in practice it is random and not known until after a default event has taken
place.

We also assume the risk-neutral distribution of default time of i th credit is

known
– it can be estimated from either credit-default-swap (CDS) spreads or the
prices of corporate bonds.

Therefore can compute qi (t), the risk-neutral probability that the i th credit
defaults before time t, for any time t.
2
 The Gaussian Copula Model

We let Xi denote the normalized asset value of the i th credit

– can think of Xi as representing the total value of assets of firm i.

We assume that q
Xi = ai M + 1 − ai2 Zi (1)
where M , Z1 , . . . , ZN are standard IID normal random variables
– note that each Xi is also a standard normal random variable.

Each of the factor loadings, ai , is assumed to lie in the interval [0, 1].

Clear that Corr(Xi , Xj ) = ai aj and that the Xi ’s are multivariate normally

distributed
– with covariance matrix equal to correlation matrix, P, where Pi,j = ai aj
for i 6= j.

3
 The Gaussian Copula Model
Assume that the i th credit has defaulted by time ti if Xi falls below some
threshold value, x̄i (ti ).

By earlier assumption must be the case that x̄i (ti ) = Φ−1 (qi (ti )) where Φ(·) is
the standard normal CDF.

Let F (t1 , . . . , tN ) denote the joint distribution of the default times of the N
credits in the portfolio. Then

F (t1 , . . . , tN ) = ΦP (x̄1 (t1 ), . . . , x̄N (tN ))

ΦP Φ−1 (q1 (t1 )), . . . , Φ−1 (qN (tN ))


=

where ΦP (·) denotes the multivariate normal CDF with mean vector 0 and
covariance matrix, P.

This is the infamous (1-factor) Gaussian copula model

– a 1-factor model because just 1 random variable, M , driving the
dependence between the Xi ’s

4
 Computing the Portfolio Loss Distribution
In order to price credit derivatives and CDOs in particular, we need to compute
the portfolio loss distribution.

First note that, conditional on M , the N default events are independent.

Then conditional on M , the default probabilities are given by

!
x̄i (t) − ai M
qi (t|M ) = Φ p
1 − ai2

where Φ(·) is the standard normal CDF.

Now let pN (l, t) denote risk-neutral probability that there are a total of l
portfolio defaults before time t.

Then may write Z ∞

pN (l, t) = pN (l, t|M ) φ(M ) dM (2)
−∞

where φ(·) is the standard normal PDF.

5
 Computing the Portfolio Loss Distribution
We can easily calculate pN (l, t|M ) using a simple iterative procedure:

Can then perform a numerical integration on the right-hand-side of (2) to

calculate pN (l, t).

If we assume that the notional, Ai , and the recovery rate, Ri , are constant
across all credits, then the loss on any given credit will be either 0 or A(1 − R).

Therefore knowing the distribution of the number of defaults is equivalent to

knowing the distribution of the total loss in the reference portfolio
– so the assumption of constant Ai ’s and Ri ’s simplifies calculations
– we will make this assumption although we could get by without it.
6
Financial Engineering & Risk Management
A Simple Example: Part I

M. Haugh G. Iyengar
Department of Industrial Engineering and Operations Research
Columbia University
 A Simple Example: A 1-Period CDO
We want to find the expected losses in a CDO with the following characteristics:
Maturity is 1 year
125 bonds in the reference portfolio
Each bond pays a coupon of one unit after 1 year if it has not defaulted
The recovery rate on each bond is zero
There are 3 tranches of interest: the equity, mezzanine and senior tranches

This example is taken from “The Devil is in the Tails: Actuarial Mathematics and the Subprime Mortgage
Crisis”, by C. Donnelly and P. Embrechts in ASTIN Bulletin 40(1), 1-33.
2
 A Simple Example: A 1-Period CDO
We make the simple assumption that the probability, q, of defaulting within 1
year is identical across all bonds.
– and that the correlation between each pair of default events is identical.
As before Xi is the normalized asset value of the i th credit and now
√ p
Xi = ρM + 1 − ρ Zi (3)
where M , Z1 , . . . , ZN are IID normal random variables.
Recall that the i th credit defaults if Xi ≤ x̄i . Since probability of default, q, is
identical across credits we have
x̄1 = · · · = x̄N = Φ−1 (q).
Moreover
P(Xi defaults | M ) = P(Xi ≤ x̄i | M )
√
P( ρM + 1 − ρ Zi ≤ Φ−1 (q) | M )
p
=
√
Φ−1 (q) − ρM
 
= P Zi ≤ √ |M
1−ρ
3
 The Conditional Default Distribution

Therefore conditional on M , the total number of defaults is Bin(N , qM ) so that

 
N l
pN (l|M ) = qM (1 − qM )N −l (4)
l

where √
Φ−1 (q) − ρM
 
qM := Φ √
1−ρ
where Φ(·) is the standard normal CDF and q is the risk-neutral probability of a
single name defaulting within 1 year.

Question: When correlations and default probabilities are not identical why is
(4) no longer valid?

Question: How do we calculate pN (l|M ) in that case?

4
 Computing the Expected Tranche Losses
We can compute the expected loss on each tranche:
EQ
0 [Equity tranche loss] = 3 × P (3 or more defaults by year end)
X2
+ k × P (k defaults by year end)
k=1

EQ
0 [Mezzanine tranche loss] = 3 × P (6 or more defaults by year end)
X2
+ k × P (k + 3 defaults by year end)
k=1

EQ
0 [Senior tranche loss] = 3 × P (9 or more defaults by year end)
X2
+ k × P (k + 6 defaults by year end)
k=1

Each probability P (·) can be calculated by integrating the binomial probabilities

with respect to M : Z ∞
p(l) = pN (l|M ) φ(M ) dM (5)
−∞

5
Financial Engineering & Risk Management
A Simple Example: Part II

M. Haugh G. Iyengar
Department of Industrial Engineering and Operations Research
Columbia University
 A Simple Example: A 1-Period CDO
We want to find the expected losses in a CDO with the following characteristics:
Maturity is 1 year
125 bonds in the reference portfolio
Each bond pays a coupon of one unit after 1 year if it has not defaulted
The recovery rate on each bond is zero
There are 3 tranches of interest: the equity, mezzanine and senior tranches

This example is taken from “The Devil is in the Tails: Actuarial Mathematics and the Subprime Mortgage
Crisis”, by C. Donnelly and P. Embrechts in ASTIN Bulletin 40(1), 1-33.
2
 Some Important Observations
Regardless of the individual default probability, q, and correlation, ρ, we see
EQ Q Q
0 [Equity tranche loss] ≥ E0 [Mezzanine tranche loss] ≥ E0 [Senior tranche loss]

– only holds when each tranche has same notional exposure

– in this case 3 units.

The expected losses in the equity tranche are always decreasing in ρ.

Mezzanine tranches are often relatively insensitive to ρ

– this has implications when it comes to model calibration.

The expected losses in super senior tranches (with upper attachment point of
100% or 125 units in our example) are always increasing in ρ
– as evidenced in figure on next slide where the senior tranche is now
exposed to all losses between 7 and 125.

Note also that in figure on next slide the total expected losses on the three
tranches, i.e. the expected losses on the index, is independent of ρ
– this is not an accident!
4
Financial Engineering & Risk Management
The Mechanics of a “Synthetic” CDO Tranche

M. Haugh G. Iyengar
Department of Industrial Engineering and Operations Research
Columbia University
 The Mechanics of a “Synthetic” CDO Tranche
Recall there are N credits in the reference portfolio.
Each credit has the same notional amount, A,
If the i th credit defaults, then the portfolio incurs a loss of A × (1 − R)
– the recovery rate, R, is assumed fixed, known and constant across credits.
A tranche is defined by the lower and upper attachment points, L and U ,
respectively.
The tranche loss function, TLL,U (l), for a fixed time, t, is a function of the
number of defaults, l, up to time t and is given by

TLtL,U (l) := max {min{lA(1 − R), U } − L, 0} .

For a given number of defaults it tells us the loss suffered by the tranche.

Example: Suppose L = 3% and U = 7% and suppose total portfolio loss is

lA(1 − R) = 5%.
Then tranche loss is 2% of total portfolio notional
– or 50% of tranche notional = 7% − 3% = 4%.
2
 The Mechanics of a “Synthetic” CDO Tranche
When an investor sells protection on the tranche she is guaranteeing to reimburse
any realized losses on the tranche to the protection buyer.

In return, the protection seller receives a premium at regular intervals from the
protection buyer
– these payments typically take place every three months.

In some cases protection buyer may also make an upfront payment in addition to,
or instead of, a regular premium
– often the case for equity tranches which have a lower attachment point of
zero.

The fair value of the CDO tranche is that value of the premium (plus upfront
payment if applicable) for which the expected value of the premium leg equals
the expected value of the default leg
– so just like a swap, the initial value of the position is zero.

Clearly then the fair value of the CDO tranche depends on the expected value of
the tranche loss function at each of the fixed time periods.
3
 The Mechanics of a “Synthetic” CDO Tranche
Indeed, for a fixed time, t, the expected tranche loss is given by
h i N
X
L,U
EQ
0 TL t = TLtL,U (l) p(l, t)
l=0
R∞
which we can compute using our expression p(l, t) = −∞
pN (l, t|M ) φ(M ) dM .

4
Financial Engineering & Risk Management
Computing the Fair Value of a CDO Tranche

M. Haugh G. Iyengar
Department of Industrial Engineering and Operations Research
Columbia University
 The Fair Value of the Premium Leg
The premium leg represents the premium payments that are paid periodically by
the protection buyer to the protection seller.

These payments are made at the end of each time interval and they are based
upon the remaining notional in the tranche
– in this sense different to a CDS since the latter contract ends as soon as a
default occurs.

Formally, the time t = 0 value of the premium leg, P0L,U , satisfies

n
X  h i
PL0L,U = S dt ∆t (U − L) − EQ
0 TL L,U
t−1 (6)
t=1

where:
n is the number of periods in the contract
dt is the risk-free discount factor for payment date t
S is the annualized spread or premium paid to the protection seller
∆t is the accrual factor for date t. e.g. ∆t = 1/4 for quarterly payments.
2
 The Fair Value of the Default Leg

The default leg represents the cash flows paid to the protection buyer upon losses
occurring in the tranche.

Formally, the time t = 0 value of the default leg, DL0L,U , satisfies

n
X  h i h i
DL0L,U = dt EQ
0 TLt
L,U
− E Q
0 TL L,U
t−1 . (7)
t=1
h i
L,U
While some programming is required, we can calculate the EQ
0 TLt ’s
numerically very quickly using the Gaussian copula model
– therefore can easily compute PL0L,U and DL0L,U
– the principal reason for the model’s popularity despite its many flaws.

3
 The Fair Value of the Tranche
The fair premium, S ∗ say, is the value of S that equates the value of the default
leg with the value of the premium leg at the beginning of the contract:
DL0L,U
S ∗ := P  h i . (8)
n L,U
t=1 dt ∆t (U − L) − EQ
0 TLt−1

As is the case with swaps and forwards, the fair value of the tranche to the
protection buyer and seller at initiation is therefore zero.

Easy to incorporate any possible upfront payments that the protection buyer
must pay at time t = 0 in addition to the regular premium payments.
Can also incorporate recovery values and notional values that vary with each
credit in the portfolio.

In practice S ∗ was seen in the market and (8) was used to calibrate an implied
correlation parameter
– unfortunately a different correlation value is inferred for each tranche
– and sometimes not possible to solve / calibrate an implied correlation
parameter at all.
4
Financial Engineering & Risk Management
Cash and Synthetic CDOs

M. Haugh G. Iyengar
Department of Industrial Engineering and Operations Research
Columbia University
 Cash CDOs
The first CDOs to be traded were all cash CDOs where the reference portfolio
actually existed and consisted of corporate bonds that the CDO issuer usually
kept on its balance sheet.
Capital requirements meant that these bonds required a substantial amount of
capital to be set aside to cover any potential losses.

To reduce these capital requirements, banks converted the portfolio into a series
of tranches and sold most of these tranches to investors.
But they usually kept the equity tranche for themselves
– thereby keeping most of the economic risk and rewards of the portfolio
– but they also succeeded in dramatically reducing the amount of capital
they needed to set aside.
First CDO deals were therefore motivated by regulatory arbitrage considerations.
But cash CDOs must be managed, and the legal documentation can be lengthy
– waterfall structures, credit enhancement etc. are all important features
– the tranches were typically “rated” by the ratings agencies.
2
 Synthetic CDOs
Soon become clear there was an appetite in the market-place for these products
– hedge funds, for example, were keen to buy the riskier tranches
– whereas insurance companies and others sought the AAA-rated senior and
super-senior tranches.

This appetite and explosion in the CDS market gave rise to synthetic tranches
– the underlying reference portfolio is no longer a physical portfolio of
corporate bonds or loans
– instead it is a fictitious portfolio consisting of a number of credits with an
associated notional amount for each credit.

3
 Mechanics of Synthetic CDOs
Mechanics of a synthetic tranche are precisely as described earlier.

But they have at least two features that distinguish them from cash CDOs:

(1) With a synthetic CDO it is no longer necessary to tranche the entire

portfolio and sell the entire “deal".
e.g. A bank could sell protection on a 3%-7% tranche and never have to
worry about selling the other pieces of the reference portfolio
– this is not the case with cash CDOs.
(2) Because the bank no longer owns the underlying bond portfolio, it is no
longer hedged against adverse price movements
– it therefore needs to dynamically hedge its synthetic tranche position
– typically does so using the CDS markets
– but hedging and risk managing CDO portfolios are very difficult with
or without a good pricing model
– there are simply too many moving parts.

4
Financial Engineering & Risk Management
Pricing and Risk Management of CDO Portfolios

M. Haugh G. Iyengar
Department of Industrial Engineering and Operations Research
Columbia University
 A Sample Synthetic CDO Portfolio

A Simple Example of a Structured Credit Portfolio

In practice structured credit portfolios could contain many, many positions with:
different reference portfolios, different maturities and counter-parties
different trading formats, i.e. upfront and / or a running spread format.

2
 Pricing and Risk Management of CDO Portfolios
Ultimate payoff of such a portfolio is very path dependent with substantial
idiosyncratic risk
– very difficult to risk-manage these portfolios properly
– can also be expensive to unwind due to wide bid-offer spreads.

Computing the mark-to-market value of these portfolios can also be very difficult
because market prices may non be transparent
– witness, for example, the “Belly of the Whale Series" on the Alphaville blog
of the Financial Times at http://ftalphaville.ft.com/
– apparently the so-called London Whale first came to attention
because price levels in the CDX IG9 index diverged too much from
other related price levels.

3
 Pricing and Risk Management of CDO Portfolios
Risk-management for structured credit portfolios is also very challenging:
1. Scenario analysis is certainly difficult
– what are the main risk factors?
– what are reasonable stress levels for these factors?
– how do we re-evaluate the portfolio in a given scenario?
2. The Greeks can be computed and used to risk-manage a portfolio
– but they don’t work well either and are model dependent
– witness the fallout from the downgrade of Ford and General
Motors in May 2005 as described in the Wall Street Journal on 12th
Sep 2005 and available at:
http://online.wsj.com/article/0,,SB112649094075137685,00.html
– but don’t believe every thing you read!
Liquidity risk and market endogeneity are key risks that must be considered.
Over-reliance on ratings agencies, models, the behavior of organizations etc. all
played an important part in the crisis.
4
 A Brief Aside on Copulas
The most famous model for pricing structured credit securities is the Gaussian
copula model.
There has also been enormous criticism aimed at this model

– most (if not all of it) justified

– but nothing we didn’t know well before the crisis!

There has been much academic work on building better and more sophisticated
models
– but none of them are really satisfactory.

Aside: A common fallacy is that the marginal distributions and correlation matrix
are sufficient for describing the joint distribution of a multivariate distribution.

This is not true! Correlation only measures linear dependence

– and the next slide provides a counter-example.

5
A Brief Aside on Copulas

Each figure shows 5000 simulated points from the Bivariate Normal and
Meta-Gumbel distributions.
Both bivariate distributions have standard normal marginal distributions.
In each case the correlation is .7.
But Meta-Gumbel is much more likely to see large joint moves.
6
Financial Engineering & Risk Management
CDO-Squared’s and Beyond

M. Haugh G. Iyengar
Department of Industrial Engineering and Operations Research
Columbia University
 CDO-Squared’s and Beyond
It should already be clear that structured credit portfolios consisting of CDO
tranches can be difficult to risk manage.

But at least there is a solid risk-sharing motivation for the creation of CDOs
– true for securitization in general.

But the structured credit market quickly ran amok with the creation and trading
of ever more complex securities
– for example, products such as CDO-squared’s or CDO2 were soon
developed
– they were difficult to justify economically
– and provided great examples of product risk, model risk, legal
risk, etc.

Before discussing CDO2 , recall first how a CDO is constructed . . .

2
 Creating a CDO2
Suppose now we have 100 different CDO’s.

We can construct a new CDO using the mezzanine tranches, for example, of
these 100 CDO’s
– and create a (synthetic) pool of mezzanine tranches.

This pool now forms the collateral for a new CDO

– which we call a CDO2 !

The CDO2 only incurs losses when the mezzanine tranche in one of the 100
underlying CDO’s incurs losses
– so a mezzanine tranche of a CDO now plays the role of a bond (or CDS).

Note that many of the same bonds act as collateral for many of the underlying
CDO’s.

5
 The Joys of CDO-Squared’s
Question: How would you price and risk-manage a CDO2 ?
Some considerations ...
1. The legal contract governing each of the mezzanine tranches in the
underlying portfolio of CDO’s is ≈ 150 pages long
– so only need to read approx 100 × 150 = 15, 000 pages of legal
documents
– must also read the contract governing the CDO2 , of course.
2. How would you keep track of CDO2 performance?
– many thousands of lines of computer code required!

Question: How would we perform a scenario analysis?

Question: How would we estimate its VaR or CVaR?

But why stop there!? There are also ABS-CDO’s, CDO-cubed’s, a.k.a. CDO3
and more ...