Credit Risk II

• Credit Risk in Derivatives Transaction

– XVAs: CVA, DVA, FVA, MVA, KVA (adjustments in derivatives values)
– CVA
• Credit valuation adjustment (CVA) is an adjustment to the no-default
value of derivatives arising from the possibility of a counterparty
default
• CVA is calculated on a counterparty-by-counterparty basis.
– Agreements for bilaterally cleared transactions typically state
that outstanding transactions are netted in the event of a
default.
» If there are two outstanding transactions worth +10 and -6
with a counterparty, the potential loss will be 4.
– DVA
• Debit (or debt) valuation adjustment is an adjustment to a dealer’s
no-default value because the dealer itself might default.
• Like CVA, DVA must be calculated on a counterparty-by-
counterparty basis
– FVA and MVA: adjustments to the value of a derivatives portfolio for the
cost of funding derivatives positions
– KVA: capital valuation adjustment
– Calculation Issues
• The calculation of all the XVAs involves very time-consuming Monte
Carlo simulations for calculating expected future exposures, future
margin requirements, and future capital requirements
– Valuing Bilaterally Cleared Derivatives Portfolios
• Value after credit adjustments is (for the dealer):
No-default value − CVA + DVA
• Why does DVA increase the value of the portfolio of transactions
with the counterparty?
– It is zero-sum game, what is worse for the counterparty is better
for the dealer.
 – CVA calculation

Time 0 t1 t2 t3 t4 tn=T

Counterparty q1 q2 q3 q4 ……………… qn
default probability

PV of dealer’s loss v1 v2 v3 v4 ……………… vn

given default

n
CVA = ∑ qi vi
i =1

• Default probabilities are calculated from credit spreads

 s (ti −1 )ti −1   s (ti )ti 

qi = exp −  − exp − 
 1− R   1− R 

The average hazard rate 𝜆𝜆�𝑖𝑖 = 𝑠𝑠(𝑡𝑡𝑖𝑖 )/(1 − 𝑅𝑅)

 – DVA calculation

Time 0 t1 t2 t3 t4 ……………… tn=T

Dealer default q1* q2* q3* q4* ……………… qn*

probability

PV of counterparty’s v1* v2* v3* v4* ……………… vn*

loss given default
n
DVA = ∑ qi ∗vi ∗
i =1
– Calculation of vi’s
• The vi is calculated by simulating the market variables underlying
the portfolio in a risk-neutral world
– If no collateral is posted, the loss on a particular simulation trial
during the ith interval is the PV of (1-R)max(Vi, 0) where Vi is
the value of the portfolio at the midpoint of the interval
– vi is the average of the values of the potential losses across all
simulation trials
• More details in actual scenarios:
– Collateral (posted by both the dealer and the counterparty),
collateral changes the risk exposures and potential loss
– Downgrade Triggers
» Collateral required if the counterparty is downgraded below
a certain credit rating
» Examples: AIG was downgraded below AA on Sept 15,
2008, this triggered a collateral requirement that AIG was
not able to meet --- needs government bailout.
– CVA Risks
• The CVA for a counterparty is regarded as a complex derivative
• Increasingly dealers are managing it like any other derivative
(measuring the risks by calculating the Greeks)
• Two sources of risk:
– Changes in counterparty spreads
– Changes in market variables underlying the portfolio
– Wrong-Way/Right-Way Risk
• So far we only consider that the probability of default qi is
independent of net exposure vi. But they are often not independent,
we need to pay attention to the wrong-way risk
• Wrong-way risk (potentially destabilizing risk) occurs when qi is
positively dependent on vi
• Right-way risk occurs when qi is negatively dependent on vi
– Wrong-Way Risk Examples:
• The counterparty uses CDS to sell protection to the dealer (AIG sold
lots of CDS in the 08’ financial crisis). The time when the credit
spread of the reference entity increases, the value of the protection
to the dealer increases (vi increases); because of default correlation
among companies, the counter party’s default probability qi also
increases.
• Counterparty is speculating by entering many similar (unhedged)
trades with one or more dealers could lead to wrong-way risk for the
dealers. If the trades move against the counterparty, the
counterparty’s probability of default is likely to move higher.

An alpha multiplier (as large as 1.4) is often used to increase the CVA
for wrong-way risk.
• 1st Simple Situation
– Suppose a portfolio with a counterparty consists of a single uncollateralized
derivative that always has a positive value to the dealer and provides a
payoff only at time T (e.g. the dealer bought an option from a counterparty)
• No payoff before maturity, the present value of the expected exposure
at time ti < T is f0, the present value of the derivative.
vi = (1-R)f0, 𝐶𝐶𝐶𝐶𝐶𝐶 = (1 − 𝑅𝑅)𝑓𝑓0 ∑𝑛𝑛𝑖𝑖=1 𝑞𝑞𝑖𝑖
The adjusted derivative value
𝑓𝑓0∗ = 𝑓𝑓0 − 𝐶𝐶𝐶𝐶𝐶𝐶 = 𝑓𝑓0 1 − 1 − 𝑅𝑅 ∑𝑛𝑛𝑖𝑖=1 𝑞𝑞𝑖𝑖
• Now consider a zero-coupon bond that ranks equally with the derivative
in case of a default (same recovery rate), then
𝑓𝑓0∗
𝐵𝐵0∗ = 𝐵𝐵0 1 − 1 − 𝑅𝑅 ∑𝑛𝑛𝑖𝑖=1 𝑞𝑞𝑖𝑖 ; = 𝐵𝐵0∗ /𝐵𝐵0
𝑓𝑓0
𝐵𝐵0 = 𝑒𝑒 −𝛾𝛾𝑇𝑇 (the price of the zero-coupon bond assuming no default)
∗ 𝑇𝑇
𝐵𝐵0 ∗ = 𝑒𝑒 −𝛾𝛾 (the price of the zero-coupon bond)
𝑓𝑓0 ∗ = (𝐵𝐵0∗ /𝐵𝐵0 )𝑓𝑓0 = 𝑒𝑒 −𝑠𝑠 𝑇𝑇 𝑇𝑇 𝑓𝑓
0 , where 𝑠𝑠 𝑇𝑇 = 𝛾𝛾 ∗ − 𝛾𝛾
– 1st Simple Situation (cont.)
• The CVA adjustment has the effect of multiplying the value of the
transaction by e-s(T)T, where s(T) is the counterparty credit spread
for maturity T.
• DVA = 0 (the position is always positive to the dealer)

• Example:
1) A 2-year uncollateralized option sold by a new counterparty to
the dealer has a Black-Scholes-Merton value of $3
Assume a 2-year zero coupon bond issued by the counterparty
has a yield of 1.5% greater than the risk-free rate
If there is no collateral and there are no other transactions
between the parties, the adjusted value of the option is
3e-0.015×2=2.91
2) If the option the dealer holds is a binary option that pays Q if
the stock price is less than K in T=2 year, the adjusted value of
the option 𝑄𝑄𝑄𝑄 −𝑑𝑑2 𝑒𝑒 −𝑟𝑟𝑟𝑟 𝑒𝑒 −0.015𝑇𝑇
– 2nd Simple Situation: Uncollateralized Long Forward with Counterparty
• For a long forward contract that matures at time T
– The value of the contract at time t is
(Ft−K)e−r(T−t) (At time 0, the value is known (F0−K)e−rT; if the
contract is initiated at time 0, the value at time 0 should be 0)
• The expected exposure at time t is
E[max((Ft-K)e-r(T-t),0)]=e-r(T-t)E[max(Ft-K,0)]
Assuming Geometric Brownian motion for Ft., the expected
exposure can be calculated as
w(t)=e-r(T-t) [F0N(d1(t))-KN(d2(t))], where
𝐹𝐹 𝐹𝐹
𝑙𝑙𝑙𝑙[ 𝐾𝐾0 ]+(𝜎𝜎2 /2)𝑡𝑡 𝑙𝑙𝑙𝑙 𝐾𝐾0 −(𝜎𝜎2 /2)𝑡𝑡
𝑑𝑑1 = ; 𝑑𝑑2 =
𝜎𝜎 𝑡𝑡 𝜎𝜎 𝑡𝑡

σ is the volatility of the forward price.

Thus we have
vi = e-rt(1-R)w(ti)=e-rT(1-R)[F0N(d1(ti))-KN(d2(ti))]
– 2nd Simple Situation (cont.)
Example：2 year Gold forward with the current forward price is $1,600
per ounce. K = 1,500, r = 5%
• Consider two one-year intervals, and as an approximation, assume
the defaults can occur only in the mid-year: t1 =0.5, t2=1.5
• Assume σ = 20% and R = 0.3
• We can calculate v1 = 92.67 and v2 = 130.65;
• Suppose we can further estimate q1 =0.02 and q2=0.03
CVA=q1v1+q2v2=0.02×92.67+0.03×130.65 = 5.77
Value the forward contract after CVA
(1600−1500)e-0.05×2 − 5.77 = 84.71

How about DVA associated with this contract?

• Measuring credit risk --- Credit Value at Risk
– defined analogously to Market Risk VaR
• For example, one-year credit VaR with a 99.9% confidence is the
credit loss level that we are 99.9% confident will not be exceeded
over one year
– Credit VaR is important for determining the regulatory capital
requirement for credit risk
– It is also based on scenario analysis --- the probability of default
estimate should be based on real-world probabilities.
– The time horizon for credit risk VaR is much longer than market risk
VaR (one year compared to a few days for market risk VaR).
– For scenario simulation, we need to know the historical average default
probability for a company with a certain rating. The simulation will also
involve a rating transition matrix (e.g. transition probability from A to
Baa, etc.), as a company rating may not stay constant.
– The one-year rating transition matrix can be produced from historical
data.
 – One-Year Rating Transition Matrix
Moody’s (1970-2016)

Initial Rating at year-end

Rating Aaa Aa A Baa Ba B Caa Ca-C Default
Aaa 90.94% 8.36% 0.59% 0.08% 0.02% 0.00% 0.00% 0.00% 0.00%
Aa 0.87% 89.68% 8.84% 0.45% 0.07% 0.04% 0.02% 0.00% 0.02%
A 0.06% 2.64% 90.90% 5.67% 0.51% 0.12% 0.04% 0.01% 0.06%
Baa 0.04% 0.16% 4.44% 90.16% 4.09% 0.75% 0.17% 0.02% 0.18%
Ba 0.01% 0.05% 0.47% 6.66% 83.03% 7.90% 0.78% 0.12% 0.99%
B 0.01% 0.03% 0.16% 0.51% 5.32% 82.18% 7.39% 0.61% 3.79%
Caa 0.00% 0.01% 0.03% 0.11% 0.46% 7.82% 78.52% 3.30% 9.75%
Ca-C 0.00% 0.00% 0.07% 0.00% 0.80% 3.19% 11.41% 51.28% 33.24%
Default 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 100.00%

• How about a two-year rating transition matrix?

Let T be the one-year transition matrix

Two-year transition matrix M = TT: 𝑀𝑀𝑖𝑖𝑖𝑖 = ∑𝑛𝑛𝑘𝑘=1 𝑇𝑇𝑖𝑖𝑖𝑖 𝑇𝑇𝑘𝑘𝑘𝑘

 – Five-Year Rating Transition Matrix
M=TTTTT

Initial Rating at end

Rating Aaa Aa A Baa Ba B Caa Ca-C Default
Aaa 62.75% 28.14% 7.58% 1.16% 0.20% 0.07% 0.02% 0.00% 0.03%
Aa 2.96% 60.27% 29.98% 5.27% 0.82% 0.33% 0.13% 0.01% 0.20%
A 0.40% 9.00% 65.74% 19.73% 3.19% 1.08% 0.32% 0.04% 0.55%
Baa 0.18% 1.45% 15.45% 63.40% 12.25% 4.33% 1.18% 0.13% 1.68%
Ba 0.06% 0.39% 3.57% 19.72% 43.71% 19.78% 5.10% 0.55% 7.16%
B 0.04% 0.16% 0.94% 3.83% 13.22% 43.21% 16.56% 1.65% 20.39%
Caa 0.01% 0.06% 0.26% 0.94% 3.46% 17.61% 34.25% 3.53% 39.88%
Ca-C 0.00% 0.03% 0.21% 0.51% 2.09% 7.30% 12.50% 4.53% 72.82%
Default 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 100.00%

• How about a half-year transition probability?

– For the transition matrix over a period of less than a year
First, we diagonalize T: T= U-1ΛU,
𝜆𝜆1 ⋯ 0
Λ = ⋮ ⋱ ⋮ ; The eigenvalues {𝜆𝜆1, …,𝜆𝜆n} should be positive
0 ⋯ 𝜆𝜆𝑛𝑛
The transition matrix Tm over a period of 1/m of a year: Tm= U-1ΛmU,
where

𝜆𝜆11/𝑚𝑚 ⋯ 0
Λ𝑚𝑚 = ⋮ ⋱ ⋮ ;
0 ⋯ 𝜆𝜆𝑛𝑛 1/𝑚𝑚

This is the correct transition matrix, because

Λ = Λ𝑚𝑚 …Λ𝑚𝑚 (m times), and
𝑇𝑇𝑚𝑚 … 𝑇𝑇𝑚𝑚 = (𝑈𝑈 −1 Λ𝑚𝑚 𝑈𝑈) … (𝑈𝑈 −1 Λ𝑚𝑚 𝑈𝑈) = 𝑈𝑈 −1 Λ𝑚𝑚 … Λ𝑚𝑚 𝑈𝑈= 𝑈𝑈 −1 Λ𝑈𝑈 = 𝑇𝑇

In particular T = Th2 Th2, Th2 is the half-year transition matrix.

– One-Month Rating Transition Matrix

Initial Rating at month end

Rating Aaa Aa A Baa Ba B Caa Ca-C Default
Aaa 99.21% 0.77% 0.02% 0.01% 0.00% 0.00% 0.00% 0.00% 0.00%
Aa 0.08% 99.08% 0.81% 0.02% 0.00% 0.00% 0.00% 0.00% 0.00%
A 0.00% 0.24% 99.19% 0.52% 0.04% 0.01% 0.00% 0.00% 0.00%
Baa 0.00% 0.01% 0.40% 99.12% 0.39% 0.05% 0.01% 0.00% 0.01%
Ba 0.00% 0.00% 0.03% 0.63% 98.42% 0.78% 0.04% 0.01% 0.07%
B 0.00% 0.00% 0.01% 0.03% 0.53% 98.32% 0.75% 0.06% 0.30%
Caa 0.00% 0.00% 0.00% 0.01% 0.02% 0.79% 97.94% 0.41% 0.82%
Ca-C 0.00% 0.00% 0.01% 0.00% 0.08% 0.32% 1.42% 94.55% 3.62%
Default 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 100.00%

• Assumption of Independence

– The assumption that credit rating transition probability in one

period is independent of that in another period is not exactly
valid, but not too unreasonable for many purposes.

– Dependency due to rating momentum – If a company has been

downgraded, then it has a higher chance of being downgraded
again compared to other companies of the same rating.
– Credit Default Correlation：
• The credit default correlation between two companies is a measure
of their tendency to default at about the same time --- Default for
different companies don’t happen independently
– A key factor in evaluating credit risk VaR
• Vasicek’s Gaussian copula model is the “standard model”
– Basil II/III internal-ratings-based capital requirements for credit
risk in the banking book are based on the model
• Credit default correlation is also important in the valuation of some
credit derivatives, e.g. a first-to-default basket CDS (credit
protection ceases to exist when the CDS on the first asset within the
portfolio is triggered) and CDO tranches.
• There is no generally accepted measure of default correlation
– In Vasicek’s model, the default time t is mapped to a Gaussian
variable U, and the default correlation is defined through U, as
there is no clear way to define a default correlation using the
default time itself.
• Given the Gaussian copula correlation ρ and the average
unconditional probability of default (PD) by time T, we can derive
the worst-case default rate that will not be exceeded at the
confidence level X.
Prob(default_rate < WCDR) = X,
𝑁𝑁−1 𝑃𝑃𝑃𝑃 + 𝜌𝜌𝑁𝑁−1 𝑋𝑋
WCDR(T, X)= 𝑁𝑁
1−𝜌𝜌
WCDR(T, X) is the worst-case default rate at confidence level X.

For an individual loan

The credit risk VaR is simply given by
VaR = WCDR x EAD x LGD
EAD --- Exposure at default
LGD – Loss given default
• An example
– A bank has $100 million of retail exposures
– The 1-year probability of default averages 2% and the recovery
rate averages 60%, The copula correlation parameter is 0.1
– 99.9% worst-case default rate is
𝑁𝑁−1 0.02 + 0.1𝑁𝑁−1 (0.999)
𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊 = 𝑁𝑁 = 0.128
1−0.1
– The one-year 99.9% credit VaR is therefore 0.128×100x(1-0.6)
or $5.13 million

Gordy extends the result to a large portfolio of n loans (assuming one-

systematic factor, and idiosynchratic risks are diversified away), each
is a small part of the portfolio. The credit risk VaR is approximately
given by
𝑛𝑛

𝑉𝑉𝑉𝑉𝑉𝑉 = � 𝑊𝑊𝑊𝑊𝐷𝐷𝐷𝐷𝑖𝑖 × 𝐸𝐸𝐸𝐸𝐸𝐸𝑖𝑖 × 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖

𝑖𝑖=1
– Credit Risk Plus
A framework for calculating CVaR --- Developed by Credit Suisse 1997
How the default rate is modeled analytically?
• If default rate q is assumed to be known we can use a Poisson
distribution to determine the probability of m defaults, which is the
𝑛𝑛! 𝑚𝑚 (1 − 𝑞𝑞)𝑛𝑛−𝑚𝑚 ,
limiting case of the binomial distribution: 𝑛𝑛−𝑚𝑚 !
𝑞𝑞
𝑚𝑚!
in the limit of small q and large n, this can be approximated by a
𝑒𝑒 −𝑛𝑛𝑛𝑛 (𝑛𝑛𝑛𝑛)𝑚𝑚
Poisson distribution, 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 𝑚𝑚 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 =
𝑚𝑚!
• If the default rate is not constant (a default correlation would lead to
a distribution of default rates) --- Assuming the default rate q has a
gamma distribution (with standard deviation ϭ, mean μ), then the
probability of m defaults can be derived to have a negative binomial
distribution
𝑝𝑝𝑚𝑚 1 − 𝑝𝑝 𝛼𝛼 Γ(𝑚𝑚 + 𝛼𝛼)
𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 𝑚𝑚 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 =
Γ 𝑚𝑚 + 1 Γ(α)
where p = σ2/(σ 2+ μ), 𝛼𝛼=μ2/σ2
This, together with the exposure and recovery rate assumption, can be
used to calculate CVaR
– Monte Carlo Simulation Approach for Credit Risk Plus
In general, we would need MC simulation for obtaining CVaR:
• Sample overall default rate (from historical distribution)
• Sample probability of default for each counterparty category (regress
default rate for each category on overall default rate)
• Sample the number of losses for each counterparty category
• Sample the size of loss for each default (sampling of recovery rate)
• Calculate total loss from defaults
– This is repeated many times to calculate a probability
distribution for the total loss and obtain credit VaR

• This can be extended to take into account serial correlation in the

default rates (we will need a model of how the default rate depends
on macroeconomic variables and the time series modeling of these
variables).
– Credit Metrics
• Developed by JP Morgan in 1997
• Take into account rating changes as well as default (Vasicek’s
model and Credit Risk Plus don’t consider rating changes)
• A Gaussian copula model is used to define the correlation between
the rating transitions of different companies (the correlation can be
estimated using equity return correlation as a proxy)
– Example: Consider a pair of A-rated and B-rated companies.
The one-year transition probabilities for the A-rated company to
Aaa, Aa, A, … for A are 0.0006, 0.0264, 0.9090, …
These transition probabilities can be sampled via a Gaussian
variable xA. If the value xA of sampled is
xA < N-1(0.0006) = -3.2389  Aaa
-3.2389 < xA < N-1(0.0006+0.0264)= -1.9268  Aa
-1.9268 < xA < N-1(0.0006+0.0264+0.9090)= 1.522  A
– The transition probabilities for B can similarly be sampled via
the Gaussian variable xB
– When the correlation is taken into account, xA and xB are jointly
sampled from a bivariate normal distribution
• The Copula Model: xA and xB are sampled from a correlated
standard bivariate normal distribution
xA<-3.2389
Aaa xB<-3.7190
Aaa
-3.2389 < x < -1.9268 -3.7190 < x < -3.3528
Aa
A
Aa B

-1.9268 < x < 1.5220 -3.3528 < x < -2.8782

A
A B
A
1.5220 < xA < 2.4422 -2.8782 < x < -2.4522
Baa B
Baa
A 2.4422 < xA < 2.8480
B -2.4522< xB < -1.5523
Ba Ba

2.8480 < xA < 3.0902

B -1.5523 < xB < 1.1856
B

3.0902< xA < 3.2389

Caa 1.1856 < xB < 1.7060
Caa
3.2389<xA <3.2905 1.7060<xA <1.7756
Ca-C Ca-C
xA > 3.2905 xB > 1.7756
Default Default
– Credit Metrics (cont.)
• To calculate credit risk VaR of one year or longer, one only needs a one-
year rating transition matrix to simulate default by the end of the year and
rating change (credit spread change) at the end of the year.
• Shorter horizon rating changes will need to be considered using the Credit
Metrics approach for simulating the contribution (in the form of credit
spread) of credit risk to VaR on a trading book, which usually is defined as
10-day 99% VaR --- a large contribution of it is due to market risk.
• Credit Metrics also allows one to simulate the performance of a constant
level of risk strategy vs. the buy-and-hold strategy
– An example of a constant level of risk strategy
» Suppose a bank has a BBB bond and uses a liquidity horizon of
3 months
» At the end of each 3-month period the bond, if it has deteriorated
is assumed to be sold and replaced with a new BBB bond
» The one-year loss is then replaced by four three-month losses
– VaR and ES are generally less when a constant level of risk strategy
is used
– Securitization and the Credit Crisis of 2007
• Securitization
– Traditionally banks have funded loans with deposits
– Securitization is a way that loans can increase much faster than
deposits
• Asset-Backed Security (Simplified)

Senior Tranche
Asset 1 Principal: $75 million
Asset 2 Return = LIBOR + 60bp
Asset 3

Mezzanine Tranche
 SPV Principal:$20 million
Return = LIBOR+ 250bp
Asset n

Principal: Equity Tranche

$100 million Principal: $5 million
Return =LIBOR+2,000bp
– The Waterfall

Asset
Cash
Flows

Senior
Tranche

Mezzanine Tranche
Equity Tranche
 – ABS CDOs or Mezz CDOs (Simplified; more tranches in reality)

The mezzanine tranche is

repackaged with other
ABSs mezzanine tranches
Assets Senior Tranche (75%)
AAA
ABS CDOs
Senior Tranche (75%)
AAA
Mezzanine Tranche (20%)
BBB

Mezzanine Tranche
(20%) BBB

Equity Tranche (5%)

Not Rated

Equity Tranche (5%)

• CDO (Collateralized Debt Obligation) --- securitization of securities

 – Losses to AAA Tranche of ABS CDO

Losses on
Losses on Losses on
Losses on Equity Losses on
Mezzanine Mezzanine Senior
Subprime Tranche Senior Tranche
Tranche of Tranche of Tranche
portfolios of ABS of ABS CDO Senior
ABS ABS CDO (75%)
CDO AAA Tranche
(75%)
10% 25% 100% 100% 0% AAA
Mezzanine
15% 50% 100% 100% 33.30% Tranche
(20%) Mezzanine
20% 75% 100% 100% 66.70% BBB Tranche
(20%) BBB
25% 100% 100% 100% 100%
Equity
Tranche (5%)
Example: BBB Tranches Not Rated Equity
Tranche
BBB tranches of ABS CDOs were often quite thin (1% wide) (5%)

The BBB tranche of the Mezz ABS CDO in this simplified example:
20%x20%=4%
This means that they have a quite different loss distribution from BBB
bonds and should not be treated as equivalent to BBB bonds
They tend to be either safe or completely wiped out (cliff risk)
This type of risk is very hard to manage
 – A More Realistic Structure
High Grade ABS
CDO

Senior AAA 88%

Junior AAA 5%
AA 3%
A 2%
BBB 1%
ABS
AAA 81%
NR 1%

AA 11%
Subprime
Mortgages
A 4% Mezz ABS CDO CDO of CDO
BBB 3% Senior AAA 62% Senior AAA 60%
BB, NR 1% Junior AAA 14% Junior AAA 27%
AA 8% AA 4%
A 6% A 3%
BBB 6% BBB 3%
NR 4% NR 2%
• U.S. Real Estate Prices, 1987 to 2016: S&P/Case-Shiller Composite-
10 Index

250.00

200.00

150.00

100.00

50.00

0.00
– What went wrong?
1. Starting in 2000, mortgage originators in the US relaxed their
lending standards and created large numbers of subprime first
mortgages.
2. This, combined with very low-interest rates, increased the demand
for real estate and prices rose.
3. To continue to attract first-time buyers and keep prices increasing
they relaxed lending standards further

• Features of the market: 100% mortgages, ARMs (adjustable-rate),

teaser rates, NINJAs, liar loans, non-recourse borrowing (personally
not liable)
• Loan approvals are “standardized”, depending largely on the
applicant’s FICO score.
• Mortgages were packaged in financial products and sold to investors
• Banks found it profitable to invest in the AAA-rated tranches
because the promised return was significantly higher than the cost of
funds and capital requirements were low
– What went wrong? (cont..)
4. In 2007 the bubble burst. Some borrowers could not afford their
payments when the teaser rates ended. Others had negative
equity and recognized that it was optimal for them to exercise their
“put options”.
5. Foreclosures increased supply and caused U.S. real estate prices
to fall. Products, created from the mortgages, that were previously
thought to be safe began to be viewed as risky (The correlation
plays a role here)
6. There was a “flight to quality” and credit spreads increased to very
high levels
7. Many banks incurred huge losses
– What the market participants didn’t anticipate
• Default correlation goes up in stressed market conditions
• Recovery rates are less in stressed market conditions
• A tranche with a certain rating cannot be equated with a bond with
the same rating. For example, the BBB tranches used to create ABS
CDOs were typically about 1% wide and had “all or nothing” loss
distributions (WCDR=100%, quite different from the BBB bond)

– Regulatory Arbitrage
• The regulatory capital banks were required to keep for the tranches
created from mortgages was less than that for the mortgages
themselves
– Incentives
• The crisis highlighted what is referred to as agency costs
– Mortgage originators (Their prime interest was in originating
mortgages that could be securitized)
– Valuers (They were under pressure to provide high valuations
so that the loan-to-value ratios looked good)
– Traders (They were focused on the next end-of-year bonus and
not worried about any longer-term problems in the market)
– A huge amount of new regulations (Basel II.5, Basel III, Dodd-
Frank, etc). For example:
• Banks are required to hold more equity capital with the definition of
equity capital being tightened
• Banks required to satisfy liquidity ratios
• CCPs and SEFs (Swap execution facilities) for OTC derivatives
• Bonuses limited in Europe
• Bonuses spread over several years
• Proprietary trading is restricted (Volcker’s rule)
– Lessons from the Crisis
• Beware irrational exuberance
• Do not underestimate default correlations in stressed markets
• The recovery rate depends on the default rate
• Compensation structures did not create the right incentives
• If a deal seems too good to be true (eg, a AAA earning LIBOR plus
100 bp) it probably is
• Do not rely on ratings alone (the default correlation is more
important)
• Transparency is important in financial markets
– ABSs and ABS CDOs were complex inter-related products
– Once the AAA-rated tranches were perceived as risky they
became very difficult to trade because investors realized they
did not understand the risks
– Other credit-related products with simpler structures (eg, credit
default swaps) continued to trade during the crisis.
• Re-securitization was a badly flawed idea