GRM - Day 3.1

Scenario Analysis and
Stress Testing
Chapter 17
Risk Management and Financial Institutions 2e, Chapter 17, Copyright © John C. Hull 2009 1
Stress Testing
l Key Questions
l How do we generate the scenarios?
l How do we evaluate the scenarios?
l What do we do with the results?
Generating the scenarios
l Stress individual variables
l Choose particularly days when there were
big market movements and stress all
variables by the amount they moved on
those days
l Form a stress testing committee of senior
management and ask it to generate the
scenarios
Core vs Peripheral Variables
l If scenario generated involves only a few
“core” variables, regress other “peripheral”
variables on the core variables to
determine their movements. (Kupiec,
1999)
l Ideally the relationship between peripheral
and core variables should be estimated for
stressed market conditions (Kim and
Finger, 2000)
Making Scenarios Complete
l Often an adverse scenario has an
immediate effect on the value of a portfolio
and a “knock on” effect
l Examples
l Credit crisis of 2007
l LTCM
l Greek Bond
l 2020 Corona Virus
Reverse Stress Testing
l Use an algorithm to search for scenarios
where large losses occur
l Can be a useful input to the stress testing
committee.
What are the Incentives of a
Financial Institution?
l If the stress testing committee comes
up with extreme scenarios more
regulatory capital is likely to be required
l The stress testing committee may
therefore has an incentive to “water
down” the scenarios they consider
Scenarios Proposed by
Regulators?
l Will regulators provide their own scenarios to

be used by all banks?
l Part of the Basel Committee’s consultative
document suggests that it is thinking about
this as a possibility
l There is a danger that, if the scenarios are
announced in advance, financial institutions
will hedge only against the scenarios (See
Business Snapshot 17.1; traffic light options)
What to do with the Results?
l Should managers place more reliance on
stress testing results or VaR results
l One idea is to ask the stress testing
committee to assign probabilities to
scenarios (e.g. 0.05% or 0.2% or 0.5%)
l The stress scenarios can then be
integrated with the historical simulation
scenarios to produce a composite VaR
Example from Chapter 12
Scenario Loss ($000s) Probability Cumul. Probability
s5 850.000 0.00050 0.00050
s4 750.000 0.00050 0.00100
h494 499.395 0.00198 0.00298
s3 450.000 0.00200 0.00498
h339 359.440 0.00198 0.00696
h329 341.366 0.00198 0.00894
s2 300.000 0.00200 0.01094
h349 251.943 0.00198 0.01292
h487 247.571 0.00198 0.01490
h131 241.712 0.00198 0.01688
s1 235.000 0.00500 0.02188
h227 230.265 0.00198 0.02386
h495 227.332 0.00198 0.02584
h441 225.051 0.00198 0.02782
…. …. …. ….
…. …. …. ….
Subjective vs Objective
Probabilities
l Objective probabilities are calculated from data
l Subjective probabilities is base don a
individual’s judgment.
l Objective probabilities are inevitably backward
looking
l The procedure just described is a way of
combining subjective and objective probabilities.
Credit Risk: Estimating
Default Probabilities
Chapter 16
Altman’s Z-score (Manufacturing
companies) page 348
l X1=Working Capital/Total Assets
l X2=Retained Earnings/Total Assets
l X3=EBIT/Total Assets
l X4=Market Value of Equity/Book Value of Liabilities
l X5=Sales/Total Assets
Z = 1.2X1+1.4X2+3.3X3+0.6X4+0.99X5
If the Z > 3.0 default is unlikely; if 2.7 < Z < 3.0 we should
be on alert. If 1.8 < Z < 2.7 there is a moderate chance of
default; if Z < 1.8 there is a high chance of default
Estimating Default Probabilities
l Alternatives:
l Use historical data
l Use bond prices or asset swaps
l Use CDS spreads (only introduction)
l Use Merton’s model X not do it
Historical Data
Historical data provided by rating agencies

can be used to estimate the probability of
default.
Banks can calculate the probability of
default of their loans from past data.
Cumulative Average Default Rates %
(1970-2010, Moody’s) Table 16.1, page 350
Time (years)
1 2 3 4 5 7 10
Aaa 0.000 0.013 0.013 0.037 0.104 0.244 0.494
Aa 0.021 0.059 0.103 0.184 0.273 0.443 0.619
A 0.055 0.177 0.362 0.549 0.756 1.239 2.136
Baa 0.181 0.510 0.933 1.427 1.953 3.031 4.904
Ba 1.157 3.191 5.596 8.146 10.453 14.440 20.101
B 4.465 10.432 16.344 21.510 26.173 34.721 44.573
Caa 18.163 30.204 39.709 47.317 53.768 61.181 72.384
Interpretation
l The table shows the probability of

default for companies starting with a
particular credit rating
l A company with an initial credit rating of
Baa has a probability of 0.181% of
defaulting by the end of the first year,
0.510% by the end of the second year,
and so on
Do Default Probabilities Increase
with Time?
l For a company that starts with a good
credit rating default probabilities tend to
increase with time
l AA rating bond .021, .038, .044, .081% for 1, 2, 3, 4, and 5 yrs
l For a company that starts with a poor

credit rating default probabilities tend to
decrease with time
l Caa rating bond 18.16, 12.041, 9.505, 7.608% and 6.451 % for 1, 2,
3, 4, and 5 yrs
Hazard Rate vs. Unconditional
Default Probability
l The hazard rate or default intensity is the
probability of default over a short period of
time conditional on no earlier default
l The unconditional default probability is the
probability of default as seen at time zero
l dV(t)/dt = - l(t) V(t)
l V(T) = e – [integration 0 to t l(t) d(t)]
l Defining Q(t) as PD, Q(t) = 1 – e [- l(t)*t]
Properties of Hazard rates
l Suppose that l(t) is the hazard rate at time t
l The probability of default between times t and
t+Dt conditional on no earlier default is l(t)Dt
l The probability of default by time t is
1 - e - l (t )t
where l (t ) is the average hazard rate between
time zero and time t
Recovery Rate
The recovery rate for a bond is usually
defined as the price of the bond 30 days
after default as a percent of its face value
From Chapter 11 Loss Given Default (LGD)
% LGD = [100% - % recovery rate ]
Recovery Rates; Moody’s: 1982 to 2010
(Table 16.2, page 352)
Class Ave Rec Rate (%)

First lien bank loan 65.8
Second lien bank loan 29.1
Senior unsecured bank loan 47.8
Senior secured bond 50.8
Senior unsecured bond 36.7
Senior subordinated bond 30.7
Subordinated bond 31.3
Junior subordinated bond 24.7
Recovery Rates Depend on
Default Rates
l Moody’s best fit estimate for the 1982 to
2007 period is
Ave Recovery Rate =
59.33 − 3.06 × Spec Grade Default Rate
l R2 of regression is about 0.5
Credit Default Swaps (page 352)
l Buyer of the instrument acquires protection from the

seller against a default by a particular company or
country (the reference entity)
l Example: Buyer pays a premium of 90 bps per year
for $100 million of 5-year protection against company
X
l Premium is known as the credit default spread. It is
paid for life of contract or until default
l If there is a default, the buyer has the right to sell
bonds with a face value of $100 million issued by
company X for $100 million (Several bonds may be
deliverable)
CDS Structure (Figure 16.1, page 354)
90 bps per year
Default Default
Protection Protection
Buyer, A Seller, B
Payoff if there is a default by
reference entity=100(1-R)
Recovery rate, R, is the ratio of the value of the bond issued

by reference entity immediately after default to the face value
of the bond
Other Details
l Payments are usually made quarterly in arrears
l In the event of default there is a final accrual
payment by the buyer
l Increasingly settlement is in cash and an auction
process determines cash amount
l Suppose payments are made quarterly in the
example just considered. What are the cash
flows if there is a default after 3 years and 1
month and recovery rate is 40%?
Attractions of the CDS Market
l Allows credit risks to be traded in the

same way as market risks
l Can be used to transfer credit risks to a
third party
l Can be used to diversify credit risks
Credit Indices (page 356-357)
l CDX IG: equally weighted portfolio of 125
investment grade North American companies
l iTraxx: equally weighted portfolio of 125
investment grade European companies
l If the five-year CDS index is bid 165 offer 166 it
means that a portfolio of 125 CDSs on the CDX
companies can be bought for 166bps per
company, e.g., $800,000 of 5-year protection on
each name could be purchased for $1,660,000
per year. When a company defaults the annual
payment is reduced by 1/125.
Use of Fixed Coupons
l Increasingly CDSs and CDS indices trade
like bonds
l A coupon and a recovery rate is specified
l There is an initial payments from the buyer
to the seller or vice versa reflecting the
difference between the currently quoted
spread and the coupon
Credit Default Swaps and Bond
Yields (page 357-358)
l Portfolio consisting of a 5-year par yield

corporate bond that provides a yield of 6% and a
long position in a 5-year CDS costing 100 basis
points per year is (approximately) a long position
in a riskless instrument paying 5% per year
l What are arbitrage opportunities in this situation
is risk-free rate is 4.5%? What if it is 5.5%?
Risk-free Rate
l The risk-free rate used by bond traders when
quoting credit spreads is the Treasury rate
l The risk-free rate traditionally assumed in
derivatives markets is the LIBOR/swap rate
l By comparing CDS spreads and bond yields it
appears that in normal market conditions traders
are assuming a risk-free rate 10 bp less than the
LIBOR/swap rates
l In stressed market conditions the gap between
the LIBOR/swap rate and the “true” risk-free rate
is liable to be much higher
Asset Swaps
l Asset swaps are used by the market as an
estimate of the bond yield relative to
LIBOR
l The present value of the asset swap
spread is an estimate of the present value
of the cost of default
Asset Swaps (page 358)
l Suppose asset swap spread for a particular corporate
bond is 150 basis points
l One side pays coupons on the bond; the other pays
LIBOR+150 basis points. The coupons on the bond are
paid regardless of whether there is a default
l In addition there is an initial exchange of cash reflecting
the difference between the bond price and $100
l The PV of the asset swap spread is the amount by which
the price of the corporate bond is exceeded by the price
of a similar risk-free bond when the LIBOR/swap curve is
used for discounting
CDS-Bond Basis
l This is the CDS spread minus the Bond
Yield Spread
l Bond yield spread is usually calculated as
the asset swap spread
l Tended to be positive pre-crisis
Using CDS Prices to Predict
Average hazard rate between time zero and
time t is to a good approximation
s (t )
l=
1- R
where s(t) is the credit spread calculated for
a maturity of t and R is the recovery rate
More Exact Calculation for Bonds
(page 361)
l Suppose that a five year corporate bond pays a coupon

of 6% per annum (semiannually). The yield is 7% with
continuous compounding and the yield on a similar risk-
free bond is 5% (with continuous compounding)
l The expected loss from defaults is 8.75. This can be
calculated as the difference between the market price of
the bond and its risk-free price
l Suppose that the unconditional probability of default is Q
per year and that defaults always happen half way
through a year (immediately before a coupon payment).
Calculations
Time Def Recovery Risk-free Loss Discount PV of Exp
(yrs) Prob Amount Value Factor Loss
0.5 Q 40 106.73 66.73 0.9753 65.08Q
1.5 Q 40 105.97 65.97 0.9277 61.20Q
2.5 Q 40 105.17 65.17 0.8825 57.52Q
3.5 Q 40 104.34 64.34 0.8395 54.01Q
4.5 Q 40 103.46 63.46 0.7985 50.67Q
Total 288.48Q
Calculations continued
l We set 288.48Q = 8.75 to get Q = 3.03%
l This analysis can be extended to allow
defaults to take place more frequently
l With several bonds we can use more
parameters to describe the default
probability distribution
Real World vs Risk-Neutral
l The default probabilities backed out of
bond prices or credit default swap spreads
are risk-neutral default probabilities
l The default probabilities backed out of
historical data are real-world default
probabilities
A Comparison
l Calculate 7-year hazard rates from the

Moody’s data (1970-2010). These are real
world default probabilities)
l Use Merrill Lynch data (1996-2007) to
estimate average 7-year default intensities
from bond prices (these are risk-neutral
default intensities)
l Assume a risk-free rate equal to the 7-year
swap rate minus 10 basis points
Real World vs Risk Neutral
Default Probabilities (7 year
averages) Table 16.4, page 363
Rating Historical Hazard Rate Hazard Rate from bonds Ratio Difference
(% per annum) (% per annum)
Aaa 0.03 0.60 17.2 0.57
Aa 0.06 0.73 11.5 0.67
A 0.18 1.15 6.5 0.97
Baa 0.44 2.13 4.8 1.69
Ba 2.23 4.67 2.1 2.44
B 6.09 8.02 1.3 1.93
Caa 13.52 18.39 1.4 4.87
Risk Premiums Earned By Bond
Traders (Table 16.5, page 364)
Rating Bond Yield Spread of risk-free Spread to Extra Risk

Spread over rate used by market compensate for Premium
Treasuries over Treasuries default rate in the (bps)
(bps) (bps) real world (bps)
Aaa 78 42 2 34
Aa 86 42 4 42
A 111 42 11 63
Baa 169 42 26 104
Ba 322 42 132 148
B 523 42 355 126
Caa 1146 42 759 345
Possible Reasons for These Results
(The third reason is the most important)
l Corporate bonds are relatively illiquid

l The subjective default probabilities of bond
traders may be much higher than the
estimates from Moody’s historical data
l Bonds do not default independently of each
other. This leads to systematic risk that
cannot be diversified away.
l Bond returns are highly skewed with limited
upside. The non-systematic risk is difficult to
diversify away and may be priced by the
market
Which World Should We Use?
l We should use risk-neutral estimates for
valuing credit derivatives and estimating
the present value of the cost of default
l We should use real world estimates for
calculating credit VaR and scenario
analysis
Merton’s Model (Section 16.8, pages 367-370)
l Merton’s model regards the equity as an

option on the assets of the firm
l In a simple situation the equity value is
max(VT –D, 0)
where VT is the value of the firm and D is
the debt repayment required
Equity vs. Assets
An option pricing model enables the value

of the firm’s equity today, E0, to be related
to the value of its assets today, V0, and the
volatility of its assets, sV
E 0 = V0 N ( d 1 ) - De - rT N ( d 2 )
where
ln (V0 D) + ( r + sV2 2) T
d1 = ; d 2 = d1 - sV T
sV T
Volatilities
¶E
s E E0 = s V V0 = N ( d 1 ) s V V0
¶V
This equation together with the option pricing

relationship enables V0 and sV to be
determined from E0 and sE
Example
l A company’s equity is $3 million and the
volatility of the equity is 80%
l The risk-free rate is 5%, the debt is $10
million and time to debt maturity is 1 year
l Solving the two equations yields V0=12.40
and sv=21.23%
Example continued
l The probability of default is N(-d2) or
12.7%
l The market value of the debt is 9.40
l The present value of the promised
payment is 9.51
l The expected loss is about 1.2%
l The recovery rate is 91%
The Implementation of Merton’s
Model to estimate real-world default
probability (e.g. Moody’s KMV)
l Choose time horizon
l Calculate cumulative obligations to time horizon. We denote it
by D
l Use Merton’s model to calculate a theoretical probability of
default
l Use historical data to develop a one-to-one mapping of
theoretical probability into real-world probability of default.
l Assumption is that the rank ordering of probability of default
given by the model is the same as that for real world
probability of default
l Distance to default is ln(V0 ) - ln( D) + (r - sV2 / 2)T
sV T
The Implementation of Merton’s
Model to estimate real-world default
probability (e.g. CreditGrades)
l Same approach can be used

l In this case the assumption is that the rank
ordering of probability of default given by
the model is the same as that for real
world probability of default
The Credit Crisis of 2007
Chapter 6
U.S. Real Estate Prices, 1987 to 2011:
S&P/Case-Shiller Composite-10 Index
Case Shiller
S&P/Case-Shiller U.S. National Home Price Index (CSUSHPISA)
What happened…
l Starting in 2000, mortgage originators in the US relaxed
their lending standards and created large numbers of
subprime first mortgages.
l This, combined with very low interest rates, increased
the demand for real estate and prices rose.
l To continue to attract first time buyers and keep prices
increasing they relaxed lending standards further
l Features of the market: 100% mortgages, ARMs, teaser
rates, NINJAs, liar loans, non-recourse borrowing
What happened...
l Mortgages were packaged in financial products and sold to investors
l 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
l 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.
l U.S. real estate prices fell and products, created from the mortgages,
that were previously thought to be safe began to be viewed as risky
l There was a “flight to quality” and credit spreads increased to very
high levels
Asset Backed Security (Simplified)
ABS
Senior Tranche
Asset 1 Principal: $75 million
Asset 2 Return = 6%
Asset 3
Mezzanine Tranche
6 SPV Principal:$20 million
Return = 10%
Asset n
A “waterfall” defines
Principal: Equity Tranche the precise rules for
$100 million Principal: $5 million allocating cash flows
Return =30% to tranches
The Waterfall
Asset
Cash
Flows
Senior
Tranche
Mezzanine Tranche
Equity Tranche
ABS CDOs or Mezz CDOs (Simplified)
ABSs
Subprime Mortgages Senior Tranches (75%)
AAA ABS CDO
Senior Tranche (75%)
Mezzanine Tranches (20%) AAA
BBB
Mezzanine Tranche
(20%) BBB
Equity Tranches (5%)
Not Rated
Equity Tranche (5%)
How much of the original portfolio of subprime

mortgages is AAA?
Losses to AAA Tranche of ABS CDO
(Table 6.1)
Losses on Losses on Losses on Losses on Losses on

Subprime Mezzanine Equity Mezzanine Senior
portfolios Tranche of Tranche of Tranche of Tranche of
ABS ABS CDO ABS CDO ABS CDO
10% 25% 100% 100% 0%
15% 50% 100% 100% 33.3%
20% 75% 100% 100% 66.7%
25% 100% 100% 100% 100%
A More Realistic Structure
(Figure 6.5)
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%
BBB Tranches
l BBB tranches of ABSs were often quite thin (1%
wide)
l This means that they have a quite different loss
distribution from BBB bonds and should not be
treated as equivalent to BBB bonds
l They tend to be either safe or completely wiped
out (cliff risk)
l What does this mean for the tranches of the
Mezz ABS CDO?
Regulatory Arbitrage
l Capital required for securities created from
a portfolio of mortgages was considerably
less than capital that would be required if
mortgages had been kept on the balance
sheet
Role of Incentives
l Arguably the incentives of valuers, the
creators of ABSs and ABS CDOs, and
rating agencies helped to create the crisis
l Compensation plans of traders created
short-term horizons for decision making
Importance of Transparency
l ABSs and ABS CDOs were complex inter-

related products
l Once the AAA rated tranches were perceived
as risky they became very difficult to trade
because investors realized they did not
understand the risks
l Other credit related products with simpler
structures (eg, credit default swaps)
continued to trade during the crisis.
Lessons from the Crisis (page 133-
134)
l Beware Irrational Exuberance (Book by Shiller 2013 Nobel)
l Do not underestimate default correlations in stressed
markets
l Recovery rate depends on default rate
l Compensation structures did not create the right
incentives
l If a deal seems too good to be true (eg, a AAA earning
LIBOR plus 100 bp) it probably is
l Do not rely on ratings
l Transparency is important in financial markets
l Resecuritization was a badly flawed idea
Credit Value at Risk
Chapter 18
Rating Transitions
l One year rating transition probabilities are
published by rating agencies.
l If we assume that the rating transition in one
period is independent of that in other periods we
can calculate the rating transition for any period
(see Appendix J and software)
l The “ratings momentum” phenomenon means
that the independence assumption is not
perfectly correct
One-Year Rating Transition
Matrix (% probability, Moody’s 1970-2010)
Table 18.1 page 401
Initial Rating at year end

Rating Aaa Aa A Baa Ba B Caa Ca-C Default
Aaa 90.42 8.92 0.62 0.01 0.03 0.00 0.00 0.00 0.00
Aa 1.02 90.12 8.38 0.38 0.05 0.02 0.01 0.00 0.02
A 0.06 2.82 90.88 5.52 0.51 0.11 0.03 0.01 0.06
Baa 0.05 0.19 4.79 89.41 4.35 0.82 0.18 0.02 0.19
Ba 0.01 0.06 0.41 6.22 83.43 7.97 0.59 0.09 1.22
B 0.01 0.04 0.14 0.38 5.32 82.19 6.45 0.74 4.73
Caa 0.00 0.02 0.02 0.16 0.53 9.41 68.43 4.67 16.76
Ca-C 0.00 0.00 0.00 0.00 0.39 2.85 10.66 43.54 42.56
Default 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 100.00
Five-Year Rating Transition
Matrix (calculated from one-year transitions)
Table 18.2 page 401
Initial Rating at end

Aaa 61.12 29.99 7.70 0.89 0.21 0.05 0.01 0.00 0.03
Aa 3.45 61.89 28.70 4.71 0.73 0.25 0.07 0.01 0.19
A 0.44 9.72 65.78 18.88 3.24 1.06 0.24 0.04 0.60
Baa 0.22 1.69 16.38 60.98 12.93 4.64 0.97 0.13 2.06
Ba 0.07 0.44 3.40 18.20 44.69 20.07 3.70 0.52 8.92
B 0.04 0.20 0.83 3.27 13.28 43.05 11.49 1.64 26.21
Caa 0.01 0.08 0.23 0.93 3.52 16.80 18.67 2.93 56.84
Ca-C 0.00 0.02 0.06 0.31 1.39 5.89 6.78 2.40 83.15
Default 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 100.00
One-Month Rating Transition
Matrix (calculated from one-year transitions)
Table 18.3 page 401
Initial Rating at month end

Aaa 99.16 0.82 0.02 0.00 0.00 0.00 0.00 0.00 0.00
Aa 0.09 99.12 0.77 0.01 0.00 0.00 0.00 0.00 0.00
A 0.00 0.26 99.18 0.51 0.04 0.01 0.00 0.00 0.00
Baa 0.00 0.01 0.44 99.05 0.41 0.06 0.02 0.00 0.01
Ba 0.00 0.00 0.02 0.59 98.46 0.79 0.03 0.01 0.09
B 0.00 0.00 0.01 0.02 0.53 98.32 0.70 0.07 0.36
Caa 0.00 0.00 0.00 0.01 0.02 1.01 96.79 0.67 1.48
Ca-C 0.00 0.00 0.00 0.00 0.04 0.28 1.53 93.23 4.92
Default 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 100.00
Credit VaR (page 321)
l Can be defined analogously to Market
Risk VaR
l A one year credit VaR with a 99.9%
confidence is the loss level that we are
99.9% confident will not be exceeded over
one year
Vasicek’s Model (Equation 18.1, page 402)
l For a large portfolio of loans, each of which has
a probability of Q(T) of defaulting by time T the
default rate that will not be exceeded at the X%
confidence level is
ìï N -1 [Q(T )] + r N -1 ( X ) üï
Ní ý
ïî 1- r ïþ
l Where r is the Gaussian copula correlation
VaR Model (Equation 18.2, page 402)
VaR = å WCDR i (T , X ) ´ EADi ´ LGDi

i
Credit Risk Plus (Section 18.3, page 403)
This calculates a loss probability distribution using a
Monte Carlo simulation where the steps are:
l Sample overall default rate
l Sample probability of default for each counterparty
category
l Sample number of losses for each counterparty category
l Sample size of loss for each default
l Calculate total loss from defaults
This is repeated many times to calculate a probability
distribution for the total loss
CreditMetrics (Section 18.4, page 405)
l Calculates credit VaR by considering
possible rating transitions
l A Gaussian copula model is used to define
the correlation between the ratings
transitions of different companies
The Copula Model : xA and xB are sampled from
correlated standard normals
Credit Risk in the Trading Book: The
Specific Risk Charge
l To calculate the specific risk charge,

banks must model credit spreads to
calculate a 10-day 99% VaR
l Alternatives:
l Historical simulation
l 10-day transition matrix
Credit Risk in the Trading Book:
Incremental Risk Charge
l Banks must calculate a one year 99.9%
VaR
l This is to ensure that capital is similar to
the capital that would be charged if the
instrument were in the banking book
l They are allowed to make a constant level
of risk assumption (minimum liquidity
horizon is three months)
Constant Level of Risk
Assumption
l Suppose a bank has a BBB bond and
uses a liquidity horizon of 3 months
l At the end of each month period the bond,
if it has deteriorated is assumed to be sold
and replaced with a new BBB bond
l The one-year loss is then replaced by four
three-month losses
Counterparty Credit
Risk in Derivatives
Chapter 17
Clearing Arrangements for OTC
Derivatives (Figure 17.2, page 381)
l Bilateral clearing: usually governed by an ISDA
Master agreement
l Central clearing: a central clearing party (CCP)
stands between the two sides
Regulations
l Standard OTC transactions must be cleared through

a CCP (some exceptions)
l Nonstandard OTC transactions continue to be
cleared bilaterally, but with much higher capital
charges
Central Clearing: Role of CCP
(Figure 17.1, page 380)
Key Questions
l How many CCPs
l Will there be interoperability?
l Will benefits of netting increase or
decrease?
Simple Example: 3 market
participants; 2 product types (Figure
17.3, page 382)
Bilateral Clearing:The ISDA
Master Agreement
l One important feature is netting
l This states that all transactions with the
counterparty are considered to be a single
transaction in the case of early termination
and for the purposes of posting collateral
Events of Default and Early
Termination
l Declaration of bankruptcy
l Failure to make payments on derivatives
as they are due
l Failure to provide collateral when it is due
l The non-defaulting party has the right to
declare an early termination event a few
days after an event of default if there has
been no resolution of outstanding issues
Collateral Arrangements
The credit support annex (CSA) of the
ISDA Master Agreement specifies
l Threshold
l Independent Amount
l Minimum Transfer Amount Eligible
Securities and Currencies
l Haircuts
Possible ISDA Clause: Downgrade
Triggers
l Specify that in the event of a downgrade
the counterparty has certain rights.
l It might specify that the counterparty can
terminate outstanding transactions or that
the counterparty can require collateral
l In AIG’s case counterparties could require
collateral in the event of a downgrade
below AA. This necessitated a huge
government bailout.
CVA
l Credit value adjustment (CVA) is the

amount by which a dealer must reduce the
value of transactions because of
counterparty default risk
The CVA Calculation
Time 0 t1 t2 t3 t4 ……………… tn=T
Default probability q1 q2 q3 q4 ……………… qn
PV of net exposure v1 v2 v3 v4 ……………… vn
n
CVA = (1 - R)å qi vi where R is the recovery rate
i =1
Copyright 2011 © John Hull, Joseph L. Rotman School of Management, University of Toronto 93
CVA continued
l The default probabilities (i.e., the qi’s) are

calculated from credit spreads
l The PV of the net exposure is calculated using
Monte Carlo simulation. Random paths are
chosen for all the market variables underlying the
derivatives and the net exposure is calculated at
the mid point of each time interval. (These are
the “default times”
l The vi is the present value of the average net
exposure at the ith default time
Calculation of Net Exposure
l If no collateralization the net exposure is the maximum of

the value of the derivatives and zero
l If collateral is posted we assume that there is a “cure
period” (= c days) immediately before a default during
which collateral is not posted.
l On each Monte Carlo trial we must calculate the value of
the portfolio c days before each default time
l This determines the collateral available at the default
time
Incremental CVA
l Results from Monte Carlo are stored so
that the incremental impact of a new trade
can be calculated without simulating all the
other trades.
CVA Risk
l The CVA for a counterparty can be

regarded as a complex derivative
l Increasingly dealers are managing it like
any other derivative
l Two sources of risk:
l Changes in counterparty spreads
l Changes in market variables underlying the
portfolio
Basel III (2010)
l Basel III requires CVA risk arising from a
parallel shift in the term structure of
counterparty credit spreads to be included
in the calculation of capital for market risk
l It does not require banks to include CVA
risk arising from the underlying market
variables
Wrong Way/Right Way Risk
l Simplest assumption is that probability of
default qi is independent of net exposure
vi.
l Wrong-way risk occurs when qi is
positively dependent on vi
l Right-way risk occurs when qi is
negatively dependent on vi
Examples
l Wrong-way risk typically occurs when
l Counterparty is selling credit protection
l Counterparty is a hedge fund taking a big
speculative positions
l Right-way risk typically occurs when
l Counterparty is buying credit protection
l Counterparty is partially hedging a major
exposure
Problems in Estimating Wrong
Way/Right Way Risk
l Knowing trades counterparty is doing with other dealers
l Knowing how different market variables influence the
fortunes of the counterparty
l Do counterparties become more likely to default when
interest rates are high or low? The evidence is mixed and
so we do not know whether receiving or paying fixed
generates wrong way risk
l Even when there appears to be right-way risk liquidity
problems can lead to a company being unable to post
collateral (e.g Ashanti)
Allowing for Wrong-Way risk
l One common approach is to use the
“alpha” multiplier to increase the v’s
l Estimates of 1.07 to 1.1 for alpha obtained
from banks
l Basel II sets alpha equal to 1.4 or allows
banks to use their own models, with a floor
of 1.2
DVA (more recent and more
controversial)
l Debit (or debt) value adjustment (DVA) is an
estimate of the cost to the counterparty of a
default by the dealer
l Same formulas apply except that v is
counterparty’s exposure to dealer and q is
dealer’s probability of default
l Accounting value of transactions with
counterparty = No default value – CVA +
DVA
DVA continued
l What happens to the reported value of
transactions as dealer’s credit spread
increases?
Expected Exposure on Pair of Offsetting
Interest Rate Swaps and a Pair of
Offsetting Currency Swaps (No collateral)
(Figure 17.2, page 317-318)
Exposure
Currency
swaps
Interest Rate
Swaps
Maturity
Interest Rate vs Currency Swaps
l The qi’s are the same for both
l The vi’s for an interest rate swap are on
average much less than the vi’s for a
currency swap
l The expected cost of defaults on a
currency swap is therefore greater.
Simple Example: Single transaction
always has positive value to dealer
CVA has the effect of multiplying value of

transaction by e-sT where s is spread
between T-year bond issued by
counterparty and risk-free T-year bond
Example 17.1 (page 391)
l A 2-year option sold by a counterparty to
the dealer has a Black-Scholes value of $3
l Assume a 2 year zero coupon bond issued
by the counterparty has a yield of 1.5%
greater than the risk free rate
l If there is no collateral and there are no
other transactions between the parties,
value of option is 3e-0.015×2=2.91
Dealer Has Single Long Forward
with Counterparty (page 392)
For a long forward contract that matures at time T the
present value of the exposure at time ti is
vi = e - rT [F0 N (d1 ) - KN (d 2 )]
where
ln( F0 / K ) + s 2ti / 2
d1,i = d 2,i = d1,i - s ti
s ti
F0 is the forward price today, K is the delivery price, s is
the volatility of the forward price, T is the time to maturity
of the forward contract, and r is the risk-free rate
Correlations and
Copulas
Chapter 11
Correlation and Covariance
l The coefficient of correlation between two
variables V1 and V2 is defined as
E (V1V2 ) - E (V1 ) E (V2 )

SD (V1 ) SD (V2 )
l The covariance is
E(V1V2)−E(V1 )E(V2)
Independence
l V1 and V2 are independent if the
knowledge of one does not affect the
probability distribution for the other
f (V2 V1 = x ) = f (V2 )
where f(.) denotes the probability density
function
Independence is Not the Same as
Zero Correlation
l Suppose V1 = –1, 0, or +1 (equally
likely)
l If V1 = -1 or V1 = +1 then V2 = 1
l If V1 = 0 then V2 = 0
V2 is clearly dependent on V1 (and vice
versa) but the coefficient of correlation
is zero
Types of Dependence (Figure 11.1, page 235)
E(Y) E(Y)
X X
(a) (b)
E(Y)
(c)
Monitoring Correlation Between
Two Variables X and Y
Define xi=(Xi−Xi-1)/Xi-1 and yi=(Yi−Yi-1)/Yi-1
Also
varx,n: daily variance of X calculated on day n-1
vary,n: daily variance of Y calculated on day n-1
covn: covariance calculated on day n-1
The correlation is
cov n
varx ,n vary ,n
Covariance
l The covariance on day n is
E(xnyn)−E(xn)E(yn)
l It is usually approximated as E(xnyn)
Monitoring Correlation continued
EWMA:
cov n = l cov n -1 + (1 - l) xn -1 yn -1
GARCH(1,1)
cov n = w + axn -1 yn -1 + b cov n -1
Positive Finite Definite Condition
A variance-covariance matrix, W, is
internally consistent if the positive semi-
definite condition
wTWw ≥ 0
holds for all vectors w
Example
The variance covariance matrix
æ 1 0 0.9ö
ç ÷
ç 0 1 0.9÷
ç ÷
è 0.9 0.9 1ø
is not internally consistent
V1 and V2 Bivariate Normal
l Conditional on the value of V1, V2 is normal with
mean
V1 - µ1
µ 2 + rs 2
s1
and standard deviation s 2 1 - r 2 where µ1,, µ2, s1,

and s2 are the unconditional means and SDs of
V1 and V2 and r is the coefficient of correlation
between V1 and V2
Multivariate Normal Distribution
l Fairly easy to handle

l A variance-covariance matrix defines
the variances of and correlations
between variables
l To be internally consistent a variance-
covariance matrix must be positive
semidefinite
Generating Random Samples for
Monte Carlo Simulation (pages 239-240)
l =NORMSINV(RAND()) gives a random

sample from a normal distribution in
Excel
l For a multivariate normal distribution a
method known as Cholesky’s
decomposition can be used to generate
random samples
Factor Models (page 240)
l When there are N variables, Vi (i = 1,
2,..N), in a multivariate normal distribution
there are N(N−1)/2 correlations
l We can reduce the number of correlation
parameters that have to be estimated with
a factor model
One-Factor Model continued
l If Ui have standard normal distributions
we can set
U i = ai F + 1 - ai2 Z i
where the common factor F and the
idiosyncratic component Zi have
independent standard normal
distributions
l Correlation between Ui and Uj is ai aj
Gaussian Copula Models:
Creating a correlation structure for variables that are not
normally distributed
l Suppose we wish to define a correlation structure between

two variable V1 and V2 that do not have normal distributions
l We transform the variable V1 to a new variable U1 that has a
standard normal distribution on a “percentile-to-percentile”
basis.
l We transform the variable V2 to a new variable U2 that has a
standard normal distribution on a “percentile-to-percentile”
basis.
l U1 and U2 are assumed to have a bivariate normal
distribution
Risk Management and Financial Institutions 3e, Chapter 11, 125

Copyright © John C. Hull 2012
The Correlation Structure Between the V’s is
Defined by that Between the U’s
-0.2 0 0.2 0.4 0.6 0.8 1 1.2 -0.2 0 0.2 0.4 0.6 0.8 1 1.2
V1 V2
One-to-one
mappings
-6 -4 -2 0 2 4 6 -6 -4 -2 0 2 4 6
U2
U1
Correlation
Assumption
Example (page 241)
V1 V2
V1 Mapping to U1
V1 Percentile U1
0.2 20 -0.84
0.4 55 0.13
0.6 80 0.84
0.8 95 1.64
V2 Mapping to U2
V2 Percentile U2
0.2 8 −1.41
0.4 32 −0.47
0.6 68 0.47
0.8 92 1.41
Example of Calculation of Joint
Cumulative Distribution
l Probability that V1 and V2 are both less

than 0.2 is the probability that U1 < −0.84
and U2 < −1.41
l When copula correlation is 0.5 this is
M( −0.84, −1.41, 0.5) = 0.043
where M is the cumulative distribution
function for the bivariate normal
distribution
Other Copulas
l Instead of a bivariate normal distribution
for U1 and U2 we can assume any other
joint distribution
l One possibility is the bivariate Student t
distribution
5000 Random Samples from the
Bivariate Normal
0
-5 -4 -3 -2 -1 0 1 2 3 4 5
-1
-2
-3
-4
-5
5000 Random Samples from the
Bivariate Student t
10
0
-10 -5 0 5 10
-5
-10
Multivariate Gaussian Copula
l We can similarly define a correlation
structure between V1, V2,…Vn
l We transform each variable Vi to a new
variable Ui that has a standard normal
distribution on a “percentile-to-percentile”
basis.
l The U’s are assumed to have a
multivariate normal distribution
Factor Copula Model
In a factor copula model the correlation
structure between the U’s is generated by
assuming one or more factors.
Credit Default Correlation
l The credit default correlation between two
companies is a measure of their tendency
to default at about the same time
l Default correlation is important in risk
management when analyzing the benefits
of credit risk diversification
l It is also important in the valuation of some
credit derivatives
Model for Loan Portfolio
l We map the time to default for company i, Ti, to a
new variable Ui and assume
U i = aF + 1- a Z i 2
l Where F and the Zi have independent standard

normal distributions
l The copula correlation is r=a2
Vasicek’s Model for Loan Portfolio
l Suppose the Bank has a large no. of Loans where PD is 1% but will
change in each year. Suppose Ti is the time firm defaults. All firms has
the same Cum. Prob. Dist, for the time to default Qi.
l The Gaussian copula method can be used to define correlation structure
between Ti’s.
l We map the time to default for company i, Ti, to a new variable Ui N(0,1)
and assume
U i = aF + 1- a Z i 2
l Where F and the Zi have independent standard normal distributions

l The copula correlation is r=a2
Vasicek’s Model for Loan Portfolio
l
Analysis
l To analyze the model we
l Calculate the probability that, conditional on the value
of F, Ui is less than some value U
l This is the same as the probability that Ti is less that T
where T and U are the same percentiles of their
distributions
This leads to
é N -1 [PD] - r F ù
Prob(Ti < T F ) = N ê ú
êë 1- r úû
where PD is the probability of default in time T
The Model continued
l The X% worst case value of F is N-1(X)
l The worst case default rate during time T with a
confidence level of X is therefore
æ N -1[Q(T )] + r N -1 ( X ) ö
WCDR(T,X) = N ç ÷
ç 1- r ÷
è ø
l The VaR for this time horizon and confidence limit
is
VaR (T , X ) = L ´ LGD ´ WCDR (T , X )
where L is loan principal and LGD is loss given default
Gordy’s Result
l In a large portfolio of M loans where each
loan is small in relation to the size of the
portfolio it is approximately true that
M
VaR (T , X ) = å Li ´ LGDi ´ WCDR i (T , X )
i =1
Estimating PD and r
l We can use data on default rates in
conjunction with maximum likelihood
methods
l The probability density function for the
default rate is
ì é ü
1 - r N (DR ) - N (PD) ù ï
2
1- r ï 1 ê -1 æ -1 -1
ö
g (DR ) = expí ( N (DR )) - çç
2
÷ úý
÷ ú
r ïî 2 êë è r ø û ïþ
Non-Normal Distributions
Elements of
Financial Risk Management
Chapter 6
Peter Christoffersen
144
Elements of Financial
Risk Management
Overview
l Third part of the Stepwise Distribution Modeling
(SDM) approach: accounting for conditional
nonnormality in portfolio returns.
l Returns are conditionally normal if the
dynamically standardized returns are normally
distributed.
l Fig.6.1 illustrates how histograms from
standardized returns typically do not conform to
normal density
l The top panel shows the histogram of the raw
returns superimposed on the normal distribution
145Elements of Financial
and
Risk the bottom panel shows the histogram of the
Management
Figure 6.1: Histogram of Daily S&P 500
Returns and Histogram of GARCH Shocks
30%
Frequency (%)
25%
20%
15%
10%
5%
0%
-0,08 -0,06 -0,05 -0,03 -0,02 0,00 0,01 0,03 0,05 0,06 0,08
Return
30%
25%
Frequency (%)
20%
15%
10%
5%
0%
-6,00 -4,00 -2,00 0,00 2,00 4,00 6,00
Standardized Return
Risk Management
Learning Objectives
l We introduce the quantile-quantile (QQ) plot,
which is a graphical tool better at describing
tails of distributions than the histogram.
l We define the Filtered Historical Simulation
approach which combines GARCH with
historical simulation.
l We introduce the simple Cornish-Fisher
approximation to VaR in non-normal
distributions.
l We consider the standardized Student’s t

Risk Management
Learning Objectives
l We extend the Student’s t distribution to a
more flexible asymmetric version.
l We consider extreme value theory for
modeling the tail of the conditional
distribution
l For each of these methods we will consider
the Value-at-Risk and the expected shortfall
formulas
Risk Management
Visualising Non-normality Using
QQ Plots
l Consider a portfolio of n assets with Ni,t
units or shares of asset i then the value of
the portfolio today is
• Yesterday’s portfolio value would be
• The log return can now be defined as
Risk Management
QQ Plots
l Allowing for a dynamic variance model we can

say
• where sPF,t is the conditional volatility forecast

• So far, we have relied on setting D(0,1) to N(0,1), but we
now want to assess the problems of the normality
assumption
Risk Management
QQ Plots
l QQ (Quantile-Quantile) plot: Plot the
quantiles of the calculated returns against
the quantiles of the normal distribution.
l Systematic deviations from the 45 degree
angle signals that the returns are not well
described by normal distribution.
l QQ Plots are particularly relevant for risk
managers who care about VaR, which itself
is essentially a quantile.
Risk Management
QQ Plots
l 1) Sort all standardized returns in ascending order
and call them zi
l 2) Calculate the empirical probability of getting a
value below the value i as (i-.5)/T
l 3) Calculate the standard normal quantiles as
l 4) Finally draw scatter plot
• If the data were normally distributed, then the scatterplot

should conform to the 45-degree line.
Risk Management
Figure 6.2: QQ Plot of Daily S&P 500 Returns
8
4
Return Quantile
0
-8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8
-2
-4
-6
-8
Normal Quantile
Risk Management
Figure 6.2: QQ Plot of Daily S&P 500 GARCH
Shocks
8
6
GARCH Shock Quantile
0
-8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8
-2
-4
-6
-8
Normal Quantile
Risk Management
Filtered Historical Simulation Approach
l We have seen the pros and cons of both
data-based and model-based approaches.
l The Filtered Historical Simulation (FHS)
attempts to combine the best of the model-
based with the best of the model-free
approaches in a very intuitive fashion.
l FHS combines model-based methods of
variance with model-free method of
distribution in the following fashion.
Risk Management
l Assume we have estimated a GARCH-type
model of our portfolio variance.
l Although we are comfortable with our
variance model, we are not comfortable
making a specific distributional
~ ( ) assumption
t d
about the standardized returns, such as a
Normal or a distribution.
l Instead we would like the past returns data
to tell us about the distribution directly
without making further assumptions.
Risk Management
l To fix ideas, consider again the simple
example of a GARCH(1,1) model
• where
• Given a sequence of past returns,

we can estimate the GARCH model.
• Next we calculate past standardized returns from
the observed returns and from the estimated
standard deviations as
Risk Management
l We will refer to the set of standardized returns as
l To calculate the 1-day VaR using the percentile

of the database of standardized residuals
• Expected shortfall (ES) for the 1-day horizon is
• The ES is calculated from the historical shocks via
Risk Management
l where the indicator function 1(*) returns a 1
if the argument is true and zero if not
l FHS can generate large losses in the

forecast period even without having
observed a large loss in the recorded past
returns
l FHS deserves serious consideration by any

risk management team
Risk Management
The Cornish-Fisher Approximation to
VaR
l We consider a simple alternative way of
calculating Value at Risk, which has certain
advantages:
l First, it allows for skewness and excess
kurtosis.
l Second, it is easily calculated from the
empirical skewness and excess kurtosis
estimates from the standardized returns.
l Third, it can be viewed as an approximation
to the VaR from a wide range of
conditionally
160Elements of Financial nonnormal distributions.
Risk Management
The Cornish-Fisher Approximation
to VaR
• Standardized portfolio returns is defined by
• where D(0,1) denotes a distribution with a mean

equal to 0 and a variance equal to 1
• i.i.d. denotes independently and identically
distributed
• The Cornish-Fisher VaR with coverage rate, p, can
be calculated as
Risk Management
to VaR
• Where
• Where is the skewness and is the excess kurtosis

of the standardized returns
• If we have neither skewness nor excess kurtosis so that
. , then we get the quantile of the normal
distribution
Risk Management
to VaR
• Consider now for example the one percent VaR,
where
• Allowing for skewness and kurtosis we can

calculate the Cornish-Fisher 1% quantile as
• and the portfolio VaR can be calculated as
Risk Management
The Cornish-Fisher Approximation to
VaR
l Thus, for example, if skewness equals –1 and

excess kurtosis equals 4, then we get
• which is much higher than the VaR number from a

normal distribution, which equals 2.33sPF,t+1
Risk Management
to VaR
• The expected shortfall can be derived as
• Where
Risk Management
to VaR
• Recall that the ES for the normal case is
• Which can be derived by setting in the

equation for .
• The CF approach is easy to implement and we avoid

having to make an assumption about exactly which
distribution fits the data best
Risk Management
The Standardized t distribution
l The Student’s t distribution is defined by
• G(*) notation refers to the gamma function

• the distribution has only one parameter d
• In the Student’s t distribution we have the following
first two moments
Risk Management
l Define Z by standardizing x so that,
• The Standardized t distribution, , is then

defined as
• where
Risk Management
l In standardized t distribution random variable z
has mean equal to zero and a variance equal to 1
l Note also that the parameter d must be larger than
two for standardized distribution to be well
defined
l In distribution, the random variable, z, is
taken to a power, rather than an exponential,
which is the case in the standard normal
• distribution
The power function
where driven by d will allow for
distribution to have fatter tails than the normal
Risk Management
l The distribution is symmetric around zero,
and the mean µ, variance s2, skewness z1, and
excess kurtosis z2 of the distribution are
Risk Management
l Note that d must be higher than 4 for the kurtosis
to be well defined.
l Note also that for large values of d the

distribution will have an excess kurtosis of zero,
and we can show that it converges to the standard
normal distribution as d goes to infinity.
l For values of d above 50, the distribution is

difficult to distinguish from the standard normal
distribution.
Risk Management
Maximum Likelihood Estimation
l We can combine dynamic volatility model such
as GARCH with the standardized t distribution to
specify our model portfolio returns as
• If we ignore the fact that variance is estimated with error,

we can treat standardized return as a regular random
variable, calculated as
Risk Management
l The d parameter can then be chosen to
maximize the log likelihood function
Risk Management
l If we want to jointly maximize over the
parameter d and we should adjust the
distribution to take into account the
variance
• We can then maximize
Risk Management
l As a simple univariate example of the difference
between QMLE and MLE consider the
GARCH(1,1)- model with leverage:
• We can estimate all the parameters in one

step using lnL2 from before, which would correspond to
exact MLE.
• Exact MLE is clearly feasible as the total number of
parameters is only five
Risk Management
An Easy Estimate of d
l There is a simple alternative estimation
procedure to the QMLE estimation
procedure above.
l If the conditional variance model has
already been estimated, then we are only
estimating one parameter, namely d.
l The simple closed-form relationship
between d and the excess kurtosis z2
suggests calculating z2, from the zt variable
and calculating d from
Risk Management
Calculating VaR and ES
l Having estimated d1 and d2 we can calculate
the VaR of the portfolio return
• as
• Where is the pth percentile of the

asymmetric t distribution.
Risk Management
Extreme Value Theory (EVT)
l Typically, the biggest risks to a portfolio is the
sudden occurrence of a single large negative
return.
l Having an as precise as possible knowledge of the
probabilities of such extremes is therefore at the
essence of financial risk management.
l Consequently, risk managers should focus
attention explicitly on modeling the tails of the
returns distribution.
l Fortunately, a branch of statistics is devoted
178exactly to the modeling of these extreme values.
Elements of Financial
Risk Management
l The central result in EVTstates that the extreme
tail of a wide range of distributions can
approximately be described by a relatively simple
distribution, the so-called Generalized Pareto
distribution.
l Virtually all results in Extreme Value Theory
assumes that returns are i.i.d. and are therefore not
very useful unless modified to the asset return
environment.
l Asset returns appear to approach normality at long
horizons, thus EVT is more important at short
horizons,
179Elements of Financialsuch as daily.
Risk Management
l Unfortunately, the i.i.d assumption is the
least appropriate at short horizons due to the
time-varying variance patterns.
l We therefore need to get rid of the variance
dynamics before applying EVT.
l Consider therefore again the standardized
portfolio returns
Risk Management
The Distribution of Extremes
l Define a threshold value u on the horizontal axis
of the histogram in Figure 6.1
l As you let the threshold u go to infinity, the
distribution of observations beyond the threshold
(y) converge to the Generalized Pareto
Distribution, where
• With b>0 and y≥u. Tail-index parameter x controls the

shape of the distribution tail
Risk Management
Estimating Tail Index Parameter, x
l If we assume that the tail parameter, x, is strictly
positive, then we can use the Hill estimator to
approximate the GPD distribution
• for y > u and x> 0. Recall now the definition of a

conditional distribution
• Note that from the definition of F(y) we have
Risk Management
• We can get the density function of y from F(y):
• The likelihood function for all observations yi larger than

the threshold, u,
• where Tu is the number of observations y larger than u
Risk Management
l The log-likelihood function is therefore
• Taking the derivative with respect to x and setting it to zero

yields the Hill estimator of tail index parameter
Risk Management
l We can estimate the c parameter by ensuring that
the fraction of observations beyond the threshold
is accurately captured by the density as in
• Solving this equation for c yields the estimate
• Cumulative density function for observations beyond u is
Risk Management
l Notice that our estimates are available in closed
form
l So far, we have implicitly referred to extreme

returns as being large gains. As risk managers,
we are more interested in extreme negative
returns corresponding to large losses
l We can simply do the EVT analysis on the

negative of returns (i.e. the losses) instead of
returns
186Elements themselves.
of Financial
Risk Management
Choosing the Threshold, u
l When choosing u we must balance two evils: bias
and variance.
l If u is set too large, then only very few
observations are left in the tail and the estimate of
the tail parameter, x, will be very noisy.
l If on the other hand u is set too small, then the
data to the right of the threshold does not conform
sufficiently well to the Generalized Pareto
Distribution to generate unbiased estimates of x.
Risk Management
Choosing the Threshold, u
l Simulation studies have shown that in typical data
sets with daily asset returns, a good rule of thumb
is to set the threshold so as to keep the largest 50
observations for estimating x
l We set Tu = 50.
l Visually gauging the QQ plot can provide useful

guidance as well.
l Only those observations in the tail that are clearly
deviating from the 45-degree line indicating the
normal distribution
should be used in the
estimation
Risk Management of the tail index parameter, x
Constructing the QQ Plot from EVT
l Define y to be a standardized loss
• The first step is to estimate x and c from the losses, yi,

using the Hill estimator
• Next, we need to compute the inverse cumulative

distribution function, which gives us the quantiles
Risk Management
l We now set the estimated cumulative probability
function equal to 1-p so that there is only a p
probability of getting a standardized loss worse
than the quantile, F-11-p
• From the definition of F(*), we can solve for the

quantile to get
Risk Management
l We are now ready to construct the QQ plot

from EVT using the relationship
• where yi is the ith sorted standardized loss
Risk Management
Figure 6.7: QQ Plot of Daily S&P 500 Tail Shocks
against the EVT Distribution
8
4
Empirical Quantiles
0
-8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8
-2
-4
-6
-8
EVT Quantile
Risk Management
Calculating VaR and ES from EVT
l We are ultimately interested not in QQ plots but
rather in portfolio risk measures such as VaR.
l Using the loss quantile F-11-p defined above by
• The VaR from the EVT combined with the variance

model is now easily calculated as
Risk Management
Calculating VaR and ES from
EVT
l We usually calculate the VaR taking F p to be the
-1
pth quantile from the standardized return so that
• But we now take F-11-p to be the (1-p)th quantile of the

standardized loss so that
• The expected shortfall can be computed using
• Where when x<1

Risk Management
EVT
l In general, the ratio of ES to VaR for fat-tailed
distribution will be higher than that of the normal.
l When using the Hill approximation of the EVT tail
the previous formulas for VaR and ES show that
we have a particularly simple relationship, namely
• so that for fat-tailed distributions where x > 0, the fatter

the tail, the larger the ratio of ES to VaR:
Risk Management
EVT
l The preceding formula shows that when x= 0.5
then the ES to VaR ratio is 2
l Thus even though the 1% VaR is the same in the

two distributions by construction, the ES measure
reveals the differences in the risk profiles of the
two distributions, which arises from one being fat-
tailed
l The VaR does not reveal this difference unless the

VaR is reported for several extreme coverage

Risk Management
Figure 6.8: Tail Shapes of the Normal
Distribution (blue) and EVT (red)
0,030
0,025
Tail Distribution Shape
0,020 1% VaR for Normal

Distribution and for EVT
0,015
0,010
0,005
0,000
-5,00 -4,75 -4,50 -4,25 -4,00 -3,75 -3,50 -3,25 -3,00 -2,75 -2,50
Return
Risk Management

GRM - Day 3.1

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

GRM - Day 3.1

Uploaded by

Copyright:

Available Formats

Scenario Analysis and

l Will regulators provide their own scenarios to

Historical data provided by rating agencies

Aa 0.021 0.059 0.103 0.184 0.273 0.443 0.619

A 0.055 0.177 0.362 0.549 0.756 1.239 2.136

Baa 0.181 0.510 0.933 1.427 1.953 3.031 4.904

Ba 1.157 3.191 5.596 8.146 10.453 14.440 20.101

B 4.465 10.432 16.344 21.510 26.173 34.721 44.573

Caa 18.163 30.204 39.709 47.317 53.768 61.181 72.384

l The table shows the probability of

l For a company that starts with a poor

% LGD = [100% - % recovery rate ]

Class Ave Rec Rate (%)

l R2 of regression is about 0.5

l Buyer of the instrument acquires protection from the

90 bps per year

Recovery rate, R, is the ratio of the value of the bond issued

l Allows credit risks to be traded in the

l Portfolio consisting of a 5-year par yield

l Suppose that a five year corporate bond pays a coupon

1.5 Q 40 105.97 65.97 0.9277 61.20Q

2.5 Q 40 105.17 65.17 0.8825 57.52Q

3.5 Q 40 104.34 64.34 0.8395 54.01Q

4.5 Q 40 103.46 63.46 0.7985 50.67Q

l Calculate 7-year hazard rates from the

Rating Bond Yield Spread of risk-free Spread to Extra Risk

l Corporate bonds are relatively illiquid

l Merton’s model regards the equity as an

An option pricing model enables the value

This equation together with the option pricing

l Same approach can be used

How much of the original portfolio of subprime

Losses on Losses on Losses on Losses on Losses on

l ABSs and ABS CDOs were complex inter-

Initial Rating at year end

Initial Rating at end

Initial Rating at month end

l Where r is the Gaussian copula correlation

VaR = å WCDR i (T , X ) ´ EADi ´ LGDi

l To calculate the specific risk charge,

l Standard OTC transactions must be cleared through

l Credit value adjustment (CVA) is the

Time 0 t1 t2 t3 t4 ……………… tn=T

Default probability q1 q2 q3 q4 ……………… qn

PV of net exposure v1 v2 v3 v4 ……………… vn

l The default probabilities (i.e., the qi’s) are

l If no collateralization the net exposure is the maximum of

l The CVA for a counterparty can be

CVA has the effect of multiplying value of

E (V1V2 ) - E (V1 ) E (V2 )

is not internally consistent

and standard deviation s 2 1 - r 2 where µ1,, µ2, s1,

l Fairly easy to handle

l =NORMSINV(RAND()) gives a random

l Suppose we wish to define a correlation structure between

Risk Management and Financial Institutions 3e, Chapter 11, 125

l Probability that V1 and V2 are both less

l Where F and the Zi have independent standard

l Where F and the Zi have independent standard normal distributions

where L is loan principal and LGD is loss given default

l We consider the standardized Student’s t

• Yesterday’s portfolio value would be

• The log return can now be defined as

l Allowing for a dynamic variance model we can