Financial Risk: Credit Risk, Lecture 1

Alexander Herbertsson

Centre For Finance/Department of Economics

School of Economics, Business and Law, University of Gothenburg

E-mail: [email protected]
or
[email protected]

Financial Risk, Chalmers University of Technology,

Göteborg
Sweden

April 25, 2017

Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 1 / 26
 Main goal of lectures and content of today’s lecture

The main goal of coming three lectures is to study the loss

distribution for a credit portfolio

The loss distribution is used to compute risk measures such as

Value-at-Risk etc.

Today’s lecture
Short discussion of the important components of credit risk

Study different static portfolio credit risk models.

Discussion of the binomial loss model

Discussion of the mixed binomial loss model

Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 2 / 26
 Definition of Credit Risk
Credit risk
− the risk that an obligor does not honor his payments

Example of an obligor:
A company that have borrowed money from a bank
A company that has issued bonds.
A household that have borrowed money from a bank, to buy a house
A bank that has entered into a bilateral financial contract (e.g an
interest rate swap) with another bank.

Example of defaults are

A company goes bankrupt.
As company fails to pay a coupon on time, for some of its issued
bonds.
A household fails to pay amortization or interest rate on their loan.
Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 3 / 26
 Credit Risk
Credit risk can be decomposed into:
arrival risk, the risk connected to whether or not a default will
happen in a given time-period, for a obligor

timing risk, the risk connected to the uncertainness of the exact

time-point of the arrival risk (will not be studied in this course)

recovery risk. This is the risk connected to the size of the actual loss
if default occurs (will not be studied in this course, we let the
recovery be fixed)

default dependency risk, the risk that several obligors jointly

defaults during some specific time period. This is one of the most
crucial risk factors that has to be considered in a credit portfolio
framework.
The coming three lectures focuses only on default dependency risk.
Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 4 / 26
 Portfolio Credit Risk is important

Portfolio credit risk models differ greatly depending on what types of

portfolios, and what type of questions that should be considered. For
example,

models with respect to risk management, such as credit Value-at-Risk

(VaR) and expected shortfall (ES)
models with respect to valuation of portfolio credit derivatives, such as
CDO´s and basket default swaps

In both cases we need to consider default dependency risk, but....

...in risk management modelling (e.g. VaR, ES), the timing risk is ignored,
and one often talk about static credit portfolio models,

...while, when pricing credit derivatives, timing risk must be carefully

modeled (not treated here)

The coming three lectures focuses only on static credit portfolio models,
Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 5 / 26
 Literature

The slides for the coming three lectures are rather self-contained, but more details
on certain topics can be found in the lecture notes.

The content of the lecture today and the next lecture is partly based on materials
presented in:
”Quantitative Risk Management” by McNeil A., Frey, R. and Embrechts, P.
(Princeton University Press)

”Credit Risk Modeling: Theory and Applications” by Lando, D . (Princeton

University Press)

”Risk and portfolio analysis - principles and methods” by Hult, Lindskog,

Hammerlid and Rehn. (Springer)

Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 6 / 26
 Static Models for homogeneous credit portfolios

Today we will consider the following static modes for a homogeneous

credit portfolio:

The binomial model

The mixed binomial model

To understand mixed binomial models, we give a short introduction of

conditional expectations

In the next two lectures we will

study three different mixed binomial models.
discuss Value-at-Risk and Expected shortfall in a mixed binomial
models.
Correlations etc in mixed binomial models.

Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 7 / 26
 The binomial model for independent defaults
Consider a homogeneous credit portfolio model with m obligors where each
obligor can default up to fixed time T , and have the same constant credit loss ℓ.

Let Xi be a random variable such that


1 if obligor i defaults before time T
Xi = (1)
0 otherwise, i.e. if obligor i survives up to time T

We assume that X1 , X2 , . . . Xm are i.i.d, that is they are independent with

identical distribution. Furthermore P [Xi = 1] = p so P [Xi = 0] = 1 − p.

The total credit loss in the portfolio at time T , called Lm , is then given by
m
X m
X m
X
Lm = ℓXi = ℓ Xi = ℓNm where Nm = Xi
i =1 i =1 i =1
thus, Nm is the number of defaults in the portfolio up to time T .

Since ℓ is a constant, we have P [Lm = kℓ] = P [Nm = k], so it is enough to

study the distribution of Nm .
Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 8 / 26
 The binomial model for independent defaults, cont.
Since X1 , X2 , . . . Xm are i.i.d with P [Xi = 1] = p we see that Nm = m
P
i =1 Xi
is binomially distributed with parameters m and p, i.e. Nm ∼ Bin(m, p).

Hence, we have  
m k
P [Nm = k] = p (1 − p)m−k
k
Pm
Recalling the binomial theorem (a + b)m = k=0 m

k m−k
k a b we see that
m m  
X X m k
P [Nm = k] = p (1 − p)m−k = (p + (1 − p))m = 1
k
k=0 k=0

proving that Bin(m, p) is a distribution.

Furthermore, E [Nm ] = mp since

" m # m
X X
E [Nm ] = E Xi = E [Xi ] = mp.
i =1 i =1

Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 9 / 26
 The binomial model for independent defaults, cont.

The portfolio credit loss distribution in the binomial model

0.2

0.18 bin(50,0.1)

0.16

0.14

0.12
probability

0.1

0.08

0.06

0.04

0.02

0
0 5 10 15 20 25 30
number of defaults

Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 10 / 26
 The binomial model for independent defaults, cont.

The binomial distribution have very thin ”tails”, that is, it is extremely
unlikely to have many losses (see figure).

For example, if p = 5% and m = 50 we have that P [Nm ≥ 8] = 1.2% and

for p = 10% and m = 50 we get P [Nm ≥ 10] = 5.5%

The main reason for these small numbers is due to the independence
assumption for X1 , X2 , . . . Xm .

To see this, recall that the variance Var(X ) measures

h the degreei of the
2
deviation of X around its mean, i.e. Var(X ) = E (X − E [X ]) .

Since X1 , X2 , . . . Xm are independent we have that

m
! m
X X
Var(Nm ) = Var Xi = Var(Xi ) = mp(1 − p) (2)
i =1 i =1

where the second equality is due the independence assumption.

Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 11 / 26
 The binomial model for independent defaults, cont.

Furthermore, by Chebyshev’s inequality we have that for any random

variable X , and any c > 0 it holds
Var(X )
P [|X − E [X ] | ≥ c] ≤ .
c2

Example: if p = 5% and m = 50 then Var(Nm ) = 50p(1 − p) = 2.375 and

E [Nm ] = 50p = 2.5.

So with p = 5% and m = 50, the probability of say, 6 more or less losses

than expected, is smaller or equal than 6.6%, since by Chebyshev
2.375
P [|Nm − 2.5| ≥ 6] ≤ = 6.6%.
36

Next we show that the deviation of the fractional number of defaults in the
portfolio, Nmm , from the constant p = E Nmm , goes to zero as m → ∞.


Nm
So m converges towards a constant as m → ∞ (the law of large numbers).

Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 12 / 26
 Independent defaults and the law of large numbers
By applying Chebyshev’s inequality to Nmm together with Equation (2) we get
1
Var Nmm


m2 Var (Nm )
 
Nm mp(1 − p) p(1 − p)
P −p ≥ε ≤ 2
= 2
= 2 2
=
m ε ε m ε mε2

Thus, P | Nmm − p| ≥ ε → 0 as m → ∞ for any ε > 0.

 

This result is called the weak law of large numbers

For our credit portfolio it means that the fractional number of defaults in
the portfolio, i.e. Nmm , converges (in probability) to the constant p, i.e the
individual default probability.

One can also show the so called strong law of large numbers, that is
 
Nm
P → p when m → ∞ = 1
m
and we say that Nmm converges almost surely to the constant p. In these
lectures we write Nmm → p to indicate almost surely convergence.
Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 13 / 26
 Independent defaults lead to unrealistic loss scenarios

We conclude that the independence assumption, or more generally, the i.i.d

assumption for the individual default indicators X1 , X2 , . . . Xm implies that
the fractional number of defaults in the portfolio Nmm converges to the
constant p almost surely.

It is an empirical fact, observed many times in the history, that defaults tend
to cluster and Nmm have often values much bigger than p.

Consequently, the empirical (i.e. observed) density for Nmm will have much
more ”fatter” tails compared with the binomial distribution.

We will therefore next look at portfolio credit models that can produce more
realistic loss scenarios, with densities for Nmm that have fat tails, which implies
that Nmm does not converges to a constant with probability 1, when m → ∞.

Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 14 / 26
 Conditional expectations
Before we continue this lecture, we need to introduce the concept of conditional
expectations
Let L2 denote the space of all random variables X such that E X 2 < ∞
 

Let Z be a random variable and let L2 (Z ) ⊆ L2 denote the space of all

random variables Y such that Y = g (Z ) for some function g and Y ∈ L2
Note that E [X ] is the value µ that minimizes the quantity E (X − µ)2 .
 

Inspired by this, we define the conditional expectation E [ X | Z ] as follows:

Definition of conditional expectations

For a random variable Z , and for X ∈ L2 , the conditional expectation E [ X | Z ] is
the random variable Y ∈ L2 (Z ) that minimizes E (X − Y )2 .
 

Intuitively, we can think of E [ X | Z ] as the orthogonal projection of X onto

the space L2 (Z ), where the scalar product hX , Y i is defined as
hX , Y i = E [XY ].

Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 15 / 26
 Properties of conditional expectations

For a random variable Z it is possible to show the following properties

1. If X ∈ L2 , then E [E [ X | Z ]] = E [X ]
2. If Y ∈ L2 (Z ), then E [ YX | Z ] = Y E [ X | Z ]
3. If X ∈ L2 , we define Var(X |Z ) as
2
Var(X |Z ) = E X 2 Z − E [ X | Z ]
 

and it holds that Var(X ) = E [Var(X |Z )] + Var (E [ X | Z ]).

Furthermore, for an event A, we can define the conditional probability P [ A | Z ] as

P [ A | Z ] = E [ 1A | Z ]

where 1A is the indicator function for the event A (note that 1A is a random
variable). An example:
 if X ∈{a, b}, let A = {X = a}, and we get that
P [ X = a | Z ] = E 1{X =a} Z .

Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 16 / 26
 The mixed binomial model

The binomial model can be extended to the mixed binomial model which
randomizes the default probability, allowing for stronger dependence.

The mixed binomial model works as follows: Let Z be a random variable

(discrete or continuous) and let p(x) ∈ [0, 1] be a function such that the
random variable p(Z ) is well-defined.

Let X1 , X2 , . . . Xm be identically distributed random variables such that

Xi = 1 if obligor i defaults before time T and Xi = 0 otherwise.

Conditional on Z , the random variables X1 , X2 , . . . Xm are independent and

each Xi have default probability p(Z ), that is P [ Xi = 1 | Z ] = p(Z )

The economic intuition behind this randomizing of the default probability

p(Z ) is that Z should represent some common background variable affecting
all obligors in the portfolio.

Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 17 / 26
 The mixed binomial model, cont

Let F (x) and p̄ be the distribution and mean of the random variable p(Z ),
that is,
F (x) = P [p(Z ) ≤ x] and E [p(Z )] = p̄. (3)

If for example Z is a continuous random variable on R with density fZ (z)

then p̄ is given by
Z ∞
p̄ = E [p(Z )] = p(z)fZ (z)dz. (4)
−∞

Since P [ Xi = 1 | Z ] = p(Z ) we get that E [ Xi | Z ] = p(Z ), because

E [ Xi | Z ] = 1 · P [ Xi = 1 | Z ] + 0 · (1 − P [ Xi = 1 | Z ]) = p(Z ).

Note that E [Xi ] = p̄ and thus p̄ = E [p(Z )] = P [Xi = 1] since

P [Xi = 1] = E [Xi ] = E [E [ Xi | Z ]] = E [p(Z )] = p̄
where the last equality is due to (3).
Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 18 / 26
 The mixed binomial model, cont
One can show that (see in the lecture notes)
Cov(Xi , Xj ) = E p(Z )2 − p̄ 2 = Var(p(Z )) (5)
 
Var(Xi ) = p̄(1 − p̄) and

Next, letting all losses be the same and constant given by, say ℓ, then the
total credit loss in the portfolio at time T , called Lm , is
m
X m
X m
X
Lm = ℓXi = ℓ Xi = ℓNm where Nm = Xi
i =1 i =1 i =1
thus, Nm is the number of defaults in the portfolio up to time T

Again, since P [Lm = kℓ] = P [Nm = k], it is enough to study Nm .

Since the random variables X1 , X2 , . . . Xm now only are conditionally

independent, given the outcome Z , we have
 
m
P [ Nm = k | Z ] = p(Z )k (1 − p(Z ))m−k
k
Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 19 / 26
 The mixed binomial model, cont.
Hence,
  
m k k
P [Nm = k] = E [P [ Nm = k | Z ]] = E p(Z ) (1 − p(Z )) (6)
k
so if Z is a continuous random variable on R with density fZ (z) then
Z ∞  
m
P [Nm = k] = p(z)k (1 − p(z))m−k fZ (z)dz. (7)
−∞ k

Furthermore, because X1 , X2 , . . . Xm no longer are independent we have that

m
! m m m
X X X X
Var(Nm ) = Var Xi = Var(Xi ) + Cov(Xi , Xj ) (8)
i =1 i =1 i =1 j=1,j6=i

and by homogeneity in the model we thus get

Var(Nm ) = mVar(Xi ) + m(m − 1)Cov(Xi , Xj ). (9)

So inserting (5) in (9) we get that

Var(Nm ) = mp̄(1 − p̄) + m(m − 1) E p(Z )2 − p̄ 2 .
 

(10)
Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 20 / 26
 The mixed binomial model, cont.
Next, it is of interest to study how our portfolio will behave when m → ∞,
that is when the number of obligors in the portfolio goes to infinity.

Recall that Var(aX ) = a2 Var(X ) so this and (10) imply that

p̄(1 − p̄) (m − 1) E p(Z )2 − p̄ 2
   

Nm Var(Nm )
Var = = + .
m m2 m m

We therefore conclude that

 
Nm
→ E p(Z )2 − p̄ 2 = Var(p(Z ))
 
Var as m → ∞ (11)
m

Note that in the case when p(Z ) is a constant, say p, so that p = p̄. we are
back in the standard binomial loss model and
 
Nm
E p(Z )2 − p̄ 2 = p 2 − p 2 = 0 so Var
 
→ 0 as m → ∞
m
i.e. the fractional number of defaults in the portfolio converge to the
constant p = p̄ as portfolio size tend to infinity (law of large numbers.)
Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 21 / 26
 The mixed binomial model, cont.
So in the mixed binomial model, we

see from (11) that the law of large
numbers do not hold, i.e. Var Nmm does not converge to 0.
Nm
Consequently, the fractional number of defaults in the portfolio m does not
converge to a constant as m → ∞.

This is due to the fact that X1 , X2 , . . . Xm , are not independent. The

dependence among X1 , X2 , . . . Xm is created by Z .

However, conditionally on Z , we have that the law of large numbers hold

(because if we condition on Z , then X1 , X2 , . . . Xm are i.i.d with default
probability p(Z )), that is
Nm
given a ”fixed” outcome of Z then → p(Z ) as m → ∞ (12)
m
Since a.s convergence implies convergence in distribution (12) implies that
for any x ∈ [0, 1] we have
 
Nm
P ≤ x → P [p(Z ) ≤ x] when m → ∞. (13)
m
Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 22 / 26
 The mixed binomial model, cont.
Note that (13) can also be verified intuitive from (12) by making the
following observation. From (12) we have that
  
Nm 0 if p(Z ) > x
P ≤x Z → as m → ∞
m 1 if p(Z ) ≤ x
that is,  
Nm
P ≤ x Z → 1{p(Z )≤x} as → ∞. (14)
m
Next, recall that
    
Nm Nm
P ≤x =E P ≤x Z (15)
m m
so (14) in (15) renders
 
Nm  
P ≤ x → E 1{p(Z )≤x} = P [p(Z ) ≤ x] = F (x) as m → ∞
m
where F (x) = P [p(Z ) ≤ x], i.e. F (x) is the distribution function of the
random variable p(Z ).
Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 23 / 26
 Large Portfolio Approximation (LPA)

Hence, from the above remarks we conclude the following important result:

Large Portfolio Approximation (LPA) for mixed binomial models

For large portfolios in a mixed binomial model, the distribution of the fractional
number of defaults Nmm in the portfolio converges to the distribution of the random
variable p(Z ) as m → ∞, that is for any x ∈ [0, 1] we have
 
Nm
P ≤ x → P [p(Z ) ≤ x] when m → ∞. (16)
m

The distribution P [p(Z ) ≤ x] is called the Large Portfolio Approximation (LPA) to

the distribution of Nmm .

The above result implies that if p(Z ) has heavy tails, then the random variable
Nm
m will also have heavy tails, as m → ∞, which then implies a strong default
dependence in the credit portfolio.

Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 24 / 26
 Examples of mixing distributions (next two lectures)
Example 1: A mixed binomial model with p(Z ) = Z where Z is a beta
distribution, Z ∼ Beta(a, b) and by definition of a beta distribution it holds
that P [0 ≤ Z ≤ 1] = 1 so that p(Z ) ∈ [0, 1].

Example 2: Another possibility for mixing distribution p(Z ) is to let p(Z )

be a logit-normal distribution. This means that
1
p(Z ) =
1 + exp (−(µ + σZ ))
where σ > 0 and Z is a standard normal. Note that p(Z ) ∈ [0, 1].

Example 3: The mixed binomial model inspired by the Merton model (will
be discussed coming lectures) with p(Z ) given by
 −1 √ 
N (p̄) − ρZ
p(Z ) = N √ (17)
1−ρ
where Z is a standard normal and N(x) is the distribution function of a
standard normal distribution. Furthermore, ρ ∈ [0, 1] and p̄ = P [Xi = 1].
Note that p(Z ) ∈ [0, 1].
Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 25 / 26
 Thank you for your attention!

Alexander Herbertsson (Univ. of Gothenburg) Financial Risk: Credit Risk, Lecture 1 April 25, 2017 26 / 26