om

i.c
CM2-19: Credit risk Page 1

ng
si
om
o m
as
.m
w
w
w
Credit risk
Syllabus objectives
4.6 Simple models for credit risk

4.6.1 Define the terms credit event and recovery rate.

4.6.2 Describe the different approaches to modelling credit risk:
 structural models
 reduced-form models
 intensity-based models.
4.6.3 Demonstrate a knowledge and understanding of the Merton model.
4.6.4 Demonstrate a knowledge and understanding of a two-state model for
credit ratings with a constant transition intensity.
4.6.5 Describe how the two-state model can be generalised to the Jarrow-Lando-
Turnbull model for credit ratings.
4.6.6 Describe how the two-state model can be generalised to incorporate a
stochastic transition intensity.

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 2 CM2-19: Credit risk

ng
si
om
o m
0 Introduction

as
.m
w
This chapter addresses credit risk – the risk that a person or an organisation will fail to make a

w
w
payment they have promised. This is quite a new area (eg the Jarrow-Lando-Turnbull (JLT) model
was only published in 1997) that has grown in importance with the introduction of credit
derivatives (which are covered in some later subjects, not in this course).

We start with some definitions to set the scene, then go on to look at the Merton model, which is
a simple model that relates the price of a bond subject to default to the price of an option. We
then look at two models that are applications of the theory of continuous-time jump processes.
The first model is a two-state model. The second is the JLT model, which is a more general
multi-state model.

The Core Reading in this chapter is adapted from course notes written by Timothy Johnson.

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-19: Credit risk Page 3

ng
si
om
o m
1 Credit events and recovery rates

as
.m
w
In earlier chapters it has been assumed that bonds are default-free.

w
w
Question

Explain what it means for a bond to be default-free.

Solution

A bond is default-free if the stream of payments due from the bond will definitely be paid in full
and on time.

Credit risk exists when a party may default on its obligations. Credit risk is usually ignored
with respect to payments by a sovereign government in its own currency, but needs to be
accommodated for if an obligation is met in a currency issued by a third-party (such as
corporate obligations, obligations by a government in a currency it does not control).

Corporate entities issuing bonds consist mainly of large companies and banks. For example, in
May 2014, Barclays had a 5¾% bond redeemable in September 2026 and Tesco had a 5½% bond
redeemable in December 2019. These companies entered into a contract to make interest
payments on set dates to the bondholders and to repay the face value of the bond on the
redemption date. Failure to do this would result in the bonds being in default.

A credit loss only exists if the counter-party defaults and the contract has value. In a
forward or swap contract, both long/receiving and short/paying parties are exposed to a
credit risk, since either party could default if the market moves against them. For options
and bonds, the purchaser of the option/bond is exposed to default by the writer/issuer, but
they do not have an obligation to the writer/issuer.

Credit risk is calculated as an expected loss:

Expected Loss = Exposure  Probability of Default  Loss Given Default

All the parameters have an implicit time dependence. The Loss Given Default (LGD) is the
percentage of the exposure that will be lost on a default, the recovery rate is the reciprocal
of the LGD (Recovery Rate = 100%-LGD). Usually some value can be recovered when a
counter-party defaults.

Credit risk changes with the market and good practice is to assess both current and
potential exposure. The current exposure is the current market value of the asset, the
future exposure should be based on a wide range of future scenarios, with different default
probabilities.

For the remainder of the chapter we’ll only be considering credit risk with respect to bonds.

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 4 CM2-19: Credit risk

ng
si
om
m
The outcome of a default may be that the contracted payment stream:

o
as
.m
 is rescheduled

w
w
 is cancelled by the payment of an amount which is less than the default-free value of the

w
original contract
 continued but at a reduced rate
 is totally wiped out.

An example of cancellation would be where the bondholders agree to accept a reduced one-off
cash payment of 75% of the face value instead of the contractual payments.

Credit events, which might result in a failure to meet an obligation (defined for the purposes
of credit derivatives), include:

 actions that are associated with bankruptcy or insolvency laws

ie the bond issuer becomes insolvent.

 downgrade by ‘Nationally Recognised Statistical Rating Organisations’, (NRSROs

such as Moody’s, S&P and Fitch)
This is of particular concern when a bond is issued with a guaranteed minimum credit
rating.

 failure to pay
ie either a coupon or the capital amount is not paid in full and on time.

 repudiation / moratorium
ie the validity of the contract is disputed or a temporary suspension of activity is imposed
on the issuer.

 restructuring – when the terms of the obligation are altered so as to make the new
terms less attractive to the debt holder, such as a reduction in the interest rate, re-
scheduling, change in principal, change in the level of seniority.

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-19: Credit risk Page 5

ng
si
om
o m
2 Approaches to modelling credit risk

as
.m
w
2.1 Structural models

w
w
Structural models are explicit models for a corporate entity issuing both equity and debt. They
aim to link default events explicitly to the fortunes of the issuing corporate entity.

Structural, or firm-value, models are used to represent a firm’s assets and liabilities and
define a mechanism for default. Typically, default occurs when a stochastic variable (or
process) hits a barrier representing default. The main example of a structural model is the
Merton model.

These models deliver an explicit link between a firm’s default and the economic conditions
and provide a sound basis for estimating default correlations amongst different firms. The
disadvantage is identifying the correct model and estimating its parameters.

These models are called ‘structural’ because they focus on the financial structure – the split
between debt and equity – of the company issuing the bond. We will discuss the Merton model
in the next section.

2.2 Reduced-form models

Reduced-form models do not attempt to deliver a representation of a firm, like structural
models do. Rather they are statistical models that use observed data, both macro and
micro, and so can usually be ‘fitted’ to data.

The market statistics most commonly used are the credit ratings issued by NRSROs. The
credit rating agencies will have used detailed data specific to the issuing corporate entity
when setting their rating. They will also regularly review the data to ensure that the rating
remains appropriate and will re-rate the bond either up or down as necessary.

Default is no longer tied to the firm value falling below a threshold-level, as in structural
models. Rather, default occurs according to some exogenous hazard rate process.

These models are called ‘reduced-form’ because they do not attempt to model the inner financial
workings of the particular company issuing the bond. Instead, they model the different levels of
creditworthiness and how companies move from one status to another.

2.3 Intensity-based models

An intensity-based model is a particular type of continuous-time reduced-form model. It
typically models the jumps between different states (usually credit ratings) using transition
intensities. The disadvantage of reduced-form models is that they sometimes lack the
clarity of structural models.

This approach uses a continuous-time multiple state model (jump process) where the states
correspond to the creditworthiness of the company. The ‘intensities’ (denoted by  ) are the
rates driving the company to switch from one state to another as time passes. Intensity-based
models are an example of a reduced-form model.

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 6 CM2-19: Credit risk

ng
si
om
o m
3 The Merton model

as
.m
w
The Merton model is a simple example of a structural model.

w
w
Classical finance defines the value of a firm F  t  as the sum of its debt, B  t  and equity
E  t  , so:

F t   B t   E t 

Merton’s model assumes that a corporate entity has issued both equity and debt such that its
total value at time t is F  t  . This value varies over time as a result of actions by the corporate
entity, which does not pay dividends on its equity or coupons on its bonds.

For example, the value of a company will change along with investors’ perceptions of the future
prospects of that company.

Assume a firm has issued a single zero-coupon bond with face value of L which matures at
time T.

Debt holders rank above shareholders in the wind-up of a company. So, provided the company
has sufficient funds to pay the debt, the shareholders will receive F (T )  L .

F( T)

F(t)
Equity F(T)-L
Equity

Debt Debt L

t T

The corporate entity will default if the total value of its assets, F T  , is less than the promised
debt repayment at time T :

ie F T   L

In this situation, the bondholders will receive F T  instead of L and the shareholders will receive
nothing.

Combining these two cases, we see that the shareholders will receive a payoff of max F (T )  L,0
at time T .

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-19: Credit risk Page 7

ng
si
om
m
This can be regarded as treating the shareholders of the corporate entity as having a European

o
as
call option on the assets of the company with maturity T and a strike price equal to the value of

.m
the debt. The Merton model can be used to estimate either the risk-neutral probability that the

w
w
company will default or the credit spread on the debt.

w
Since debt is senior to equity, the value of equity at maturity:

E T   max F (T )  L,0

and so the value of a firm’s equity is a call option on the value of the firm with a strike of the
debt.

Credit spread is a measure of the excess of the yield on a risky security over a risk-free yield. It
largely relates to the expected cost of default. However, in practice it will also typically reflect
other factors, such as a risk premium relating to the risk of default and a liquidity premium.

Because of the risk of default, a bond issued by a company will have a lower market price than a
similar bond issued by a government. So the yield on the company bond will be higher than the
yield on the government bond. The credit spread refers to the difference between these two
rates.

Question

Suggest how this model could be used to calculate the value of the risky corporate bonds at
time t.

Solution

Since we are viewing the equity as a call option on the total value of the company, we could use
an option pricing method, such as the Black-Scholes option pricing formula, to calculate how
much the equity at time T is worth now (ie at time t ).

The value of the bonds could then be found by subtracting this from the current value of the
company F (t) , ie:

B(t)  F (t)  E (t) .

where B(t) and E (t) are the current value of the company’s risky corporate bonds and the equity
respectively.

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 8 CM2-19: Credit risk

ng
si
om
m
Consequently,

o
as
.m
E  t   F (t )(d1)  Le  r (T t )(d 2 ) (1)

w
w
w
with:

log  F (t ) / L   (r  21  F2 )(T  t )
d1  , d 2  d1   F T  t
F T  t

where  F is the volatility of the firm value.

The value of the debt today is F (t )  E (t ) .

Unfortunately, F (t ) (and  F ) are unobservable, since they depend on the market’s

assessment of B(t ) . However, if we assume that the value of the firm and the equity both
follow geometric Brownian motion, and E (t )  f (F (t )) then by Ito:

 E (t ) 1 2 2  2E (t )  E (t )
dE (t )   F F (t )   F F (t ) 2  dt   F F (t ) dWt
  F (t ) 2 F (t )  F (t )

where the values F and  F come from the stochastic differential equation for F (t) :

dF (t)  F (t)  F dt   F dWt 

However, E (t) also has its own stochastic differential equation:

dE (t )  E (t )  E dt   E dWt 

Comparing terms leaves:

E (t )
 E E (t )   F F (t )   F F (t )  d1  (2)
F (t )

Now E (t) and  E can be observed from market data.

Using equations (1) and (2) we have:

 E E (t )   F F (t )  d1 


  F E  t   Le  r (T  t )(d 2 ) 
 E E (t )
 F 
E (t )  Le  r (T  t )(d 2 )

This is not trivial since  F is used to calculate d 2 .

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-19: Credit risk Page 9

ng
si
om
m
Recall that, in the Black-Scholes world, (d 2 ) is the risk-neutral probability that a call

o
as
option will be exercised, that is it expires in the money. In this context, this means that:

.m
w
1  (d 2 )  (d 2 )

w
w
is the risk-neutral probability that the firm is in default at time T (but not that it has defaulted
in [0,T ) and then recovered). This method provides a rough estimate of the probability of
default.

In general it may be possible for default bonds to ‘recover’, ie to re-start coupon payments.
However, the Merton model is only concerned with the state of the bond at time T. In the next
section it is assumed that once a bond defaults then it always remains in such a state.

One limitation is that the default probability is given in the abstract risk-neutral world. The
real-world probability can be derived using:

log  F (t ) / L   (F  21  F2 )(T  t )

d1  , d 2  d1   F T  t
F T  t

where the real-world drift, F , replaces the risk-less drift. However, F is not observable.

The pricing equation (1) uses the risk-neutral probability measure, under which F (t) is expected
to grow at the risk-free rate. This is the reason why the unobservable quantity F is not required,
and the above calculations have been to find only  F (which is used in (1)).

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 10 CM2-19: Credit risk

ng
si
om
o m
4 Two-state models for credit risk

as
.m
w
4.1 Interest rates as hazard rates

w
w
In Europe, before the Reformation of the Catholic Church in the sixteenth century, the
charging of interest was only permissible as a compensation for the risk that the lender
took on; this is captured in the opening observation of the Black-Scholes paper:

It should not be possible to make a risk-less profit.

We will now consider a two-state intensity-based model, which is the simplest continuous-time
reduced-form model.

A model can be set up, in continuous time, with two states:

1. N = not previously defaulted

2. D = defaulted.

If the transition intensity, under the risk-neutral measure Q , from N to D at time t is denoted
by  (t) , this model can be represented as:

 (t)
No default, N Default, D

and D is an absorbing state.

This is a two-state continuous-time Markov model. It has the same structure as the two-state
mortality model, with ‘No default’ corresponding to ‘Alive’, ‘Default’ corresponding to ‘Dead’ and
 (t) corresponding to the force of mortality.

Question

Write down a formula for the probability (under Q ) that a company that is in state N at time t
will remain in state N until time T .

Use the survival probability formula t px  exp     x  sds  for the corresponding
t
Hint:
 0 
two-state mortality model from page 32 of the Tables.

Solution

The probability we are interested in can be written as Q  X (T )  " N "| X (t)  "N "  .

The formula given in the hint tells us the probability that a person who is alive at age x will still be
alive t years later.

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-19: Credit risk Page 11

ng
si
om
m
So the corresponding formula here is:

o
as
.m
Q  X (T )  "N "| X (t)  "N "  exp     (s)ds 
T

w
 

w
t

w
Consider a lender lending a sum of money, L. The lender is concerned that the borrower
will not default, a hopefully rare event, and will eventually pay back the loan. Poisson
worked out that if the rate of a rare event occurring was  then the chance of there being k
rare events in n time periods was given by:

( n)k e   n
P (k wins in n rounds) 
k!

Say the lender assesses that the borrower will default at a rate of  defaults a day – known
as the hazard rate – and the loan will last T days. The lender might also assume that they
will get all their money back, providing the borrower makes no defaults in the T days, and
nothing if the borrower makes one or more defaults.

In this model the borrower can make at most one default.

On this basis the lender’s mathematical expectation of the value of the loan is:

E  value of loan  P (No default)  L + P (Default)  0

Using the Law of Rare Events, the probability of no defaults is given when k  0 , so:

(T )0 e  T
E  value of loan   L + P (Default)  0
0!

We can ignore the second expression, since it is zero, then:

E  value of loan  Le  T

So, the lender is handing over L with the expectation of only getting Le  T  L back. To
make the loan equal the expected repayment, the banker needs to inflate the expected
repayment by e T :

 Le  e 
T  T
L

Another way to view this is to keep the payout L the same, and to reduce the amount the lender
is prepared to give for it, ie: Le  T . This discounting can be viewed as the interest rate obtained
on the debt.

4.2 Incorporating recovery rates

Consider a simple situation whereby an asset is due to pay-out at some time T, but is
subject to credit-risk and may default at a time  . In the event of default, the investor
recovers a fraction  at time T.

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 12 CM2-19: Credit risk

ng
si
om
m
So, for a zero-coupon bond that is due to pay 1 at time T , the actual payment at time T will be:

o
as
.m
 1 if T  

w
payment  
 if   T

w
w
For example, if   0.9 , we are assuming that, once the company has gone into default, all future
interest payments and the redemption payment will be reduced by 10% (since 1    0.1) . So
bondholders will receive 90% (the recovery rate) of the full amounts.

Question

Write down a single formula (ie one without if’s) for the payment at time T .

Solution

We can use the indicator variables 1T   and 1 T  to do this. The correct formula for the
payment is:

1T     1 T  .

If default has not occurred by time T , (ie T   ) then 1T    1 and 1 T   0 , and the formula
gives 1, ie the full payment will be made.

If default has occurred (ie   T ), then 1T    0 and 1 T   1 and the formula gives  , ie the
reduced or ‘recovered’ payment of  will be made.

So a bond with credit risk can be viewed as a derivative, whose future payout is uncertain. This
allows the use of the general risk-neutral pricing formula:

value of the loan at time t  e  r (T t )EQ payout | Ft 

assuming a constant risk-free rate of interest r, and a risk-neutral probability measure Q.

More generally, let Ct represent a money market cash account which is a unit of currency
invested at time 0 and accumulated at the (possibly varying) risk-free rate. In the case of a
constant risk-free rate we have Ct  ert . Then we have:

 payout | Ft 
value of the loan at time t  Ct EQ  
 CT 

The value, at time t, of this asset to its holder is:

 (t ,T )  e r (T t)EQ 1{T  }   1{ T } |Ft 

1   1{ T } |Ft 
 Ct EQ  {T  } 
 CT 

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-19: Credit risk Page 13

ng
si
om
m
This implies that:

o
as
.m
 (t ,T )  1{T  }   1{ T } 

w
 EQ  | Ft 

w
Ct  CT 

w
Assume that the probability of default is independent of the money market account.

This means that the CT doesn’t need to be inside the expectation operator. Let P(t ,T ) be the
price at time t of a risk-free zero-coupon bond that pays out 1 at future time T.

Question

Write P(t ,T ) in terms of the money market cash account.

Solution

C
P(t ,T )  t
CT

Then the value of the credit-risky asset can be written as:

Ct
 (t ,T )  EQ 1{T  }   1{ T } |Ft 
CT

 P (t ,T ) 1  q (t ,T )   q (t ,T )

where q is the probability, in the risk-neutral measure, that default occurs before T, ie:
Q(  T ) .

This is algebraically equivalent to:

 (t ,T )  P(t ,T )1  (1   )q(t ,T )

where q(t ,T ) is the risk-neutral probability of default. We saw earlier that the risk-neutral

probability of not defaulting (ie surviving) is exp     (s)ds  . So we have:

T
 t 

  
 (t ,T )  P(t ,T ) 1  (1   )  1  exp     (s)ds   
T

   t  

This expression implies that the probability of default, q, is given by:

1   (t ,T ) 
q (t ,T )   1 
1   P (t ,T ) 

Employing this model is not as straightforward as it appears because it is practically

difficult to identify the risk-less bond, P (t ,T ) , pertaining to the credit-risky asset,  (t ,T ) , and
the recovery rate,  , at some specific maturity, T.

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 14 CM2-19: Credit risk

ng
si
om
m
To get around this, credit ratings are employed, and it is assumed that the same yield curve

o
as
is applied to firms in the same credit rating. This allows bonds issued by different firms in

.m
the same risky class to be considered as if there were a single issuer.

w
w
w
Dividing companies into just two states – ‘No default’ (N) and ‘Default’ (D) – is a rather crude,
black and white approach. In reality, the ‘N’ state is a heterogeneous category and we can
improve this model by introducing some shades of grey. This involves subdividing the ‘N’ state.

In the next section we will extend the two-state model to cover multiple credit ratings.

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-19: Credit risk Page 15

ng
si
om
o m
5 The Jarrow-Lando-Turnbull (JLT) model

as
.m
w
There are several established credit rating agencies, such as Standard & Poor’s and Moody’s, who

w
w
publish credit ratings for all the major companies. On Standard & Poor’s scale, companies that
have already defaulted are Grade D. Companies that have not defaulted are given one of the
seven grades: AAA, AA, A, BBB, BB, B, CCC (which can be fine-tuned further with +’s and –’s). For
example, in September 2014, the Barclays bond mentioned earlier had a credit rating of BBB–
with Standard & Poor’s.

Using credit ratings develops the two-state (default/no default) model into an n-state model.
The state dynamics are represented by a time-homogenous Markov chain. The probability
of going from one state to another depends only on the two states themselves (the Markov
property) and this transition probability is assumed to be independent of time (time
homogeneity).

The first state is the best credit quality and the n th state represents default, which is an
absorbing state – there is no chance of recovering from default – and a payment of  is
made at maturity. The n  n transition matrix, Q(t ,T ) , of the Markov chain, can be obtained
from credit-ratings agencies along with information on recovery rates.

So, for example, we could model the Standard & Poor’s rating system if we used n  8 . This
would give n  1  7 credit ratings, from AAA (= State 1) down to CCC (= State 7), for companies
that are not already in default. The n th state (= State 8) would be for companies that are already
in default (which we assume they stay in for ever).

The following Core Reading table gives an example of a transition matrix, showing the
probabilities (%) of jumping from one credit rating to another (and ultimately into ‘default’).

AAA AA A BBB BB B CCC Default

AAA 90.81 8.33 0.6 0.06 0.12 0.08 0.00 0.00
AA 0.70 90.65 7.79 0.64 0.06 0.14 0.02 0.00
A 0.09 2.27 91.05 5.52 0.74 0.26 0.01 0.06
BBB 0.02 0.33 5.95 86.93 5.30 1.17 0.12 0.18
BB 0.03 0.14 0.67 7.73 80.53 8.84 1.00 1.06
B 0.00 0.12 0.24 0.43 6.48 83.46 4.07 5.20
CCC 0.22 0.00 0.22 1.30 2.38 11.24 64.86 19.78

This means that the value of a credit-risky asset issued by a firm with rating i is given by:

 i (t ,T )  P (t ,T ) 1  qin (t ,T )   qin (t ,T )

where qin is the risk-neutral probability of a bond in State i at time t being in State n at time T (ie
having defaulted). So bonds with the same credit rating and term are equally priced.

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 16 CM2-19: Credit risk

ng
si
om
m
We define the transition intensity, under the risk-neutral measure Q , from State i to State j at

o
as
time t to be ij  t  . If the transition intensities ij  t  are assumed to be deterministic, then this

.m
w
model for default risk can be represented by the following diagram:

w
w
In this n-state model, transfer is possible between all states except for the default State n, which
is absorbing. The  ’s here correspond to the entries in the generator matrix. As shorthand we
write:

ii  t     ij  t 
j i

The value of  when the two subscripts are equal (ie ii ) does not correspond to an actual
transition – if the process starts and ends in State i , it hasn’t actually made a jump. However,
defining ii to equal minus the sum of all the actual transition rates out of State i allows us to
write the set of equations for this process in a compact form.

Question

For a particular 4-state version of the JLT model:

12  21  0.1 , 23  0.2 , 24  0.1 , 32  0.1 , 34  0.4 , 13  14  31  0

Construct the complete generator matrix.

Solution

Entering the values we’ve been given, we get:

1 2 3 4
1  ? 0.1 0 0 
2 0.1 ? 0.2 0.1 

 
3  0 0.1 ? 0.4 
 
4  ? ? ? ? 

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-19: Credit risk Page 17

ng
si
om
m
We know that State 4 (the default state) is absorbing. So the first three entries along the bottom

o
as
row will be zero.

.m
w
w
The entries down the diagonal are equal to minus the sum of the other entries for that row. For

w
example, 22  (0.1  0.2  0.1)  0.4 .

So the final generator matrix looks like this:

1 2 3 4
1  0.1 0.1 0 0 
2  0.1 0.4 0.2 0.1 
 
3  0 0.1 0.5 0.4 
 
4  0 0 0 0 

The generator matrix can be used to write down a set of differential equations (the Kolmogorov
differential equations), which can then be solved to find the probabilities of making specified
transitions over specified periods.

In fact it is possible to calculate the transition probabilities directly from the generator matrix,
using a method involving matrix calculations, which we will describe now.

We introduce the following notation:

 i , j 1
n
   t  is an n  n generator matrix,   t   ij  t 

 qij  s,t   P  X  t   j X  s   i  for t  s

 i , j 1 is the matrix of transition probabilities

n
 Q  s,t   qij  s,t 

Question

State in words what (t) , qij (s ,t) and Q(s,t) represent.

Solution

(t) is the generator matrix at time t . The positive entries in this matrix represent the transition
rates (intensities) from one state to another at that time. The negative entries on the main
diagonal are notional values equal to minus the sum of the other entries in the same row (which
means that the entries in each row add up to zero).

qij (s ,t) is the probability that a company that is in State i at a particular time s will be in State j
at a specified future time t . If s is the current time and t is the time a bond payment is due, this
probability tells us how likely it is that the company (currently in State i ) will be in each state –
and hence how likely it is to default. So these are what we want to use the model to work out.

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 18 CM2-19: Credit risk

ng
si
om
m
Q(s,t) is a matrix showing the complete set of n  n transition probabilities for the period from

o
as
time s to time t . It gives us the same information as working out all the individual probabilities.

.m
w
w
w
It can be shown that:

t 
Q  s,t   exp     u  du 
 s 

where, for an n  n matrix M , we define:


1 k
eM  I   M
k 1 k !

The proof of this is not required for this course.

The e x function on your calculator can be generalised to an exponential function for matrices,
written as exp(M) or eM . You can see that its definition corresponds to the familiar series
e x  1  x  2!
1 x2   .

1 0  0
0 1   
In the matrix version, I    is the identity matrix, which has 1’s down the main
    0
 
0  0 1
diagonal and zeros everywhere else. The matrix exponential function shares most of the familiar
properties of the scalar version, such as exp(0)  I , where 0 is a matrix consisting entirely of
zeros, and exp( A)exp( A)  exp(2 A) .

Question

In a particular two-state intensity-based model with constant transition intensities, the integrated
t  a a 
generator matrix M   (u)du has the form M    , where a  0.2(t  s) and
s
 b b 
b  0.1(t  s) .

(i) Show that, in this case, M 2  (a  b)M .

(ii) Hence deduce an explicit formula for eM in terms of a and b .

(iii) Hence deduce a set of formulae for the transition probabilities qij (s ,t) .

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-19: Credit risk Page 19

ng
si
om
o m
Solution

as
.m
 a a 

w
If M  

w
(i)  , we find that:
 b b 

w
 a a  a a 
M2    
 b b  b b 
 (a)(a)  ab (a)a  a(b) 
 
 b(a)  (b)b ba  (b)(b) 
 a2  ab a2  ab 
 
 ba  b2 ba  b2 

 a(a  b) a(a  b) 
 
 b(a  b) b(a  b) 
 a a 
 (a  b)  
 b b 
 (a  b)M

(ii) Extending this to higher powers, we see that:

M 3  M 2M  (a  b)MM  (a  b)M 2  (a  b)2 M

and M k  [(a  b)]k 1 M

So, in this case, the exponential function is:


1 k
exp(M)  I   M
k 1 k !


1
I  [(a  b)]k 1 M
k 1 k !

1   1 k
I   [(a  b)]  M
a  b  k 1 k ! 
 e (ab)  1   1  e (ab) 
 I  M  I   M
 ab   ab 
   

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 20 CM2-19: Credit risk

ng
si
om
m
(iii) We can then substitute the values of a and b to find the matrix of transition

o
as
probabilities:

.m
w
 1  e (ab) 

w
w
Q(s,t)  I   M
 ab 
 

 1 0   1  e (ab)   a a 
   
 0 1   a  b   b b 

 1 0   1  e 0.3(t  s)   0.2 0.2 

  
 0 1   0.3   0.1 0.1 

 1  2 e 0.3(t  s) 2  2 e 0.3(t  s) 
3 3 3 3 
 1  1 e 0.3(t  s) 2  1 e 0.3(t  s) 
3 3 3 3 

So the transition probabilities are:

q11 (s,t)  13  23 e 0.3(t  s) , q12 (s,t)  23  23 e 0.3(t  s)

q21 (s,t)  13  13 e 0.3(t  s) , q22 (s,t)  23  13 e 0.3(t  s)

This seems rather trivial, however the approach can be enhanced if it is assumed, as in
Jarrow-Lando-Turnbull, that the transition matrix, Q(t ,T ) , obeys the following relationship:

Q(t ,T )  e  (T t )

In addition, assume that  , a hazard rate, is the generator matrix and is ‘diagonalisable’,
meaning it can be written:

   D 1

where D is a diagonal matrix of the eigenvalues of  and  is a matrix whose columns are
the eigenvectors of  . This means:

Q(t ,T )  e D (T  t ) 1

The elements, ij of  have the following properties:

 ij  0

n
  ij  0
j 1

 nj  0

 state i  1 is always more risky that state i

So in   jn for all i  j .

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-19: Credit risk Page 21

ng
si
om
m
As a result:

o
as
.m
  ijˆ jn  e 
n 1
d j (T  t )
qin (t ,T )   1 , 1 i  n  1

w
w
j 1

w
where:

  ij is an element of 

 ˆ jn is an element of  1

 dj is an eigenvalue of  .

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 22 CM2-19: Credit risk

ng
si
om
o m
6 Stochastic transition probabilities

as
.m
w
The Jarrow-Lando-Turnbull approach assumes that the transition intensities between states are

w
w
deterministic.

Although we have allowed the values of the transition intensities  (t) between any two states to
vary over time, we have assumed that the functions involved are known with certainty at the
outset. In real life, however, economic conditions can change unpredictably. If, for example, a
recession struck, we would expect the  (t) ’s corresponding to jumps to a higher-numbered state
(ie a state closer to the default State n ) to increase significantly, as companies struggled to
remain profitable.

An alternative approach would be to assume that the transition intensity between states,   t  , is
stochastic and dependent on a separate state variable process U  t  .

By using a stochastic approach,   t  , can be allowed to vary with company fortunes and other
economic factors. For example, a rise in interest rates may make default more likely, so U  t 
could include appropriate allowance for changes in interest rates. This approach can be used to
develop models for credit risk that combine the structural modelling and intensity-based
approaches.

In the previous section we had a generator matrix of the form:

   D 1

The approach now is to allow this matrix to vary in a stochastic way.

This additional level of complexity means that Λ can be made stochastic:

  DU (t ) 1

where U (t ) is a stochastic process. This means that the transition probabilities can be
conditioned on U (t ) :

  T  
n 1   d j  U (s )ds  
qin (t ,T ) | U (t )   ij ˆ jn  E e t   1 , 1  i  n  1
j 1    
   

The process U (t ) can be made as complex as required, generating multi-factor credit-risk

models.

Question

In the trivial case where U  t   1 for all t, show that the above result recovers the deterministic
version of qin (t ,T ) found in the previous section.

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-19: Credit risk Page 23

ng
si
om
o m
Solution

as
.m
w
  T  

w
n1   d j  1ds  

w
qin (t ,T )|(U(t)  1)    ijˆ jn  E e t   1 
j 1    
   
n1
  ijˆ jn  E e  1 
d j (T t ) 

j 1
 

  ijˆ jn e 
n1
d j (T t )
 1
j 1

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 24 CM2-19: Credit risk

ng
si
om
o m
as
.m
w
w
w
This page has been left blank so that you can keep the chapter
summaries together for revision purposes.

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-19: Credit risk Page 25

ng
si
om
o m
as
Chapter 19 Summary

.m
w
w
Credit events and recovery rates

w
A credit event is an event that will trigger the default of a bond. In the event of a default, the
fraction  of the defaulted amount that can be recovered through bankruptcy proceedings
or some other form of settlement is known as the recovery rate.

Modelling credit risk

Structural models are models for a company issuing both shares and bonds, which aim to link
default events explicitly to the fortunes of the issuing company.

Reduced-form models are statistical models that use observed market statistics such as
credit ratings.

Intensity-based models are a particular type of continuous-time reduced-form models. They

typically model the ‘jumps’ between different states (usually credit ratings) using transition
intensities.

The Merton model

Merton’s model is a structural model. It assumes that a company has issued both equity and
debt such that its total value at time t is F  t  . The total value of the bonds issued and the
shareholders’ interests equals F  t  .

The shareholders of the company can be regarded as having a European call option on the
assets of the company with maturity T and a strike price equal to the face value ( L ) of the
debt.

The Merton model is tractable and gives us some insight into the nature of default and the
interaction between bondholders and shareholders. It can be used to estimate either the
risk-neutral probability that the company will default or the credit spread on the debt.

The Actuarial Education Company © IFE: 2019 Examinations

 om
i.c
Page 26 CM2-19: Credit risk

ng
si
om
o m
as
The two-state model for credit ratings

.m
w
 State N = not previously defaulted

w
w
 State D = previously defaulted

Let   t ,T  be the price at time t of a risky zero-coupon bond that matures at time T .
Then:

  t ,T   e  EQ payoffattimeT Ft 
 r T t

 e  EQ 1T     1 T  Ft 

 r T t
 

 r T t   
 e   1  (1   ) 1  exp     (s)ds  
T

   t  

where:
●  is the recovery rate
● 1{} is the indicator function

●   s  is the risk-neutral transition rate or intensity.

The Jarrow-Lando-Turnbull (JLT) model

© IFE: 2019 Examinations The Actuarial Education Company

 om
i.c
CM2-19: Credit risk Page 27

ng
si
om
o m
If
Inathis n -state model,
zero-coupon bond transfer
maturingisatpossible pays: all states except for State n (default),
time T between

as
 t yet

.m
which is1absorbing. If Xnot
if default has is the state or credit rating at time t , then, for i  1, 2,..., n  1 ,
occurred

w
w
the transition probabilities over the time interval (s,t) are:
  if default has occurred

w
qij  s,t rating
and the credit 
  Q Xof tthe
  underlying 
j X  s   i corporate
for t  s entity is i , the fair price of the bond is:

The matrix of transition probabilities is:

 
B  t ,T , X  t   i   P  t ,T  1  1    PQ  X T   n X (t)  i 
 i , j 1
n
Q  s,t   qij  s,t 
Two-state models with stochastic transition intensity
then:
The transition intensity  t   t  can
 be allowed to vary stochastically to reflect other economic
Q  s,tas
factors, such  the   (of
explevel du 
u)interest rates. This is done by introducing a separate state
 s 
variable process X (t) .

where:
This approach can be used to develop models for credit risk that combine the structural
modelling and intensity-based approaches.
 i , j 1
n
  t   ij  t  is the matrix of transition intensities.

The Actuarial Education Company © IFE: 2019 Examinations