Foundations of Financial Engineering

The Merton Structural Model

Martin B. Haugh
Department of Industrial Engineering and Operations Research
Columbia University
 Structural Models
Structural approach to credit modeling began with Merton (1974).

It is based on the fundamental accounting equation:

Assets = Debt + Equity (1)

- states that asset-value of a firm equals value of firm’s debt + equity

- we are assuming no taxes.

(1) follows because all profits generated by a firm’s assets will ultimately accrue
to the debt- and equity-holders.

2
 Structural Models
The capital structure of a firm is such that debt-holders are senior to equity
holders
- implies that in event of bankruptcy debt-holders must be paid off in full
before equity-holders can receive anything.

This insight allowed Merton to write time T equity value, ET , as a call option on
firm value, VT , with strike equal to face value of debt, DT .

Merton’s model therefore implies

ET = max (0, VT − DT ) (2)

with default occurring if VT < DT .

3
 Merton’s Structural Model
Note that (2) implicitly assumes that the firm is wound up at time T and that
default can only occur at that time
- not very realistic assumptions
- they have been relaxed in many directions since Merton’s original work
- nonetheless, can gain many insights from working with (2).

Can take Vt to be value of a traded asset (why?) so that risk-neutral pricing

applies.

If firm does not pay dividends then could assume that Vt ∼ GBM(r, σ)
- so Et = Black-Scholes price of a call option with maturity T , strike DT and
underlying security value Vt .

Can compute other quantities such as the (risk-neutral) probability of default etc.

4
 e.g. A Merton Lattice Model
Consider following example:

V0 = 1, 000, T = 7 years, µ = 15%, σ = 25% and r = 5%.

# of time periods = 7.
Face value of debt is 800 and coupon on the debt is zero.

First task is to construct lattice model for Vt . Can do this following our usual
approach to lattice construction:
ν = (µ − σ 2 /2)
q
2
ln u = σ 2 ∆t + (ν∆t)
d = 1/u
risk-neutral probability of an up-move is q = (e r∆t − d)/(u − d).

5
 A Merton Lattice Model
Firm Price Lattice
6940.6
5262.6 3990.2
3990.2 3025.5 2294.0
3025.5 2294.0 1739.4 1318.9
2294.0 1739.4 1318.9 1000.0 758.2*
1739.4 1318.9 1000.0 758.2 574.9 435.9*
1318.9 1000.0 758.2 574.9 435.9 330.5 250.6*
1000.0 758.2 574.9 435.9 330.5 250.6 190.0 144.1*

t = 0 t = 1 t = 2 t = 3 t = 4 t = 5 t = 6 t = 7

Note default only possible at time T .

6
 Merton’s Model
Now ready to price the equity and debt, i.e. corporate bonds, of the firm.

We price the equity first by simply viewing it as a regular call option on VT with
strike K = 800 and using risk-neutral backward evaluation.

The bond or debt price can then be computed similarly or by simply observing
that it must equal the difference between the firm-value and equity value at each
time and state.

7
Equity Lattice
6140.6
4501.6 3190.2
3266.4 2264.5 1494.0
2336.9 1570.2 978.4 518.9
1640.8 1054.7 603.6 258.0 0.0
1127.8 687.1 358.4 128.3 0.0 0.0
758.6 435.7 207.1 63.8 0.0 0.0 0.0
499.7 269.9 117.4 31.7 0.0 0.0 0.0 0.0

t = 0 t = 1 t = 2 t = 3 t = 4 t = 5 t = 6 t = 7

Debt Lattice
800.0
761.0 800.0
723.9 761.0 800.0
688.6 723.9 761.0 800.0
653.2 684.7 715.3 742.0 758.2*
611.6 631.7 641.6 630.0 574.9 435.9*
560.3 564.3 551.1 511.1 435.9 330.5 250.6*
500.3 488.3 457.5 404.2 330.5 250.6 190.0 144.1*

t = 0 t = 1 t = 2 t = 3 t = 4 t = 5 t = 6 t = 7
8
 Merton’s Model
We see the initial values of the equity and debt are 499.7 and 500.3, respectively.

The yield-to-maturity, y, of the bond satisfies 500.3 = e −yT × 800 which implies
y = 6.7%.

The credit spread is then given by c = y − r = 1.7% or 170 basis points.

9
 Merton’s Model
Can easily compute the true or risk-neutral probability of default by constructing
an appropriate lattice.

Also easy to handle coupons: if debt pays a coupon of C per period, then we
write ET = max(0, VT − DT − C ).

And in any earlier period we have

Et = max 0, qE u + (1 − q)E d /R − C
 

where R = e r∆t .

10
 Foundations of Financial Engineering
The Black-Cox Structural Model

Martin B. Haugh
Department of Industrial Engineering and Operations Research
Columbia University
 The Black-Cox Model
Black-Cox model generalizes Merton model by allowing default to also occur
before time T .

We now assume default occurs first time firm value falls below face value of debt.

In that case we can compute the value of the equity by placing 0 in those cells
where default occurs
- and updating other cells using usual backwards evaluation approach.

Debt value at a given cell in the lattice given by difference between the firm and
equity values in that cell.

We obtain the following equity and debt lattices:

2
Equity Lattice
6140.6
4501.6 3190.2
3266.4 2264.5 1494.0
2336.9 1570.2 978.4 518.9
1640.8 1054.7 603.6 258.0 0.0
1115.8 660.7 300.1 0.0 0.0 0.0
703.9 328.5 0.0 0.0 0.0 0.0 0.0
350.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0

t = 0 t = 1 t = 2 t = 3 t = 4 t = 5 t = 6 t = 7

Debt Lattice
800.0
761.0 800.0
723.9 761.0 800.0
688.6 723.9 761.0 800.0
653.2 684.7 715.3 742.0 758.2*
623.6 658.2 699.9 758.2* 574.9* 435.9*
614.9 671.5 758.2* 574.9* 435.9* 330.5* 250.6*
650.0 758.2* 574.9* 435.9* 330.5* 250.6* 190.0* 144.1*

t = 0 t = 1 t = 2 t = 3 t = 4 t = 5 t = 6 t = 7
3
 The Black-Cox Model
We see the debt-holders have benefitted from this new default regime: their value
increased from 500.3 to 650.

Of course this increase has come at the expense of the equity holders whose value
has fallen from 499.7 to 350.

In this case the credit spread on the bond is -200 basis points!
- unreasonable value of credit spread is evidence against the realism of default
assumption made here.

Negative credit spreads generally not found in practice but have occurred here
because the debt holders essentially own a down-and-in call option on the value
of the firm with zero strike and barrier equal to the face value of the debt.

While it is true that a firm can default at any time, the barrier would generally be
much lower than the face value of the long-term debt of 800.

Note that we could easily use a different and time-dependent default barrier to
obtain a more realistic value of the credit spread.
4
 Foundations of Financial Engineering
Ratings Models

Martin B. Haugh
Department of Industrial Engineering and Operations Research
Columbia University
 Ratings Models
An alternative approach to modeling default events is via ratings transitions.

Ratings are intended to signify the credit-worthiness of financial instruments with

safer securities having higher ratings
- e.g. a security rated AAA is believed to be very secure and have almost no
default risk.

These ratings are updated periodically as the prospects of the firms (or
governments!) change.

Can model these periodic updates via a Markov chain and use historical ratings
transitions to estimate the transition matrix, P.

2
 Ratings Models

– A sample transition matrix for Standard & Poors ratings.

Pi,j = probability a firm with current rating i will be rated j one year or one
quarter from now.

Final row of matrix omitted since a firm in default assumed to stay in default.

3
 Ratings Models
Ratings transition models were popularized by CreditMetrics and J.P. Morgan in
the late 1990’s.

Their approach to credit risk was to assume the credit rating of a company was
well-modeled by a Markov chain with transition matrix P.

Then easy to compute the probability of default (or indeed losses / gains due to a
deterioration / improvement in credit quality) over any period of time.

e.g. easy to see that Pk = k-period transition matrix.

4
 Ratings Models
More generally, could use a database of ratings transitions to estimate a
continuous-time Markov model so that ratings transitions could occur at any
time instant – this of course is more realistic.

To compute risk measures such as Value-at-Risk (VaR) for credit portfolios, it is

necessary to model the joint ratings transition of many companies
- can be achieved using copula methods
- Monte-Carlo methods can then be used to estimate quantities of interest.

5
 Foundations of Financial Engineering
Credit-Default Swaps

Martin B. Haugh
Department of Industrial Engineering and Operations Research
Columbia University
 Credit-Default Swaps
Credit default swaps (CDS’s) are a very important class of derivative instrument
- developed in late 1990’s
- and now ubiquitous in the financial markets.

A CDS allow investors (or speculators) to hedge (or take on) the default risk of a
firm or government.

A CDS is structured like an insurance policy between two parties:

Party A agrees to pay party B a fixed amount every period (typically every
quarter), in return for protection against the default of a third party, C.

These payments constitute the premium leg and size of these payments are
proportional to the notional amount, N .

When a default occurs, party B must pay party A the difference between N and
the market value of the reference bond (with notional N ) issued by party C.

2
 Pricing a CDS in General
Also possible to have physical settlement: upon default the protection buyer
delivers the reference bond to the protection seller and receives the face value of
the bond in return
- this payment constitutes the default leg of the CDS.

3
 Credit-Default Swaps
The three parties (A, B and C) are referred to as the protection buyer, the
protection seller and the reference entity, respectively.

The CDS has a maturity date, T , and all payments cease at min(τ, T ) where τ
is the default time of the reference entity, party C
- not true for a CDS written on an index containing multiple reference entities
- in the index case there can be many default-leg payments, one for each
reference entity
- notional of the CDS, however, is reduced appropriately after each default.

We price a CDS by equating the risk-neutral value of the premium and default
legs and from this determining the fair annual spread that the protection buyer
must pay to the protection seller.

4
 Pricing a CDS: The Premium Leg
The Premium Leg:
Suppose the premium leg of the CDS has n payment times, t1 , . . . , tn = T
- we assume that default can only occur at one of these times.
Then fair value of the premium leg at time t = 0 is given by
n
X
P0 = s0T N Z0ti αi P(τ > ti ) (3)
i=1

where Z0ti is the discount factor, N is the notional and αi = ti − ti−1 .

Annualized CDS spread is then s0T

- use superscript ‘T ’ since CDS spread is maturity-dependent in practice.

5
 Pricing a CDS: The Default Leg
The Default Leg:
The fair value at time t = 0 of the default leg satisfies
n
X
D0 = (1 − R)N Z0ti P(τ = ti ) (4)
i=1

where R is the so-called recovery rate.

R = % of face value of the reference bond that is recovered by a bond-holder

upon default of the reference entity.

Common to assume that R is a fixed constant, e.g. 40%

- but in practice it is stochastic.

6
 Obtaining the Fair CDS Spread
The spread, s0T , is obtained by equating P0 with D0 so that the CDS has zero
value for both parties at the beginning of the contract.

This implies Pn ti
(1 − R) i=1 Z0 P(τ = ti )
s0T = Pn ti . (5)
i=1 Z0 αi P(τ > ti )

Can (and often should) make simple adjustments to (5) to allow for default
possibility in (ti−1 , ti ).

A plot of s0T against T then shows the term-structure of CDS spreads.

7
 Foundations of Financial Engineering
Structural Models for Pricing a CDS

Martin B. Haugh
Department of Industrial Engineering and Operations Research
Columbia University
 Structural Models for Pricing a CDS
Recall that CDS spread calculated as
Pn ti
(1 − R) i=1 Z0 P(τ = ti )
s0T = Pn ti . (6)
i=1 Z0 αi P(τ > ti )

To compute a specific value for s0T , must have a (risk-neutral) default model in
order to compute the various probabilities in (6).

Can certainly compute s0T using structural models as described earlier.

2
 Structural Models for Pricing a CDS
Recall the debt value lattice from Black-Cox model:
Debt Lattice
800.0
761.0 800.0
723.9 761.0 800.0
688.6 723.9 761.0 800.0
653.2 684.7 715.3 742.0 758.2*
623.6 658.2 699.9 758.2* 574.9* 435.9*
614.9 671.5 758.2* 574.9* 435.9* 330.5* 250.6*
650.0 758.2* 574.9* 435.9* 330.5* 250.6* 190.0* 144.1*

t = 0 t = 1 t = 2 t = 3 t = 4 t = 5 t = 6 t = 7

Can use lattice to build separate lattices for premium and default leg cash-flows
- trial and error could be used to find the fair premium, s0T .
Alternatively could use lattices to compute the required default probabilities
- can then use (6) to calculate s0T .
3
 Structural Models for Pricing a CDS
One of the main weaknesses of the structural approach to pricing CDS’s is that
the default event is said to be predictable.

Example: Consider the Black-Cox model where default is impossible in next

time period unless firm value is very close to the default boundary.

And if firm value is far from the default boundary then default in the next
several periods will also be impossible.

In contrast, if we let the length of a time period, ∆t, go to zero and the firm
value process is extremely close to the default boundary then default will
happen for sure very soon.

This is what we mean when we say default is predictable.

4
 Structural Models for Pricing a CDS
In the real-world, however, default is not predictable and the default of firms can
come as a complete surprise.

This is true for firms such as Enron and Parmalat where default was caused by
the discovery of huge accounting frauds.

Even when the market sees that a firm, e.g. Lehman Brothers, is in financial
difficulty so that the CDS spreads of that firm have widened, the actual default
event itself is still a surprise
- and unpredictable (in a mathematical sense).

Intensity models are commonly used to circumvent this problem

- default events in these models are unpredictable
- they do not model the economic value of the firm and hence are often
termed reduced-form models.

(Inhomogeneous) Poisson process intensity models are examples of deterministic

intensity models.

Cox process models are examples of stochastic intensity models.

5
 Foundations of Financial Engineering
Using (Inhomogeneous) Poisson Processes to Model Default

Martin B. Haugh
Department of Industrial Engineering and Operations Research
Columbia University
 Using Poisson Processes to Model Default
CDS spread now calculated as
Pn
(1 − R) i=1 Z0ti P(ti−1 < τ ≤ ti )
s0T = Pn ti (7)
i=1 Z0 αi P(τ > ti )

since we assume default event can take place any time in (ti−1 , ti ] for any i.

Suppose the arrival of default follows a Poisson process with parameter λ.

Then P(τ ≤ t) = 1 − e −λt and so we can easily compute s0T using (7).

2
 Using Poisson Processes to Model Default
Can also compute time t = 0 value, V0 , of a zero-coupon bond with face value F
and maturity T .

If the recovery-rate upon default is a known value, R, that is paid at the time of
default, τ , then we have
" Z T #
V0 = E0 e −rT F 1{τ >T} +
Q
e −rt RF 1{τ =t}
0
Z T
= e −(r+λ)T F + RF λe −(r+λ)t dt
0
λ  
= e −(r+λ)T F + RF 1 − e −(r+λ)T . (8)
r +λ

3
 Using an Inhomogeneous Poisson Process
A clear weakness with the Poisson model is that there is just a single parameter,
λ, that we are free to choose.

In practice, however, there are typically liquid CDS spreads s0T , for several values
of T
- typical values are T = 1, 3, 5, 7 and 10 years.

Unless term structure of credit spreads is constant then not possible to choose λ
so that model values of s0T coincide with corresponding market values for all
values of T .

Can resolve this problem using an inhomogeneous Poisson process

- arrival rate, λt , is now a deterministic function of time, t.

For such a process, probability of zero arrivals in interval (t, t + ∆t) is then
approximately e −λt ∆t .

4
 Using an Inhomogeneous Poisson Process
Using this we obtain the survival probability
Rt
− λs ds
P(τ > t) = e 0 . (9)

Can now use (9) to compute fair CDS spread according to our expression
Pn
T (1 − R) i=1 Z0ti P(ti−1 < τ ≤ ti )
s0 = Pn ti . (10)
i=1 Z0 αi P(τ > ti )

Can also calibrate the inhomogeneous Poisson process model to the CDS spreads
observed in the market.

In particular, can assume λt is piecewise constant on the intervals (0, T1 ],

(T1 , T2 ], . . . , (Tn−1 , Tn ] where T1 < T2 < · · · < Tn are CDS market maturities.

Can then use (10) to first calibrate λ0,T1 and to then calibrate each of
λT1 ,T2 , . . . , λTn−1 ,Tn in turn.
5
 Foundations of Financial Engineering
Stochastic Intensity Models

Martin B. Haugh
Department of Industrial Engineering and Operations Research
Columbia University
 Stochastic Intensity Models
While the inhomogeneous Poisson model can be calibrated to the CDS spreads
seen in the market, it has a glaring weakness:
credit spreads at all future times are known today at time t = 0
- follows because the intensity, λt , is deterministic.

In practice, however, credit spreads are stochastic

- so we would like to consider models where the default intensity is stochastic.

Will consider one important and tractable class of stochastic intensity models:
the doubly-stochastic or Cox process
- a Cox process is in fact an entire class of (point) processes.

2
 Stochastic Intensity Models
Before defining such a process let Nt be the number of “arrivals” up to time t.

When Nt follows a Poisson process we know Nt has a Poisson distribution with

parameter λt
Rt
- or parameter 0 λu du in the case of an inhomogeneous Poisson process.

A Cox process generalizes this by allowing the intensity λu to be stochastic.

In a Cox process, however, conditional on knowing {λu }0≤u≤T , the # of arrivals

in interval [s, t] where 0 ≤ s < t ≤ T has an inhomogeneous Poisson distribution
Rt
with parameter s λu du
- such a process remains very tractable.

3
 Cox Processes
Example: If τ is the first arrival of a Cox process then we can compute the
survival probability as follows:
 
P(τ > t) = E 1{τ >t}
  
= E E 1{τ >t} | {λu }0≤u≤t
 Rt 
− λ du
= E e 0 u (11)

where (11) follows from the fact that the point process is an inhomogenous
Poisson process on [0, t] conditional on {λu }0≤u≤t .

4
 Pricing a Bond with a Cox Process Default Model
Recall our earlier expression for the price of a zero-coupon bond with face value
F and maturity T :
λ  
V0 = e −(r+λ)T F + RF 1 − e −(r+λ)T . (12)
r +λ
– assumes recovery-rate is R is known and paid at time of default, τ .

Assuming a (risk-neutral) Cox process, can generalize (12) to obtain

" Z T #
Q −rT −rt
V0 = E0 e F 1{τ >T} + e RF 1{τ =t}
0
"Z #
 RT  T Rt
Q − (r+λu ) du Q − (r+λs ) ds
= E0 e 0 F + RF E0 λt e 0 dt
0
 RT  Z T  Rt 
− (r+λu ) du − (r+λs ) ds
= EQ
0 e 0 F + RF EQ
0 λ t e 0 dt. (13)
0

5
 Pricing a Bond with a Cox Process Default Model
A Cox process is only fully specified once we specify dynamics for λt .

We can consider a lattice model for λt but we also note that there are
continuous-time models
- e.g. the CIR model, for which (13) can be computed analytically.

Parameters of these models would then be chosen so that the model CDS
spreads, s0T , match market CDS spreads for different maturities, T
- this is the process of calibration.