Bjork Notes

Interest Rate Theory
Toronto 2010
Tomas Bjork
Tomas Bjork, 2010
Contents
1. Mathematics recap. (Ch 10-12)
2. Recap of the martingale approach. (Ch 10-12)
3. Incomplete markets (Ch 15)
4. Bonds and short rate models. (Ch 22-23)
5. Martingale models for the short rate. (Ch 24)
6. Forward rate models. (Ch 25)
7. Change of numeraire. (Ch 26)
8. LIBOR market models. (Ch 27)
Bjork,T. Arbitrage Theory in Continuous Time.
3:rd ed. 2009. Oxford University Press.
Tomas Bjork, 2010 1
1.
Mathematics Recap
Ch. 10-12
Tomas Bjork, 2010 2
Contents
1. Conditional expectations
2. Changing measures
3. The Martingale Representation Theorem
4. The Girsanov Theorem
Tomas Bjork, 2010 3
1.
Conditional Expectation
Tomas Bjork, 2010 4
Conditional Expectation
If F is a sigma-algebra and X is a random variable
which is F-measurable, we write this as X F.
If X F and if G F then we write E [X| G] for
the conditional expectation of X given the information
contained in G. Sometimes we use the notation E
G
[X].
The following proposition contains everything that we
will need to know about conditional expectations within
this course.
Tomas Bjork, 2010 5
Main Results
Proposition 1: Assume that X F, and that G F.
Then the following hold.
The random variable E [X| G] is completely determined by
the information in G so we have
E [X| G] G
If we have Y G then Y is completely determined by G so
we have
E [XY | G] = Y E [X| G]
In particular we have
E [Y | G] = Y
If H G then we have the law of iterated expectations
E [E [X| G]| H] = E [X| H]
E [X] = E [E [X| G]]
Tomas Bjork, 2010 6
2.
Changing Measures
Tomas Bjork, 2010 7
Absolute Continuity
Denition: Given two probability measures P and Q
on F we say that Q is absolutely continuous w.r.t.
P on F if, for all A F, we have
P(A) = 0 Q(A) = 0
We write this as
Q << P.
If Q << P and P << Q then we say that P and Q
are equivalent and write
Q P
Tomas Bjork, 2010 8
Equivalent measures
It is easy to see that P and Q are equivalent if and
only if
P(A) = 0 Q(A) = 0
or, equivalently,
P(A) = 1 Q(A) = 1
Two equivalent measures thus agree on all certain
events and on all impossible events, but can disagree
on all other events.
Simple examples:
All non degenerate Gaussian distributions on R are
equivalent.
If P is Gaussian on R and Q is exponential then
Q << P but not the other way around.
Tomas Bjork, 2010 9
Absolute Continuity ctd
Consider a given probability measure P and a random
variable L 0 with E
P
[L] = 1. Now dene Q by
Q(A) =
_
A
LdP
then it is easy to see that Q is a probability measure
and that Q << P.
A natural question is now if all measures Q << P
are obtained in this way. The answer is yes, and the
precise (quite deep) result is as follows.
Tomas Bjork, 2010 10
The Radon Nikodym Theorem
Consider two probability measures P and Q on (, F),
and assume that Q << P on F. Then there exists a
unique random variable L with the following properties
1. Q(A) =
_
A
LdP, A F
2. L 0, P a.s.
3. E
P
[L] = 1,
4. L F
The random variable L is denoted as
L =
dQ
dP
, on F
and it is called the Radon-Nikodym derivative of Q
w.r.t. P on F, or the likelihood ratio between Q and
P on F.
A simple example
The Radon-Nikodym derivative L is intuitively the local
scale factor between P and Q. If the sample space
is nite so = {
1
, . . . ,
n
} then P is determined by
the probabilities p
1
, . . . , p
n
where
p
i
= P(
i
) i = 1, . . . , n
Now consider a measure Q with probabilities
q
i
= Q(
i
) i = 1, . . . , n
If Q << P this simply says that
p
i
= 0 q
i
= 0
and it is easy to see that the Radon-Nikodym derivative
L = dQ/dP is given by
L(
i
) =
q
i
p
i
i = 1, . . . , n
If p
i
= 0 then we also have q
i
= 0 and we can dene
the ratio q
i
/p
i
arbitrarily.
If p
1
, . . . , p
n
as well as q
1
, . . . , q
n
are all positive, then
we see that Q P and in fact
dP
dQ
=
1
L
=
_
dQ
dP
_
1
as could be expected.
Computing expected values
A main use of Radon-Nikodym derivatives is for the
computation of expected values.
Suppose therefore that Q << P on F and that X is
a random variable with X F. With L = dQ/dP on
F then have the following result.
Proposition 3: With notation as above we have
E
Q
[X] = E
P
[L X]
The Abstract Bayes Formula
We can also use Radon-Nikodym derivatives in order to
compute conditional expectations. The result, known
as the abstract Bayes Formula, is as follows.
Theorem 4: Consider two measures P and Q with
Q << P on F and with
L
F
=
dQ
dP
on F
Assume that G F and let X be a random variable
with X F. Then the following holds
E
Q
[X| G] =
E
P
_
L
F
X
E
P
[L
F
| G]
Dependence of the -algebra
Suppose that we have Q << P on F with
L
F
=
dQ
dP
on F
Now consider smaller -algebra G F. Our problem
is to nd the R-N derivative
L
G
=
dQ
dP
on G
We recall that L
G
is characterized by the following
properties
1. Q(A) = E
P
_
L
G
I
A
A G
2. L
G
0
3. E
P
_
L
G
= 1
4. L
G
G
A natural guess would perhaps be that L
G
= L
F
, so
let us check if L
F
satises points 1-4 above.
By assumption we have
Q(A) = E
P
_
L
F
I
A
A F
Since G F we then have
Q(A) = E
P
_
L
F
I
A
A G
so point 1 above is certainly satised by L
F
. It is
also clear that L
F
satises points 2 and 3. It thus
seems that L
F
is also a natural candidate for the R-N
derivative L
G
, but the problem is that we do not in
general have L
F
G.
This problem can, however, be xed. By iterated
expectations we have, for all A G,
E
P
_
L
F
I
A
= E
P
_
E
P
_
L
F
I
A

Since A G we have
E
P
_
L
F
I
A
= E
P
_
L
F
I
A
Let us now dene L
G
by
L
G
= E
P
_
L
F
We then obviously have L

G
G and
Q(A) = E
P
_
L
G
I
A
A G
It is easy to see that also points 2-3 are satised so we
have proved the following result.
A formula for L
G
Proposition 5: If Q << P on F and G F then,
with notation as above, we have
L
G
= E
P
_
L
F

The likelihood process on a ltered space
We now consider the case when we have a probability
measure P on some space and that instead of just
one -algebra F we have a ltration, i.e. an increasing
family of -algebras {F
t
}
t0
.
The interpretation is as usual that F
t
is the information
available to us at time t, and that we have F
s
F
t
for s t.
Now assume that we also have another measure Q,
and that for some xed T, we have Q << P on F
T
.
We dene the random variable L
T
by
L
T
=
dQ
dP
on F
T
Since Q << P on F
T
we also have Q << P on F
t
for all t T and we dene
L
t
=
dQ
dP
on F
t
0 t T
For every t we have L
t
F
t
, so L is an adapted
process, known as the likelihood process.
The L process is a P martingale
We recall that
L
t
=
dQ
dP
on F
t
0 t T
Since F
s
F
t
for s t we can use Proposition 5 and
deduce that
L
s
= E
P
[L
t
| F
s
] s t T
and we have thus proved the following result.
Proposition: Given the assumptions above, the
likelihood process L is a P-martingale.
Where are we heading?
We are now going to perform measure transformations
on Wiener spaces, where P will correspond to the
objective measure and Q will be the risk neutral
measure.
For this we need dene the proper likelihood process L
and, since L is a P-martingale, we have the following
natural questions.
What does a martingale look like in a Wiener driven
framework?
Suppose that we have a P-Wiener process W and
then change measure from P to Q. What are the
properties of W under the new measure Q?
These questions are handled by the Martingale
Representation Theorem, and the Girsanov Theorem
respectively.
3.
The Martingale Representation Theorem
Intuition
Suppose that we have a Wiener process W under
the measure P. We recall that if h is adapted (and
integrable enough) and if the process X is dened by
X
t
= x
0
+
_
t
0
h
s
dW
s
then X is a a martingale. We now have the following
natural question:
Question: Assume that X is an arbitrary martingale.
Does it then follow that X has the form
X
t
= x
0
+
_
t
0
h
s
dW
s
for some adapted process h?
In other words: Are all martingales stochastic integrals
w.r.t. W?
Answer
It is immediately clear that all martingales can not be
written as stochastic integrals w.r.t. W. Consider for
example the process X dened by
X
t
=
_
0 for 0 t < 1
Z for t 1
where Z is an random variable, independent of W,
with E [Z] = 0.
X is then a martingale (why?) but it is clear (how?)
that it cannot be written as
X
t
= x
0
+
_
t
0
h
s
dW
s
for any process h.
Intuition
The intuitive reason why we cannot write
X
t
= x
0
+
_
t
0
h
s
dW
s
in the example above is of course that the random
variable Z has nothing to do with the Wiener process
W. In order to exclude examples like this, we thus need
an assumption which guarantees that our probability
space only contains the Wiener process W and nothing
else.
This idea is formalized by assuming that the ltration
{F
t
}
t0
is the one generated by the Wiener
process W.
The Martingale Representation Theorem
Theorem. Let W be a P-Wiener process and assume
that the ltation is the internal one i.e.
F
t
= F
W
t
= {W
s
; 0 s t}
Then, for every (P, F
t
)-martingale X, there exists a
real number x and an adapted process h such that
X
t
= x +
_
t
0
h
s
dW
s
,
i.e.
dX
t
= h
t
dW
t
.
Proof: Hard. This is very deep result.
Note
For a given martingale X, the Representation Theorem
above guarantees the existence of a process h such that
X
t
= x +
_
t
0
h
s
dW
s
,
The Theorem does not, however, tell us how to nd
or construct the process h.
4.
The Girsanov Theorem
Setup
Let W be a P-Wiener process and x a time horizon
T. Suppose that we want to change measure from P
to Q on F
T
. For this we need a P-martingale L with
L
0
= 1 to use as a likelihood process, and a natural
way of constructing this is to choose a process g and
then dene L by
_
dL
t
= g
t
dW
t
L
0
= 1
This denition does not guarantee that L 0, so we
make a small adjustment. We choose a process and
dene L by
_
dL
t
= L
t
t
dW
t
L
0
= 1
The process L will again be a martingale and we easily
obtain
L
t
= e
R
t
0

s
dW
s
1
2
R
t
0

2
s
ds
Thus we are guaranteed that L 0. We now change
measure form P to Q by setting
dQ = L
t
dP, on F
t
, 0 t T
The main problem is to nd out what the properties
of W are, under the new measure Q. This problem is
resolved by the Girsanov Theorem.
The Girsanov Theorem
Let W be a P-Wiener process. Fix a time horizon T.
Theorem: Choose an adapted process , and dene
the process L by
_
dL
t
= L
t
t
dW
t
L
0
= 1
Assume that E
P
[L
T
] = 1, and dene a new mesure Q
on F
T
by
dQ = L
t
dP, on F
t
, 0 t T
Then Q << P and the process W
Q
, dened by
W
Q
t
= W
t
_
t
0
s
ds
is Q-Wiener. We can also write this as
dW
t
=
t
dt +dW
Q
t
Changing the drift in an SDE
The single most common use of the Girsanov Theorem
is as follows.
Suppose that we have a process X with P dynamics
dX
t
=
t
dt +
t
dW
t
where and are adapted and W is P-Wiener.
We now do a Girsanov Transformation as above, and
the question is what the Q-dynamics look like.
From the Girsanov Theorem we have
dW
t
=
t
dt +dW
Q
t
and substituting this into the P-dynamics we obtain
the Q dynamics as
dX
t
= {
t
+
t
t
} dt +
t
dW
Q
t
Moral: The drift changes but the diusion is
unaected.
The Converse of the Girsanov Theorem
Let W be a P-Wiener process. Fix a time horizon T.
Theorem. Assume that:
Q << P on F
T
, with likelihood process
L
t
=
dQ
dP
, on F
t
0, t T
The ltation is the internal one .i.e.
F
t
= {W
s
; 0 s t}
Then there exists a process such that
_
dL
t
= L
t
t
dW
t
L
0
= 1
2.
The Martingale Approach
Ch. 10-12
Financial Markets
Price Processes:
S
t
=
_
S
0
t
, ..., S
N
t
Example: (Black-Scholes, S
0
:= B, S
1
:= S)
dS
t
= S
t
dt +S
t
dW
t
,
dB
t
= rB
t
dt.
Portfolio:
h
t
=
_
h
0
t
, ..., h
N
t
h
i
t
= number of units of asset i at time t.
Value Process:
V
h
t
=
N
i=0
h
i
t
S
i
t
= h
t
S
t
Self Financing Portfolios
Denition: (intuitive)
A portfolio is self-nancing if there is no exogenous
infusion or withdrawal of money. The purchase of a
new asset must be nanced by the sale of an old one.
Denition: (mathematical)
A portfolio is self-nancing if the value process
satises
dV
t
=
N
i=0
h
i
t
dS
i
t
Major insight:
If the price process S is a martingale, and if h is
self-nancing, then V is a martingale.
NB! This simple observation is in fact the basis of the
following theory.
Arbitrage
The portfolio u is an arbitrage portfolio if
The portfolio strategy is self nancing.
V
0
= 0.
V
T
0, P a.s.
P (V
T
> 0) > 0
Main Question: When is the market free of arbitrage?
First Attempt
Proposition: If S
0
t
, , S
N
t
are P-martingales, then
the market is free of arbitrage.
Proof:
Assume that V is an arbitrage strategy. Since
dV
t
=
N
i=0
h
i
t
dS
i
t
,
V is a P-martingale, so
V
0
= E
P
[V
T
] > 0.
This contradicts V
0
= 0.
True, but useless.
Example: (Black-Scholes)
dS
t
= S
t
dt +S
t
dW
t
,
dB
t
= rB
t
dt.
(We would have to assume that = r = 0)
We now try to improve on this result.
Choose S
0
as numeraire
Denition:
The normalized price vector Z is given by
Z
t
=
S
t
S
0
t
=
_
1, Z
1
t
, ..., Z
N
t
The normalized value process V

Z
is given by
V
Z
t
=
N
0
h
i
t
Z
i
t
.
Idea:
The arbitrage and self nancing concepts should be
independent of the accounting unit.
Invariance of numeraire
Proposition: One can show (see the book) that
S-arbitrage Z-arbitrage.
S-self-nancing Z-self-nancing.
Insight:
If h self-nancing then
dV
Z
t
=
N
1
h
i
t
dZ
i
t
Thus, if the normalized price process Z is a P-
martingale, then V
Z
is a martingale.
Second Attempt
Proposition: If Z
0
t
, , Z
N
t
are P-martingales, then
the market is free of arbitrage.
True, but still fairly useless.
Example: (Black-Scholes)
dS
t
= S
t
dt +S
t
dW
t
,
dB
t
= rB
t
dt.
dZ
1
t
= ( r)Z
1
t
dt +Z
1
t
dW
t
,
dZ
0
t
= 0dt.
We would have to assume risk-neutrality, i.e. that
= r.
Arbitrage
Recall that h is an arbitrage if
h is self nancing
V
0
= 0.
V
T
0, P a.s.
P (V
T
> 0) > 0
Major insight
This concept is invariant under an equivalent change
of measure!
Martingale Measures
Denition: A probability measure Q is called an
equivalent martingale measure (EMM) if and only
if it has the following properties.
Q and P are equivalent, i.e.
Q P
The normalized price processes
Z
i
t
=
S
i
t
S
0
t
, i = 0, . . . , N
are Q-martingales.
Wan now state the main result of arbitrage theory.
First Fundamental Theorem
Theorem: The market is arbitrage free
i
there exists an equivalent martingale measure.
Comments
It is very easy to prove that existence of EMM
imples no arbitrage (see below).
The other imnplication is technically very hard.
For discrete time and nite sample space the hard
part follows easily from the separation theorem for
convex sets.
For discrete time and more general sample space we
need the Hahn-Banach Theorem.
For continuous time the proof becomes technically
very hard, mainly due to topological problems. See
the textbook.
Proof that EMM implies no arbitrage
This is basically done above. Assume that there exists
an EMM denoted by Q. Assume that P(V
T
0) = 1
and P(V
T
> 0) > 0. Then, since P Q we also have
Q(V
T
0) = 1 and Q(V
T
> 0) > 0.
Recall:
dV
Z
t
=
N
1
h
i
t
dZ
i
t
Q is a martingale measure
V
Z
is a Q-martingale
V
0
= V
Z
0
= E
Q
_
V
Z
T
> 0
No arbitrage
Choice of Numeraire
The numeraire price S
0
t
can be chosen arbitrarily. The
most common choice is however that we choose S
0
as
the bank account, i.e.
S
0
t
= B
t
where
dB
t
= r
t
B
t
dt
Here r is the (possibly stochastic) short rate and we
have
B
t
= e
R
t
0
r
s
ds
Example: The Black-Scholes Model
dS
t
= S
t
dt +S
t
dW
t
,
dB
t
= rB
t
dt.
Look for martingale measure. We set Z = S/B.
dZ
t
= Z
t
( r)dt +Z
t
dW
t
,
Girsanov transformation on [0, T]:
_
dL
t
= L
t
t
dW
t
,
L
0
= 1.
dQ = L
T
dP, on F
T
Girsanov:
dW
t
=
t
dt +dW
Q
t
,
where W
Q
is a Q-Wiener process.
The Q-dynamics for Z are given by
dZ
t
= Z
t
[ r +
t
] dt +Z
t
dW
Q
t
.
Unique martingale measure Q, with Girsanov kernel
given by
t
=
r
.
Q-dynamics of S:
dS
t
= rS
t
dt +S
t
dW
Q
t
.
Conclusion: The Black-Scholes model is free of
arbitrage.
Pricing
We consider a market B
t
, S
1
t
, . . . , S
N
t
.
Denition:
A contingent claim with delivery time T, is a random
variable
X F
T
.
At t = T the amount X is paid to the holder of the
claim.
Example: (European Call Option)
X = max [S
T
K, 0]
Let X be a contingent T-claim.
Problem: How do we nd an arbitrage free price
process
t
[X] for X?
Solution
The extended market
B
t
, S
1
t
, . . . , S
N
t
,
t
[X]
must be arbitrage free, so there must exist a martingale
measure Q for (B
t
, S
t
,
t
[X]). In particular
t
[X]
B
t
must be a Q-martingale, i.e.
t
[X]
B
t
= E
Q
_
T
[X]
B
T
F
t
_
Since we obviously (why?) have
T
[X] = X
we have proved the main pricing formula.
Risk Neutral Valuation
Theorem: For a T-claim X, the arbitrage free price is
given by the formula
t
[X] = E
Q
_
e
R
T
t
r
s
ds
X
F
t
_
Example: The Black-Scholes Model
Q-dynamics:
dS
t
= rS
t
dt +S
t
dW
Q
t
.
Simple claim:
X = (S
T
),
t
[X] = e
r(Tt)
E
Q
[(S
T
)| F
t
]
Kolmogorov
t
[X] = F(t, S
t
)
where F(t, s) solves the Black-Scholes equation:
_
_
F
t
+rs
F
s
+
1
2
2
s
2
2
F
s
2
rF = 0,
F(T, s) = (s).
Problem
Recall the valuation formula
t
[X] = E
Q
_
e
R
T
t
r
s
ds
X
F
t
_
What if there are several dierent martingale measures
Q?
This is connected with the completeness of the
market.
Hedging
Def: A portfolio is a hedge against X (replicates
X) if
h is self nancing
V
T
= X, P a.s.
Def: The market is complete if every X can be
hedged.
Pricing Formula:
If h replicates X, then a natural way of pricing X is
t
[X] = V
h
t
When can we hedge?
Existence of hedge
Existence of stochastic integral

representation
Fix T-claim X.
If h is a hedge for X then
V
Z
T
=
X
B
T
h is self nancing, i.e.
dV
Z
t
=
K
1
h
i
t
dZ
i
t
Thus V
Z
is a Q-martingale.
V
Z
t
= E
Q
_
X
B
T
F
t
_
Lemma:
Fix T-claim X. Dene martingale M by
M
t
= E
Q
_
X
B
t
F
t
_
Suppose that there exist predictable processes
h
1
, , h
N
such that
M
t
= x +
N
i=1
_
t
0
h
i
s
dZ
i
s
,
Then X can be replicated.
Proof
We guess that
M
t
= V
Z
t
= h
B
t
1 +
N
i=1
h
i
t
Z
i
t
Dene: h
B
by
h
B
t
= M
t
i=1
h
i
t
Z
i
t
.
We have M
t
= V
Z
t
, and we get
dV
Z
t
= dM
t
=
N
i=1
h
i
t
dZt
i
,
so the portfolio is self nancing. Furthermore:
V
Z
T
= M
T
= E
Q
_
X
B
T
F
T
_
=
X
B
T
.
Second Fundamental Theorem
The second most important result in arbitrage theory
is the following.
Theorem:
The market is complete
i
the martingale measure Q is unique.
Proof: It is obvious (why?) that if the market
is complete, then Q must be unique. The other
implication is very hard to prove. It basically relies on
duality arguments from functional analysis.
Black-Scholes Model
Q-dynamics
dS
t
= rS
t
dt +S
t
dW
Q
t
,
dZ
t
= Z
t
dW
Q
t
M
t
= E
Q
_
e
rT
X
F
t
,
Representation theorem for Wiener processes
there exists g such that

M
t
= M(0) +
_
t
0
g
s
dW
Q
s
.
Thus
M
t
= M
0
+
_
t
0
h
1
s
dZ
s
,
with h
1
t
=
g
t
Z
t
.
Result:
X can be replicated using the portfolio dened by
h
1
t
= g
t
/Z
t
,
h
B
t
= M
t
h
1
t
Z
t
.
Moral: The Black Scholes model is complete.
Special Case: Simple Claims
Assume X is of the form X = (S
T
)
M
t
= E
Q
_
e
rT
(S
T
)
F
t
,
Kolmogorov backward equation M
t
= f(t, S
t
)
_
f
t
+rs
f
s
+
1
2
2
s
2
2
f
s
2
= 0,
f(T, s) = e
rT
(s).
Ito
dM
t
= S
t
f
s
dW
Q
t
,
so
g
t
= S
t
f
s
,
Replicating portfolio h:
h
B
t
= f S
t
f
s
,
h
1
t
= B
t
f
s
.
Interpretation: f(t, S
t
) = V
Z
t
.
Dene F(t, s) by
F(t, s) = e
rt
f(t, s)
so F(t, S
t
) = V
t
. Then
_
_
_
h
B
t
=
F(t,S
t
)S
t
F
s
(t,S
t
)
B
t
,
h
1
t
=
F
s
(t, S
t
)
where F solves the Black-Scholes equation
_
F
t
+rs
F
s
+
1
2
2
s
2
2
F
s
2
rF = 0,
F(T, s) = (s).
Main Results
The market is arbitrage free There exists a
martingale measure Q
The market is complete Q is unique.
Every X must be priced by the formula
t
[X] = E
Q
_
e
R
T
t
r
s
ds
X
F
t
_
for some choice of Q.
In a non-complete market, dierent choices of Q
will produce dierent prices for X.
For a hedgeable claim X, all choices of Q will
produce the same price for X:
t
[X] = V
t
= E
Q
_
e
R
T
t
r
s
ds
X
F
t
_
Completeness vs No Arbitrage
Rule of Thumb
Question:
When is a model arbitrage free and/or complete?
Answer:
Count the number of risky assets, and the number of
random sources.
R = number of random sources
N = number of risky assets
Intuition:
If N is large, compared to R, you have lots of
possibilities of forming clever portfolios. Thus lots
of chances of making arbitrage prots. Also many
chances of replicating a given claim.
Rule of thumb
Generically, the following hold.
The market is arbitrage free if and only if
N R
The market is complete if and only if
N R
Example:
The Black-Scholes model.
dS
t
= S
t
dt +S
t
dW
t
,
dB
t
= rB
t
dt.
For B-S we have N = R = 1. Thus the Black-Scholes
model is arbitrage free and complete.
Stochastic Discount Factors
Given a model under P. For every EMM Q we dene
the corresponding Stochastic Discount Factor, or
SDF, by
D
t
= e
R
t
0
r
s
ds
L
t
,
where
L
t
=
dQ
dP
, on F
t
There is thus a one-to-one correspondence between
EMMs and SDFs.
The risk neutral valuation formula for a T-claim X can
now be expressed under P instead of under Q.
Proposition: With notation as above we have
t
[X] =
1
D
t
E
P
[D
T
X| F
t
]
Proof: Bayes formula.
Martingale Property of S D
Proposition: If S is an arbitrary price process, then
the process
S
t
D
t
is a P-martingale.
Proof: Bayes formula.
3.
Incomplete Markets
Ch. 15
Derivatives on Non Financial Underlying
Recall: The Black-Scholes theory assumes that the
market for the underlying asset has (among other
things) the following properties.
The underlying is a liquidly traded asset.
Shortselling allowed.
Portfolios can be carried forward in time.
There exists a large market for derivatives, where the
underlying does not satisfy these assumptions.
Examples:
Weather derivatives.
Derivatives on electric energy.
CAT-bonds.
Typical Contracts
Weather derivatives:
Heating degree days. Payo at maturity T is
given by
Z = max {X
T
30, 0}
where X
T
is the (mean) temperature at some place.
Electricity option:
The right (but not the obligation) to buy, at time
T, at a predetermined price K, a constant ow of
energy over a predetermined time interval.
CAT bond:
A bond for which the payment of coupons and
nominal value is contingent on some (well specied)
natural disaster to take place.
Problems
Weather derivatives:
The temperature is not the price of a traded asset.
Electricity derivatives:
Electric energy cannot easily be stored.
CAT-bonds:
Natural disasters are not traded assets.
We will treat all these problems within a factor model.
Typical Factor Model Setup
Given:
An underlying factor process X, which is not the
price process of a traded asset, with dynamics under
the objective probability measure P as
dX
t
= (t, X
t
) dt + (t, X
t
) dW
t
.
A risk free asset with dynamics
dB
t
= rB
t
dt,
Problem:
Find arbitrage free price
t
[Z] of a derivative of the
form
Z = (X
T
)
Concrete Examples
Assume that X
t
is the temperature at time t at the
village of Peniche (Portugal).
Heating degree days:
(X
T
) = 100 max {X
T
30, 0}
Holiday Insurance:
(X
T
) =
_
_
_
1000, if X
T
< 20
0, if X
T
20
Question
Is the price
t
[] uniquely determined by the P-
dynamics of X, and the requirement of an arbitrage
free derivatives market?
NO!!
WHY?
Stock Price Model Factor Model
Black-Scholes:
dS
t
= S
t
dt +S
t
dW
t
,
dB
t
= rB
t
dt.
Factor Model:
dX
t
= (t, X
t
)dt +(t, X
t
)dW
t
,
dB
t
= rB
t
dt.
What is the dierence?
Answer
X is not the price of a traded asset!
We can not form a portfolio based on X.
1. Rule of thumb:
N = 0, (no risky asset)
R = 1, (one source of randomness, W)
We have N < R. The exogenously given market,
consisting only of B, is incomplete.
2. Replicating portfolios:
We can only invest money in the bank, and then sit
down passively and wait.
We do not have enough underlying assets in order
to price X-derivatives.
There is not a unique price for a particular
derivative.
In order to avoid arbitrage, dierent derivatives
have to satisfy internal consistency relations.
If we take one benchmark derivative as given,
then all other derivatives can be priced in terms of
the market price of the benchmark.
We consider two given claims (X
T
) and (X
T
). We
assume they are traded with prices
t
[] = f(t, X
t
)
t
[] = g(t, X
t
)
Program:
Form portfolio based on and . Use Ito on f and
g to get portfolio dynamics.
dV = V
_
u
f
df
f
+u
g
dg
g
_
Choose portfolio weights such that the dW term
vanishes. Then we have
dV = V kdt,
(synthetic bank with k as the short rate)
Absence of arbitrage implies
k = r
Read o the relation k = r!
From Ito:
df = f
f
dt +f
f
dW,
where
_

f
=
f
t
+f
x
+
1
2
2
f
xx
f
,
f
=
f
x
f
.
Portfolio dynamics
dV = V
_
u
f
df
f
+u
g
dg
g
_
.
Reshuing terms gives us
dV = V
_
u
f
f
+u
g
g
_
dt +V
_
u
f
f
+u
g
g
_
dW.
Let the portfolio weights solve the system
_
u
f
+u
g
= 1,
u
f
f
+u
g
g
= 0.
u
f
=

g
f

g
,
u
g
=

f
f

g
,
Portfolio dynamics
dV = V
_
u
f
f
+u
g
g
_
dt.
i.e.
dV = V
_
f

f
f

g
_
dt.
Absence of arbitrage requires
f

f
f

g
= r
which can be written as
g
r
g
=

f
r
f
.
g
r
g
=

f
r
f
.
Note!
The quotient does not depend upon the particular
choice of contract.
Result
Assume that the market for X-derivatives is free of
arbitrage. Then there exists a universal process ,
such that
f
(t) r
f
(t)
= (t, X
t
),
holds for all t and for every choice of contract f.
NB: The same for all choices of f.
= Risk premium per unit of volatility
= Market Price of Risk (cf. CAPM).
= Sharpe Ratio
Slogan:
On an arbitrage free market all X-derivatives have
the same market price of risk.
The relation
f
r
f
=
is actually a PDE!
Pricing Equation
_
_
_
f
t
+{ } f
x
+
1
2
2
f
xx
rf = 0
f(T, x) = (x),
P-dynamics:
dX = (t, X)dt +(t, X)dW.
Can we solve the PDE?
No!!
Why??
Answer
Recall the PDE
_
_
_
f
t
+{ } f
x
+
1
2
2
f
xx
rf = 0
f(T, x) = (x),
In order to solve the PDE we need to know .
is not given exogenously.
is not determined endogenously.
Question:
Who determines ?
Answer:
THE MARKET!
Interpreting
Recall that the f dynamics are
df = f
f
dt +f
f
dW
t
and is dened as
f
(t) r
f
(t)
= (t, X
t
),
measures the aggregate risk aversion in the
market.
If is big then the market is highly risk averse.
If is zero then the market is risk netural.
If you make an assumption about , then you
implicitly make an assumption about the aggregate
risk aversion of the market.
Moral
Since the market is incomplete the requirement of
an arbitrage free market will not lead to unique
prices for X-derivatives.
Prices on derivatives are determined by two main
factors.
1. Partly by the requirement of an arbitrage free
derivative market. All pricing functions satises
the same PDE.
2. Partly by supply and demand on the market.
These are in turn determined by attitude towards
risk, liquidity consideration and other factors. All
these are aggregated into the particular used
(implicitly) by the market.
We recall the PDE
_
_
_
f
t
+{ } f
x
+
1
2
2
f
xx
rf = 0
f(T, x) = (x),
Using Feynman-Kac we obtain a risk neutral valuation
formula.
f(t, x) = e
r(Tt)
E
Q
t,x
[(X
T
)]
Q-dynamics:
dX
t
= { } dt +dW
Q
t
Price = expected value of future payments
The expectation should not be taken under the
objective probabilities P, but under the risk
adjusted probabilities Q.
Interpretation of the risk adjusted
probabilities
The risk adjusted probabilities can be interpreted as
probabilities in a (ctuous) risk neutral world.
When we compute prices, we can calculate as if
we live in a risk neutral world.
This does not mean that we live in, or think that
we live in, a risk neutral world.
The formulas above hold regardless of the attitude
towards risk of the investor, as long as he/she prefers
more to less.
Diversication argument about
If the risk factor is idiosyncratic and diversiable,
then one can argue that the factor should not be
priced by the market. Compare with APT.
Mathematically this means that = 0, i.e. P = Q,
i.e. the risk neutral distribution coincides with
the objective distribution.
We thus have the actuarial pricing formula
f(t, x) = e
r(Tt)
E
P
t,x
[(X
T
)]
where we use the objective probabiliy measure P.
Modeling Issues
Temperature:
A standard model is given by
dX
t
= {m(t) bX
t
} dt +dW
t
,
where m is the mean temperature capturing
seasonal variations. This often works reasonably
well.
Electricity:
A (naive) model for the spot electricity price is
dS
t
= S
t
{m(t) a lnS
t
} dt +S
t
dW
t
This implies lognormal prices (why?). Electricty
prices are however very far from lognormal, because
of spikes in the prices. Complicated.
CAT bonds:
Here we have to use the theory of point processes
and the theory of extremal statistics to model
natural disasters. Complicated.
Martingale Analysis
Model: Under P we have
dX
t
= (t, X
t
) dt + (t, X
t
) dW
t
,
dB
t
= rB
t
dt,
We look for martingale measures. Since B is the only
traded asset we need to nd Q P such that
B
t
B
t
= 1
is a Q martingale.
Result: In this model, every Q P is a martingale
measure.
Girsanov
dL
t
= L
t
t
dW
t
P-dynamics
dX
t
= (t, X
t
) dt + (t, X
t
) dW
t
,
dL
t
= L
t
t
dW
t
dQ = L
t
dP on F
t
Girsanov:
dW
t
=
t
dt +dW
Q
t
Martingale pricing:
F(t, x) = e
r(Tt)
E
Q
[Z| F
t
]
Q-dynamics of X:
dX
t
= {(t, X
t
) + (t, X
t
)
t
} dt + (t, X
t
) dW
Q
t
,
Result: We have
t
=
t
, i.e,. the Girsanov kernel
equals minus the market price of risk.
Several Risk Factors
We recall the dynamics of the f-derivative
df = f
f
dt +f
f
dW
t
and the Market Price of Risk
f
r
f
= , i.e.
f
r =
f
.
In a multifactor model of the type
dX
t
= (t, X
t
) dt +
n
i=1
i
(t, X
t
) dW
i
t
,
it follows from Girsanov that for every risk factor W
i
there will exist a market price of risk
i
=
i
such
that
f
r =
n
i=1
i
Compare with CAPM.
4.
Bonds and Interest Rates.
Short Rate Models
Ch. 22-23
Denitions
Bonds:
T-bond = zero coupon bond, paying $1 at the date of
maturity T.
p(t, T) = price, at t, of a T-bond.
p(T, T) = 1.
Main Problem
Investigate the term structure, i.e. how prices of
bonds with dierent dates of maturity are related
to each other.
Compute arbitrage free prices of interest rate
derivatives (bond options, swaps, caps, oors etc.)
Risk Free Interest Rates
At time t:
Sell one S-bond
Buy exactly p(t, S)/p(t, T) Tbonds
Net investment at t: $0.
At time S:
Pay $1
At time T:
Collect $p(t, S)/p(t, T) 1
Net Eect
The contract is made at t.
An investment of 1 at time S has yielded
p(t, S)/p(t, T) at time T.
The equivalent constant rates, R, are given as the
solutions to
Continuous rate:
e
R(TS)
1 =
p(t, S)
p(t, T)
Simple rate:
[1 +R (T S)] 1 =
p(t, S)
p(t, T)
Continuous Interest Rates
1. The forward rate for the period [S, T],
contracted at t is dened by
R(t; S, T) =
log p(t, T) log p(t, S)
T S
.
2. The spot rate, R(S, T), for the period [S, T] is
dened by
R(S, T) = R(S; S, T).
3. The instantaneous forward rate at T, conracted
at t is dened by
f(t, T) =
log p(t, T)
T
= lim
ST
R(t; S, T).
4. The instantaneous short rate at t is dened by
r(t) = f(t, t).
Simple Rates (LIBOR)
1. The simple forward rate L(t;S,T)for the period
[S, T], contracted at t is dened by
L(t; S, T) =
1
T S

p(t, S) p(t, T)
p(t, T)
2. The simple spot rate, L(S, T), for the period
[S, T] is dened by
L(S, T) =
1
T S

1 p(S, T)
p(S, T)
Practical Formula (LIBOR)
The simple spot rate, L(T, T + ), for the period
[T, T +] is given by
p(T, T +) =
1
1 +L(T, T +)
i.e.
L =
1

1 p
p
Bond prices forward rates
p(t, T) = p(t, s) exp
_
_
T
s
f(t, u)du
_
,
p(t, T) = exp
_
_
T
t
f(t, s)ds
_
.
Interest Rate Options
Problem:
We want to price, at t, a European Call, with exercise
date S, and strike price K, on an underlying T-bond.
(t < S < T).
Naive approach: Use Black-Scholess formula.
F(t, p) = pN [d
1
] e
r(St)
KN [d
2
] .
d
1
=
1
S t
_
ln
_
p
K
_
+
_
r +
1
2
2
_
(S t)
_
,
d
2
= d
1
S t.
where
p = p(t, T)
Is this allowed?
p shall be the price of a traded asset. OK!
The volatility of p must be constant. Here we have
a problem because of pull-to-par, i.e. the fact that
p(T, T) = 1. Bond volatilities will tend to zero as
the bond approaches the time of maturity.
The short rate must be constant and
deterministic. Here the approach collapses
completely, since the whole point of studying
bond prices lies in the fact that interest rates are
stochastic.
There is some hope in the case when the remaining
time to exercise the option is small in relation to the
remaining time to maturity of the underlying bond
(why?).
Deeply felt need
A consistent arbitrage free model for the
bond market
Stochastic interest rates
We assume that the short rate r is a stochastic process.
Money in the bank will then grow according to:
_
dB(t) = r(t)B(t)dt,
B(0) = 1.
i.e.
B(t) = e
R
t
0
r(s)ds
Models for the short rate
Model: (In reality)
P:
dr = (t, r)dt +(t, r)dW,
dB = r(t)Bdt.
Question: Are bond prices uniquely determined
by the P-dynamics of r, and the requirement of an
arbitrage free bond market?
NO!!
WHY?
Stock Models Interest Rates
Black-Scholes:
dS = Sdt +Sdw,
dB = rBdt.
Interest Rates:
dr = (t, r)dt +(t, r)dW,
dB = r
t
Bdt.
Question: What is the dierence?
Answer: The short rate r is not the price of a
traded asset!
1. Meta-Theorem:
N = 0, (no risky asset)
R = 1, (one source of randomness, W)
We have M < R. The exongenously given market,
consisting only of B, is incomplete.
2. Replicating portfolios:
We can only invest money in the bank, and then sit
down passively and wait.
We do not have enough underlying assets in order
to price bonds.
There is not a unique price for a particular
Tbond.
In order to avoid arbitrage, bonds of dierent
maturities have to satisfy internal consistency
relations.
If we take one benchmark T
0
-bond as given, then
all other bonds can be priced in terms of the market
price of the benchmark bond.
Assumption:
p(t, T) = F(t, r
t
; T)
p(t, T) = F
T
(t, r
t
),
F
T
(T, r) = 1.
Program
Form portfolio based on T and Sbonds. Use Ito
on F
T
(t, r(t)) to get bond- and portfolio dynamics.
dV = V
_
u
T
dF
T
F
T
+u
S
dF
S
F
S
_
Choose portfolio weights such that the dW term
vanishes. Then we have
dV = V kdt,
(synthetic bank with k as the short rate)
Absence of arbitrage k = r .
Read o the relation k = r!
From Ito:
dF
T
= F
T
T
dt +F
T
T
d

W,
where _
_
_
T
=
F
T
t
+F
T
r
+
1
2
2
F
T
rr
F
T
,
T
=
F
T
r
F
T
.
Portfolio dynamics
dV = V
_
u
T
dF
T
F
T
+u
S
dF
S
F
S
_
.
Reshuing terms gives us
dV = V
_
u
T
T
+u
S
S
_
dt+V
_
u
T
T
+u
S
S
_
dW.
Let the portfolio weights solve the system
_
u
T
+u
S
= 1,
u
T
T
+u
S
S
= 0.
_
u
T
=

S
S
,
u
S
=

T
S
,
Portfolio dynamics
dV = V
_
u
T
T
+u
S
S
_
dt.
i.e.
dV = V
_
T

T
T

S
_
dt.
Absence of arbitrage requires
T

T
T

S
= r
which can be written as
S
(t) r(t)
S
(t)
=

T
(t) r(t)
T
(t)
.
S
(t) r(t)
S
(t)
=

T
(t) r(t)
T
(t)
.
Note!
The quotient does not depend upon the particular
choice of maturity date.
Result
Assume that the bond market is free of arbitrage. Then
there exists a universal process , such that
T
(t) r(t)
T
(t)
= (t),
holds for all t and for every choice of maturity T.
NB: The same for all choices of T.
= Risk premium per unit of volatility
= Market Price of Risk (cf. CAPM).
Slogan:
On an arbitrage free market all bonds have the same
market price of risk.
The relation
T
r
T
=
is actually a PDE!
The Term Structure Equation
_
_
_
F
T
t
+{ } F
T
r
+
1
2
2
F
T
rr
rF
T
= 0,
F
T
(T, r) = 1.
P-dynamics:
dr = (t, r)dt +(t, r)dW.
=

T
r
T
, for all T
In order to solve the TSE we need to know .
General Term Structure Equation
Contingent claim:
X = (r(T))
Result:
The price is given by
[t; X] = F(t, r(t))
where F solves
F
t
+{ } F
r
+
1
2
2
F
rr
rF = 0,
F(T, r) = (r).
In order to solve the TSE we need to know .
Who determines ?
THE MARKET!
Moral
Since the market is incomplete the requirement of
an arbitrage free bond market will not lead to unique
bond prices.
Prices on bonds and other interest rate derivatives
are determined by two main factors.
1. Partly by the requirement of an arbitrage free
bond market (the pricing functions satises the
TSE).
2. Partly by supply and demand on the market.
These are in turn determined by attitude towards
risk, liquidity consideration and other factors. All
these are aggregated into the particular used
(implicitly) by the market.
Using FeynmacKac we obtain
F(t, r; T) = E
Q
t,r
_
e
R
T
t
r(s)ds
1
_
.
Q-dynamics:
dr = { }dt +dW
[t; X] = E
Q
t,r
_
e
R
T
t
r(s)ds
X
_
Q-dynamics:
dr = { }dt +dW
Price = expected value of future payments
The expectation should not be taken under the
objective probabilities P, but under the risk
adjusted probabilities Q.
Interpetation of the risk adjusted
probabilities
The risk adjusted probabilities can be interpreted as
probabilities in a (ctuous) risk neutral world.
When we compute prices, we can calculate as if
we live in a risk neutral world.
This does not mean that we live in, or think that
we live in, a risk neutral world.
The formulas above hold regardless of the attitude
towards risk of the investor, as long as he/she prefers
more to less.
Martingale Analysis
Model: Under P we have
dr
t
= (t, r
t
) dt + (t, r
t
) dW
t
,
dB
t
= rB
t
dt,
We look for martingale measures. Since B is the only
traded asset we need to nd Q P such that
B
t
B
t
= 1
is a Q martingale.
Result: In a short rate model, every Q P is a
martingale measure.
Girsanov
dL
t
= L
t
t
dW
t
P-dynamics
dr
t
= (t, r
t
) dt + (t, r
t
) dW
t
,
dL
t
= L
t
t
dW
t
dQ = L
t
dP on F
t
Girsanov:
dW
t
=
t
dt +dW
Q
t
Martingale pricing:
[t; Z] = E
Q
_
e
R
t
0
r
s
ds
Z
F
t
_
Q-dynamics of r:
dr
t
= {(t, r
t
) + (t, r
t
)
t
} dt + (t, r
t
) dW
Q
t
,
Result: We have
t
=
t
, i.e,. the Girsanov kernel
equals minus the market price of risk.
5.
Martingale Models for the Short Rate
Ch. 24
Martingale Modelling
Recall:
t
[X] = E
Q
_
e
R
t
0
r
s
ds
X
F
t
_
All prices are determined by the Q-
dynamics of r.
Model dr directly under Q!
Problem: Parameter estimation!
Pricing under risk adjusted probabilities
Q-dynamics:
dr = (t, r)dt +(t, r)dW
where W denotes a Q-Wiener process.
t
[X] = E
Q
_
e
R
T
t
r
s
ds
X
F
t
_
p(t, T) = E
Q
_
e
R
T
t
r
s
ds
1
F
t
_
The case X = (r
T
):
price given by
t
[X] = F (t, r
t
)
_
F
t
+F
r
+
1
2
2
F
rr
rF = 0,
F(T, r) = (r(T)).
1. Vasicek
dr = (b ar) dt +dW,
2. Cox-Ingersoll-Ross
dr = (b ar) dt +
rdW,
3. Dothan
dr = ardt +rdW,
4. Black-Derman-Toy
dr = (t)rdt +(t)rdW,
5. Ho-Lee
dr = (t)dt +dW,
6. Hull-White (extended Vasicek)
dr = {(t) ar} dt +dW,
Bond Options
European call on a T-bond with strike price K and
delivery date S.
X = max [p(S, T) K, 0]
X = max
_
F
T
(S, r
S
) K, 0
We have
t
[X] = F(t, r
t
)
where F solves the PDE
_
_
_
F
t
+F
r
+
1
2
2
F
rr
rF = 0,
F(S, r) = (r).
and where is dened by
(r) = max
_
F
T
(S, r) K, 0

To solve the pricing PDE for F we need to calculate
(r) = max
_
F
T
(S, r) K, 0
and for this we need to compute the theoretical bond

price function F
T
. We thus also need to solve the
PDE
_
_
_
F
T
t
+F
T
r
+
1
2
2
F
T
rr
rF
T
= 0,
F
T
(T, r) = 1.
Lots of equations!
Need analytic solutions.
Ane Term Structures
We have an Ane Term Structure if
F(t, r; T) = e
A(t,T)B(t,T)r
,
where A and B are deterministic functions.
Moral: If you want to obtain analytical formulas, then
you must have an ATS.
Problem: How do we specify and in order to have
an ATS?
Main Result for ATS
Proposition: Assume that and are of the form
(t, r) = (t)r +(t),
2
(t, r) = (t)r +(t).
Then the model admits an ane term structure
F(t, r; T) = e
A(t,T)B(t,T)r
,
where A and B satisfy the system
_
B
t
(t, T) = (t)B(t, T) +
1
2
(t)B
2
(t, T) 1,
B(T; T) = 0.
_
A
t
(t, T) = (t)B(t, T)
1
2
(t)B
2
(t, T),
A(T; T) = 0.
Parameter Estimation
Suppose that we have chosen a specic model, e.g.
H-W . How do we estimate the parameters a, b, ?
Naive answer:
Use standard methods from statistical theory.
WRONG!!
The parameters are Q-parameters.
Our observations are not under Q, but under P.
Standard statistical techniques can not be used.
We need to know the market price of risk ().
Who determines ?
The Market!
We must get price information from the market
in order to estimate parameters.
Inversion of the Yield Curve
Q-dynamics with parameter list :
dr = (t, r; )dt +(t, r; )dW
Theoretical term structure

p(0, T; ); T 0
Observed term structure
p
(0, T); T 0.
Requirement:
A model such that the theoretical prices of today
coincide with the observed prices of today. We want
to choose tha parameter vector such that
p(0, T; ) p
(0, T); T 0
Number of equations = (one for each T).
Number of unknowns = number of parameters.
Need:
Innite parameter list.
The time dependent function in Hull-White is
precisely such an innite parameter list (one parameter
for every t).
Result
Hull-White can be calibrated exactly to any initial term
strucutre. The calibrated model has the form
p(t, T) =
p
(0, T)
p
(0, t)
e
C(t,r
t
)
where C is given by
B(t, T)f
(0, t)

2
2a
2
B
2
(t, T)
_
1 e
2aT
_
B(t, T)r
t
There are analytical formulas for interest rate options.
Short rate models
Pro:
Easy to model r.
Analytical formulas for bond prices and bond
options.
Con:
Inverting the yield curve can be hard work.
Hard to model a exible volatility
structure for forward rates.
With a one factor model, all points on the yield
curve are perfectly correlated.
6.
Forward Rate Models
Ch. 25
Heath-Jarrow-Morton
Idea:
Model the dynamics for the entire yield curve.
The yield curve itself (rather than the short rate r)
is the explanatory variable.
Model forward rates. Use observed yield curve as
boundary value.
Dynamics:
df(t, T) = (t, T)dt +(t, T)dW(t),
f(0, T) = f
(0, T).
One SDE for every xed maturity time T.
Existence of martingale measure
f(t, T) =
log p(t, T)
T
p(t, T) = exp
_
_
T
t
f(t, s)ds
_
Thus:
Specifying forward rates.
Specifying bond prices.

Thus:
No arbitrage
restrictions on and .
Strategy
Start with P-dynamics for the forward rates
df(t, T) = (t, T)dt +(t, T)d

W(t)
where

W is P-Wiener.
Compute the corresponding bond price dynamics.
Do a Girsanov transformation P Q.
The Q-dynamics must then have the form
dp(t, T) = r
t
p(t, T)dt +p(t, T)v(t, T)dW(t)
where W is Q-Wiener.
Practical Toolbox
df(t, T) = (t, T)dt +(t, T)d

W
dp(t, T) = p(t, T)
_
r(t) +A(t, T) +
1
2
S(t, T)
2
_
dt
+ p(t, T)S(t, T)d

W
_
_
A(t, T) =
_
T
t
(t, s)ds,
S(t, T) =
_
T
t
(t, s)ds
Girsanov
_
dL(t) = L(t)g
(t)d

W(t),
L(0) = 1.
From Girsanov we have
d

W
t
= gdt +dW
t
where W is Q-Wiener.
From the toolbox, the Q-dynamics are obtained as
dp(t, T) = p(t, T)r(t)dt
+
_
A(t, T) +
1
2
||S(t, T)||
2
+S(t, T)g(t)
_
dt
+ p(t, T)S(t, T)dW(t),
Proposition:
There exists a martingale measure
There exists process g(t) = [g

1
(t), g
d
(t)]
s.t.
A(t, T) +
1
2
||S(t, T)||
2
+S(t, T)g(t) = 0, t, T
Taking T-derivatives we obtain the alternative formula
(t, T) = (t, T)
_
T
t
(t, s)ds (t, T)g(t), t, T

Moral for Modeling
Specify arbitrary volatilities (t, T).
Fix d benchmark maturities T
1
, , T
d
. For these
maturities, specify drift terms (t, T
1
), (t, T
1
).
The Girsanov kernel is uniquely determined (for each
xed t) by
d
i=1
i
(t, T
j
)g
i
(t) =
d
i=1
i
(t, T
j
)
_
T
0
i
(t, s)ds
(t, T
j
), j = 1, d.
Thus Q is uniquely determined.
All other drift terms will be uniquely dened by
(t, T) = (t, T)
_
T
t
(t, s)ds(t, T)g(t), t, T
Martingale Modelling
Q-dynamics:
df(t, T) = (t, T)dt +(t, T)dW(t)
Specifying forward rates specifying bond prices.
Under Q all bond prices have r as the local rate of
return.
Thus, martingale modeling = restrictions on
and .
Which?
Martingale modeling
Recall:
(t, T) = (t, T)
_
T
t
(t, s)ds (t, T)g(t), t, T
Martingale modeling g = 0
Theorem: (HJM drift Condition) The bond market is
arbitrage free if and only if
(t, T) = (t, T)
_
T
t
(t, s)ds.
Moral: Volatility can be specied freely. The forward
rate drift term is then uniquely determined.
Musiela parametrization
Parameterize forward rates by the time to maturity x,
rather than time of maturity T.
Def:
r(t, x) = f(t, t +x).
Q-dynamics:
dr(t, x) = (t, x)dt +(t, x)dW.
What are the relations between and under Q?
Compare with HJM!
df(t, T) = (t, T)dt +
0
(t, T)dW.
where we use
0
to denote the HJM volatility.
dr(t, x) = d [f(t, t +x)]
= df(t, t +x) +f
T
(t, t +x)dt
= {(t, t +x) +r
x
(t, x)} dt +
0
(t, t +x)dW
(t, x) = (t, t +x) +r
x
(t, x)
(t, x) =
0
(t, t +x).
HJM-condition:
(t, T) =
0
(t, T)
_
T
t
0
(t, s)ds.
Substitute!
The Musiela Equation
dr(t, x) =
_

x
r(t, x) +(t, x)
_
x
0
(t, y)dy
_
dt
+ (t, x)dW
When is deterministic this is a linear equation
in innite dimensional space. Connections to control
theory.
Forward Rate Models
Pro:
Easy to model exible volatility structure for forward
rates.
Easy to include multiple factors.
Con:
The short rate will typically not be a Markov process.
Computational problems.
7.
Change of Numeraire
Ch. 26
Change of Numeraire
Valuation formula:
t
[X] = E
Q
_
e
R
T
t
r
s
ds
X
F
t
_
Hard to compute. Double integral.
Note: If X and r are independent then
t
[X] = E
Q
_
e
R
T
t
r
s
ds
F
t
_
E
Q
[X| F
t
]
= p(t, T) E
Q
[X| F
t
] .
Nice! We do not have to compute p(t, T). It can be
observed directly on the market!
Single integral!
Sad Fact: X and r are (almost) never independent!
Idea
Use T-bond (for a xed T) as numeraire. Dene the
T-forward measure Q
T
by the requirement that
(t)
p(t, T)
is a Q
T
-martingale for every price process (t).
Then
t
[X]
p(t, T)
= E
T
_
T
[X]
p(T, T)
F
t
_
T
[X] = X, p(T, T) = 1.
t
[X] = p(t, T)E
T
[X| F
t
]
Do such measures exist?.
The forward measure takes care of the stochastics
over the interval [t, T].
Enormous computational advantages.
Useful for interest rate derivatives, currency derivatives
and derivatives dened by several underlying assets.
General change of numeraire.
Idea: Use a xed asset price process S
t
as numeraire.
Dene the measure Q
S
by the requirement that
(t)
S
t
is a Q
S
-martingale for every arbitrage free price process
(t).
Constructing Q
S
Fix a T-claim X. From general theory:
0
[X] = E
Q
_
X
B
T
_
Assume that Q
S
exists and denote
L
t
=
dQ
S
dQ
, on F
t
Then
0
[X]
S
0
= E
S
_
T
[X]
S
T
_
= E
S
_
X
S
T
_
= E
Q
_
L
T
X
S
T
_
Thus we have
0
[X] = E
Q
_
L
T
X S
0
S
T
_
,
For all X F
T
we thus have
E
Q
_
X
B
T
_
= E
Q
_
L
T
X S
0
S
T
_
Natural candidate:
L
t
=
dQ
S
t
dQ
t
=
S
t
S
0
B
t
Proposition:
(t) /B
t
is a Q-martingale.
(t) /S
t
is a Q
-martingale.
Proof.
E
S
_
(t)
S
t
F
s
_
=
E
Q
_
L
t
(t)
S
t
F
s
_
L
s
=
E
Q
_
(t)
B
t
S
0
F
s
_
L
s
=
(s)
B(s)S
0
L
s
=
(s)
S(s)
.
Result
t
[X] = S
t
E
S
_
X
S
t
F
t
_
We can observe S
t
directly on the market.
Example: X = S
t
Y
t
[X] = S
t
E
S
[Y | F
t
]
Several underlying
X = [S
0
(T), S
1
(T)]
Assume is linearly homogeous. Transform to Q
0
.
t
[X] = S
0
(t)E
0
_
[S
0
(T), S
1
(T)]
S
0
(T)
F
t
_
= S
0
(t)E
0
[(Z
T
)| F
t
]
(z) = [1, z] , Z
t
=
S
1
(t)
S
0
(t)
Exchange option
X = max [S
1
(T) S
0
(T), 0]
t
[X] = S
0
(t)E
0
[max [Z(T) 1, 0]| F
t
]
European Call on Z with strike price K. Zero interest
rate.
Piece of cake!
Identifying the Girsanov Transformation
Assume Q-dynamics of S known as
dS
t
= r
t
S
t
dt +S
t
v
t
dW
t
L
t
=
S
t
S
0
B
t
From this we immediately have
dL
t
= L
t
v
t
dW
t
.
and we can summarize.
Theorem:The Girsanov kernel is given by the
numeraire volatility v
t
, i.e.
dL
t
= L
t
v
t
dW
t
.
Forward Measures
Use price of T-bond as numeraire.
L
T
t
=
p(t, T)
p(0, T)B
t
dp(t, T) = r
t
p(t, T)dt +p(t, T)v(t, T)dW
t
,
dL
T
t
= L
T
t
v(t, T)dW
t
Result:
t
[X] = p(t, T)E
T
[X| F
t
]
Common Conjecture: The forward rate is an
unbiased estimator of the future spot rate:
Lemma:
f(t, T) = E
T
[r
t
| F
t
]
A new look on option pricing
European call on asset S with strike price K and maturity T.
X = max [S
T
K, 0]
Write X as
X = (S
T
K) I {S
T
K} = S
T
I {S
T
K} KI {S
T
K}
Use Q
S
on the rst term and Q
T
on the second.
0
[X] = S
0
Q
S
[S
T
K] K p(0, T) Q
T
[S
T
K]
Analytical Results
Assumption: Assume that Z
S,T
, dened by
Z
S,T
(t) =
S
t
p(t, T)
,
has dynamics
dZ
S,T
(t) = Z
S,T
(t)m
S
T
(t)dt +Z
S,T
(t)
S,T
(t)dW,
where
S,T
(t) is deterministic.
We have to compute
Q
T
[S
T
K]
and
Q
S
[S
T
K]
Q
T
(S
T
K) = Q
T
_
S
T
p(T, T)
K
_
= Q
T
(Z
S,T
(T) K)
By denition Z
S,T
is a Q
T
-martingale, so Q
T
-dynamics
are given by
dZ
S,T
(t) = Z
S,T
(t)
S,T
(t)dW
T
,
with the solution
Z
S,T
(T) =
S
0
p(0, T)
exp
_
1
2
_
T
0
2
S,T
(t)dt +
_
T
0
S,T
(t)dW
T
_
Lognormal distribution!
The integral
_
T
0
S,T
(t)dW
T
is Gaussian, with zero mean and variance
2
S,T
(T) =
_
T
0
S,T
(t)
2
dt
Thus
Q
T
(S
t
K) = N[d
2
],
d
2
=
ln
_
S
0
Kp(0,T)
_
1
2
2
S,T
(T)
_
2
S,T
(T)
Q
S
(S
t
K) = Q
S
_
p(T, T)
S
t
1
K
_
= Q
S
_
Y
S,T
(T)
1
K
_
,
Y
S,T
(t) =
p(t, T)
S
t
=
1
Z
S,T
(t)
.
Y
S,T
is a Q
S
-martingale, so Q
S
-dynamics are
dY
S,T
(t) = Y
S,T
(t)
S,T
(t)dW
S
.
Y
S,T
= Z
1
S,T
S,T
(t) =
S,T
(t)
Y
S,T
(T) =
p(0, T)
S
0
exp
_
1
2
_
T
0
2
S,T
(t)dt
_
T
0
S,T
(t)dW
S
_
,
Q
S
(S
t
K) = N[d
1
],
d
1
= d
2
+
_
2
S,T
(T)
Proposition: Price of call is given by
0
[X] = S
0
N[d
2
] K p(0, T)N[d
1
]
d
2
=
ln
_
S
0
Kp(0,T)
_
1
2
2
S,T
(T)
_
2
S,T
(T)
d
1
= d
2
+
_
2
S,T
(T)
2
S,T
(T) =
_
T
0
S,T
(t)
2
dt
Hull-White
Q-dynamics:
dr = {(t) ar} dt +dW.
Ane term structure:
p(t, T) = e
A(t,T)B(t,T)r
t
,
B(t, T) =
1
a
_
1 e
a(Tt)
_
.
Check if Z has deterministic volatility
Z
t
=
S
t
p(t, T
1
)
, S
t
= p(t, T
2
)
Z
t
=
p(t, T
2
)
p(t, T
1
)
,
Z
t
= exp{A(t) B
t
r
t
} ,
Z(t) = exp{A(t) B
t
r
t
} ,
A(t) = A(t, T
2
) A(t, T
1
),
B
t
= B(t, T
2
) B(t, T
1
),
dZ(t) = Z(t) { } dt +Z(t)
z
(t)dW,
z
(t) = B
t
=

a
e
at
_
e
aT
1
e
aT
2
Deterministic volatility!
8
LIBOR Market Models
Ch. 27
Problems with innitesimal rates
Innitesimal rates can never be observed in real life.
Calibration to cap- or swaption data is dicult.
Disturbing facts from real life:
The market uses Black-76 to quote caps and
swaptions. Behind Blavk-76 are the assumptions
The short rate is constant.
The LIBOR rates are lognormally distributed.
Logically inconsistent!
Despite this, the market happily continues to use
Black-76 for quoting purposes.
Project
Construct a logically consistent model which (to
some extent) justies market practice.
Construct an arbitrage free model with the
property that caps, oors and/or swaptions are
priced with a Black-76 type formula.
Main models
LIBOR market models (Miltersen-Sandmann-
Sondermann, Brace-Gatarek-Musiela)
Swap market models (Jamshidian).
Instead of modeling instantaneous rates, we model
discrete market rates, such as
LIBOR rates (LIBOR market models)
Forward swap rates (swap market models).
Under a suitable numeraire the market rates can be
modeled lognormally.
The market models with thus produce pricing
formulas of the type Black-76.
By construction the market models are very easy to
calibrate to market data, i.e. to:
Caps and oors (LIBOR market model)
Swaptions (swap market model)
Exotic derivatives has to be priced numerically.
Caps
Resettlement dates:
T
0
< T
1
< . . . < T
n
,
Tenor:
= T
i+1
T
i
, i = 0, . . . , n 1.
Typically = 1/4, i.e. quarterly resettlement.
LIBOR forward rate for [T
i1
, T
i
]:
L
i
(t) =
1
p
i1
(t) p
i
(t)
p
i
(t)
, i = 1, . . . , N.
where we use the notation
p
i
(t) = p(t, T
i
)
Denition:
A cap with cap rate R and resettlement dates
T
0
, . . . , T
n
is a contract which at each T
i
give the
holder the amount
X
i
= max [L
i
(T
i1
) R, 0] , i = 1, . . . , N
The cap is thus a portfolio of caplets X
1
, . . . , X
n
.
Black-76
The Black-76 formula for the caplet
X
i
=
i
max [L(T
i1
, T
i
) R, 0] , (1)
is given by
Capl
B
i
(t) = p
i
(t) {L
i
(t)N[d
1
] RN[d
2
]}
where
d
1
=
1
T
i
t
_
ln
_
L
i
(t)
R
_
+
1
2
2
i
(T t)
_
,
d
2
= d
1
i
_
T
i
t.
Black-76 presupposes that each LIBOR rate is
lognormal.
The constants
1
, . . . ,
N
are known as the Black
volatilities
Market price quotes
Market prices are quoted in terms of Implied Black
volatilities: The can be quoted in two dierent ways.
Flat volatilities
Spot volatilities (also known as forward
volatilities)
Market Price Data
For each i = 1, . . . , N:
Cap
m
i
(t) = market price of cap with resettlement
dates T
0
, T
1
, . . . , T
i
Implied market prices of caplets:
Capl
m
i
(t) = Cap
m
i
(t) Cap
m
i1
(t),
with the convention Cap
m
0
(t) = 0
Dening Implied Black Volatility
Given market price data as above, the implied Black
volatilities are dened as follows.
The implied at volatilities
1
, . . . ,
N
are dened
as the solutions of the equations
Cap
m
i
(t) =
i
k=1
Capl
B
k
(t;
i
), i = 1, . . . , N. (2)
The implied forward or spot volatilities
1
, . . . ,
N
are dened as solutions of the equations
Capl
m
i
(t) = Capl
B
i
(t;
i
), i = 1, . . . , N. (3)
The sequence
1
, . . . ,
N
is called the volatility term
structure.
Theoretical Price of a Caplet
By risk neutral valuation:
Capl
i
(t) = E
Q
t
_
e
R
T
i
0
r(s)ds
max [L
i
(T
i1
) R, 0]
_
,
Better to use T
i
forward measure
Capl
i
(t) = p
i
(t)E
T
i
[max [L
i
(T
i1
) R, 0]| F
t
] ,
The crucial point is the distribution of L
i
under Q
i
where Q
i
= Q
T
i
Important Fact: L
i
is a martingale under Q
i
L
i
(t) =
1
p
i1
(t) p
i
(t)
p
i
(t)
Idea: Model L
i
as GBM under Q
i
:
dL
i
=
i
L
i
dW
i
LIBOR Market Model Denition
Dene, for each i, the dynamics of L
i
under Q
i
as
dL
i
(t) = L
i
(t)
i
(t)dW
i
(t), i = 1, . . . , N,
where
1
(t), . . . ,
N
(t) are deterministic and W
i
is
Q
i
-Wiener.
The initial term structure L
1
(0), . . . , L
N
(0) is observed
on the market.
Pricing Caps in the LIBOR Model
L
i
(T) = L
i
(t) e
R
T
t

i
(s)dW
i
(s)
1
2
R
T
t

i
(s)
2
ds
.
Lognormal!
Theorem: Caplet prices are given by
Capl
i
(t) =
i
p
i
(t) {L
i
(t)N[d
1
] RN[d
2
]} ,
where
d
1
=
1
i
(t, T
i1
)
_
ln
_
L
i
(t)
R
_
+
1
2
2
i
(t, T
i1
)
_
,
d
2
= d
1
i
(t, T
i1
),
2
i
(t, T) =
_
T
t
i
(s)
2
ds.
Moral: Each caplet price is given by a
Black-76 formula with
i
as the Black volatility.
Practical Handling of the LIBOR Model
We are standing at time t = 0.
Collect implied caplet volatilities

1
, . . . ,
N
from the market.
Choose model volatilities
1
(), . . . ,
N
()
such that

i
=
1
T
i
_
T
i1
0
2
i
(s)ds, i = 1, . . . , N.
Now the model is calibrated.
Use numerical methods to compute prices of exotics.
Terminal Measure dynamics
Dene the Likelihood process
j
i
as
j
i
(t) =
dQ
j
dQ
i
, on F
t
Can show that
d
i1
i
(t) =
i1
i
(t)

i
L
i
(t)
1 +
i
L
i
(t)
i
(t)dW
i
(t).
Girsanov gives us
dW
i
(t) =

i
L
i
(t)
1 +
i
L
i
(t)
i
(t)dt +dW
i1
(t).
Proposition The Q
N
dynamics of the LIBOR rates are
dL
i
(t) = L
i
(t)
_
_
N
k=i+1
k
L
k
(t)
1 +
k
L
k
(t)
k
(t)
i
(t)
_
_
dt
+ L
i
(t)
i
(t)dW
N
(t),
Computational aspects
The terminal measure dynamics of the system of
LIBOR rates are quite messy.
Various approximations for the drift term have been
suggested.
Numerical work.

Bjork Notes

Caricato da

Informazioni sul documento

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Bjork Notes

Caricato da

Copyright:

Formati disponibili

Interest Rate Theory

Tomas Bjork, 2010 17

We then obviously have L

Tomas Bjork, 2010 19

The normalized value process V

Existence of stochastic integral

there exists g such that

Tomas Bjork, 2010 139

and for this we need to compute the theoretical bond

Theoretical term structure

Specifying bond prices.

There exists process g(t) = [g

(t, s)ds (t, T)g(t), t, T

Potrebbero piacerti anche