Swing Options Valuation With Jumps2

Risk-neutral valuation of Swing options
in presence of jumps
Jess Prez Colino
jpcolino@gmail.com
January 2011
Jess Prez Colino (jpcolino@gmail.com) Swinging between jumps January 2011 1 / 18
Goal of this presentation
The main goal of this presentation is to show the basic framework for the
risk-neutral pricing of swing options valuation where the logarithm of the
spot price is the sum of a deterministic seasonal trend and the
Ornstein-Uhlenbeck process driven by a Lvy process (jump-diusion).
This presentation has been designed for quants analyst, and the content of
this presentation can hurt your feelings if you do not used to work with
stochastic dierential equations, PDE-PIDE and some measure theory.
Questions addressed in this presentation:
Which is the stochastic dierential equation that better ts the "most
correct" dynamics of the commodities involved in a swing option?
Can we build a risk-neutral model that allow us to model the forward
prices?
Could we build an analytical pricing framework that allows us to estimate
not only the prices but also the Greeks for the swing option hedging?
How can we obtain the most robust parameters that we have to introduce
in the model such that guarantee the most stable possible pricing and
hedging?
Agenda
1. Introduction
Introduction
Denition
A Swing option is a nancial contract with the following payo characteristics:
1
Maturity contract: runs over [0, T]
2
Strike: xed price K Eur/MWh
3
Swing action times: nite set of dates T
n
N
n=1
with
0 _ T
1
< T
2
< ... < T
N
< T
4
Swing action: At each swing action date T
n
the holder decides on the
amount of energy B
d
n
MWh to be bought at xed price K Eur/MWh over
each of the D periods
_
T
d
n
, T
d+1
n
, 1 _ d _ D.
5
Total and partial boundaries: assume that B
d
n
O _ [0, ) where O is
either a closed interval O = [B
,

B] or a discrete set. Additionally, the holder
must buy at least M
MWh, and the most

M MWh in total
6
Settlement: All swing options are nancially settled.
Introduction
The outline of this presentation (excluding this introduction) is the following:
2. Section 1: Deng (2000) spot price model
- This section introduce the model developed by Deng (2000) where the
logarithm of the spot price is the sum of a seasonality term and a
Ornstein-Uhlenbeck process driven by a jump diusion.
3. Section 2: Model calibration
- Section 2 is devoted to the calibration or estimation of the parameters of the
model, using a combination between a least-squares method (Lucia and
Schwartz (2002)) and the Fourier transform based in the maximum likelihood
approach by Singleton (2001).
4. Section 3: Swing option price estimation and numerical algorithm
- In the last section, we proceed to develop the swing option pricing algorithm
based in the discretization of a PIDE using a nite dierence method, similar
as Cont and Tankov (2003).
Agenda
1. Introduction
2. Section 1:
Deng (2000) spot price model
Agenda
1. Introduction
2. Section 1:
Section 1: Deng (2000) spot price model
We consider that power spot prices S = S
t
: 0 _ t _ T
?
lives in a
continuous-time trading economy in [0, T
?
] driven by a compensated Lvy
process L on a suciently rich stochastic basis (, /, L, P) .
Denition
Assume that the power-spot prices follows the P-dynamics:
_
S
t
= exp(f (t) + X
t
)
dX
t
= X
t
dt + dL
t
(1)
where > 0 is xed, f (t) is a deterministic seasonal trend and L
t
is a
compensated Lvy process that accept the canonical Lvy-Ito decomposition
with triplet (, , 0) such that
L(t) =
_
t
0
W (ds) +
_
t
0
_
[x[>
x (J
X
) (ds, dx)
= W
t
+ U
x
t
E
P
[x] t (2)
The solution S
t
to the SDE (1) started in t
0
is given by
S
t
= S
t
0
e
(tt
0
)
E
P
[x]
_
1 e
(tt
0
)
_
+
_
t
t
0
e
(tt
0
)
dW
s
+
N
t
i=N
t
0
e
(tt
i
)
X
i
(3)
The density of S
t
is generally not known explicitly, but using the FT for any
t R
+

t
(z) = e
(tt
0
)(z)
with z R
d
(4)
where
j
(z) is the Lvy-Khitchine exponent with the following
representation
(z) = izxe
(tt
0
)

2
z
2
4
_
1 e
2(tt
0
)
_
+
_
R
d
_
e
izx
1 izxe
(tt
0
)
_
(dx) (5)
Fact
Forward prices, under a diusive risk-neutral measure, has the following
expression:
F(t, T) = exp
_
f (T) + (logS
t
f (t)) e
(Tt)
_
exp
_
_
1 e
(Tt)
_
+

2
4
_
1 e
2(Tt)
_
_
exp
_
_
R
d
_
e
x
1 xe
(Tt)
_
(dx)
_
(6)
= F
season
F
diffusion
F
jump
Proof.
That can be obtained following Lucia and Schwartz(2002)
Agenda
1. Introduction
2. Section 1:
3. Section 2:
Model Calibration
Agenda
1. Introduction
2. Section 1:
3. Section 2:
Model Calibration
Agenda
1. Introduction
2. Section 1:
3. Section 2:
Model Calibration
Section 2: Model Calibration
1
Let us dene a periodic seasonal trend f ( t[ ) with
= (A
0
, A
n
, B
n
)
N
n=1
such that
f ( t[ ) = A
0
+
N
n=1
A
n
cos (2f
n
t + B
n
) (7)
where the parameter vector is unknown and is to be estimated from data.
2
Notice that the stationary covariance function of the lnS
t
is given by
Cov (lnS
t+
, lnS
t
) = e
Var (lnS
t
) so is estimated from sample
covariances and variances.
3
Next, we subtract the seasonal component f ( t[ ) from the lnS
t
, and we
estimate the remaining parameters with the maximum likelihood method.
Notice that if we unknown the density function we may form the log
likelihood function using the inverse Fourier transformation of the
characteristic function, following Deng(2000) or Singleton(2001).
4
Finally, to estimate the market price of risk , we minimized the distance
between market and model prices, such that
= arg min
K
k=1

F (t, T) F (t, T)
2
(8)
Agenda
1. Introduction
2. Section 1:
Deng(2000) spot price model
3. Section 2:
Model Calibration
4. Section 3:
Swing option price estimation
Agenda
1. Introduction
2. Section 1:
3. Section 2:
Model Calibration
4. Section 3:
Agenda
1. Introduction
2. Section 1:
3. Section 2:
Model Calibration
4. Section 3:
Agenda
1. Introduction
2. Section 1:
3. Section 2:
Model Calibration
4. Section 3:
Section 3: Swing option price estimation
Let us remind that at time T
n
, the swing option holder decide to buy B
d
n
MWh at K EUR/MWh.
A swing action is described as a D-vector
_
B
1
n
...B
D
n
_
, and dened as
n
=
D
d=1
B
d
n
such that
n
S = [0, D
B] . A swing action means choosing
n
without violating the contract constraints, and receiving the amount
g (T
n
, s,
n
) .
Finally, let us dene Z
t
=
j
n=1
n
for t
_
T
j
, T
j+1
.
Denition
We can dene the value of a swing option at moment t, when s = S
t
and
z = Z
t
, as V (t, s, z) given by
V (t, s, z) =
_
_
sup
N
n=j
e
r
(
T
n
T
j
)
E
_
g (T
n
, S
T
n
,
n
)[ T
T
j
_
t = T
j
e
r
(
T
n
T
j
)
E
_
V
_
T
j
, S
T
j
, z
_
T
T
j
_
T
j
< t < T
j
, j > 1
t < T
1
(9)
Theorem
There exists at least one optimal swing action strategy
+
n
N
n=j
such that the
supremum is attainable.
According with the Feynman-Kac theorem, the swing option price V (t, s, z) is
the unique solution to the following parabolic partial integro-dierential equation
(PIDE)
V
t
+/
x
V rV = 0 (10)
where the operator /
x
ca be splitted into two parts (integral and dierential) such
that /
x
= T
x
+ J
x
with
T
x
V (t, x, z) =

2
2
2
V
x
2
( + x)
V
x
J
x
V (t, x, z) =
_
R
_
V (t, x
l
+ y, z) V (t, x
l
, z) y
V
x
_
f (y)dy
(11)
that can be solved numerically splitting the diusion or continuous part, from the
integral or discontinuous part.
The space derivatives are discretized using nite dierences:
_
V
x
_
n
-
_
V
n+1
V
n
x
if + x < 0
V
n
V
n1
x
if + x _ 0
_
2
V
x
2
_
n
-
V
n+1
2V
n
V
n1
(x)
2
In order to approximate the integral terms one can use the trapezoidal
quadrature rule with the same grid resolution x. More specically, if we
dene:
V
n
=
_
x
(
n+
1
2
)
x
(
n
1
2
)
f (y)dy
then J
x
V (t, x, z) may approximated as:
J
x
V (t
k+1
, x
l
, z) -
N
n=N
_
V
l+n
k+1
V
l
k+1
n
2
_
V
l+1
k+1
V
l1
k+1
__
n
Answers:
1. We have introduced a stochastic process L
t
, that has "some good"
properties for the commodities price modelling.
2. We have built the risk-neutral dynamics for forward prices where we have
included seasonality, mean-reversion, Brownian motion and jumps.
3. We have proposed a calibration methodology, that include the market
price of risk for the diusion part.
4. We have dened a pricing framework for swing option based in a PIDE.
The solution of the PIDE can be found with a numerical pricing algorithm
based in nite dierences (using a combination of explicit-implicit
schemes).
References
R. Cont and P. Tankov. Financial Modelling with Jump Processes. Chapman
& Hall, London (2003)
S.Deng. Stochastic models of energy commodity prices and their applications:
mean reversion with jumps and spikes, UC Energy Institute (1998)
P.Jaillet, E.Ronn and S.Tompaidis. Valuation of commodity-based swing
options. Management Science, 50(7) (2004)
J.Jacod and A.N.Shiryaev. Limit Theorems for Stochastic Processes. Springer
(1987)
J.Lucia and E.Schwartz. Electricity prices and power derivatives: evidence
from the nordic power exchange. Review of Derivatives Research (2002)
M.Kjaer. Pricing of swing options in a mean reverting model with jumps.
Goteborg University. (2007)
T. Kluge. Pricing Swing options and other Electricity Derivatives. Oxford
(2006)
... among others.

Swing Options Valuation With Jumps2

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Swing Options Valuation With Jumps2

Uploaded by

Copyright:

Available Formats

Risk-neutral valuation of Swing options

MWh, and the most

B] . A swing action means choosing

You might also like