An Introduciton To Levy Processes With Fin Modelling in Mind - SLIDES

Motivation
The structure of L
evy processes
Modelling and inference
References
An introduction to L
evy processes
with financial modelling in mind
Matthias Winkel
Department of Statistics, University of Oxford
27 May 2008
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Motivation
The structure of L
evy processes
Examples of Levy processes
Infinite divisibility and the Levy-Khintchine formula
Construction and simulation of Levy processes
Parametric families of Levy processes

General modelling with Levy processes
Modelling financial price processes
Quadratic variation and realised power variation
Application: Volatility inference in the presence of jumps
References
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Motivation
Historically
Continuous-time analogues and limits of random walks,
generalisations of Brownian motion
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Motivation
Historically
Prototypes of Markov processes: simple dependence structure, spatial
as well as temporal homogeneity
Prototypes of semimartingales: suitable for stochastic calculus
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Motivation
Historically
Applications to insurance ruin, storage problems, dams etc.
Subordination of Markov processes by increasing Levy processes
Relationships to other classes of Markov processes: branching
processes, self-similar Markov processes
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Motivation
Historically
Applications to insurance ruin, storage problems, dams etc.
Subordination of Markov processes by increasing Levy processes
Relationships to other classes of Markov processes: branching
processes, self-similar Markov processes
More recently
Financial modelling: non-Normal returns, volatility smiles
Financial modelling: stochastic volatility, leverage effects
Population and phylogenetic models: trees encoded in Levy processes
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
Infinite divisibility and the L
evy-Khintchine formula
Construction and simulation of L
evy processes
Parametric families of L
evy processes
The structure of L
evy processes
Definition (L
evy process)
A Levy process is a continuous-time stochastic process (Xt , t 0) with
stationary increments
independent increments
and càdlàg paths
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
The structure of L
evy processes
Definition (L
evy process)
Xt+s Xt has the same distribution as Xs for all t 0, s 0,
and càdlàg paths
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
The structure of L
evy processes
Definition (L
evy process)
Xtj Xtj1 , j = 1, . . . , n, are independent for all 0 t0 < . . . < tn
and càdlàg paths
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
The structure of L
evy processes
Definition (L
evy process)
and càdlàg paths
t 7 Xt is a.s. right-continuous with left limits.
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
The structure of L
evy processes
Definition (L
evy process)
and càdlàg paths
t 7 Xt is a.s. right-continuous with left limits.
This definition makes sense essentially for processes with values in a locally
compact topological group. We will focus on the real-valued case,
extensions to k-dimensional Euclidean space are straightforward.
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
Examples
1. Brownian motion Bt Normal(t, 2 t).
0.02
0.04
2. Poisson process Nt Poisson(t) for some intensity (0, ).
0.04
0.02
space
0.00
space
10
10
time
time
Their moment generating functions exists, for all R:

Z

1
2
2
Bt
E(e ) =
e x
e (x) /(2 ) dx = exp t + 12 2 2
2
2
X
(t)n t
E(e Nt ) =
e n
e
= exp {t (e 1)} .
n!
n=0
Matthias Winkel
evy processes
k=1
where Ak , k 1, independent
identically distributed
0.5
space
0.0
Example
3. Compound Poisson process
Nt
X
Ct =
Ak ,
Examples of L
evy processes
evy processes
evy processes
0.5
Motivation
The structure of L
evy processes
References
and (Nt , t 1) is an independent Poisson process.
10
time
Its moment generating function exists if and only if E(e A1 ) exists and then
( n
)!
X
X
E(e Ct ) =
E exp
Ak
P(Nt = n)
=
n=0

X
n=0
k=1
n (t)n

o
n
E e A1
e t = exp t E e A1 1 .
n!

Matthias Winkel
evy processes
k=1
where Ak , k 1, independent
identically distributed
0.5
space
0.0
Example
3. Compound Poisson process
Nt
X
Ct =
Ak ,
Examples of L
evy processes
evy processes
evy processes
0.5
Motivation
The structure of L
evy processes
References
and (Nt , t 1) is an independent Poisson process.
10
time
Its moment generating function exists if and only if E(e A1 ) exists and then
( n
)!
X
X
E(e Ct ) =
E exp
Ak
P(Nt = n)
=
n=0

X
n=0
k=1
n (t)n
o
n
E e A1
e t = exp t E e A1 1
.
n!

Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
Definition (Infinite divisibility)

A random variable Y (or its distribution) is called infinitely divisible if
(n)
Y Y1
(n)
+ . . . + Yn
(n)
(n)
for all n 2 and independent and identically distributed Y1 , . . . , Yn .

(n)
The distribution of Yj
will depend on n, but not on j.
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
Definition (Infinite divisibility)

A random variable Y (or its distribution) is called infinitely divisible if
(n)
Y Y1
(n)
+ . . . + Yn
(n)
(n)
for all n 2 and independent and identically distributed Y1 , . . . , Yn .

(n)
The distribution of Yj
will depend on n, but not on j.
Proposition
Xt is infinitely divisible for a Levy process (Xt , t 0).
Proof.
We recall the stationarity and independence of increments and take
(n)
Yj = Xjt/n X(j1)t/n .
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
Theorem (L
evy-Khintchine)
A random variable Y is infinitely divisible if and only if

Z

1 2 2
iY
ix
E(e ) = exp i +
e 1 ix1{|x|1} (dx)
2
R
R
for some R, 2 0, and on R \ {0} such that R (1 x 2 )(dx) < .
Furthermore, if E(e Y ) < , then

Z

1 2 2
Y
x
E(e ) = exp + +
e 1 x1{|x|1} (dx) .
2
R
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
Theorem (L
evy-Khintchine)

Z

1 2 2
iY
ix
E(e ) = exp i +
e 1 ix1{|x|1} (dx)
2
R
R

1
Y
E(e ) = exp + 2 2
2
Matthias Winkel

.
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
Theorem (L
evy-Khintchine)

Z

1 2 2
iY
ix
E(e ) = exp i +
e 1 ix1{|x|1} (dx)
2
R
R

Z
Y
E(e ) = exp
Matthias Winkel
evy processes

(dx) .
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
Theorem (L
evy-Khintchine)

Z

1 2 2
iY
ix
E(e ) = exp i +
e 1 ix1{|x|1} (dx)
2
R
R

Z
Y
E(e ) = exp
x1{|x|1}
Matthias Winkel
evy processes

(dx) .
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
Theorem (L
evy-Khintchine)

Z

1 2 2
iY
ix
E(e ) = exp i +
e 1 ix1{|x|1} (dx)
2
R
R

Z

1 2 2
Y
x
E(e ) = exp + +
e 1 x1{|x|1} (dx) .
2
R
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
Theorem (L
evy-Khintchine)

Z

1 2 2
iY
ix
E(e ) = exp i +
e 1 ix1{|x|1} (dx)
2
R
R

Z

1 2 2
Y
x
E(e ) = exp + +
e 1 x1{|x|1} (dx) .
2
R
In fact, Y B1 + C1 + M1 , where = ([1, 1]c ), Ak 1 |[1,1]c and

Z
()
2
M1 = L -lim C1 x1{<|x|1} (dx) .
0
Matthias Winkel
R
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
Construction of L
evy processes
Every Levy process is infinitely divisible. For every infinitely divisible
distribution there is a Levy process, the limit as 0 of
Z
()
()
()
()
Xt = t + Wt + Ct + Mt ,
Mt = Ct t
x1{<|x|1} (dx),
R
where
(Wt , t 0) is standard Brownian motion with Wt Normal(0, t),
(Ct , t 0) is an independent compound Poisson process with
= ((1, 1)c ), Ak 1 |[1,1]c ,
()
(Ct , t 0) is an independent compound Poisson process with

()
= ( < |x| 1), Ak 1
|{<|x|1} .
()
Note that Ct
()
has jumps of sizes Ak { < |x| 1}.

Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
For a sequence a0 = 1, an 0, and In = (an , an1 ], In = [an1 , an ),

Z

X [n]
0 = 0, n =
Xt = t + Wt +
Ct n t ,
x(dx).
In
nZ
Convergence of this series means, we can approximate Xt by only

considering In0 , . . . , In0 for sufficiently high n0 .
Is compensation of small jumps necessary? In general, yes.
R
However, if R (|x| 1)(dx) < , then we do not need compensation

Z

1 2 2
ix
iX1
e 1ix1{|x|1} (dx) ;
E(e
) = exp i +
2
R
R
if R (|x|2 |x|)(dx) < , then we can also compensate big jumps

Z

1 2 2
ix
iX1
e 1ix1{|x|1} (dx) .
E(e
) = exp i +
2
R
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes

Z

X [n]
0 = 0, n =
Xt = t + Wt +
Ct n t ,
x(dx).
In
nZ

R

Z

1 2 2
0
ix
iX1
e 1
(dx) ;
E(e
) = exp i +
2
R
R

Z

1 2 2
ix
iX1
e 1ix1{|x|1} (dx) .
E(e
) = exp i +
2
R
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes

Z

X [n]
0 = 0, n =
Xt = t + Wt +
Ct n t ,
x(dx).
In
nZ

R

Z

1 2 2
0
ix
iX1
e 1
(dx) ;
E(e
) = exp i +
2
R
R

Z

1 2 2
ix
iX1
e 1ix1{|x|1} (dx) .
E(e
) = exp i +
2
R
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes

Z

X [n]
0 = 0, n =
Xt = t + Wt +
Ct n t ,
x(dx).
In
nZ

R

Z

1 2 2
0
ix
iX1
e 1
(dx) ;
E(e
) = exp i +
2
R
R

Z

1 2 2
ix
00
iX1
e 1ix
(dx) .
E(e
) = exp i +
2
R
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
Example (Compensation)
For 00 = 2 = 0, (dx) = |x|5/2 1{3x<0} dx, the compensating drifts
R3
x(dx) are 0.845, 2.496, 5.170 and 18.845 for = 1, = 0.3, = 0.1
and = 0.01. In the simulation, the slope increases (to infinity, as 0):
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
Method (Simulation by time discretisation)

Fix a lag > 0. Denote F1 (u) = inf{x R : P(X x) > u}. Simulate
n
X
(1,)
Xt
= S[t/] ,
where Sn =
Yk and Yk = F1 (Uk ).
k=1
Method (Simulation by throwing away small jumps)

Let 2 = 0. Fix a jump size threshold > 0 so that = (|x| > ) > 0.
c
Denote H1 (u) = inf{x R : 1
((, x] [, ] ) > u}. Simulate
n
X
Nt = #{n 1 : Tn t}, where Tn =
Zk and Zk = 1
ln(U2k1 ),
(2,)
Xt
= SNt b t,
where Sn =
k=1
n
X
Ak and Ak = H1 (U2k ),
k=1
with b =
{xR:<|x|1} x(dx).
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
Theorem (Asmussen and Rosinski)

Let (Xt ,Rt 0) be a Levy process with characteristics (a, 0, ). Denote
v () = x 2 (dx). If v ()/2 as 0, then

(2,)
Xt Xt
p
v ()
Wt
in distribution as 0
for an independent Brownian motion (Wt )t0

If v ()/2 , it is well-justified to adjust Method 2 to set
(2+,)
Xt
(2,)
= Xt
v ()Wt
for an independent Brownian motion. We thereby approximate the

small jumps (that we throw away) by an independent Brownian motion.
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
Parametric families
We can construct distributions that are infinitely divisible and consider
associated Levy processes.
1. Brownian motion Bt Normal(t, 2 t), two parameters R,
2 0.
2. Gamma process Gt Gamma(t, ), two parameters > 0, > 0.
3. Generalised Variance Gamma process Vt = Gt Ht for
Gt Gamma(+ t, + ) and Ht Gamma( t, ) independent,
four parameters > 0, > 0.
(2)
(1)
4. Normal Inverse Gaussian Zt = BIt , It = inf{s 0 : Bs

(i)
Bs
independent Brownian motions

parameters 1 0, 2 R, 22
Normal(i s, i2 s),
0, w.l.o.g. 12 = 1.
> t} for two

three
Advantages: explicit densities, useful for parameter estimation.

Disadvantages: not much modelling freedom, may lead to poor fit
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
space
0
0.0
0.2
0.4
space
0.6
0.8
1.0
Gamma processes for parameters = {0.1, 1, 10, 100}.
10
10
time
Gamma process with shape parameter 1 and scale parameter 1
space
0
space
10
time
Gamma process with shape parameter 0.1 and scale parameter 0.1
10
time
Matthias Winkel
time
evy processes
10
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
2 = {1, 10, 100, 1000}.
space
2.0
1.0
1.5
space
0.5
0.0
Variance gamma processes for parameters
10
10
time
Variance Gamma process with shape parameter 50 and scale parameter 10
space
0.5
1.0
0.5
0.0
0.5
0.0
space
1.0
1.5
1.5
2.0
2.0
time
Variance Gamma process with shape parameter 0.5 and scale parameter 1
10
time
Variance Gamma process with shape parameter 5000 and scale parameter 100
Matthias Winkel
10
time
Variance Gamma process with shape parameter 5e+05 and scale parameter 1000
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
More flexibly, we can specify characteristics (, 2 , ).

1. Stable processes (St , t 0) of index (0, 1) with 0 = 2 = 0,
(dx) = |x|1 (c+ 1{x>0} + c 1{x<0} )dx for further parameters
c 0. They are the only Levy processes with Sat = a1/ St .
2. Stable processes (St , t 0) of index (1, 2) with 00 = 2 = 0,
(dx) = |x|1 (c+ 1{x>0} + c 1{x<0} )dx for further parameters
c 0. They are the only Levy processes with Sat = a1/ St . Slight
variation for = 1. Brownian motion = 2.
3. CGMY process (Xt , t 0) with 2 = 00 = 0 and (dx) = g (x)dx

C+ exp{G |x|}|x|Y 1 x > 0
g (x) =
C exp{M|x|}|x|Y 1 x < 0
for five parameters C > 0, G > 0, M > 0, Y [0, 2).
Advantages: more modelling freedom, can beR easily generalised further,
since any measure is admissible subject to R (1 |x|2 )(dx) < .
Disadvantages: probability density function not available explicitly
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
space
600
40
400
20
space
200
20
Symmetric stable processes for parameters {0.5, 1, 1.5, 1.8}.
10
10
time
Stable process with index 1 and cplus= 1 and cminus= 1
space
6
4
0
2
space
10
12
time
Stable process with index 0.5 and cplus= 1 and cminus= 1
10
time
Matthias Winkel
time
evy processes
10
Motivation
The structure of L
evy processes
References
Examples of L
evy processes
evy processes
evy processes
20
10000
40
space
20000
space
60
80
30000
Stable processes with no negative jumps for {0.5, 0.8, 1.5, 1.8}.
10
10
time
6
space
4
0
5
10
space
10
15
20
10
time
10
time
Matthias Winkel
time
evy processes
10
Motivation
The structure of L
evy processes
References
General modelling with L

evy processes

Levy processes are semimartingales. As a consequence, stochastic integrals
Z t
Z t
f (s)dXs
and
Q(s)dXs
0
are well-defined for L2 -functions f and predictable processes Q that are

appropriately L2 also.
Example ((Non-Gaussian) Ornstein-Uhlenbeck process)
Let (Xt , t 0) be a Levy process and (0, ). Then we call
Z t
t
Yt = e
y0 +
e (ts) dXs ,
t 0,
0
the associated Ornstein-Uhlenbeck process. The Ornstein-Uhlenbeck

process is a Markov process and possesses a stationary regime under a
mild log-moment condition.
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References

evy processes
Further classes of Levy-driven stochastic processes can be obtained

by stochastic differential equations
dYt = F (Yt )dt + G (Yt )dXt .
For F (y ) = y and G (y ) 1 this is again the Ornstein-Uhlenbeck
process.
by time-change/subordination
Xu(s)
or
X (s)
e.g. for a deterministic increasing function u, or for an increasing

stochastic process , either independent of X or such that (s) is a
stopping time for all s.
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References

evy processes

Definition (Black-Scholes model)
The Black Scholes model for a financial price process is the solution to the
stochastic differential equation dSt = St dt + St dWt , S0 > 0, which is
solved by

1 2
St = S0 exp
t + Wt .
2
There is abundant empirical evidence that financial price processes do not
have Normally distributed returns log(St+s St ).
Definition (L
evy market)
In a Levy market, the price process is taken as St = S0 exp {Xt } for a Levy
process (Xt , t 0).
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References

evy processes
Stochastic volatility models

Levy markets give good fit to return distributions, but there is also
evidence against independence of increments. Many stochastic volatility
models have been considered.
E.g., Barndorff-Nielsen and Shephard introduced the following Levy-driven
stochastic volatility model:
dYt = t dWt ,
dt = t dt + dZt ,
where Z is an increasing Levy process and W an independent Brownian

motion.
The price process (Yt , t 0) has continuous sample paths.
This means that the volatility process (t , t 0) is a non-Gaussian
Ornstein-Uhlenbeck process.
Interpretation: upward jumps correspond to new information leading
to increased activity that then slows down.
The volatility process (t , t 0) is not (directly) observable.
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References

evy processes

Rt
Rt
Let Yt = 0 as ds + 0 s dWs be a (continuous) Brownian semimartingale
(a stochastic process with stochastic volatility (s , s 0)).
To make inference on to unobserved volatility process from the observed
price process, consider
Z t
[t/]m m
X Y
(m)
2/m
CiP {Y }t =
(Y(j+k) Y(j+k1)2n ) dm
s2 ds = dm [Y ]t
k=0
k=1
in probability as 0. The left-hand side is called realised m-power

variation (or realised variation for m = 1), while the right-hand side is
called quadratic variation. This convergence holds under very general
assumptions. Under slightly more restrictive assumptions,
Z t

(m)
1/2
CLT
{Y }t dm [Y ]t cm
s2 ds
0
in distribution for an independent Brownian motion .

Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References

evy processes
Application: Volatility inference with jumps

Suppose now that the price process includes a discontinuous component
St = Yt + Xt where
Rt
Rt
Yt = 0 as ds + 0 s dWs is a Brownian semimartingale
and (Xt , t 0) is a Levy process without Brownian term, not
necessarily independent of Y .
R
Define = inf{ 0 : [1,1] |x| (dx) < } [0, 2].
We are still interested in making inference on (t , t 0).
Theorem (Inference in the presence of jumps)
1
If < 2 and m 2, then CiP is valid with Y replaced by S.
If < 1 and /(2 ) < 2/m < 1, then CLT is valid with Y
replaced by S.
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
1
2
5
6
Applebaum, D.: Levy processes and stochastic calculus. CUP 2004

Asmussen, S. and Rosinski, J.: Approximations of small jumps of
Levy processes with a view towards simulation. J. Appl. Prob. 38
(2001), 482493
Barndorff-Nielsen, O.E., Shephard, N.: Realised power variation
and stochastic volatility. Bernoulli 9 (2003), 243265. Correction
published in pages 11091111
Barndorff-Nielsen, O.E., Shephard, N. and Winkel, M.: Limit
theorems for multipower variation in the presence of jumps. Stoch.
Proc. Appl. 116 (2006), 796806
Bertoin, J.: Levy processes. CUP 1996
Kyprianou, A.E.: Introductory lectures on fluctuations of Levy
processes with applications. Springer 2006
Sato, K.: Levy processes and infinitely divisible distributions.
CUP 1999
Schoutens, W.: Levy processes in finance. Wiley 2001
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
R code for some of the simulations

psum <- function(vector){
b=vector;
b[1]=vector[1];
for (j in 2:length(vector)) b[j]=b[j-1]+vector[j];
b}
gammarw <- function(a,p){
unif=runif(10*p,0,1)
pos=qgamma(unif,a/p,a);
space=psum(pos);
time=(1/p)*1:(10*p);
plot(time,space,
pch=".",
sub=paste("Gamma process with shape parameter",a,
"and scale parameter",a))}
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
vgammarw <- function(a,p){

unifpos=runif(10*p,0,1)
unifneg=runif(10*p,0,1)
pos=qgamma(unifpos,a*a/(2*p),a);
neg=qgamma(unifneg,a*a/(2*p),a);
space=psum(pos-neg);
time=(1/p)*1:(10*p);
plot(time,space,
pch=".",
sub=paste("Variance Gamma process with shape
parameter",a*a/2,"and scale parameter",a))}
Simulations can now been carried out with various values of parameters
a > 0 and steps per time unit p = 1/ in gammarw(a,p), e.g.
gammarw(10,100)
vgammarw(10,1000)
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
stableonesided <- function(a,c,eps,p){

f=c*eps^(-a)/a;
n=rpois(1,10*f);t=runif(n,0,10);
y=(eps^(-a)-a*f*runif(n,0,1)/c)^(-1/a);
ytemp=1:n;res=(1:(10*p))/100;{
for (k in 1:(10*p)){{for (j in 1:n){
if(t[j]<=k/p)ytemp[j]<-y[j] else ytemp[j]<-0}};
res[k]<-sum(ytemp)}};
res}
stable <- function(a,cp,cn,eps,p){
pos=stableonesided(a,cp,eps,p);
neg=stableonesided(a,cn,eps,p);
space=pos-neg;time=(1/p)*1:(10*p);
plot(time,space,pch=".",
sub=paste("Stable process with index",a,
"and cplus=",cp,"and cminus=",cn))}
Matthias Winkel
evy processes
Motivation
The structure of L
evy processes
References
stableonesidedcomp <- function(a,c,eps,p){

f=(c*eps^(-a))/a;
n=rpois(1,10*f);
t=runif(n,0,10);
y=(eps^(-a)-a*f*runif(n,0,1)/c)^(-1/a);
ytemp=1:n;
res=(1:(10*p))/100;{
for (k in 1:(10*p)){{if (n!=0)for (j in 1:n){
if (t[j]<=k/p)ytemp[j]<-y[j] else ytemp[j]<-0}};{
if (n!=0)res[k]<-sum(ytemp)-(c*k/(p*(a-1)))*(eps^(1-a))
else res[k]<--c*k/(p*(a-1))*(eps^(1-a))}}};
res}
This R code to simulate stable processes can be refined to more
general measures with smooth densities using the rejection method.
Matthias Winkel
evy processes

An Introduciton To Levy Processes With Fin Modelling in Mind - SLIDES

Enviado por

Dados do documento

Descrição original:

Título original

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

An Introduciton To Levy Processes With Fin Modelling in Mind - SLIDES

Enviado por

Direitos autorais:

Formatos disponíveis

Motivation

Modelling and inference

2. Poisson process Nt Poisson(t) for some intensity (0, ).

Their moment generating functions exists, for all R:

and (Nt , t 1) is an independent Poisson process.

and (Nt , t 1) is an independent Poisson process.

Definition (Infinite divisibility)

for all n 2 and independent and identically distributed Y1 , . . . , Yn .

will depend on n, but not on j.

Definition (Infinite divisibility)

for all n 2 and independent and identically distributed Y1 , . . . , Yn .

will depend on n, but not on j.

(Ct , t 0) is an independent compound Poisson process with

has jumps of sizes Ak { < |x| 1}.

For a sequence a0 = 1, an 0, and In = (an , an1 ], In = [an1 , an ),

Convergence of this series means, we can approximate Xt by only

For a sequence a0 = 1, an 0, and In = (an , an1 ], In = [an1 , an ),

Convergence of this series means, we can approximate Xt by only

For a sequence a0 = 1, an 0, and In = (an , an1 ], In = [an1 , an ),

Convergence of this series means, we can approximate Xt by only

For a sequence a0 = 1, an 0, and In = (an , an1 ], In = [an1 , an ),

Convergence of this series means, we can approximate Xt by only

Method (Simulation by time discretisation)

Method (Simulation by throwing away small jumps)

Theorem (Asmussen and Rosinski)

v () = x 2 (dx). If v ()/2 as 0, then

for an independent Brownian motion (Wt )t0

for an independent Brownian motion. We thereby approximate the

4. Normal Inverse Gaussian Zt = BIt , It = inf{s 0 : Bs

independent Brownian motions

> t} for two

Advantages: explicit densities, useful for parameter estimation.

Gamma processes for parameters = {0.1, 1, 10, 100}.

2 = {1, 10, 100, 1000}.

Variance gamma processes for parameters

More flexibly, we can specify characteristics (, 2 , ).

Symmetric stable processes for parameters {0.5, 1, 1.5, 1.8}.

General modelling with L

Modelling and inference

are well-defined for L2 -functions f and predictable processes Q that are

the associated Ornstein-Uhlenbeck process. The Ornstein-Uhlenbeck

General modelling with L

Further classes of Levy-driven stochastic processes can be obtained

e.g. for a deterministic increasing function u, or for an increasing

General modelling with L

Modelling financial price processes

General modelling with L

Stochastic volatility models

where Z is an increasing Levy process and W an independent Brownian

General modelling with L

Quadratic variation and realised power variation

in probability as 0. The left-hand side is called realised m-power

in distribution for an independent Brownian motion .

General modelling with L

Application: Volatility inference with jumps

If < 2 and m 2, then CiP is valid with Y replaced by S.

Applebaum, D.: Levy processes and stochastic calculus. CUP 2004

R code for some of the simulations

vgammarw <- function(a,p){

stableonesided <- function(a,c,eps,p){

stableonesidedcomp <- function(a,c,eps,p){

Você também pode gostar