Doubly Stochastic Poisson processes are generalizations of Compound Poisson processes, in the sense that the intensity of the (simple) counting process, N t , is stochastic. The (random) intensity function is defined as λ(t ) = lim h→0 P[N t +h - N t =1 | σ(N u ) u<t ] h = lim h→0 E[N t +h - N t | σ(N u ) u<t ] h The Doubly Stochastic Poisson Process, or Cox process satisfy: P [N t - N s = n |F s ∨ σ(λ u ) u∈[s ,t ] = exp - t s λ u du ( t s λ u du) n n! P [N t - N s = n |F s ]= E exp - t s λ u du ( t s λ u du) n n! F s Where does the randomness of the intensity function may come from? What is the difference in the two equations above? Jonathan Ch´ avez (University of Calgary) Doubly Stochastic Poisson Processes 18/05/2016 1/5
5
Embed
Doubly Stochastic Poisson Processespeople.ucalgary.ca/~aswish/Doubly_StochasticPP.pdf · Doubly Stochastic Poisson processes are generalizations of Compound Poisson processes, in
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Doubly Stochastic Poisson processes are generalizations of Compound Poisson processes, in thesense that the intensity of the (simple) counting process, Nt , is stochastic.The (random) intensity function is defined as
λ(t) = limh→0
P[Nt+h − Nt = 1 | σ(Nu)u<t ]h = lim
h→0
E[Nt+h − Nt | σ(Nu)u<t ]h
The Doubly Stochastic Poisson Process, or Cox process satisfy:
� P [Nt − Ns = n | Fs ∨ σ(λu)u∈[s,t]]
= exp(−∫ t
s λudu) (∫ t
sλudu)n
n!
� P [Nt − Ns = n | Fs ] = E[
exp(−∫ t
s λudu) (∫ t
sλudu)n
n!
∣∣∣∣ Fs
]Where does the randomness of the intensity function may come from?What is the difference in the two equations above?
Jonathan Chavez (University of Calgary) Doubly Stochastic Poisson Processes 18/05/2016 1 / 5
Some examples of possible (random) intensity functions are:Feller process:
dλt = κ(θ − λt)dt + η√λtdWt
O-U process:dλt = −κλtdt + γdJt
Jump-diffusion:dλt = κ(θ − λt)dt + η
√λtdWt + γdJt
Hawkes process:
dλt =∫ t
0g(t − s)dNs
where J is an independent compound Poisson process with intensity λJ and i.i.d. jump size ε ∼ Fand ε > 0 a.s.
Jonathan Chavez (University of Calgary) Doubly Stochastic Poisson Processes 18/05/2016 2 / 5
Compensator (Doob-Meyer decomposition) of a Doubly Stochastic process Nt is∫ t
0 λsds. That is,
Nt −∫ t
0λsds
is a martingale.
If J is the compensated version of J , that is, J = J − E[ε]JλJ t, then we can rewrite the O-U andjump-diffusion intensity processes above as:
O-U process:dλt = κ
(γλJκE[ε]− λt
)dt + γdJt
Jump-diffusion:dλt = κ(θ + γλJ
κE[ε]− λt)dt + η
√λtdWt + γdJt
Jonathan Chavez (University of Calgary) Doubly Stochastic Poisson Processes 18/05/2016 3 / 5
Let µ be a discrete measure with support on {a1, a2, . . . , aN} with N ≤ ∞. Assume thatµ(aj) = hj ∈ R for each j ∈ {1, 2, . . . ,N}. What is the definition of
∫A f (x)dµ(x)?
Can a counting process be seen as a discrete measure?
Thus, we have (it is NOT a definition) that for a doubly stochastic PP, Nt , with intensity λt ,
∫ t
0g(s−)dNs =
Nt∑τk
g(τ−k ),
where τk are the jumping times of N on (0, t). Also, for Nt the compensated doubly stochastic PP,(why?) ∫ t
0g(s−)dNs =
Nt∑τk
g(τ−k )−∫ t
0gsλsds
Jonathan Chavez (University of Calgary) Doubly Stochastic Poisson Processes 18/05/2016 4 / 5
Jonathan Chavez (University of Calgary) Doubly Stochastic Poisson Processes 18/05/2016 5 / 5