2080 CHAPTER 62. STOCHASTIC PROCESSES

62.7 Doob Optional Sampling Continuous Case62.7.1 Stopping TimesLet X (t) be a stochastic process adapted to a filtration {Ft} for t ∈ [0,T ]. We will assumetwo things. The stochastic process is right continuous and the filtration is normal.

Definition 62.7.1 A normal filtration is one which satisfies the following :

1. F0 contains all A ∈F such that P(A) = 0. Here F is the σ algebra which containsall Ft .

2. Ft = Ft+ for all t ∈ I where Ft+ ≡ ∩s>tFs.

For an F measurable [0,∞) valued function τ to be a stopping time, we want to havethe stopped process Xτ defined by Xτ (t)(ω)≡ X (t ∧ τ (ω))(ω) to be adapted whenever Xis right continuous and adapted. Thus a stopping time is a measurable function which canbe used to stop the process while retaining the property of being adapted. We want to finda simple criterion which will ensure that this happens.

Let X (t) be adapted. Let O be an open set in some metric space where X has its values.This could probably be generalized. Then we need to have

Xτ (t)−1 (O) ∈Ft

Thus we need to have

[τ ≤ t]∩[X (τ)−1 (O)

]∪ [τ > t]∩

[X (t)−1 (O)

]∈Ft

How does this happen? Consider τk (ω)≡∑∞n=0 X

τ−1((n2−k,(n+1)2−k]) (ω)(n+1)2−k. Thusfor each ω,τk (ω) ↓ τ (ω). Since X is right continuous for each ω, it follows that, since Ois open,

[τ ≤ t]∩[X (τ)−1 (O)

]= [τ ≤ t]∩

(∪m∩k≥m

[X (τk)

−1 (O)])

= ∪m∩k≥m∪∞n=0τ

−1((n2−k,(n+1)2−k ∧ t]

)∩[

X((n+1)2−k

)−1(O)

]the last union being a disjoint union. Now

τ−1((n2−k,(n+1)2−k ∧ t]

)∩[

X((n+1)2−k

)−1(O)

]is a set of F(n+1)2−k intersected with

[τ ∈ (n2−k,(n+1)2−k]

]∩ [τ ∈ (0, t]] . If we assume

[τ ≤ t] ∈Ft , for all t, then this shows that the above expression is a set of Ft+2−k . Since

this is true for each k, and the filtration is normal, this implies [τ ≤ t]∩[X (τ)−1 (O)

]∈Ft .

Also with this assumption, [τ > t] = [τ ≤ t]C ∈Ft and so we get Xτ (t)−1 (O) ∈Ft . Thisis why we define stopping times this way. It is so that when you have a right continuousadapted process, then the stopped process is also adapted.

2080 CHAPTER 62. STOCHASTIC PROCESSES62.7 Doob Optional Sampling Continuous Case62.7.1 Stopping TimesLet X (t) be a stochastic process adapted to a filtration {.¥,} for ¢ € [0,7]. We will assumetwo things. The stochastic process is right continuous and the filtration is normal.Definition 62.7.1 A normal filtration is one which satisfies the following :1. ¥o contains all A € ¥ such that P(A) =0. Here F is the o algebra which containsall F;.2. F¥:= F,,4 for allt € I where Fy4. = Oss Fs.For an ¥ measurable [0,c¢) valued function t to be a stopping time, we want to havethe stopped process X* defined by X* (t) (@) =X (t A T(@)) (@) to be adapted whenever Xis right continuous and adapted. Thus a stopping time is a measurable function which canbe used to stop the process while retaining the property of being adapted. We want to finda simple criterion which will ensure that this happens.Let X (t) be adapted. Let O be an open set in some metric space where X has its values.This could probably be generalized. Then we need to haveXt(t) (O) EF,Thus we need to haveIt <1]A [x (2) (0)| Ulr>tn [x (tr)! (0)| CF,How does this happen? Consider t; (@) = Yo 2-1 ((n2-,(n41)2-4)) (w)(n+1)2~*. Thusfor each @, 7 (@) | t(@). Since X is right continuous for each @, it follows that, since Ois Open,[It <2]n [x (rt)! (0)| =[t<1]n (Un Ak>m [x (t) | (0)])= Un Mm U2 ot! ((n2-%, (n+1)2-* At) A Ix ((n+ 1) 2*) - (0)the last union being a disjoint union. Now1 k k k\!T ((n2- (n+1)2> Aq) a x ((n +1)2- ) (0)is a set of F,,,1)>-« intersected with [t € (n2~*, (n+ 1)2~*]] M[t € (0,1]] . If we assume[t <t] € F,, for all t, then this shows that the above expression is a set of .F,,5-«. Sincethis is true for each k, and the filtration is normal, this implies [t < r]M [x (rt)! (0)| € F;.Also with this assumption, [t > ¢] = [t <1]© € F, and so we get X*(t) | (O) € .F,. Thisis why we define stopping times this way. It is so that when you have a right continuousadapted process, then the stopped process is also adapted.