The Mathematics of Arbitrage

(Tina Meador) #1
14.4 The GeneralRd-valued Case 303

Denote byDthe predictable set

D=


k≥ 1

[[Tk]]⊆Ω×R+

and splitSintoS=Sa+Si,where


Sa= (^1) D·S and Si= (^1) (Ω×R+)\D·S
where the letters “a” and “i” refer to “accessible” and “inaccessible”.Saand
Siare well-defined semi-martingales and in view of the above constructionSi
is quasi-left-continuous.
Denote byCaandCithe cones inL∞(Ω,F,P) associated by (14.1) toSa
andSi, and observe thatCaandCiare subsets ofC(obtained by considering
only integrands supported byDor (Ω×R+)\Drespectively) hence
EQ 1 [f]≤ 0 , forf∈Caand forf∈Ci.
HenceSisatisfies the assumptions of Proposition 14.4.4 with respect to
the probability measureQ 1 and we therefore may find a probability measure,
now denoted byQ̂,Q̂∼P,whichturnsSiinto a sigma-martingale and such
that, for each predictable stopping timeT,wehave
dQ̂|FT
dQ 1 |FT


=


dQ̂|FT−
dQ 1 |FT−

. (14.2)


By assumption we have, for eachk=1, 2 ,..., and for each admissible
integrandHsupported by [[Tk]] , t h a t


EQ 1 [(H·S)∞]=EQ 1

[


HTk(STk−S(Tk)−)

]


≤ 0.


Noting that the inequality remains true if we replaceHbyH (^1) A, for any
F(Tk)−-measurable set A, and using (14.2) we obtain
EQ̂[(H·S)∞]=EQ̂


[


HTk(STk−S(Tk)−)

]


≤ 0 (14.3)


for each admissible integrand supported by [[Tk]].
We now shall proceed inductively onk: suppose we have chosen, fork≥0,
probability measuresQ ̃ 0 =Q̂,Q ̃ 1 ,...,Q ̃ksuch that


EQ ̃k

[


STj|F(Tj)−

]


=S(Tj)−,j=1,...,k

and such that, for


εj=

ε
2 j+1

∧inf

{


2 −jQ ̃j[A]
P[A]






A∈F,P[A]≥ 2 −j

}


,


we have

Free download pdf