4 Strong robust representation of convex risk measures by prob-
ability measures
We want to look for conditions which induce a strong robust representation of ρ by probability measures in the
sense that
ρ(X) = Pm∈Max (-EP [X] - αρ(P))
holds for any X ∈ X. The considerations are reduced to a Stonean vector lattice X being stable w.r.t. countable
convex combinations of antitone sequences of financial positions. In this case the following result gives a complete
answer to the problem of strong robust representations.
Theorem 4.1 Let X be a Stonean vector lattice and let us assume that for every antitone sequence (Xn)n in
∞∞
X with Xn ∖ 0 and each sequence (λn)n in [0,1] with ∑ λn = 1 there is some pointwise limit ɪɪ λnXn of
n=1 n=1
m
( λnXn)m belonging to X. Then the following statements are equivalent:
n=1
.1 ρ(X) = max (-EP [X] - αρ (P)) holds for every X ∈ X.
.2 ρ(Xn) ∖ ρ(x) for Xn 7 X.
.3 lim ρ(-λ(X — n) + ) = ρ(0) hold for arbitrary nonnegative X ∈ X,λ> 0, and ρ(Xn) ∖ ρ(X) for any isotone
n→∞
sequence (Xn)n of bounded positions from X with Xn / X, X being bounded.
In any of these cases every linear form from the domain of βρ is representable by a probability measure. Moreover,
for any Stonean vector lattice L ⊆ X which contains C as well as generates the same σ-algebra as X and induces
∞
the set system S consisting of all sets T Xn-1 ([xn , ∞[) (Xn ∈ L ∩ Xb nonnegative, xn > 0) any of the statements
n=1
.1 - .3 is implied by
.4 lim ρ(-λ(X — n) + ) = ρ(0) hold for arbitrary nonnegative X ∈ X,λ> 0, and inf p(λZ) ∖ ρ(λ) for
n→∞ 1An ≥Z∈X
∞
λ> 0 whenever (1An)n is an isotone sequence of indicator mappings of subsets An ∈ S with (J An = Ω.
We have even equivalence of the statements .1 - .4 if the indicator mappings 1A (A ∈ S) belong to X.
The proof may be found in section 8.
For bounded one-period positions, Theorem 4.1 enables us to give an equivalent characterization of convex risk
measures that admit strong robust representations by probability measures.
Corollary 4.2 Let F denote some σ-algebra on Ω, and let X consist of all bounded F—measurable real-valued
mappings. Then the following statements are equivalent:
.1 ρ(X) = max (—EP [X] — αρ (P)) holds for every X ∈ X.
11