Optimal risk sharing for law invariant monetary utility functions
Summary (3 min read)
1 Introduction
- The purpose of this paper is to obtain an explicit characterization of the optimal risk sharing in the context of monetary utility functions, see Definition 2.1 below.
- The class of monetary utility functions which satisfy these two conditions include the so-called entropic utility and the semi-deviation utility (see e.g. [23]).
- In particular this shows that Agent 0 takes the extremal risks, and that the AV@R measure of risk is not so prudent.
- The authors believe that the above stated results provide an additional justification for the existence of options in financial markets, and stop-loss contracts, deductibles and layers in insurance markets.
2.1 Monetary utility functions
- Observe that the cash invariance and the monotonicity of U imply that U is finite and Lipschitz-continuous on L∞. Monetary utility functions can be identified with convex risk measures by the formula ρ(ξ) = −U(ξ).
- Assuming the Fatou property, suppose that a monetary utility function U is positively homogeneous, i.e. U(ξ) = −ρ(ξ) for some coherent risk measure ρ.
- For later use, the authors provide the following well-known properties and their proof for completeness, see e.g. [6].
2.2 Law-invariance
- L∞ for any pair (U0, U1) of law-invariant monetary utility functions.
- Given two law-invariant monetary utility functions U0 and U1, the authors know from Lemma 2.1 and (2.2) that U0 U1 is a monetary utility function.
- An interesting characterization of law-invariant monetary utility functions, obtained by Kusuoka [32] and further extended in [22] and [29].
- (Theorem 2.1 and equation (14)), expresses any law-invariant coherent risk measure in terms of the quantile function.
- C↘ is not continuous at this point (see Remark 2.4 below).
3.1 Pareto optimal allocations
- The authors first recall some classical notions from economic theory, see e.g. Gerber [26].
- X} is called the set of attainable allocations.
- Theorem 3.1. Let U0 and U1 be two monetary utility functions with associated convex conjugate functions V0 and V1 defined on (L∞)∗.
- Equivalence with (vi) in the law-invariant context is new.
3.2 The existence result
- X} be the subset of admissible allocations which increase with the corresponding aggregate risk.
- By the Denneberg’s lemma [16], the authors observe that A↗(X) is the subset of A(X) consisting of all comonotone allocations.
- The following result is proved in Section 6, where the authors also discuss the fact that the law-invariance assumption can not be dropped.
- Theorem 3.2. Let U0 and U1 be two law-invariant monetary utility functions.
3.3 Two concrete examples
- After having established the existence of Pareto optimal allocations, the authors now give an explicit characterization of Pareto optimal allocations in A↗(X) for some specific lawinvariant monetary utility functions.
- The authors first consider the case where both agents are defined by comonotone law-invariant monetary utility functions.
- The following easy application relates this result to options or stop-loss contracts, layers and deductibles in insurance.
- L∞+ \{0}, with a strict risk-aversion property to be defined below.
- Before turning to the precise definition of the latter restriction, the authors state their second main concrete example.
3.4 The semi-deviation utility
- This can be easily checked in the following example.
- For p = 1, the semi-deviation utility is not strictly risk averse conditionally on lower tail events.
3.6 Optimal risk sharing allocations
- The explicit Pareto optimal allocations obtained in Propositions 3.1 and 3.2 induce optimal risk sharing rules.
- Obviously, under the condition (IR) of individual rationality, the agents can agree to exchange the risk ξ if and only if p0(ξ) ≤ p1(ξ).
- In view of the discussion preceding the theorem, the authors only need to show that p0(ξ∗) ≤ p1(ξ∗).
- Of course, the value of ε depends strongly on the design of these experiments, and the real-life situation to which the present model is applied.
4 Super-gradient of law-invariant monetary utility functions
- The main purpose of this section is to prove Proposition 3.1.
- The authors start by considering law-invariant monetary utility functions, before specializing the discussion to the comonotone case.
- This contradiction shows that (i) holds true.
- The authors next specialize the discussion to the comonotone case.
5 Strict risk aversion conditionally on lower-tail events
- The crucial arguments are isolated in the two following lemmas.
- Let U be a law-invariant monetary utility function which is strictly risk averse conditionally on lower tail events.
- The authors are now ready for the Proof of Proposition 3.2 Let (ξ0, ξ1) ∈ A↗(X) be a Pareto optimal allocation.
6.1 A duality argument
- This provides a partial proof of Theorem 3.2 which states in addition that a Pareto optimal allocation exists in the smaller set A↗(X).
- L∞ for law-invariant monetary utility functions U0 and U1.
- As a by-product, the authors obtain Lemma 6.2 which states that the existence statement is actually equivalent to the identity (6.2) which, in particular, yields a proof for (6.2).
- Notice that that the law-invariance assumption cannot be dropped, as shown in Subsection 6.3 below.
6.2 A direct existence argument
- Lemma 6.1. Let U0 and U1 be two law-invariant monetary utility functions.
- The authors first recall that any attainable allocation (ξ0, ξ1) ∈ A(X) is dominated by some comonotone attainable allocation (ξ̂0, ξ̂1) ∈ A(X) in the sense of second order stochastic dominance.
- This result was proved by Landsberger and Meilijson [33] in the context of a finite probability space, and further extended to L∞ allocations in a general probability space by Dana and Meilijson [14].
- By Corollary 4.59 in [25], any law-invariant monetary utility function preserves second order stochastic dominance, which provides Ui ( ξ̂i ) ≥ Ui (ξi) .
- This justifies the duality-based existence argument reported in the beginning of Section 6. Lemma 6.2. Proof. (ii)=⇒(i) : Under (ii), the argument in the beginning of this section provides the required existence result.
Did you find this useful? Give us your feedback
Citations
252 citations
205 citations
150 citations
144 citations
132 citations
Cites background from "Optimal risk sharing for law invari..."
...DOI: 10.1111/j.1467-9965.2010.00450.x C© 2010 Wiley Periodicals, Inc. 743 Frittelli and Rosazza Gianin (2004), Staum (2004), Filipović and Kupper (2008), and Jouini, Schachermayer, and Touzi (2008)....
[...]
References
8,651 citations
"Optimal risk sharing for law invari..." refers background in this paper
...It was stressed in [2] that the V@R criterion for risk measuring leads to incoherent results because of the lack of sub-additivity....
[...]
...It was stressed in Artzner et al. (1999) that the V@R criterion for risk measuring leads to incoherent results because of the lack of sub-additivity....
[...]
...We also refer to Heath and Ku (2004) for analyzing Pareto optimal risk between banks defined by coherent risk measures in a finite probability space....
[...]
...In particular, positively homogeneous monetary utility functions can be identified with coherent risk measures, introduced in Artzner, Delbaen, Eber and Heath [2], while the general notion of monetary utility functions corresponds to convex risk measures as introduced independently by Föllmer and Schied [24], and Frittelli and Rossaza-Gianin [21]....
[...]
...In particular, positively homogeneous monetary utility functions can be identified with coherent risk measures, introduced in Artzner et al. (1999), while the general notion of monetary utility functions corresponds to convex risk measures as introduced independently by Föllmer and Schied (2002)…...
[...]
2,019 citations
1,809 citations
1,596 citations
1,302 citations
Related Papers (5)
Frequently Asked Questions (11)
Q2. What is the value of a measure preserving transformation of (,F,P?
A measure preserving transformation of (Ω,F ,P) is a bi-measurable bijection τ : (Ω,F ,P) → (Ω,F ,P) leaving P invariant, i.e., τ(P) = P.
Q3. What is the definition of Pareto optimal allocations?
Pareto optimal allocations are only defined up to a constant, i.e. given a Pareto optimal allocation (ξ0, ξ1), the attainable allocation (ξ0 + c, ξ1− c) is also Pareto optimal for every c ∈ R, 2. Let (ξ0, ξ1) be an attainable allocation which is not Pareto optimal.
Q4. What is the proof for the comonotone law-invariant monetary utility?
Let U0 and U1 be two comonotone law-invariant monetary utility functions, and let ϕ̄0, ϕ̄1 ∈ C↘ be the associated functions in the representation (2.3).
Q5. What is the comonotonicity of the pairs?
By Lemma 4.2, the authors see thatqξ0 is constant on [β, 1] . (5.4)Since U1 is strictly risk-averse conditionally on lower tail events, it follows again from the comonotonicity of the pairs (−Z, ξi), i = 0, 1, together with Lemma 5.1 thatqξ1 is constant on [0, β] .
Q6. What is the proof for the monetary utility function U0 and V 1i?
By Remark 2.3, the functions Ui and V 1i = χCϕ̄i are 〈L ∞,L1(P)〉−conjugate, where χ is the indicator function in the sense of convex analysis.
Q7. What is the proof of equivalence with (iii)?
2The characterization (ii) in the above Theorem 3.1 has a well-known extension for general utility functions (see e.g. [26]), which involves two Lagrange multipliers (simply leaving out the last sentence in the above proof of (i)=⇒(ii)).
Q8. Why is the set of Pareto optimal allocations empty?
In this example due to F. Delbaen [20], the authors show that the set of Pareto optimal allocations might be empty in the context where all monetary utility functions U0, U1, U0 U1, are positively-homogeneous, and have the Fatou property.
Q9. What is the proof of the maximization problem in the definition of U0 U1?
The authors first start by proving that the maximization problem in the definition of the supconvolution U0 U1 can be restricted to pairs (ξ0, ξ1) ∈ A↗(X).
Q10. what is the subset of all functions in C which are continuous at the point t?
D↘ } and its pointwise closureC↘ = {ϕ : [0, 1] −→ [0, 1] non-decreasing, concave, ϕ(0) = 0 and ϕ(1) = 1} .Clearly, C↘ is the subset of all functions in C↘ which are continuous at the point t = 0.
Q11. What is the simplest example of a Pareto optimal?
This example was triggered by a question of N. El Karoui, and shows that, even in a finite-dimensional setting, it may happen that the set of Pareto optimal allocations is empty for any possible aggregate risk X ∈ L∞.