Risk statistic is a critical factor not only for risk analysis but also for financial application. However, the traditional risk statistics may fail to describe the characteristics of regulator-based risk. In this paper, we consider the regulator-based risk statistics for portfolios. By further developing the properties related to regulator-based risk statistics, we are able to derive dual representation for such risk.
Department of Education of Guangdong Province2019KQNCX1561. Introduction
Risk measure is a popular topic in both financial application and theoretical research. The quantitative calculation of risk involves two problems: choosing an appropriate risk model and allocating risk to individual institutions. This has led to further research on risk statistics. In a seminal paper, Sun et al. and Liu et al. [1, 2] first introduced the class of natural risk statistics with representation results. Furthermore, Ahmed et al. [3] derived an alternative proof for the natural risk statistics. Later, Tian and Jiang and Tian and Suo [4, 5] obtained the representation results for convex risk statistics and quasiconvex risk statistics, respectively.
However, traditional risk statistics may fail to describe the characteristics of regulator-based risk. Therefore, the study of regulator-based risk statistics is particularly interesting. On the other hand, in the abovementioned research on risk statistics, the set-valued risk was never studied. Jouini et al. [6] pointed out that a set-valued risk measure is more appropriate than a scalar risk measure especially in the case where several different kinds of currencies are involved when one is determining capital requirements for the portfolio. Indeed, a natural set-valued risk statistic can be considered as an empirical (or a data-based) version of a set-valued risk measure. More recent studies of set-valued risk measures include those of [7–14] and the references therein.
The main focus of this paper is regulator-based risk statistics for portfolios. In this context, both empirical versions and data-based versions of regulator-based risk measures are discussed. By further developing the properties related to regulator-based risk statistics, we are able to derive their dual representations. Indeed, this class of risk statistics can be considered as an extension of those introduced in [15–17].
The remainder of this paper is organized as follows. In Section 2, we briefly introduce some preliminaries. In Section 3, we state the main results of regulator-based risk statistics, including the dual representations. Section 4 investigates the data-based versions of regulator-based risk measures. Finally, in Section 5, the main proofs in this paper are discussed.
2. Preliminary Information
In this section, we briefly introduce some preliminaries that are used throughout this paper. Let d≥1 be a fixed positive integer. The space ℝd×n represents the set of financial risk positions. With positive values of X∈ℝd×n, we denote the gains while the negative denotes the losses. Let nj be the sample size of D=X1,…,Xd in the jth scenario, j=1,…,l. Let n≔n1+⋯+nl. More precisely, suppose that the behavior of D is represented by a collection of data X=X1,…,Xd∈ℝn×⋯×ℝn, where Xi=Xi,1,…,Xi,l∈ℝn and Xi,j=x1i,j,…,xnji,j∈ℝnj is the data subset that corresponds to the jth scenario with respect to Xi. For each j=1,…,l, h=1,…,nj, Xhj≔xh1,j,xh2,j,…,xhd,j is the data subset that corresponds to the hth observation of D in the jth scenario.
In this paper, an element z of ℝd is denoted by z≔z1,…,zd. An element X of ℝd×n is denoted by X≔X1,…,Xd≔x11,1,…,xn11,1,…,x11,l,…,xnl1,l,…,x1d,1,⋯xn1d,1,…,x1d,l,…,xnld,l. Let K be a closed convex polyhedral cone of ℝd where K⊇ℝ++d≔x1,…,xd∈ℝd;xi>0,1≤i≤d and K∩ℝ−d=∅ where ℝ−d≔x1,…,xd∈ℝd;xi≤0,1≤i≤d. Let K+ be the positive dual cone of K, that is, K+≔u∈ℝd:utrv≥0 for any v∈K, where utr means the transpose of u. For any X=X1,…,Xd,Y=Y1,…,Yd∈ℝd×n, X+Y stands for X1+Y1,…,Xd+Yd and aX stands for aX1,…,aXd for a∈ℝ. Denote K1n≔z11n,z21n,…,zd1n:z∈K and z1n≔z,z,…,z:z∈ℝ∈ℝn where 1n≔1,…,1∈ℝn. By K1n+, and we denote the positive dual cone of K1n in ℝd×n, i.e., K1n+≔w∈ℝd×n:wztr≥0 for any z∈K. The partial order with respect to K is defined as a≤Kb, which means b−a∈K where a,b∈ℝd, and X≤K1nY means Y−X∈K1n where X,Y∈ℝd×n.
Let M≔ℝm×0d−m be the linear subspace of ℝd for 1≤m≤d. The introduction of M was considered in [6, 9]. Denote M+≔M∩ℝ+d, where ℝ+d≔x1,…,xd∈ℝd;xi≥0,1≤i≤d and M⊥≔0m×ℝd−m. Therefore, a regulator can only accept security deposits in the first m reference instruments. Denoting KM≔K∩M by the closed convex polyhedral cone in M, KM+≔u∈M:utrz≥0 for any z∈KM the positive dual cone of KM in M and int KM is the interior of KM in M. We denote QMt≔A⊂M:A=clcoA+KM and QM+t≔A⊂KM:A=clcoA+KM, where the clcoA represents the closed convex hull of A.
By [18], a set-valued risk statistic is any map ρ,(1)ρ:ℝd×n⟶2M,which can be considered as an empirical (or a data-based) version of a set-valued risk measure. The axioms related to this set-valued risk statistic are organized as follows:
[A0] Normalization: KM⊆ρ0 and ρ0∩−intKM=ϕ
[A1] Monotonicity: for any X,Y∈ℝd×n, X−Y∈K1n implies that ρX⊇ρY
[A2] M-translative invariance: for any X∈ℝd×n and z∈ℝd, ρX−z1n=ρX+z
[A3] Convexity: for any X,Y∈ℝd×n and λ∈0,1, ρλX+1−λY⊇λρX+1−λρY
[A4] Positive homogeneity: ρλX=λρX for any X∈ℝd×n and λ>0
[A5] Subadditivity: ρX+Y⊇ρX+ρY for any X,Y∈ℝd×n
We end this section with more notations. A function ρ:ℝd×n⟶2M is said to be proper if dom ρ≔X∈ℝd×n:ρX≠∅≠∅ and ρX≠M for all X∈dom ρ. ρ is said to be closed if graph ρ is a closed set. For the properties of the graphs, see [19–21].
3. Empirical Versions of Regulator-Based Risk Measures
In this section, we state the dual representations of regulator-based risk statistics, which are the empirical versions of regulator-based risk measures. Firstly, for any X∈ℝd×n, X∧K1n0 is defined as follows:(2)X∧K1n0≔X,X∉K1n,0,X∈K1n.
Therefore, the position that belongs to K regarded is as 0 position. Next, we derive the properties related to regulator-based risk statistics.
Definition 1.
A regulator-based risk statistic is a function ϱ:ℝd×n⟶QM+t that satisfies the following properties:
[P0] Normalization: KM⊆ϱ0 and ϱ0∩−intKM=ϕ
[P1] Cash cover: for any z∈KM, z∈ϱ−z1n
[P2] Monotonicity: for any X,Y∈ℝd×n, X−Y∈ℝd×n∩K1n implies that ϱX⊇ϱY
[P3] Regulator-dependence: for any X∈ℝd×n, ϱX=ϱX∧K1n0
[P4] Convexity: for any X,Y∈ℝd×n and λ∈0,1, ϱλX+1−λY⊇λϱX+1−λϱY
Remark 1.
The property of P1 means any fixed negative risk position −z can be canceled by its positive quality z; P2 says that if X1 is bigger than X2 for the partial order in K, then the X1 need less capital requirement than X2, so ϱX1 contains ϱX2; P3 means the regulator-based risk statistics start only from the viewpoint of regulators which only care about the positions that need to pay capital requirements, while the positions that belong to K regarded as 0 position.
We now construct an example for regulator-based risk statistics.
Example 1.
The coherent risk measure AV@R was studied by Föllmer and Schied [22] in detail. They have given several representations and many properties such as law invariance and the Fatou property [12]. First, they introduced set-valued AV@R, where the representation result is derived. Moveover, they also proved that it is a set-valued coherent risk measure. We now define the regulator-based average value at risk. For any X∈ℝd×n and 0<α<1, we define ϱX as(3)ϱX≔AV@RαlossX,≔infz∈ℝd1α−X∧K1n0M+z+−z+ℝ+m,
It is clear that ϱ satisfies the cash cover, monotonicity, regulator dependence properties, and convexity, so ϱ is a regulator-based risk statistic.
Definition 2.
Let Y∈ℝd×n, u∈M. Define a function SY,uX:ℝd×n⟶2M as(4)SY,uX≔z∈M:XtrY≤utrz.
In fact, the SY,uX is the support function of X. Before we derive the dual representations of regulator-based risk statistics, the Legendre–Fenchel conjugate theory ([9]) should be recalled.
Lemma 1 (see [9], Theorem 2).
Let R:ℝd×n⟶QMt be a set-valued closed convex function. Then, the Legendre–Fenchel conjugate and the biconjugate of R can be defined, respectively, as(5)−R∗Y,u≔cl∪X∈ℝd×nRX+SY,u−X,Y∈ℝd×n,u∈ℝd,(6)RX=R∗∗X≔∩Y,u∈ℝd×n×KM+\0−R∗Y,u+SY,uX,X∈ℝd×n.
Definition 3.
(indicator function). For any Z⊆ℝd×n, the QMt-valued indicator function IZ:ℝd×n⟶QMt is defined as(7)IZX≔clKM,X∈Z,ϕ,X∉Z.
Remark 2.
The conjugate of QMt-valued indicator function IZ is(8)−IZ∗Y,u≔cl∪X∈ZSY,u−X.
Remark 3.
It is easy to see that the regulator-based risk statistic ϱ does not have cash additivity, see [9]. However, ϱ has cash subadditivity introduced in [23, 24]. Indeed, from Theorem 2 of [10], ϱ satisfies the Fatou property. Then, considering any X∈ℝd×n and z∈KM, for any ε∈0,1, we have(9)ϱ1−εX−z1n=ϱ1−εX+ε−zε1n,⊇1−εϱX+εϱ−zε1n,⊇1−εϱX+z,where the last inclusion is due to the property P1. Using the arbitrariness of ε, we have the following lemma.
Lemma 2.
Assume that ϱ is a regulator-based risk statistic. For any z∈ℝ+d, X∈ℝd×n,(10)ϱX−z1n⊇ϱX+z,which also implies(11)ϱX+z1n⊆ϱX−z.
Proposition 1.
Let ϱ:ℝd×n⟶QM+t be a proper closed convex regulator-based risk statistic with u∈−∑j=1l∑h=1njYh1,j,…,−∑j=1l∑h=1njYhd,j+M⊥∩KM+\0. Then,(12)−ϱ∗Y,u=cl∪X∈ℝd×nSY,u−X,Y∈−ℝ+d×n∩K+1n,M,elsewhere.
Now, we state the main result of this paper, the dual representations of regulator-based risk statistics.
Theorem 1.
If ϱ:ℝd×n⟶QM+t is a proper closed convex regulator-based risk statistic, then there is a −α:−ℝ+d×n∩K+1n×KM+0⟶QM+t, which is not identically M of the set(13)W=Y,u∈−ℝ+d×n∩K+1n×KM+\0u∈−∑j=1l∑h=1njYh1,j,…,−∑j=1l∑h=1njYhd,j+M⊥,such that for any X∈ℝd×n,(14)ϱX=∩Y,u∈W−αY,u+SY,uX∧K1n0.
4. Alternative Data-Based Versions of Regulator-Based Risk Measures
In this section, we develop another framework, the data-based versions of regulator-based risk measures. This framework is a little different from the previous one. However, almost all the arguments are the same as those in the previous section. Therefore, we only state the corresponding notations and results and omit all the proofs and relevant explanations.
We replace M by M˜∈ℝd×n that is a linear subspace of ℝd×n. We also replace K by K˜∈ℝd×n that is a is a closed convex polyhedral cone where K˜⊇ℝ++d×n. The partial order with respect to K˜ is defined as X≤K˜Y, which means Y−X∈K˜. Let M˜+≔M˜∩ℝ+d×n. Denoting K˜M˜≔K˜∩M˜ by the closed convex polyhedral cone in M˜, K˜M˜+≔u˜∈M:u˜trz˜≥0 for any z˜∈K˜M˜ is the positive dual cone of K˜M˜ in M˜ and intK˜M˜ is the interior of K˜M˜ in M˜. We denote QM˜t≔A˜⊂M˜:A˜=clcoA˜+K˜M˜ and QM˜+t≔A˜⊂K˜M˜:A˜=clcoA˜+K˜M˜. We still start from the viewpoint of regulators which only care about the positions that need to pay capital requirements. Therefore, for any X∈ℝd×n, we define X∧K˜0 as(15)X∧K˜0≔X,X∉K˜,0,X∈K˜,
Then, we state the axioms related to regulator-based risk statistics.
Definition 4.
A regulator-based risk statistic is a function ϱ˜:ℝd×n⟶QM˜+t that satisfies the following properties:
[Q0] Normalization: K˜M˜⊆ϱ˜0 and ϱ˜0∩−intK˜M˜=ϕ
[Q1] Cash cover: for any z˜∈K˜M˜, z˜∈ϱ˜−z˜
[Q2] Monotonicity: for any X1,X2∈ℝd×n, X1−X2∈ℝd×n∩K˜ implies that ϱ˜X1⊇ϱ˜X2
[Q3] Regulator-dependence: for any X∈ℝd×n, ϱ˜X=ϱ˜X∧K˜0
[Q4] Convexity: for any X,Y∈ℝd×n, λ∈0,1, ϱ˜λX+1−λY⊇λϱ˜X+1−λϱ˜Y
We need more notations. Let Y∈ℝd×n, u˜∈M˜. Define a function SY,u˜X:ℝd×n⟶2M˜ as(16)SY,u˜X≔z˜∈M˜:XtrY≤u˜trz˜,let R˜:ℝd×n⟶QM˜t be a set-valued closed convex function. Then, the Legendre–Fenchel conjugate and the biconjugate of R˜ can be defined, respectively, as(17)−R˜∗Y,u≔cl∪X∈ℝd×nR˜X+SY,u˜−X,Y∈ℝd×n,u˜∈ℝd×n,(18)R˜X=R˜∗∗X≔∩Y,u˜∈ℝd×n×K˜M˜+\0−R˜∗Y,u˜+SY,u˜X,X∈ℝd×n.
For any Z˜⊆ℝd×n, the QM˜t-valued indicator function IZ˜:ℝd×n⟶QM˜t is defined as(19)IZ˜X≔clK˜M˜,X∈Z˜,ϕ,X∉Z˜.
The conjugate of QM˜t-valued indicator function IZ˜ is(20)−IZ˜∗Y,u˜≔cl∪X∈Z˜SY,u˜−X.
Assume that ϱ˜ is a regulator-based risk statistic. For any z˜∈ℝ+d×n, X∈ℝd×n,(21)ϱ˜X−z˜⊇ϱ˜X+z˜,which also implies(22)ϱ˜X+z˜⊆ϱ˜X−z˜.
Next, we state the dual representations of regulator-based risk statistics.
Proposition 2.
Let ϱ˜:ℝd×n⟶QM˜+t be a proper closed convex regulator-based risk statistic with u˜∈−∑j=1l∑h=1njYh1,j,…,−∑j=1l∑h=1njYhd,j+M˜⊥∩K˜M˜+\0. Then,(23)−ϱ˜∗Y,u˜=cl∪X∈ℝd×nSY,u˜−X,Y∈−ℝ+d×n∩K˜+,M˜,elsewhere.
Theorem 2.
If ϱ˜:ℝd×n⟶QM˜+t is a proper closed convex regulator-based risk statistic, then there is a −α:−ℝ+d×n∩K˜+×K˜M˜+\0⟶QM˜+t, that is not identically M˜ of the set(24)W˜=Y,u˜∈−ℝ+d×n∩K˜+×K˜M˜+\0:u˜∈−∑j=1l∑h=1njYh1,j,…,−∑j=1l∑h=1njYhd,j+M˜⊥such that for any X∈ℝd×n,(25)ϱ˜X=∩Y,u˜∈W−αY,u˜+SY,u˜X∧K˜0.
5. Proofs of Main ResultsProof of Lemma 2.
the proof of Lemma 2 is straightforward from Remark 3.
Proof of Proposition 1.
if Y∉−ℝ+d×n∩K+1n, there exits an X¯∈ℝd×n∩K1n such that X¯trY>0. Using the definition of SY,u, we have SY,u−tX¯=z∈M:−tX¯trY≤utrz for t>0. Therefore,(26)cl∪X∈ℝd×nSY,u−X⊇∪t>0SY,u−tX¯=M.
The last equality is due to −tX¯trY⟶−∞ when t⟶+∞. Using the definition of SY,u, we conclude that cl ∪X∈ℝd×nSY,u−X⊆M. Hence,(27)cl ∪X∈ℝd×nSY,u−X=M,Y∉−ℝd×n∩K+1n.
It is easy to check that for any X∈ℝd×n and v∈M,(28)SY,u−X−v1n=z∈M:−XtrY≤utrz+Ytrv1n,=z−v∈M:−XtrY≤utrz−v+Y+u1ntrv1n+v,=z∈M:−XtrY≤utrz+Y+u1ntrv1n+v,when −∑j=1l∑h=1njYh1,j,…,−∑j=1l∑h=1njYhd,j+u∈M⊥, and we have SY,u−X−v1n=SY,u−X+v. However, u∉−∑j=1l∑h=1njYh1,j,…,−∑j=1l∑h=1njYhd,j+M⊥. Therefore, −∑j=1l∑h=1njYh1,j,…,−∑j=1l∑h=1njYhd,j+u∉M⊥, and we can find v∈M such that for any z∈M,(29)−XtrY≤utrz+Y+u1ntrv1n.
Therefore, we have(30)z+v∈SY,u−X−v1n.
Therefore,(31)∪z,v∈Mz+v⊂∪v∈MSY,u−X−v1n.
Therefore,(32)M⊂∪v∈MSY,u−X−v1n.
From the definition of SY,u, the inverse inclusion is always true. So, we conclude that(33)M=∪v∈MSY,u−X−v1n.
It is also easy to check that(34)−ϱ∗Y,u=cl∪X∈ℝd×n,v∈MϱX+v1n+SY,u−X−v1n,=cl∪X∈ℝd×n,v∈MϱX+v1n+M,=M,where the last equality comes from the fact that M is a linear space and ϱX⊆M. We now derive that −ϱ∗Y,u=cl∪X∈ℝd×nSY,u−X. In this context, from −ϱ∗Y,u=cl∪X∈ℝd×nϱX+SY,u−X, we derive it in two cases.
Case 1.
When X∧K1n0=0, using the definition, we have ϱX=ϱ0∋0. Hence,(35)cl∪X∈ℝd×nϱX+SY,u−X⊃cl∪X∈ℝd×nSY,u−X.
Case 2.
When X∧K1n0=X, we can always find an α∈KM such that α∈ϱX. Then,(36)ϱX+SY,u−X⊇α+SY,u−X=SY,u−X−α1n=SY,u−β,where β=X+α1n. It is relatively simple to check that β∈ℝd×n. Therefore,(37)cl∪X∈ℝd×nϱX+SY,u−X⊇cl∪z∈ℝd×nSY,u−z,that is,(38)−ϱ∗Y,u⊇cl∪X∈ℝd×nSY,u−X.
Consequently, we have(39)−ϱ∗Y,u⊇cl∪X∈ℝd×nSY,u−X.
We now need only to derive that −ϱ∗Y,u⊆cl∪X∈ℝd×nSY,u−X. In fact, for any z∈ϱX and X∈ℝd×n, X+z1n∈ℝd×n. Therefore,(40)cl∪X∈ℝd×nSY,u−X=cl∪X∈ℝd×nSY,u−X⊇SY,u−X−z1n=SY,u−X+z.
Using the arbitrariness of z, we have(41)ϱX+SY,u−X⊆cl∪X∈ℝd×nSY,u−X.
Therefore,(42)−ϱ∗Y,u⊆cl∪X∈ℝd×nSY,u−X.
Proof of Theorem 1.
the proof is straightforward from Lemma 1 and Proposition 1.
Data Availability
No data and code were generated or used during the study.
Disclosure
This manuscript has been released as a preprint at arXiv: 1904.08829v4.
Conflicts of Interest
The authors declare no conflicts of interest.
Acknowledgments
This work was supported by funds from Education Department of Guangdong (2019KQNCX156).
HeydeC. C.KouS. G.PengX. H.2007New York City, NY, USAColumbia UniversityKouS.PengX.HeydeC. C.External risk measures and basel accords201338339341710.1287/moor.1120.05772-s2.0-84878965180AhmedS.FilipovićD.SvindlandG.A note on natural risk statistics200836666266410.1016/j.orl.2008.06.0092-s2.0-54249132481TianD.JiangL.Quasiconvex risk statistics with scenario analysis20159211112110.1007/s11579-014-0136-y2-s2.0-84940005816TianD.SuoX.A note on convex risk statistic201240655155310.1016/j.orl.2012.09.0112-s2.0-84869239706JouiniE.MeddebM.TouziN.Vector-valued coherent risk measures20048453155210.1007/s00780-004-0127-62-s2.0-24144464232AraratC.HamelA. H.RudloffB.Set-valued shortfall and divergence risk measures2017https://arxiv.org/abs/1405.4905FarkasW.Koch-MedinaP.MunariC.Measuring risk with multiple eligible assets20159132710.1007/s11579-014-0118-02-s2.0-84939895804HamelA. H.A duality theory for set-valued functions i: fenchel conjugation theory200917215318210.1007/s11228-009-0109-02-s2.0-73549118156HamelA. H.HeydeF.Duality for set-valued measures of risk201011669510.1137/0807434942-s2.0-79958772235HamelA. H.HeydeF.RudloffB.Set-valued risk measures for conical market models20115112810.1007/s11579-011-0047-02-s2.0-79957663662HamelA. H.RudloffB.YankovaM.Set-valued average value at risk and its computation20137222924610.1007/s11579-013-0094-92-s2.0-84876154066LabuschagneC. C. A.Offwood-Le RouxT. M.Representations of set-valued risk measures defined on the $$l$$ l -tensor product of banach lattices201418361963910.1007/s11117-013-0267-z2-s2.0-84957434024MolchanovI.CascosI.Multivariate risk measures: a constructive approach based on selections201626486790010.1111/mafi.120782-s2.0-84907979207ChenY.SunF.HuY.Coherent and convex loss-based risk measures for portfolio vectors201822139941410.1007/s11117-017-0517-62-s2.0-85024485760ContR.DeguestR.HeX. D.Loss-based risk measures201330213316710.1524/strm.2013.11322-s2.0-85010044072SunF.ChenY.HuY.Set-valued loss-based risk measures201822385987110.1007/s11117-017-0550-52-s2.0-85039979835ChenY.HuY.Set-valued risk statistics with scenario analysis2017131253710.1016/j.spl.2017.08.0042-s2.0-85029522918LiuJ.-B.WangC.WangS.WeiB.Zagreb indices and multiplicative zagreb indices of eulerian graphs2019421677810.1007/s40840-017-0463-22-s2.0-85056620341LiuJ.ZhaoJ.MinJ.CaoJ.On the hosoya index of graphs formed by a fractal graph2019278193510.1142/s0218348x19501354LiuJ.-B.ZhaoJ.HeH.ShaoZ.Valency-based topological descriptors and structural property of the generalized sierpiński networks201917761131114710.1007/s10955-019-02412-2FöllmerH.SchiedA.2011ThirdBerlin, GermanyWalter de Gruyter BerlinEl KarouiN.RavanelliC.Cash subadditive risk measures and interest rate ambiguity200919456159010.1111/j.1467-9965.2009.00380.x2-s2.0-72449149340SunF.HuY.Set-valued cash sub-additive risk measures201933224125710.1017/s02699648180001412-s2.0-85045155372