Abstract
We investigate capital allocation based on the higher moment risk measure at a confidence level \(q\in (0,1)\). To reflect the excessive prudence of today’s regulatory frameworks in banking and insurance, we consider the extreme case with \(q\uparrow 1\) and study the asymptotic behavior of capital allocation for heavy-tailed and asymptotically independent/dependent risks. Some explicit asymptotic formulas are derived, demonstrating that the capital allocated to a specific line is asymptotically proportional to the Value at Risk of the corresponding individual risk. In addition, some numerical studies are conducted to examine their accuracy.
Similar content being viewed by others
Data Availability
Not applicable.
References
Asimit AV, Furman E, Tang Q, Vernic R (2011) Asymptotics for risk capital allocations based on conditional tail expectation. Insur Math Econ 49(3):310–324
Bingham NH, Goldie CM, Teugels JL (1987) Regular Variation. Cambridge University Press, Cambridge
Chen Y, Liu J (2022) An asymptotic study of systemic expected shortfall and marginal expected shortfall. Insur Math Econ 105:238–251
Chen Y, Yuen KC (2009) Sums of pairwise quasi-asymptotically independent random variables with consistent variation. Stoch Model 25(1):76–89
de Haan L, Resnick SI (1981) On the observation closest to the origin. Stoch Process Appl 11(3):301–308
Embrechts P, Klüppelberg C, Mikosch T (1997) Modelling extremal events: for insurance and finance. Springer, Berlin
Fougeres AL, Mercadier C (2012) Risk measures and multivariate extensions of Breiman’s theorem. J Appl Probab 49(2):364–384
Gómez F, Tang Q, Tong Z (2022) The gradient allocation principle based on the higher moment risk measure. J Bank Finance 143:106544
Goovaerts MJ, Kaas R, Dhaene J, Tang Q (2004) Some new classes of consistent risk measures. Insur Math Econ 34(3):505–516
Haezendonck J, Goovaerts M (1982) A new premium calculation principle based on Orlicz norms. Insur Math Econ 1(1):41–53
Kalkbrener M (2005) An axiomatic approach to capital allocation. Math Financ 15(3):425–437
Kley O, Klüppelberg C, Reinert G (2016) Risk in a large claims insurance market with bipartite graph structure. Oper Res 64(5):1159–1176
Krokhmal PA (2007) Higher moment coherent risk measures. Quant Finance 7(4):373–387
Li J (2018) A revisit to asymptotic ruin probabilities for a bidimensional renewal risk model. Stat Probab Lett 140:23–32
McNeil AJ, Frey R, Embrechts P (2015) Quantitative risk management: concepts, techniques and tools. Princeton University Press, Princeton
Resnick SI (1987) Extreme values, regular variation and point processes. Springer, New York
Resnick SI (2007) Heavy-tail phenomena: probabilistic and statistical modeling. Springer Science & Business Media, Berlin
Shi X, Tang Q, Yuan Z (2017) A limit distribution of credit portfolio losses with low default probabilities. Insur Math Econ 73:156–167
Tang Q, Yang F (2012) On the Haezendonck-Goovaerts risk measure for extreme risks. Insur Math Econ 50(1):217–227
Tang Q, Yuan Z (2013) Asymptotic analysis of the loss given default in the presence of multivariate regular variation. N Am Actuar J 17(3):253–271
Tang Q, Yuan Z (2014) Randomly weighted sums of subexponential random variables with application to capital allocation. Extremes 17:467–493
Tasche D (2004) Allocating portfolio economic capital to sub-portfolios. In: Dev A (ed) Contribution to economic capital: a practitioner guide. Risk Books
Rockafellar RT, Uryasev S (2000) Optimization of conditional value-at-risk. J Risk 2:21–42
Rockafellar RT, Uryasev S (2002) Conditional value-at-risk for general loss distributions. J Bank Finance 26(7):1443–1471
Acknowledgements
The authors would like to thank two anonymous referees for their careful reading and valuable comments, which have helped significantly improve the quality of the manuscript. The research of Yiqing Chen was supported by a Summer Research Grant from the Zimpleman College of Business at Drake University. The research of Jiajun Liu was supported by the National Natural Science Foundation of China (NSFC: 12201507; 72171055), and the XJTLU Research Enhancement Fund (REF-22-02-003).
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix
Appendix
1.1 Lemmas
The following first lemma is a restatement of Proposition 2.2.3 of [2]:
Lemma A.1
If \(f\in \textrm{RV}_{-\alpha }\) for some \(\alpha \ge 0\), then for arbitrary \(0<\varepsilon <1\) there is some \(x_{0}>0\) such that the Potter’s bounds
hold whenever \(x\ge x_{0}\) and \(xy\ge x_{0}\).
The next lemma is closely related to Theorem 3.1 of [4], but neither of them covers the other:
Lemma A.2
Under the conditions of Theorem 3.1, we have
Proof
We only need to prove the first relation, which immediately implies the second one as \({\overline{F}}\in \textrm{RV}_{-\alpha }\). Denote by \({\mathcal {I}}\) the subset of \(\{1,\ldots ,d\}\) such that \(\theta _{i}>0\) for \(i\in {\mathcal {I}}\) and \(\theta _{i}=0\) for \(i\in {\mathcal {I}}^{c}=\{1,\ldots ,d\}\backslash {\mathcal {I}}\), and denote by \(d_{1}\) and \(d_{2}=d-d_{1}\) the cardinalities of the sets \({\mathcal {I}}\) and \({\mathcal {I}}^{c}\), respectively. Write \(S_{{\mathcal {I}}}=\sum _{i\in {\mathcal {I}}}X_{i}\) and \(S_{{\mathcal {I}} ^{c}}=\sum _{i\in {\mathcal {I}}^{c}}X_{i}\). Note that \({\mathcal {I}}\ne \emptyset \) since \(\Theta >0\). By Theorem 3.1 of [4] and the conditions in (3.1),
Thus, we will have proved the lemma if \({\mathcal {I}}^{c}=\emptyset \). Now assume \({\mathcal {I}}^{c}\ne \emptyset \). Then for every small \(\varepsilon \in (0,1)\), applying (A.1) we obtain
On the other hand,
By (A.1), the first term on the right-hand side is asymptotic to \( (1+\varepsilon )^{-\alpha }\left( \sum _{i\in {\mathcal {I}}}\theta _{i}\right) {\overline{F}}(x)\). With \(c=\frac{1+\varepsilon }{d_{1}}\wedge \frac{ \varepsilon }{d_{2}}\), each summand in the second term is no more than
since \(X_{i}\) and \(X_{j}\) are QAI. In summary,
By the arbitrariness of \(\varepsilon \) and the fact that \(\Theta =\sum _{i\in {\mathcal {I}}}\theta _{i}\) since \(\sum _{i\in {\mathcal {I}}^{c}}\theta _{i}=0\), we complete the proof. \(\square \)
In the following lemma, we establish a novel inequality in the spirit of Potter’s upper bound in Lemma A.1 but in a multidimensional scenario. Notably, it does not require any dependence information among the individual random variables.
Lemma A.3
Consider d real-valued random variables \(X_{1}\), ..., \( X_{d}\) distributed by \(F_{1}\), ..., \(F_{d}\), respectively, and denote by S their sum (1.1). Assume that there is some reference distribution F on \({\mathbb {R}}_{+}\), with a tail \({\overline{F}}\in \textrm{RV }_{-\alpha }\) for some \(\alpha \ge 0\), such that the limit
exists for each \(i=1,\ldots ,d\). Then for every small \(0<\varepsilon <1\), there are some large \(K>0\) and \(x_{1}>0\) dependent on \(\varepsilon \) such that the inequality
holds whenever \(ux\ge x_{1}\) and \(vx\ge x_{1}\).
Proof
In this proof, for ease of exposition we use K to denote a large positive constant, which depends on \(\varepsilon \) but not on the working variables u, v, and x, and which may vary from place to place. By (A.2),
For every small \(\varepsilon \in (0,1)\), we derive
Choose \(x_{1}=\frac{d-1}{\varepsilon }x_{0}\ge x_{0}\), where \(x_{0}\) is specified in Lemma A.1, so that whenever \(ux\ge x_{1}\) and \(vx\ge x_{1}\) we have \(ux\ge x_{0}\) and \(\frac{\varepsilon }{d-1} vx\ge x_{0}\). For the first term \(I_{k}\) in (A.5), by (A.4 ) and Lemma A.1,
For the second term in (A.5), if \(u\ge v\), we have
while if \(u<v\), we have
Thus, in any case,
Plugging (A.6)–(A.7) into (A.5) yields (A.3). \(\square \)
The following last lemma bears a resemblance to Lemma A.3 of [3], but we come up with a much shorter proof in the current setting:
Lemma A.4
Under the conditions of Theorem 3.1(ii), it holds for every \(u,v>0\) that
Proof
To obtain a suitable upper bound, we start from (A.5) in which \(\varepsilon \in (0,1)\) is arbitrarily fixed. For the first term \( I_{k} \), by (3.1) for \(F_{k}\) for which \(\theta _{k}>0\), and by \( {\overline{F}}\in \textrm{RV}_{-\alpha }\),
For the second term, since \(X_{k}\) and \(X_{i}\) are QAI, we have
as in the second half of the proof of Lemma A.2. Substituting these results into (A.5) yields
The corresponding lower bound can be obtained in a similar way. We derive
Based on these upper and lower bounds and by the arbitrariness of \( \varepsilon \), we obtain (A.8). \(\square \)
1.2 Proof of Theorem 3.1
(i) By Lemma A.2, \(\overline{F_{S}}\in \textrm{RV}_{-\alpha }\). Thus, for \(p\ge 1\), by Theorem 4.1 of [19],
where in the last step we have applied \(F_{S}^{\leftarrow }({\tilde{q}})\sim \Theta ^{1/\alpha }F^{\leftarrow }({\tilde{q}})\) stated in Lemma A.2. This proves (3.2).
(ii) We need to formulate the proof into two parts: \(p=1\) and \(p>1\).
\(\underline{{\hbox {Step 1:\ Assume}} p=1.}\)
For this case, \({\tilde{q}}=q\). We start from (2.5). For brevity, denote \(x=F_{S}^{\leftarrow }(q)\), which tends to \(\infty \) as \(q\uparrow 1\). We expand
with
with \(\delta >0\) introduced to be arbitrarily fixed and small.
For \(I_{1}\), by Lemma A.3, for every small \(0<\varepsilon <(\alpha -1)\wedge 1\), there is some large \(K>0\) such that, for all large x and all \(u\ge \delta \),
Note that \(\int _{\delta }^{\infty }\left( u\vee 1\right) ^{-\alpha +\varepsilon }\textrm{d}u<\infty \). Thus, by the dominated convergence theorem
where the last step is due to Lemma A.4.
For \(I_{2}\), we examine
where \({\tilde{u}}=u\wedge \frac{1+u}{d-1}\). As \(x\rightarrow \infty \), each conditional probability \(P\left( \left. X_{k}<-{\tilde{u}}x\right| X_{i}> {\tilde{u}}x\right) \) tends to 0 uniformly for all \({\tilde{u}}\ge \delta \wedge \frac{1+\delta }{d-1}\) since \(X_{k}\) and \(X_{i}\) are QAI. Moreover,
By (A.4), \(\frac{P\left( X_{i}>ux\right) }{{\overline{F}}(ux)}\) is uniformly bounded for all \(u>\delta \) and \(x>0\). By Lemma A.1, for arbitrarily fixed and small \(0<\varepsilon <\left( \alpha -1\right) \wedge 1\) and for all large \(x>0\),
while \(\int _{\delta }^{\infty }\left( u^{-\alpha -\varepsilon }\vee u^{-\alpha +\varepsilon }\right) \textrm{d}u<\infty \). Thus, by the dominated convergence theorem,
For \(I_{3}\), we derive
where the last step is due to Lemma A.2.
Finally, substituting (A.10)–(A.12) into (A.9) and keeping in mind the arbitrariness of \(\delta \), we obtain
where the last step follows from \(F_{S}^{\leftarrow }(q)\sim \Theta ^{1/\alpha }F^{\leftarrow }(q)\), as stated in Lemma A.2. This proves (3.3) for \(p=1\).
\(\underline{{\hbox {Step 2:\ Assume}} p>1.}\)
We start from (2.6) in which \(s_{*}\) is the unique solution to equation (2.4). By Lemma 2.2 of [19], as \( q\uparrow 1\) we have \(s_{*}\uparrow \infty \), and by relation (4.4) of [19], we have
We need to explicitize the asymptotics for both the numerator and the denominator of (2.6). For the numerator, we expand
Similar to the proof in step 1, we split the double integral above into three parts as
with
with \(\delta >0\) introduced to be arbitrarily fixed and small. As in step 1, by virtue of Lemmas A.3–A.4, we can verify the applicability of the dominated convergence theorem in the following derivations, but we skip details:
and
We deal with \(I_{3}\) as follows:
where the last step is due to Lemma A.2 and the condition \( {\overline{F}}\in \textrm{RV}_{-\alpha }\). Substituting (A.15)–(A.17) into (A.14) and keeping in mind the arbitrariness of \( \delta \), we obtain
For the denominator of (2.6), we derive
where in the second last step we have applied the dominated convergence theorem justified by Lemmas A.1–A.2.
Finally, we conclude that
where the last step is due to (A.13). This proves (3.3) for \(p>1\).
1.3 Proof of Theorem 3.2
(i) First of all, as pointed out in Remark 2.1, under the conditions of Theorem 3.2, in particular (3.5), when considering the limit behavior of the probability of \(\frac{{\textbf{X}}}{ x}\) over a Borel set \(B\subset [-\varvec{\infty },\varvec{ \infty }]\backslash \{{\textbf{0}}\}\), we can always restrict \(\frac{{\textbf{X}} }{x}\) to \(B\cap [{\textbf{0}},\varvec{\infty }]\backslash \{\textbf{ 0}\}\). Thus, it holds for every \(u\ge -\infty \) and \(v>0\) that
In fact, by Lemma A.1 of [18], one can easily verify that \(\mu \) assigns mass zero to the boundary of the set \(\left( {\textbf{x}}\in [{\textbf{0}},\varvec{\infty }]:x_{k}>u_{+},\sum _{i=1}^{d}x_{i}>1+v\right) \). Relation (A.18) will be essentially used in the proof of (ii).
To complete the proof of (i), we take \(u=-\infty \) and \(v=1\) in (A.18) and obtain
The limit result (A.19) implies that \(\overline{F_{S}}\in \textrm{RV} _{-\alpha }\) and further that
similar to Lemma A.2. Then by Theorem 4.1 of [19],
This proves (3.6).
(ii) The same as the proof of Theorem 3.1, we formulate the proof into two parts: \(p=1\) and \(p>1\).
\(\underline{{\hbox {Step 1:\ Assume}} p=1.}\)
For this case, \({\tilde{q}}=q\). We start from (2.5) and denote \( s=F_{S}^{\leftarrow }(q)\), which tends to \(\infty \) as \(q\uparrow 1\). We expand
with \(\delta >0\) introduced to be arbitrarily fixed and small. For \(I_{1}\), by (A.19),
For \(I_{2}\), by (A.18) with \(v=1\), subject to an dominated convergence argument based on Lemma A.3, we obtain
For \(I_{3}\), following Remark 2.1, we have
For a direct verification, we derive
where the last step is due to (A.19) and (3.5). For the integral \(\int _{\delta }^{\infty }P\left( X_{k}>ys\right) \textrm{d}y\), by the dominated convergence theorem justified by Lemma A.1, we have
This verifies (A.24). Substituting (A.22)–(A.24) into (A.21) and bearing in mind the arbitrariness of \(\delta \), we arrive at
where the last step is due to (A.20). This proves (3.7) for \( p=1\).
\(\underline{{\hbox {Step 2:\ Assume}} p>1.}\)
We start from (2.6) in which \(s_{*}\) is the unique solution to equation (2.4). By Lemma 2.2 of [19], as \( q\uparrow 1\) we have \(s_{*}\uparrow \infty \), and by relation (4.4) of [19],
As in the corresponding proof of Theorem 3.1, we need to explicitize the asymptotics for both the numerator and the denominator of (2.6). For the numerator, we still follow the expansion (A.14). As in step 1, by virtue of Lemma A.3 and relation (A.18), we can verify the applicability of the dominated convergence theorem in the following derivations, but we skip details:
and
We deal with \(I_{3}\) as follows:
Since \(\delta >0\) can be arbitrary small, it follows that
For the denominator of (2.6), we derive
where the second last step is subject to a dominated convergence argument justified by Lemma A.1 and relation (A.19). Substituting (A.25)–(A.26) into (2.6), we obtain (3.7) for \(p>1\).
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Chen, Y., Liu, J. Asymptotic capital allocation based on the higher moment risk measure. Eur. Actuar. J. (2024). https://doi.org/10.1007/s13385-024-00378-4
Received:
Revised:
Accepted:
Published:
DOI: https://doi.org/10.1007/s13385-024-00378-4
Keywords
- Capital allocation
- The higher moment risk measure
- Asymptotic independence/dependence
- Univariate/multivariate regular variation