Abstract
A new characterization for survival functions of multivariate failure-times arising in exogenous shock models with non-negative, continuous, and unbounded shocks is presented. These survival functions are the product of their ordered and individually transformed arguments. The involved transformations may depend on the specific order of the arguments and must fulfill a monotonicity condition. Conversely, every survival function of that form can be constructed using an exogenous shock model with independent and non-homogeneous shocks.
Similar content being viewed by others
Notes
The functional equation of the lack-of-memory property is another starting point for generalizations, see Muliere and Scarsini (1987).
The interpretation \(\lambda _I=0 \,\Leftrightarrow \, {\mathbb {P}}(Z_I=\infty )=1\) requires the marginal-finiteness condition
to make the resulting vector \((\tau _1,\ldots ,\tau _d)\) well defined.
For readability, the necessary conditions on the transformations \(g_i^\pi \) are omitted here and the reader is referred to the full statement in Theorem 1.
This reflects the exchangeability of \({\bar{F}}\), which is assumed here.
If \(\Lambda \) has also stationary increments, it is called a Lévy subordinator.
A Bernstein function is a non-negative, infinitely often differentiable function \(\psi : [0, \infty ) \rightarrow [0, \infty )\) with \((-1)^{n+1}\psi ^{(n)} \ge 0\). Standard literature, see, e.g., Berg et al. (1984), Schilling et al. (2012), states that the class of Bernstein functions is represented as \(\{ x\mapsto a 1_{(0,\infty )} (x) + b x + \int _{0, \infty } (1-\exp \{-x s\}) \nu ({\text {ds}}) : a, b \ge 0, \nu \) is a Lévy-measure}.
These are (0.2995, 1.401, 1), (0.2, 2.4, 2), and (0.0151, 0.994749, 0.01).
For a definition of (regenerative) composition structures and an introduction of the notation which is used hereinafter, the interested reader is referred to Gnedin and Pitman (2005).
The independence of the specific choice of \(\{ \pi _{J} \}_{J \subseteq I_{2}}\) can also be derived without resorting to the probabilistic interpretation by using the assumption that \({\bar{F}}\) has a well-defined representation as in Eq. (6).
For a non-increasing function f, its generalized inverse is defined by \(f^\leftarrow (x) :=\inf \{ x : f(x) \le y\}\) and for a non-decreasing function f, its generalized inverse is defined by \(f^\leftarrow (x) :=\inf \{ y : f(y) \ge x\}\).
If g is a continuous and non-increasing function, then \(g^{\leftarrow } (x) = ( -g )^{\leftarrow } ( - x )\), where the generalized inverse on the l.h.s. is for non-increasing and on the r.h.s. for non-decreasing functions. As \((-g)^{\leftarrow }\) is a right-inverse of \(-g\), see Embrechts and Hofert (2013, p.425 sq., prop. 1 (4)), this implies that \(g^{\leftarrow }\) is a right-inverse of g.
References
Berg C, Christensen JPR, Ressel P (1984) Harmonic analysis on semigroups. Graduate texts in mathematics, vol 100. Springer, New York. https://doi.org/10.1007/978-1-4612-1128-0
Bielecki TR, Cousin A, Crépey S, Herbertsson A (2013) A bottom-up dynamic model of portfolio credit risk: part II: common-shock interpretation, calibration and hedging issues. available at SSRN: https://ssrn.com/abstract=2245130 or https://doi.org/10.2139/ssrn.2245130
Durante F, Quesada-Molina JJ, Úbeda Flores M (2007) On a family of multivariate copulas for aggregation processes. Inf Sci 177(24):5715–5724. https://doi.org/10.1016/j.ins.2007.07.019
Durante F, Kolesárová A, Mesiar R, Sempi C (2008) Semilinear copulas. Fuzzy Sets Syst 159(1):63–76. https://doi.org/10.1016/j.fss.2007.09.001
Elouerkhaoui Y (2007) Pricing and hedging in a dynamic credit model. Int J Theor Appl Finance 10(04):703–731. https://doi.org/10.1142/S0219024907004408
Embrechts P, Hofert M (2013) A note on generalized inverses. Math Methods Oper Res 77(3):423–432
Giesecke K (2003) A simple exponential model for dependent defaults. J Fixed Income 13(3):74–83. https://doi.org/10.3905/jfi.2003.319362
Gnedin A, Pitman J (2005) Regenerative composition structures. Ann Probab 33(2):445–479. https://doi.org/10.1214/009117904000000801
Joe H (1997) Multivariate models and multivariate dependence concepts. CRC Press, Boca Raton
Klein JP, Keiding N, Kamby C (1989) Semiparametric Marshall–Olkin models applied to the occurrence of metastases at multiple sites after breast cancer. Biometrics 45(4):1073–1086
Li H (2009) Orthant tail dependence of multivariate extreme value distributions. J Multivar Anal 100(1):243–256. https://doi.org/10.1016/j.jmva.2008.04.007
Li X, Pellerey F (2011) Generalized Marshall–Olkin distributions and related bivariate aging properties. J Multivar Anal 102(10):1399–1409. https://doi.org/10.1016/j.jmva.2011.05.006
Lin J, Li X (2014) Multivariate generalized Marshall–Olkin distributions and copulas. Methodol Comput Appl Probab 16(1):53–78. https://doi.org/10.1007/s11009-012-9297-4
Lindskog F, McNeil AJ (2003) Common Poisson shock models: applications to insurance and credit risk modelling. Astin Bull 33(2):209–238. https://doi.org/10.1017/S0515036100013441
Mai JF (2010) Extendibility of Marshall–Olkin distributions via Lévy subordinators and an application to portfolio credit risk. Dissertation, Technische Universität München. http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:91-diss-20100628-969547-1-8
Mai JF (2014) Multivariate exponential distributions with latent factor structure and related topics. Habilitation thesis, Technische Universität München
Mai JF, Scherer M (2017) Simulating copulas: stochastic models, sampling algorithms and applications. Series in quantitative finance, vol 6, 2nd edn. World Scientific, Singapore
Mai JF, Schenk S, Scherer M (2016) Exchangeable exogenous shock models. Bernoulli 22(2):1278–1299. https://doi.org/10.3150/14-BEJ693
Marshall AW (1996) Copulas, marginals, and joint distributions. In: Rüschendorf L, Schweizer B, Taylor MD (eds) Distributions with fixed marginals and related topics. Lecture notes, monograph series, vol 28. Institute of Mathematical Statistics, Hayward, pp 213–222. https://doi.org/10.1214/lnms/1215452620
Marshall AW, Olkin I (1967) A multivariate exponential distribution. J Am Stat Assoc 62(317):30–44. https://doi.org/10.2307/2282907
Muliere P, Scarsini M (1987) Characterization of a Marshall–Olkin type class of distributions. Ann Inst Stat Math 39(1):429–441. https://doi.org/10.1007/BF02491480
Schenk S (2016) Exchangeable exogenous shock models. Dissertation, Technical University of Munich, Munich. Available at https://mediatum.ub.tum.de/doc/1278138/1278138.pdf
Schilling RL, Song R, Vondracek Z (2012) Bernstein functions. De Gruyter studies in mathematics, vol 37, 2nd edn. De Gruyter, Berlin
Schweizer B, Sklar A (1983) Probabilistic metric spaces. Elsevier/North-Holland, New York
Sklar A (1959) Fonctions de répartition a n dimensions et leurs marges. Publ Inst Stat Univ Paris 8:229–231
Sun Y, Mendoza-Arriaga R, Linetsky V (2017) Marshall–Olkin distributions, subordinators, efficient simulation, and applications to credit risk. Adv Appl Probab 49(2):481–514
Acknowledgements
We thank the two anonymous reviewers as well as the editor and associate editor who provided helpful comments.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no conflict of interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Proof of Theorem 1
The theorem will be proven in four steps. Particularly, it is shown that \(3. \Rightarrow 4. \Rightarrow 1. \Rightarrow 2. \Rightarrow 3.\) Proofs of used auxiliary results are deferred to “Appendix B”.
Remark 1
Under the assumptions of Theorem 1, particularly the representation of \({\bar{F}}\) in Eq. (6), the expression
is invariant for different permutations with coinciding images of \([i-1]\) and i. If the first statement of the theorem is fulfilled, then \(g_i^\pi \) has the interpretation of a conditional probability, i.e.
Hence, the function \(g_i^\pi \) only depends on \(\pi ([i-1])\) and \(\pi (i)\) and it is justified to work with \({\tilde{g}}^{\pi ([i]), \pi (i)}\).
Proof of\(3. \Rightarrow 4.\) First observe that 4. is a special case of 3., hence \(3. \Rightarrow 4.\) follows directly. \(\square \)
Proof of\(4. \Rightarrow 1.\) Let 4. from Theorem 1 be fulfilled and define for independent random variables \(Z_{I} \sim {\bar{S}}_{I}\), \(\emptyset \not = I \subseteq [d]\), the random vector \(\varvec{\tau }\) by
For \(\varvec{t} \ge 0\) and \(\pi \in {\mathcal {S}}_{d}\) with \(t_{\pi (1)} \ge \cdots \ge t_{\pi (d)}\), using the independence of the shock variables and reordering the factors, it holds that
For \(i \in [d]\) and \(\pi (i) \in I \subseteq \pi (\{ i, \ldots , d\})\), by assumption, the survival function \({\bar{S}}_{I} \equiv {\bar{S}}_{I}^{\pi (i)}\) has a representation as in Eq. (10) with \(m = \pi (i)\) and
Fix \(K \subseteq [d]\) with \(\pi \left( [i] \right) \subseteq K\); then \(i \le |K |= k \le d\) and \(1 \le j \le k - i + 1\). The expression \({\tilde{g}}^{K, \pi (i)}(t_{\pi (i)})\) with an exponent of \((-1)^{j - 1}\) appears \(\left( {\begin{array}{c}k - i\\ j - 1\end{array}}\right) \) times, as there are exactly \(\left( {\begin{array}{c}k - i\\ j - 1\end{array}}\right) \) possible choices for J with \(\pi (i) \in J \subseteq K \backslash \pi ([i-1])\). Hence, the overall exponent of the expression \({\tilde{g}}^{K, \pi (i)}(t_{\pi (i)})\) is
where the latter expression follows with the binomial formula. Finally, it follows that
\(\square \)
In the following, \(I_{1}\), \(I_{2}\), \(\{ \pi _{J} \}_{J \subseteq I_{2}}\), s and t (or a subset of these elements) fulfill the usual conditions if
-
1.
\(s > t \ge 0\),
-
2.
\(I_{1}, I_{2} \subseteq [d]\) with \(I_{1} \cap I_{2} = \emptyset \) and \(I_{2} \not = \emptyset \),
-
3.
for \(J \subseteq I_{2}\) one has
-
(a)
\(\pi _{J} \left( \{ 1, \ldots , |I_{1} |\} \right) = I_{1}\) (if \(I_{1} \not = \emptyset \)),
-
(b)
\(\pi _{J} \left( \{ |I_{1} |+ 1, \ldots , |I_{1} \cup J |\} \right) = J\),
-
(c)
\(\pi _{J} \left( \{ |I_{1} \cup J |+ 1, \ldots , |I_{2} |\} \right) = I_{2} \backslash J\).
-
(a)
If only a specific permutation \(\pi \) is used, it is assumed that it fulfills this property for \(J = I_{2}\).
Proof of\(1. \Rightarrow 2.\) Let 1. in Theorem 1 be fulfilled and let \(I_{1}\), \(I_{2}\), \(\{ \pi _{J} \}_{J \subseteq I_{2}}\), s and t fulfill the usual conditions. First assume that for arbitrary \(\pi \in {\mathcal {S}}_{d}\) and \(i \in [d]\) the functions \(g_{i}^{\pi }\) are strictly positive on \({\mathbb {R}}_{+}\). Then
where it is used that by 1. the diagonal of marginal survival functions of \(\varvec{\tau }_{I_{1}}\) can be represented with every \(\pi \) fulfilling \(\pi (\{1, \ldots , |I_{1}|\}) = I_{1}\). Particularly, it holds that
Subsequently, the numerator of Eq. (17) can be rewritten using the principle of inclusion and exclusion as
where
It follows that
and subsequently that \(G_{I_{1}, I_{2}}^{\{ \pi _{J} \}_{J \subseteq I_{2}}} (s, t)\) is non-negative and does not depend on the specific choice of \(\{ \pi _{J} \}_{J \subseteq I_{2}}\).Footnote 9
Now, by induction over i, the strict positivity, continuity, and non-increasingness of \(g_{i}^{\pi }\) is proven for all \(\pi \in {\mathcal {S}}_{d}\). This implies that \(G_{I_{1}, I_{2}} (s, t)\) is continuous in s and t. For \(i = 1\) and \(\pi \in {\mathcal {S}}_{d}\), the assumptions of Theorem 1 imply that \(g_{1}^{\pi }\) is strictly positive, continuous, and non-increasing. Let the claim be fulfilled for \(j < i\), i.e. \(g_{j}^{\pi }\) is strictly positive, continuous, and non-increasing for \(j \le i -1\) and \(\pi \in {\mathcal {S}}_{d}\).
Right-continuity and left-limits It is well known, see, e.g., Schweizer and Sklar (1983, Chp. 6), that copulae are Lipschitz-continuous with constant one. Hence, by exploiting the copula/survival function decomposition, it holds that
and right-continuity as well as left-limits of \({\bar{F}}\) are inherited from the margins. For \(\pi \in {\mathcal {S}}_{d}\) the survival function \(t \mapsto {\mathbb {P}} \left( \min _{j\le i} \tau _{\pi (j)} > t \right) \) is right-continuous with left-limits and with
right-continuity with left-limits for \(g_{i}^{\pi }\) follows with the induction hypothesis.
Non-increasingness For \(\pi \in {\mathcal {S}}_{d}\) and \(s \ge t \ge 0\) define the vector \(\varvec{u} (s, t)\) by
Then, by monotonicity of the measure \({\mathbb {P}}\), one has
where the induction hypothesis, i.e. \(g_{j}^{\pi }\) is strictly positive for all \(j < i\), is used.
Strict positivity Assume for \(\pi \in {\mathcal {S}}_{d}\) that there exists a finite upper bound \(s^{\star }\) for strict positivity of \(g_{i}^{\pi }\), i.e. \(s^{\star } :=\inf {\{ u > 0 : g_{i}^{\pi } (u) = 0 \}} < \infty \), and as \(g_{i}^{\pi }\) is right-continuous and non-increasing we have that \(g_{i}^{\pi } (s^{\star }) = 0\). For \(t < s^{\star }\), let \(I_{1} =\pi \left( \{ 1 , \ldots , i - 2\} \right) \) and \(I_{2} = \pi (\{ i - 1, i \})\). Furthermore, let \({\tilde{\pi }}\) be the permutation which switches the positions of \(i - 1\) and i in \(\pi \), i.e. \({\tilde{\pi }} = \pi (i - 1, i)\). Assume w.l.o.g. that \(s^{\star } \le u^{\star }\) for \(u^{\star } :=\inf {\{ u > 0 : g_{i}^{{\tilde{\pi }}} (u) = 0 \}} \in \bar{{\mathbb {R}}}_{+}\) (else switch the roles of \(\pi \) and \({\tilde{\pi }}\) and prove the contradiction for \({\tilde{\pi }}\) first). Then, with the induction hypothesis it holds that \(g_{j}^{\pi }, g_{j}^{{\tilde{\pi }}} > 0 \ \forall j < i\) and, for \(\pi _{\emptyset } \in \{ \pi , {\tilde{\pi }} \}\), that
The last expression in Eq. (18) becomes negative if t is sufficiently close to \(s^{\star }\):
-
1.
If \(u^{\star } > s^{\star }\), choose \(\pi _{\emptyset } = \pi \). Then for \(t \nearrow s^{\star }\) Eq. (18) approaches \(-g_{i - 1}^{{\tilde{\pi }}}(s^{\star }) g_{i}^{{\tilde{\pi }}}(s^{\star }-)\).
As \(g_{i - 1}^{{\tilde{\pi }}} (s^{\star }) > 0\) by the induction hypothesis and \(g_{i}^{{\tilde{\pi }}} (t) > 0 \ \forall t < u^{\star }\) with \(s^{\star } < u^{\star }\) by the assumption made above it holds that
$$\begin{aligned} 0 \le -g_{i - 1}^{{\tilde{\pi }}} (s^{\star }) g_{i}^{{\tilde{\pi }}} (s^{\star }-) < 0 . \end{aligned}$$ -
2.
If \(s^{\star } = u^{\star }\) and \(g_{i}^{\pi _{\emptyset }} (s^{\star }-) > g_{i}^{\pi _{\emptyset }} (s^{\star }) = 0\) for at least one \(\pi _{\emptyset } \in \{ \pi , {\tilde{\pi }} \}\), then for \(t \nearrow s^{\star }\) Eq. (18) approaches \(-g_{i - 1}^{\pi _{\emptyset }}(s^{\star }) g_{i}^{\pi _{\emptyset }}(s^{\star }-)\).
As \(g_{i - 1}^{\pi _{\emptyset }} (s^{\star }) > 0\) by the induction hypothesis and \(g_{i}^{\pi _{\emptyset }} (s^{\star }-) > 0\) by the assumption made above it holds that
$$\begin{aligned} 0 \le -g_{i - 1}^{\pi _{\emptyset }} (s^{\star }) g_{i}^{\pi _{\emptyset }} (s^{\star }-) < 0 . \end{aligned}$$ -
3.
Otherwise, as \(g_{j}^{\pi _{\emptyset }}\) for \(j \in \{ i - 1, i\}\) have left-limits by the induction hypothesis, for every sequence \(t_{k} \nearrow s^{\star }\) with \(t_k \ne s^\star \), non-negative sequences \(\{ a_{j, k}^{\pi _{\emptyset }} \}_{k \in {\mathbb {N}}}\) with \(a_{j, k}^{\pi _{\emptyset }} (s^{\star } - t_{k}) \rightarrow 0\) for \(k \rightarrow \infty \) can be found s.t.
$$\begin{aligned} g_{j}^{\pi _{\emptyset }} (t_{k}) = g_{j}^{\pi _{\emptyset }} (s^{\star }-) + a_{j, k}^{\pi _{\emptyset }} (s^{\star } - t_{k}) ,\quad \ j\in \{i-1, i\} ,\quad \ k \in {\mathbb {N}} . \end{aligned}$$By the assumption on \(s^\star \), it holds that \(a_{i, k}^{\pi _\emptyset } > 0\) for all \(k\in {\mathbb {N}}\) and \(\pi _\emptyset \in \{\pi , {\tilde{\pi }}\}\). If \(s^{\star } = u^{\star }\) and \(g_{i}^{\pi _{\emptyset }} (s^{\star }-) = g_{i}^{\pi _{\emptyset }} (s^{\star }) = 0\) for all \(\pi _{\emptyset } \in \{ \pi , {\tilde{\pi }} \}\), it follows from Eq. (18) and (left-)continuity of \(g_{i - 1}^{\pi _{\emptyset }}\) that
$$\begin{aligned} 0 \le {\left\{ \begin{array}{ll} a_{i - 1, k}^{\pi } a_{i, k}^{\pi } (s^{\star } - t_{k})^2 - g_{i - 1}^{{\tilde{\pi }}} (s^{\star }) a_{i, k}^{{\tilde{\pi }}} (s^{\star } - t_{k}) , {}&{} \pi _{\emptyset } = \pi \\ a_{i - 1, k}^{{\tilde{\pi }}} a_{i, k}^{{\tilde{\pi }}} (s^{\star } - t_{k})^2 - g_{i - 1}^{\pi } (s^{\star }) a_{i, k}^{\pi } (s^{\star } - t_{k}) , &{} \pi _{\emptyset } = {\tilde{\pi }} \end{array}\right. } \end{aligned}$$or equivalently
$$\begin{aligned} 0 \le {\left\{ \begin{array}{ll} a_{i - 1, k}^{\pi } (s^{\star } - t_{k}) \frac{a_{i, k}^{\pi }}{a_{i, k}^{{\tilde{\pi }}}} - g_{i - 1}^{{\tilde{\pi }}} (s^{\star }) , {}&{} \pi _{\emptyset } = \pi \\ a_{i - 1, k}^{{\tilde{\pi }}} (s^{\star } - t_{k}) \frac{a_{i, k}^{{\tilde{\pi }}}}{a_{i, k}^{\pi }} - g_{i - 1}^{\pi } (s^{\star }) , &{} \pi _{\emptyset } = {\tilde{\pi }} . \end{array}\right. } \end{aligned}$$Now choose k sufficiently large and \(\pi _{\emptyset }\) s.t. the fraction appearing in the upper equation is smaller or equal to 1, then
$$\begin{aligned} \begin{aligned} 0 {}&\le {\left\{ \begin{array}{ll} a_{i - 1, k}^{\pi } (s^{\star } - t_{k}) - g_{i - 1}^{{\tilde{\pi }}} (s^{\star }) , {}&{} a_{i, k}^\pi \le a_{i, k}^{{\tilde{\pi }}} \\ a_{i - 1, k}^{{\tilde{\pi }}} (s^{\star } - t_{k}) - g_{i - 1}^{\pi } (s^{\star }) , &{} a_{i, k}^\pi > a_{i, k}^{{\tilde{\pi }}} \end{array}\right. } \\&< 0 , \end{aligned} \end{aligned}$$
where it is used that the respective first summand converges for \(k \rightarrow \infty \) to 0 and the last summand is negative. Hence, a contradiction is found for each case and therefore \(g_{i}^{\pi }(t) > 0 \ \forall t \in {\mathbb {R}}_{+}\).
Left-continuity Let \(I_{1}\) and \(I_{2}\) as well as \(\pi \), \({\tilde{\pi }}\), and \(\pi _{\emptyset }\) be as above. Then, for all \(s > t \ge 0\) the function
has left-limits in t. Assume that there exists \(s^{\dagger } \in {\mathbb {R}}_{+}^{\times }\) with \(g_{i}^{\pi }(s^{\dagger }-) > g_{i}^{\pi }(s^{\dagger })\), then
where it is used in \((\star )\), that the first and third summand cancel out, when using that \(g_{i - 1}^{{\tilde{\pi }}}\) is continuous under the induction hypothesis. This is a contradiction—hence \(g_{i}^{\pi }\) is left-continuous. \(\square \)
Remark 2
The induction in the second part of the proof can be performed on the basis of statement 2. (instead of 1.) from Theorem 1 if the parts on right-continuity with left-limits and non-increasingness are replaced by the following lemma (as they rely on the survival function assumption of 1.). In particular, 2. implies \(g_{i}^{\pi } \in \bar{{\mathcal {G}}}\) for all \(i \in [d], \pi \in {\mathcal {S}}_{d}\).
Lemma 1
Let 2. from Theorem 1 be fulfilled and \(g_{j}^{\pi }\) be right-continuous with left-limits, non-increasing, and strictly positive for all \(j \le i - 1\) and \(\pi \in {\mathcal {S}}_{d}\). Then \(g_{i}^{\pi }\) is right-continuous with left-limits and non-increasing for all \(\pi \in {\mathcal {S}}_{d}\).
Lemma 2
Assume that statement 2. of Theorem 1 is fulfilled and let \(I_{1}\) and \(I_{2}\) fulfill the usual conditions. Then for each \(m \in I_{2}\), \({\bar{S}}_{I_{1}, I_{2}}^{m}\) is an \({\mathbb {R}}_{+}\)-valued, positive, and continuous function on \({\mathbb {R}}_{+}\). Furthermore, \({\bar{S}}_{I_{1}, I_{2}}^{m}\) does not depend on \(m\in I_2\), i.e.
Lemma 3
Let \(I_{1}\) and \(I_{2}\) fulfill the usual conditions and assume that \({\bar{S}}_{I_{1} \cup I_{2} \backslash J, J}^{m_{1}} = {\bar{S}}_{I_{1} \cup (I_{2} \backslash J), J}^{m_{2}} \in \bar{{\mathcal {G}}}\) for all \(\emptyset \not = J \subseteq I_{2}\) and \(m_{1}, m_{2} \in J\). Then for \(s > t \ge 0\)
where
with independent random shocks \(\check{Z}_{J} \sim {\bar{S}}_{I_{1} \cup I_{2} \backslash J, J}\) for \(\emptyset \not = J \subseteq I_{2}\).
The essence of the previous Lemma is the following: Let \(I_1\) and \(I_2\) fulfill the usual conditions, \(Z_I\sim {\bar{S}}_I \in \bar{{\mathcal {G}}}\), \(\emptyset \ne I\subseteq [d]\), \(\varvec{\tau }\) be defined as in Eq. (1), and \(\check{\varvec{\tau }} \in {\mathbb {R}}_+^{|I_2|}\) be defined by
Then
Lemma 4
Let \(I_{1}\) and \(I_{2}\) fulfill the usual conditions. Then, for a specific family of permutions, \(\{ \pi _{J} \}_{J \subseteq I_{2}}\), the function \(G_{I_{1}, I_{2}}^{\{ \pi _{J} \}_{J \subseteq I_{2}}}\) depends on \(g_{i}^{\pi _{J}}, |I_{1} |+ 1 \le i \le |I_{1} \cup I_{2} |\), \(J \subseteq I_{2}\). Therefore, write
Assume that \(g_{i}^{\pi _{J}}, |I_{1} |+ 1 \le i \le |I_{1} \cup I_{2} |, J \subseteq I_{2}\) are positive. Then it holds for all \(s \ge t \ge 0\) that
for an arbitrary function \({\hat{g}}_{|I_{1} |+ 1}^{\pi _{\emptyset }}\) which is positive on \({\mathbb {R}}_{+}\), where
which are by definition positive functions on \({\mathbb {R}}_{+}\).
Lemma 5
For \(k \in {\mathbb {N}}_{0}\), \(j \ge 2\), let the functions \({\bar{F}}_{1, k}, \ldots , {\bar{F}}_{j, k}: [0, \infty ) \rightarrow (0, 1]\) as well as \({\bar{F}}_{1, k + 1}, \ldots , {\bar{F}}_{j - 1, k + 1}: [0, \infty ) \rightarrow (0, 1]\) be non-increasing with \({\bar{F}}_{l, k} = \frac{{\bar{F}}_{l - 1, k}}{{\bar{F}}_{l - 1, k + 1}}\) for \(l \in \{ 2 , \ldots , j\}\). Then it holds that for \(s\ge t \ge 0\)
Proof of\(2. \Rightarrow 3.\) Let statement 2. in Theorem 1 be fulfilled, then due to Remark 2, Lemmas 1 and 2:
-
For \(i = 1, \ldots , d\) and \(\pi \in {\mathcal {S}}_{d}\), it holds that \(g_{i}^{\pi } \in \bar{{\mathcal {G}}}\).
-
For \(I_{1}\) and \(I_{2}\) fulfilling the usual conditions and \(m \in I_{2}\), the function \({\bar{S}}_{I_{1}, I_{2}}^{m}\) is well-defined as well as positive and continuous. Moreover, it does not depend on the specific \(m \in I_{2}\) chosen, hence write \({\bar{S}}_{I_{1}, I_{2}}\).
It is left to prove that \({\bar{S}}_{I_{1}, I_{2}}\) is non-increasing for all \(I_{1}, I_{2}\) fulfilling the usual conditions.
The claim is proven by induction over \(|I_{2} |\). For \(I_{2} = \{ m \}\), let \(I_{1}\) and \(I_{2}\) fulfill the usual conditions, then \({\bar{S}}_{I_{1}, I_{2}} = {\tilde{g}}^{I_{1} \cup I_{2}, m} \in \bar{{\mathcal {G}}}\). Now let \(p > 1\) and assume that for all \(I_{1}\) and \(I_{2}\) fulfilling the usual conditions with \(|I_{2} |< p\) it holds that \({\bar{S}}_{I_{1}, I_{2}} \in \bar{{\mathcal {G}}}\). Let \(I_{1}, I_{2}\), \(\{ \pi _{J} \}_{J \subseteq I_{2}}\), s, and t fulfill the usual conditions and \(|I_{2} |= p\) and define the function \({\hat{g}}_{|I_{1} |+ 1}^{\pi _{\emptyset }} :=g_{|I_{1} |+ 1}^{\pi _{\emptyset }} / {\bar{S}}_{I_{1}, I_{2}} \), which is continuous and positive. With Lemma 4 it follows that
where \({\hat{g}}_{|I_{1} |+ 1}^{\pi _{J}} :=g_{|I_{1} |+ 1}^{\pi _{J}} / {\bar{S}}_{I_{1}, I_{2}}\) for \(J \subseteq I_{2}\).
In light of Lemma 3, it makes sense to derive an exogenous shock model from
Hence one has to check, that for \(\emptyset \not = J \subseteq I_{2}\) if \(\bar{{\hat{S}}}_{I_{1} \cup I_{2} \backslash J, J} \in \bar{{\mathcal {G}}}\). Note that
As \({\bar{S}}_{I_{1} \cup I_{2} \backslash J, J} \in \bar{{\mathcal {G}}}\) by the induction step for \(\emptyset \not = J \subsetneq I_{2}\) and \(\bar{{\hat{S}}}_{I_{1}, I_{2}} \equiv 1 \in \bar{{\mathcal {G}}}\), Lemma 3 can be used. Write for \(s > t \ge 0\)
where
with independent \({\hat{Z}}_{I} \sim {\hat{S}}_{I_{1} \cup I_{2} \backslash I, I}\) for \(\emptyset \not = I \subseteq I_{2}\). Let \(s > t \ge 0\) and define
Since \({\hat{Z}}_{I_{2}} = \infty \), there are at least two different sets \(\emptyset \not = I, J \subsetneq I_{2}\) for which the respective shocks \({\hat{Z}}_{I}\), \({\hat{Z}}_{J}\) are minimal for one of their components. Moreover, this implies
From the sub-additivity of the probability measure \({\mathbb {P}}\), it follows that
where we used that for \(\emptyset \not = I \subsetneq I_{2}\)
Note that for \(\emptyset \not = J \subseteq I \subsetneq I_{2}\) and \(m, n \in J\), \(m \not = n\)
Writing \(b :=\left( {\begin{array}{c}2^{|I_{2} |} - 2\\ 2\end{array}}\right) \) and using Lemma 5 for ascending sequences \(\emptyset \not = J_{1} \subsetneq \cdots \subsetneq J_{|I |} = I \subseteq I_{2}\) with \(|J_I|= |I |\) as well as
-
1.
\({\bar{F}}_{|J_{l} |, |I_{1} \cup (I_{2} \backslash I) |} \equiv {\bar{S}}_{I_{1} \cup (I_{2} \backslash I), J_{l}}\) for \(l \in \left[ |I |\right] \) and
-
2.
\({\bar{F}}_{|J_{l} |, |I_{1} \cup (I_{2} \backslash I) \cup (J_{l + 1} \backslash J_{l}) |} \equiv {\bar{S}}_{I_{1} \cup (I_{2} \backslash I) \cup (J_{l + 1} \backslash J_{l}), J_{l}}\) for \(l \in \left[ |I |- 1\right] \)
it follows that
Now let \(\emptyset \not = I \subsetneq I_{2}\), \(k = |I_{1} \cup (I_{2} \backslash I)|\), \(J_{1} = \{ m\}\) and \(\pi \in {\mathcal {S}}_{d}\) be a permutation fulfilling \(\pi \left( \{ 1, \ldots , k \} \right) = I_{1} \cup (I_{2} \backslash I)\), \(\pi (k + 1) = m\). Denote with \({\tilde{\pi }}\) the permutation, which switches the positions of m and \(\pi (k)\), i.e. \({\tilde{\pi }} = \pi (k, k + 1)\). Then
which is equivalent to
This yields inductively the following inequality
Subsequently,
with
where \(\Pi _{I_{1}, I_{2}, I}\) is the set of permutations fulfilling the conditions stated above and
Let \(s_{0} \ge s > t \ge t_{0} \ge 0\); the non-increasingness of the involved survival functions \({\bar{S}}_{I_{1} \cup (I_{2} \backslash I) \cup (J_{l + 1} \backslash J_{l}), J_{l}} (s)\), \({\tilde{g}}^{\pi (\{ [l] \}) \cup \{ m \}, m} (t)\), and \({\tilde{g}}^{\pi (\{ [l] \}), \pi (l)} (s)\) implies
Define for \(s \ge t \ge 0\)
As \({\tilde{g}}^{\{ m\}, m}, m \in I_{2}\) are non-negative and non-increasing and \(q_{I_{2}} (s, t) \ge 0\) all summands are non-negative and
Hence
Now, the proof proceeds analogously as for copulas in the exchangeable case (see Mai et al. 2016, 1296 sq.) or bivariate exchangeable case (see Durante et al. 2008, 67).
The function \({\bar{S}}_{I_{1}, I_{2}}\) splits in positive and negative powers in the product terms and
where the monotonicity of \({\tilde{g}}^{I, m}\) is used in \((\star )\). Assume that \({\bar{S}}_{I_{1}, I_{2}}\) is not non-increasing, i.e. there exists \(s_{0} > t_{0} \ge 0\) s.t. \({\bar{S}}_{I_{1}, I_{2}} (s_{0}) > {\bar{S}}_{I_{1}, I_{2}} (t_{0})\).
Case\(q_{I_{1}} (s_{0}, t_{0}) = 0\): From Eq. (21) we get
which is a contradiction.
Case\(q_{I_{1}} (s_{0}, t_{0}) > 0\): Let
then we can write
For all \(k \ge 1\), one can find \(s_{k}, t_{k} \in [t_{0}, s_{0}]\) with \(s_{k} > t_{k}\) and
as well as
This can be seen by setting \(t^{(0, k)} :=t_{0}\), \(t^{(k, k)} :=s_{0}\), and
where \({\leftarrow }\) denotes the generalized inverse for non-increasing functionsFootnote 10 and for \(k \in \{ 0, \ldots , k\}\)
As \({\tilde{g}}^{\{ m \}, m}\) are continuous and non-decreasing the generalized inverse is a right-inverseFootnote 11 and
Assume that for all \(j \in \{ 1, \ldots , k\}\) the following inequality holds
Then,
which is a contradiction. Hence, with \(t_{k} = t^{(j - 1, k)}\), \(s_{k} = t^{(j, k)}\) for some \(j \in \{ 1, \ldots , k\}\), Eq. (22) is fulfilled and \(s_k > t_k\).
Combining Eq. (21) with these results gives for feasible \(t_{k}\), \(s_{k}\) (chosen as above)
In particular, if the latter inequality is multiplied by k and the limit \(k \rightarrow \infty \) is taken, then
which leads to a contradiction. \(\square \)
Proofs of supporting lemmas
Proof of Lemma 1
Let \(I_{1}\), \(I_{2}\), and \(\pi \) fulfill the usual conditions with \(|I_{2} |= 2\) and \(|I_{1} |= i - 2\) and define \({\tilde{\pi }} = \pi (i - 1, i)\).
Right-continuity Let \(s + h> s > t \ge 0\). As \(G_{I_{1}, I_{2}} (s, t)\) is right-continuous in s it holds that
where it is used that under the induction hypothesis all but two terms cancel out.
Left-limits Let \(s> s - h > t \ge 0\). As \(G_{I_{1}, I_{2}} (s, t)\) and \(g_{i - 1}^{\pi _{\emptyset }} (s)\), \(\pi _{\emptyset } \in \{ \pi , {\tilde{\pi }}\}\) have left-limits in s and \(g_{i - 1}^{\pi }\) is positive by induction hypothesis it follows that \(g_{i}^{\pi }\) has left-limits:
Non-increasingness Now, let \(I_{1}\), \(I_{2}\), and \(\pi \) fulfill the usual conditions with \(I_{2} = \{ \pi (i) \}\) and \(I_{1} = \pi ([i - 1])\). As \(G_{I_{1}, I_{2}}\) is non-negative, it holds for all \(s > t \ge 0\) that
\(\square \)
Proof of Lemma 2
For \(\pi \in {\mathcal {S}}_{d}\), due to Remark 2 and Lemma 1, it follows that the functions \(g_{i}^{\pi }\), \(i = 1, \ldots , d\) are positive, continuous functions on \({\mathbb {R}}_{+}\). Hence \({\bar{S}}_{I_{1}, I_{2}}^{m}\) is an \({\mathbb {R}}_{+}\)-valued, positive, and continuous function for every \(I_{1}, I_{2}\) fulfilling the usual conditions with \(m \in I_{2}\).
In the following, it is proven, by induction over \(|I_{2} |\), that Eq. (19) holds and furthermore, that for all \(I_{1}\) and \(I_{2}\) fulfilling the usual conditions
for all \({\tilde{\pi }}, {\hat{\pi }} \in {\mathcal {S}}_{d}\) fulfilling \(\pi ([|I_{1} |]) = I_{1}\) and \(\pi ([|I_{1} \cup I_{2} |] \backslash [|I_{1} |]) = I_{2}\) for \(\pi \in \{ {\tilde{\pi }}, {\hat{\pi }} \}\). For \(|I_{2} |= 1\) both claims are naturally fulfilled. Let both claims be fulfilled for \(|I_{2} |< p\) and let \(I_{1}\), \(I_{2}\) as well as \(\pi \) fulfill the usual conditions with \(|I_{2} |= p\), \(m \in I_{2}\) as well as \(\pi (|I_{1} |+ 1) = m\), then for \(t \ge 0\)
where the factors in \((\star )\) are regrouped in a similar sense as for the alternative representation for the GMO survival function.
Now for \(i \in [d]\) fix \(\pi (\{ 1, \ldots , |I_{1}|+ i\}) \subseteq K \subseteq I_{1} \cup I_{2}\) and define \(k = |K |\) as well as \(1 \le l \le k - |I_{1} |- i + 1\). The expression \({\tilde{g}}_{k}^{K, \pi (|I_{1}|+ i)} (t)\) with exponent \((-1)^{l - 1}\) appears \(\left( {\begin{array}{c}k - |I_{1} |- i\\ l - 1\end{array}}\right) \) times and the overall exponent for \({\tilde{g}}_{k}^{K, \pi (|I_{1} |+ i)}\) is
Hence, as it holds for \(k = |I_{1} |+ i\) that \(K = \pi (\{ 1, \ldots , |I_{1} |+ i\})\) and
or equivalently,
By induction, the factors of the denominator of the r.h.s. in Eq. (24), \({\bar{S}}_{I_{1} \cup (I_{2} \backslash J), J}^{\pi (\min _{j \in J}{\pi ^{-1} (j)})}\), are independent of \(\pi (\min _{j \in J}{ \pi ^{-1} (j)})\) and subsequently also of m. Moreover, for arbitrary \(I_{1}\), \(I_{2}\) and \(\{ \pi _{J} \}_{J \subseteq I_{2}}\) fulfilling the usual conditions and \(s \ge 0\)
By induction and assumption, the r.h.s. does not depend on the specific family of permutations, \(\{ \pi _{J} \}_{J \subseteq I_{2}}\), chosen, therefore Eq. (23) holds for \(|I_{2} |= p\). In conclusion, the nominator in Eq. (24) does not depend on the specific \(\pi \), and subsequently m, chosen and Eq. (19) holds for \(|I_{2} |= p\). \(\square \)
Proof of Lemma 3
As in the proof of 4. to 1. one can derive analogously for \(\varvec{t} \ge 0\) and \(\pi \in {\mathcal {S}}_{d}\) with \(t_{\pi (1)} \ge \cdots \ge t_{\pi (d)}\) as well as \(\pi \left( \{ 1, \ldots , |I_{1} |\} \right) = I_{1}\) and \(\pi \left( \{ |I_{1} |+ 1, \ldots , |I_{1} \cup I_{2} |\} \right) = I_{2}\) that
where for \(\check{I}_{2} = \{ 1, \ldots , |I_{2} |\}\), \({\check{\pi }} \in {\mathcal {S}}_{|I_{2} |}\) is defined by
and \(\check{g}_{j}^{{\check{\pi }}} :=g_{|I_{1} |+ j}^{\pi }\) as well as \(\check{t}_{{\check{\pi }} (j)} :=t_{\pi (|I_{1} |+ j)}\). Then, it holds for all \(0 \le t < s\) that
where \(\check{G}_{\emptyset , \check{I}_{2}}\) corresponds to Eq. (8) w.r.t. \(\{ \check{g}_{j}^{{\check{\pi }}} \}_{j \in \check{I}_{2}, {\check{\pi }} \in {\mathcal {S}}_{|I_{2}|}}\). \(\square \)
Proof of Lemma 4
Every summand corresponding to a non-empty interval \(\emptyset \not = J \subseteq I_{2}\) contains a term \(g_{|I_{1} |+ 1}^{\pi _{J}} (s)\). Therefore the result follows by multiplying \(G_{I_{1}, I_{2}}\) with \(\frac{g_{|I_{1} |+ 1}^{\pi _{\emptyset }} (s)}{{\hat{g}}_{|I_{1} |+ 1}^{\pi _{\emptyset }} (s)}\) and its reciprocal, whereas the first summand in Eq. (20) is a correction term for the summand belonging to \(J = \emptyset \). \(\square \)
Proof of Lemma 5
This is a direct corollary of Mai et al. (2016, lem. B.2 on p. 1295). \(\square \)
Rights and permissions
About this article
Cite this article
Scherer, M., Sloot, H. Exogenous shock models: analytical characterization and probabilistic construction. Metrika 82, 931–959 (2019). https://doi.org/10.1007/s00184-019-00715-8
Received:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00184-019-00715-8