UNIFORM ASYMPTOTICS FOR DISCOUNTED AGGREGATE CLAIMS IN DEPENDENT RISK MODELS

Transcription

1 Applied Probability Trust 2 October 2013 UNIFORM ASYMPTOTICS FOR DISCOUNTED AGGREGATE CLAIMS IN DEPENDENT RISK MODELS YANG YANG, Nanjing Audit University, and Southeast University KAIYONG WANG, Southeast University DIMITRIOS G. KONSTANTINIDES, University of the Aegean Abstract In this paper, we consider some non-standard renewal risk models with some dependent claim sizes and stochastic return, where an insurance company is allowed to invest her/his wealth in financial assets, and the price process of the investment portfolio is described as a geometric Lévy process. When the claim-size distribution belongs to some classes of heavy-tailed distributions and a constraint is imposed on the Lévy process in terms of its Laplace exponent, we obtain some asymptotic formulas for the tail probability of discounted aggregate claims and ruin probabilities holding uniformly for some finite or infinite time horizons. Keywords: Discounted aggregate claims; dependence; Lévy process; consistently varying tail; dominatedly varying tail; long tail; uniformity 2010 Mathematics Subject Classification: Primary 91B30 Secondary 60G51; 60K05 1. Introduction We consider a non-standard renewal risk model in which successive claim sizes {X n, n 1} form a sequence of identically distributed but not necessarily independent Postal address: School of Mathematics and Statistics, Nanjing Audit University, Nanjing , China. address: yyangmath@gmail.com Postal address: Department of Mathematics, Southeast University, Nanjing , China. Postal address: Department of Mathematics, University of the Aegean, Karlovassi, GR Samos, Greece. address: konstant@aegean.gr 1

2 2 Y. YANG, K. WANG AND D.G. KONSTANTINIDES nonnegative random variables r.v.s with common distribution F and the inter-arrival times {θ n, n 1} form another sequence of independent and identically distributed i.i.d. nonnegative r.v.s, which are independent of {X n, n 1}. We suppose that the claim arrival times τ n = n θ k, n 1, constitute a renewal counting process Nt = sup{n 0 : τ n t}, t 0, which represents the number of claims up to time t, with a finite mean function λt = E[Nt] as t. Suppose that the insurer is allowed to make risk-free and risky investments. The price process of the investment portfolio is described as a geometric Lévy process {e Rt, t 0} with {R t, t 0} being a Lévy process, which starts from zero and has independent and stationary increments. This assumption on price processes is widely used in mathematical finance. The reader is referred to [18], [19], [20], [11], [4], [26], [27], [24] [15] among others. As usual, we assume that {X n, n 1}, {θ n, n 1} and {R t, t 0} are mutually independent. The discounted aggregate claims up to time t 0 can be expressed as Dt = X k e Rτ k 1{τk t}, 1.1 where 1 A denotes the indicator function of an event A. Then, for any t 0, the discounted value of the surplus process with stochastic return on investments of an insurance company is described as Ut = x + cs e Rs ds Dt, 1.2 where x 0 is the initial risk reserve of the insurance company, and ct denotes the density function of premium income at time t. Throughout the paper, we assume that the premium density function ct is bounded, i.e. 0 ct M for some constant M > 0 and all t 0. In this way, in the above renewal risk model, for any t 0, the finite-time ruin probability can be defined as ψx, t = P inf Us < 0 U0 = x, s [0, t] and the infinite-time ruin probability as ψx, = P inf Us < 0 U0 = x. s 0

3 Uniform asymptotics for discounted aggregate claims 3 In the present paper, we shall investigate the asymptotics for the tail probability of discounted aggregate claims or ruin probabilities, holding uniformly for all t, such that λt is positive. For this purpose, as in [22], define Λ = {t : 0 < λt } = {t : Pθ 1 t > 0} with t = inf{t : λt > 0} = inf{t : Pθ 1 t > 0}. Clearly, [t, ], if Pθ 1 = t > 0, Λ = t, ], if Pθ 1 = t = 0. For the above risk model with a constant premium rate c > 0 i.e. ct = c for all t > 0 and a constant interest force δ > 0 e.g. R t = δ t for all t > 0, if the claim sizes {X n, n 1} and the inter-arrival times {θ n, n 1} are, respectively, i.i.d. r.v.s, some earlier works on ruin probabilities can be found in [1], [12], [10], [13], [21] among others; Tang [22] and Hao and Tang [8] derived some uniform results in Λ. If the claim sizes {X n, n 1} follow a certain dependence structure, ruin probabilities have also been investigated by many researchers. Chen and Ng [5] considered a dependent risk model, where the claim sizes are pairwise negatively quadrant dependent NQD, see the definition in Section 2.1 with common distribution F, belonging to the class of extended regularly varying distributions. They obtained the following asymptotics for the infinite-time ruin probability: ψx, F x e δu λdu. Further, assuming some restrictive dependence structure and heavy-tailed claim sizes, Yang and Wang [25] obtained the relation ψx, t holding for each fixed t Λ or uniformly for all t Λ. F x e δu λdu, 1.3 Two recent interesting papers Tang et al. [24] and Li [15] investigated the above renewal risk model with stochastic return, allowing the price process of the investment portfolio being a geometric Lévy process under independence and dependence structures, respectively. We remark that the paper [24] investigated the independent renewal risk model, where the claim sizes {X n, n 1} and the inter-arrival times {θ n, n 1} are two sequences of i.i.d. nonnegative r.v.s, respectively, and they are mutually independent; whereas the recent paper [15] considered a more general

4 4 Y. YANG, K. WANG AND D.G. KONSTANTINIDES time-dependent renewal risk model, where {X n, θ n, n 1} are assumed to be i.i.d. random vectors but a certain dependence exists between X n and θ n for each fixed n. Motivated by these two papers, the main goal of this paper is to establish some asymptotic formulas of the tail probability of discounted aggregate claims and ruin probabilities, holding uniformly for some finite or infinite time horizons, under the conditions that the claim sizes {X n, n 1} are dependent and have common distribution belonging to the class of long and dominatedly-varying tailed distributions or to the class of consistently-varying tailed distributions. The rest of the paper is organized as follows. Section 2 presents the four main results on the uniform asymptotics for the tail probability of discounted aggregate claims and ruin probabilities after introducing some preliminaries; Sections 3 and 4 prove these results, respectively, after preparing a series of lemmas. 2. Preliminaries and main results Hereafter, all limit relationships hold for x tending to, unless stated otherwise. For two positive functions ax and bx, we write ax bx if lim ax/bx = 1; write ax bx or bx ax if lim sup ax/bx 1; ax = obx if lim ax/bx = 0; ax = Obx if lim sup ax/bx < ; and ax bx if ax = Obx and bx = Oax. Furthermore, for two positive bivariate functions ax, t and bx, t, we write ax, t bx, t uniformly for all t in a nonempty set A, if lim sup ax, t x bx, t 1 = 0, t A and write ax, t bx, t or bx, t ax, t uniformly for all t A, if 2.1. Dependence structures ax, t lim sup sup x t A bx, t 1. Since what we are interested in is some actual dependent risk models, we start this section by introducing some dependence structures. A sequence of r.v.s {ξ n, n 1} is said to be pairwise negatively quadrant dependent NQD, if, for any i j 1 and real x, y, Pξ i > x, ξ j > y Pξ i > xpξ j > y.

5 Uniform asymptotics for discounted aggregate claims 5 The pairwise NQD structure was introduced by Lehmann [14], and is weaker and more verifiable than the commonly used notions of the upper/lower negative dependence see [3], and the negative association see [9]. A more general dependence structure, namely upper tail asymptotic independence UTAI structure, was proposed by Maulik and Resnick [16]. A sequence of r.v.s {ξ n, n 1} is said to be UTAI, if Pξ n > x > 0 for all x R, n 1, and for any i j 1, lim Pξ i > x ξ j > y = 0. min{x, y} Clearly, if a sequence of r.v.s is pairwise NQD, then it is also UTAI Some classes of heavy-tailed distributions We shall restrict the claim-size distribution F to some class of heavy-tailed distributions, whose moment generating functions do not exist. An important class of heavytailed distributions is D, which consists of all distributions F = 1 F with dominated variation. A distribution F on R belongs to the class D, if lim sup F x y/f x < for any 0 < y < 1. A slightly smaller class is C of consistently varying distributions. A distribution F on R belongs to the class C, if lim y 1 lim sup x F x y/f x = 1. Closely related is a wider class L of long-tailed distributions. A distribution F on R belongs to the class L, if lim F x + y/f x = 1 for any y R. There are some other heavy-tailed subclasses, the class ERV α, β of distributions with extended regularly varying tails, and the class R α of distributions with regularly varying tails. A distribution F on R belongs to the class ERV α, β, if there are some constants 0 < α β < such that y β lim inf F x y/f x lim sup F x y/f x y α for any y 1. If α = β, F belongs to the class R α. For a distribution F on [0,, denote its upper Matuszewska index by J + F = lim log F y y log y with F y := lim inf x F x y F x for y > 1. The presented definitions yield, that the following assertions are equivalent for details, see [2]: i F D, ii F y > 0 for some y > 1, iii J + F <. The lemma below is from Proposition of [2].

6 6 Y. YANG, K. WANG AND D.G. KONSTANTINIDES Lemma 2.1. For a distribution F D on [0, and any p > J + F, there are positive constants C 1 and D 1, such that the inequality F y y p F x C 1, x holds for all x y D 1. The next lemma follows immediately from Lemma 2.1 see Lemma 3.5 of [23]. Lemma 2.2. For a distribution F D on [0, and for any p > J + F, the asymptotics x p = of x holds Main results Suppose that the Lévy process {R t, t 0} in 1.1 is right continuous with left limit. Let E[R 1 ] > 0, so that R t drifts to almost surely as t. The Laplace exponent for the Lévy process {R t, t 0} is defined as φz = log E[e z R1 ], z R. If φz is finite, then it holds for any t 0 that see, e.g. Proposition 3.14 of [7]. E[e z Rt ] = e t φz <, To simplify the discussion, we assume throughout the paper that t = 0. For any T Λ and ɛ Λ, set Λ T = [0, T ] and Λ ɛ = [ɛ, ]. Note that ɛ can be chosen to be 0 if Pθ 1 = 0 > 0. We are now ready to state the main results of this paper. The first two results below consider the UTAI claim sizes, requiring that the Lévy process {R t, t 0} is almost surely nonnegative, which means that the insurer invests her/his wealth only into a risky-free market. We obtain two uniform asymptotic formulas for the tail probability of discounted aggregate claims and ruin probabilities, when the claim sizes have both long and dominatedly varying tails, or consistently varying tails, respectively. Theorem 2.1. In the non-standard renewal risk model, described in Section 1, let assume that the claim sizes {X n, n 1} are UTAI nonnegative r.v.s with common distribution F L D. If R t 0 a.s. for any t 0, then, for any fixed T > 0, PDt > x PX 1 e Rs > x λds, 2.1

7 Uniform asymptotics for discounted aggregate claims 7 holds uniformly for all t Λ T. Corollary 2.1. Under the conditions of Theorem 2.1, if F C, then, for any fixed T > 0, ψx, t holds uniformly for all t Λ T. PX 1 e Rs > x λds, 2.2 Our next two results restrict the claim sizes to be pairwise NQD r.v.s with extended regularly varying tails, but allow the insurer to make risk-free and risky investments. This means the Lévy process {R t, t 0} can be real-valued. Theorem 2.2. In the non-standard renewal risk model described in Section 1, let assume that the claim sizes {X n, n 1} are pairwise NQD nonnegative r.v.s with common distribution F ERV α, β, 0 < α β <. If φr < 0 for some r > β, then, 2.1 holds uniformly for all t Λ ɛ. Corollary 2.2. Under the conditions of Theorem 2.2, 2.2 holds uniformly for all t Λ ɛ. 3. Proof of Theorem 2.1 Before proving our first two results, we cite two lemmas. The following lemma is due to [22]. Lemma 3.1. For the renewal counting process {Nt, t 0}, described in Section 1, for any v > 0 and any fixed T > 0 it holds lim sup λ 1 te[n v t 1 {Nt>x} ] = 0. x t Λ T The second lemma was proven by Maulik and Zwart [16]. Lemma 3.2. Consider the exponential functional of the Lévy process {R t, t 0} defined as Z = 0 e Rs ds. For every v > 0 satisfying φv < 0, it holds E[Z v ] <. Proof of Theorem 2.1. Since the positive Lévy process i.e. a subordinator {R t, t 0} has non-decreasing paths, it holds uniformly for all t Λ T that PX 1 e Rs > x λds PX 1 e R T > x λt F x λt,

8 8 Y. YANG, K. WANG AND D.G. KONSTANTINIDES where the second step follows from Theorem 3.3 iv in [6] and F D. Hence, there exists some positive constant C 2, such that for sufficiently large x and all t Λ T, PX 1 e Rs > x λds C 2 F x λt. 3.1 For any integer m 1, t Λ T and x > 0, m n PDt > x = + P X k e Rτ k > x, Nt = n =: I 1 + I n=1 n=m+1 For I 2, by R s 0 a.s. for all s 0, Lemma 2.1, Markov s inequality and Lemma 2.2, it holds uniformly for all t Λ T that I 2 n P X k > x PNt = n m<n x/d 1 + n>x/d 1 x n F PNt = n + P Nt > n xd1 m<n x/d 1 p+1 x C 1 F x n p+1 PNt = n + E[N p+1 t 1 {Nt>x/D1}] D 1 m<n x/d 1 C 1 F x E[N p+1 t 1 {Nt>m} ], 3.3 for some p > J + F. Hence, from 3.1, 3.3 and Lemma 3.1, we obtain lim m lim sup x I 2 sup t Λ T PX 1 e Rs > x λds C 1 lim C sup λ 1 t E[N p+1 t 1 {Nt>m} ] = m t Λ T We mainly deal with I 1. Let Hy 1,..., y n+1 be the joint distribution of the random vector τ 1,..., τ n+1, n 1. Clearly, for 1 n m and t Λ T, x > 0, n P = X k e Rτ k > x, Nt = n {0 s 1... s n t, s n+1>t} n P X k e Rs k > x Hds 1,..., ds n Similarly to 3.1, there exist some 0 < C 3 < 1 and large x 1 > 0, both depending only on F, such that for all 1 k n, 0 s k t T, when x x 1, it holds that PX k e Rs k > x C3 F x. 3.6

9 Uniform asymptotics for discounted aggregate claims 9 Next, we aim to show, that for some > 0, which can be arbitrarily small, there exists some x, depending only on F,, and n, such that for all x x and 0 s 1... s n t T, 1 n m, 1 n n PX k e Rs k > x P X k e Rs k > x 1 + n PX k e Rs k > x. 3.7 For any ε 0, 1 2, by Psup s [0, T ] R s = = 0, there exists a δ = δε 0, 1 such that P sup R s < log δ s [0, T ] which implies, that for all 1 k n, 0 s k t T and x > 0, 1 ε, x PX k e Rs k > x F Pe δ u Rs k du x x F P sup R s < log δ 1 ε F. 3.9 δ s [0, T ] δ By F L, there exists some positive, increasing and slowly varying function lx such that lx/x 0 and for any fixed constant K, F x K lx F x, 3.10 which implies, that for the above ε > 0, there exists some x 2 x 1, depending only on F and ε, such that for all x x 2, F x lx F x lx 1 + ε F x δ Since X 1,..., X n are UTAI, and by F D, there exists some x 3 x 2, depending only on F, ε and n, such that for all x x 3 and all 1 i j n 1 n m, PX i > x, X j > x P X i > lx n 1, X j > x ε n n C 3 F x Let choose x = max{x 3, D 1 /1 δ}, which also depends only on F, ε and n, then for x x, it holds that x lx 1 δx and for some p > J + F, F 1 δx F x C 1 1 δ p, 3.13

10 10 Y. YANG, K. WANG AND D.G. KONSTANTINIDES together with equations 3.11, For the lower bound of 3.7, by Bonferroni inequality, 3.12 and 3.6 we have, that for all x x and 0 s k t T, 1 k n, n n P X k e Rs k > x PX k e Rs k > x PX i > x, X j > x 1 ε 1 i<j n n PX k e Rs k > x For the upper bound of 3.7, n P X k e Rs k > x n PX k e Rs k > x lx + 1 i j n P X i > lx n 1, X j > x, 3.15 n where lx is defined in Since lx is infinitely increasing, and by 3.11, 3.13, 3.8 and 3.9, we have, that for all x x, 0 s k t T and 1 k n, 1 δ x lx PX k e Rs k > x lx = + F Pe Rs k du δ 0 u 1 x F δ u lx/u δ 1 δ x Pe Rs k du + F Pe Rs k du δ 0 u 1 x δ x 1 + ε F Pe δ u Rs k du + C1 1 δ p F Pe 0 u Rs k du x 1 + ε PX k e Rs k > x + C1 1 δ p F P sup R s > log δ δ s [0, T ] 1 + ε + ε C 1 1 δ p PX k e Rs k > x ε By 3.12 and 3.6, the second sum on the right-hand side of 3.15 can be bounded from above by ε n PX k e Rs k > x Combining , we obtain, that for all x x and 0 s k t T, 1 k n, n P X k e Rs k > x 1 + ε 2 + C 1 1 δ p n PX k e Rs k > x ε Let = ε 2 + C 1 1 δ p 1 ε 1 > 0, which can be arbitrarily small because of the arbitrariness of ε. Therefore, 3.7 follows from 3.14 and 3.18.

11 Uniform asymptotics for discounted aggregate claims 11 By 3.5 and 3.7, we have, that for all x x depending only on F, and n, 1 n m and t Λ T, 1 n PX k e Rτ k > x, Nt = n n P X k e Rτ k > x, Nt = n 1 + n PX k e Rτ k > x, Nt = n. This means, that it holds uniformly for all t Λ T and 1 n m n n P X k e Rτ k > x, Nt = n PX k e Rτ k > x, Nt = n. Thus, it holds uniformly for all t Λ T that m n PX k e Rτ k > x, Nt = n where Since I 1 = n=1 n=1 n=m+1 n PX k e Rτ k > x, Nt = n =: I3 I 4, 3.19 I 3 = PX k e Rτ k > x, Nt k = PX 1 e Rs > x λds I 4 F x n=m+1 npnt = n = F x E[Nt 1 {Nt>m} ], similarly to the proof of 3.4, and by 3.1, Lemma 3.1, we obtain lim m 1 C 2 lim sup x I 4 sup t Λ T PX 1 e Rs > x λds lim sup λ 1 t E[Nt 1 {Nt>m} ] = m t Λ T Therefore, by 3.4 and the desired 2.1 holds uniformly for all t Λ T and thus we conclude the proof. Proof of Corollary 2.1. The upper bound of ψx, t is trivial. Indeed, by Theorem 2.1 it holds uniformly for all t Λ T ψx, t PDt > x PX 1 e Rs > x λds. 3.22

12 12 Y. YANG, K. WANG AND D.G. KONSTANTINIDES We next turn to the asymptotic lower bound of ψx, t. For any 0 < ε < 1, since the positive Lévy process {R t, t 0} has non-decreasing paths, and according to 0 cu M, 2.1, we have, that for all t Λ T and sufficiently large x, ψx, t PDt > x + MT PDt > 1 + ε x ε x F Pe Rs duλds 0 u F 1+ε x 1 u x inf u 0, 1] F F Pe x u Rs du λds F 1 + ε u 0 PX 1 e Rs > x λds Noting F 1 + ε 1 as ε 0 because of F C, the desired lower bound follows from This completes the proof of Corollary Proof of Theorem 2.2 We start this section by a series of lemmas. The first two lemmas are special cases of Lemmas 3.2 and 3.3 in [15] under the independence structure between {X n, n 1} and {θ n, n 1}. Lemma 4.1. Under the conditions of Theorem 2.2, for each k 1, it holds uniformly for all t Λ F x Pτ k t = O1PX k e Rτ k 1{τk t} > x. Lemma 4.2. Under the conditions of Theorem 2.2, for any positive function ax = x/lx, with the slowly varying function lx, and each k 1, it holds uniformly for all t Λ Pe Rτ k 1{τk t} > ax = o1px k e Rτ k 1{τk t} > x. The next lemma plays an important role in the proof of Theorem 2.2. Lemma 4.3. Under the conditions of Theorem 2.2, for any fixed 0 < y < 1 and each k 1, it holds uniformly for all t Λ ɛ PX k e Rτ k 1{τk t} > xy y β PX k e Rτ k 1{τk t} > x.

13 Uniform asymptotics for discounted aggregate claims 13 Proof. We firstly show that there exists some > 0, not depending on t, such that Pe Rτ k 1{τk t} > > holds for all t Λ ɛ. Indeed, if Pθ 1 = 0 > 0, then for any 0, 1 and all t Λ, Pe Rτ k 1{τk t} > Pe Rτ k 1{τk =0} > = Pθ 1 = 0 k > 0. If Pθ 1 = 0 = 0, then ɛ/k Λ. Hence, For all t Λ ɛ and x 0, Pτ k ɛ P θ 1 ɛ k > k Pe Rτ k 1{τk t} > x ɛ Pe Rs ɛ P > x Pτ k ds sup R s < log x s [0,ɛ] Pτ k ds. 4.3 Noting Psup [0, ɛ] R s = = 0, there exists some > 0, not depending on t, such that C 4 := P sup R s < log s [0,ɛ] > 0. Plugging this into 4.3, and by 4.2, we have that for all t Λ ɛ, Pe Rτ k 1{τk t} > C 4 Pτ k ɛ C 4 P θ 1 ɛ k > 0. k We secondly prove that Pe Rτ k 1{τk t} > x = of x, 4.4 holds uniformly for all t Λ. Indeed, for all t Λ, by Markov s inequality, φr < 0 and Lemma 2.2, we have Pe Rτ k 1{τk t} > x Pe Rs > x Pτ k ds x r e s φr Pτ k ds = of x. The claim 4.4 implies, that there exists some increasing function lx, not depending on t, such that x/lx and Pe Rτ k 1{τk t} > x/lx = of x, 4.5

14 14 Y. YANG, K. WANG AND D.G. KONSTANTINIDES holds uniformly for all t Λ. Then, we have that for all x >, any fixed y 0, 1 and all t Λ ɛ, PX k e Rτ k 1 {τk t} > xy PX k e Rτ k 1 {τk t} > x x/lx x/lx + F x y/upe Rτ k 1 {τk t} du F x/u Pe Rτ k 1 {τk t} du x/lx F x y/u Pe Rτ k 1 {τk t} du F x/u Pe Rτ k 1 {τk t} du F y z sup z lx F z + Pe Rτk 1 {τk t} > x/lx F x/ Pe Rτ k 1 {τk t} >, which, combined with 4.1, 4.5, leads to the desired result. Lemma 4.4. Under the conditions of Theorem 2.2, for each j > i 1, it holds uniformly for all t Λ PX i e Rτ i 1{τi t} > x, X j e Rτ j 1{τj t} > x = o1 PX k e Rτ k 1{τk t} > x. k=i,j Proof. By Lemma 4.2, we have that uniformly for all t Λ, PX i e Rτ i 1{τi t} > x, X j e Rτ j 1{τj t} > x PX i e Rτ i > x, Xj e Rτ j > x, τi t, e Rτ j ax + Pe Rτ j 1{τj t} > ax =: J 1 + o1px j e Rτ j 1{τj t} > x, 4.6 where ax is the function defined in Lemma 4.2. pairwise NQD, it holds uniformly for all t Λ J 1 P X i e Rτ i > x, τi t, X j > x ax = P X i > x u, X j > x ax = F 0 0 x ax Pe Rs For J 1, since {X n, n 1} are du Pτ i ds x x F F Pe u Rs du Pτ i ds ax PX i e Rτ i 1{τi t} > x = o1px i e Rτ i 1{τi t} > x. Plugging this estimate into 4.6, we finish the proof of the lemma. Lemma 4.5. Under the conditions of Theorem 2.2, for each n 1, it holds uniformly for all t Λ ɛ n P X k e Rτ k 1{τk t} > x n PX k e Rτ k 1{τk t} > x.

15 Uniform asymptotics for discounted aggregate claims 15 Proof. For any 0 < ε < 1, we have n P X k e Rτ k 1{τk t} > x +P n n PX k e Rτ k 1{τk t} > 1 ε x 4.7 X k e Rτ k 1{τk t} > x, max 1 k n X k e Rτ k 1{τk t} 1 ε x We firstly deal with J 2. By Lemma 4.3 we have =: J 2 + J 3. J 2 1 ε β n PX k e Rτ k 1{τk t} > x, 4.8 holds uniformly for all t Λ ɛ. For J 3, by Lemmas 4.4 and 4.3, we have n J 3 P {X x} i e Rτ i 1{τi t} > x n, X j e Rτ j 1{τj t} > ε n i=1 i=1 1 j n, j i = o1 1 j n, j i P X i e Rτ i 1{τi t} > ε x n, X j e Rτ j 1{τj t} > εx n n PX i e Rτ i 1{τi t} > x, 4.9 i=1 holds uniformly for all t Λ ɛ. From , it holds uniformly for all t Λ ɛ n n P X k e Rτ k 1{τk t} > x 1 ε β PX k e Rτ k 1{τk t} > x Then, the upper bound follows from 4.10 and the arbitrariness of ε. For the lower bound, according to Bonferroni inequality, n n { } P X k e Rτ k 1{τk t} > x P X k e Rτ k 1{τk t} > x n PX k e Rτ k 1{τk t} > x 1 i<j n PX i e Rτ i 1{τi t} > x, X j e Rτ j 1{τj t} > x n PX k e Rτ k 1{τk t} > x, holds uniformly for all t Λ, where we used Lemma 4.4 in the last step. The last lemma can be found in [15].

16 16 Y. YANG, K. WANG AND D.G. KONSTANTINIDES Lemma 4.6. Under the conditions of Theorem 2.2, it holds P k=n+1 lim lim sup sup X k e Rτ k 1 {τk t} > x n x t Λ PX 1 e Rτ 1 1 {τ1 t} > x k=n+1 PX k e Rτ k 1 {τk t} > x = lim n lim sup sup x t Λ PX 1 e Rτ 1 1 {τ1 t} > x = 0. Proof of Theorem 2.2. For any ε > 0, according to Lemma 4.6, there exists some sufficiently large integer n 0 such that uniformly for all t Λ, { } max P X k e Rτ k 1{τk t} > x, PX k e Rτ k 1{τk t} > x k=n 0+1 k=n 0+1 εpx 1 e Rτ 1 1{τ1 t} > x For the upper bound, by Lemma 4.5, 4.11 and Lemma 4.3, we have that for any 0 < δ < 1, PDt > x P n 0 + P k=n 0+1 X k e Rτ k 1{τk t} > 1 δ x n0 X k e Rτ k 1{τk t} > δ x PX k e Rτ k 1{τk t} > 1 δ x + εpx 1 e Rτ 1 1{τ1 t} > δ x n 0 1 δ β PX k e Rτ k 1{τk t} > x + ε δ β PX 1 e Rτ 1 1{τ1 t} > x 1 δ β + ε δ β PX 1 e Rs > x λds, 4.12 holds uniformly for all t Λ ɛ. Thus, the upper bound of 2.1 follows from 4.12 by letting firstly ε 0, then δ 0. For the lower bound, we obtain from Lemma 4.5 and 4.11 n0 PDt > x P X k e Rτ k 1{τk t} > x 1 ε k=n 0+1 PX k e Rτ k 1{τk t} > x PX 1 e Rs > x λds, 4.13 holds uniformly for all t Λ ɛ. It ends the proof of the theorem.

17 Uniform asymptotics for discounted aggregate claims 17 Proof of Corollary 2.2. Clearly, Theorem 2.2 implies, that the upper bound 3.22 holds uniformly for all t Λ ɛ. We estimate the lower bound for the ruin probability. For any ε > 0 and the integer n 0 defined in the proof of Theorem 2.2, we have ψx, t PDt M Z > x 4.14 n0 P X k e Rτ k 1{τk t} > 1 + ε x P Z > ε x =: J 4 J 5, M where Z = e Rs ds and M > 0 is the upper bound of the intensity of premium payments. As done in 4.13, by Lemmas 4.5, 4.3 and 4.11, we get that J 4 n 0 PX k e Rτ k 1{τk t} > 1 + ε x n ε β PX k e Rτ k 1{τk t} > x 1 + ε β 1 ε PX 1 e Rs > x λds, 4.15 holds uniformly for all t Λ ɛ. For J 5, by Markov s inequality and Lemmas 3.2, 2.2, we have J 5 r M E[Z r ] x r = of x ε If Pθ 1 = 0 > 0, then by Lemma 4.1, it holds that uniformly for all t Λ F x Pθ 1 = 0 1 F x Pθ 1 t = O1PX 1 e Rτ 1 1{τ1 t} > x, which, together with 4.16, yields that uniformly for all t Λ, J 5 = o1px 1 e Rτ 1 1{τ1 t} > x = o1 PX 1 e Rs > x λds Similarly, if Pθ 1 = 0 = 0, also by Lemma 4.1, it holds uniformly for all t Λ ɛ F x Pθ 1 ɛ 1 F x Pθ 1 t = O1PX 1 e Rτ 1 1{τ1 t} > x, which implies, that 4.17 holds uniformly for all t Λ ɛ. Combining and noticing the arbitrariness of ε, we can derive the lower bound.

18 18 Y. YANG, K. WANG AND D.G. KONSTANTINIDES Acknowledgements The authors are most grateful to the referee for her/his very thorough reading of the paper and valuable suggestions, which greatly improve the original results and presentation of this paper. The first and second authors were supported by the National Natural Science Foundation of China grant numbers , References [1] Asmussen, S Subexponential asymptotics for stochastic processes: extremal behavior, stationary distributions and first passage probabilities. Ann. Appl. Probab. 8, [2] Bingham, N. H., Goldie, C. M. and Teugels, J. L Regular Variation. Cambridge University Press, Cambridge. [3] Block, H. W., Savits, T. H. and Shaked, M Some concepts of negative dependence. Ann. Probab. 10, [4] Cai, J Ruin probabilities and penalty functions with stochastic rates of interest. Stochastic Process. Appl. 112, [5] Chen, Y. and Ng, K. W The ruin probability of the renewal model with constant interest force and negatively dependent heavy-tailed claims. Insurance Math. Econom. 40, [6] Cline, D. B. H. and Samorodnitsky, G Subexponentiality of the product of independent random variables. Stochast. Process. Appl. 49, [7] Cont, R. and Tankov, P Financial Modelling with Jump Processes. Chapman & Hall/CRC, Boca Raton, FL. [8] Hao, X. and Tang, Q A uniform asymptotic estimate for discounted aggregate claims with sunexponential tails. Insurance Math. Econom. 43, [9] Joag-Dev, K. and Proschan, F Negative association of random variables with application. Ann. Statist. 11, [10] Kalashnikov, V. and Konstantinides, D Ruin under interest force and subexponential claims: a simple treatment. Insurance Math. Econom. 27, [11] Kalashnikov, V. and Norberg, R Power tailed ruin probabilities in the presence of risky investments. Stochastic Process. Appl. 98, [12] Klüppelberg, C. and Stadtmüller, U Ruin probabilities in the presence of heavy-tails and interest rates. Scand. Actuar. J. 1,

19 Uniform asymptotics for discounted aggregate claims 19 [13] Konstantinides, D., Tang, Q. and Tsitsiashvili, G Estimates for the ruin probability in the classical risk model with constant interest force in the presence of heavy tails. Insurance Math. Econom. 31, [14] Lehmann, E. L Some concepts of dependence. Ann. Math. Statist. 37, [15] Li, J Asymptotics in a time-dependent renewal risk model with stochastic return. J. Math. Anal. Appl. 387, [16] Maulik, K. and Resnick, S Characterizations and examples of hidden regular variation. Extremes 7, [17] Maulik, K. and Zwart, B Tail asymptotics for exponential functionals of Lévy processes. Stochastic Process. Appl. 116, [18] Paulsen, J Risk theory in a stochastic economic environment. Stochastic Process. Appl. 46, [19] Paulsen, J On Cramér-like asymptotics for risk processes with stochastic return on investments. Ann. Appl. Probab. 12, [20] Paulsen, J. and Gjessing, H. K Ruin theory with stochastic return on investments. Adv. in Appl. Probab. 29, [21] Tang, Q The finite time ruin probability of the compound Poisson model with constant interest force. J. Appl. Probab. 42, [22] Tang, Q Heavy tails of discounted aggregate claims in the continuous-time renewal model. J. Appl. Probab. 44, [23] Tang, Q. and Tsitsiashvili, G Precise estimates for the ruin probability in finite horizon in a discrete-time model with heavy-tailed insurance and financial risks. Stochastic Process. Appl. 108, [24] Tang, Q., Wang, G. and Yuen, K. C Uniform tail asymptotics for the stochastic present value of aggregate claims in the renewal risk model. Insurance Math. Econom. 46, [25] Yang, Y. and Wang, Y Asymptotics for ruin probability of some negatively dependent risk models with a constant interest rate and dominatedly-varying-tailed claims. Stat. Probab. Lett. 80, [26] Yuen, K. C., Wang, G. and Ng, K. W Ruin probabilities for a risk process with stochastic return on investments. Stochastic Process. Appl. 110, [27] Yuen, K. C., Wang, G. and Wu, R On the renewal risk process with stochastic interest. Stochastic Process. Appl. 116,