On asymptotic behaviour of probabilities of small deviations for compound Cox processes

We derive logarithmic asymtotics for probabilities of small deviations for compound Cox processes in the space of trajectories. We find conditions under which these asymptotics are the same as those for sums of independent identically distributed random variables and homogeneous processes with indep...

Повний опис

Збережено в:
Бібліографічні деталі
Дата:2008
Автор: Frolov, A.N.
Формат: Стаття
Мова:Англійська
Опубліковано: Інститут математики НАН України 2008
Онлайн доступ:https://nasplib.isofts.kiev.ua/handle/123456789/4548
Теги: Додати тег
Немає тегів, Будьте першим, хто поставить тег для цього запису!
Назва журналу:Digital Library of Periodicals of National Academy of Sciences of Ukraine
Цитувати:On asymptotic behaviour of probabilities of small deviations for compound Cox processes / A.N. Frolov // Theory of Stochastic Processes. — 2008. — Т. 14 (30), № 2. — С. 19–27. — Бібліогр.: 10 назв.— англ.

Репозитарії

Digital Library of Periodicals of National Academy of Sciences of Ukraine
_version_ 1860059819779555328
author Frolov, A.N.
author_facet Frolov, A.N.
citation_txt On asymptotic behaviour of probabilities of small deviations for compound Cox processes / A.N. Frolov // Theory of Stochastic Processes. — 2008. — Т. 14 (30), № 2. — С. 19–27. — Бібліогр.: 10 назв.— англ.
collection DSpace DC
description We derive logarithmic asymtotics for probabilities of small deviations for compound Cox processes in the space of trajectories. We find conditions under which these asymptotics are the same as those for sums of independent identically distributed random variables and homogeneous processes with independent increments. We show that if these conditions do not hold, the asymptotics of small deviations for compound Cox processes are quite different.
first_indexed 2025-12-07T17:03:40Z
format Article
fulltext Theory of Stochastic Processes Vol. 14 (30), no. 2, 2008, pp. 19–27 UDC 519.21 ANDREI N. FROLOV ON ASYMPTOTIC BEHAVIOUR OF PROBABILITIES OF SMALL DEVIATIONS FOR COMPOUND COX PROCESSES We derive logarithmic asymtotics for probabilities of small deviations for compound Cox processes in the space of trajectories. We find conditions under which these asymptotics are the same as those for sums of independent identically distributed random variables and homogeneous processes with independent increments. We show that if these conditions do not hold, the asymptotics of small deviations for compound Cox processes are quite different. Introduction The asymptotic behaviour of probabilities of small deviations has been investigated for various classes of stochastic processes. Asymptotics of probabilities of small deviations for sums of independent random variables and homogeneous processes with independent increments were found in Mogul’skii [1], Borovkov and Mogul’skii [2] and references therein. Note that the last class of processes includes stable, Poisson and compound Poisson processes. Various results for Gaussian processes and references may be found in surveys by Ledoux [3], Li and Shao [4] and Lifshits [5]. Further results were obtained for various stochastic processes, generated by sums of random numbers of independent random variables. In this case, the number of summands is a stochastic process which is usually independent with the summands. Increments of such the processes may be dependent. These processes are called compound processes. Now we give definitions for some of them. Let X,X1, X2, . . . be a sequence of independent, identically distributed random vari- ables. Put Sn = X1 +X2 + · · · +Xn for n � 1, S0 = 0. Let ν(t) be a standard Poisson process, independent with the sequence {Xk}. Then the stochastic process η(t) = Sν(λt), λ > 0, is called a compound Poisson process. If δ(t) is a renewal process, independent with {Xk}, then ζ(t) = Sδ(t) is called a compound renewal process. When the renewal times have an exponential distribution, the compound renewal process coincides with the compound Poisson process. Small deviations of the renewal and compound renewal processes has been studied in Frolov, Martikainen, Steinebach [6]. Now we turn to the definition of the compound Cox process. We start with the defi- nition of a Cox process which we borrow from the paper of Embrechts and Klüppelberg [7]. Let Λ(t), t > 0, be a random measure, i.e. a.s. (almost surely) Λ(0) = 0, Λ(t) < ∞ for all t > 0 and Λ(t) has non-decreasing trajectories. Assume that Λ(t) does not depend on the standard Poisson process ν(t). The point process N(t) = ν(Λ(t)) is called a Cox process. If, in particular, the trajectories of Λ(t) are continuous a.s., then for every 2000 AMS Mathematics Subject Classification. Primary 60F15; Secondary 60K05. Key words and phrases. Small deviation, limit theorems, compound Cox processes. This article was partially supported by RFBR, grant 05–01–00486. 19 20 ANDREI N. FROLOV realization λ(t) of the measure Λ(t) the process N(t) is a non-homogeneous Poisson process with the intensity measure λ(t). Now we define the compound Cox process in the same way as we introduce the pro- cesses η(t) and ζ(t) above. Let N(t) be a Cox process, independent with the sequence {Xk}. The stochastic process S(t) = SN(t) is called a compound Cox process. Compound Cox processes play an important role in actuarial and financial mathemat- ics. They describe, for example, the processes of total claims of an insurance company in a collective risk model (cf. [7]). Note that under additional conditions (cf.,e.g. [7]), the Cox processes are renewal processes. Nevertheless, in the sequel, we assume that Λ(t) is such that our Cox process will be the renewal process only if it is a Poisson process. So, we do not consider renewal processes here. The logarithmic asymptotics of small deviations of the compound Cox processes has been investigated by Frolov [8]. In there, we have described the behaviour of PT = P ( sup 0�t�T ∣∣∣S(t) − cΛ(t) ∣∣∣ � xT ) , where {xT } is a positive real function such that xT → ∞ and x2 T = o(f(T )) as T → ∞, f(T ) is a function, depending on properties of the measure Λ(t), c = EX for EX < ∞ and c = 0 otherwise. Note that if EX <∞, then S(t) is centered by a random function, generally speaking. Nevertheless, in the case of the homogeneous Poisson process, Λ(t) = λt and our center- ing coincides with ES(t). The same holds true when N(t) is a non-homogeneous Poisson process. Moreover, centering functions of this type are used in appropriate models of the risk process. It turns out that fluctuations of risk in such models have to be compensate by insurance premiums of random amounts since, otherwise, the ruin probability of in- surance company may be separated from zero and may be non-decreasing with increasing of initial capital of insurance company (cf. [7]). Therefore we consider this centering. Here, we present generalizations of the results in Frolov [8]. PT is the probability that trajectories of the compound Cox process, centered at cΛ(t) and normed by xT , lie in a strip of width 2 around zero. We consider below more general sets instead of strips and derive logarithmic asymtotics for such probabilities. We find conditions under which these asymptotics are the same as those for sums of independent identically distributed random variables and homogeneous processes with independent increments. We show that the asymptotics of small deviations for compound Cox processes are quite different, if these conditions do not hold. 1. Results Let X,X1, X2, . . . be a sequence of independent, identically distributed random vari- ables. Put Sn = X1 +X2 + · · · +Xn for n � 1, S0 = 0. Let ν(t) be a standard Poisson process, independent with the sequence {Xk}. Denote ξ(t) = Sν(t). Note that ξ(t) is a compound Poisson process and, therefore, it is a homogeneous process with independent increments. Let Λ(t), t > 0, be a random measure, i.e. a.s. (almost surely) Λ(0) = 0, Λ(t) < ∞ for all t > 0 and Λ(t) has non-decreasing trajectories. Assume that the trajectories of Λ(t) are a.s. continuous and Λ(∞) = ∞ a.s. Suppose that Λ(t) is independent with the process ν(t) and the sequence {Xk}. Define the Cox process N(t) and the compound Cox process S(t) by the relations N(t) = ν(Λ(t)) and S(t) = SN(t). Let xT be a real function with xT → ∞ as T → ∞. SMALL DEVIATIONS FOR COMPOUND COX PROCESSES 21 Let g1(t), g2(t), t ∈ [0, 1], be continuous functions such that g1(0) < 0 < g2(0), g1(t) is non-increasing and g2(t) is non-decreasing. Put PT = P ( g1 ( Λ(t) Λ(T ) ) xT � S(t) − cΛ(t) � g2 ( Λ(t) Λ(T ) ) xT for all t ∈ [0, T ] ) . The probability PT is well defined since trajectories of S(t) are jump functions. In the sequel, we will describe the asymptotic behaviour of probabilities of small deviations PT . In general case, gi(Λ(t)/Λ(T )) in the definition of PT are random. We mentioned above that the random centering of S(t) is well motivated. So, random normings may also be used. Nevertheless, there are important cases in which gi(Λ(t)/Λ(T )) are non-random functions. The first case is that g1(t) ≡ −1 and g2(t) ≡ 1. In this case the results have been obtained in Frolov [8]. If N(t) is a Poisson process, then Λ(t) is a continuous increasing function. This is the second case. The third case arises when Λ(t) = Λf(t), where Λ is a positive random variable and f(t) is a continuous function. Then gi(Λ(t)/Λ(T )) = gi(f(t)/f(T )). If, in addition, f(t) = tβ , β > 0, we have PT = P ( g3 ( t T ) xT � S(t) − cΛ(t) � g4 ( t T ) xT for all t ∈ [0, T ] ) = P (g3(t)xT � S(T t) − cΛ(T t) � g4(t)xT for all t ∈ [0, 1]) , where gi+2(t) = gi(tβ), i = 1, 2. Note that in this last case, one can consider more general sets in the definition of PT . To this end, consider a family of processes {sT (t) = S(T t)/xT ; 0 � t � 1, T ∈ (0,∞)}. The trajectories sT (·) belong to the Skorohod space D[0, 1] and we can introduce P (sT (·) ∈ G), where G ∈ G and G is a class of subsets of D[0, 1]. One can define this class in the same way as in Mogul’skii [1] with the following additional assumption: on the first step of the definition on p. 757, functions L1(t) and L2(t) (L1(t) > L2(t)) have to be non-decreasing and non-increasing correspondingly. Of course, the class G will be a subclass of the class in Mogul’skii [1], but it will be wide enough. It seems that the most interesting set is the set G0 = {g ∈ D[0, 1] : g(0) = 0, g3(t) < g(t) < g4(t), for all t ∈ [0, 1]}, where g3(t), g4(t), t ∈ [0, 1], are continuous functions such that g3(0) < 0 < g4(0), g3(t) is non-increasing and g4(t) is non-decreasing. The asymptotics of P (sT (·) ∈ G0) may be derived from our results below. Our first result is the following theorem. Theorem 1. Assume that there exist a positive, increasing, continuous function f(T ), f(T ) → ∞ as T → ∞, and a non-negative random variables Λ such that the dis- tributions of Λ(T )/f(T ) converge weakly to the distribution Λ as T → ∞. Denote aT = ess inf Λ(T )/f(T ), a = ess inf Λ. Assume that aT → a as T → ∞ and a > 0. If EX <∞, then put c = EX. Put c = 0 otherwise. Suppose that the distribution of the random variable ξ1 = ξ(1)− c belongs to a domain of attraction of a strictly stable law Fα with the index α ∈ (0, 2], i.e. the distributions of (ξ(n) − cn)/Bn converge weakly to Fα as n → ∞, where {Bn} is a sequence of positive constants. Assume that Fα is not concentrated on the half of line. Then for every positive function xT with xT → ∞ and xT = o(B[f(T )]) as T → ∞, the following relation holds (1) logPT = −CHαa f(T ) xα T L(xT ) ( 1 + o(1) ) as T → ∞, 22 ANDREI N. FROLOV where L(x) = xα−2 Eξ21I{|ξ1| < x} is a slowly varying at infinity function, C is an abso- lute positive constant, depending only on the distribution Fα, Hα = ∫ 1 0 (g2(t)−g1(t))−αdt. If α = 2, then C = π2/8. Here and in the sequel, I{B} denotes the indicator of the event B, [x] denotes the integer part of x. Theorem 1 for g1(t) ≡ −1 and g2(t) ≡ 1 has been obtained in Frolov [8]. Conditions, necessary and sufficient for belonging of the distribution of ξ1 to a domain of attraction of Fα, are well known. These conditions are usually stated in terms of asymptotic behaviours for tails or truncated moments (cf., for example, Feller [9], Chapter XVII, §5). Nevertheless, to check the conditions of Theorem 1, it is more convenient to apply Theorem 2.6.5, p. 103 from Ibragimov and Linnik [10] where such conditions are given in terms of an asymptotic behaviour of a characteristic function at zero. Since Eeitξ1 = exp { −itc+ EeitX − 1 } , one can easily check these conditions. The simplest example of Λ(t) is Λ(t) = Λf(t), where the random variable Λ and the function f(t) satisfy the conditions of Theorem 1.(Note that the random variable Λ may be degenerate and we will deal in this case with the non-homogeneous Poisson process N(t) and the corresponding process S(t).) We will arrive to another examples, if Λ(t) will be a stochastic process which satisfies the law of large numbers and has appropriate trajectories. Theorem 1 yields that if a > 0, then the behaviour of logPT is the same as that for probabilities of small deviations for sums of independent identically distributed random variables and homogeneous processes with independent increments. Now we turn to the case a = 0. In this case the asymptotic of logPT may be quite different. Theorem 2. Assume that the conditions of Theorem 1 hold and a = 0. Then for every positive function xT with xT → ∞ and xT = o(B[f(T )]) as T → ∞, the following relation holds (2) logPT = o ( f(T ) xα T L(xT ) ) as T → ∞, where L(x) is the function from Theorem 1. In the case g1(t) ≡ −1 and g2(t) ≡ 1, Theorem 2 has been proved in Frolov [8]. Theorem 2 does not give the exact asymptotic of logPT . In the next result we find this asymptotic which depends on the behaviour of the distribution functions of Λ(T )/f(T ) and Λ at zero. Theorem 3. Assume that the conditions of Theorem 1 hold and aT = a = 0 for all T . Put FT (λ) = P (Λ(T ) < λf(T )) and F (λ) = P (Λ < λ). For every positive function xT with xT → ∞ and xT = o(B[f(T )]) as T → ∞, define εT by the relation εT = sup { ε > 0 : ε − logFT (ε) � xα T CHαf(T )L(xT ) } , where the function L(x) and the constants C and Hα are from Theorem 1. Assume that εT is equivalent to a continuous decreasing function and FT (εT ) → 0 as T → ∞. Suppose that for every τ > 0, the following relation holds (3) logFT (τεT ) ∼ logFT (εT ) as T → ∞. Then (4) logPT = logFT (εT ) ( 1 + o(1) ) = −εTCHα f(T ) xα T L(xT ) ( 1 + o(1) ) as T → ∞. SMALL DEVIATIONS FOR COMPOUND COX PROCESSES 23 Here εT → 0 as T → ∞. Theorem 3 for g1(t) ≡ −1 and g2(t) ≡ 1 and continuous FT (λ) and F (λ) has been obtained in Frolov [8]. Note that if for all T the distribution functions FT (λ) and F (λ) are continuous and positive in the non-degenerate interval [0, λ0], then in Theorem 3, εT is a continuous decreasing function and FT (εT ) → 0 as T → ∞. Indeed, FT (εT ) � Δ + F (εT ), where Δ = sup0�x�λ0 |FT (x) − F (x)| → 0 as T → ∞ by the weak convergence of FT (λ) to F (λ) and the continuity of the limit function. We now show that that the asymptotic of logPT in (1) and (4) are quite different. To this goal, suppose that FT (x) ≡ F (x) for all T . If, for example, F (x) = xp for x ∈ [0, 1], where p > 0, then logFT (εT ) ∼ −p log(f(T ) xα T L(xT )) as T → ∞. If F (x) = (− log x)−p for x ∈ (0, e−1], where p > 0, then logFT (εT ) ∼ −p log log(f(T ) xα T L(xT )) as T → ∞. It turns out that the condition (3) can not be omitted in Theorem 3. This follows from the next result. Theorem 4. Assume that all the conditions of Theorem 3 hold except the condition (3). Assume that for all τ > 0, the following relation holds logFT (τεT ) ∼ τp logFT (εT ) as T → ∞, where p > 0. Then (5) logPT = o ( εT f(T ) xα T L(xT ) ) as T → ∞. Theorem 4 for g1(t) ≡ −1 and g2(t) ≡ 1 and continuous FT (λ) and F (λ) has been proved in Frolov [8]. 2. Proofs Put ξ̄(t) = ξ(t) − ct. In what follows, we will use the following result. Lemma 1. If the function f(t) satisfies the conditions of Theorem 1, then the probability P (g1(t)xT � ξ̄(λf(T )t) � g2(t)xT for all t ∈ [0, 1]) is non-increasing in λ. Proof of Lemma 1. Take λ > 1. Then P (g1(t)xT � ξ̄(λf(T )t) � g2(t)xT for all t ∈ [0, 1]) = P ( g1 ( u λf(T ) ) xT � ξ̄(u) � g2 ( u λf(T ) ) xT for all u ∈ [0, λf(T )] ) � P ( g1 ( u λf(T ) ) xT � ξ̄(u) � g2 ( u λf(T ) ) xT for all u ∈ [0, f(T )] ) � P ( g1 ( u f(T ) ) xT � ξ̄(u) � g2 ( u f(T ) ) xT for all u ∈ [0, f(T )] ) = P (g1(t)xT � ξ̄(f(T )t) � g2(t)xT for all t ∈ [0, 1]). In the last inequality we have used that g1(t) is non-increasing and g2(t) is non-decreasing. We also need the following result on asymptotics of small deviations for compound Poisson process ξ(t). Lemma 2. Let g(T ) be an increasing, continuous function with g(T ) → ∞ as T → ∞. 24 ANDREI N. FROLOV If the conditions of Theorem 1 hold, then for every positive function xT with xT → ∞ and xT = o(B[g(T )]) as T → ∞, the following relation holds (6) logP ( g1(t)xT � ξ̄(g(T )t) � g2(t)xT for all t ∈ [0, 1] ) = −CHα g(T ) xα T L(xT ) ( 1 + o(1) ) as T → ∞, where L(x) and C,Hα are the function and the constants from Theorem 1. Proof of Lemma 2. Take a sequence {Tk} such that Tk ↗ ∞ as k → ∞. Since ξ(t) is a homogeneous process with independent increments, we have by Theorem 4 from Mogul’skii [1] and Lemma 1 P ( g1(t)xTk � ξ̄(g(Tk)t) � g2(t)xTk for all t ∈ [0, 1] ) �P ( g1(t)xTk � ξ̄([g(Tk)]t) � g2(t)xTk for all t ∈ [0, 1] ) = exp { −CHα [g(Tk)] xα Tk L(xTk )(1 + o(1)) } as k → ∞. The lower bound for the probability in (6) may be derived in the same way. Taking into account that the sequence {Tk} may be chosen arbitrarily, we get (6). Proof of Theorem 1. Put FT (λ) = P (Λ(T ) < λf(T )). Taking into account Lemma 1 and the independence of ξ̄(t) and Λ(t), we have PT = P ( g1 ( Λ(t) Λ(T ) ) xT � ξ̄(Λ(t)) � g2 ( Λ(t) Λ(T ) ) xT for all t ∈ [0, T ] ) = P (g1(t)xT � ξ̄(Λ(T )t) � g2(t)xT for all t ∈ [0, 1]) = ∫ ∞ 0 P (g1(t)xT � ξ̄(λf(T )t) � g2(t)xT for all t ∈ [0, 1])dFT (λ)(7) = ∫ ∞ aT P (g1(t)xT � ξ̄(λf(T )t) � g2(t)xT for all t ∈ [0, 1])dFT (λ) � P (g1(t)xT � ξ̄(aT f(T )t) � g2(t)xT for all t ∈ [0, 1]). Take ε ∈ (0, a). Then aT � a − ε for all sufficiently large T . By Lemma 1 it follows that (8) PT � P (g1(t)xT � ξ̄((a− ε)f(T )t) � g2(t)xT for all t ∈ [0, 1]). for all sufficiently large T . The norming constants Bn may be chosen such that Bn = n1/αL1(n), where L1(x) is a slowly varying at infinity function (cf., for example, [10], p. 48). It follows that the condition xT = o(B[f(T )]) as T → ∞ is equivalent to the condition xT = o(B[bf(T )]) as T → ∞, where b is an arbitrary fixed positive constant. By Lemma 2 logP ( g1(t)xT � ξ̄((a− ε)f(T )t) � g2(t)xT for all t ∈ [0, 1] ) = −CHα(a− ε) f(T ) xα T L(xT ) ( 1 + o(1) ) as T → ∞. The latter and (8) yield that lim sup T→∞ xα T f(T )L(xT ) logPT � −CHα(a− ε). Taking in the last inequality the limit as ε→ 0, we get (9) lim sup T→∞ xα T f(T )L(xT ) logPT � −CHαa. SMALL DEVIATIONS FOR COMPOUND COX PROCESSES 25 Take ε > 0. Since aT → a as T → ∞, we get from (7) and Lemma 1 that PT � ∫ (1+ε)2aT aT P (g1(t)xT � ξ̄(λf(T )t) � g2(t)xT for all t ∈ [0, 1])dFT (λ) � P (g1(t)xT � ξ̄((1 + ε)2aT f(T )t) � g2(t)xT for all t ∈ [0, 1])FT ((1 + ε)2aT ) � P (g1(t)xT � ξ̄((1 + ε)3af(T )t) � g2(t)xT for all t ∈ [0, 1])FT ((1 + ε)a) for all sufficiently large T . Choose ε such that (1 + ε)a is a point of continuity for F (λ) = P (Λ < λ). By the weak convergence of FT (λ) to F (λ) we have (10) PT � 1 2 P (g1(t)xT � ξ̄((1 + ε)3af(T )t) � g2(t)xT for all t ∈ [0, 1])F ((1 + ε)a) for all sufficiently large T . Since xT = o(B[(1+ε)3af(T )]) as T → ∞, by Lemma 2 logP ( g1(t)xT � ξ̄((1 + ε)3af(T )t) � g2(t)xT for all t ∈ [0, 1] ) = −C(1 + ε)3aHα f(T ) xα T L(xT ) ( 1 + o(1) ) as T → ∞. It follows from (10) that (11) lim inf T→∞ xα T f(T )L(xT ) logPT � −CaHα(1 + ε)3. Taking in the last inequality the limit as ε→ 0, we get lim inf T→∞ xα T f(T )L(xT ) logPT � −CaHα, which together with (9) yields (1). Proof of Theorem 2. It is clear that logPT � 0 and we need only prove the lower bound. Take ε > 0. Using (7) and Lemma 1, we have PT � ∫ (1+ε)2ε ε P (g1(t)xT � ξ̄(λf(T )t) � g2(t)xT for all t ∈ [0, 1])dFT (λ) � P (g1(t)xT � ξ̄((1 + ε)2εf(T )t) � g2(t)xT for all t ∈ [0, 1])(FT ((1 + ε)2ε) − FT (ε)) for all sufficiently large T . In the same way as in the proof of Theorem 1, the last implies (11) with ε instead of a. Taking the limit as ε→ 0, we get (2). Proof of Theorem 3. Put bT = CHα f(T ) xα T L(xT ). The condition xT = o(B[f(T )]) as T → ∞ and formulae for the norming constants Bn, which may be chosen to satisfy (12) nL(Bn) Bα n → d as n→ ∞, (cf., for example, [9], Chapter XVII, §5), imply that bT → ∞ as T → ∞. We will first prove that εT → 0 as T → ∞. Suppose that there exists a sequence {Tk} such that Tk ↗ ∞ as k → ∞ and εTk > ε > 0 for all sufficiently large k, where ε is a point of continuity of F (λ). Then − logFTk (εTk ) � − logFTk (ε) � − log(F (ε)/2) < ∞ for all sufficiently large k in view of the weak convergence of FT (λ) to F (λ). Hence 1/bTk � −εTk / logFTk (εTk ) > −ε/ log(F (ε)/2) > 0 which contradicts to the relation bT → ∞ as T → ∞. By the definition of εT , we get −3εT/ log(FT (3εT )) > 1/bT . This and (3) imply that −2εT/ log(FT (εT )) � 1/bT for all sufficiently large T . The latter and the relation 26 ANDREI N. FROLOV FT (εT ) → 0 as T → ∞ give 1/bT = o(εT ) as T → ∞. The last relation, the definition of bT and (12) with n = [f(T )εT τ ], τ > 0, yield that xα TL(B[f(T )εT τ ]) Bα [f(T )εT τ ]L(xT ) → 0 as T → ∞. Using the well known representation of a slowly varying function (cf., for example, [9] Chapter VIII, §9), we have that L(B[f(T )εT τ ]) L(xT ) = l(B[f(T )εT τ ]) l(xT ) exp {∫ B[f(T )εT τ] xT �(u) u du } , where l(u) → l <∞ and �(u) → 0 as u→ ∞. This yields that L(B[f(T )εT τ ]) L(xT ) � 1 2 ( xT B[f(T )εT τ ] )−α/2 for all sufficiently large T . It follows that xT = o(B[f(T )εT τ ]) as T → ∞, where τ > 0. Without loss of generality we will assume in the rest of the proof that εT is continuous. Otherwise, one can replace εT by an equivalent function in the sequel. We have from (7) and Lemma 1 PT = ∫ εT 0 P (g1(t)xT � ξ̄(λf(T )t) � g2(t)xT for all t ∈ [0, 1])dFT (λ) + ∫ ∞ εT P (g1(t)xT � ξ̄(λf(T )t) � g2(t)xT for all t ∈ [0, 1])dFT (λ) � FT (εT ) + P (g1(t)xT � ξ̄(εT f(T )t) � g2(t)xT for all t ∈ [0, 1]) � e−εT bT + P (g1(t)xT � ξ̄(εT f(T )t) � g2(t)xT for all t ∈ [0, 1]). Applying Lemma 2, we get the upper bound in (4). We now turn to the lower bound. Take τ > 0. By Lemmas 1 and 2 PT � ∫ τεT 0 P (g1(t)xT � ξ̄(λf(T )t) � g2(t)xT for all t ∈ [0, 1])dFT (λ) � P (g1(t)xT � ξ̄(τεT f(T )t) � g2(t)xT for all t ∈ [0, 1])FT (τεT ) = e−τεT bT (1+o(1))FT (τεT ) as T → ∞. By the definition of εT we get logFT ((1 + τ)εT ) > −(1 + τ)εT bT . It follows from the last inequality and (3) that PT � e−τεT bT (1+o(1))+log FT ((1+τ)εT )(1+o(1)) � e−(1+2τ)εT bT (1+o(1))(13) as T → ∞. Since τ may be chosen arbitrarily small, we get the lower bound in (4). Proof of Theorem 4. As in the proof of Theorem 2, we need only prove the lower bound. In the same way as in proof of (13), we get PT � e−(τp(1+τ)1−p+τ)εT bT (1+o(1)) as T → ∞ which yields (5). SMALL DEVIATIONS FOR COMPOUND COX PROCESSES 27 Bibliography 1. A.A. Mogul’skii, Small deviations in a space of trajectories, Theory Probab. Appl. 19 (1974), no. 4, 726-736. (Russian) 2. A.A. Borovkov, A.A. Mogul’skii, On probabilities of small deviations for stochastic processes, Sib. Adv. Math. 1 (1991), 39-63. 3. M. Ledoux, Isoperimetry and Gaussian analysis, Lectures on Probability Theory and Statistics, Lecture Notes in Mathematics, Vol. 1648, Springer, Berlin (1996), 165-294. 4. W.V. Li, Q.-M. Shao, Gaussian processes: inequalities, small ball probabilities and applications, in: C.R.Rao, D. Shanbhag (Eds.), Stochastic Processes: Theory and Methods, Handbook of Statistics, Vol. 19, North-Holland, Amsterdam, (2001), 533-597. 5. M.A. Lifshits, Asymptotic behavior of small ball probabilities, in: B. Grigelionis (Ed.), Pro- ceedings of the Seventh Vilnius Conference on Probability Theory and Mathematical Statistics, VSP/TEV. Vilnius (1999,), 453-468. 6. A.I. Martikainen, A.N. Frolov, J. Steinebach, On probabilities of small deviations for compound renewal processes, Theory Probab. Appl. 52 (2007), no. 2, 366-375. (Russian) 7. P. Embrechts, C. Klüppelberg, Some aspects of insurance mathematics, Theory Probab. Appl. 38 (1993), 374-416. (Russian) 8. A.N. Frolov, On probabilities of small deviations for compound Cox processes, Zapiski Nauchnyh seminarov POMI 339 (2006), 163-175. (Russian) 9. W. Feller, Introduction to probability theory and its applications. V. 2, Mir, Moscow, 1967. (Russian) 10. I.A. Ibragimov, Yu.V. Linnik, Independent and stationary dependent random variables, Nauka, Moscow, 1965. (Russian) '��������� � %���������� ��� %��������� ��� ������0��6 ���"����� � ���"����������� ������ � � ���� ������� � ��� ������0��6� #* 4�$� 7����� E-mail : Andrei.Frolov@pobox.spbu.ru
id nasplib_isofts_kiev_ua-123456789-4548
institution Digital Library of Periodicals of National Academy of Sciences of Ukraine
issn 0321-3900
language English
last_indexed 2025-12-07T17:03:40Z
publishDate 2008
publisher Інститут математики НАН України
record_format dspace
spelling Frolov, A.N.
2009-12-03T16:34:21Z
2009-12-03T16:34:21Z
2008
On asymptotic behaviour of probabilities of small deviations for compound Cox processes / A.N. Frolov // Theory of Stochastic Processes. — 2008. — Т. 14 (30), № 2. — С. 19–27. — Бібліогр.: 10 назв.— англ.
0321-3900
https://nasplib.isofts.kiev.ua/handle/123456789/4548
519.21
We derive logarithmic asymtotics for probabilities of small deviations for compound Cox processes in the space of trajectories. We find conditions under which these asymptotics are the same as those for sums of independent identically distributed random variables and homogeneous processes with independent increments. We show that if these conditions do not hold, the asymptotics of small deviations for compound Cox processes are quite different.
This article was partially supported by RFBR, grant 05–01–00486.
en
Інститут математики НАН України
On asymptotic behaviour of probabilities of small deviations for compound Cox processes
Article
published earlier
spellingShingle On asymptotic behaviour of probabilities of small deviations for compound Cox processes
Frolov, A.N.
title On asymptotic behaviour of probabilities of small deviations for compound Cox processes
title_full On asymptotic behaviour of probabilities of small deviations for compound Cox processes
title_fullStr On asymptotic behaviour of probabilities of small deviations for compound Cox processes
title_full_unstemmed On asymptotic behaviour of probabilities of small deviations for compound Cox processes
title_short On asymptotic behaviour of probabilities of small deviations for compound Cox processes
title_sort on asymptotic behaviour of probabilities of small deviations for compound cox processes
url https://nasplib.isofts.kiev.ua/handle/123456789/4548
work_keys_str_mv AT frolovan onasymptoticbehaviourofprobabilitiesofsmalldeviationsforcompoundcoxprocesses