Table Of Content

On several two-boundary problems for a particular class of Lévy processes T. Kadankova ∗and N. Veraverbeke Hasselt University, Belgium Key words: firstexit time; value of the overshoot; first entry time ; compound Poisson process with positive and negative jumps 7 0 Running head: Two-boundary problems for certain Lévy processes 0 2 Abstract n Severaltwo-boundaryproblemsaresolvedforaspecialLévyprocess: thePoissonprocess a J with anexponentialcomponent. The jumps ofthis processarecontrolledby ahomogeneous 1 Poisson process, the positive jump size distribution is arbitrary, while the distribution of 3 the negative jumps is exponential. Closed form expressions are obtained for the integral transformsofthejointdistributionofthefirstexittimefromanintervalandthevalueofthe ] R overshoot through boundaries at the first exit time. Also the joint distribution of the first entry time into the interval and the value of the process at this time instant are determined P . in terms of integral transforms. h t a m 1 Introduction [ 1 We assume that all random variables and stochastic processes are defined on (Ω, F,{Ft},P), v a filtered probability space, where the filtration {Ft} satisfies the usual conditions of right- 4 continuity and completion. A Lévy process is a F -adapted stochastic process {ξ(t); t ≥ 0} 2 which has independent and stationary increments and whose paths are right-continuous with 9 1 left limits [1]. Under the assumption that ξ(0) = 0 the Laplace transform of the process 0 {ξ(t); t ≥ 0} has the form E[e−pξ(t)] = etk(p), Rep = 0, where the function k(p) is called 7 the Laplace exponent and is given by the formula ([6]) 0 / h 1 1 ∞ px t k(p) = lnE[e−pξ(t)] = p2σ2−αp+ e−px−1+ Π(dx). (1) a t 2 1+x2 m Z−∞(cid:18) (cid:19) : Here α, σ ∈ R and Π(·) is a measure on the real line. The introduced process is a space v homogeneous, strong Markov process. Note, that the distribution of the first exit time from an i X interval plays a crucial role in applications and its knowledge also allows to solve a number of r a other two-boundary problems. Let us fix B > 0 and define the variable χ(y) = inf{t : y+ξ(t) ∈/ [0,B]}, y ∈ [0,B], the first exit time from the interval [0,B] by the process y + ξ(t). The random variable χ(y) is a Markov time and P [χ(y) < ∞] = 1 [6]. Exit from the interval [0,B] can take place either through the upper boundary B, or through the lower boundary 0. Introduce events: AB = {ω : ξ(χ(y)) > B}, i.e. the exit takes place through the upper boundary; A = {ω : ξ(χ(y)) < 0}, i. e. the exit takes place through the lower boundary. Define 0 X(y) = (ξ(χ(y))−B)I +(−ξ(χ(y)))I , P [AB +A ]= 1, AB A0 0 ∗ Hasselt University,Center for Statistics, Agoralaan, Building D,3590 Diepenbeek, Belgium, tel.: +32(0)11 26 82 97, e-mail: [email protected] 1 thevalueoftheovershootthroughoneoftheboundariesatthefirstexittime,where I = I (ω) A A is the indicator of the event A. The first two-boundary problem for Lévy processes with the Laplace exponent of the general form (1) has been solved by Gihman and Skorokhod ([6], p.306- 311). These authors have determined the joint distribution of {ξ−(t),ξ(t),ξ+(t)}, where ξ+(t) = sup ξ(u), ξ−(t) = inf ξ(u), t ≥ 0. For a spectrally positive Lévy process the joint dis- u≤t u≤t tribution of {χ(y),X(y)} has been studied by many authors among which Emery [5], Suprun and Shurenkov [19]. The first exit time for a spectrally one-sided Lévy process has been consid- eredbyBertoin [2], Pistorius[15], [16], Kyprianou[12]andothers. Kadankov andKadankova [8] have suggested another approach for determining the joint distribution of {χ(y),X(y)} for the Lévy process with Laplace exponent (1). Their method is based on application of one-boundary functionals {τx, Tx}, {τ , T }, x ≥ 0, where x x τx = inf{t : ξ(t) > x}, Tx = ξ(τx)−x, τ = inf{t : ξ(t) < −x}, T = −ξ(τ )−x. x x x Integral transforms of these joint distributions have been obtained in 60’s in papers of Pecher- skii and Rogozin [17],[13], Borovkov [3], Zolotarev [20]. Kadankov and Kadankova [8] have used probabilistic methods (the total probability law, space homogeneity and the strong Markov property of the process) to determine the integral transforms E[e−sχ(y);X(y) ∈ du,AB], E[e−sχ(y);X(y) ∈ du,A ] of the joint distribution of {χ(y),X(y)}. For a spectrally positive 0 Lévy process several two-boundary problems have been solved in [9]-[11]. In this paper we obtain the integral transforms of the distributions of a number of two- boundary functionals associated with the first exit time (Section 3) and the first entry time (Section 4) for an important particular class of Lévy processes described in Section 2. The advantage is that these are closed formulas for the transforms, with no recursions, typical for the general case. 2 The Poisson process with negative exponential exponent We now give a formal definition of the process which we consider. Let η ∈ (0,∞) be a positive randomvariable,and γ beanexponentialvariablewithparameter λ> 0 : P [γ > x]= e−λx, x≥ 0. Introduce the random variable ξ ∈ R by its distribution function F(x) =aexλI{x ≤ 0}+(a+(1−a)P [η ≤ x])I{x > 0}, a∈ (0,1), λ > 0. N(t) Consider a right-continuous compound Poisson process ξ(t) = ξ , t ≥ 0, where k=0 k {ξ ; k ≥ 1} are independent random variables identically distributed as ξ, ξ = 0, and k 0 P N(t) is a homogeneous Poisson process with intensity c > 0. Then its Laplace exponent is of the form ∞ p k(p) = c (e−xp−1)dF(x) = a +a (E[e−pη]−1), c > 0, Rep = 0, (2) 1 2 λ−p −∞ Z where a = ac, a = (1−a)c. Here and in the sequel we will call such process the Poisson 1 2 process with a negative exponential component. Note, that inter-arrival times of the jumps of the process {ξ(t); t ≥ 0} are exponentially distributed with parameter c. With probability 1−a thereoccur positive jumpsof size η, and with probability a there occur negative jumps of value γ that is exponentially distributed with parameter λ. The first term of (2) is the simplest case of a rational function, while the second term is nothing but the Laplace exponent ofamonotonePoissonprocesswithpositivejumpsofvalue η. Itiswellknownfact([7]), thatin this case the equation k(p)−s = 0, s > 0 has a unique root c(s) ∈ (0,λ), in the semi-plane Rep > 0. Denote by ν an exponentially distributed random variable with parameter s > 0, s independent of the process. The introduction of ν allows the following short-hand notation s 2 ∞ for the doubleLaplace transforms of the process, i.e. se−stE[e−pξ(t)]dt =E[e−pξ(νs)]. For the 0 double Laplace transforms of the processes ξ+(·), ξ−R(·) the following formulae hold E[e−pξ−(νs)] = c(s) λ−p , Rep ≤ 0, λ c(s)−p sλ E[e−pξ+(νs)]= (p−c(s))R(p,s), Rep ≥ 0, (3) c(s) where R(p,s)= a p+(p−λ)[s−a (E[e−pη]−1)] −1, Rep ≥ 0, p 6= c(s). (4) 1 2 Observethatthefun(cid:0)ction R(p,s) isanalyticinthesem(cid:1)i-plane Rep > c(s) and lim R(p,s)= p→∞ 0. Therefore, it allows a representation in the form of an absolutely convergent Laplace integral ([4]) ∞ R(p,s)= e−pxR (s)dx, Rep > c(s). (5) x Z0 We will call the function R (s), x ≥ 0 the resolvent of the Poisson process with a nega- x tive exponential component. We assume that R (s) = 0, for x < 0. Note, that R (s) = x 0 lim pR(p,s)= (c+s)−1, and the equalities (3) imply p→∞ c(s) λ s P [ξ−(ν )= 0] = , P [ξ+(ν )= 0] = . s s λ c(s) s+c The second formula of (3) yields c(s) 1 R(p,s)= E[e−pξ+(νs)], Rep > c(s). (6) sλ p−c(s) The functions 1 ∞ = e−u(p−c(s))du, Rep > c(s), p−c(s) Z0 ∞ E[e−pξ+(νs)]= e−updP [ξ+(ν )< u], Rep ≥ 0, s Z0 which enter the right-hand side of (6), are the Laplace transforms for Rep > c(s). Therefore, the original functions of the left-hand side and the right-hand side of (6) coincide, and c(s) x R (s) = ec(s)(x−u)dP [ξ+(ν ) < u], x≥ 0, (7) x s sλ Z−0 which is the resolvent representation of the Poisson process with a negative exponential component. Note, that the representation for the resolvent of the spectrally one-sided Lévy process similar to (7) was obtained by Suprunand Shurenkov [18],[19]. This representation implies that R (s), x≥ 0 is a positive, monotone, continuous, increasing function of an exponential order, x i.e. there exists 0 < A(s) < ∞ such that R (s) < A(s)exc(s), for all x≥ 0. Therefore, x ∞ R (s)e−αxdx < ∞, α> c(s). x Z0 Moreover, in the neighborhood of any x ≥ 0 the function R (s) has bounded variation. x Hence, the inversion formula is valid 1 α+i∞ R (s) = expR(p,s)dp, α > c(s). (8) x 2πi Zα−i∞ 3 The latter equality together with (5) determines the resolvent of the Poisson process with a negative exponential component. To derive the joint distribution of the first exit time and the value of the overshoot at the first exit time for a Poisson process with a negative exponential component, we apply a general theorem for the Lévy processes which has been proved in [8]. Before stating the theorem we mention the following results E[e−sτx−pTx] = E[e−pξ+(νs)] −1E[e−p(ξ+(νs)−x); ξ+(ν ) > x], Rep ≥ 0, s E[e−sτx−pTx]= (cid:16)E[epξ−(νs)] −(cid:17)1E[ep(ξ−(νs)+x); −ξ−(ν ) > x], Rep ≥ 0. (9) s (cid:16) (cid:17) The formulae (9) have been obtained by Pecherskii and Rogozin [13]. A simple proof of these equalities is given in [8]. After some calculations it follows from (3) and (9) that the integral transforms of the joint distributions of {τ , T }, {τx, Tx} of the Poisson process with a x x negative exponential component satisfy the equalities E[e−sτx; T ∈ du] = (λ−c(s))e−xc(s)e−λudu = E[e−sτx]P [γ ∈ du], (10) x ∞ 1 p+z−c(s) R(p+z,s) e−pxE[e−sτx−zξ(τx)]dx = 1− , Rep > 0, Rez ≥ 0. p z−c(s) R(z,s) Z0 (cid:18) (cid:19) Thefirstequality of (10) yields that τ and T are independent. Moreover, for all x≥ 0 the x x value of the overshoot through the lower level T is exponentially distributed with parameter x λ. This feature characterizes the Poisson process with a negative exponential component. Now we state the main results on two-sided exit problems. 3 The first exit from an interval We now derive the joint distribution of the first exit time and the value of the overshoot at the first exit. The following result is true for the general Lévy processes ([8]), and for convenience it is stated here as a lemma. Lemma 1. Let {ξ(t); t ≥ 0}, ξ(0) = 0 be a real-valued Lévy process whose Laplace exponent is given by (1), B > 0, y ∈ [0,B], x= B −y. Let χ(y)= inf{t : y+ξ(t) ∈/ [0,B]}, X(y) = (ξ(χ(y))−B)I +(−ξ(χ(y)))I AB A0 be, respectively, the first exit time from the interval [0,B] by the process y+ξ(t) and the value of the overshoot through the boundary at the first exit time. For s> 0, the Laplace transforms of the joint distribution of {χ(y),X(y)} satisfy the following equations ∞ E[e−sχ(y);X(y) ∈ du,AB]= fs(x,du)+ fs(x,dv)Ks(v,du), + + + Z0 ∞ E[e−sχ(y);X(y) ∈ du,A ] = fs(y,du)+ fs(y,dv)Ks(v,du), (11) 0 − − − Z0 where ∞ fs(x,du) = E[e−sτx;Tx ∈ du]− E[e−sτy;T ∈ dv]E[e−sτv+B;Tv+B ∈ du], + y Z0 ∞ fs(y,du) = E[e−sτy;T ∈ du]− E[e−sτx;Tx ∈ dv]E[e−sτv+B;T ∈ du]; − y v+B Z0 ∞ and Ks(v,du) = K(n)(v,du,s), v ≥ 0 are the seriesof the successive iterations of kernels ± ± n=1 K±(v,du,s). ThesPe kernels are given by ∞ K (v,du,s) = E[e−sτv+B;T ∈ dl]E[e−sτl+B;Tl+B ∈ du], + v+B Z0 ∞ K−(v,du,s) = E[e−sτv+B;Tv+B ∈ dl]E[e−sτl+B;Tl+B ∈ du], (12) Z0 4 and their successive iterations (n ∈ N = {1,2,...}) are defined by ∞ (1) (n+1) (n) K (v,du,s) =K (v,du,s), K (v,du,s) = K (v,dl,s)K (l,du,s). (13) ± ± ± ± ± Z0 We apply now the formulae of Lemma 1 for the case when the underlying process is the Poisson process with an exponentially distributed negative component. Theorem 1. Let {ξ(t); t ≥ 0}, ξ(0) = 0 be a Poisson process with a negative exponential component whose Laplace exponent is given by (2), B > 0, y ∈ [0,B], x = B−y. Let χ(y)= inf{t : y+ξ(t) ∈/ [0,B]}, X(y) = (ξ(χ(y))−B)I +(−ξ(χ(y)))I AB A0 be, respectively, the first exit time from the interval and the value of overshoot through one of the boundaries. Then for s > 0, 1) the integral transforms of the joint distribution of {χ(y), X(y)} satisfy the following equations E[e−sχ(y);X(y) ∈du,A ]= e−λu(λ−c(s))e−yc(s) 1−E[e−sτx−c(s)ξ(τx)] K(s)−1du, (14) 0 E[e−sχ(y);X(y) ∈du,AB]= E[e−sτx;Tx ∈ du]−E(cid:16)[e−sχ(y);A ]E[e−sτγ+B(cid:17);Tγ+B ∈ du], 0 where K(s)= 1−E[e−sτB]E[e−sτγ+B−c(s)Tγ+B] ∞ E[e−sτγ+B−c(s)Tγ+B ] = λ e−λuE[e−sτu+B−c(s)Tu+B]du. Z0 In particular, c(s) E[e−sχ(y);A ]= 1− e−yc(s) 1−E[e−sτx−c(s)ξ(τx)] K(s)−1, (15) 0 λ (cid:18) (cid:19) (cid:16) (cid:17) E[e−sχ(y);AB]=E[e−sτx]−E[e−sχ(y);A ]E[e−sτγ+B]; 0 2) the Laplace transforms of the random variable χ(y) satisfy the following representations R (s) 1 R (s) E[e−sχ(y);X(y) ∈ du,A ] =e−λ(u+B) x du, E[e−sχ(y);A ]= e−λB x , 0 Rˆ (λ,s) 0 λ Rˆ (λ,s) B B R (s) 1 E[e−sχ(y);AB] =1− x e−λB +sλSˆ (λ,s) +sλS (s), Rˆ (λ,s) λ B x B (cid:20) (cid:21) ∞ R (s) e−stP[χ(y) > t]dt =λ x Sˆ (λ,s)−λS (s), (16) Rˆ (λ,s) B x Z0 B where R (s), x ≥ 0 is the resolvent of the process, defined by (5), (8); x x ∞ ∞ S (s) = R (s)du, Rˆ (λ,s) = e−λuR (s)du, Sˆ (λ,s) = e−λuS (s)du. x u B u B u Z0 ZB ZB Proof. For the Poisson process with a negative exponential component, equalities of Lemma 1 take a simplified form. Using the equalities (10) and the defining formulae (12) for the kernels K (v,du,s) we obtain ± c(s) K (v,du,s) = 1− e−c(s)(v+B)E[e−sτγ+B;Tγ+B ∈ du], + λ (cid:18) (cid:19) K (v,du,s) = e−λu(λ−c(s))e−c(s)BE[e−sτv+B−c(s)Tv+B]du, − 5 where γ is an exponentially distributed random variable with the parameter λ, independent of the process under the consideration. Using these equalities, the method of mathematical induction and the formulae (13), we obtain the successive iterations K(n)(v,du,s), n ∈ N of ± the kernels K (v,du,s) : ± K(n)(v,du,s) =E[e−sτv+B−c(s)Tv+B] E[e−sτB] n E[e−sτγ+B−c(s)Tγ+B] n−1 λe−λudu, − K(n)(v,du,s) =e−vc(s) E[e−sτB] n (cid:0)E[e−sτγ+B(cid:1)−c(cid:16)(s)Tγ+B ] n−1E[e−sτγ(cid:17)+B;Tγ+B ∈ du]. + (cid:16) (cid:17) (cid:0) (cid:1) The series Ks(v,du) of the successive iterations K(n)(v,du,s) are nothing but geometric ± ± series, and their sums are given by ∞ Ks(v,du) = K(n)(v,du,s) = E[e−sτv+B−c(s)Tv+B]E[e−sτB] K(s)−1λe−λudu, − − n=1 X ∞ Ks(v,du) = K(n)(v,du,s) = e−vc(s)E[e−sτB]K(s)−1 E[e−sτγ+B;Tγ+B ∈du]. + + n=1 X Substituting the obtained expressions for Ks(v,du) into (11) yields the formulae (14) of The- ± orem 1. Integrating the formulae (14) with respect to u ∈ R leads to formula (15) of the + theorem. Now, utilizing the definition of the resolvent (5), (8) and the equalities (10) we derive the resolvent representation for the functions E[e−sτx−c(s)ξ(τx)], E[e−sτx] : E[e−sτx−c(s)ξ(τx)]= 1−e−xc(s) R (s)r(c(s),s), x sλ E[e−sτx]= 1− R (s)+sλS (s), x x c(s) where x d S (s)= R (s)du, r(c(s),s) = R(p,s)−1 . x u dp Z0 (cid:12)p=c(s) (cid:12) Substituting these expressions into (15), we obtain the representation(cid:12)s (16) of Theorem 1. (cid:12) 4 The first entry time into an interval The knowledge of the joint distribution of {χ(y), X(y)} allows to solve another two-boundary problem, namely to determine the integral transforms of the joint distribution of the first entry time into the fixed interval by the Lévy process and the value of the process at this time. We obtain the result in Theorem 2 below for the general Lévy processes. In Theorem 3 closed expressions for the integral transforms in case of a Poisson process with a negative exponential exponent are given. Theorem 2. Let {ξ(t); t ≥ 0}, ξ(0) = 0 be a Lévy process whose Laplace exponent is given def by (1), B > 0, χ(y) = 0 for y ∈/ [0,B]. Let χ(y)= inf{t > χ(y) : y+ξ(t)∈ [0,B]}, X(y) = y+ξ(χ(y)) ∈ [0,B], y ∈R be, respectively, the first entry time into the interval [0,B] by the process y +ξ(t) and the value of the process at this time. For s > 0, the integral transforms of the joint distribution of 6 {χ(y),X(y)}, y ∈ R satisfy the following equations ∞ bv(du,s) d=ef E[e−sχ(v+B);X(v+B)∈ du] = Qs (v,dl)E[e−sτl;B −T ∈ du] + l Z0 ∞ ∞ + Qs (v,dl) E[e−sτl;T −B ∈ dν]E[e−sτν;Tν ∈ du], v > 0, + l Z0 Z0 ∞ b (du,s) d=ef E[e−sχ(−v);X(−v) ∈ du] = Qs (v,dl)E[e−sτl;Tl ∈ du] (17) v − Z0 ∞ ∞ + Qs (v,dl) E[e−sτl;Tl−B ∈dν]E[e−sτν;B −T ∈ du], v > 0, − ν Z0 Z0 ∞ b(y,du,s) d=ef E[e−sχ(y);X(y) ∈ du] = E[e−sχ(y);X(y) ∈ dv,AB]bv(du,s) Z0 ∞ + E[e−sχ(y);X(y) ∈ dv,A ]b (du,s), y ∈ [0,B], 0 v Z0 where δ(x), x∈ R is the delta function and ∞ Qs (v,du) = δ(v−u)du+ Q(n)(v,du,s), v > 0. (18) ± ± n=1 X The functions Q(n)(v,du,s), n ∈ N are defined by ± ∞ (1) (n+1) (n) Q (v,du,s) = Q (v,du,s), Q (v,du,s) = Q (v,dl,s)Q (l,du,s); (19) ± ± ± ± ± Z0 and they are the successive iterations of the kernels Q (v,du,s), which are given by ± ∞ Q (v,du,s) = E[e−sτv;T −B ∈ dl]E[e−sτl;Tl−B ∈ du], + v Z0 ∞ Q−(v,du,s) = E[e−sτv;Tv −B ∈ dl]E[e−sτl;Tl −B ∈ du]. (20) Z0 Proof. For the functions bv(du,s), b (du,s), v > 0 according to the total probability law, v space homogeneity of the process and the fact that τ , τv are Markov times, the following v system of equations is valid ∞ bv(du,s) =E[e−sτv;B −T ∈ du]+ E[e−sτv;T −B ∈ dl]b (du,s), v v l Z0 ∞ b (du,s) =E[e−sτv;Tv ∈ du]+ E[e−sτv;Tv −B ∈ dl]bl(du,s). (21) v Z0 This system is similar to a system of linear equations with two variables. Substituting the expression for b (du,s) from the right-hand side of the second equation into the first equation v yields ∞ bv(du,s) = E[e−sτv;B −T ∈ du]+ E[e−sτv;T −B ∈ dl]E[e−sτl;Tl ∈ du] v v Z0 ∞ ∞ + E[e−sτv;T −B ∈ dl] E[e−sτl;Tl−B ∈ dν]bν(du,s). v Zl=0 Zν=0 Changing the order of integration in the third term of the second equation implies for the function bv(du,s), v > 0 ∞ bv(du,s) = Q (v,dν,s)bν(du,s) (22) + Z0 ∞ +E[e−sτv;B −T ∈ du]+ E[e−sτv;T −B ∈ dl]E[e−sτl;Tl ∈du], v v Z0 7 which is a linear integral equation with the following kernel ∞ Q (v,du,s) = E[e−sτv;T −B ∈ dl]E[e−sτl;Tl−B ∈du], v > 0. + v Z0 We now show, that for all v,u > 0, s> s > 0 this kernel enjoys the following property 0 Q (v,du,s) < λ, λ= E[e−sτB]E[e−sτB], s > 0. + 0 Indeed, for all s > 0 it follows from B E[e−sτv;Tv −B ∈ du] = E[e−sτv+B;Tv+B ∈ du]− E[e−sτv;Tv ∈ dl]E[e−sτB−l;TB−l ∈ du], Z0 that the following chain of inequalities holds E[e−sτv;Tv −B ∈ du]≤ E[e−sτv+B;Tv+B ∈ du] ≤ E[e−sτv+B] ≤ E[e−sτB]. Analogously we establish, that E[e−sτv;Tv −B ∈ du] ≤ E[e−sτv+B;Tv+B ∈ du] ≤ E[e−sτv+B] ≤ E[e−sτB]. Thesechainsoftheinequalitiesimplythefollowingestimationforthekernel Q (v,du,s), + for all v,u > 0, s > s > 0 0 ∞ Q (v,du,s) = E[e−sτv;T −B ∈ dl]E[e−sτl;Tl−B ∈ du] + v Z0 ≤ E[e−sτB]E[e−sτB]< λ= E[e−s0τB]E[e−s0τB], s > 0. 0 This bound and the method of mathematical induction yield that the successive iterations (n) Q (v,du,s) (19) of the kernels Q (v,du,s) for all v,u > 0, s > s > 0 obey the inequal- + + 0 ∞ ity Q(n+1)(v,du,s) < λn+1,n ∈ N. Therefore,theseriesofsuccessiveiterations Q(n)(v,du,s) < + + n=1 λ(1−λ)−1 converges uniformly for all v,u > 0, s > s > 0. Utilizing nowPthe method of 0 successive iterations ([14], p. 33) to solve the integral equation (22) yields the first equality of the theorem. The second equality of the theorem can be verified analogously. It is not difficult to establish the third equality of the theorem using the total probability law and the fact that χ(y) is the Markov time. Denote by ∞ ms(du) = λe−λxE[e−sτx; Tx ∈du]dx, P(λ,du) = e−λB(ms(du)+λeλudu). γ γ Z0 Theorem 3. Let {ξ(t); t ≥ 0}, ξ(0) = 0 be a Poisson process with a negative exponential def component as specified above, B > 0, χ(y) = 0, for y ∈/ [0,B], and let χ(y) = inf{t > χ(y) : y+ξ(t) ∈ [0,B]}, X(y) = y+ξ(χ(y)) ∈ [0,B], y ∈ R be,respectively, the first entry time into the interval [0,B] by the process y+ξ(t) and the value of the process at this time. The integral transforms of the joint distribution of {χ(y),X(y)}, y ∈ R for s > 0 satisfy the following equations c(s) bv(du,s) = e−vc(s) 1− T(s)−1P(λ,du), v > 0, λ (cid:18) (cid:19) c(s) b (du,s) = ms(du)+eBc(s) 1− Tˆs(c(s))T(s)−1P(λ,du), v > 0, v v λ v (cid:18) (cid:19) c(s) R (s) e−B(λ−c(s)) b(y,du,s) = 1− T(s)−1 ec(s)(B−y) − B−y P(λ,du) (cid:18) λ (cid:19) " RˆB(λ,s) λ−c(s) # 1 R (s) + B−y (T(s)−1−1)P(λ,du), y ∈ [0,B], (23) λ Rˆ (λ,s) B 8 where ms(du) = E[e−sτx; Tx ∈ du], Tˆs(c(s)) = E[e−sτx−c(s)Tx; Tx > B], x≥ 0, x x ∞ c(s) Tˆs(c(s)) = λ e−λxTˆs(c(s))dx, T(s)= 1− 1− Tˆs(c(s))e−B(λ−c(s)). γ x λ γ Z0 (cid:18) (cid:19) Proof. We apply now the equalities (17) of Theorem 2 to obtain the formulae (23). For this we have to calculate for the Poisson process with a negative component the kernels Q (v,dl,s), ± and the successive iterations Q(n)(v,dl,s), n ∈ N, and the series Qs(v,dl). Utilizing the ± ± defining formula of the kernels (21) and the formulae (10) yields c(s) Q (v,dl,s) = e−vc(s) 1− e−λBE[e−sτγ;Tγ −B ∈ dl], v > 0, + λ (cid:18) (cid:19) c(s) Q (v,dl,s) = Tˆs(c(s))e−B(λ−c(s)) 1− λe−λldl, v > 0, (24) − v λ (cid:18) (cid:19) where Tˆs(c(s)) = E[e−sτx−c(s)Tx;Tx > B], x ≥ 0. Using the defining formula (19) for the x successive iterations and the method of mathematical induction it follows from (24) that for n ∈ N, c(s) n−1 Q(n)(v,dl,s) = e−vc(s) 1− e−λB T˜s(c(s)) E[e−sτγ;Tγ −B ∈ dl],, + λ γ (cid:18) (cid:19) (cid:16) (cid:17) c(s) n−1 Q(n)(v,dl,s) = Tˆs(c(s))e−B(λ−c(s)) 1− T˜s(c(s)) λe−λldl, − v λ γ (cid:18) (cid:19) (cid:16) (cid:17) where ∞ T˜s(c(s)) = e−B(λ−c(s))(λ−c(s)) e−λxTˆs(c(s))dx. γ x Z0 Theseries Qs(v,dl) of thesuccessive iterations Q(n)(v,dl,s) (see(18)) arejustthegeometric ± ± series, and their sums are given by c(s) Qs(v,dl) = δ(v−l)dl+e−vc(s) 1− e−λBT(s)−1E[e−sτγ; Tγ −B ∈ dl], v > 0, + λ (cid:18) (cid:19) c(s) Qs(v,dl) = δ(v−l)dl+Tˆs(c(s))e−B(λ−c(s)) 1− T(s)−1λe−λldl, v > 0, − v λ (cid:18) (cid:19) where T(s) = 1−T˜s(c(s)). Substituting in the equalities (17) of Theorem 2 the expressions γ for the functions Qs(v,dl), and the expressions for the functions E[e−sχ(y); X(y) ∈ dv, AB], ± E[e−sχ(y); X(y) ∈ dv, A ], which are given by the formulae of Theorem 1, we obtain the 0 formulae (23) of Theorem 3. Acknowledgements. The authors acknowledge support of the Belgian Federal Science Policy. (Interuniversity Attraction Pole Programme P6/STADEC). References [1] Bertoin, J. (1996). Lévy processes, Cambridge University Press. [2] Bertoin, J. (1997). Exponential decay and ergodicity of completely assymetric Lévy process in a finite interval. Ann.Appl. Probab, 7, 156-169. [3] Borovkov, A.A. (1976). Stochastic processes in queueing theory, Springer Verlag. 9 [4] Ditkin, V.A., Prudnikov, A.P. (1966). Operational calculus, Moscow, Russian edition. [5] Emery, D.J. (1973). Exit problem for a spectrally positive process. Adv. Appl. Prob. 5, 498–520. [6] Gihman, I.I., Skorokhod, A. V. (1975). Theory of stochastic processes, Vol. 2, Springer Verlag, translated from the Russian by S. Kotz. [7] Bratiychuk, N.S., Gusak D.V. (1990). Boundary problems for processes with independent increments, Kiev, Naukova Dumka, in Russian. [8] Kadankov, V. F., Kadankova, T. V. (2005). On the distribution of the first exit time from an interval and the value of overshoot through the boudaries for processes with independent increments and random walks. Ukr. Math. J. 10 (57), 1359-1384. [9] Kadankov,V.F.Kadankova,T.V.(2004).Onthedisributionofdurationofstayinaninterval ofthesemi-continuousprocesswithindependentincrements.Random Oper. andStoch. Equa. (ROSE) 12(4), 365–388. [10] Kadankova, T.V. (2003). On the distribution of the number of the intersections of a fixed interval by the semi-continuous process with independentincrements. Theor. of Stoch. Proc. 1-2, 73–81. [11] Kadankova, T.V. (2004). On the joint distribution of supremum, infimum and the magni- tude of a process with independent increments. Theor. Prob. and Math. Statist. 70, 54–62. [12] Kyprianou, A.E. (2003). A martingale review of some fluctuation theory for spectrally negative Lévy processes, Research report, Utrecht University. [13] Pecherskii, E.A., Rogozin B.A.(1969). Onjointdistributionsofrandomvariablesassosiated with fluctuations of a process with independent increments. Theor. Prob. and its Appl. 14, 410–423. [14] Petrovskii,I.G.(1965).Lecturesonthetheoryofintegralequations,Moscow,Nauka,Russian edition. [15] Pistorius, M.R. (2004). A potential theoretical review of some exit problems of spectrally negative Lévy processes. Séminaire de Probabilités, 38, 30–41. [16] Pistorius, M. R. (2004). On exit and ergodicity of the spectrally negative Lévy process reflected in its infimum. J. Theor. Prob. 17, 183–220. [17] Rogozin, B.A. (1966). On distributions of functionals related to boundary problems for processes with independent increments. Theor. Prob. and its Appl. 11(4), 656-670. [18] Suprun, V.N. (1976). Ruin problem and the resolvent of a terminating process with independent increments. Ukr. Math. J. 28(1), 53–61, (English transl.). [19] Suprun, V.N., Shurenkov, V.M. (1976). On the resolvent of a process with independent increments terminating at the moment when it hits the negative real semiaxis. In Studies in the Theory of Stochastic processes, Institute of Mathematics, Academy of Sciences of UKrSSR, Kiev, 170-174. [20] Zolotarev, V. M. (1964). The first passage time of a level and the behaviour at infinity for a class of processes with independent increments. Theor. Prob. and its Appl. 9(4), 653–664. 10