ASYMPTOTIC BEHAVIOR OF FINITE-TIME RUIN PROBABILITY IN A BY-CLAIM RISK MODEL WITH CONSTANT INTEREST RATE

Similar documents
Asymptotic Behavior of Finite-Time Ruin Probability in a By-Claim Risk Model with Constant Interest Rate

Series of New Information Divergences, Properties and Corresponding Series of Metric Spaces

Dr. Junchao Xia Center of Biophysics and Computational Biology. Fall /21/2016 1/23

Mathematical Statistics. Chapter VIII Sampling Distributions and the Central Limit Theorem

8. Queueing systems. Contents. Simple teletraffic model. Pure queueing system

Control Systems (Lecture note #6)

CHAPTER: 3 INVERSE EXPONENTIAL DISTRIBUTION: DIFFERENT METHOD OF ESTIMATIONS

Ruin Probability in a Generalized Risk Process under Rates of Interest with Homogenous Markov Chain Claims

Reliability analysis of time - dependent stress - strength system when the number of cycles follows binomial distribution

Chap 2: Reliability and Availability Models

Inference on Curved Poisson Distribution Using its Statistical Curvature

A NOVEL DIFFERENCE EQUATION REPRESENTATION FOR AUTOREGRESSIVE TIME SERIES

Almost unbiased exponential estimator for the finite population mean

Chapter 4. Continuous Time Markov Chains. Babita Goyal

Lecture 12: Introduction to nonlinear optics II.

Gauge Theories. Elementary Particle Physics Strong Interaction Fenomenology. Diego Bettoni Academic year

3.4 Properties of the Stress Tensor

Least squares and motion. Nuno Vasconcelos ECE Department, UCSD

Almost Unbiased Exponential Estimator for the Finite Population Mean

MECE 3320 Measurements & Instrumentation. Static and Dynamic Characteristics of Signals

Improvement of the Reliability of a Series-Parallel System Subject to Modified Weibull Distribution with Fuzzy Parameters

The Poisson Process Properties of the Poisson Process

Reliability of time dependent stress-strength system for various distributions

Existence of Nonoscillatory Solutions for a Class of N-order Neutral Differential Systems

On nonnegative integer-valued Lévy processes and applications in probabilistic number theory and inventory policies

On the Possible Coding Principles of DNA & I Ching

Algorithms to Solve Singularly Perturbed Volterra Integral Equations

Improved Exponential Estimator for Population Variance Using Two Auxiliary Variables

Two-Dimensional Quantum Harmonic Oscillator

Total Prime Graph. Abstract: We introduce a new type of labeling known as Total Prime Labeling. Graphs which admit a Total Prime labeling are

Major: All Engineering Majors. Authors: Autar Kaw, Luke Snyder

Consider a system of 2 simultaneous first order linear equations

On the Class of New Better than Used. of Life Distributions

Anouncements. Conjugate Gradients. Steepest Descent. Outline. Steepest Descent. Steepest Descent

Let's revisit conditional probability, where the event M is expressed in terms of the random variable. P Ax x x = =

Phys Nov. 3, 2017 Today s Topics. Continue Chapter 2: Electromagnetic Theory, Photons, and Light Reading for Next Time

Extinction risk depends strongly on factors contributing to stochasticity

(1) Cov(, ) E[( E( ))( E( ))]

Periodic Solutions of Periodic Delay Lotka Volterra Equations and Systems

Improved Exponential Estimator for Population Variance Using Two Auxiliary Variables

COMPLEX NUMBERS AND ELEMENTARY FUNCTIONS OF COMPLEX VARIABLES

Survival Analysis for Randomized Clinical Trials II Cox Regression. Ziad Taib Biostatistics AstraZeneca February 26, 2008

Mellin Transform Method for the Valuation of the American Power Put Option with Non-Dividend and Dividend Yields

On the Existence and uniqueness for solution of system Fractional Differential Equations

Introduction to logistic regression

Asymptotic Behavior of Solutions of Nonlinear Delay Differential Equations With Impulse

Quantum Harmonic Oscillator

Poisson Arrival Process

Summary: Solving a Homogeneous System of Two Linear First Order Equations in Two Unknowns

The Variance-Covariance Matrix

LECTURE 6 TRANSFORMATION OF RANDOM VARIABLES

Independent Domination in Line Graphs

CHAPTER Let "a" denote an acceptable power supply Let "f","m","c" denote a supply with a functional, minor, or cosmetic error, respectively.

Repeated Trials: We perform both experiments. Our space now is: Hence: We now can define a Cartesian Product Space.

Advanced Queueing Theory. M/G/1 Queueing Systems

Aotomorphic Functions And Fermat s Last Theorem(4)

Unbalanced Panel Data Models

Almost all Cayley Graphs Are Hamiltonian

14. Poisson Processes

k of the incident wave) will be greater t is too small to satisfy the required kinematics boundary condition, (19)

Some Probability Inequalities for Quadratic Forms of Negatively Dependent Subgaussian Random Variables

The probability of Riemann's hypothesis being true is. equal to 1. Yuyang Zhu 1

A Simple Representation of the Weighted Non-Central Chi-Square Distribution

Math Tricks. Basic Probability. x k. (Combination - number of ways to group r of n objects, order not important) (a is constant, 0 < r < 1)

Convolution of Generated Random Variable from. Exponential Distribution with Stabilizer Constant

Akpan s Algorithm to Determine State Transition Matrix and Solution to Differential Equations with Mixed Initial and Boundary Conditions

An N-Component Series Repairable System with Repairman Doing Other Work and Priority in Repair

Second Handout: The Measurement of Income Inequality: Basic Concepts

APPENDIX: STATISTICAL TOOLS

In 1991 Fermat s Last Theorem Has Been Proved

NHPP and S-Shaped Models for Testing the Software Failure Process

Institute of Actuaries of India

Special Curves of 4D Galilean Space

Poisson Arrival Process

Chapter 5 Transient Analysis

Introduction to Laplace Transforms October 25, 2017

CHAPTER 7. X and 2 = X

Complex Numbers. Prepared by: Prof. Sunil Department of Mathematics NIT Hamirpur (HP)

Least Squares Fitting (LSQF) with a complicated function Theexampleswehavelookedatsofarhavebeenlinearintheparameters

Different types of Domination in Intuitionistic Fuzzy Graph

STRUCTURAL FAULT DETECTION OF BRIDGES BASED ON LINEAR SYSTEM PARAMETER AND MTS METHOD

Department of Mathematics and Statistics Indian Institute of Technology Kanpur MSO202A/MSO202 Assignment 3 Solutions Introduction To Complex Analysis

t=0 t>0: + vr - i dvc Continuation

Computing OWA weights as relevance factors

1973 AP Calculus BC: Section I

(Reference: sections in Silberberg 5 th ed.)

8. Queueing systems lect08.ppt S Introduction to Teletraffic Theory - Fall

Delay-Dependent State Estimation for Time Delay Systems

Continuous Time Markov Chains

ON A RISK MODEL WITH A CONSTANT DIVIDEND AND DEBIT INTEREST

Binary Choice. Multiple Choice. LPM logit logistic regresion probit. Multinomial Logit

An Indian Journal FULL PAPER. Trade Science Inc. A stage-structured model of a single-species with density-dependent and birth pulses ABSTRACT

The real E-k diagram of Si is more complicated (indirect semiconductor). The bottom of E C and top of E V appear for different values of k.

Moments of Order Statistics from Nonidentically Distributed Three Parameters Beta typei and Erlang Truncated Exponential Variables

Power Spectrum Estimation of Stochastic Stationary Signals

Quantum Theory of Open Systems Based on Stochastic Differential Equations of Generalized Langevin (non-wiener) Type

Chapter 6. pn-junction diode: I-V characteristics

Continous system: differential equations

The Linear Regression Of Weighted Segments

Boyce/DiPrima 9 th ed, Ch 7.6: Complex Eigenvalues

Transcription:

Joural of Mahmacs ad Sascs 3: 339-357 4 ISSN: 549-3644 4 Scc Publcaos do:.3844/mssp.4.339.357 Publshd Ol 3 4 hp://www.hscpub.com/mss.oc ASYMPTOTIC BEHAVIOR OF FINITE-TIME RUIN PROBABILITY IN A BY-CLAIM RISK MODEL WITH CONSTANT INTEREST RATE L Wag Dparm of Mahmacs Th Uvrsy of Souhr Msssspp USA Rcvd 4-5-7; Rvsd 4-6-9; Accpd 4-8-4 ABSTRACT Ths sudy vsgas h ru probably of a rwal rs modl wh cosa rs ra ad by-clam pars. W assum ha h clam sz ad h r-arrval m sasfy a cra dpd srucur wh som addoal assumpos o hr dsrbuo fucos. I parcular w sudy h asympoc bhavor of PR δ > whch holds uformly a f rval. I hs way w sgfcaly d h L s rsul rgardg h parws srog quas-asympocally dpd radom varabls. Kywords: Rwal Rs Modl Subpoal Dsrbuo Uform Asympoc Parws Srog Quas-Asympocally Scc Publcaos. INTRODUCTION Rs hory plays a mpora rol facal mahmacs ad acuaral scc. I has b sudd by may domsc ad forg scholars. A vary of rs modls wh may spcal faurs also hav b vsgad by may paprs h lraur. Thy provd us dffr prspcvs o udrsad h rs modl ad rlad hors. Abou a cury ago Ludbrg 93 lad h foudao of acuaral rs modl hs Uppsala hss. Wars ad Paparadafylou 983 roducd dlay clams slms a dscr-m rs modl ad appld margal chqus o drv uppr bouds for ru probabls. I addo Yu al. 5 appld h probably-grag fucos o oba ru probabls for h compoud bomal modl wh dscr dlay m for by-clam. I addo Tag 5 vsgad a smpl asympoc formula for h ru probably of h rwal rs modl wh cosa rs forc ad rgularly varyg ald clams. Rcly Wg al. was rsd h al probably of h Posso sho os procss ad sablshd som asympoc formulas for h f 339 ad f ru probabls of a couous m rs modl. Corrspodg rsuls ca also b foud Ch ad Ng 7; Wag 8; L al. 9; Yag ad Wag ad som ohrs. Cosdr rwal rs modl wh h oal capal rsrv up o m dod by R δ as prssd by h followg qulbrum Equao.: δ = δ δ s δ c ds X { } = δ T Z { } T = R. whr dos h al capal of h surac compay δ> s h cosa rs ra ad c s h cosa gross prmum ra δ dos h oal capal afr m. W hav alrady ow h gral formula of h compoud-rs: δ m P = m

L Wag / Joural of Mahmacs ad Sascs 3: 339-357 4 P s h amou of moy accumulad afr yars s h prcpal amou s h umbr of yars h amou s dposd or borrowd for m s h umbr of ms h rs s compoudd pr yar ad δ s h aual ra of rs. Th w sar h compoudg mor ad mor h frqucy of compoudg clud yarly half-yarly quarrly mohly daly v f gos o fy ad h w drv ha: Scc Publcaos δ m P= lm m = δ m I hs rwal rs modl h drmsc lar fuco c dos h oal amou of prmums accumulad up o m. Th δ s c δ cɶ = c ds = dos oal capal δ grad by h prmums by h m. Cosdr h rs modl whch h clam szs ad h arrval ms of succssv clams fulfll h followg rqurms: Th clams szs: X cosu a squc of o cssarly dpd ad ogav radom varabls wh commo dsrbuo H: H = H = P{ } > for all > Th arrval ms of succssv clams ar: N = Y =. Th r-arrval m {Y ; } form a squc of radom varabls wh commo dsrbuo fuco V bu ar o cssarly dpd. Th arrval ms of succssv clams ca gra a rwal coug procss Equao.: = = { }. whr A s h dcaor fuco of a v A. Th N also dscrbs h oal umbr of clams f rval [ ]. Do h rwal fuco by λ = EN ad assum ha λ< for all << ha: [ ] P Y P Y Λ = > ] = = W assum ha {X ; } {Y ; } ad {c; } ar muually dpd 34 I our rs modl hr ar wo pars of muually dpd clams amly ma clams ad byclams Equao.3: δ δ = X { } = δ T Z { } T = R.3 W rfr Z as by-clams or dlayd clams h rwal rs modl. Thy ar dcally dsrbud wh commo dsrbuo F. Thy ar usually ducd by h ma clam wh som probably ad h occurrc of a by-clam may b dlayd dpdg o assocad ma clams amou. If h ma clam occurs a h h h by-clam occurs a h T. L T b h corrspodg dlay ms of h by-clam ad hy ar dcally dsrbud wh commo dsrbuo fuco G ad form a squc radom varabls whch ar ogav bu possbly grad a. I hs sudy w assum ha h {X Z ; } ad {T ; } ar muually dpd. Th clams ca produc h dpd fluc o ach ohr ad som addoal damags ad coss such as a orado hurrca ad havy ra-sorm ad so o. W df hm as by-clam modl. Hc our rwal modl wh by-clam pars ca br rflc h ruh. Th ulma ru probably h f m s dfd as Equao.4: Φ = P Rδ s < for soms.4 Ad h ru probably h f m rval [ ] s gv by Equao.5: Φ = P Rδ s < for som s.5 Th ru probably h f m mas h surr s capal falls blow zro h f m rval [] ha s h oal clam cds h al capal plus prmum com. W also vsga h asympoc bhavor of h ru probably h f m. Oc h capal s lss ha zro h ru occurs ad h compay wll barup.. PRELIMINARIES W frs roduc som oaos. Throughou h papr all lm rlaoshps ar for dg o f ulss ohrws sad. Df:

L Wag / Joural of Mahmacs ad Sascs 3: 339-357 4 Scc Publcaos = f lm a o b a / b = Furhrmor for wo posv bvara fucos a ad b Equao.: a a = b b f lm sup. Clarly h asympoc rlao a~b holds uformly for : Ad: a lm sup sup b a lm f f b As s h cas for may rc rfrcs h flds of rs hory w ar rsd ru probabls f rval udr h assumpo ha H s havyald. I parcular h grad-al dsrbuo of H s subpoal. For dsrbuo H o h - H = H s h al dsrbuo of h fuco H. W do h uppr ad lowr Mauszwsa d of H: log H J lm L H H L log H y = lm f for y > H log H J lm U H HU log H y = lm sup for y > H A d.f. H wh suppor s subpoal f for all : H lm = H whr h H dos h -fold covoluo of H. Tag 6. 34 Th class of subpoal dsrbuo fucos wll b dod by S. Th class of subpoal dsrbuo plays a crucal rol havy-ald dsrbuo. I h surac dusry pracors usually choos havy-ald radom varabls o modl larg clams so w hav cludd for a mor dald aalyss of hr proprs. A suffc codo for subpoaly f hr s a gr such ha: H lm H Th H S: Thrfor udr h hypohss of h formula w oba: H lm H Th class of domad varyg dsrbuo s dfd as: H y D = Hdf o : lm sup < for ay y > H y If w suppos H D h for ay > J H hr s wo posv cosas c ad d such ha wh y d: H y c H η y η W hav alrady courd mmbrs of h followg hr famls: Th dfo of h class of log-ald dsrbuo: H y L = Hdf o : lm sup = for ay y > H y W ca s ha a dsrbuo H L ad oly f hr ss a fuco l : [ [ such ha l l = o ad: H y H Holds uformly for all y l.

L Wag / Joural of Mahmacs ad Sascs 3: 339-357 4 Accordg o h dfo w may df a ll bggr dsrbuo H ERV o [ f hr ar som <α β< such ha: β H s H s α s lmf lmsup s forall s H H Whch dod by H ERV -α -β. If α = β w say ha H blogs o h rgular varao class ad wr H R- α. I s wll ow ha h followg propr cluso rlaoshp should hold for h dsrbuo of havy-al: Scc Publcaos a L D D R ERV α β Embrchs al. 997; Klupplbrg ad Sadmullr 998; Tag ad Tssashvl 3; L al. 9; Hao ad Tag 8 From h sudy of may cos ad lraurs w asly foud ha h rwal rs modl wh cosa rs ra maly volvs h dpd srucur bw h clam szs ad arrval ms of succssv clams; hs lms h usfulss of h obad rsuls o som. Howvr h roduco of dpd srucur o rs modls has capurd mor ad mor rsarchrs ao rc yars ad provds a spcal prspcv for h ru probabls hory. May prvous paprs hav alrady word o hs w opc for ampl Yag ad Wag ; Lu al. ; Wag al. 3 ad ohrs. Th dpd srucur also allows h udrlyg radom varabls o b posv or gav. Hc w smply summarz h curr corrspodg rsuls ad ma clar h rlaoshp bw hm. I s also cssary for our proof. W furhr d h sudy o h dpd cas ad g svral smlar rsuls abou h dpd radom varabls. W may df radom varabls {ξ } as Lowr Ngavly Dpd LND ad Uppr Ngavly Dpd UND f for ach ad all : = { ξ } ξ P P = = { ξ > } ξ > P P = If h squc ca sasfy boh h LND ad UND w ca am Ngavly Dpd ND srucur. Wh = h LND UND ad ND srucurs ar quval Lhma 966. 34 W say ha wo radom varabls {ξ } ar parws Ngavly Quadra Dpd NQD f for all posv grs ξ ad ξ ar NQD: P ξ ξ P ξ P ξ Or quvally: P ξ > ξ > P ξ > P ξ > Addoally w also amd h LND as h NLOD L al. 9 wh dffr oaos ad dffr formulas. W ca df ha {ξ } ar WUOD wdly uppr orha dpd. If hr ss a f ral squc {g U } sasfyg for ach ad for all : = { ξ > } U ξ > P g u P = W ca also df ha {ξ } ar WLOD wdly lowr orha dpd. If hr ss a f ral squc {g L } sasfyg for ach ad for all : = { ξ } L ξ > P g P = Accordgly w would l rmar ha f h {ξ } ca hold.9 ad. s also sad o b WOD wdly orha dpd Wag al. 3: Ad h w should b famlar wh som mpora proprs of h WUOD ad WLOD Frsly w suppos {ξ } sasfy h WLOD or WUOD If {f } ar o-dcrasg h {f ξ } ar sll WLOD or WUOD Ivrsly f {f } ar o-crasg h {f ξ } ar WUOD or WLOD Scodly f {ξ } ar ogav ad WUOD h for ach : ξ E g Eξ U = = I parcular f h {ξ } ar WUOD h for ach ad ay v>:

L Wag / Joural of Mahmacs ad Sascs 3: 339-357 4 Scc Publcaos { ξ } = E p v g Epvξ u I h followg w wll us h assumpos ha for ay ε> Equao. ad.3: g =. lm U ε Ad g =.3 lm L ε W df h squc of h ral valu radom varabls {ξ } as Parws Quas-Asympocally Idpd PQAI for ay : ξ ξ ξ ξ lm z P > z > z = W also df h squc of h ral valu radom varabls {ξ } as parws Srog Quas- Asympocally Idpd psqai for ay : ξ ξ lm z z P > z > z = W ls svral corrspodg rsuls ad rmar h mhods usd som paprs mod abov havly rly o h..d assumpo o h clam sz ad h arrval ms of succssv clams. By obsrvao ad aalyss w compar h advaags ad dsadvaags of abov dpd srucurs. Th dpd cas of WUOD ad WLOD ca allow som gavly ad posvly radom varabls. Th psqai srucur ca clud h WUOD ad PQAI. I addo wh h radom varabls ar ogav h wo srucurs of psqai ad PQAI radom varabls ar quval ad h psqai srucur s a mor gral dpd cas ha h WUOD. Th ND srucur s a rlavly wa codo h asympoc bhavor of h ru probably s o ssv o h ND srucur. Rsul Thorm of Ch ad Ng 7. Cosdr h rwal rs modl sco f h clam szs {X ; } ar parws ND wh commo dsrbuo H ERV h r-arrval ms Y ar..d radom varabls ad h {c } s a drmsc lar fuco ad h h asympoc for h ulma ru probably Φ: 343 Rsul Φ H δ dλ Thorm of L al. 9. Cosdr h rwal rs modl Sco f h clam szs {X ; } ar parws NQD wh commo dsrbuo H D h rarrval ms {Y ; } ar NLOD ad h{c } s a drmsc lar fuco. I parcular f H L ad J > w oba h Φ: H Rsul 3 Φ H δ dλ Thorm of Sh ad L 8. Cosdr h rwal rs modl f h clam szs {X ; } ar NOD radom varabls wh commo dsrbuo H L D h r-arrval ms h r-arrval ms {Y ; } ar..d wh commo poal dsrbuo {N } s a homogous Posso procss: Rsul 4 Φ δ H dλ Thorm. of Wag al. 3. Cosdr h rwal rs modl Sco. If h clam szs {X ; } ar WUOD wh commo dsrbuo H L D h r-arrval ms {Y ; } ar WLOD. Also holds h rlaos. ad.. Th for ay f T Λ h rlao.5 holds uformly for Λ [T] ad h w oba h quval form for h ΦT: Rsul 5 Φ H δ dλ Thorm. of Lu al.. Cosdr h rwal rs modl Sco. If h clam szs {X ; } ar UTAI wh commo dsrbuo H L D h r-arrval ms {Y ; } ar WLOD such ha h rlao. holds. Th for ay fd T Λ h: Φ H δ dλ

L Wag / Joural of Mahmacs ad Sascs 3: 339-357 4 Rsul 6 Thorm 3. of L 3. Cosdr h by-clam modl Sco assumg ha {X X ; } {θ ; } ad {T ; } ar muually dpd X Y X Y ar PQAI ad radom pars X Y X Y... ar dcally dsrbud. L dsrbud. L H ERV ad F also ERV h w oba: Φ Scc Publcaos δ H λ δ s F d G s dλ For h rs modls ad rsuls w may dscuss hm varous aspcs accordg o h movao of rsarch such as h gral rs modl or rwal rs modl dpd srucur or dpd srucur som commo havy-ald dsrbuo classs h cosa rs ra or o ad so o. By aalyss w foud ha h clam szs ad h r-arrval ms rsuls of L al. 9; Wag al. 3; L 3 sasfd h dffr dpd srucurs s a srogr rsrco ha h..d codo rsul of Ch ad Ng 7. Bu amog h dffr dpd srucurs w may hav dffr chocs dffr rs modls ad h lad o dffr rsuls such as Lu al. h rqurd boh h commo dsrbuo of clam szs ad r-arrval ms follow h rsco class bu may cass h auhor chos a mor mld codo ERV. Furhrmor rms of commo dsrbuo som paprs volvs a mor complcad cas L al. 9 papr ad Yag ad Wag h rmard h uppr ad lowr Mauszwsa d w also cosdr h uppr ad lowr Mauszwsa d h rwal rs modl. Bu Wag al. 3 papr h cacld h codo J H I parcular bacgroud sco w roducd h rlao. ad. Wag al. 3 cosdrd hm [Rsul 4] ad w wll dscuss hm our rwal rs modl. I addo h [Rsul ] o [Rsul 5] maly vsga h asympoc bhavor of ru probably f m ad h h [Rsul ] ad [Rsul 6] word o h formula of ulma ru probably rs modl. Grally spag h prmum fuco c s a gral sochasc procss bu som papr s assumd ha h c s a drmsc lar fuco such as L al. 9 ad Ch ad Ng 7. Furhrmor w do o always rqur δ> ad h r-arrval ms may o hav a poal dsrbuo bu mos cass w 344 df ha h δ s cosa rs ra somms δ yld ad h r-arrval ms may follow a commo poal dsrbuo. Fally w cosdr h N facor rs modl sco w df N o cosu a rwal coug procss bu h rsul of Sh ad L 8 h N s a homogous Posso procss whch follow h Posso dsrbuo wh assocad paramr λ. 3. MAIN RESULTS I hs sudy w sll vsga h rwal rs modl ad rqur h clam szs ad h r-arrval ms sasfy h psqai ad WLOD srucur. W ca g a srogr rsul udr mld assumpo whch h H ad F blog o L D. I addo h rs modl volvg by-clam pars ca also lad o a dffr rsul. So w hav h followg rsuls: Lmma 3. If P s a probably fuco of A ad A ar ay s h Equao 3.: = m m P A P A P A A = m 3. Proof W us mahmacal duco o prov h rlao 3.. Wh m = w ca asly draw h cocluso ha Equao 3.: P A P A 3. Assum ha s ru ha wh m =.. Equao 3.3: = P A P A P A A = 3.3 Wh m = w ca us h basc probably formula PA B = PAPB-PA B Equao 3.4: = = = A A P = A P A P = A A P A P = 3.4 Cosquly by duco assumpo Equao 3.5: = A = P A P = A A P P A P A A 3.5

L Wag / Joural of Mahmacs ad Sascs 3: 339-357 4 Thrfor combg h rlao from 3. o 3.5 w fally g ha Equao 3.6: = P A P A P A A 3.6 = Scc Publcaos Ths ds h proof of h Lmma 3.. Accordg o h Thorm Chapr of Kabaov al. 986 w hav: Lmma 3. Igrao by pars Suppos f ad g ar rgh couous o-dcrasg ad wh lf-had lm fucos o [a b] whr a < b <R. Th Equao 3.7: b g df = g b f b g a f a f dg 3.7 a Lmma 3.3 Cosdrg h rwal coug procss {N } dfd.. Suppos ha {Y } sasfy h WLOD srucur ad h also holds h. rlao. For ay T Λ ad ay γ> w oba ha Equao 3.8: γ lm sup Λ[T] λ EN {N > } 3.8 Proof S h proof of Lmma. Wag al. 3. Lmma 3.4 L {X ; } ad {Z ; } b h muual dpd squcs wh commo dsrbuo fucos H ad F blog o h class L D uformly for Λ[ T] = or. Th Equao 3.9 ad : P X δ { < } δ { < } P X > ± l / N P > / N = b a = / N < > = δ T { T } < > ± N = δ T Z { T } P Z l / Lmma 3.5 3.9 3. Udr h assumpo of Lmma 3.4 X ad Z b h muual dpd squcs wh commo dsrbuo fucos H ad F blog o h L D ad 345 boh sasfy h psqai srucur. For ay holds uformly Λ[T]: δ T P X T δ { < } Z { < } > l N = lm f f Λ[T] δ P X N Proof { < } > = δ T { T < } P Z > N = By h Lmma 3. ad h formula PA B PAPB w ca asly foud: δ δ T { < } { T < } P X Z > l N = δ { < } δ T { T < } P X > l N = Z > N = δ { < } = P X > l N = δ T { T < } P Z > N = P X δ { < } δ T { T < } δ { < } δ T { T < } δ { < } δ T { T < } > l N = Z > N = P X > l N = P Z > N = P X > l Z > N = From h las sp bcaus of h fac ha h dpd rlaoshp amog X Z ad w coclud: δ { < } P X > l Z > N = δ { < } = P X > l N = P Z > N = Bcaus h basc propry of probably ad h l s larg ough: Lmma 3.6 lm P Z > N = = L X ad Z b h muual dpd squcs wh commo dsrbuo fucos H ad F blog o h L D ad boh sa ad boh sasfy h psqai srucur. For ay holds uformly Λ [T] Equao 3.:

L Wag / Joural of Mahmacs ad Sascs 3: 339-357 4 lm δ δ T P X { } < Z { T } < > l N = δ T > = > = sup sup P X N P Z N δ Λ[T] { < } { T < } 3. For h covc of proof w df ha: J = P X Z T > l N = δ δ T { < } { < } δ δ T J = P X N P Z N A = B = { } < > = { T } < > = δ δ T { X { } < Z { T } < > l } δ δ { X { < } > l N = } { Z T { T } < > l N = } Proof Tha s o say: Scc Publcaos J lm sup sup Λ[T] J Frsly f s larg ough w cosdr h dfo of l ad h formula PA = PAB PAB c s obvously ru: δ J P X { < } Z { T < } > l δ { < } δ T { T < } δ P X { < } Z { T < } > l δ { < } δ T { δ T X > l N = Z > l N = X l Z δ T T < } l N = Followd by h abov qualy w apply h smpl formula ha P A = B P = B P B = ad Bool s qualy w fd: δ J P X { < } > l N = δ T { T < } P Z > l N = δ l P X { < } > δ T { T < } Z > l N = δ T l P Z { T < } > δ { < } X > l N = 346 Cosdrg h fac ha -δt ad -δ blog o h rval [ -δ ] w may coclud ha h s δ l δ T { } X < > Z { T } < > l δ l X { < } > Z > l ad h s δ T l δ Z { T } { } < > X < > l δ T l Z { T } < > X > l So w ca apply h commo probably formula f A B h PA<PB o g h followg prsso: δ J P X { < } > l N = δ T { T < } P Z > l N = δ l P X { < } > Z > l N = δ T l P Z { T < } > > = X l N By h dpd rlaoshp amog h X Z ad { T } w hav:??? < δ T { T < } J P X > l N = P Z > l N = δ l P X { < } > N = P Z > l δ T l P Z { T < } > = > N P X l By h propry of class D ad Lmma 3.4 w ca g:

L Wag / Joural of Mahmacs ad Sascs 3: 339-357 4 lm lm δ l P X { } < > N = Scc Publcaos δ { } < > = P X l N δ { < } P X > N = δ T l P Z { T } < > N = δ T { T < } δ T { T < } P Z > l N = P Z > N = Thus f s larg ough: l P X > N = P Z > l = δ { < } δ T P Z { T < } l > N = P X > l = Hc w may fd: J lm sup sup Λ[T] J Cosquly w hold h rlao 3.. Ths ds h proof of Lmma 3.6. Lmma 3.7 For h rwal rs modl roducd Rs Thory sco w hav: δ P X { } < > N = δ T { T } < P Z > N δ = H dλ F dg s d Proof δ s λ Frsly w should df h commo dsrbuo of as V Equao 3.: δ { } < > δ T P Z { T < } N P X N > = H F δ dv δ s dg s dv 3. 347 Accordg o h o-dcrasg codo Lmma 3. w rorgaz h rlao 3.3: δ { } < > N δ T P Z { T < } N P X > δ dv = H F δ s dg s dv δ V V H d V = V G G δ s V F d G s d δ V V H d V = V V G G H δ s F d G s dv δ s F G dv δ V V H V H V = δ H G G V dh V V δ F G dv δ s F G d V δ s Gs d F d V δ s F G d V V d H H V δ s s G d F dv δ F G d V δ δ W should ma clar ha V P = < so h s quval o h N ad < s quval o h N> w ca drv:

L Wag / Joural of Mahmacs ad Sascs 3: 339-357 4 Ad: = = V = P N = EN < - = = V = P N = EN < I h rs modl sco w hav alrady show ha EN s a o-dcrasg ad rgh couous fuco. Accordg o h Lmma 3. w ca asly oba ha: V = δ δ { } < > T P Z T < N = δ { } = d H H P X N > δ = V δ δ s Gs d F d V = δ δ F G V = = EN d H H EN G δ s s d F den δ F G EN EN δ { } < > T P Z { } T < N P X N = δ > δ δ s = H d λ F d G s d λ I ds h proof of Lmma 3.7. Our ma rsul s for h appromao of f ru probably of h rwal rs modl wh cosa rs ra. I h followg sco w wll gv proofs of som rlad Lmmas mod h followg sco. Lmma 3.8 Cosdr h rs modl Sco. Assum ha h {X Y ; }{ ; } ad {T ; } ar muually dpd. L clam sz {X ; } ad by-clam pars {Z ; } b psqai dpd srucur wh commo dsrbuo fucos H ad F blog o h L D ad h r-arrval m {Y ; } b WLOD wh commo dsrbuo V ad also hold h rlao. ad.. Th for ay fd T Λ holds uformly Λ[T]:..: Φ δ H d λ δ s F d G s dλ W should apply h grao by pars aga. W g: Scc Publcaos δ { } < > T P Z { T < } N P X N = δ > δ δ s λ d H Gs d F d λ δ δ H EN H d λ = = δ F G F Thrfor: EN EN δ s d G s d λ 348 Φ lm sup Λ[T] = δ H dλ δ s F d G s dλ Lmma 4. 4. PROOFS Cosdr h rs modl roducd Sco. L X b h squc of psqai srucur radom varabls wh a commo dsrbuo fuco of H ad Z b h squc of psqai srucur radom varabls wh a commo dsrbuo fuco of F. Th X ad Z ar H L D ad F L D for ay fd holds uformly Λ[T] Equao 4.:

L Wag / Joural of Mahmacs ad Sascs 3: 339-357 4 δ δ T P { } X { } < Z T < > N = = δ T P X { } { } < > N = P Z T < > N = = = δ 4. Suppos ha: ˆ δ δ T J = P { } X { } < Z T < > N = = ˆ δ δ T J = P X { } { } < > N = P Z T < > N = = = Whch s udrsood as Equao 4.: Jˆ lm sup Λ[T] = Jɶ Proof Scc Publcaos 4. W d o prov h wo quals Equao 4.3 ad 4.4: Jˆ lm sup Λ[T] Jɶ Jˆ lm f f Λ[T] Jɶ 4.3 4.4 Frsly w ar rady o prov h rlao 4.4 rcallg h dfo of h l sasfyg h.7 ad Lmma 3. w hav Equao 4.5: δ X { ˆ } < J P = = δ T Z { < } > T δ X { < } Z { T < } > l N = δ X { < } Z { T < } > l N = δ T δ T δ X { < } δ T Z { } < > T l By obsrvao w add h qualy: 4.5 349 δ X { < } δ T Z > T { < } l δ δ { < } T X Z { T < oh } > = l N Thus w g: δ X { < } P = δ T Z { < } > T δ δ { < } T X Z { T < } > = l N δ X { < } P T Z { T < } = δ δ { } δ T < T < > l X Z { } > l N = If w apply h basc probably formula PB A c = PB -PB A ad PAB = PA BPB w hav: Jˆ = δ δ { < } T X Z { T < P } > = l N δ X { < } P = δ T Z T δ δ { < } T X Z { T < } > = l N { < } l

L Wag / Joural of Mahmacs ad Sascs 3: 339-357 4 < δ P X { < } δ T δ Z > l X { T < } { < } δ T Z { } T < l N > = > = P X Z l J l J l J l 3 Bcaus of H ad F blog o h class L h dpd rlaoshp bw X ad Z ad h Lmma 3.5 w fd Equao 4.6: J l lm f f Λ[T] Jɶ J lm f f Λ[T] J 4.6 Thus w ca coclud ha h rlao 3.9 ca lad o h rlao 4.7. For J l cosdrg h dpd rlaoshp amog X Z ad N ad h dpd prcpl of probably PA A B PA B PA B w drv: δ δ T l { } X < Z { T } < J l P = < δ δ { } T X { } < Z T < > l N = δ l δ δ { } { } { } T P X X Z < < T < > l N = = < δ T l δ δ { } { } { } T P Z T < X < Z T < > l N = = < δ l δ l P X { < } X { } < > N = δ l δ { } T l P X { } < Z T < > N = δ T l δ l P Z { } { } T < X < > N = δ T l δ { } T l P Z T < Z { T } < > N = δ l δ l P X { } { } X < < > N = = < δ l δ { } T l P X { } < Z T < > N = δ T l δ l P Z { } { } T < X < > N = δ T l δ { } T l P Z { } T < Z T < > N = For h las four rm of rlao 4.8 w ca apply h formula PAB = PA BPB ad h dfo of psqai rspcvly: δ l δ l δ l sup sup P X { } { } > P { } > = Λ[ T] X X lm < < < δ l δ δ sup sup { } T l { } > P T l P X Z T { } > = Λ[ T] Z T lm < < < Scc Publcaos 35

L Wag / Joural of Mahmacs ad Sascs 3: 339-357 4 lm lm δ T l δ l δ l sup sup P Z { T < } { } > P { } > = X X Λ[ T] < < δ T l δ sup sup { } T l P Z T { } > Z T Λ[ T] < < δ P T l Z { T < } > = Fally J l also gos o zro. Ad h w ur o J 3 l. < < T < δ δ δ 3 { } { } T J l P X l X Z { } > l N = < Z l X P P X Z > l > = δ T δ { } T < { < } δ T Z { } T < l N δ δ P X { } < l X { } < < > l N = δ δ { } T P X { } < l Z T < l N δ T δ P Z { } { } T < l X < l N δ T δ { } T P Z { } T < l Z T < l N > = > = > = W ry o apply h dfo of psqai ad h codo of H L ad F L aga: sup sup P X l X > l = Λ[ T] δ δ { } { } lm < < δ δ T { } { } lm < < sup sup P X l Z T > l = Λ[ T] δ T δ { } { } lm < sup sup P Z T < l X > l = Λ[ T] lm sup sup P Z δ T δ T T l Z < T < > l = Λ T [ ] { } { } Cosquly J 3 l also gos o zro. Thus a combao of J l J l ad J 3 l ca hold h rlao 4.4. I h co of h abov proof w ca cou our proof of rlao 4.3. If w cosdr h followg oao rlao 4.3. w df: δ δ T A = X { < } = Z { T < } > = Z > { } { } < T < δ δ T B X l = W ca a h sam logc rasog of h gral cas of h probably ha PA = PAB PAB c o h followg rlao w fd Equao 4.7: Scc Publcaos 35

L Wag / Joural of Mahmacs ad Sascs 3: 339-357 4 ˆ δ δ T J = P X { } { } < = Z T < > = N δ δ T = X Z > = < l N < { } { } T δ δ T { } { } < T < P X Z > N = = X Z l { } { } N = T δ δ T = < < 4.7 Afr applyg h P = B P B AB = = o rlao 4.9 Thrfor w oba: ˆ δ δ T J P X Z > l N = = { } { } < < T δ δ T { } { } < T < P X Z > N = < X Z l = Afr h rasformao of h qualy from: < X Z l o X Z l < Ad addg o h s of : = W drv: Jˆ = Scc Publcaos δ δ T X { } < Z { T < } > δ δ T X { < } Z { T < P } > = l N δ X { < } δ T P { } = Z T < < l N = δ δ T X Z > { } { } = < N T < Bcaus of h fac ha: δ δ T { X { } { } } < Z T < > l δ { X { } } < Z > l Thrfor: 35 δ δ T X { < } Z { T < P } > = l N δ P X Z > l { < } I addo w all hav ow ha h Bool s qualy ca lad o h cocluso: δ X { < } P δ T l Z { } < < = T N P δ X { < } δ T Z { T < } < l N = Th w apply h smpl logc rasog ad D Morga rul w may fd: P δ X { < } δ T Z { T < } > l N = δ X { < } P δ T l Z { } T < > = N Hc w g:

L Wag / Joural of Mahmacs ad Sascs 3: 339-357 4 δ δ { < } { < } > ˆ δ { } < > = T l X Z T P X Z l P = δ δ T X Z > { } { } = < N < T Accordg o h Bool s qualy: Th w hav: δ δ T l P { } { } < X Z T < > = N δ δ T l P X { } { } < Z T < > = N δ δ T { } { } < T < Jˆ P X Z > l = = δ δ T l P X { } { } < Z T < > = N δ δ T X Z > { } { } = < N < T δ δ T T < > l P X Z { } { < } = δ l δ δ T P X { } { } { } < > X Z < > N = < T = δ δ T l δ T { } { } { } Z < > > = T < < X Z N T = δ < Z > l P X { } δ l δ P X { } { } < > X > N = < δ l δ T { } { } P X < > > = Z N < = T δ δ T l P Z { } { } < > > = T < X N δ T l δ { } T { } P Z T < > > = = 4 5 < Z N J l J l T By h H ad F blogg o h class L w should cosdr Lmma 3.6 for vry : J4 l lmsup sup lmsup sup Λ[T] Jɶ Λ[T] Jˆ lmsup sup Λ[T] Jɶ = δ P X { } < Z > l Jɶ Scc Publcaos 353

L Wag / Joural of Mahmacs ad Sascs 3: 339-357 4 For J 5 f h H ad F blog o h class L ad h propry of psqai dpd srucur w drv ha: δ l δ lmsup P X { } { } < > X > N = < = δ l δ lmsup T P X { } { } < > Z > N = T < = δ δ lmsup T l P Z { } { } T < > > X > N = < = δ δ lmsup T l T P Z { } { } T Z < T < N > > > = Thus combg h J 4 l ad J 5 l w hold h rlao 4.3. Th followg sco wll dscuss h asympoc bhavor of P Rδ > dal. Bcaus w ar o Scc Publcaos oly rsd h asympoc bhavor of P Rδ > volvd h dpd srucur of h clam sz ad h arrval ms of succssv clams bu also h dpd srucur h rwal rs modl. I ds h proof of Lmma 4.. Lmma 4. Assum ha h {X Y ; }{ ; } ad {T ; } ar muually dpd. Cosdr h rwal rs modl w also assum ha X ad Y o cssarly dpd. L clam sz {X ; } ad by-clam pars {Z ; } follows psqai dpd srucur wh commo dsrbuo fucos H L D ad F L D. Th r-arrval m {Y ; } wh commo dsrbuo V ar WLOD ad also suppos h rlao. ad. hold. Th for ay fd T Λ Equao 4.8: > δ δ λ P R H d δ s F d G s dλ Holds uformly Λ [T]. Proof: δ m = m P R > = δ δ T X { < } Z P = I I { } < > = T N 4.8 354 Frsly w dal wh h I accordg o h Lmma 4. holds uformly for Λ [T]. W go: I ~ m P X = = m = = δ { } < > N = δ T { } T < N P Z > = m δ P X { } N = = m < δ T Z { } T < N I3 I4 = > = P >> = = By usg h Lmma 3. ad h dpd rlaoshp bw X Z ad w ca chag h ad m m = = = = δ 3 P X = δ T P Z δ H d λ w ca g: I = > = { < } N > = = { T < } N δ s F d G s dλ Ad h for I 4 bcaus h dpd rlaoshp amog X Z ad ad H F D : δ { < } > = δ T P Z { } T < > = N δ = P X { } < > I4 P X N = m = = m = δ T P Z = { T < } > P N =

L Wag / Joural of Mahmacs ad Sascs 3: 339-357 4 > > = m P X P Z P N = = = H P N F = m = m P N = { N > m } E { N > m } = H EN F N By h codo H D ad Lmma 3.3 w drv: P > a E E N a / d Ad combg h rlao.6 w oba: F / v v c c F H / v H ad c c v I lm 4 m sup sup Λ[T] = δ H dλ δ s F d G s dλ Bfor w ur o h I w should roduc som basc probably hory rlad o I. Bcaus of h fac ha Py> P>/ Py>/ w g: P X Z P X > > / = = = P Z > / = W may also fd wo cosas c ad d such ha for γ J ad γ J udr h codo of H D ad > H > F rlao.6 m /d holds uformly for Λ[T]. Thrfor from orgal qualy I should yld: m < /d I H / P N = m < /d F / P N = P N > / d m < / d γ m < / d γ γ = γ {N > / d } cf ch P N = cf d P N EN γ γ {N > m } {N > m } ch EN EN Scodly h basc rul probably s ha ay v of probably should o grar ha : I Thus: Scc Publcaos P X Z > = = m < / d / d < P X Z > P = = = P X > / m < /d = N P Z > / P N = P N = = / d < Applyg h Marov s qualy ad ma corrspodg subsuo w drv: 355 Hc by h guarad codo of H ad F blogg o h class D ad applyg Lmma 3.3 o h followg qualy aga w oba: lm lm [ ] m Λ T F δ s F dλ δt δt Λ[ ] γ {N > m } m Λ T T T sup EN γ {N m } cf EN > m sup [ ] δ δ H λ F λ c lm H H F λ m EN δ H dλ lm sup m T γ {N > m } = d G s ch I

L Wag / Joural of Mahmacs ad Sascs 3: 339-357 4 Cosquly w compl h proof of Lmma 4.. Proof of Thorm 3.8 Followd by h L 3 approach w foud ha: ɶ δ Rδ = Rδ = δ s δ c ds X { } = δ T Z { } = T Th w rwr as: Scc Publcaos Rɶ δ = cɶ Rδ Cosdrg h ru probably.5 f m Rs Modl sco: Φ = P Rδ s < for som s Thrfor: δ Φ = P Rδ s < for som s I follows ha: δ < ɶ Φ δ > P R c P R From whch w ca s ha: δ δ ɶ c Rδ Rδ R δ I follows ha: δ P Rδ c P Rδ > Φ > δ W hav alrady prov h rlao 4. so w ca rwr : δ c δ δ ~ λ c P Rδ > H d δ δ δ δ s λ c F d G s d δ 356 Th for ay ε> du o h codo.6 prlmary sco w fd ha: ε δ δ ε λ c P Rδ > H d δ δ s F ε d G s d λ η δ H d λ δ F d G s d λ By h arbrarss of ε> follows ha: c > δ δ δ λ δ F d G s d λ P Rδ H d Ad h smlarly w ca also oba h rmag par: > δ y δ δ F d G s dy P R H dy Ths compls h proof of Thorm 3.8. 5. CONCLUSION Elghd by h rsuls of L 3; Wag al. 3 w obad som ovl rsuls rgardg h psqai ad WLOD radom varabls wh h class L D. Our ma rsuls cocrd h appromao for cosa rs ra ad by-clam modl. I addo w furhr prov h ma rsuls ad corrspodg assumpos. Th asympoc bhavor of P Rδ > s a y rol our proof par. Fally w apply h obad rsuls o a d of clam-dpd rs modl ad drv a mor prcs ad mor gral asympoc formula for ru probably f m. 6. ACKNOWLEDGEMENT Th rsarchr would l o ha som comms ad cosrucv suggsos of my advsor ad all h

L Wag / Joural of Mahmacs ad Sascs 3: 339-357 4 mmbrs of comm. Svral smulag dscussos allowd m o dvlop orgal das ad mprov my papr. 7. REFERENCE Ch Y. ad K.W. Ng 7. Th ru probably of h rwal modl wh cosa rs forc ad gavly dpd havy-ald clams. Isur. Mah. Eco. 4: 45-43. DOI:.6/.smahco.6.6.4 Embrchs P. C. Klupplbrg ad T. Mosch 997. Modllg Ermal Evs: For Isurac ad Fac. s Ed. Sprgr Scc ad Busss Mda Brl ISBN-: 3546938 pp: 645. Hao X. ad Q. Tag 8. A uform asympoc sma for dscoud aggrga clams wh subpoal als. Isur. Mah. Eco. 43: 6-. DOI:.6/.smahco.8.3.9 Klupplbrg C. ad U. Sadmullr 998. Ru probabls h prsc of havy-als ad rs ras. Scad. Acuar. J. 998: 49-58. DOI:.8/34638.998.4399 Lhma E.L. 966. Som cocps of dpdc. Aals Mah. Sa. 37: 37-53. L J. K. Wag ad Y. Wag 9. F-m ru probably wh NQD domad varyg-ald clams ad NLOD r-arrval ms. J. Sys. Sc. Compl. : 47-44. DOI:.7/s44-9-973-7 L J. 3. O parws quas-asympocally dpd radom varabls ad hr applcaos. Sa. Probab. L. 83: 8-87. DOI:.6/.spl.3.5.3 Lu X. Q. Gao ad Y. Wag. A o o a dpd rs modl wh cosa rs ra. Sa. Probab. L. 8: 77-7. DOI:.6/.spl...6 Ludbrg F. 93. Cofgurao appromad fro of h probably fuco. II. FBK orsargav collcv rss Almqvs ad Wsll Uppsala. Sh X. ad Z. L 8. Prcs larg dvaos for radomly wghd sums of gavly dpd radom varabls wh cossly varyg als. Sa. Probab. L. 78: 3-39. DOI:.6/.spl.8.6.7 Tag Q. ad G. Tssashvl 3. Radomly wghd sums of subpoal radom varabls wh applcao o ru hory. Erms 6: 7-88. DOI:.3/B:EXTR.378.959.57 Tag Q. 6. Issvy o gav dpd of h asympoc bhavor of prcs larg. Elcro J. Probab. : 7-. Tag Q. 5. Th f m ru probably of h compoud Posso modl wh cosa rs forc. J. Appld Pro. 4: 68-69. Wag D. 8. F-m ru probably wh havyald clams ad cosa rs ra. Soch. Modls 4: 4-57 8. DOI:.8/53634786898 Wag K. Y. Wag ad Q. Gao 3. Uform asympocs for h f-m ru probably of a dpd rs modl wh a cosa rs ra. Mhodol. Compu. Appld Probab. 5: 9-4. DOI:.7/s9--96-y Wg C. Y. Zhag ad S. Ta. Tal bhavor of posso sho os procsss udr havy-ald shocs ad acuaral applcaos. Mhodol. Compu. Appld Probab. 5: 655-68. DOI:.7/s9--974-3 Yag Y. ad Y. Wag. Asympocs for ru probably of som gavly dpd rs modls wh a cosa rs ra ad domadly-varyg-ald clams. Sa. Probab. L. 8: 4354. DOI:.6/.spl.9.9.3 Yu K. J. Guo ad K. Wag 5. O ulma ru a dlayd-clams rs modl. J. Appld Prob. 4: 63-74. DOI:.39/ap/38378 Wars H.R. ad A. Paparadafylou 983. Ru probabls allowg for dlay clams slm. Isur. Mah. Eco. 4: 3-. DOI:.6/67-66878595-8 Kabaov Y.M. R.S. Lpsr ad A.N. Shryav 986. O h varao dsac for probably masurs dfd o a flrd spac. Probab. Thory Rlad Flds 7: 9-35. DOI:.7/BF3667 Scc Publcaos 357