A note on almost sure behavior of randomly weighted sums of φ-mixing random variables with φ-mixing weights

Similar documents
A CHARACTERIZATION OF ADDITIVE DERIVATIONS ON VON NEUMANN ALGEBRAS

Econ107 Applied Econometrics Topic 3: Classical Model (Studenmund, Chapter 4)

The Order Relation and Trace Inequalities for. Hermitian Operators

3.1 Expectation of Functions of Several Random Variables. )' be a k-dimensional discrete or continuous random vector, with joint PMF p (, E X E X1 E X

THE WEIGHTED WEAK TYPE INEQUALITY FOR THE STRONG MAXIMAL FUNCTION

STEINHAUS PROPERTY IN BANACH LATTICES

More metrics on cartesian products

arxiv: v1 [math.co] 1 Mar 2014

TAIL PROBABILITIES OF RANDOMLY WEIGHTED SUMS OF RANDOM VARIABLES WITH DOMINATED VARIATION

Dirichlet s Theorem In Arithmetic Progressions

Continuous Time Markov Chain

Lecture 17 : Stochastic Processes II

Math 426: Probability MWF 1pm, Gasson 310 Homework 4 Selected Solutions

Randić Energy and Randić Estrada Index of a Graph

Another converse of Jensen s inequality

PROCEEDINGS OF THE AMERICAN MATHEMATICAL SOCIETY Volume 125, Number 7, July 1997, Pages 2119{2125 S (97) THE STRONG OPEN SET CONDITION

General viscosity iterative method for a sequence of quasi-nonexpansive mappings

The Multiple Classical Linear Regression Model (CLRM): Specification and Assumptions. 1. Introduction

APPENDIX A Some Linear Algebra

Appendix B. Criterion of Riemann-Stieltjes Integrability

Restricted divisor sums

Uniqueness of Weak Solutions to the 3D Ginzburg- Landau Model for Superconductivity

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 12 10/21/2013. Martingale Concentration Inequalities and Applications

Affine transformations and convexity

Asymptotic Properties of the Jarque-Bera Test for Normality in General Autoregressions with a Deterministic Term

Perfect Competition and the Nash Bargaining Solution

A New Refinement of Jacobi Method for Solution of Linear System Equations AX=b

FACTORIZATION IN KRULL MONOIDS WITH INFINITE CLASS GROUP

Durban Watson for Testing the Lack-of-Fit of Polynomial Regression Models without Replications

Some basic inequalities. Definition. Let V be a vector space over the complex numbers. An inner product is given by a function, V V C

ON A DETERMINATION OF THE INITIAL FUNCTIONS FROM THE OBSERVED VALUES OF THE BOUNDARY FUNCTIONS FOR THE SECOND-ORDER HYPERBOLIC EQUATION

Y. Guo. A. Liu, T. Liu, Q. Ma UDC

Convergence of random processes

Using T.O.M to Estimate Parameter of distributions that have not Single Exponential Family

Markov chains. Definition of a CTMC: [2, page 381] is a continuous time, discrete value random process such that for an infinitesimal

UNIVERSITY OF TORONTO Faculty of Arts and Science. December 2005 Examinations STA437H1F/STA1005HF. Duration - 3 hours

Lecture 4: September 12

Games of Threats. Elon Kohlberg Abraham Neyman. Working Paper

Lecture 3. Ax x i a i. i i

Lecture 4. Instructor: Haipeng Luo

STAT 511 FINAL EXAM NAME Spring 2001

P exp(tx) = 1 + t 2k M 2k. k N

Errors for Linear Systems

Excess Error, Approximation Error, and Estimation Error

Asymptotics of the Solution of a Boundary Value. Problem for One-Characteristic Differential. Equation Degenerating into a Parabolic Equation

A Hybrid Variational Iteration Method for Blasius Equation

FREQUENCY DISTRIBUTIONS Page 1 of The idea of a frequency distribution for sets of observations will be introduced,

princeton univ. F 13 cos 521: Advanced Algorithm Design Lecture 3: Large deviations bounds and applications Lecturer: Sanjeev Arora

APPROXIMATE PRICES OF BASKET AND ASIAN OPTIONS DUPONT OLIVIER. Premia 14

Analysis of Discrete Time Queues (Section 4.6)

Foundations of Arithmetic

Strong Markov property: Same assertion holds for stopping times τ.

Lectures on Stochastic Stability. Sergey FOSS. Heriot-Watt University. Lecture 5. Monotonicity and Saturation Rule

2E Pattern Recognition Solutions to Introduction to Pattern Recognition, Chapter 2: Bayesian pattern classification

ANSWERS. Problem 1. and the moment generating function (mgf) by. defined for any real t. Use this to show that E( U) var( U)

First Year Examination Department of Statistics, University of Florida

Google PageRank with Stochastic Matrix

Introduction to Econometrics (3 rd Updated Edition, Global Edition) Solutions to Odd-Numbered End-of-Chapter Exercises: Chapter 13

Expected Value and Variance

Projective change between two Special (α, β)- Finsler Metrics

Lectures - Week 4 Matrix norms, Conditioning, Vector Spaces, Linear Independence, Spanning sets and Basis, Null space and Range of a Matrix

Notes on Frequency Estimation in Data Streams

Systems of Equations (SUR, GMM, and 3SLS)

Maximizing the number of nonnegative subsets

1. Inference on Regression Parameters a. Finding Mean, s.d and covariance amongst estimates. 2. Confidence Intervals and Working Hotelling Bands

Lecture 3: Probability Distributions

Math 217 Fall 2013 Homework 2 Solutions

j) = 1 (note sigma notation) ii. Continuous random variable (e.g. Normal distribution) 1. density function: f ( x) 0 and f ( x) dx = 1

Probability and Random Variable Primer

THE CHVÁTAL-ERDŐS CONDITION AND 2-FACTORS WITH A SPECIFIED NUMBER OF COMPONENTS

Random Partitions of Samples

Salmon: Lectures on partial differential equations. Consider the general linear, second-order PDE in the form. ,x 2

arxiv: v1 [math.co] 12 Sep 2014

Infinitely Split Nash Equilibrium Problems in Repeated Games

DIFFERENTIAL FORMS BRIAN OSSERMAN

k t+1 + c t A t k t, t=0

Supplement to Clustering with Statistical Error Control

Stanford University CS254: Computational Complexity Notes 7 Luca Trevisan January 29, Notes for Lecture 7

A Note on Bound for Jensen-Shannon Divergence by Jeffreys

Econ Statistical Properties of the OLS estimator. Sanjaya DeSilva

Supplementary material: Margin based PU Learning. Matrix Concentration Inequalities

THE FUNDAMENTAL THEOREM OF CALCULUS FOR MULTIDIMENSIONAL BANACH SPACE-VALUED HENSTOCK VECTOR INTEGRALS

COMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS

EXPONENTIAL ERGODICITY FOR SINGLE-BIRTH PROCESSES

TAIL BOUNDS FOR SUMS OF GEOMETRIC AND EXPONENTIAL VARIABLES

Eigenvalues of Random Graphs

Convexity preserving interpolation by splines of arbitrary degree

Self-complementing permutations of k-uniform hypergraphs

Volume 18 Figure 1. Notation 1. Notation 2. Observation 1. Remark 1. Remark 2. Remark 3. Remark 4. Remark 5. Remark 6. Theorem A [2]. Theorem B [2].

n α j x j = 0 j=1 has a nontrivial solution. Here A is the n k matrix whose jth column is the vector for all t j=0

Perron Vectors of an Irreducible Nonnegative Interval Matrix

Case Study of Markov Chains Ray-Knight Compactification

ON THE EXTENDED HAAGERUP TENSOR PRODUCT IN OPERATOR SPACES. 1. Introduction

10-801: Advanced Optimization and Randomized Methods Lecture 2: Convex functions (Jan 15, 2014)

SUPER PRINCIPAL FIBER BUNDLE WITH SUPER ACTION

See Book Chapter 11 2 nd Edition (Chapter 10 1 st Edition)

Additional Codes using Finite Difference Method. 1 HJB Equation for Consumption-Saving Problem Without Uncertainty

Finding Primitive Roots Pseudo-Deterministically

Linear, affine, and convex sets and hulls In the sequel, unless otherwise specified, X will denote a real vector space.

The equation of motion of a dynamical system is given by a set of differential equations. That is (1)

Transcription:

ACTA ET COMMENTATIONES UNIVERSITATIS TARTUENSIS DE MATHEMATICA Volume 7, Number 2, December 203 Avalable onlne at http://acutm.math.ut.ee A note on almost sure behavor of randomly weghted sums of φ-mxng random varables wth φ-mxng weghts Marcn Przystalsk Abstract. Randomly weghted sums play an mportant role n varous appled and theoretcal problems, e.g., n actuaral mathematcs or statstcs. The almost sure convergence of randomly weghted sums s usually studed under the assumpton that sequences are ndependent and dentcally dstrbuted. In ths note, we assume that both sequences are φ-mxng. Under some addtonal condtons, we prove a strong law of large numbers for sequences of randomly weghted sums.. Introducton Let {Y, } be a sequence of random varables defned on a probablty space (Ω, F, P ). Let Fj k be a σ-algebra generated by the random varables Y l, l j,..., k. Defne the φ-mxng coeffcent (unform mxng coeffcent) φ (m) sup { P (B A) P (B) }, where the supremum s taken over A F k, B F k+m n, P (A) 0, k n m. The unform mxng coeffcent was ntroduced ndependently by Rozanov and Volkonsk [] and Ibragmov [4]. Snce then many authors have studed sequences of φ-mxng random varables, and a number of useful results have been establshed. In [8], Nagaev proved probablty and maxmal nequaltes for φ-mxng random varables. The summablty of φ-mxng random varables was studed by Kesel n [5, 6], whereas strong laws of large numbers were obtaned, e.g., n [5, 6, 7, 2]. Receved May 2, 202. 200 Mathematcs Subject Classfcaton. 60F5. Key words and phrases. Randomly weghted sums, φ-mxng sequences of random varables, strong law of large numbers. http://dx.do.org/0.2097/acutm.203.7. 27

28 MARCIN PRZYSTALSKI Randomly weghted partal sums n A X play an mportant role n varous appled and theoretcal problems. In actuaral mathematcs, f X s regarded as the net loss wthn the tme perod of the company and each A as the dscount factor from tme to tme 0, then n A X can be nterpreted as the total dscounted amount of the net loss from tme 0 to tme n. For example, n the feld of queueng theory, the n A X can be used to represent the total output for customer beng served by n machnes. In statstcs, Arenal-Gutérez et al. [] obtaned a strong law of large numbers for the bootstrap mean, assumng that {X, } s a sequence of parwse ndependent and dentcally dstrbuted random varables. The latter was further generalzed by Rosalsky and Sreehar n [0]. In ths note, we study strong lmt theorems of randomly weghted partal sums n A X, assumng that both sequences are weakly dependent, whch generalze the results obtaned n [, 0]. In contrast to [], we assume that {X, } s φ-mxng. On the sequence of weghts {A, }, we assume that ths sequence s a sequence of postve, dentcally dstrbuted (.d.) φ-mxng random varables such that A and X are ndependent, for each. We establsh strong laws of large numbers for a non-dentcally dstrbuted sequence {X, } usng the noton of a regular cover, whch was ntroduced by Pruss n [9]. Defnton.. Let X,..., X n be random varables, and X be a random varable possbly defned on a dfferent probablty space. Then X,..., X n are sad to be a regular cover of X provded E (G (X)) E (G (X )), () n for any measurable functon G for whch both sdes make sense. 2. Techncal lemmas Let {Y, } be a sequence of φ-mxng random varables. Set S k k Y and M n max k n S k. Defne φ + (m) sup {P (B A) P (B)}, where the supremum s taken over A F k, B F k+m n, P (A) 0, k n m. In [8], t was ponted out that φ + (n) < φ (n). Assume that φ + () < and let δ > 0 satsfy the condton δ + φ + () <. Set ρ δ + φ + (). Let α be a number such that the followng condton s satsfed: P (2M n > α) < δ. Under the above notaton, Nagaev [8] proved the followng nequalty.

RANDOMLY WEIGHTED SUMS 29 Lemma 2.. For any p > 0 and 0 < ε < 6 such that s (ε) > p, EMn p < c (p) E Y p + c 2 (p) α p, where s (ε) log ρ/ log ( + ε), c (p) < 2p+ ε 3p+ B (p +, s (ε) p + ), ρ c 2 (p) < ρ ε p B (p +, s (ε) p) p +, and B (, ) s the Euler Beta functon. In the proof of the man result we wll need the followng lemma. Lemma 2.2. Let {A, A, } be a sequence of postve.d. varables wth EA p <, for some p 2. Then max n A n 0 a.s. random Proof. Note that the condton EA p < mples n P (A n > ɛn) <, for every ɛ > 0. Hence, by the Borel Cantell lemma we have that A n n 0 a.s. Thus, by Lemma n [3] we get the asserton. Throughout ths paper, C and C 2 always stand for postve constants whch may dffer from one place to another. 3. Man results Theorem 3.. Let {A, A, } be a sequence of postve.d. φ-mxng random varables wth EA p <, for some p 2, and let {X, } be a sequence of φ-mxng random varables that s ndependent of {A, A, }. Let X be a random varable, possbly defned on a dfferent probablty space, satsfyng condton (). Moreover, addtonally assume that EX n 0, for all n. Let b 0 0 and be an ncreasng sequence of postve numbers satsfyng n and bp n O (n). (2) b p n If s (ε) > p, for some 0 < ε < 6, and E X p <, then lm n n /p A X 0 a.s. (3)

30 MARCIN PRZYSTALSKI Proof. For n, set X X I ( X b ) and X X I ( X > b ). Then A X ( A X EX ) + A X + A EX. In order to show that ( n /p ) n A X 0 a.s., we only need to show that all terms above are o ( n /p ) a.s. Frst, we show that ( ( P A X EX ) ) > n /p ɛ <, (4) n for all ɛ > 0. Because X X I ( X b ) s also φ-mxng, by Theorem 5.2 n [2], we have that {A X, } s also φ-mxng. Hence, by Markov s nequalty and Lemma 2., we have that ( ( P A X EX ) ) >n /p ɛ n n C E n A (X EX ) p ɛ p n + C 2 C EA p n n I + I 2. E X EX p E X p + C 2 n Note that the second part of (2) ensures that I 2 <. Hence, t remans to show that I <. Because s ncreasng, we have that b, for all n, and I C C n n E X p E X p I ( X ).

RANDOMLY WEIGHTED SUMS 3 Let G (x) x p I [ x ]; then by the defnton of regular cover I C C where X XI ( X ). Further, I C C n n n E X p I ( X ) E X p, E X p C n n n b p P (b < X b ) E X p I [b < X b ] C b p P (b < X b ). (5) Then, by (5) and (2), n I C P (b < X b ) C P ( X > b ) C E X p <, b p and (4) holds. Thus, ( n /p ) n A (X EX ) converges completely to 0, whch mples that n A (X EX ) s o ( n /p ) a.s. Next, by the defnton of regular cover, condton E X p <, and (2), we have that P ( X n > ) P (b < X n b ) n n n n n P (b < X n b ) EI [b < X n b ] EI [b < X b ]

32 MARCIN PRZYSTALSKI P (b < X b ) E X p b p <. Hence, by the Borel Cantell lemma, n X s bounded a.s. By (2) and Lemma 2.2, t follows that n /p A X n /p max b A X n n n /p max n A X 0 a.s. n Fnally, by Markov s nequalty, Lemma 2., condton EA p <, and the defnton of regular cover, ( ) P A EX > n/p ɛ C EA p EX p + C2 n C n n E X p + C2 n n Hence, usng the same arguments as n the estmaton of I and I 2, we obtan that ( ) P A EX > n/p ɛ <, n for every ɛ > 0. Thus, ( n /p ) n A EX converges completely to 0, whch mples that n A EX s o ( n /p ) a.s. Ths completes the proof. In [9], t was ponted out that every.d. sequence of random varables satsfes the regular cover condton () wth X X. Thus, from Theorem 3. we have the followng corollary. Corollary 3.2. Let {A, A, } be a sequence of postve.d. φ-mxng random varables wth EA p <, for some p 2, and let {X, X, } be a sequence of.d. φ-mxng random varables that s ndependent of {A, A, }. Moreover, addtonally assume that EX n 0, for all n. Let b 0 0 and be an ncreasng sequence of postve numbers satsfyng (2). If s (ε) > p, for some 0 < ε < 6, and E X p <, then (3) holds. It s known that every sequence of ndependent and dentcally dstrbuted (..d.) random varables s φ-mxng wth φ (n) 0, for each n. Thus, from Theorem 3. we obtan the followng corollary..

RANDOMLY WEIGHTED SUMS 33 Corollary 3.3. Let {A, A, } be a sequence of postve..d. random varables wth EA p <, for some p 2, and let {X, } be a sequence of φ-mxng random varables that s ndependent of {A, A, }. Let X be a random varable, possbly defned on a dfferent probablty space, satsfyng condton (). Moreover, addtonally assume that EX n 0, for all n. Let b 0 0 and be an ncreasng sequence of postve numbers satsfyng (2). If s (ε) > p, for some 0 < ε < 6, and E X p <, then (3) holds. We conclude wth some remarks. Remark. Usng Theorem 5.2 n [2], under some addtonal condtons, one can obtan the counterpart of Theorem 3. for other mxng coeffcents. Remark 2. It should be stressed that the assumpton of ndependence of {A, } and {X, } n Theorem 3. s very crucal. Assumng only that both sequences are φ-mxng does not guarantee that {A X, } wll be φ-mxng (see [2, Theorem 5.2], and dscusson below the theorem). References [] E. Arenal-Gutérez, C. Matrán, and J. A. Cuesta-Albertos, On the uncondtonal strong law of large numbers for bootstrap mean, Statst. Probab. Lett. 27 (996), 49 60. [2] R. C. Bradley, Basc propertes of strong mxng condtons. A survey and some open questons, Probab. Surv. 2 (2005), 07 44. [3] J. H. J. Enmahl and A. Rosalsky, General weak laws of large numbers for bootstrap sample means, Stochastc Anal. Appl. 23 (2005), 853 869. [4] I. A. Ibragmov, Some lmt theorems for statonary processes, Theory Probab. Appl. 7 (962), 349 382. [5] R. Kesel, Strong laws and summablty for sequences of φ-mxng random varables n Banach spaces, Electron. Comm. Probab. 2 (997), 22 4. [6] R. Kesel, Summablty and strong laws of φ-mxng sequences, J. Theoret. Probab. (998), 209 224. [7] A. Kuczmaszewska, On the strong law of large numbers for φ-mxng and ρ-mxng random varables, Acta Math. Hungar. 32 (20), 74 89. [8] S. V. Nagaev, On probablty and moment nequaltes for dependent random varables, Theory Probab. Appl. 45 (2000), 52 60. [9] A. R. Pruss, Randomly sampled Remann sums and complete convergence n the law of large numbers for a case wthout dentcal dstrbutons, Proc. Amer. Math. Soc. 24 (996), 99 929. [0] A. Rosalsky and M. Sreehar, On the lmtng behavor of randomly weghted partal sums, Statst. Probab. Lett. 40 (998), 403 40. [] Y. A. Rozanov and V. A. Volkonsk, Some lmt theorems for random functon, Theory Probab. Appl. 4 (959), 86 207. [2] D. Q. Tuyen, A strong law of φ-mxng random varables, Perod. Math. Hungar. 38 (999), 3 36. Research Center for Cultvar Testng, 63-022 S lupa Welka, Poland E-mal address: marprzyst@gmal.com