Precise Rates in Complete Moment Convergence for Negatively Associated Sequences

Similar documents
A central limit theorem for moving average process with negatively associated innovation

A Note on the Kolmogorov-Feller Weak Law of Large Numbers

Complete Convergence for Asymptotically Almost Negatively Associated Random Variables

Review Article Complete Convergence for Negatively Dependent Sequences of Random Variables

Convergence of random variables. (telegram style notes) P.J.C. Spreij

Mi-Hwa Ko and Tae-Sung Kim

HAJEK-RENYI-TYPE INEQUALITY FOR SOME NONMONOTONIC FUNCTIONS OF ASSOCIATED RANDOM VARIABLES

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.436J/15.085J Fall 2008 Lecture 19 11/17/2008 LAWS OF LARGE NUMBERS II THE STRONG LAW OF LARGE NUMBERS

Asymptotic distribution of products of sums of independent random variables

MAXIMAL INEQUALITIES AND STRONG LAW OF LARGE NUMBERS FOR AANA SEQUENCES

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 21 11/27/2013

Research Article On the Strong Laws for Weighted Sums of ρ -Mixing Random Variables

This section is optional.

LECTURE 8: ASYMPTOTICS I

7.1 Convergence of sequences of random variables

7.1 Convergence of sequences of random variables

Central limit theorem and almost sure central limit theorem for the product of some partial sums

Notes 5 : More on the a.s. convergence of sums

Research Article Moment Inequality for ϕ-mixing Sequences and Its Applications

1 = δ2 (0, ), Y Y n nδ. , T n = Y Y n n. ( U n,k + X ) ( f U n,k + Y ) n 2n f U n,k + θ Y ) 2 E X1 2 X1

1 Convergence in Probability and the Weak Law of Large Numbers

An almost sure invariance principle for trimmed sums of random vectors

Lecture 19: Convergence

Distribution of Random Samples & Limit theorems

Bull. Korean Math. Soc. 36 (1999), No. 3, pp. 451{457 THE STRONG CONSISTENCY OF NONLINEAR REGRESSION QUANTILES ESTIMATORS Seung Hoe Choi and Hae Kyung

Equivalent Conditions of Complete Convergence and Complete Moment Convergence for END Random Variables

1 Introduction, definitions and assumption

This exam contains 19 pages (including this cover page) and 10 questions. A Formulae sheet is provided with the exam.

Statistical Analysis on Uncertainty for Autocorrelated Measurements and its Applications to Key Comparisons

Advanced Stochastic Processes.

Sequences and Series of Functions

THE STRONG LAW OF LARGE NUMBERS FOR STATIONARY SEQUENCES

EE 4TM4: Digital Communications II Probability Theory

Probability 2 - Notes 10. Lemma. If X is a random variable and g(x) 0 for all x in the support of f X, then P(g(X) 1) E[g(X)].

Lecture 2: Concentration Bounds

Chapter 3. Strong convergence. 3.1 Definition of almost sure convergence

Research Article Strong and Weak Convergence for Asymptotically Almost Negatively Associated Random Variables

4. Partial Sums and the Central Limit Theorem

ECE 901 Lecture 12: Complexity Regularization and the Squared Loss

for all x ; ;x R. A ifiite sequece fx ; g is said to be ND if every fiite subset X ; ;X is ND. The coditios (.) ad (.3) are equivalet for =, but these

The standard deviation of the mean

Product measures, Tonelli s and Fubini s theorems For use in MAT3400/4400, autumn 2014 Nadia S. Larsen. Version of 13 October 2014.

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 6 9/23/2013. Brownian motion. Introduction

Berry-Esseen bounds for self-normalized martingales

Hyun-Chull Kim and Tae-Sung Kim

2.2. Central limit theorem.

Probability and Statistics

Lecture 3. Properties of Summary Statistics: Sampling Distribution

Regression with an Evaporating Logarithmic Trend

Notes 27 : Brownian motion: path properties

MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.265/15.070J Fall 2013 Lecture 3 9/11/2013. Large deviations Theory. Cramér s Theorem

Dimension-free PAC-Bayesian bounds for the estimation of the mean of a random vector

Law of the sum of Bernoulli random variables

<, if ε > 0 2nloglogn. =, if ε < 0.

ECE 330:541, Stochastic Signals and Systems Lecture Notes on Limit Theorems from Probability Fall 2002

On the convergence rates of Gladyshev s Hurst index estimator

Lecture 20: Multivariate convergence and the Central Limit Theorem

Estimation of the essential supremum of a regression function

The random version of Dvoretzky s theorem in l n

Gamma Distribution and Gamma Approximation

On Weak and Strong Convergence Theorems for a Finite Family of Nonself I-asymptotically Nonexpansive Mappings

Research Article On the Strong Convergence and Complete Convergence for Pairwise NQD Random Variables

Supplementary Material for Fast Stochastic AUC Maximization with O(1/n)-Convergence Rate

STA Object Data Analysis - A List of Projects. January 18, 2018

Probability and Statistics

STAT Homework 1 - Solutions

Rademacher Complexity

Definition 4.2. (a) A sequence {x n } in a Banach space X is a basis for X if. unique scalars a n (x) such that x = n. a n (x) x n. (4.

lim za n n = z lim a n n.

Lecture 7: Properties of Random Samples

Statistically Convergent Double Sequence Spaces in 2-Normed Spaces Defined by Orlicz Function

Self-normalized deviation inequalities with application to t-statistic

ST5215: Advanced Statistical Theory

MA Advanced Econometrics: Properties of Least Squares Estimators

A Weak Law of Large Numbers Under Weak Mixing

Kolmogorov-Smirnov type Tests for Local Gaussianity in High-Frequency Data

Comparison of Minimum Initial Capital with Investment and Non-investment Discrete Time Surplus Processes

Random Matrices with Blocks of Intermediate Scale Strongly Correlated Band Matrices

5 Birkhoff s Ergodic Theorem

Research Article Complete Convergence for Maximal Sums of Negatively Associated Random Variables

1 Lecture 2: Sequence, Series and power series (8/14/2012)

1 Introduction to reducing variance in Monte Carlo simulations

Econ 325/327 Notes on Sample Mean, Sample Proportion, Central Limit Theorem, Chi-square Distribution, Student s t distribution 1.

January 25, 2017 INTRODUCTION TO MATHEMATICAL STATISTICS

J. Stat. Appl. Pro. Lett. 2, No. 1, (2015) 15

Exercise 4.3 Use the Continuity Theorem to prove the Cramér-Wold Theorem, Theorem. (1) φ a X(1).

An Introduction to Randomized Algorithms

ON STATISTICAL CONVERGENCE AND STATISTICAL MONOTONICITY

Entropy Rates and Asymptotic Equipartition

Journal of Multivariate Analysis. Superefficient estimation of the marginals by exploiting knowledge on the copula

Singular Continuous Measures by Michael Pejic 5/14/10

COMPLETE CONVERGENCE AND COMPLETE MOMENT CONVERGENCE FOR ARRAYS OF ROWWISE END RANDOM VARIABLES

SDS 321: Introduction to Probability and Statistics

Mathematics 170B Selected HW Solutions.

Research Article Approximate Riesz Algebra-Valued Derivations

Information Theory and Statistics Lecture 4: Lempel-Ziv code

On the optimality of McLeish s conditions for the central limit theorem

Chapter 5. Inequalities. 5.1 The Markov and Chebyshev inequalities

The log-behavior of n p(n) and n p(n)/n

Lecture 8: Convergence of transformations and law of large numbers

Transcription:

Commuicatios of the Korea Statistical Society 29, Vol. 16, No. 5, 841 849 Precise Rates i Complete Momet Covergece for Negatively Associated Sequeces Dae-Hee Ryu 1,a a Departmet of Computer Sciece, ChugWoo Uiversity Abstract Let {X, 1} be a egatively associated sequece of idetically distributed radom variables with mea zeros ad positive fiite variaces. Set S = i=1 X i. Suppose that < σ 2 = EX 2 1 + 2 i=2 CovX 1, X i ) <. We prove that, if EX 2 1 log+ X 1 ) δ < for ay < δ 1, the log ) δ 1 =2 2 ES I 2 S ɛσ log ) = E N 2δ+2, δ where N is the stadard ormal radom variable. We also prove that if S is replaced by M = max 1 k S k, the the precise rate still holds. Some results i Fu ad Zhag 27) are improved to the complete momet case. Keywords: Precise rates, complete momet covergece, egatively associated, law of the logarithm. 1. Itroductio A fiite sequece of radom variables {X i, 1 i } is said to be egatively associatedna), if for every disjoit subsets A ad B of {1, 2,..., }, we have Cov f X i ; i A), gx j ; j B)), wheever f o R A ad g o R B are coordiatewise odecreasig fuctios ad the covariace exists. A ifiite sequece of radom variables is NA if every fiite subsequece is NA. The otio of NA was itroduced by Alam ad Saxea 1981). Joag-Dev ad Proscha 1983) showed that may well kow multivariate distributios possess the NA property. Some examples iclude: a) the multiomial, b) the covolutio of ulike multiomials, c) the multivariate hypergeometric distributio, d) the Dirichlet, e) the Dirichlet compoud multiomial, f) the egatively correlated ormal distributio, g) the permutatio distributio, h) the radom samplig without replacemet, ad i) the joit distributio of raks. Because of its wide applicatios i multivariate statistical aalysis ad system reliability, the otio of NA has received cosiderable attetio recetly. We refer to Joag-Dev ad Proscha 1983) for fudametal properties, Shao ad Su 1999) for the law of the iterated logarithm, Shao 2) for momet iequalities ad the maximal iequalities of the partial sum, Liag 2) for complete covergece, Kim et al. 21) for the estimatio of empirical distributio, Li ad Zhag 24) for complete momet covergece, Fu ad Zhag 27) for the precise rates of i the law of the logarithm ad Ko 29) for cetral limit theorem of a liear process based o the egatively associated process i a Hilbert space. Set S = i=1 X i ad deote log x = lx e). Whe {X, 1} is a sequece of i.i.d. radom variables Liu ad Li 26) proved as follows: Let {X, 1} be a sequece of i.i.d. radom variables. Suppose that EX 1 =, < EX1 2 = σ2 ad EX1 2 log + X 1 )δ <, 1.1) 1 Professor, Departmet of Computer Sciece, ChugWoo Uiversity, Chugam 351-71, Korea. E-mail: rdh@chugwoo.ac.kr

842 Dae-Hee Ryu for < δ 1. The log ) δ 1 =2 where N is the stadard ormal radom variable. Coversely, if 1.2) is true, the 1.1) holds. Let where 2 ρ) =: sup k 1 ES I 2 S ɛ log ) = σ2+2δ E N 2δ+2, 1.2) δ sup X L 2 Fk ) Y L 2 F + EXY EXEY sup, k+ ) VarX)VarY) F = σx i ; 1 i ) ad F + = σx i ; i ). The the sequece {X, 1} is said to be ρ-mixig, if ρ) as see e.g. Peligrad, 1987). Zhao 28) proved that 1.2) is true for ρ-mixig sequeces uder appropriate coditios as follows: Let {X, 1} be a strictly statioary sequece of ρ-mixig radom variables with EX 1 = ad EX1 2 <. Suppose that ES 2 lim = σ 2 >, ρ 2 q 2 ) <, 1.3) =1 for q > 2δ + 2 ad EX 2 1 log+ X 1 ) δ < for ay < δ 1. The log ) δ 1 =2 2 ES I 2 S ɛσ log ) = E N 2δ+2. 1.4) δ Furthermore, Zhao 28) proved that if S is replaced by M = max 1 k S k the the precise rate still holds as follows: Let {X, 1} be a strictly statioary sequece of ρ-mixig radom variables satisfyig above coditios. The, log ) δ 1 =2 2 EM 2 I M ɛσ log ) = 2E N 2δ+2 δ = 1). 1.5) 2 + 1) 2δ+2 The purpose of this paper is to show that, for egatively associated radom variables 1.4) ad 1.5) still hold uder appropriate coditios. 2. Prelimiaries We itroduce some prelimiary results which are eeded i provig the mai results. Lemma 1. Newma, 1984) Assume that {X, 1} is a strictly statioary sequece of egatively associated radom variables with EX 1 = ad EX1 2 <. If < σ2 = EX1 2 +2 i=2 CovX 1, X i ) <, the S σ D N, 1) as, 2.1) where D idicates covergece i distributio, N a stadard ormal distributio.

Precise Rates i Complete Momet Covergece for NA Sequeces 843 Lemma 2. Shao, 2) Let {X, 1} be a strictly statioary sequece of egatively associated radom variables with EX 1 = ad EX1 2 <. If < σ2 = EX1 2 + 2 i=2 CovX 1, X i ) <, the W W, where W is a stadard Wieer measure, W t) = S [t] /σ, t 1, ad meas weak covergece i D[, 1] with Skorohod topology. I particular, M σ D sup Wt), 2.2) t 1 where M = max 1 k S k, 1 ad {Wt); t } is a stadard Wieer process. Lemma 3. Shao, 2) Let {Y i, 1 i } be a sequece of NA radom variables with mea zeros ad fiite variaces. Deote S k = k i=1 Y i, 1 k, B = i=1 EYi 2. The, for ay u >, v >. ) ) ) { } u P max S k u 2P max Y k v + 4 exp u2 12v B + 4. 2.3) 1 k 1 k 8B 4uv + B ) Propositio 1. Zhao, 28) Suppose that N is a stadard ormal radom variable. The for ay < δ 1, +2 log ) δ P N ɛ log ) = E N 2δ+2 δ + 1. 2.4) Proof: For the proof see Theorem 1.3 i Huag ad Zag 25). =2 Propositio 2. Zhao, 28) Suppose that N is a stadard ormal radom variable. The for ay < δ 1, log ) δ 1 2xP N x ) = E N 2δ+2 δδ + 1). 2.5) =2 2 ɛ log Proof: See the proof of Propositio 5.1 i Liu ad Li 26). The followig result is similar to oe of Propositio 3.2 i Fu ad Zhag 27). Propositio 3. Let {X ; 1} be a egatively associated sequece of idetically distributed radom variables with EX 1 = ad EX1 2 <. Suppose that < σ2 = EX1 2 + 2 i=2 CovX 1, X i ) = 1, ad EX1 2log+ X 1 ) δ < for ay < δ 1. The +2 log ) δ P S ɛ log ) P N ɛ log ) =. 2.6) =2 Proof: Usig the stadard method, set Hɛ) = [expm/ɛ 2 )], where M > 4, < ɛ < 1/4. I fact, we get log ) δ P S ɛ log ) P N ɛ log ) =2 log ) δ = P S ɛ log ) P N ɛ log ) + log ) δ P S ɛ log ) P N ɛ log ) =: I + II. 2.7)

844 Dae-Hee Ryu By Lemma 3.4 i Fu ad Zhag 27) we have Obviously, we have, for the secod part II, II log ) δ P N ɛ log ) + +2 I =. 2.8) log ) δ P S ɛ log ) = III + IV. 2.9) Notice that Hɛ) 1 Hɛ) for M > 4 ad < ɛ < 1/4, a easy calculatio leads to uiformly with respect to < ɛ < 1/4. For IV, by Lemma 3 we have ɛ 2δ+2 III ɛ 2δ+2 log ) δ P N ɛ log ) C y 2δ+1 P N > y) as M, 2.1) M 2 lim lim M ɛ2δ+2 IV lim lim M ɛ2δ+2 log ) δ P max S k ɛ ) log =. 2.11) 1 k See Lemma 3.7 i Fu ad Zhag 27).) From 2.7) 2.11), 2.6) follows. 3. Results Propositio 4. Set Hɛ) = [expm/ɛ 2 )] ad let {X, 1} be a sequece of idetically distributed NA radom variables. If EX 2 1 log+ X 1 ) δ < for ay < δ 1. The we obtai log ) δ 1 2 ɛ log 2xP S x)dx 2xP N x ) dx =. 3.1) ɛ log Proof: Deote = sup x P S x) P N x). Assume that x = y + ɛ) log. By itegral formula ad trasformatio, it is eough to show that C 2 log ) δ 1 ɛ log 1 log ) δ 2xP S x) dx 2xP N x ) dx ɛ log 2y + ɛ) P S y + ɛ) log ) P N y + ɛ) log )

Precise Rates i Complete Momet Covergece for NA Sequeces 845 C 1 log ) δ [ 1/ log 1 4 2y + ɛ)p { N y + ɛ) log } 1 1/ log 4 + 2y + ɛ) P { S y + ɛ) log } P { N y + ɛ) log } + 2y + ɛ)p { S 1/ log 1 y + ɛ) log } ] 4 log )δ =: C Λ 1 + Λ 2 + Λ 3 ). 3.2) The estimates of Λ 1 ad Λ 2 are similar to those of Propositio 5.2 i Liu ad Li 26), so we omit them. It remais to estimate Λ 3, takig θ = 1/EX1 2, u = y + ɛ) log, v = ay + ɛ) log ad a = 1/{12δ + 2)} i Lemma 2.4, which yields 1 log ) δ Λ 3 C + C + C log ) δ 1/ log 1 4 4y + ɛ)p { X 1 > ay + ɛ) log } 1 log ) δ 8y + ɛ) exp 1/ log 1 4 { 1 log ) δ 8y + ɛ) 1/ log 1 4 { θ2 y + ɛ) 2 } log 8 /θ 2 4ay + ɛ) 2 log + /θ 2 } 1 12a =: Λ 4 + Λ 5 + Λ 6. 3.3) Note that {log Hɛ)} δ = M δ /ɛ 2δ ad EX1 2IX 1 ) as. By Toeplitz s lemma, it follows that ) ɛ 2δ Λ 4 ɛ 2δ Clog ) δ X1 E 4y + ɛ)i log y + ɛ 1/ log 1 4 a Cɛ 2δ ) CM δ 1 log Hɛ)) δ Observe that exp θ 2 /8 1/2 ɛ 2δ Λ 5 ɛ 2δ E X 1 2 I X 1 ) log ) δ 1 ) M δ 1 log Hɛ)) δ C a 2 E X 1 2 I X 1 ) log ) δ 1 as ɛ. 3.4) a 2 ) as. Usig Toeplitz s lemma agai, we have ) 32 C 1 log ) δ 1 exp θ 2 θ2 8 1 2 ) exp θ2 log ) δ 1 8 2 1 as ɛ. 3.5)

846 Dae-Hee Ryu Observe that δ+1)/2 as. Usig Toeplitz s lemma agai, we have { θ ɛ 2δ Λ 6 ɛ 2δ C 1 log ) δ 2 y + ɛ) 2 } δ+2) log 8y + ɛ) 1/ log 1 4 3δ + 2) { ɛ 2δ C 1 log ) δ θ 2 } δ+2) 3δ + 2) 8y + ɛ) { y + ɛ) 2 log } δ+2) 1/ log 4 1 ) CM δ 1 δ+1 2 log ) δ 1 log Hɛ) δ as ɛ. 3.6) Combiig 3.2) 3.6), cosequetly, we obtai 3.1). Propositio 5. Let {X, 1} be a sequece of idetically distributed NA radom variables. If EX1 2log+ X 1 ) δ < for ay < δ 1. The we obtai, we have log ) δ 1 2xP N x ) dx =, 3.7) 2 log ) δ 1 2 ɛ log ɛ log 2xP S x)dx =. 3.8) Proof: The proof of 3.7) is quite routie, we omit it. Applyig Lemma 2.4, the proof of 3.8) are exposed as follows: For a = 1/{12δ + 2)}, we have C C 2 log ) δ 1 ɛ log 2xP{ S x}dx 1 log ) δ 2x + ɛ)p { S x + ɛ) log } dx 1 log ) δ [ {2P X1 2x + ɛ) > ax + ɛ) log } { θ 2 x + ɛ) 2 } log + 4 exp + 4 { 4θ 2 ax + ɛ) 2 log } ] 1 12a dx 8 =: C 1 log ) δ 2x + ɛ)ii 1 + II 2 + II 3 )dx. 3.9) Recallig the momet coditio, it suffices to prove that 1 log ) δ 2x + ɛ)ii 1 dx C CE log ) δ E ɛ X 2 1 x ɛ 4xI X 1 ax log ) dx log X 1 log x δ 1 I X1 x ) I X 1 M ) dx

Precise Rates i Complete Momet Covergece for NA Sequeces 847 Hece, for II 1, we have CEX 2 1 log X 1 log ɛ δ I X1 M ) CEX 2 1 log X 1 ) δ + CEX 2 1 log ɛ δ <. 3.1) 1 log ) δ 2x + ɛ)ii 1 dx. Next, for II 2 θ 1 log ) δ 2x + ɛ)ii 2 dx C 1 log ) δ 2 x 2 ) log 8x exp dx ɛ 8 ) C 1 log ) δ 32 exp θ2 ɛ 2 ) log θ 2 log 8 ) 32 C 1 θ2 ɛ 2 θ 2 8 <. 3.11) Hece 1 log ) δ 2x + ɛ)ii 2 dx as ɛ. 3.12) Fially, for II 3 { θ 1 log ) δ 2x + ɛ)ii 3 dx C 1 log ) δ 2 x 2 } δ+2) log 8x dx ɛ 3δ + 2) ) { C 1 log ) 2 4 θ 2 } δ+2) ɛ 2δ 2. 3.13) δ + 1 3δ + 2) Hece 1 log ) δ 2x + ɛ)ii 3 dx Cɛ 2 4 δ + 1 Cɛ 2 uiformly with respect to ɛ. Hɛ) ) δ ) δ+2) 3 2 + 1 Hɛ) 1 log ) 2 dx xlog x) 2 CM 1 as M, Theorem 1. Let {X, 1} be a sequece of idetically distributed NA radom variables. EX1 2log+ X 1 ) δ < for ay < δ 1, the 1.4) holds. Proof: I fact, oe ca easily get log ) δ 1 ES I 2 S 2 ɛ log ) =1 = ɛ 2 log ) δ P S ɛ log ) log ) δ 1 + =1 =1 2 ɛ log 2xP S x) dx =: I 1 + I 2. 3.14) If

848 Dae-Hee Ryu Cosequetly to verity Theorem 1 we oly eed to cosider I 1 ad I 2, respectively. From Propositios 1 ad 3 we have I 1 = E N 2δ+2 δ + 1. 3.15) It follows from Propositios 4 ad 5 that log ) δ 1 2 2xP S x) dx 2xP N ) x dx =. 3.16) ɛ log ɛ log =1 Hece, by Propositio 2 ad 3.16) we also have I 2 = E N 2δ+2 δδ + 1), 3.17) which completes the proof together with 3.15). Next we will show that if S is replaced by M = max 1 k S k, the Theorem 1 still holds. Propositio 6. Fu ad Zhag, 27) Suppose that {Wt); t } is a stadard Wieer processbrowiia motio). The, for ay < δ 1, +2 log ) δ =1 Proof: Refer to Huag ad Zhag 25). P sup Ws) ɛ ) log = 2E N 2δ+2 δ 1 δ + 1 = 1) 2 + 1) 2δ+2. Propositio 7. Zhao, 28) Suppose that {Wt); t } is a stadard Wieer process. The, for ay < δ 1, log ) δ 1 =2 2 ɛ log 2xP sup Wt) x ) = 2E N 2δ+2 t 1 δδ + 1) = 1) 2 + 1) 2δ+2. Theorem 2. Let {X, 1} be a sequece of idetically distributed NA radom variables. EX1 2log+ X 1 ) δ < for ay < δ 1, the 1.5) holds. Proof: Note that Theorem 2 is the maximal versio of Theorem 1. Hece, if we make some modificatio of the proof of Theorem 1, Theorem 2 will follow. As i 3.14), ideed, it suffices to stu that log ) δ 1 EM I 2 M 2 ɛ log ) =2 = ɛ 2 log ) δ P M ɛ log ) log ) δ 1 + =2 =: I 3 + I 4. =2 2 ɛ log 2xP M x) dx To pave the way for the proofs of I 3 of I 4 alog the same lies as that of the proof of Theorem 1, together with Lemmas 2 ad 3 ad Propositios 6 ad 7. If

Precise Rates i Complete Momet Covergece for NA Sequeces 849 4. Cocludig Remark It is of iterest to show that the precise rate result i this kid of complete momet covergece also holds for movig average process. I the future stu, we ivestigate the precise rate of covergece i complete momet of movig average processes based NA radom variables by extedig Theorems 1 ad 2 to the movig average processes. Refereces Alam, K. ad Saxea, K. M. L. 1981). Positive depedece i multivariate distributios, Commuicatios i Statistics - Theory ad Methods, 1, 1183 1196. Joag-Dev, K. ad Proscha, F. 1983). Negative associatio of radom variables with applicatios, The Aals of Statistcis, 11, 286 295. Fu, K. A. ad Zhag, L. X. 27). Precise rates i the law of the logarithm for egatively associated radom variables, Computers & Mathematics with Applicatios, 54, 687 698. Huag, W. ad Zhag, L. X. 25). Precise rates i the law of the logarithm i the Hilbert space, Joural of Mathematical Aalysis ad Applicatios, 34, 734 758. Kim, T. S., Lee, S. W. ad Ko, M. H. 21). O the estimatio of empirical distributio fuctio for egatively associated processes, The Korea Commuicatio i Statistics, 8, 229 236. Ko, M. H. 29). A cetral limit theorem for the liear process i a Hilbert space uder egative associatio, Commuicatio of the Korea Statistical Society, 16, 687 696. Li, Y. X. ad Zhag, L. X. 24). Complete momet covergece of movig- average processes uder depedece assumptios, Statistics & Probability Letters, 7, 191 197. Liag, H. Y. 2). Complete covergece for weighted sums of egatively associated radom variables, Statistics & Probability Letters, 48, 317 325. Liu, W. D. ad Li, Z. Y. 26). Precise asymptotics for a ew kid of complete momet covergece, Statistics & Probability Letters, 76, 1787 1799. Newma, C. M. 1984). Asymptotic idepedece ad limit theorems for positively ad egatively depedet radom variables, Y. L. Tog, ed. Iequalities i Statistics ad Probability: Proceedigs of the Symposium o Iequalities i Statistics ad Probability, 5, 127 14. Peligrad, M. 1987). The covergece of momets i the cetral limit theorem for ρ-mixig sequece of radom variables, I Proceedigs of the America Mathematical Society, 11, 142 148. Shao, Q. M. ad Su, C. 1999). The law of the iterated logarithm for egatively associated radom variables, Stochastic Processes ad their Applicatios, 83, 139 148. Shao, Q. M. 2). A compariso theorem o momet iequalities betwee egatively associated ad idepedet radom variables, Joural of Theoretical Probability, 13, 343 356. Zhao, Y. 28). Precise rates i complete momet covergece for ρ-mixig sequeces, Joural of Mathematical Aalysis ad Applicatios, 339, 553 565. Received July 29; Accepted August 29