We present the Bernstein-type inequality for widely dependent random variables. By using the Bernstein-type inequality and the truncated method, we further study the strong consistency of estimator of fixed design regression model under widely dependent random variables, which generalizes the corresponding one of independent random variables. As an application, the strong consistency for the nearest neighbor estimator is obtained.
1. Introduction
Let {Xn,n≥1} be a sequence of random variables defined on a fixed probability space (Ω,ℱ,P). It is well known that the Bernstein-type inequality for the partial sum ∑i=1nXi plays an important role in probability limit theory and mathematical statistics. The main purpose of the paper is to present the Bernstein-type inequality, by which, we will further investigate the strong consistency for the estimator of nonparametric regression models based on widely dependent random variables.
1.1. Brief Review
Consider the following fixed design regression model:
(1)Yni=g(xni)+εni,i=1,2,…,n,
where xni are known fixed design points from A, where A⊂ℝp is a given compact set for some p≥1, g(·) is an unknown regression function defined on A, and εni are random errors. Assume that for each n≥1, (εn1,εn2,…,εnn) have the same distribution as (ε1,ε2,…,εn). As an estimator of g(·), the following weighted regression estimator will be considered:
(2)gn(x)=∑i=1nWni(x)Yni,x∈A⊂ℝp,
where Wni(x)=Wni(x;xn1,xn2,…,xnn), i=1,2,…,n are the weight functions.
The above estimator was first proposed by Georgiev [1] and subsequently has been studied by many authors. For instance, when εni are assumed to be independent, consistency and asymptotic normality have been studied by Georgiev and Greblicki [2], Georgiev [3] and Müller [4] among others. Results for the case when εni are dependent have also been studied by various authors in recent years. Fan [5] extended the work of Georgiev [3] and Müller [4] in the estimation of the regression model to the case where it forms an Lq-mixingale sequence for some 1≤q≤2. Roussas [6] discussed strong consistency and quadratic mean consistency for gn(x) under mixing conditions. Roussas et al. [7] established asymptotic normality of gn(x) assuming that the errors are from a strictly stationary stochastic process and satisfying the strong mixing condition. Tran et al. [8] discussed again asymptotic normality of gn(x) assuming that the errors form a linear time series, more precisely, a weakly stationary linear process based on a martingale difference sequence. Hu et al. [9] studied the asymptotic normality for double array sum of linear time series. Hu et al. [10] gave the mean consistency, complete consistency, and asymptotic normality of regression models with linear process errors. Liang and Jing [11] presented some asymptotic properties for estimates of nonparametric regression models based on negatively associated sequences. Yang et al. [12] generalized the results of Liang and Jing [11] for negatively associated sequences to the case of negatively orthant dependent sequences and obtained the strong consistency for the estimator of the nonparametric regression models based on negatively orthant dependent errors. Wang et al. [13] studied the complete consistency of the estimator of nonparametric regression models based on ρ~-mixing sequences, and so forth. The main purpose of this paper is to investigate the strong consistency for the estimator of the nonparametric regression models based on widely dependent random variables, which contains independent random variables, negatively associated random variables, negatively orthant dependent random variables, extended negatively orthant dependent random variables, and some positively dependent random variables as specials cases. For more details about the strong consistency for the estimator of g(·), Ren and Chen [14] obtained the strong consistency for the least squares estimator of β and the nonparametric estimator of g(·) based on negatively associated samples, Baek and Liang [15] studied the strong consistency for the weighted least squares estimator of β and nonparametric estimator of g(·) in a semi-parametric model under negatively associated samples, which extended the corresponding one on independent random error settings, Liang et al. [16] also studied the strong consistency in a in semiparametric model for a linear process with negatively associated innovations and established the convergence rate, they also pointed out that their results on nonparametric estimator of g(·) can attain the optimal convergence rate, and so forth.
1.2. Concepts of Wide Dependence
In this section, we will present some wide dependence structures introduced in Wang et al. [17].
Definition 1.
For the random variables {εn,n≥1}, if there exists a finite real sequence {gU(n),n≥1} satisfying for each n≥1 and for all xi∈(-∞,∞), 1≤i≤n,
(3)P(ε1>x1,ε2>x2,…,εn>xn)≤gU(n)∏i=1nP(εi>xi),
then we say that the random variables {εn,n≥1} are widely upper orthant dependent (WUOD); if there exists a finite real sequence {gL(n),n≥1} satisfying for each n≥1 and for all xi∈(-∞,∞), 1≤i≤n,
(4)P(ε1≤x1,ε2≤x2,…,εn≤xn)≤gL(n)∏i=1nP(εi≤xi),
then we say that the {εn,n≥1} are widely lower orthant dependent (WLOD, in short); if they are both WUOD and WLOD, then we say that the {εn,n≥1} are widely orthant dependent (WOD).
WUOD, WLOD, and WOD random variables are called by a joint name wide dependent (WD) random variables, and gU(n), gL(n), n≥1, are called dominating coefficients.
For examples of WD random variables with various dominating coefficients, we refer the reader to Wang et al. [17]. These examples show that WD random variables contain some common negatively dependent random variables, some positively dependent random variables, and some others. For details about WD random variables, one can refer to Wang et al. [17], Wang and Cheng [18], Wang et al. [19], Chen et al. [20], and so forth.
In what follows, denote g(n)=max{gU(n),gL(n)}. Recall that when gL(n)=gU(n)=1 for any n≥1 in (3) and (4), the random variables {εn,n≥1} are called negatively upper orthant dependent (NUOD) and negatively lower orthant dependent (NLOD), respectively. If they are both NUOD and NLOD, then we say that the random variables {εn,n≥1} are negatively orthant dependent (NOD) (see, e.g., Ebrahimi and Ghosh [21], Block et al. [22], Joag-Dev and Proschan [23], Wang et al. [24–26], Wu [27, 28], Wu and Jiang [29], or Wu and Chen [30]).
If both (3) and (4) hold when gL(n)=gU(n)=M for some constant M, the random variables {Xn,n≥1} are called extended negatively upper orthant dependent (ENUOD) and extended negatively lower orthant dependent (ENLOD), respectively. If they are both ENUOD and ENLOD, then we say that the random variables {εn,n≥1} are extended negatively orthant dependent (ENOD) (see, e.g., Liu [31]). The concept of general extended negative dependence was proposed by Liu [31, 32] and further promoted by Chen et al. [33, 34], Shen [35], Wang and Chen [18], S. J. Wang and W. S. Wang [36] and Wang et al. [37], and so forth.
Wang et al. [17] obtained the following properties for WD random variables, which will be used to prove the main results of the paper.
Proposition 2.
(1) Let {εn,n≥1} be WLOD (WUOD) with dominating coefficients gL(n), n≥1(gU(n),n≥1). If {fn(·),n≥1} are nondecreasing, then {fn(εn),n≥1} are still WLOD (WUOD) with dominating coefficients gL(n), n≥1(gU(n),n≥1); if {fn(·),n≥1} are nonincreasing, then {fn(εn),n≥1} are WUOD (WLOD) with dominating coefficients gL(n), n≥1(gU(n),n≥1).
(2) If {εn,n≥1} are nonnegative and WUOD with dominating coefficients gU(n), n≥1, then for each n≥1,
(5)E∏i=1nεi≤gU(n)∏i=1nEεi.
In particular, if {εn,n≥1} are WUOD with dominating coefficients gU(n), n≥1, then for each n≥1 and any s>0,
(6)Eexp{s∑i=1nεi}≤gU(n)∏i=1nEexp{sεi}.
By Proposition 2, we can get the following corollary immediately.
Corollary 3.
(1) Let {εn,n≥1} be WD. If {fn(·),n≥1} are nondecreasing (or nonincreasing), then {fn(εn),n≥1} are still WD.
(2) If {Xn,n≥1} are WD, then for each n≥1 and any s∈ℝ,
(7)Eexp{s∑i=1nεi}≤g(n)∏i=1nEexp{sεi}.
In this paper, we will present the Bernstein-type inequality for WD random variables. By using the Bernstein-type inequality, we will further investigate the strong consistency for the estimator of nonparametric regression models based on WD errors.
This work is organized as follows: the Bernstein-type inequality for WD random variables is provided in Section 2 and strong consistency for the estimator of nonparametric regression models based on WD errors is investigated in Section 3.
Throughout the paper, C denotes a positive constant not depending on n, which may be different in various places. an=O(bn) represents an≤Cbn for all n≥1. Let ⌈x⌉ denote the integer part of x and I(A) be the indicator function of the set A. Denote x+=xI(x≥0) and x-=-xI(x<0). Let {εn,n≥1} be a sequence of WD random variables. Denote Sn=∑i=1nεi. In the sequel, we will use the following different assumptions in different situations:
(8)limn→∞g(n)e-anc=0,(9)limn→∞g(n)e-dlog3/2n=0,
where a, c, and d are finite positive constants.
2. Bernstein-Type Inequality for WD Random Variables
In this section, we will present the Bernstein-type inequality for WD random variables, which will be used to prove the strong consistency for estimator of the nonparametric regression model based on WD random variables.
Theorem 4.
Let {εn,n≥1} be a sequence of WD random variables with Eεi=0 and |εi|≤b for each i≥1, where b is a positive constant. Denote σi2=Eεi2 and Bn2=∑i=1nσi2 for each n≥1. Then for any ε>0,
(10)P(Sn≥ε)≤gU(n)exp{-ε22Bn2+(2/3)bε},(11)P(|Sn|≥ε)≤2g(n)exp{-ε22Bn2+(2/3)bε}.
Proof.
For any t>0, by Taylor’s expansion, EXi=0 and the inequality 1+x≤ex for x∈ℝ, we can get that for i=1,2,…,n,
(12)Eexp{tεi}=1+∑j=2∞E(tεi)jj!≤1+∑j=2∞tjE|εi|jj!=1+t2σi22∑j=2∞tj-2E|εi|j(1/2)σi2j!≐1+t2σi22Fi(t)≤exp{t2σi22Fi(t)},
where
(13)Fi(t)=∑j=2∞tj-2E|εi|j(1/2)σi2j!,i=1,2,…,n.
Denote C=b/3 and Mn=bε/3Bn2+1. Choosing t>0 such that tC<1 and
(14)tC≤Mn-1Mn=CεCε+Bn2.
It is easy to check that for i=1,2,…,n and j≥2,
(15)E|εi|j≤σi2bj-2≤12σi2Cj-2j!,
which implies that for i=1,2,…,n,
(16)Fi(t)=∑j=2∞tj-2E|εi|j(1/2)σi2j!≤∑j=2∞(tC)j-2=(1-tC)-1≤Mn.
By Markov’s inequality, Corollary 3, (12), and (16), we can get
(17)P(Sn≥ε)≤e-tεEexp{tSn}≤e-tεgU(n)∏i=1nEexp{tεi}≤gU(n)exp{-tε+t2Bn22Mn}.
Taking t=ε/Bn2Mn=ε/(Cε+Bn2). It is easily seen that tC<1 and tC=Cε/(Cε+Bn2). Substituting t=ε/Bn2Mn into the right-hand side of (17), we can obtain (10) immediately. By (10), we have
(18)P(Sn≤-ε)=P(-Sn≥ε)≤gL(n)exp{-ε22Bn2+(2/3)bε},
since {-εn,n≥1} is still a sequence of WD random variables. The desired result (11) follows from (10) and (18) immediately.
By Theorem 4, we can get the following complete convergence for WD random variables immediately.
Corollary 5.
Let {εn,n≥1} be a sequence of WD random variables with Eεi=0 and |εi|≤b for each i≥1, where b is a positive constant. Assume that ∑i=1∞Eεi2<∞. r>0. Let the dominating coefficients gU(n), gL(n), n≥1 satisfy (8) with any finite positive constant a and c=r. Then
(19)n-rSn⟶0,completely,asn⟶∞.
Proof.
For any ε>0, it follows from (11) that
(20)∑n=1∞P(|Sn|≥nrε)≤2∑n=1∞g(n)exp{-n2rε22∑i=1∞EXi2+(2/3)bnrε}≤C1∑n=1∞[exp(-C)]nr<∞,
which implies (19). Here C and C1 are positive constants not depending on n.
3. The Strong Consistency for the Estimator of Nonparametric Regression Models Based on WD Errors
Unless otherwise specified, we assume throughout the paper that gn(x) is defined by (2). For any function g(x), we use c(g) to denote all continuity points of the function g on A. The norm ∥x∥ is the Euclidean norm. For any fixed design point x∈A, the following assumptions on weight functions Wni(x) will be used:
|∑i=1nWni(x)-1|=O(n-1/4);
∑i=1n|Wni(x)|≤C for all n≥1 and max1≤i≤n|Wni(x)|=O(n-1/2log-3/2n);
∑i=1n|Wni(x)|·|g(xni)-g(x)|I(∥xni-x∥>σn-1/4)=O(n-1/4) for some σ>0.
Theorem 6.
Let {εn,n≥1} be a sequence of mean zero WD random variables such that supn≥1Eεn2<∞. Suppose that the conditions (A1)–(A3) hold true and (9) holds for any positive constant d. Assume that g(x) satisfies a local Lipschitz condition around the point x. Then for any x∈A,
(21)gn(x)⟶g(x),asn⟶∞,a.s.
Proof.
For x∈A, we have by (1) and (2) that
(22)|Egn(x)-g(x)|≤∑i=1n|Wni(x)|·|g(xni)-g(x)|I(∥xni-x∥≤σn-1/4)=+∑i=1n|Wni(x)|·|g(xni)-g(x)|I(∥xni-x∥>σn-1/4)=+|g(x)|·|∑i=1nWni(x)-1|.
By (22), the conditions (A1)–(A3) and the assumption on g(x), we have
(23)|Egn(x)-g(x)|=O(n-1/4),x∈A.
Hence, to prove (21), we only need to show that
(24)gn(x)-Egn(x)⟶0,asn⟶∞,a.s.
For fixed design point x∈A, without loss of generality, we assume that Wni(x)>0 in what follows (otherwise, we use Wni+(x) and Wni-(x) instead of Wni(x), respectively, and note that Wni(x)=Wni+(x)-Wni-(x)). Let
(25)ε1,i(n)=-i1/2I(εni<-i1/2)+εniI(|εni|≤i1/2)+i1/2I(εni>i1/2),ε2,i(n)=(εni-i1/2)I(εni>i1/2),ε3,i(n)=(εni+i1/2)I(εni<-i1/2),ε1,i=-i1/2I(εi<-i1/2)+εiI(|εi|≤i1/2)+i1/2I(εi>i1/2),ε2,i=(εi-i1/2)I(εi>i1/2),ε3,i=(εi+i1/2)I(εi<-i1/2).
Since Eεni=Eεi=0 for each n, it is easy to see that
(26)gn(x)-Egn(x)=∑i=1nWni(x)εni=∑i=1nWni(x)[ε1,i(n)-Eε1,i(n)]=+∑i=1nWni(x)[ε2,i(n)-Eε2,i(n)]=+∑i=1nWni(x)[ε3,i(n)-Eε3,i(n)]=:Tn1+Tn2+Tn3.
By the condition (A2), we can see that
(27)max1≤i≤n|Wni(x)(ε1,i-Eε1,i)|≤2n1/2max1≤i≤n|Wni(x)|≤Clog-3/2n,∑i=1nVar[Wni(x)(ε1,i-Eε1,i)]≤∑i=1nWni2(x)Eεi2≤Cmax1≤i≤n|Wn,i(x)|∑i=1n|Wni(x)|≤Cn-1/2log-3/2n.
For fixed x∈A and n, since (εn1,εn2,…,εnn) have the same distribution as (ε1,ε2,…,εn) and {Wni(x)(ε1,i-Eε1,i),1≤i≤n} are WD with mean zero, we have by applying Theorem 4 that for every ϵ>0,
(28)P(|Tn1|≥ϵ)=P(|∑i=1nWni(x)[ε1,i(n)-Eε1,i(n)]|≥ϵ)=P(|∑i=1nWni(x)[ε1,i-Eε1,i]|≥ϵ)≤2g(n)exp{-ϵ2Cn-1/2log-3/2n+Cϵlog-3/2n}≤2g(n)exp{-Clog3/2n}≤Cn-2,========iifornlargeenough,
which implies
(29)Tn1=∑i=1nWni(x)[ε1,i(n)-Eε1,i(n)]⟶0,asn⟶∞,a.s.
by Borel-Cantelli lemma.
Next, we will estimate Tn2 and Tn3. It can be checked by supn≥1Eεn2<∞ that
(30)∑i=1∞Eε2,i(n)i1/2log5/4(2i)=∑i=1∞Eε2,ii1/2log5/4(2i)≤∑i=1∞E[εiI(εi>i1/2)]i1/2log5/4(2i)≤∑i=1∞Eεi2ilog5/4(2i)<∞,
which implies
(31)∑i=1∞ε2,i(n)i1/2log5/4(2i)<∞,a.s.
Consequently, by Kronecker’s lemma, we have that
(32)1n1/2log5/4(2n)∑i=1nε2,i(n)⟶0,a.s.
Thus, by the condition (A2), it is easy to see that
(33)|∑i=1nWni(x)ε2,i(n)|≤max1≤i≤n|Wni(x)|∑i=1nε2,i(n)≤Cn-1/2log-3/2n∑i=1nε2,i(n)=o(log-1/4n),a.s.
By supn≥1Eεn2<∞ and (A2) again, we have
(34)|∑i=1nWn,i(x)Eε2,i(n)|=|∑i=1nWn,i(x)Eε2,i|≤max1≤i≤n|Wni(x)|∑i=1nE[|εi|I(|εi|≥i1/2)]≤Cn-1/2log-3/2n∑i=1ni-1/2E[εi2I(|εi|≥i1/2)]=O(log-3/2n).
Combining (33) and (34), it follows that
(35)|Tn2|=|∑i=1nWni(x)[ε2,i(n)-Eε2,i(n)]|=o(log-1/4n),a.s.
Likewise, by supn≥1Eεn2<∞, we can see that
(36)∑i=1∞E|ε3,i(n)|i1/2log5/4(2i)=∑i=1∞E|ε3,i|i1/2log5/4(2i)≤∑i=1∞-E[εiI(εi<-i1/2)]i1/2log5/4(2i)≤∑i=1∞Eεi2ilog5/4(2i)<∞,
which implies
(37)∑i=1∞|ε3,i(n)|i1/2log5/4(2i)<∞,a.s.
Hence, by Kronecker’s lemma,
(38)1n1/2log5/4(2n)∑i=1n|ε3,i(n)|⟶0,a.s.
Consequently, we have by (A2) that
(39)|∑i=1nWni(x)ε3,i(n)|≤max1≤i≤n|Wni(x)|∑i=1n|ε3,i(n)|=o(log-1/4n),a.s.
On the other hand, by (A2) and supn≥1Eεn2<∞ again, we can see that
(40)|∑i=1nWni(x)Eε3,i(n)|=|∑i=1nWni(x)Eε3,i|≤max1≤i≤n|Wni(x)|∑i=1nE[|εi|I(|εi|>i1/2)]≤Cn-1/2log-3/2n∑i=1ni-1/2E[εi2I(|εi|>i1/2)]=O(log-3/2n).
From the statements above, we have
(41)|Tn3|=|∑i=1nWn,i(x)[ε3,i(n)-Eε3,i(n)]|=o(log-1/4n),a.s.
Therefore, (24) follows from (26), (29), (35), and (41) immediately. This completes the proof of the theorem.
Theorem 7.
Let {εn,n≥1} be a sequence of mean zero WD random variables such that supn≥1Eεn4<∞. Suppose that the conditions (A1)–(A3) hold true and (9) holds for any positive constant d. Assume that g(x) satisfies a local Lipschitz condition around the point x. Then for any x∈A,
(42)gn(x)-g(x)=O(n-1/4),a.s.
Proof.
According to (23), we can see that in order to prove (42), we only need to show that
(43)|gn(x)-Egn(x)|=O(n-1/4),a.s.
We still assume that Wni(x)>0 in what follows. The proof is similar to that of Theorem 6. We use the same notations εq,i(n), εq,i and Tnq for q=1,2,3 as those in Theorem 6, where i1/2 is replaced by i1/4. Obviously supn≥1Eεn4<∞ implies supn≥1Eεn2<∞. It follows by (A2) that
(44)max1≤i≤n|Wni(x)(ε1,i-Eε1,i)|≤2n1/4max1≤i≤n|Wn,i(x)|≤Cn-1/4log-3/2n,∑i=1nVar[Wni(x)(ε1,i-Eε1,i)]≤∑i=1nWn,i2(x)Eεi2≤Cn-1/2log-3/2n.
By applying Theorem 4 and (9), we can see that for every ϵ>0,
(45)P(|Tn1|≥ϵn-1/4)=P(∑i=1nWni(x)[ε1,i(n)-Eε1,i(n)]≥ϵn-1/4)=P(∑i=1nWni(x)[ε1,i-Eε1,i]≥ϵn-1/4)≤2g(n)exp{-ϵ2n-1/2Cn-1/2log-3/2n+Cϵn-1/2log-3/2n}≤2g(n)exp{-Clog3/2n}≤Cn-2,fornlargeenough,
which implies by Borel-Cantelli lemma that
(46)n1/4Tn1⟶0,a.s.
Meanwhile, it can be checked by supn≥1Eεn4<∞ that
(47)∑i=1∞Eε2,i(n)i1/4log3/2(2i)=∑i=1∞Eε2,ii1/4log3/2(2i)≤∑i=1∞E[εiI(εi>i1/4)]i1/4log3/2(2i)≤∑i=1∞Eεi4ilog3/2(2i)<∞,
which implies
(48)∑i=1∞ε2,i(n)i1/4log3/2(2i)<∞,a.s.
Then, we have by Kronecker’s lemma that
(49)1n1/4log3/2(2n)∑i=1nε2,i(n)⟶0,a.s.
Consequently, it follows by (A2) that
(50)|∑i=1nWni(x)ε2,i(n)|≤max1≤i≤n|Wni(x)|∑i=1nε2,i(n)=o(n-1/4),a.s.,(51)|∑i=1nWni(x)Eε2,i(n)|=|∑i=1nWni(x)Eε2,i|≤max1≤i≤n|Wni(x)|∑i=1nE[|εi|I(|εi|>i1/4)]≤Cn-1/2log-3/2n∑i=1ni-3/4E[εi4I(|εi|>i1/4)]=O(n-1/4log-3/2n).
On the other hand, it can be checked that
(52)∑i=1∞E|ε3,i(n)|i1/4log3/2(2i)=∑i=1∞E|ε3,i|i1/4log3/2(2i)≤∑i=1∞-E[εiI(εi<-i1/4)]i1/4log3/2(2i)≤∑i=1∞Eεi4ilog3/2(2i)<∞,
which implies
(53)∑i=1∞|ε3,i(n)|i1/4log3/2(2i)<∞,a.s.
So, by Kronecker’s lemma,
(54)1n1/4log3/2(2n)∑i=1n|ε3,i(n)|⟶0,a.s.
Consequently, we have by (A2) that
(55)|∑i=1nWni(x)ε3,i(n)|≤max1≤i≤n|Wni(x)|∑i=1n|ε3,i(n)|=o(n-1/4),a.s.,(56)|∑i=1nWni(x)Eε3,i(n)|=|∑i=1nWni(x)Eε3,i|≤max1≤i≤n|Wni(x)|∑i=1nE[|εi|I(|εi|>i1/4)]≤Cn-1/2log-3/2n∑i=1ni-3/4E[εi4I(|εi|>i1/4)]=O(n-1/4log-3/2n).
Finally, similar to the proof of (21), we can get (43) immediately by (46)–(56). This completes the proof of the theorem.
As an application of Theorems 6 and 7, we give the strong consistency for the nearest neighbor estimator of g(x). Without loss of generality, put A=[0,1], taking xni=i/n, i=1,2,…,n. For any x∈A, we rewrite |xn1-x|,|xn2-x|,…,|xnn-x| as follows:
(57)|xR1(x)(n)-x|≤|xR2(x)(n)-x|≤⋯≤|xRn(x)(n)-x|,
if |xni-x|=|xnj-x|, then |xni-x| is permuted before |xnj-x| when xni<xnj.
Let 1≤kn≤n, the nearest neighbor weight function estimator of g(x) in model (1) is defined as follows:
(58)g~n(x)=∑i=1nW~ni(x)Yni,
where
(59)W~ni(x)={1kn,if|xni-x|≤|xRkn(x)(n)-x|,0,otherwise.
Based on the notations above, we can get the following result by using Theorems 6 and 7.
Corollary 8.
Let {εn,n≥1} be a sequence of mean zero WD random variables and (9) holds for any positive constant d. Assume that g(x) satisfies a local Lipschitz condition around the point x. Denote kn=⌈n5/8⌉.
If supn≥1Eεn2<∞, then (21) holds for any x∈A.
If supn≥1Eεn4<∞, then (42) holds for any x∈A.
Proof.
It suffices to show that the conditions (A1)–(A3) are satisfied. For any x∈[0,1], it follows from the definitions of Ri(x) and W~ni(x) that
(60)∑i=1nW~ni(x)=∑i=1nW~nRi(x)(x)=∑i=1kn1kn=1,max1≤i≤nW~ni(x)=1kn≤Cn-5/8,W~ni(x)≥0,∑i=1n|W~ni(x)|·|g(xni)-g(x)|I(|xni-x|>σn-1/4)≤C∑i=1n(xni-x)2|W~ni(x)|σ2n-1/2=C∑i=1kn(xRi(x)(n)-x)2n1/2knσ2≤C∑i=1kn(i/n)2n1/2knσ2≤C(knnσ)2n1/2≤Cn-1/4,∀a>0.
Hence, conditions (A1)–(A3) are satisfied. By Theorems 6 and 7, we can get (i) and (ii) immediately. This completes the proof of the corollary.
Acknowledgments
The authors are most grateful to the Editor Juan J. Trujillo and anonymous referee for careful reading of the paper and valuable suggestions which helped in improving an earlier version of this paper. This work was supported by the National Natural Science Foundation of China (11201001, 11171001), the Natural Science Foundation of Anhui Province (1308085QA03, 11040606M12, 1208085QA03), the 211 Project of Anhui University, the Youth Science Research Fund of Anhui University, Applied Teaching Model Curriculum of Anhui University (XJYYXKC04), and the Students Science Research Training Program of Anhui University (KYXL2012007, kyxl2013003).
GeorgievA. A.GrossmannW.Local properties of function fitting estimates with application to system identification1985Dordrecht, The NetherlandsReidel141151Proceedings of the 4th Pannonian Symposium on Mathematical Statistics, September 1983MR851050ZBL0588.62059GeorgievA. A.GreblickiW.Nonparametric function recovering from noisy observations198613111410.1016/0378-3758(86)90114-XMR822121ZBL0596.62041GeorgievA. A.Consistent nonparametric multiple regression: the fixed design case198825110011010.1016/0047-259X(88)90155-8MR935297ZBL0637.62044MüllerH.-G.Weak and universal consistency of moving weighted averages198718324125010.1007/BF01848087MR902526ZBL0596.62040FanY.Consistent nonparametric multiple regression for dependent heterogeneous processes: the fixed design case1990331728810.1016/0047-259X(90)90006-4MR1057055ZBL0698.62040RoussasG. G.Consistent regression estimation with fixed design points under dependence conditions198981415010.1016/0167-7152(89)90081-3MR1006420ZBL0674.62026RoussasG. G.TranL. T.IoannidesD. A.Fixed design regression for time series: asymptotic normality199240226229110.1016/0047-259X(92)90026-CMR1150613ZBL0764.62073TranL.RoussasG.YakowitzS.VanB. T.Fixed-design regression for linear time series199624397599110.1214/aos/1032526952MR1401833ZBL0862.62069HuS. H.ZhuC. H.ChenY. B.WangL. C.Fixed-design regression for linear time series2002221918MR1883874ZBL1010.62085HuS. H.PanG. M.GaoQ. B.Estimation problems for a regression model with linear process errors20031818190MR1960506LiangH. Y.JingB. Y.Asymptotic properties for estimates of nonparametric regression models based on negatively associated sequences200595222724510.1016/j.jmva.2004.06.004MR2170396ZBL1070.62022YangW. Z.WangX. J.WangX. H.HuS. H.The consistency for estimator ol nonparametric regression model based on NOD errors20122012article 14010.1186/1029-242X-2012-140WangX. J.XiaF. X.GeM. M.HuS. H.YangW. Z.Complete consistency of the estimator of nonparametric regression models based on ρ˜-mixing sequences201220121290728610.1155/2012/907286MR3004858RenZ.ChenM. H.Strong consistency of a class of estimators in a partial linear model for negative associated samples20021816066MR1953560BaekJ.-I.LiangH. Y.Asymptotics of estimators in semi-parametric model under NA samples2006136103362338210.1016/j.jspi.2005.01.008MR2256272ZBL1112.62032LiangH. Y.MammitzschV.SteinebachJ.On a semiparametric regression model whose errors form a linear process with negatively associated innovations200640320722610.1080/02331880600688163MR2236204ZBL1098.62044WangK.WangY.GaoQ.Uniform asymptotics for the finite-time ruin probability of a new dependent risk model with a constant interest rate20131511091242-s2.0-7995585785610.1007/s11009-011-9226-yWangY.ChengD.Basic renewal theorems for random walks with widely dependent increments2011384259760610.1016/j.jmaa.2011.06.010MR2825210ZBL1230.60095WangY.CuiZ.WangK.MaX.Uniform asymptotics of the finite-time ruin probability for all times2012390120822310.1016/j.jmaa.2012.01.025MR2885767ZBL1237.91139ChenY.WangL.WangY.Uniform asymptotics for the finite-time ruin probabilities of two kinds of nonstandard bidimensional risk models2013401111412910.1016/j.jmaa.2012.11.046MR3011252EbrahimiN.GhoshM.Multivariate negative dependence1981104307337MR61240010.1080/03610928108828041ZBL0506.62034BlockH. W.SavitsT. H.ShakedM.Some concepts of negative dependence1982103765772MR65954510.1214/aop/1176993784ZBL0501.62037Joag-DevK.ProschanF.Negative association of random variables, with applications198311128629510.1214/aos/1176346079MR684886ZBL0508.62041WangX. J.HuS. H.YangW. Z.LingN. X.Exponential inequalities and inverse moment for NOD sequence2010805-645246110.1016/j.spl.2009.11.023MR2593586ZBL1186.60015WangX. J.HuS. H.VolodinA. I.Strong limit theorems for weighted sums of NOD sequence and exponential inequalities201148592393810.4134/BKMS.2011.48.5.923MR2867181ZBL1234.60035WangX. J.HuS. H.ShenA. T.YangW. Z.An exponential inequality for a NOD sequence and a strong law of large numbers201124221922310.1016/j.aml.2010.09.007MR2735145ZBL1205.60068WuQ. Y.A strong limit theorem for weighted sums of sequences of negatively dependent random variables20102010838380510.1155/2010/383805MR2678912ZBL1202.60044WuQ. Y.A complete convergence theorem for weighted sums ol arrays ol rowwise negatively dependent random variables20122012article 50WuQ. Y.JiangY. Y.The strong consistency of M estimator in a linear model for negatively dependent random samples201140346749110.1080/03610920903427792MR2765842ZBL1208.62039WuQ. Y.ChenP. Y.An improved result in almost sure central limit theorem for self-normalized products of partial sums20132013article 12910.1186/1029-242X-2013-129MR3049949LiuL.Precise large deviations for dependent random variables with heavy tails20097991290129810.1016/j.spl.2009.02.001MR2519013ZBL1163.60012LiuL.Necessary and sufficient conditions for moderate deviations of dependent random variables with heavy tails20105361421143410.1007/s11425-010-4012-9MR2658603ZBL1196.60040ChenY.ChenA.NgK. W.The strong law of large numbers for extended negatively dependent random variables2010474908922MR275290210.1239/jap/1294170508ZBL1213.60058ChenY.YuenK. C.NgK. W.Precise large deviations of random sums in presence of negative dependence and consistent variation201113482183310.1007/s11009-010-9194-7MR2851839ZBL1242.60027ShenA. T.Probability inequalities for END sequence and their applications20112011article 9810.1186/1029-242X-2011-98MR2853339WangS. J.WangW. S.Extended precise large deviations of random sums in the presence of END structure and consistent variation2012201212436531MR2898068ZBL1236.6003110.1155/2012/436531WangX. J.WangS. J.HuS. H.LingJ. M.WeiY. F.On complete convergence of weighted sums for arrays of rowwise extended negatively dependent random variables201273699610.1080/17442508.2012.736996