Volume 2010, Article ID 956907,30pages doi:10.1155/2010/956907
Research Article
QML Estimators in Linear Regression Models with Functional Coefficient Autoregressive Processes
Hongchang Hu
School of Mathematics and Statistics, Hubei Normal University, Huangshi 435002, China
Correspondence should be addressed to Hongchang Hu,retutome@163.com Received 30 December 2009; Revised 19 March 2010; Accepted 6 April 2010 Academic Editor: Massimo Scalia
Copyrightq2010 Hongchang Hu. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
This paper studies a linear regression model, whose errors are functional coefficient autoregressive processes. Firstly, the quasi-maximum likelihoodQMLestimators of some unknown parameters are given. Secondly, under general conditions, the asymptotic propertiesexistence, consistency, and asymptotic distributionsof the QML estimators are investigated. These results extend those of Maller2003, White1959, Brockwell and Davis1987, and so on. Lastly, the validity and feasibility of the method are illuminated by a simulation example and a real example.
1. Introduction
Consider the following linear regression model:
ytxTtβεt, t1,2, . . . , n, 1.1
whereyt’s are scalar response variables,xt’s are explanatory variables,βis ad-dimensional unknown parameter, and theεt’s are functional coefficient autoregressive processes given as ε1η1, εtftθεt−1ηt, t2,3, . . . , n, 1.2
whereηt’s are independent and identically distributed random errors with zero mean and finite variance σ2, θ is a one-dimensional unknown parameter and ftθ is a real valued function defined on a compact setΘwhich contains the true valueθ0as an inner point and is a subset ofR1. The values ofθ0andσ2are unknown.
Model1.1includes many special cases, such as an ordinary linear regression model whenftθ≡0; see1–11 . In the sequel, we always assume thatftθ/0,for someθ∈Θ, is a linear regression model with constant coefficient autoregressive processeswhen ftθ θ, see Maller 12 , Pere 13 , and Fuller 14 , time-dependent and functional coefficient autoregressive processes when β 0, see Kwoun and Yajima 15 , constant coefficient autoregressive processes when ftθ θand β 0, see White 16, 17 , Hamilton 18 , Brockwell and Davis 19 , and Abadir and Lucas 20 , time-dependent or time-varying autoregressive processeswhenftθ atandβ 0, see Carsoule and Franses21 , Azrak and M´elard22 , and Dahlhaus23 , and so forth.
Regression analysis is one of the most mature and widely applied branches of statistics. Linear regression analysis is one of the most widely used statistical techniques.
Its applications occur in almost every field, including engineering, economics, the physical sciences, management, life and biological sciences, and the social sciences. Linear regression model is the most important and popular model in the statistical literature, which attracts many statisticians to estimate the coefficients of the regression model. For the ordinary linear regression modelwhen the errors are independent and identically distributed random variables, Bai and Guo 1 , Chen 2 , Anderson and Taylor 3 , Drygas 4 , Gonz´alez- Rodr´ıguez et al.5 , Hampel et al.6 , He7 , Cui8 , Durbin9 , Hoerl and Kennard10 , Li and Yang11 , and Zhang et al.24 used various estimation methodsLeast squares estimate method, robust estimation, biased estimation, and Bayes estimationto obtain estimators of the unknown parameters in1.1and discussed some large or small sample properties of these estimators.
However, the independence assumption for the errors is not always appropriate in applications, especially for sequentially collected economic and physical data, which often exhibit evident dependence on the errors. Recently, linear regression with serially correlated errors has attracted increasing attention from statisticians. One case of considerable interest is that the errors are autoregressive processes and the asymptotic theory of this estimator was developed by Hannan and Kavalieris 25 . Fox and Taqqu26 established its asymptotic normality in the case of long-memory stationary Gaussian observations errors. Giraitis and Surgailis 27 extended this result to non-Gaussian linear sequences. The asymptotic distribution of the maximum likelihood estimator was studied by Giraitis and Koul in 28 and Koul in29 when the errors are nonlinear instantaneous functions of a Gaussian long-memory sequence. Koul and Surgailis 30 established the asymptotic normality of the Whittle estimator in linear regression models with non-Gaussian long-memory moving average errors. When the errors are Gaussian, or a function of Gaussian random variables that are strictly stationary and long range dependent, Koul and Mukherjee31 investigated the linear model. Shiohama and Taniguchi32 estimated the regression parameters in a linear regression model with autoregressive process.
In addition toconstant or functional or random coefficientautoregressive model, it has gained much attention and has been applied to many fields, such as economics, physics, geography, geology, biology, and agriculture. Fan and Yao 33 , Berk 34 , Hannan and Kavalieris35 , Goldenshluger and Zeevi36 , Liebscher37 , An et al.38 , Elsebach39 , Carsoule and Franses 21 , Baran et al.40 , Distaso 41 , and Harvill and Ray 42 used various estimation methodsthe least squares method, the Yule-Walker method, the method of stochastic approximation, and robust estimation methodto obtain some estimators and discussed their asymptotic properties, or investigated hypotheses testing.
This paper discusses the model 1.1-1.2 including stationary and explosive processes. The organization of the paper is as follows. InSection 2some estimators ofβ, θ,
and σ2 are given by the quasi-maximum likelihood method. Under general conditions, the existence and consistency the quasi-maximum likelihood estimators are investigated, and asymptotic normality as well, inSection 3. Some preliminary lemmas are presented in Section 4. The main proofs are presented inSection 5, with some examples inSection 6.
2. Estimation Method
Write the “true” model as
ytxTtβ0et, t1,2, . . . , n, 2.1 e1 η1, etftθ0et−1ηt, t2,3, . . . , n, 2.2
whereftθ0 dftθ/dθ|θθ0/0, andηt’s are i.i.d errors with zero mean and finite variance σ02. Define−1
i0ft−iθ0 1, and by2.2we have
ett−1
j0
j−1
i0
ft−iθ0
ηt−j. 2.3
Thusetis measurable with respect to theσ-fieldHgenerated byη1, η2, . . . , ηt, and
Eet0, Varet σ02 t−1
j0
j−1
i0
ft−i2 θ0
. 2.4
Assume at first that theηt’s are i.i.d.N0, σ2. Using similar arguments to those of Fuller14 or Maller12 , we get the log-likelihood ofy2, y3, . . . , ynconditional ony1:
Ψn
β, θ, σ2
logLn−1
2n−1logσ2− 1 2σ2
n t2
εt−ftθεt−12
− 1
2n−1log2π. 2.5
At this stage we drop the normality assumption, but still maximize 2.5 to obtain QML estimators, denoted byσn2, βn, θnwhen they exist:
∂Ψn
∂σ2 −n−1 2σ2 1
2σ4 n
t2
εt−ftθεt−12
, 2.6
∂Ψn
∂θ 1 σ2
n t2
ftθ εt−ftθεt−1
εt−1, 2.7
∂Ψn
∂β 1 σ2
n t2
εt−ftθεt−1 xt−ftθxt−1
. 2.8
Thusσn2, βn, θnsatisfy the following estimation equations:
σn2 1
n−1 n
t2
εt−ft
θn
εt−12
, 2.9
n t2
εt−ft
θn
εt−1 ft
θn
εt−10, 2.10
n t2
εt−ft
θn
εt−1 xt−ft
θn
xt−1
0, 2.11
where
εtyt−xTtβn. 2.12
Remark 2.1. If ftθ θ, then the above equations become the same as Maller’s 12 . Therefore, we extend the QML estimators of Maller12 .
To calculate the values of the QML estimators, we may use the grid search method, steepest ascent method, Newton-Raphson method, and modified Newton-Raphson method.
In order to calculate inSection 6, we introduce the most popular modified Newton-Raphson method proposed by Davidon-Fletcher-Powellsee Hamilton18 .
Letd2×1 vector→−
θm σm2, βm, θmdenote an estimator of→−
θ σ2, β, θthat has been calculated at themth iteration, and letAm denote an estimation of H→−
θm −1. The new estimator→−
θm1is given by
−
→θm1→−
θmsAmg →−
θm
2.13
forsthe positive scalar that maximizesΨn{→−
θmsAmg→−
θm},whered2×1 vector
g →−
θm
∂Ψn
→− θ
∂→− θ |→−
θ→− θm
⎛
⎜⎜
⎜⎜
⎜⎜
⎝
∂Ψn
∂σ2|σ2σm2
∂Ψn
∂β |ββm
∂Ψn
∂θ |θθm
⎞
⎟⎟
⎟⎟
⎟⎟
⎠
2.14
andd2×d2symmetric matrix
H →−
θm
−∂2Ψn
→− θ
∂→− θ ∂→−
θT
|→− θ→−
θm
⎛
⎜⎜
⎜⎜
⎜⎜
⎜⎝
∂2Ψn
∂σ22
∂2Ψn
∂σ2∂β
∂2Ψn
∂σ2∂θ
∗ ∂2Ψn
∂β∂βT
∂2Ψn
∂β∂θ
∗ ∗ ∂2Ψn
∂θ2
⎞
⎟⎟
⎟⎟
⎟⎟
⎟⎠
|→− θ→−
θm , 2.15
where
∂2Ψn
∂σ22 n−1 2σ4 − 1
σ6 n
t2
εt−ftθεt−12 ,
∂2Ψn
∂σ2∂β − 1 σ4
n t2
εt−ftθεt−1 xt−ftθxt−1T ,
∂2Ψn
∂σ2∂θ − 1 σ4
n t2
εt−ftθεt−1 ftθ,
2.16
∂2Ψn
∂β∂βT − 1 σ2
n t2
xt−ftθxt−1 xt−ftθxt−1T
, 2.17
∂2Ψn
∂β∂θ − 1 σ2
n t2
ftθεt−1xtftθεtxt−1−2ftθftθxt−1εt−1 ,
∂2Ψn
∂θ2 − 1 σ2
n t2
ft2θ ftθftθ
ε2t−1−ftθεtεt−1 .
2.18
Once→−
θm1and the gradient at→−
θm1have been calculated, a new estimationAm1 is found from
Am1Am−Am Δgm1 Δgm1T
Am Δgm1T
Am Δgm1 −
Δ→− θm1
Δ→−
θm1 T
Δgm1T Δ→−
θm1
, 2.19
where
Δ→−
θm1→−
θm1−→−
θm, Δgm1g →−
θm1
−g →−
θm
. 2.20
It is well known that least squares estimators in ordinary linear regression model are very good estimators, so a recursive algorithms procedure is to start the iteration with β0, σ02 which are least squares estimators ofβ and σ2, respectively. Take θ0 such that ftθ0 0. Iterations are stopped if some termination criterion is reached, for example, if
→−
θm1−→− θm
→− θm
< δ, 2.21
for some prechosen small numberδ >0.
Up to this point, we obtain the values of QML estimators when the functionftθ ft, θis known. However, the functionftθis never the case in practice; we have to estimate it. By2.12and1.2, we obtain
f t,θn
εt
εt−1, t2,3, . . . , n. 2.22
Based on the dataset{ft, θn, t2,3, . . . , n}, we may obtain the estimation functionft,θn of ft, θ by some smoothing methods see Simonff 43 , Fan and Yao 33 , Green and Silverman44 , Fan and Gijbels45 , etc.
To obtain our results, the following conditions are sufficient.
A1Xnn
t2xtxTt is positive definite for sufficiently largenand
n→ ∞lim max
1≤t≤nxTtXn−1xt0, 2.23
lim sup
n→ ∞ |λ|max
Xn−1/2ZnXn−T/2
<1, 2.24
whereZn 1/2n
t2xtxTt−1xt−1xTtand|λ|max·denotes the maximum in absolute value of the eigenvalues of a symmetric matrix.
A2There is a constantα >0 such that
t j1
j−1
i0
ft−i2 θ
≤α 2.25
for anyt∈ {1,2, . . . , n}andθ∈Θ.
A3The derivativesftθ dftθ/dθ, ftθ dftθ/dθexist and are bounded for anytandθ∈Θ.
Remark 2.2. Maller 12 applied the condition A1, and Kwoun and Yajima15 used the conditionsA2andA3. Thus our conditions are general.A1delineates the class ofxtfor which our results hold in the sense required. It is further discussed by Maller in12 . Kwoun and Yajima15 call{et}stable if Varetis bounded. ThusA2implies that{et}is stable.
However,{et}is not stationary. In fact, by2.3, we obtain that
Covet, etk σ02 k−1
i0
ftk−iθ0 ftθ0k
i0
ftk−iθ0 · · ·t−2
l0
ft−lθ0tk−2
i0
ftk−iθ0
, 2.26
which is dependent oft.
For ease of exposition, we will introduce the following notations which will be used later in the paper.
Defined1-vectorϕ β, θ, and
Sn ϕ
σ2∂Ψn
∂ϕ σ2 ∂Ψn
∂β ,∂Ψn
∂θ
, Fn ϕ
−σ2 ∂2Ψn
∂ϕ∂ϕT. 2.27
By2.7and2.8, we get
Fn ϕ
⎛
⎜⎜
⎜⎝
Xnθ n
t2
ftθεt−1xtftθεtxt−1−2ftθftθxt−1εt−1
∗ n
t2
ft2θ ftθftθ
εt−12 −ftθεtεt−1
⎞
⎟⎟
⎟⎠, 2.28
whereXnθ −σ2∂2Ψn/∂β∂βTand the∗indicates that the element is filled in by symmetry.
Thus,
DnE Fn ϕ0
⎛
⎜⎜
⎝
Xnθ0 0
∗ n
t2
ft2θ0 ftθ0ftθ0
Eet−12 −ftθ0Eetet−1
⎞
⎟⎟
⎠
⎛
⎜⎜
⎝
Xnθ0 0
∗ n
t2
ft2θ0Eet−12
⎞
⎟⎟
⎠
Xnθ0 0
∗ Δnθ0, σ0
,
2.29
where
Δnθ0, σ0
n t2
ft2θ0Ee2t−1 σ02 n
t2
ft2θ0t−2
j0
j−1
i0
ft−i2 θ
On. 2.30
3. Statement of Main Results
Theorem 3.1. Suppose that conditions (A1)–(A3) hold. Then there is a sequenceAn ↓ 0 such that, for eachA >0, asn → ∞, the probability
P
there are estimatorsϕn,σn2 withSn ϕn
0, and ϕn,σn2
∈NnA
−→1. 3.1
Furthermore,
ϕn,σn2
−→p
ϕ0, σ02
, n−→ ∞, 3.2
where, for eachn1,2, . . . , A >0 andAn∈0, σ02; define neighborhoods NnA
ϕ∈Rd1: ϕ−ϕ0
T
Dn ϕ−ϕ0
≤A2 , NnA NnA∩
σ2∈
σ02−An, σ20An
.
3.3
Theorem 3.2. Suppose that conditions (A1)–(A3) hold. Then 1
σnFnT/2 ϕn ϕn−ϕ0
−→DN0, Id1, n−→ ∞. 3.4
Remark 3.3. Forθ∈Rm, m∈N, our results still hold.
In the following, we will investigate some special cases in the model 1.1-1.2.
Although the following results are directly obtained from Theorems3.1and3.2, we discuss these results in order to compare with the corresponding results.
Corollary 3.4. Letftθ θ. If condition (A1) holds, then, for|θ|/1,3.1,3.2, and3.4hold.
Remark 3.5. These results are the same as the corresponding results of Maller12 . Corollary 3.6. Ifβ0 andftθ θ, then, for|θ| /1,
n
t2ε2t−1 σn
θn−θ0
−→DN0,1, n−→ ∞, 3.5
where
σ2n 1
n−1 n
t2
εt−θnεt−12
, θn n
t2εtεt−1 n
t2ε2t−1 . 3.6
Remark 3.7. These estimators are the same as the least squares estimatorssee White16 . For|θ|>1,{εt}are explosive processes. In the case, the corollary is the same as the results of White17 . While|θ|<1, notice thatσn2→pσ02and1/n−1n
t2ε2t−1→pEεt2σ02/1−θ20, and byCorollary 3.6we obtain
√n
θn−θ0
−→DN
0,1−θ02
. 3.7
The result was discussed by many authors, such as Fujikoshi and Ochi46 and Brockwell and Davis19 .
Corollary 3.8. Letβ0. If conditions (A2) and (A3) hold, then F1/2n
θn
σn
θn−θ0
−→DN0,1, n−→ ∞, 3.8
where
Fn θn
n
t2
ft2
θn ft
θn ft
θn
ε2t−1−ft θn
εtεt−1 ,
σn2 1
n−1 n
t2
εt−ft
θn εt−12
.
3.9
Corollary 3.9. Letftθ at. If condition (A1) holds, then
1 σn
n
t2
xt−atxt−1xt−atxt−1T T/2
βn−β0
−→DN0, Id, n−→ ∞. 3.10
Remark 3.10. Let at 0. Note that n
t2xtxtT O√n and σn2→pσ02; we easily obtain asymptotic normality of the quasi-maximum likelihood or least squares estimator in ordinary linear regression models from the corollary.
4. Some Lemmas
To prove Theorems3.1and3.2, we first introduce the following lemmas.
Lemma 4.1. The matrix Dn is positive definite for large enough n with ESnϕ0 0 and VarSnϕ0 σ02Dn.
Proof. It is easy to show that the matrixDnis positive definite for large enoughn. By2.8, we have
σ02E ∂Ψn
∂β |ββ0
n
t2
E et−ftθ0et−1 xt−ftθ0xt−1
n
t2
xt−ftθ0xt−1
Eηt0.
4.1
Note thatet−1andηtare independent of each other; thus by2.7andEηt0, we have
σ02E ∂Ψn
∂θ |θθ0
n
t2
E
et−ftθ0et−1
ftθ0et−1
n
t2
ftθ0E ηtet−1 0.
4.2
Hence, from4.1and4.2,
E Sn ϕ0
σ02E ∂Ψn
∂β |ββ0,∂Ψn
∂θ |θθ0
0. 4.3
By2.8and2.17, we have
Var
σ02∂Ψn
∂β |ββ0
Var n
t2
et−ftθ0et−1 xt−ftθ0xt−1
Var n
t2
ηt xt−ftθ0xt−1
σ02Xnθ0.
4.4
Note that{ftθ0ηtet−1, Ht}is a martingale difference sequence with Var ftθ0ηtet−1
ft2θ0Eη2tEe2t−1σ02ft2θ0Ee2t−1, 4.5
so
Var
σ02∂Ψn
∂θ |θθ0
Var n
t2
ηtftθ0et−1
n
t2
ft2θ0Ee2t−1σ02Δnθ0, σ0.
4.6
By2.7and2.8and noting thatet−1andηtare independent of each other, we have
Cov
σ02∂Ψn
∂β |ββ0, σ02∂Ψn
∂θ |θθ0
E
σ02∂Ψn
∂β |ββ0, σ02∂Ψn
∂θ |θθ0
E n
t2
ηt2 xt−ftθ0xt−1
ftθ0et−1
E n
t3
ηt xt−ftθ0xt−1t−1
s2
ηsfsθ0es−1
E n
s3
ηsfsθ0es−1
s−1 t2
ηt xt−ftθ0xt−1 0.
4.7
From4.4–4.7, it follows that VarSnϕ0 σ02Dn.
Lemma 4.2. If condition (A1) holds, then, for anyθ ∈ Θ, the matrixXnθis positive definite for large enoughn, and
nlim→ ∞max
1≤t≤nxTtX−1n θxt0. 4.8
Proof. Letλ1andλdbe the smallest and largest roots of|Zn−λXn|0. Then from the study of Rao in47, Ex 22.1 ,
λ1 ≤ uTZnu
uTXnu ≤λd 4.9
for unit vectors u. Thus by 2.24, there are some δ ∈ max{0,1 − 1 min2≤t≤n| ft2θ|/max2≤t≤n|ftθ|},1andn0δsuch thatn≥N0implies that
uTZnu≤1−δuTXnu. 4.10
By4.10, we have
uTXnun
t2
uT xt−ftθxt−12
n
t2
uTxt2
ft2θ
uTxt−12
−ftθuTxt−1xTtu−ftθuTxtxTt−1u
≥n
t2
uTxt2
min
2≤t≤n
ft2θn
t2
uTxt−12
−max
2≤t≤nftθuTZnu
≥uTXnumin
2≤t≤n
ft2θuTXnu−max
2≤t≤nftθuTZnu
≥
1min
2≤t≤n
ft2θ−max
2≤t≤nftθ1−δ
uTXnu Cθ, δuTXnu.
4.11
By the study of Rao in47, page 60 and2.23, we have
uTxt
2
uTXnu −→0. 4.12
From4.12andCθ, δ>0,
xTtXn−1θ sup
u
uTxt
2
uTXnθu
≤sup
u
uTxt
2
Cθ, δuTXnu
−→0. 4.13
Lemma 4.3see48 . LetWnbe a symmetric random matrix with eigenvaluesλjn,1 ≤j ≤ d.
Then
Wn−→pI⇐⇒λjn−→p1, n−→ ∞. 4.14
Lemma 4.4. For eachA >0,
sup
ϕ∈NnA
D−1/2n Fn ϕ
Dn−T/2−Φn−→p0, n−→ ∞, 4.15
and also
Φn−→DΦ, 4.16
c→lim0lim sup
A→ ∞ lim sup
n→ ∞ P
ϕ∈NinfnAλmin
D−1/2n Fn ϕ Dn−T/2
≤c 0, 4.17
where
Φn
⎛
⎜⎝
Id 0
0 n
t2ft2θ0e2t−1 Δnθ0, σ0
⎞
⎟⎠, Φ Id1. 4.18
Proof. LetXnθ0 X1/2n θ0XnT/2θ0be a square root decomposition ofXnθ0. Then
Dn
Xn1/2θ0 0
∗ !
Δnθ0, σ0
XnT/2θ0 0
∗ !
Δnθ0, σ0
D1/2n DT/2n . 4.19
Letϕ∈NnA. Then
ϕ−ϕ0
T
Dn ϕ−ϕ0
β−β0
T
Xnθ0 β−β0
θ−θ02Δnθ0, σ0≤A2. 4.20
From2.28,2.29, and4.18,
Dn−1/2Fn ϕ
D−T/2n −Φn
⎛
⎜⎜
⎜⎜
⎜⎝
X−1/2n θ0XnθXn−T/2θ0−Id Xn−1/2θ0n
t2 ftθεt−1xtftθεtxt−1−2ftθftθεt−1xt−1
!Δnθ0, σ0
∗
n t2
ft2θ ftθftθ
ε2t−1−ftθεtεt−1
−n
t2 ft2θ0e2t−1
!Δnθ0, σ0
⎞
⎟⎟
⎟⎟
⎟⎠. 4.21
Let
NβnA
β: β−β0
T
X1/2n θ02 ≤A2 , 4.22 NnθA
θ:|θ−θ0| ≤ A
!Δnθ0, σ0
. 4.23
In the first step, we will show that, for eachA >0,
sup
θ∈NθnA
X−1/2n θ0XnθXn−T/2θ0−Id−→0, n−→ ∞. 4.24
In fact, note that
X−1/2n θ0XnθXn−T/2θ0−IdXn−1/2θ0Xnθ−Xnθ0Xn−T/2θ0
Xn−1/2θ0T1T2−T3Xn−T/2θ0, 4.25
where
T1n
t2
ftθ0−ftθ
xt−1 xt−ftθ0xt−1T
,
T2n
t2
xt−ftθ0xt−1 xTt−1,
T3n
t2
ftθ0−ftθ2
xt−1xTt−1.
4.26
Letu, v ∈ Rd, |u| |v| 1, and letuTn uTXn−1/2θ0, vnT Xn−T/2θ0v. By Cauchy- Schwartz inequality,Lemma 4.2, conditionA3, and noting thatθ∈NnθA, we have that
uTnT1vn
n t2
ftθ0−ftθ
uTnxt−1 xt−ftθ0xt−1T
vn
≤max
2≤t≤nftθ0−ftθ
n t2
uTnxt−1 xt−ftθ0xt−1T
vn
≤max
2≤t≤nftθ0−ftθn
t2
uTnxt−1xTt−1un
1/2
· n
t2
vnT xt−ftθ0xt−1 xt−ftθ0xt−1T vn
1/2
≤max
2≤t≤nftθ0−ftθn
t2
uTnxtxTtun 1/2
≤max
2≤t≤n
ft
θ|θ0−θ| ·√ nmax
1≤t≤n
xtTX−1n θ0xt
≤C
"
n
Δnθ0, σ0o1−→0.
4.27
Hereθaθ 1−aθ0for some 0≤a≤1. Similar to the proof ofT1, we easily obtain that uTnT2vn−→0. 4.28
By Cauchy-Schwartz inequality,Lemma 4.2, conditionA3, and noting thatNnθA, we have that
uTnT3vn uTn
n t2
ftθ0−ftθ2
xt−1xt−1T vn
≤max
2≤t≤nftθ0−ftθ2 n
t2
uTnxtxTtun n
t2
vTnxtxTtvn 1/2
≤nmax
2≤t≤n
ft
θ2|θ0−θ|2max
1≤t≤n
xTtXn−1θ0xt
≤ nA2
Δnθ0, σ0o1−→0.
4.29
Hence,4.24follows from4.25–4.29.
In the second step, we will show that Xn−1/2θ0n
t2 ftθεt−1xtftθεtxt−1−2ftθftθεt−1xt−1
!Δnθ0, σ0 −→p0. 4.30
Note that
εtyt−xtTβxtT β0−β et, εt−ftθ0εt−1 xt−ftθ0xt−1T
β0−β ηt.
4.31
Consider
J n
t2
ftθεt−1xtftθεtxt−1−2ftθftθεt−1xt−1
n
t2
εt−1ftθ xt−ftθ0xt−1
ftθ εt−ftθ0εt−1 xt−1 2ftθ ftθ0−ftθ
εt−1xt−1 T1T2T3T42T52T6,
4.32
where
T1n
t2
xTt−1ftθ β0−β xt−ftθ0xt−1
, T2n
t2
ftθet−1 xt−ftθ0xt−1 ,
T3n
t2
ftθ xt−ftθ0xt−1T
β0−β
xt−1, T4n
t2
ftθηtxt−1,
T5n
t2
ftθ ftθ0−ftθ
xTt−1 β0−β
xt−1, T6n
t2
ftθ ftθ0−ftθ
et−1xt−1. 4.33
Forβ∈NnβAand eachA >0, we have β0−βT
xt2 β0−βT
X1/2n θ0Xn−1/2θ0xtxtTX−T/2n θ0XnT/2θ0 β0−β
≤max
1≤t≤n
xTtXn−1θ0xt β0−βT
Xnθ0 β0−β
≤A2max
1≤t≤n
xTtXn−1θ0xt
.
4.34