Upload
pandurang-m
View
212
Download
0
Embed Size (px)
Citation preview
This article was downloaded by: [University of Glasgow]On: 20 December 2014, At: 10:11Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,37-41 Mortimer Street, London W1T 3JH, UK
Communications in Statistics - Theory and MethodsPublication details, including instructions for authors and subscription information:http://www.tandfonline.com/loi/lsta20
Large sample estimation and prediction for explosivegrowth curve modelsPandurang M. Kulkarni aa Department of Statistics , La Trobe University , Melbourne, AustraliaPublished online: 27 Jun 2007.
To cite this article: Pandurang M. Kulkarni (1987) Large sample estimation and prediction for explosive growth curvemodels, Communications in Statistics - Theory and Methods, 16:9, 2677-2696, DOI: 10.1080/03610928708829532
To link to this article: http://dx.doi.org/10.1080/03610928708829532
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) containedin the publications on our platform. However, Taylor & Francis, our agents, and our licensors make norepresentations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose ofthe Content. Any opinions and views expressed in this publication are the opinions and views of the authors,and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be reliedupon and should be independently verified with primary sources of information. Taylor and Francis shall not beliable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilitieswhatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out ofthe use of the Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematicreproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in anyform to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions
COMMUN. STATIST.-THEORY METH., 1 6 ( 9 ) , 2 6 7 7 - 2 6 9 6 ( 1 9 8 7 )
LARGE SAMPLE ESTIMATION AND PREDICTION FOR EXPLOSIVE GROWTH CURVE MODELS
Pandurang M. Kulkarni
Department of S t a t i s t i c s La Trobe Universi ty
Melbourne, Aus t ra l i a
Key Words and Phrases: asymptotic prediction error; regression wi th come Zated errors; maximwn l i k e Zihood estimation; growth curves as time-series; exp Zosive grouth.
ABSTRACT
Asymptotic d i s t r i b u t i o n s of maximum l ikel ihood es t imators f o r
the parameters i n explosive growth curve models a re derived.
Limit d i s t r i b u t i o n s of predic t ion e r r o r s when the parameters a r e
est imated a re a l s o obtained. The growth curve models are viewed
a s mul t ivar ia te t ime-series models, and t h e usual t ime-series
methods a r e used f o r p red ic t ion . Estimation constrained by a
hypothesis of homogeneity of growth r a t e s is a l so considered.
1. INTRODUCTION
Large sample p roper t i e s of es t imators and t e s t s f o r the para-
meters i n l i n e a r growth curve (or response curve) models with
co r re l a t ed e r r o r s a r e now well documented, see Sandland and
McGilchrist (1979) and Hudson (1983). Hudson (1983), i n p a r t i c u l a r ,
has s tudied the asymptotic p rope r t i e s of c e r t a i n t e s t s t a t i s t i c s
using a t@e-ser ies approach.
The main purpose of t h i s paper i s t o study large sample
p roper t i e s of p red ic t ions when the parameters a r e est imated, and
Copyright O 1987 by Marcel Dekker, Inc.
Dow
nloa
ded
by [
Uni
vers
ity o
f G
lasg
ow]
at 1
0:11
20
Dec
embe
r 20
14
2678 KULKARNI
i n p a r t i c u l a r , t o examine t h e e f f e c t of est imation of t h e para-
meters on t h e asymptotic variance of the predic t ion e r ro r . It w i l l
be shown t h a t f o r t h e s t a t iona ry models (non-explosive regressors)
t h e e f f e c t of est imation i s asymptotically negl ig ib le whereas f o r
t h e explosive growth curve models the est imation e r r o r cannot be
ignored. Large sample p roper t i e s of predic t ion e r r o r s when t h e
parameters a re estimated a r e studied by Basawa (1986) in a
d i f f e r e n t context f o r non-ergodic models when s imi lar phenomenon
regarding t h e e f f e c t of est imation on predic t ion e r r o r variance has
been observed.
In t h i s paper we study two a l t e r n a t i v e growth curve models
based on (a) a regression model wi th autoregressive e r r o r s (Model I)
and (b) an autoregressive model with a regression component (Model
11). Model I was a l so considered by Hudson (1983) in a d i f f e r e n t
context. The two models exh ib i t s imi l a r asymptotic explosive
behaviour while they i n f a c t d i f f e r i n severa l r e spec t s a s regards
parametr iza t ion, asymptotic p rope r t i e s of t h e es t imates , e t c . We
emphasize t h e case when the regression ( x . ) a t time j increase 3
t o , a t e i t h e r a l i n e a r o r an exponential r a t e a s j -t , and
r e f e r t o such models a s explosive. Explosiveness of regressors
e i t h e r l i n e a r l y o r exponential ly in economic analys is is not un-
common. Simi lar types of models have been s tudied by Granger
(1986) who a l s o d iscusses severa l p r a c t i c a l appl ica t ions of such
models.
Glesser and Olkin (1972) have studied a general regression
model s imi l a r t o our Model I. However, i n t h i s model the
regressors do not depend on time ' n ' . For regression depending
on ' n ' ,.Anderson (1971) has considered est imation f o r both types
of models I and 11, discussed i n our paper. Anderson (1971) has
imposed c e r t a i n condit ions on the regression va r i ab les t o
guarantee s t a t i o n a r i t y (non-explosiveness) and t o obta in t h e
asymptotic normality of the es t imators of the parameters. Using
s imi la r techniques we obta in r e s u l t s on asymptotic normality, e t c . ,
f o r explosive growth curve models by su i t ab ly modifying the con-
d i t i o n s imposed on the regressors .
Dow
nloa
ded
by [
Uni
vers
ity o
f G
lasg
ow]
at 1
0:11
20
Dec
embe
r 20
14
EXPLOSIVE GROWTH CURVES 2679
A p r e l i m i n a r y d i s c u s s i o n o f t h e two models is g iven i n
S e c t i o n 2 , and S e c t i o n 3 c o n t a i n s t h e r e s u l t s on l i m i t d i s t r i b u t i o n s
o f t h e maximum l i k e l i h o o d e s t i m a t o r s f o r t h e e x p l o s i v e growth
curves . Asymptotic, j o i n t d i s t r i b u t i o n s o f a f i n i t e number o f
s u c c e s s i v e p r e d i c t i o n e r r o r s a f t e r a s u i t a b l e normal iza t ion a r e
d e r i v e d i n S e c t i o n s 4 and 5 f o r t h e two models. S p e c i f i c examples
a r e cons idered a t t h e end o f S e c t i o n s 4 and 5.
2 . FORMULATION O F THE BASIC EXPMSIVE MODELS
L e t Y denote t h e random o b s e r v a t i o n a t t ime j , j
1 S j S n , { x i j , l ~ i S q 1 a s e t of " f i x e d " r e g r e s s o r s cor responding
t o Y and E be independent N (0 , 0 2 ) v a r i a t e s ( i n n o v a t i o n s ) . j j
Consider t h e f o l l o w i n g models:
Model A . Regression w i t h a u t o r e g r e s s i v e e r r o r s :
where t h e e r r o r s Z a r e assumed t o s a t i s f y a f i r s t - o r d e r auto- j
r e g r e s s i o n r e l a t i o n ,
Model B . Autoregress ion w i t h a r e g r e s s i o n component
eixij + p Y . + E Y o = O . 1 j ' (3)
i=l
The parameters p , a , (Bi , lSi%) , p and c r 2 i n t h e above
models a r e assumed unknown. We assume throughout t h e paper t h a t
t h e a u t o r e g r e s s i o n parameter p s a t i s f i e s j p 1 < 1 and 10 1 < .
The "growth" component i n t h e above models is g iven by t h e
r e g r e s s i o n f u n c t i o n E (Y. ) , and we have 3
Dow
nloa
ded
by [
Uni
vers
ity o
f G
lasg
ow]
at 1
0:11
20
Dec
embe
r 20
14
KULKARNI
f o r model A
la[&] + 1 1 ~i~k- lx i , j -k+ l f o r model B 1-p i=l k=l
Thus f o r model A t he growth is l i n e a r in the parameters while it is
non-linear f o r model B . 2j
v a r ( ~ . lxij) = 0 2
1 1 - P 2 I , f o r both t h e models A and B 3
It w i l l be assumed i n what follows t h a t f o r each i , 0 < xij I. rn a s j + rn . It i s then c l e a r t h a t , f o r both the
models
where x = max{xij ,l 6 i 5 q} . A growth process s a t i s f y i n g (6) j
and z : t a s j + - w i l l be r e fe r red t o a s an explosive process. 3
I n s p i t e of t h e d i f ference i n the spec i f i ca t ion of t h e two
models both t h e processes exh ib i t t he same asymptotic explosive
behaviour i n t h e sense of (6) . Also, f o r both the models it is
c l e a r from (4 ) t h a t
We s h a l l now use the bas ic models discussed above t o formulate
t h e growth curve models i n the following section.
- Note: 0 i n (6) implies Y when divided by X is bounded i n - P j j p robab i l i ty .
Dow
nloa
ded
by [
Uni
vers
ity o
f G
lasg
ow]
at 1
0:11
20
Dec
embe
r 20
14
EXPLOSIVE GROWTH CURVES
3 . GROWTH CURVE: MODELS AND ESTIMATION OF P W T E R S
Let Y U v ( j ) denote the observation a t time j , on the v t h
indiv idual belonging t o uth group, 1 6 u < m , 1 < v < N , and
1 6 j 6 n . The t o t a l number of indiv iduals is given by
N = N1 + ... + Nm , and on each of these N indiv iduals observ-
a t i o n s a t n time p o i n t s a r e made. Assume t h a t ';yij and {ykk1
a r e independent for i f R or j # k . The growth curve models
based on t h e bas ic models A and B of t h e previous sec t ion a re then
defined by the following.
Model I
where
and E ( j 1 a r e independent N (0 ,a t ) var i a t e s . uv
Model I1
YUv(0) = 0 , and E ( j ) a r e a s inModel I . uv
We now consider t he maximum l ike l ihood es t imators of t he
parameters, and ob ta in t h e i r l i m i t d i s t r i b u t i o n s a s n + " .
3.1. Estimation f o r Model I
The log-l ikelihood function based on the observations f o r t h e
u t h group f o r Model I is given by, ( ignoring constant te rms) ,
Dow
nloa
ded
by [
Uni
vers
ity o
f G
lasg
ow]
at 1
0:11
20
Dec
embe
r 20
14
KULKARNI
where
The estimating equations f o r Put Biu , 1 S i 5 q , and pU are
given by
where E ( j ) i s defined by (12). F ina l ly a 2 can be estimated uv u
by
where E ( j ) denotes cUV ( j ) with a l l the parameters replaced by uv
t h e i r estimators.
and
Assume t h a t the following condition i s s a t i s f i e d : Dow
nloa
ded
by [
Uni
vers
ity o
f G
lasg
ow]
at 1
0:11
20
Dec
embe
r 20
14
E X P L O S I V E GROWTH CURVES
(h) lirn{cil/(bibl) '~ = fiL, 1 5 i, k ' q . n*
Remark: Condition l ( a ) ensures t h e exis tence of t he l imi t ing
covariance between t h e mle 's of t h e parameters vU and AU T
fl = (0 Lu , . . . , 8 ) , and condit ion l ( b ) t h a t of t h e covariance Nu qu between
Oiu and 0 . ( 1 , 1 . These condit ions a re
xu needed because of the explosiveness of the regressors . I t can be
seen t h a t even when t h e r eg res so r s a re not explosive these
condit ions a r e enough t o ensure t h e asymptotic normality of t h e
mle 's .
Also, l e t I (n) = diag{n,b b ,bq,n}, where IU(n ) i s U 2 'T. -
a (q+2) x(q+2) diagonal matr ix , and & = (!-I 1€J1ute2u,. . - 1 0 ,P I u qu u
where & i s (q+2) x1 vector of t h e parameters and T denotes t h e
transpose. Let F denote a qxq matrix with ( i , k ) t h element given
by fig . Note t h a t t he diagonal elements of F a re a l l equal t o T
uni ty . Denote M_ = ( f l , . . . , f ) where M i s a q x l vector. The 9
Fisher information matrix corresponding t o the parameter i s
then given by J,(&) ,
where
LA'(&) denotes the matrix of t h e second de r iva t ives of LU wi th
respect t o the components of . It i s e a s i l y v e r i f i e d t h a t
Let $(n) denote t h e es t imate of & obtained by solv ing t h e
equation (13). The following theorem can be v e r i f i e d v i a standard
asymptotics.
Dow
nloa
ded
by [
Uni
vers
ity o
f G
lasg
ow]
at 1
0:11
20
Dec
embe
r 20
14
2684 KULKARNI
Theorem 3.1. Under c o n d i t i o n 1, and Model I , a s n + , and f o r e a c h u , 1 < u < m ,
4 Iu (n) (&in) -&) + Nqi2 (0,~;' (h) ) , i n d i s t r i b u t i o n ,
p rov ided JU is non-singular .
3.2. Es t imat ion f o r Model I1
The log- l ike l ihood f u n c t i o n f o r model I1 based on observa-
t i o n s from t h e u t h group i s g iven by LU
a s d e f i n e d i n (11) , where
t h e e r r o r terms E ( j ) a r e now d e f i n e d by uv
The l i k e l i h o o d e s t i m a t i n g e q u a t i o n s f o r u U r O i u , l S i S q ,
and pU a r e g iven by
1 1 e U V ( j ) = 0 , 1 1 c U V ( j v j v j
and 1 x ~ ~ ~ ( j ) Y ~ ~ ( j - l ) = 0 v j
where E. ( j ) a r e g iven by (15) uv
x ( i , j ) = O , 1 5 i S q , uv
(16)
F o r e a s e of p r e s e n t a t i o n we now
t a k e q = l s o t h a t t h e r e i s o n l y one r e g r e s s o r x U v ( j ) , and (15) is
r e p l a c e d by
Denote a; (n) = x xUv ( j ) , b: (n) = x x2 (j) , uv
v j v j
and suppose t h e f o l l o w i n g c o n d i t i o n i s s a t i s f i e d .
Remark: Condi t ion 2 i s t o o b t a i n t h e l i m i t i n g covar iance between
u and (Burpu) . It can be no ted t h a t c o n d i t i o n 2 i s s i m i l a r u
t o c o n d i t i o n 1 (a) i n S e c t i o n 3.1.
Dow
nloa
ded
by [
Uni
vers
ity o
f G
lasg
ow]
at 1
0:11
20
Dec
embe
r 20
14
EXPLOSIVE GROWTH CURVES
Now, denote t h e m a t r i x o f norms a s fo l lows .
1: (n) = d i a g { n , b: (n) , b: (n) 1 , where I; (n) i s a (3x3) d i a g o n a l
m a t . l e t = (au ,6u ,pu)T be t h e (3x1) v e c t o r o f parameters .
It can be shown, u s i n g T o e p l i t z r s l e m , t h a t t h e F i s h e r in format ion
m a t r i x p e r o b s e r v a t i o n cor responding t o t h e e s t i m a t i o n o f is
where f* i s d e f i n e d by c o n d i t i o n 2 ,
c = l i m i I : a (n) /I: xUV (n) 1 , d = l i m ( 1 a2 (n) /I: x:v (n) 1 , and uv
n* v v n* v uv v n-1
e = lim{E a uv
7 n-@x (R). uv (n) xuv (n) f i x i v (n) 1 where a (n) = e U L P U uv n* v v !L=1
Under t h e assumptions, 0 < xUv (n) + , and c o n d i t i o n 2 , it can
e a s i l y be v e r i f i e d t h a t a l l t h e l i m i t s f*, c , d , and e i n J* u
above e x i s t wi thout f u r t h e r c o n d i t i o n s .
S p e c i f i c examples f o r t h e s e models w i l l be d i s c u s s e d i n t h e
fo l lowing s e c t i o n .
W e s t a t e t h e fo l lowing theorem, which can be proved us ing t h e
u s u a l asympto t ic t h e o r y .
Theorem 3.2. Under Model 11, and c o n d i t i o n 2 , a s n -+ f o r e a c h
u , 1 S u S m , we have
i n d i s t r i b u t i o n , where $ * (n) i s t h e s o l u t i o n of (16) . '-4
We now c o n s i d e r parameter e s t i m a t i o n f o r a s p e c i a l c a s e of
Model I1 under t h e hypothes i s t h a t t h e r e g r e s s i o n parameters f o r
d i f f e r e n t groups are equa l . Denote by Model IIH , t h e model I1
under t h e n u l l h y p o t h e s i s s p e c i f i e d above.
Dow
nloa
ded
by [
Uni
vers
ity o
f G
lasg
ow]
at 1
0:11
20
Dec
embe
r 20
14
KULKARNI
3.3. Model 11, : Parameter E s t i m a t i o n
Consider Model I1 w i t h m=2 ( i . e . two groups) , q = l , a =0, U
02=1, u=1,2, and l e t t h e n u l l hypothes i s 11 s p e c i f y t h a t 8 =8 u 1 2
(homogeneity r e g a r d i n g t h e r e g r e s s i o n p a r a m e t e r s ) . Then, under H ,
t h e parameters o f t h e model a r e 0 , (where O =O = 8 ) , pl and 1 2 P2 '
We s h a l l r e f e r t o t h e model under H by model IIH . It is t o be
noted t h a t t h e e s t i m a t e s of 8 , pl and p2 under H use
o b s e r v a t i o n s from both t h e groups u n l i k e i n t h e case of t h e models
I and 11 , where o b s e r v a t i o n s o n l y from a s i n g l e group a r e needed
t o e s t i m a t e t h e parameters o f t h a t group.
Thus, model 1111 is s p e c i f i e d by
The log- l ike l ihood f u n c t i o n f o r Model I1 i s g iven by H
( i g n o r i n g t h e te rms f r e e from t h e parameters )
T Denote = (p1,p2,8) , and c o n s i d e r t h e ( 3 x 3 ) m a t r i x of norm w
I (n) = d i a g i b * (n) ,b* (n) ,b* (n) + b* (n) 1 where b* ( n ) = 1 Z x k v ( j ) , U 1 2 1 2 U v j
u=1,2 . With E ( j ) s p e c i f i e d by (19) , t h e e s t i m a t i n g e q u a t i o n s uv
a r e
1 ~ ~ ~ ( j ) Y ~ ~ ( j - l ) = 0 , u = l r 2 , and 1 1 1 ~ ~ ~ ( j ) x ~ ~ ( j ) = 0. (20) V I u v j
Le t
n-1 n-1-R where a (n) = 0
uv L P u x (9.) , and l e t u s suppose t h a t uv
Condi t ion 3 is s a t k $ i e d .
Dow
nloa
ded
by [
Uni
vers
ity o
f G
lasg
ow]
at 1
0:11
20
Dec
embe
r 20
14
EXPLOSIVE GROWTH CURVES 2687
Condi t ion 3 1imC.E 1x2 ( j ) / (b; (n)+b; ( n ) ) 1 = a* i s f i n i t e , u=1,2 uv u
n- v j
where b; (n) i s d e f i n e d a s above.
Remark: Note t h a t c o n d i t i o n (3) reduces t o u n i t y f o r t h e c a s e
of s i n g l e group i . e . u=l . We need c o n d i t i o n (3) because it i s
c l e a r t h a t e s t i m a t i o n f o r model I1 needed o b s e r v a t i o n s on ly from
a s i n g l e group, whereas here f o r model 11 e s t i m a t i o n of H
pa rameters i n v o l v e s o b s e r v a t i o n s from a l l groups.
Now, u s i n g T o e p l i t z ' s lemma one can show t h a t t h e F i s h e r
in format ion m a t r i x p e r o b s e r v a t i o n is g iven by
where R* (u=1,2) a r e a s s p e c i f i e d i n (21) and c o n d i t i o n dul u r U
( 3 ) .
The f o l l o w i n g theorem can be v e r i f i e d us ing s t a n d a r d
asympto t ics .
Theorem 3.3. Under t h e Model II and c o n d i t i o n 3, a s n + OD we
have
where (n ) i s t h e e s t i m a t e of $ obta ined by s o l v i n g (20) . N U N U
4 . PREDICTION AND ASYMPTOTIC DISTRIBUTION OF
PREDICTION ERRORS FOR MODEL I
4.1. L i m i t d i s t r i b u t i o n of p r e d i c t i o n e r r o r s
I n t h i s s e c t i o n we s h a l l d i s c u s s t h e p r e d i c t i o n f o r Model I , assuming t h a t t h e f i r s t n o b s e r v a t i o n s a r e known, and t h e n o b t a i n
Dow
nloa
ded
by [
Uni
vers
ity o
f G
lasg
ow]
at 1
0:11
20
Dec
embe
r 20
14
2688 KULKARNI
t h e asympto t ic d i s t r i b u t i o n o f t h e p r e d i c t i o n e r r o r w i t h e s t i m a t e d
parameters .
L e t us denote t h e t - s t e p ahead p r e d i c t o r (1 < t 5 k) o f
Y . ( n + t ) given o b s e r v a t i o n s YUv ( j ) , 1 5 j < n , by uv
& ( n + t ) = E [Y ( n + t ) I h(1) , . . .flU(n) I -u
= E [Y ( n + t ) 1 (n ) 1 , where 5 ( n + t ) i s a N x l v e c t o r -u 4 U
w i t h i t s v t h element given by
L e t (n-kt) dermte & ( n + t ) wi.th a l l t h e parameters r e p l a c e d by
t h e i r e s t i m a t e s . Also , l e t e (11-kt) = 5 (ni . t ) - Y ( n + t ) -u NU hZ1
and z U ( n + t ) = 5 (n+t ) - Yw ( n + t ) be t h e p r e d i c t i o n e r r o r s when t h e W
parameters a r e lcnown and when they a r e es t imated r e s p e c t i v e l y , 2 t
and denote n ( n + t ) = v a r (Y (n+k) ]Y_U (n) ) = L (1-p,, ) a{/ (1-pt) I I , u RU
where . I is an (N xN ) i d e n t i t y mat r ix . Then t h e r e s p e c t i v e u u
s t a n d a r d i z e d p r e d i c t i o n e r r o r s a r e def ined by
Now
Using T a y l o r ' s expansion f o r E; (n+t ) we o b t a i n W
where F;: (n+t ) denotes t h e (N ) x (q+2) m a t r i x of p a r t i a l d e r i v a t i v e s u
of L ( n + t ) w i t h r e s p e c t t o t h e components of and I U b ) i s
t h e m a t r i x of norms d e f i n e d i n S e c t i o n 3.1.
Dow
nloa
ded
by [
Uni
vers
ity o
f G
lasg
ow]
at 1
0:11
20
Dec
embe
r 20
14
EXPLOSIVE GROWTH CURVES
Let 442 (n) be a (KNUxl) v e c t o r w i t h i t s e lements
(n) -+ N [E,B (&) ] i n d i s t r i b u t i o n ,
*u
whpre B ( 4 ) = ( (gts)) is a (KNU X mu) mat r ix w i t h hZ1
I is (NUxNU) i d e n t i t y mat r ix .
The l i m i t d i s t r i b u t i o n o f t h e p r e d i c t i o n e r r o r w i t h parameters
known is t h u s given by (24) .
F u r t h e r , i n o r d e r t o o b t a i n t h e d i s t r i b u t i o n of p r e d i c t i o n
e r r o r w i t h e s t i m a t e d p a r a m e t e r s , denote t h e tth column of GU(n)
by
and assume t h a t t he c o n d i t i o n 4 holds.
Dow
nloa
ded
by [
Uni
vers
ity o
f G
lasg
ow]
at 1
0:11
20
Dec
embe
r 20
14
2690 KULKARNI
Condi t ion 4. lim[w ( i , n ) ] = g U v ( i , t ) i s f i n i t e uv
Remark: Condit ion (4) i s needed t o o b t a i n t h e convergence i n
d i s t r i b u t i o n o f p r e d i c t i o n e r r o r , whereas c o n d i t i o n (1) f o r t h e
convergence i n d i s t r i b u t i o n of t h e e s t i m a t e s o f t h e parameters .
I n g e n e r a l , e i t h e r o f t h e s e c o n d i t i o n s does n o t imply t h e o t h e r .
It is c l e a r i n examples i n t h e fo l lowing s e c t i o n s t h a t t h e s e
c o n d i t i o n s a r e n o t hard t o ach ieve .
Then we have
'I' and & ( i , t ) = ( g u l ( i , t ) ,... g U r N ( i , t ) ) ,
Nuxl u
w i t h guv (i,t) d e f i n e d in Condit ion 4 .
L e t GU = [& , . . . G (K) I and ;$ (n) be a (lUi x l ) v e c t o r w i t h -u -u u
its e lements
Now, us ing (25) , Theorem (3.1) and n o t i n g t h a t t h e two te rms
on t h e r i g h t o f (24) a r e u n c o r r e l a t e d , it is easy t o v e r i f y t h a t
n + N (2, ( ) i n d i s t r i b u t i o n (28) u
- 1 w i t h B (k) , GU and JU a s s p e c i f i e d i n (24) , (26) -(27) and
Theorem 3 . 1 r e s p e c t i v e l y .
Dow
nloa
ded
by [
Uni
vers
ity o
f G
lasg
ow]
at 1
0:11
20
Dec
embe
r 20
14
EXPLOSIVE GROWTH CURVES 2691
Thus, (28) g i v e s t h e asymptot ic d i s t r i b u t i o n o f t h e
p r e d i c t i o n e r r o r w i t h e s t i m a t e d parameters .
4.2. Examples
TO i l l u s t r a t e t h e above r e s u l t s , we c o n s i d e r some s p e c i f i c
examples, f o r Model I .
I n each c a s e we o b t a i n 2, F and G ( t ) , which can r e s p e c t -
i v e l y be used t o g e t J (2) d e f i n e d i n (15) and $ (iU) d e f i n e d
i n (28) . Le t h = [ ( l - ~ ~ ~ ) 0 2 ] - * [1 -p2 l f .
Example 1. Take m=l, N =1 i n model I . u
Case (i). (Exponential growth) x . = A' , A > 1 . I
It can be shown t h a t
- 1 J (A) = cr-2 d i a g [ (1-p ) ', 1, (1-p * ) ] which i s a ( 3 x 3 ) d i a g o n a l
m a t r i x and
T t t
(t) f w
-' ( A 2 - 1 ) . = [ O , g ( t ) , O ] h , where g ( t ) =----- ( A - P )
Case (ii) . (Linear growth) x . = j . I
6 (1- ) T We g e t M = - , F=l , and G ( t ) = [0,0,01 h . 2 l + p
T h i s l e a d s t o t h e conc lus ion t h a t t h e e s t i m a t i o n o f parameters
does n o t c o n t r i b u t e t o t h e asymptot ic v a r i a n c e of t h e p r e d i c t i o n
e r r o r .
Example 2. Le t m=l., N =1, q=2 i n Model I . Let h be a s u
d e f i n e d i n example 1.
j Case (i) . (Both growth components e x p o n e n t i a l ) x . . = 1 . , hi > 1, 11 1
i=1 ,2 we g e t = (0,O) ,
Dow
nloa
ded
by [
Uni
vers
ity o
f G
lasg
ow]
at 1
0:11
20
Dec
embe
r 20
14
KULKARNI
T and G ( ~ ) = [O, g ( i , t ) ,0] h ,
N lSi.52 t t f - 1
where . g ( i , t ) = (Ai-p ) (A:-1) (Ai-p) .
Case ( i i ) . (Exponential and l i n e a r growth components) x =A: , 1 j
A > 1 and x = j . 1 2 j
We o b t a i n = [ 0 , 6 / 2 1 , F = I = i d e n t i t y m a t r i x , and 2x2
t t f -1 G ( ~ ) = [0 , g ( t ) , 0 , O I T h , where g ( t ) = (A-p )(A2-1) (A-p) . N
We may n o t e t h a t t h e c o n t r i b u t i o n s due t o t h e e s t i m a t i o n o f t h e
parameter O2 (corresponding t o t h e l i n e a r growth) i n t h e
asympto t ic v a r i a n c e o f p r e d i c t i o n e r r o r is z e r o , while t h e
e s t i m a t i o n o f O1 corresponding t o t h e e x p o n e n t i a l growth r e s u l t s
i n an i n c r e a s e i n t h e p r e d i c t i o n e r r o r var iance .
5. PREDICTION AND ASYMPTOTIC DISTRIBUTION O F
PREDICTION ERRORS FOR MODEL 11
The l i m i t i n g d i s t r i b u t i o n of p r e d i c t i o n e r r o r s w i t h e s t i m a t e d
parameters f o r Models I1 can be ob ta ined on t h e same l i n e s a s f o r
Model I d i s c u s s e d i n S e c t i o n 4.
5.1. Limit d i s t r i b u t i o n o f p r e d i c t i o n e r r o r s
,We s h a l l use t h e same n o t a t i o n a s i n S e c t i o n 4 , f o r d e f i n i n g
t h e p r e d i c t i o n e r r o r , e r r o r v a r i a n c e , e t c . For s i m p l i c i t y l e t us
t a k e p = l i n Model I1 , and proceed ing a s i n S e c t i o n 4 , we g e t ,
A.
f ;* ( n i t ) = (n+t ) + ( n ) ( t ) I (n) J I n ( n - (29) h21
Dow
nloa
ded
by [
Uni
vers
ity o
f G
lasg
ow]
at 1
0:11
20
Dec
embe
r 20
14
EXPLOSIVE GROWTH CURVES 2693
where e * ( n t t ) = ( n + t ) k ( n + t ) , % ( n + t ) = & (n+t ) - & ( n + t ) , CZ1
5: (n+t ) denotes t h e (NUx3) m a t r i x o f p a r t i a l d e r i v a t i v e s o f
5 ( n + t ) w i t h r e s p e c t t o t h e components of . And , I:(n) -Il
a r e a s d e f i n e d i n Sec t ion 3 . 2 .
Also,
where
L e t us assume t h a t t h e c o n d i t i o n (5) ho lds
Condit ion 5.
S i m i l a r remark a s i n Sec t ion 4 ho lds f o r c o n d i t i o n s ( 2 ) and (5 ) .
Then we have,
where oT = ( 0 , . . -0) , hU a s d e f i n e d above, G N U
Dow
nloa
ded
by [
Uni
vers
ity o
f G
lasg
ow]
at 1
0:11
20
Dec
embe
r 20
14
) = lirnl.2 ( n ) ] , v=1, ... N n-%o uv u -
Assuming t h a t c o n d i t i o n s (3) and (5)
same arguments a s i n S e c t i o n 4 , it can be
h o l d , and us ing t h e
shown t h a t
(1) = r T 1 , - . . n I N B ) i n d i s t r i b u t i o n KN
ICNU u (32)
where B (z) = B (k), B (&) d e f i n e d i n (24) .
Hence t h e asymptot ic v a r i a n c e of p r e d i c t i o n e r r o r s w i t h parameters
known, is t h e same f o r bo th t h e Models I and 11 .
Also,
j%(n) + N , ) ) i n d i s t r i b u t i o n
T -1 where $ (G) = B (,Q + GU J: GU ,
GU = [GU , . . . G:) 1 , G:) l S t U , d e f i n e d i n (31) . and
J*-I i s a s d e f i n e d i n Theorem 3.2. u
We now c o n s i d e r an example f o r model I1 .
Example 4. L e t m=l, N =1, and q = l , i n Model II . u
Case (i). x. = A' , A > 1 . Then we have, 7
u s i n g t h e s e J* (G) d e f i n e d i n (18) can be ob ta ined .
f t-1 t-1 i t-i-1 And g ( t ) = [ O (,A2-1) l i t (A-p)-l + ( t - i ) A p 1
i=l t t f
= e ( A -p ) ( A ~ - 1 ) ( A - P ) - ~ , - 4 t t f , and g * ( t ) = (A -p ) (,A2-1) (A-p)
- 1
Dow
nloa
ded
by [
Uni
vers
ity o
f G
lasg
ow]
at 1
0:11
20
Dec
embe
r 20
14
EXPLOSIVE GROWTH CURVES 2695
By comparing t h i s w i t h Example 1 c a s e ( i ) , it is c l e a r t h a t
t h e asymptot ic v a r i a n c e o f p r e d i c t i o n e r r o r w i t h es t imated
parameter i s d i f f e r e n t f o r t h e Models I and I1 .
Case (ii). L e t x = j . Then c = ~ ( l - p ) - ' , f * = 6 / 2 , d = c 2 , j
e = c . Using t h e s e r e s u l t s J ($$) d e f i n e d i n (18) can be T
o b t a i n e d , and a s expected we g e t g ( t ) = [O ,0,01 h .
Remark: D e t a i l s of o b t a i n i n g t h e asymptot ic p r e d i c t i o n e r r o r
d i s t r i b u t i o n f o r Model 1111 a r e o m i t t e d s i n c e t h e y can e a s i l y be
o b t a i n e d on t h e same l i n e s a s i n S e c t i o n s 4 and 5.
ACKNOWLEDGEMENT
The au thor would l i k e t o thank D r . I . V . Basawa f o r s u g g e s t i n g
t h e t o p i c . Also he would l i k e t o thank t h e e d i t o r and t h e r e f e r e e s
f o r t h e i r h e l p f u l comments.
BIBLIOGRAPFN
Anderson, T.W. (1971). The S t a t i s t i c a l Analys i s of Time S e r i e s New York, John Wiley and Sons, Inc.
Basawa, I . V . (1986). Asymptotic d i s t r i b u t i o n s of p r e d i c t i o n e r r o r s and r e l a t e d t e s t s o f f i t f o r n o n s t a t i o n a r y p r o c e s s e s . To appear i n Annals of S t a t i s t i c s .
Chakravor ty , S.R. (1976). Maxinun l i k e l i h o o d e s t i m a t e s of t h e g e n e r a l growth curve model. Ann. I n s t . S t a t i s t . Math, 28 349-357.
G l e s s e r , L. J. and O l k i n , I. (1972). E s t i m a t i o n f o r a r e g r e s s i o n model w i t h an unknown covar iance mat r ix . Proc . S i x t h Berkeley Symposium, Vol. 1, 541-568.
Granger, C.W.J. (1986). Models t h a t g e n e r a t e t r e n d s . Presen ted a t t h e 26 th summer r e s e a r c h i n s t i t u t e of t h e A u s t r a l i a n Mathe- m a t i c a l S o c i e t y , Canberra , January 1986.
Hudson,, I .L. (1983) . Asymptotic t e s t s f o r growth curve models w i t h a u t o r e g r e s s i v e e r r o r s . A u s t r a l . J. S t a t i s t . , 25, 413-424.
Sandland, R.L. and McGilchr i s t , C.A. (1979). S t o c h a s t i c growth curve a n a l y s i s . B iomet r ics , 2, 255-271.
Dow
nloa
ded
by [
Uni
vers
ity o
f G
lasg
ow]
at 1
0:11
20
Dec
embe
r 20
14
Rec&Lved Novemberr, 1985 ; RevAed June, 1 9 t i 7 .
Recommended AnonymowLy.
Redefiecd Anonymounly.
KULKARNI
Dow
nloa
ded
by [
Uni
vers
ity o
f G
lasg
ow]
at 1
0:11
20
Dec
embe
r 20
14