Probab. Theory Relat.

Fields 92, 275 311 (1992)

P r o b a b i l i t y
T h e o r y ..d
Related Fields
9 Springer-Verlag 1992
Asymptotic expansions of maximum likelihood
estimators for small diffusions via the theory
of Malliavin-Watanabe*
Nakahiro Yoshida
The Institute of Statistical Mathematics, 4-6-7 Minami-Azabu, Minato-ku, Tokyo 106, Japan
Received October 29, 1990; in revised form November 12, 1991
Summary. The asympt ot i c expansi ons of the probabi l i t y di st ri but i ons of statistics
for the small diffusion are deri ved by means of the Mal l i avi n calculus. Fr om this the
second or der efficiency of the maxi mum l i kel i hood est i mat or is proved.
Mathematics Subject Classifications (1980)" 62E20, 62F12, 62M05
1 Introduction
Let X be a diffusion process defined by the stochastic differential equat i on
dXt = Vo( Xt, O)dt + eV( Xt)dwt, t E [0, T] , e~ (0, 1]
( 1 . 1 )
X 0 = X 0 ,
where a k-di mensi onal unknown par amet er 0 e O: a bounded convex domai n of
R k, T is a fixed value, Xo is a const ant , V = ( V1 , . . . , V,.) is an Rd|
smoot h funct i on defined on R e, V0 is an R<val ued smoot h funct i on defined on
Rdx O with bounded x-derivatives and w is an r-di mensi onal st andar d Wi ener
process. The unknown par amet er 0 requires to be est i mat ed f r om t he observat i on
{ X~; 0 < t < T}. We are i nt erest ed in the asympt ot i c propert i es of t he maxi mum
l i kel i hood est i mat or 0~ of 0 when e--* 0. The maxi mum l i kel i hood est i mat or is
consistent, asympt ot i cal l y nor mal and efficient in the first order, see Kut oyant s [6].
The not i ons of the second or der efficiency of est i mat ors were i nt r oduced by
R.A. Fisher, C.R. Rao [11], [12], Takeuchi - Akahi r a, and Ghos h- Subr a ma nya m
[2] mai nl y for i ndependent observations. Aka hi r ~Ta ke uc hi [1] ment i ons t o an
aut oregressi ve process. Tani guchi [13], [14] studied for Gaussi an ARMA pro-
cesses. When we consi der Takeuchi - Akahi r a' s cri t eri on of the second or der effici-
ency of estimators, the requi red mat hemat i cal tools are the asympt ot i c expansi ons
of t he pr obabi l i t y di st ri but i ons of the est i mat ors and some rel at ed statistics.
For t he small diffusion we will pr ove the asympt ot i c expansi ons of probabi l i t y
di st ri but i ons by means of the Malliavin calculus, ext endi ng Yoshi da [18]. For this
* The research was supported in part by Grant-in-Aid for Encouragement of Young Scientists
from the Ministry of Education, Science and Culture
276 N. Yoshida
theory see Malliavin [9], [10], Watanabe [16], Kusuoka- St r ook [5], Ikeda and
Wat anabe [3], [4]. Wat anabe [15] introduced the notion of the generalized
Wiener functional, the pull-back of Schwartz distribution under Wiener mappings,
and in his celebrated work [17] he exploited the asymptotic expansion of the
generalized Wiener functionals in some Sobolev space and derived from this
various expansion formulas for heat kernels. In the present paper this method will
prove to be useful to show quite directly the asymptotic expansions for statistical
estimators.
In the next section, we present the results used later on. Section 3 gives
expansion formulas for the bias corrected maximum likelihood estimator and the
likelihood ratio statistic. In Section 4 we present non-degeneracy of the Wiener
functionals and we will derive the asymptotic expansions of the probability distri-
butions of statistics considering the composite functions of the indicator functions
together with those statistical Wiener functionals. In the last section the second
order efficiency of the maximum likelihood estimator is proved.
2 Fundame nt al resul ts
Let ( W, P) be the r-dimensional Wiener space and let H be the Cameron-Martin
subspace of W endowed with the inner product
T
( hi , h2) = ~/~1,t"/~2.t dt
0
for hi, h2~H. For a Hilbert space E, I[' lip denotes the LV(E)-norm of E-valued
Wiener functional, i.e., for Wiener functional f : ( W, P) ~ E
Ihfl[~= ~ ]flf~P(dw).
W
Let L be the Ornstein-Uhlenbeck operator and define I I f Ilp,s for E-valued Wiener
functional f, s e R, p e (1, oe ) by
[l/lip,, = H(I - L)S/2f[lp.
The Banach space D;(E) is the completion of the totality P( E) of E-valued
polynomials on the Wiener space ( W, P) with respect to I I " I l p , s . Let D~(E) be the
set of Wiener test functionals of Wat anabe [-16]:
Then,
and
s >0 l <p<m
E)=U U
s >O l <p< cc
0
s >O l <p<~
Asymptotic expansion for small diffusion 277
are the spaces of generalized Wiener functionals. Moreover, let
N U D;(E).
s >0 l <p< co
We suppress R when E = R. The Fr6chet space S(R e) is the t ot al i t y of rapidly
decreasing smoot h functions on R e and S' (R e) is its dual. The space Cak,
k = 0 , __1, + 2 , . . . i s the completion of S(R e) with respect to the nor m
It u ll2k =s upx I A k u ( x ) l , w h e r e A = 1 + I x l 2 - 8 9 We owe t h e f o l l o w i n g t h e o r e m
to S. Wat anabe.
Theorem 2.1 Let F e D~ ( R a) and ~e D ~. Let t~: R ~ R be a smooth function such
that 0 < O(x) < l for x e R , O(x) = l for Ixl _-< 89 O(x) = O for Ixl _-> 1 . Suppose
that for any p e(1, oo ), the Malliavin covariance ae o f F satisfies
(2.1) E[I{Ir __<1} (detov) -p] < o0.
A
Then, there exists a linear mapping Te S ' ( R e) --+ TED- ~176 satisfying the following
conditions:
(1) / f Te S( R e ) then T = t f l ( ~) T( F) eD ~~
(2) for k = O, 1, . . . and pe(1, go) there exists a constant C(p, k) such that
II T l l p , - 2 k < C(p, k) II T [ 1 - 2 k
for Te C_ 2k. This mapping is uniquely determined.
Proof. Let Te S(Ra). Using i nt egrat i on by parts formula, which is applicable by the
condi t i on (2.1), we see t hat for J e D ~~
[( O( ~) T( F) , J ) I = I ( t P( ~) Ak A- k T( F) , J ) I
= ] ( A- k T( F) , e ( w; J ) ) l < C][ Tll-2kllJllq.2k
for a smoot h functional 4~(w;J), any q > 1 and some C> 0 . So t hat for
k = 0, 1 , . . . and p > 1 there exists C(p, k) > 0 such t hat
II ~'(~) T( F ) li p. - 2k <---- C (p, k)II TI 1 - 2 k
for TeS(Re). Defining ;r as the cont i nuous extension of O( ~) T( F) we have the
result. []
If F is nondegenerat e in the usual sense of Malliavin,
D-~ ( ?, J )D ~ = D-~ ( T( F) , O( { ) J )D ~
for J e D ~~ Thus 7 ~ is denot ed by ~( ~) To F or ~({) T( F) if there is no confusion.
Let us consider a family of E-valued Wiener functionals (or generalized Wi ener
functionals) {F~(w)}, ge(0, 1). For k > 0 if
lim ]lF, llv,s < go
~SoUP ~
we say f ~( w) = O(e k) in D~(E) as aS0. It is said t hat F~( w) eD~( E) has the
asympt ot i c expansion
F,(w) ~f o + ef, +' "
278 N. Yoshida
in D ~(E) as e ~ 0 with f0, f l . . . . ~ D ~ (E) if for every p > 1, s > 0 and k = 1, 2 . . . .
F~(w) - (fo + ~f~ +" " + ~- ~A- ~) = 0(~ ~)
in O~p(E) as e $ 0. We say t hat F~(w)~ D~( E) has the asympt ot i c expansion
F,(w) ~f o + 8f~ +. . .
i n/ )~~ as e ~ 0 wi t hf o, f~ . . . . ~/ )~176 if for every s > 0 and k = 1, 2 . . . . there
exists p > 1 such t hat F~(w), f o, f l . . . . ~D~(E) and
F~(w) - (fo + efl +" " + ~ k - l f k - 1 ) = 0 ( ~ k)
in D~p(E) as e J, 0. Moreover, we say t hat F~( w) e D- ~( E) has the asympt ot i c
expansion
F~(w) ~ f o + efl +" "
in D- ~ as e+0 wi t hf o, f ~ , . . , e D- ~1 7 6 if for every k = 1 , 2 , . . . there exist
p > 1 a n d s > 0 such t hat F ~(w), f o, f~ . . . . eD ~( E ) and
F~(w) - (fo + ef l +" " + ~k-l f k- 1) = 0(~ k)
in D~ ( E ) as e+0. Similarly, we say t hat F~( w) e D- ~( E) has the asympt ot i c
expansion
F~(w) ~ f o + ef~ +" "
in / ) - ~176 as e$0 with f 0 , f ~ , . . . e / ) - ~ ( E ) if for every k = 1,2 . . . . there exists
s > 0 such t hat for every p > 1 F~(w), f o, f l , 9 9 . ~D; ~( E) and
F~( w) - (fo + efa +" " + ek- l f k- 1) = O(e k)
in DpS(E) as e $ 0. The generalized means of these expansions yield the ordi nary
asympt ot i c expansions.
The following t heorem will be our fundament al tool.
Theorem 2.2 Let ~b be a function defined in Theorem 2.1. Let A be an index set.
Suppose that families {F~(w); eE(0, 11} c D~(Rd), {~(w); ~6(0, 11} ~ D ~~ and
{ Tz; 2 E A } ~ S' (R d) satisfy the following conditions.
1) For any p ~ (1, oo )
sup E[ l {l : j _<l ) ( det a~) -p] < Go .
~ ( 0 , 1]
2) {F~(w); e~(0, 13 } has the asymptotic expansion
F~( w) ~f o + ~j'z +' ' " in D~( R d) ase,LO
with f~ ~ D 0o (Rd).
3) {~(w); e6(0, 1]} has the asymptotic expansion in D ~ as e,~ O.
4) For any n = l , 2 . . . . .
lira s-~ P{'~.~] > ~} = 0 9
~$0
Asympt ot i c expansion for small diffusion 279
5) Fo r any n = 1, 2 , . . . , t he r e e x i s t s a n o n n e g a t i v e i nt e ge r m s uch t hat A - m Tz e
C~( Ra) f o r al l ) o e A and
sup Y' I I O " A - m T ~ I I ~ o < 0 0 ,
~ A In/ < n
wh e r e n = (na . . . . . nd) is a mu l t i - i n d e x , ]hi = nl + " ' " + na, c ~n = 0 1 1 . . . c~ d, c~i =
O/ ~x ~, i = 1 , . . . , d. Th e n t he c o mp o s i t e f u n c t i o n a l r ~ ) T~(F~) ~ D - ~ is wel l -
de f i ne d and has t he a s y mp t o t i c e x p a n s i o n
r ~ ' ~ , o + e , ~ , ~ + " " in ~- ~ as ~0
uni f or ml y in 2 ~ A wi t h Cz, o, r a . . . . e / ~ - ~ d e t e r mi n e d b y t he f o r ma l Ta y l o r
e x p a n s i o n
1
T z ( f o + [ s f a + e2f2 + ' " "]) = ~ ~ . r T z ( f o ) [ s f a + e2f2 + ' " "]"
n
= r + ~r + " ' ' ,
wh e r e n! = nl ! . . . he!, a n = a] ~ . . . a"J f o r a s R d. I n par t i cul ar ,
cP~,o = T ~ ( f o ) ,
d
9 ~,~ - - ~ f ~ 8 ~ T ~ ( f o ) ,
i = 1
4~, 2 = f ~ ~?~ Ta ( f o ) + ~ f ~ f { 0~ Oj T z ( f o ) , . . . .
i = a i , j =a
P r o o f We f o l l o w t h e p r o o f o f T h e o r e m 2.3 o f Wa t a n a b e [ 1 7 ] . F i x
k = 1, 2 . . . . a r b i t r a r i l y . F o r Tz ~ S' ( Rd) , t h e r e e xi s t s a p o s i t i v e i n t e g e r m = r e ( A, k)
a n d r C~( R ~) s u c h t h a t T~ = A ~ Cz f o r al l 2 ~ A . F o r J ~ D ~,
D ~ ( ~ ( ~ ) Tz ( F~) , J>D ~ = E [ O; ~( F, ) l ~( J) 3
b y t h e i n t e g r a t i o n b y p a r t s f o r mu l a , wh e r e
2 m
l ~( J) = ~ < P ~ , i , D ' J ) H | |
i = 0
a n d P~,i i s a p o l y n o mi a l i n F~, ~ ( ~ ) , 7(~) = ~/7~ ~, L - d e r i v a t i v e s o f F , a n d t h e i r
H- d e r i v a t i v e s . He r e we u s e d t h e c o n d i t i o n 1). F r o m t h e r e g u l a r i t y o f q~,
0 z ( F ~ ) = ~ _1 an r [ F~ - f o ] " + V k z ~ ,
I n l _-< k 1 n ! ' '
t h e n f o r p ' ~ ( 1 , oo) , t h e r e e xi s t s a c o n s t a n t ca i n d e p e n d e n t o f 2 ~ A a n d ~ ( 0 , 1]
s u c h t h a t
II V~ . ~ , ~ l l ~ , - 5 c ~ ~
f o r e e (0, 1]. L e t q' s a t i s f y l i p ' + 1/ q' = 1 a n d l e t q > q'. T h e n
Ill~(J)llq, <: c2ll J Ilq,2m
280 N. Yoshida
for all e~(0, lJ, J e D ~, where c 2 is a const ant i ndependent of 2EA, e6(0, lJ and
J ~ D ~. Therefore,
I g [ v k , Z,~l~( J ) ] l < II gk,~,zllp,]ll~( J)lla,
<-_ cl c2 [I J Ilq, zme k
for at ( 0, 1] and J f f D ~176
On the other hand, as [F~ - f o ] n l ~ ( J ) has the asympt ot i c expansion,
2 -1 ~ " r = Z ~ o +e Z ~ , ~ + 9 9 9 +~ - I Z ~ , ~ _ , +U~ ,
In[ _-< k - 1 n ! ' ' '
where
2 m
Zx, i = ~ ( Qx , l,j, O J J ) u | 1 7 4 i = O , . . . , k - 1 ,
)=o
Q~,i,j is a H| | pol ynomi al in c ~n~z ( f o) , f o, f l . . . . . ~Po, ~ i , . . .
( O( ~ ) - Oo + e O, + ' " ) , Vo , ? ~ , - - . ( 7 ( e ) ~ ? o + e T ~ + ' " ) , L-derivatives of
F~ and their H-derivatives; and
I g [ u ~. , ~, ~] I = c ~ ~ II J I1~, 2m,
where c3 is a const ant i ndependent of 2, e and J. Let
(2.2) eb~,i = ~ D *~ Qx, l , j ,
J
t hen
By the duality,
k - 1
I D- ~( O( ~ ) Tz ( F~), J )D ~ - - D ~( ~i =0 ei ebz, i, J )D~I
_-< IE[U~, ~, k31 + I E [ V~, ~, ~t ~(J)]l
< (CLC2 + c 3 ) ~ l l J IIq,2m.
k - 1
]I ~P( ~) Tz( F~) - ~ ei~z,i]jp,-2,, < ( cl c2 + c3)e k 9
i = 0
Here we t ake p such t hat l i p + 1/ q = 1. p ~ (1, oo ) is arbitrary, so t hat
t p ( ~ ) T ~ ( F ~ ) ~ rbz, o + eebz, 1 + . . . in D - ~
with ~bz, o, ~z, 1 . . . . ~/~-~o uni forml y in 2~A. Define the mappi ng T~
S' (R d) ~ c b i ( T ) ED - ~ as in (2.2). Again by the argument of duality, we see t hat
for positive integer m', there exist s > 0, p~(1, oo) and C > 0 such t hat
II ~ i ( T ) Il p, -s _-< c IT TII- 2m' for Te C_ 2 m " For r s S(R a) let 4~, ~ , . . . be the coef-
ficients defined by the (formal) Tayl or expansion of r + [F~ - f o ] ) . Then,
O ( ~ ) ~ ( F ~ ) 0 ( )~ + + )
0 5 ; + e , P i + - . 9
in D ~. The last equivalence is by the fact t hat for any pe(1, oo ) and any s > 0,
FI 1 - 0 ( ~ ) I 1 ~ , ~ = o ( e " )
Asymptotic expansion for small diffusion 281
for n = 1, 2 , . . . . Therefore, cb~(qS) = ~b~, i = 0, 1 , . . . . By cont i nui t y we have t he
result. []
Le mma 2.1 For n = 1, 2 , . . . , there exi st s a positive integer m such that
sup ~ II~"A-mlBllo~ < ~,
B e B d I n l _-< n
where B d denotes the Borel a-field o f R a.
Pr oof The oper at or A- a is an i nt egral oper at or
A - i f ( x ) = ~ A( x , y ) f ( y ) d y ,
R a
wi t h kernel
A( x , y) = ~ e - ' p l ( t , x, y ) dt ,
0
where pl ( t , x, y) is t he t r ansi t i on densi t y of a Br owni an mot i on cor r espondi ng t o
the f or war d equat i on
0pl _ 1
c?~- - ~ Apl - l Y l 2 p l ,
see pp. 474- 475 of I keda and Wa t a na be [4]. We not e t hat for B s B d
~ " A- " I B( x ) = ~ 3 ~ A - " 6 r ( x ) d y .
B
I nt egr abi l i t y of O~A " 6y(x) follows f r om a di rect est i mat e usi ng above r epr esent a-
t i on of A- 1. Then, we see t hat
C
S I ~ A - " 6 y ( x ) l d y < (1 + Ixl) p
R a
for some m and p > 0, whi ch compl et es the proof. []
The composi t e funct i on of a measur abl e funct i on and a Wi ener funct i onal has
a usual meani ng.
Le mma 2.2 For O, 4, F given in Theorem 2.1 and any measurable f unct i on f ( x ) o f
polynomial growth order,
~ ( ~ ) f o F = O( ~. ) f ( F)
in D - ~ .
Pr o o f For si mpl i ci t y let d = 1. For any measur abl e funct i on f ( x ) of pol ynomi al
gr owt h or der and any e > 0, t here exist k e N and ~beS(R 1) such t hat
and hence
sup [ A - k f ( x ) -- q~(x)] < e
x ~ R 1
I l f - AkqS]l-2k < ~.
282 N. Yoshida
I n fact, this f ol l ows f r om t he i nequal i t y
I A - l f ( x ) l < e - t d t p l ( t , x , y ) d y
0 --cO
I - ] 1 / 2
x Ye- t dt o -co~ P l ( t ' x ' y ) f 2 ( y ) d y J
- - e - - y ) d y .
- co , , / 2
Ther ef or e, t her e exi st a s equence ~b,(x)~ S( R 1) a nd s ome k > 0 s uch t hat q~, ~ f i n
C_2k. So we have
O ( ~ ) f o V = O ( ~ ) f ( F )
i n / ) - ~ by def i ni t i on of c ompos i t e f unct i onal s. D
3 Expansion of statistics
Let X ~' 0 be t he s ol ut i on of t he s t ochas t i c di fferent i al e q u a t i o n (1.1) for 0. Let P~, 0 be
t he i nduc e d me a s ur e f r om P on C( [ 0, T] , R d) by t he ma ppi ng w -+ X ! ' ~ The
Ra d o n - Ni k o d y m der i vat i ve of P~, 0 wi t h r es pect t o P~, 0o is gi ven by t he f or mul a (e.g.,
Li pt s er a nd Shi r ya ye v [ 8] )
A~(O; X) &( 0o; x) -~ ,
wher e
A~(O; X) = exp e - 2 V ' o ( V V ' ) + ( X , , O) d X ,
201i ~. -2V~o(gV' )+ V o ( X , , O ) d t }
He r e A de not e s t he Mo o r e - P e n r o s e gener al i zed i nverse mat r i x of mat r i x A. We
a s s ume t ha t Vo( x , O) - Vo( x, 0 o ) ~ M{ V(x)}: t he l i near mani f ol d ge ne r a t e d by
c o l u mn vect or s of V(x), f or each x, 0 a nd 0o. Wh e n 0o e O is t rue, t he ma x i mu m
l i kel i hood e s t i ma t or O~(w; 0o) is def i ned by
A~(J~(w; 0o); X~' ~1 7 6 = ma x A~(O; X ~ ' ~ 1 7 6
0~0
Xt de not e s Xc . Let X ~ be Next , we pr e pa r e sever al not at i ons . Fi x a ny 0o ~ O a nd ~ ~.0o
t he s ol ut i on of t he or di na r y di fferent i al e q u a t i o n
d X ~ Vo( X~ t~EO, T]
dt
X 0 = x 0 .
Asymptotic expansion for small diffusion 283
Let an Ra| process Y ~ ( w ) be the solution of the stochastic differential
equation
d Y~ ~ 0o ~
=0Vo(Xt' , O o ) Y f d t + e c 3 V ~ ( X ~ ' ~ 1 7 6 d w ~ , te[0, T] ,
c ~ = l
Y~o = I a ,
where [ ~?V~] i j = ~ j V i , Oj = ~?/c3x j , i , j = 1 , . . . , d , ~ = O, 1 . . . . r . Then, Yt:= yO is
a deterministic Re| process. For a function f ( x , O) we abbreviate
f ~ ( O ) = f ( X ~ , 0 ) ,
a ~ f ~ ( O ) = a ~ f ( X ~ , O)
and similarly ( S i f t , 6 / S t f ~ ( O ) , c 3 ~6 j f ~( O) . . . . , where 6 j = ~?/c?O ~. It is known that
,,~ 0o ~X~' 0o
~a t ' is smooth. In particular, Dr.- ~ ~=o satisfies the stochastic differen-
tial equation
d D r = c 3 V ~ t ( O o ) D r d t + V ~ d w t , t el 0, Tl ,
D o = O .
Then, Dt is represented by
t
D r = ~ Y t Y s o d w ~ , t e [ O , T ] .
0
Put
O" 0
as = y; 1 vo vs (vs v~ ~v~
= Y21 ~ v%(Oo),
where I r a ( O) - ] j l = ~3aJ(O)/OOl, j = 1, . . . , d , 1 = 1 , . . . , k , for a( O) = ( a l ( O ) . . . . . a~(O)) '.
The Fisher information matrix I ( 0 o ) = ( I i j ( O o ) ) is defined by
T
I ~ i ( 0 o ) = ~ 6~ V ~ t ( Oo ) ' ( V V ' ) + ~ f j V ~
0
for i , j = 1 , . . . , k . Let
T
m( ( i i , . . . , i p ) / O) = ~ [ ~ i l . . . ~ip g ' o ( g g ' ) + g]~
0
T d
m ( ( i l . . . . . iv)~1 ) = 5 ~ [~,{6~1... 6~, V ' o ( V V ' ) + V } ] ~
0 1 = 1
T
n ( ( i ~ . . . . . i p) ( j ~ . . . . . j q ) / O ) = S [ ~ . . 9 6~, V ' o ( V V ' ) + g)J, " " 6 J , V o l ~ d r ,
0
and
T d
n ( ( i 1 . . . . . i p) ( j l , . . . , j q ) / 1 ) = ~ E 1 - ~ / { 6 t l " ' ' (~ip Vt o( V V t ) 4r
0 / = 1
~ J l . . . ~Jo V o } ] ~ a t -
284 N. Yoshida
L e t 6 ~ = 6 p a~ ~ . . . 6~ ~ a n d l et I v l = v~ + Y 2 -~- " " " + Y k f o r v = ( v 1 , Y 2 , 9 9 9 , Y k ) . I n
t hi s a r t i c l e we a s s u me t h e f o l l o wi n g c o n d i t i o n s .
(1) Vo, V a n d ( V V ' ) + a r e s mo o t h i n (x, 0).
(2) T h e r e exi s t s a c o n s t a n t C s u c h t h a t
I No(x, 0)1 ~ c 0 + Ixl)
f o r al l x a n d 0.
(3) F o r Inl >= 1, F = Vo , V , ( V V ' ) + , s u p ~ , o t O " F I < c o .
(4) F o r I v l _>- 1 a n d I n l _ -> 0 , a c o n s t a n t C~,. exi s t s a n d
s u p 10"6 ~ gol ~ C~,.(1 + I x l ) Q . ,
0
f o r al l x.
(5) I ( O) , 0 s O, a r e p o s i t i v e def i ni t e. F o r 0o e O, t h e r e exi s t s ao > 0 s u c h t h a t
T
[ V ~ - V ~ V V ' ) + ~ 1 7 6 - V~ d t > ao[O - 0ol 2
0
f o r 0 s O .
R e m a r k 3. 1. T o d e r i v e t h e r e s ul t s i n t hi s ar t i cl e, we c a n r e l a x t h e a b o v e c o n d i t i o n s .
By l a r g e - d e v i a t i o n a r g u me n t , t h e y c a n be r e p l a c e d b y a c e r t a i n s et o f r e g u l a r i t y
c o n d i t i o n s a b o u t V0 a n d V n e a r t h e n e i g h b o r h o o d o f t h e p a t h o f X ~
L e t
1~(0; 0o) = l ogAn( 0; X ~'~176 .
F o r ma t r i x A = (AU), l et l A 12 = 521,~1A ulz. C h o o s e 7 s o t h a t 0 < ~ < 89 L e t too, no s
N s at i s f y rno > ~ + 2. T h e S o b o l e v s p a c e wm~176 is t h e B a n a c h s p a c e en-
d o we d wi t h t h e n o r m
) l / 2 n o
2no
II u II r v . . . . . . ( o ) = Y ~ II 6 " u II L 2 , o ( O ) ,
Inl < mo
wh e r e L p ( O ) d e n o t e s t h e Lp s p a c e o n O wi t h r e s p e c t t o t he Le b e s g u e me a s u r e .
Th e n , b y So b o l e v ' s l e mma , t h e i n c l u s i o n W" ~ 1 7 6 c C 2 ( 0 ) is c o n t i n u o u s , t h a t
is, t h e r e exi s t s a p o s i t i v e c o n s t a n t C (mo, 2no, 2, O) s u c h t h a t
I t u [I c; () < C ( mo, 2 n o , 2 , O ) I I u II w . . . . . . ( 0 )
f o r u s Wm~176 wh e r e
]lullEd(O) = ~ sup l a " u ( 0 ) l .
Inr _-< 2 0 s o
F o r 0o s O l et
T
Q-ilz(W, e, O) = ~ &j ~'~o,t(O)'( V V ' ) +~ &z I Pf ) , , ( 0) dt ,
0
j , l = 1 . . . . . k, wh e r e
1
6 P S , , ( o ) = .[ 6 v M ( O o + u ( O - O o ) ) d u
o
As y mp t o t i c expans i on f or small diffusion 285
a n d l e t Ql ( w, e, 0) = ( Q{ l ( w, ~, 0) ) . M o r e o v e r , l e t
T
Q . ~( w , ~, o ) = ~ E W o , , ( o ) - ~ ' ' +~ V o , t ( O o ) ] ( V V ) t [ V~o, t(O) - Veo,t(Oo) "] d t .
o
W h e n ~ = 0, Q I ( W , O, O) a n d Qz ( w, O, O) a r e d e t e r m i n i s t i c a n d d e n o t e d b y Q t ( 0 , 0)
a n d Q2 ( 0 , 0), r e s p e c t i v e l y . F o r ~ / e ( 0 , 1), d e f i n e
a n d
f {t(w, e, r h O) = Q~( w, ~le, O) - Q{t(O, 0 ) ,
f 2 ( w , e , rl, O) = Q 2 ( w , ~ , O) - ( 2 2 ( 0 , O)
T
f 3( w, ~, tl, O) = ~ i - 2~ ~ [ V"o~t(O) - Vno~t(Oo)]' ( V V ' ) +~ V"t"dwt .
0
F o r 0o ~ O a n d c > 0 l e t
k 3
c W 2 n o
r , , , ( ) = c ~ I l f { Z ( w, ~ , t l , ' ) w . . . . . . ( o ) + c ~ I I f ~ ( w, ~ , r / , ' ) 2~o
W,.O,2~ .
j , l = l i = 2
F o r e a c h e s ( 0 , 1 ] , c > 0 a n d w ~ W, t h e f u n c t i o n ~ / ~ r ~ , , ( w) , t / s ( 0 , 1), i s s m o o t h .
L e t
Rg( w) I l r L ( w ) 2
= I I w i , ~ ( ( o , I ) ) 9
T h e r e e x i s t s a p o s i t i v e c o n s t a n t C ( 1 , 2, 0 , ( 0 , 1) ) s u c h t h a t
II U IlCt,((O, 1)) ~ C ( 1 , 2, 0 , ( 0 , 1))II u I} wl.~I(oA))
f o r u e W1' 2( ( 0, 1)), w h e r e I l u l l c ~ , o , l ~ = s up, ~( O, l ~ l u @ l . L e t
Co = C( mo, 2 n o , 2, O ) C ( 1 , 2, 0 , ( 0 , 1)) 1/2"~ .
T h e r e e x i s t s d l > 0 s u c h t h a t {0; l 0 - 0ol < d l } c O ,
1
s u p I Q l ( O , O ) - I ( O o ) l < ~ 2 1
1 0 - 0 o l = < d l
a n d
I ( 0 o ) 1
s u p 6 2 Q 2 ( 0 , O) - < ~ 21 ,
IO-Ool <dx
w h e r e 21 = i nf l r I = 1 ~' 1( 0o) ~, t h e m i n i m u m e i g e n v a l u e o f 1(0o).
L e m m a 3 . 1 L e t Co s at i s f y C o c o p~ < m i n { ~ 2 1 , 88 a o d 2 } , wher e Pl = 2@o. Suppos e
c > Co. Then,
( 1) I f e ~ < d l and R~( w) < 1, a ma x i mu m l i kel i hood e s t i mat e 0~,(w; 0o ) e x i s t s and is in
( 0 ; 10 - 0o1 < e ' ~} f or e' < e.
( 2) I r e ~ < d l and R~( w) < 1,
s u p ~ ' ~ 2 z ' 2 l e ' ( O ; 0 o ) ~ < 1
[ r = - ~ 2 1 '
I 0 - - 0 o 1 _-<d~
f o r ~' < ~.
(3) I f e ~ < dl and R~( w) < 1, t he ma x i mu m l i kel i hood e s t i mat e O~,(w; 0 o ) is a uni que
s ol ut i on in { 0 ; 10 - 0o1 < d r } o f t he equat i on cSl~,(O; 0o) = O f o r e' < e.
286 N. Yoshida
(4) L e t Of ( w) = O( R a ~ ( w) ) i f ~ < d l a n d l et O~( w) = 0 i f e ~ > d l . T h e n ,
0 < O~( w) < 1 a n d ~ ( w ) 6 D ~176
(5) 0~(w; 0o) c a n b e e x t e n d e d t o a f u n c t i o n a l o n W a n d O~(w)0~(w; 0 o ) ~ D~ ( Rk ) .
(6) F o r a n y n = l , 2, . . . , O~( w) = l - O( e ") in D ~ as e $ O.
Pr o o f . (1) S u p p o s e e ~ < d l a n d R ~ ( w ) < 1. By d e f i n i t i o n o f RC( w) a n d So b o l e v ' s
i n e q u a l i t y , we h a v e
s up [I QJll( w, e', ") - Q{l(O, ) l l c ~( o ~ <= C o e - ~ ,
a n d
s up 1] Q2 ( w, e', " ) - Q2(O, ' )]lc~(ot <= C o c - p ~
W
e ' I - 2 ' ~ ~ [" # ' V~ ' d wt
Vo,~(Oo)3 ( v v ' ) ? ~' < Vo, ~( ) - ~' ' = C o c - ~ ' s u p D
e'<=~ o C~(O)
Si nce c > Co, f r o m C o n d i t i o n (5), we see t h a t
i nf e' - 2~Q2(w, e', O) > i n f
d a <1 0 - 0 o l d l < 1 0 - 0 o [
~ ' - ~ ' [ Q 2 ( O , 0 ) - C o c - ~ ]
> e - 2 ~ [ a o d 2 _ Co c - P ~ ]
= > ~ a o d ~
a n d
i n f
#~<[O-Oo[<dl
e ' - z r Q2 ( w, e', 0 ) = i n f
~'~<_lO-Ool <_dl
~" <<g
8 ' - 2 7 ( 0 - - O0)' Q I ( W , F/, O) ( 0 - - 0 0 )
>
i n f
I r
g'<=lO-Ool<=d~
~ ' Q l ( w , e ' , 0 ) ~
> i nf
I~1=1
e ~< 0 - 0 o <d
' = 1 I = 1
~' QI ( O, 0 ) 3 - k C o c -p~
1
>-_ ~ it1 - k C o c -p~
3
Th e r e f o r e ,
i nf
e',__< [ 0 - 0 o I
e ' - 2 ~ Q 2 ( w , e', O) > mi n a o d 2 , g q = : C1 9
Asymptotic expansion for small diffusion 287
Then,
sup
~'"_-<- 10- 0ol
0 e O , g < ~
1
e'2(~-~)l-l~,(O; 0 o ) - I~,(0o; 0o)3 ~ Co c - p~ - ~ Ca
1 2
< ma x { - g a o d l , 1 ),
- 1 6 ~ j < 0 .
Therefore, if s < d~ and R ~ ( w ) < 1, there exists a maxi mum likelihood estimate
O~,(w; 0o)e{0; 10 - 0o1 < e '7} for e' < e. This proves (1).
(2) If e 7 < dl and R ~ ( w ) < 1,
Consequently,
sup ]~'zo21~,(0; 0o) + I(0o)l
1 0 - 0 o l <d~
1 I ( 0 o )
< = 2 C o c - V * + sup - ~62Q2(0, 0) +
[O-Oo]<=a~
1
<=~ ,h
1
sup ~' (~2~'21e'(0; 00) ~ ~ -- ~ ~i 9
IO-Ool <dt
(3) Since 0 ~ l~,(O; 0o) is smooth, 0~,(w; 0o) is a root of the estimating equat i on
6 I ~ , ( 0 ; 0 o ) = 0 . Fr om (2) we see t hat this root is a unique solution in
{ I o - O o i =< d l } .
(4) is easy.
(5) Let 0~(W;0o)=0o if s or s and R ~ ( w ) > l . Suppose e y < d l .
For h e H , if R~( w) <1 , there exists t o ( w , h ) > 0 such t hat R ~ ( w + t h) < 1 for
Itl < to(W, h). Then, IO~(w + th; 0o) - 0ol < e ~ and
(~l~(O~(w + th; 0o); 0o) [w + t h ] = O .
By the uni form non-degeneracy of the bilinear forms e262l~(0; 0o) and the Tayl or
formula, we have
10~(w + th; 0o) - 0~(w; 0o)l < 22111e281~(O~(w + th; 0o); 0o) [w]l
= 22;ZleZ61~(O~(w + th; 0o); 0o) [w]
- e261~(O~(w + th; 0o); 0o)[w + t h ] [
(3.1)
< 22i -1 sup]e2(~/~(0; Oo)[W] - eZc~le(0; 0o)[-w + th]l .
0 ~ O
Let H1 be any bounded set in (H, H " Iln). We define the process X* = supu_<s[Xul for
process X. Fr om the stochastic differential equat i on (1.1), we have for h e Ha,
t
X ; ( w + h) - X ~ = i ~ V ( X ~ d s + S { [ Vo(X](w + h), 0o) - Vo(X ~ 0o)]
0 0
+ ~[ v(X~(w + h ) ) - v(x~ ds
t
+ ~ e V ( X ( w + h l ) d w s .
0
288 N. Yoshida
By Le mma 3.2 (1) be l ow we have
(3.2) sup IIX~*(w + h) l l zp< oo
~ ( 0 , 1)
h e l l 1
f or p__> 1.
Next , l et Z~(t, 7, h) = X~ ( w + t~h) - X~( w) for s e [ 0 , T] , e e ( 0, 1), t, t / e( 0, 1)
a nd h e l l . Then, Z~(t, tl, h) satisfies
Z~(t, tl, h) = i e t ~ V ( X ~ ( w + t~h))t~,du
0
+ i [ Vo(X~,( w + ttlh), 0o) - Vo(X~(w), 0o)3 du
0
+ e i [ V(X~. ( w + trlh)) - V ( X ~ ( w) ) ] d w . .
0
Us i ng Le mma 3.2 (1) be l ow we have for p e N , > 2,
H Z ( t , tl, h)[Iv =< C1 l tgt/[
f or s ome C1 > 0 de pe ndi ng on Condi t i ons (2), (3), d, r, p, Xo, T a nd I I h [[. Mo r e -
over l et
0
W: ( t , 1"1, h) = ~ X : ( w + t~h) ,
s e [ 0 , T] , e e ( 0, 1), t, t / e( 0, 1) a nd h e l l . Then, W~(t, tl, h) satisfies t he s t ochas t i c
i nt egr al e q u a t i o n
W~(t, rl, h) = i t e V ( X ~ ( w + trlh))h, du
0
+ i [ ~ ~i Vo(X~u(W + trlh), Oo) [ W~(t, rl, h)-]i
+ terl ~ ai V ( X ~ ( w + trlh)) [ W~(t, rl, h ) ] i h . 1 du
i A
+ e i ~ ai V ( X ~ ( w + tqh)) [ W~(t, 7, h ) f d w , .
0 i
Agai n by Le mma 3.2 (1) a nd (3.2) we have
I1W~*(t, tt, h) llp < C2t e
f or p e N , p > 2, wher e Cz de pe nds on Condi t i ons (2), (3), d, r, p, x0, T a nd I[ h [[n,
Aft er all we o b t a i n
sup E s up ( X ] ( w + tqh) - X] ( w) ) ~ 0 as t ~ 0
t i e(0, 1) O < s < T
Asymptotic expansion for small diffusion 289
for I = 0, 1, h e H, each e and all p > 2. Fr om this we obtain, using the Burkh61der-
Davi s- Gundy inequality and (4),
sup E ( ~ o 6~(6/~(0; 0o)[w + t r l h ] - 61~(0; 0o)[w]) p-~ 0
~o,~) \ a ~j
as t --* 0 for Vo, v ~ . . . . , Vk > O, Vo + V~ + 9 9 9 + Vk < 1, h e H , each e and all p > 1.
Consequently,
E 6~(6/,(0; 0o)[w + t r l h ] - 61~(0; 0o)[W]) dr/d0 --*0
,o, V, /
a s t ~ 0 f o r %, v ~ . . . . , Vk > O, v o 4- v l + " " " + v k < l , h ~ H H , e a c h e a n d a l l p > l .
By Sobolev' s lemma, we see t hat there exists a sequence ( t q } ~ 0 , q = 1, 2 , . . . , such
t hat
sup 1 6 1 ~ ( 0 ; 0o)[W + t q r l h ] - 61~(0; 0o)[W]l ~ 0 , a.s.
tt~ ( 0 , 1 )
0 ~ O
as q ~ ~ for h s H H , each e. Consequently, we see t hat the right hand side of (3.1)
tends to zero a.s. as t ~ 0 and obt ai n
lim O~(w + t h ; 0o) = 0~(w; 0o), a . s .
t ~ 0
if R ~ ( w ) < 1.
If R ~ ( w ) > 1, R ~ C ( w ) > 1 and hence for any h e I - I and some h ( w , h ) > O,
O , ( w -I- t h ) = 0 if [tl < t l ( W , h ) . Therefore, limt-.o0~(w + t h ) O , ( w + t h ; 0 o ) =
O ~ ( w ) O~(w; 0o) for h e l l , we W, ee(0, 1). Funct i onal s O , ( w ) O ~ ( w ; 0o) are bounded
and in (~p> 1Lp( W, P). Next, we calculate H-derivatives of these functionals. For
R ~ ( w ) < 1, h e H and It[ < t o ( W , h ) ,
e261~(O~(w + t h ; Oo ); 0o) [w + t h ] - e261~(O~(w; 0o); 0o) [-w 4- t h ]
= - e z 6 1 ~ ( O ~ ( w ; 0o); 0o)[w + t h ] + ez61~(O~(w; 0o); Oo)[W].
For i = 1 . . . . . k, there exist 0 and ~" such t hat
1 5 - O~(w; 0o)1 < IO~(w + t h ; 0o) - 0~(w; 0o)1,
[~[ < ]t[ and
e 2 6 6 i l ~ ( 0 ; 0 0 ) [ w + t h ] ( O~( w + t h ; 00) - 0,(w; 00))
= - - t ( D h e 2 6 i l ~ ) ( O ~ ( w ; 00); 00) [W + ~h] .
Dividing bot h sides by t and t --* 0 we have
Dh O~ ( w; 0 0 ) = - - Ee2621~(O~(w; 00); 00)Ew]-l-1(Dhe261~)(O~(w; 00); 00)[w]
for h s I - I if ~ < dl and R ~ ( w ) < 1. Then H-derivative of O~(w)0,(w; 00) exists and
Dh E ~ k ~ ( w) O~ ( w; 0o)] = E D h O ~ ( w ) ] O ~ ( w ; 0 o ) + O ~ ( w ) D h O , ( w ; 0 o ) .
290 N. Yoshida
When R~(w) > 1, R~C(w) > 1 and O, (w + th) = 0 if It[ < t l ( w, h ) . Therefore,
D h [ # I e ( w ) O ~ ( w ; 00) ] exists and equals zero. Thus we see t hat ~(w)0~(w; 0o) is
//-differentiable on whole W. The/ / - der i vat i ve is in Lp( W, P), p > 1. In fact, this
follows from uniform non-degeneracy of ~2621~(0; 0o) and the integrability of
sup0~o I De261~(O; Oo)lus, which is obt ai ned by estimating the Sobolev nor m
E[ l[ [DeZ61~('; 0o) [w] 12s I[p ~,~(o)] 9
For a representation of Dale(O; 0o) see Section 4. Similarly we can verify existence
and integrability of higher order H-derivatives, which completes the pr oof of (5).
(6) Fr om the following Lemma 3.2 (3) for any n e N and cz >0 ,
P( R~( w) > Cl) = O(e ") as e $ 0. Fr om this fact we can show (6) in view of the chain
rule for H-derivatives. This completes the proof. []
L e m m a 3 . 2 . (1) L et ~(t), te [ 0 , T ] , e ~ ( O , 1), 0 e O (an i ndex set), be d-di mensi onal
nonant i ci pat i ve processes given by
~o(t) = q~o(t) + ~ :do(S)~(s)ds + ~ fl~o(s)dws ,
0 0
where tfo is an Rd-valued cont i nuous nonant i ci pat i ve process, Cdo is an R d | Rm-valued
nonant i ci pat i ve process, fi~o is an Rd | Rr-val ued nonant i ci pat i ve process, ~ is an
Rm-valued f unct i on sat i sf yi ng
T
j i~(s)12 d s < oo
0
and w is an r-di mensi onal Wi ener process. Suppose t hat there exi st positive const ant s
Ki , i = 1, 2, 3, such t hat
and
Ifi~(s)l ~ K2(l~g~(s)[ + K 3 ) ,
a.s. f o r any ee(0, 1), 0 ~ 0 , 0 <_ s <_ T. As s ume t hat f o r all n e N, e~(0, 1) and O~O,
[ _ O<=t <=T
Then, f or any n >= 2, there exi st s positive const ant Cn = Cn(T, d, r, K1, KE) such t hat
n
s upE sup r~g(t)[" < C~exp Cn tO(s)12ds
0~0 O<- - t <- T 0
f o r ~c(O, 1).
Asymptotic expansion for small diffusion 291
(2) Under Conditions (1), (2) and (3), f or h E N there exist positive constants c l , c2
independent o f e such that
and
s u p E ( s up I X " ~ " - X ~
q ~ ( O , 1) O<-t<-T
/
s u p E s u p X y ~ ~ C2 8n
q e ( O , 1) \ O<=t <- T
f or s e ( 0 , 1).
(3) F o r n e N , c > O a n d a > O ,
Proof (1) L e t
P(R~(w) > a) = O(&) .
aN = i n f { t > 0; [ ~ ( t ) ] > N} A T
f o r N 8 N. F o r s o me Co > 0
I~g(t)l 2 < Co ]r/~(t)] 2 + K 2 I~g(s)] 2 ds + fi~o(s)dws ,
0 0
wh e r e Ko = ( ~ ] 0 ( s ) [ 2 d s ) l / 2 . By t he B u r k h 6 1 d e r - Da v i s - Gu n d y i n e q u a l i t y
E fl~o(s) < C l ( n , T , K 2 , d , r ) K ~ + E l ~ g ( s A a N) l " d s ,
0 aN 0
t ~ [-0, T ] , ~ ~ (0, 1), 0 E O, f o r s o me C1 (n, T, K2 , d, r) > 0. Us i n g a s s u mp t i o n s we
h a v e
g , n z ~ , t l
E(I~oI,A~N) < C2(1 + g ~ ) g(l~gl:*") + E(l~ol . . . . ) d s + g ~ " ,
0 /
r e [ 0 , T ] , e s ( 0 , 1), 0 ~ O , f o r s o me C2 = C2(n, T, K t , K2, d, r) > 0. By Gr o n wa l l ' s
l e mma we h a v e
E( ~ *~ = e c~(I +K'~T
I ~ o ] r ~ , , ) < C2(1 + / g ) E E ( I t / ~ l * " ) + K ~ e " ] ,
e ~ (0, 1), 0 ~ O. L e t t i n g N ~ oo we h a v e t h e r es ul t .
(2) W e k n o w t h a t ( X T ~ - X ~
f d ( X ] ~ - X ~ = [ Vo( X] ~, 00) - Vo(X ~ 00) ] d t + tl~V(X]~)dwt
d OT-X]~=c~Vo(Xnt~,Oo) OT-X't~dt + e ~ t l c ? V ( X ~ ) OT-Xnt~+ V(Xnt~)ldwt
o~l o~l L o~ j
X, o~ _ X O = O
~
X~ z O.
292 N. Yoshida
Since 0Vo(' , 0o) and ~V are bounded and Vo(', 0o) and V are of linear growt h
order, we can apply (1) to this case and obt ai n the result.
(3) It suffices to show t hat for any m ~ N there exists n ~ N such t hat E [IR~(w) l"] =
O(e m) as e ~ 0. We can estimate the norms of the parts of R~(w) corresponding to
f i t and f2 by (2). For the part corresponding to f3, we obt ai n a similar estimate
using (2) and the Burkh61der-Davis-Gundy inequality, which completes the
proof. []
In the context of the higher order statistical asympt ot i c t heory we need to
modi fy the maxi mum likelihood estimators to get efficient estimators. We call an
estimator 0* a bias corrected maxi mum likelihood est i mat or if
where b(O) is a bounded smoot h function. Then, O~(w)@ is well-defined. The
following l emma gives its expansion formula.
Lemma 3.3 O~(w) e- t (@ (w; 0o) - 0o) e D oo (R k) has t he asympt ot i c expansi on
~*( w; 0o) - 0o
O~(w) "~ f o + ef l + " " in D~ ( R k) as e $ O
wi t h f o, f l . . . . ~ D ~~ (Rk). I n particular,
f o = I ( 0o) - ~ B ,
f l = - b ( Oo ) + 89 - l ( Oo ) F + ) I - l ( Oo ) QI - l ( Oo ) B . . . . .
where B = (Bi), F = ( F i) and Q = (Qf, j ) are def i ned as f ol l ows.
B ~=m( ( i ) / O) , i =l . . . . , k ,
r ~=2 m( ( i ) / 1 ) , i =l , . . . , k ,
k
1Bl m N. .
Qi, j = ~ [ I ( 0o) - ~ , , a, ~+2Ai . j , i , j = l , . . . , k ,
m=l
whe r e
N, , j . , , = - [ n( ( j i ) ( m) / O) + n( ( i m) ( j ) / O) + n( ( mj ) ( i ) / O) ] , i , j , m = 1 . . . . . k ,
A ~ . j =m( ( j i ) / O) - n ( ( j ) ( i ) / 1 ) , i,j--= 1, . . . , k .
Pr o o f Suppose e~ < dl and R~o(W) < 1. Let F(O, 8, w) = eZal~(O; 0o) [w]. The map-
ping (e, 0)--+ F(O, e, w) is smoot h and A(O, e, w):= 6F(O, e, w) is non-si ngul ar for
e < eo and 0 satisfying 10 - 0ol < da uniformly. Moreover, by Lemma 3.2 (1), it can
be proved t hat for a.s. w e W and e > 0,
lim sup IF(O, 8 + u, w) - F(O, e, w)l = 0 .
u--+0 0~O
Therefore, the mappi ng e --+ 6~(w; 0o) is continuous. Hence, we have
OF (O~(w; 0o), 8, w)
a ~ (w; 0o) = - A ( J ~ ( w ; 0o), ~, w) - 1 (3.3) t?-~
Asymptotic expansion for small diffusion 293
f o r s < So. I n p a r t i c u l a r ,
f o = l i r a ~ O*(w; 0o) = I ( Oo) - l B
Di f f e r e n t i a t e (3.3) o n c e mo r e we get ~ 0*(w; 0o) a n d f l . Si mi l a r l y, we h a v e
@ (w; 0o), i = 3, 4 , . . . a n d t h e i r l i mi t s i n t e r ms o f A, F, B, t h e i r d e r i v a t i v e s
wi t h r e s p e c t t o 0 a n d s, a n d 0 j w; 0o). T h e r e f o r e , @( w; 0o) i s s mo o t h o n [ 0, so).
T h e n we h a v e t h e e x p a n s i o n
S S 2
6 * ~ ( w ; Oo) = Oo + ~ (ao)o6.*(w; Oo) + ~ (ao)o:&(w; Oo) + . . .
S j - 1
+ - - (ao){ &(w; Oo)
( j - - 1)[
I 1 ( 1 - - S) j - 1 (150)i~ 6 * ( W; 0 0 ) d s
~J
+
~9 ( j - - 1)!
f o r s _<_ So, wh e r e (6o)~ d e n o t e s t h e s - d e r i v a t i v e a t s. He n c e ,
S 2
O~(w)6*~(w; Oo) = ~,~(w)Oo + O~(w) ~ ( a o ) o & ( w ; O o ) + O~(w) ~. ( a o ) o ~ & ( w ; O o )
1 1
+ " " + ~kj w) s j ! ~ (1 - s) j -~ ( 5o) i ~&( w; 0 o ) d s .
Th i s e x p a n s i o n h o l d s o n W. T h e H- d e r i v a t i v e s o f t h e i n t e g r a n d s o f t h e r e s i d u a l
t e r m t a k e t h e f o r m o f (1 - s) i - 1 G( 0~j w; 0o), w), wh e r e G(O, w) is a n H | 9 9 9 |
/ - / - va l ue d r a n d o m f i el d o n O. E s t i ma t i n g t h e n o r m
E [ II I G ( ' w ) l ~ s " ,
, I lw~o,~.o(o)]
p > 1, b e f o r e s u b s t i t u t i n g 0~j w; 0o) f o r 0, we c a n s h o w
Oe(W)S-1(6~(W; 0 0 ) - - 0 0 ) "~ ~r 1 ((~0)0 6.~(W; 0 0 ) @ ~ ( W ) ~ ((~0)g6:l:(W; 0 0 )
1 ! A !
1
(ao)o&(W; Oo) + ~~(ao),~6*(w ; Oo) + . . .
+ . . . . 1-~
i n D ~ ( R k) as e $ 0 s i nce ~kjw) = 1 - O(e') i n D ~~ as e $ 0 f o r n ~ N . Th i s c o mp l e t e s
t h e p r o o f . [ ]
Remarks 3.2. (1) F o r ma l l y , we ma y wr i t e
Bi OF i
= l i m - - ( 0 , ( W ; 0 o ) , e , w ) , i = 1 , . . . k,
~; o ( ~
0 2 F ~
F ~ -= l i m ~T-e2 ( 0 j w; 0o), e, w), i = 1 . . . . . k,
294 N. Yoshida
Qi , j = l i m d [ A( 0~( w; 0o), e, w) ] i ' i
~+ode
0 2 F i
+ l i m Z ~ ( 0 ~ ( W ; 0 o ) , e , w ) , i , j = l , . . . , k ,
e, ko
N~,j,m = l i r a 6~ 61F~(O~(w; 0o), e, w), i, j , m = 1 . . . . , k ,
~,L0
CZ F i
A i , j - - l i m m (O~(w; 0o), ~, w), i , j = 1, k
~ o CeCOJ " " ' "
(2) K u t o y a n t s [ 7 ] p r o v e d a n e x p a n s i o n f o r mu l a f o r t h e m a x i m u m l i k e l i h o o d
e s t i ma t o r o f o n e d i me n s i o n . We wi l l n e e d a s m o o t h t r u n c a t i o n t o a p p l y Ma l l i a v i n ' s
c a l c ul us .
F o r h ~ R e we d e n o t e t h e l o g l i k e l i h o o d r a t i o
T h e n ,
l~,h(w; 0o) = l~(Oo + eh; 0o) - l~(Oo; 0 o ) .
T
l~,h(w; 0o) = ~ e- t [ V~o, t ( Oo + eh) ~ ' ' +~ - Vo , t ( 0 o ) ] ( V V ) t V~d wt
0
1 r
v v )3 [ Vo, , ( 0o + ~h) - v k ~ ( 0 o ) ] d t 2 e - Z[ v ~~1 7 6 + eh) - Vf~, ~(0o)]' ( ' +~ ~
T h e f o l l o wi n g l e m m a gi ve s t h e e x p a n s i o n o f t h e l i k e l i h o o d r a t i o , wh i c h i s r a t h e r
e a s y t o s h o w a n d t h e p r o o f i s o mi t t e d .
L e m m a 3. 4 The l o g l i kel i hood rat i o l ~, h( w; Oo) ~D ~~ and has t he as y mpt ot i c
e x pans i on
l ~ , h ( W ; 00) ~ f ~ + e f L + ' ' " in D ~ as e ,L O,
wi t h f ~, f ~, . . . ~ D ~~ I n part i cul ar,
f ~ = h' B - 8 9
f ~ = ~ h~m((i )/ 1) + ~ h~hJm((iJ)/O)
i =1 i , j = l
k
k 1 ~ hi hJhmn( ( i j ) ( m) / O) . . . . .
1 ~, h i h J n ( ( i ) ( j ) / 1 ) _ ~ i , j , m = l
2 i , j = l
Thi s e x pans i on hol ds uni f or ml y in any c ompac t s et o f R e.
4 As y mp t o t i c e x pa ns i o ns o f probabi l i t y
Fi r s t , we k n o w t h a t f o r h ~ H
i w w - 1 V~l~sds DhX~ "~176 = e ~ t ~
0
Asymptotic expansion for small diffusion 295
For h ~ H, the H-derivative of F( O, e, w ) in the direction of h is
T
Dh F ( O, e, w ) = e ~ [ 6 V ' o ( V V ' ) + V ] ; ( O ) t ~ t d t
0
T
+ ~ I Y. [ ~ { 6 V ' o ( V V ' ) + ~ ~,oo,~
V } ] , ( O ) D h X t d w t
0 l
T
- - ~ ~ , [ ~ { 6 V ' o ( V V ' ) + ( Vo ( 9 , O) - V o ( ' , Oo)) } ] ; ( O ) D h X ; ' ~ 1 7 6 d t
0 1
T T
= e ~ [ 6 V ' o ( V V ' ) + V ] ; ( O ) h t d t + ~2 ~ K ] , ~ ( O ) h s d s
0 0
T
- - e Z (O - - Oo)J ~ K ~, ; , , ( O ) h ~ds ,
j o
where
and
T
K ] , s ( O ) = 2 ~ [ ~ I { ( ~ V ~ o ( V V ' ) + V}]~(0) [ Y; ] lm d wt [ y s e - 1 V~s]m.
m , I S
T 1
/ < ~ , j , ~ ( 0 ) = g ~ [ e ~ { 6 y o ( . , o ) ' ( v v ' ) + S 6jVo
l s 0
( ' , Oo + u( O - 0o))du}]~[Y~]t ' dt y ] - 1 V] .
Let ~( w) = O ( R 3 ~ ( w ) ) for some fixed c > Co. Then, if ~( w) > 0,
Dh ) ~ ( w ) O~ ( w ; 0o) = --A(0~(w; 0o), e, W ) - ' DhF( d~( W; 0 o ) , e, W) .
Therefore, the Malliavin covariance of O~(w) e- 1 ((~(w; 0o) - 0o) is
T
= A ( O ~ ( w ; 0o) , e, w) - ~ ~ {[ a V ' o ( V V ' ) + V ] ~( d~( w; 0o) )
0
+ ~ : ~ , ~ ( d ~ ( w ; O o ) ) - Y [ ~ ( w ; O o ) - O o ] J / ~ , j , s ( 6 ~ ( w ; O o ) ) }
J
{ . . . } ' d s A ( O~ ( w; 0o) , e, w ) - 1 ,
and this can be denot ed by
A (0~(w; 0o), e, w) - i { I~(w) + eR] (w) + R~ (w) (0~(w; 0o) -- 0o) } A (0~(w; 0o),e, w) - i ,
where
T
U ( w ) = ~ [ 6 V ' o ( V V ' ) + 6Vo]](J~(w; 0o) ) ds,
0
R~z(W) is a 3-liner form and R"z(W)(X) = R ~ ( w ) I x , . , . ] .
2 9 6 N. Y o s h i d a
T h e Ma l l i a v i n c o v a r i a n c e o f ~ 5 - l ( 0 * ( w ; 0o) - 0o) is gi ve n b y
~rO,,-q*(w; oo) = ( I k - - 525b(0e(w; 0 o ) ) ) %, 8 ~g,(w; O o ) ( I k - - 5 Z ( ~ b ( O e ( w ; 0 o ) ) ) '
i f Ct(w) > 0.
Le t
f o( w, 5, O) = Qo(w, 5, O) - Qo(O, 0 ) ,
wh e r e
De f i n e
T
Q o ( W, 5, o ) = S [ 6 V ' o ( V V ' ) + ~ V o ] : ( o ) d s .
0
R ~o " t( w) = c ' l l I f o ( w , e, " ) [ 2 2. ~
W, ~ , 2n~(O)
k
f o r c' > 0, m~, n I E N , m I > - -
2n1"
L e t
~ ( w) = ]51/2R] (w)[ 2 + ]57/2R~z(W)]2 + R~o"t(w) + R ~ ( w ) .
Th e n , ~ t ( w ) e D ~, s i nce we c a n r e p l a c e 0t(w; 0o) i n t h e r i g h t h a n d si de o f t h e
r e p r e s e n t a t i o n o f %, ~ %(w; 0o) b y r 00).
L e m m a 4.1 Fo r s ome eo > 0 and c' > O,
s up E[ I {I ~I __<1} ( de t aO, ~ 'g*(w; Oo)) - p ] < oe
t<gO
f o r al l p > 1.
P r o o f I f [~t[ < 1, 10t(w; 0o) - 0o1 < e ' a n d
l i t ( w) - I ( 0o) t < [No(w, ~, 0~(w; 0o))l + ]Qo(0, 0~(w; 0o) ) - I(0o)1
5~ C; r + s up IQ0(0, 0) - Q0(0, 0 0 ) l ,
IO - 0 o l < ~
wh e r e C~ s t e ms f r o m t h e S o b o l e v i n e q u a l i t y f o r O. Mo r e o v e r , i f [~t[ -<_ 1,
leR~ (w)l < ~:/2
a n d
IR~(w)(6~(w; 0o) - 0o)l < eVlR%(w)l < 52 .
F r o m t h e p r o o f o f L e m m a 3.1 (2), we see t h a t i f 5 ~ < dl a n d R~(w) < 1, t h e o p e r a t o r
n o r m o f A(O~(w; 0o), e, w) < 2k + 89 wh e r e Zk is t h e m a x i m u m e i g e n v a l u e o f I ( 0o) .
T h u s t h e r e exi s t s ma l l 5o, l a r g e c' a n d s o me c" > 0 s u c h t h a t
d e t o - ~ - %.(w; 0o) > c "
f o r e < 5o i f ~t(w) < 1. Th i s c o mp l e t e s t he p r o o f . [ ]
Asymptotic expansion for small diffusion 297
Lemma 4.2 ~(w) has an asympt ot i c expansi on in D + and f o r h e N and c 1 > 0,
P [ I ~ I > c l ] = 0 ( 5 9
a s ~ J , O.
Proof. Fr om Lemma 3.2 (2) we see t hat for n e N and ca > 0
P[R~o"~(w) > c l ] = 0 ( 5 " )
as e J, 0. Applying Lemma 3.2 (1) to Y~ - Yt and Yf-1 _ y;-1 we have
L O<t <T I L O<t<=T
for all n e N. For n e N, ca > 0 and c~ > 0
P i e ~ sup ,K~z,j,s(O), > c a ] = O( ~ " )
i st [ 0, T]
0cO
as e ~ 0 for j = 1, . . , k. Applying the Bur kh61der - Davi s- Gundy i nequal i t y to
T T
S [ a , { 6 V ' o ( V V ' ) + v } ] ; ( 0 ) [ ~ ' ~ Y, ] dw, = ~ [~3, {6V' o(VV' ) + ~ ~ ' ' V } ] t ( O) [ Y , ] dw,
s 0
- i [ O, { 6V' o( VV' ) + V}]~(0)[ Y ~ ] ' ' d w,
0
and its derivatives in 0, we have
I , ]
sup g s up l i Ke, s ( ) l l w' , =' - <o) <
ee(O, 1) L se[O, T]
for l, m, n e N. By Sobolev' s l emma we see t hat for any ~ > O, h e N and cl > 0
P [ ~ sup s u p , K ] , s ( 0 ) , > C l ] = O ( e ' )
se[O, T] OeO
as ~ J, O. Since
P[ I ~ I > ca] < P[ I ~I > c~, R~(w) < 1] + P[R~( w) > 1]
and I O~(w; 0o) - 0ol < e ~ if R~(w) < 1, it is not difficult t o show this l emma by
Lemma 3.2 (3). []
Let qS(x;/~, S) be t he pr obabi l i t y density funct i on of the k-di mensi onal nor mal
di st ri but i on wi t h mean /~ and covari ance mat ri x S. Let I - a ( O o ) = I - l = (IU).
Then, we have the following t heorem.
Theor em 4.1 The probabi l i t y di st ri but i on o f t he bias correct ed max i mum likelihood
est i mat or @(w; 0o) has t he asympt ot i c expansi on
0 o ]
e A ~ ~ p o ( x ) d x + e ~ p a ( x ) d x + " " as s ~ O, Ae B k
A A
298 N. Yoshida
The e x p a n s i o n is uni f or m in Bo r e l s e t s A E B k. I n par t i c ul ar ,
po( x ) = O(x; O, I - i ) ,
Pl(X) = I 2 I i J A i , j , l x l - { - S I i J B i , l,J x l - S b J ( O o ) l j tXl
, , j , l i , j , l j , I
- ~, Ai , j , t x i x J x l - ~ ~ B i , j , t x ~ x J x ~ ~b(x;0, I 1 ) , . . . ,
i , j , l i , j , l
wh e r e
and
Ai, j , , 2 o o l,~=1
P r o o f Le t
T t d
= Z
0 0 1, m = $
a n d
T t d
B , , j , ~ = n ( ( i j ) ( l ) / O ) .
[ a , { 6 , V , o ( V V , ) + v } ] O ( O o ) [ , , + - o , ~ ~ ~ V ( V V ) Os Vo ] , ( Oo ) Y , g, d s d t
A * * n = ~ ~ 2 [ ( ~ i V ~ o ( V V ' ) + Vl~ [ O , { V ' ( V V ' ) + c~j V o } ] ~ y~mg~nn d s d t .
0 0 l , m= l
T h e n , A-*- = A-*.* a n d
, , = ~ ( A i , j , n + , , j ,
Ai j , 1 . A.**. , ) .
I t suf f i ces t o ve r i f y t h e c o n d i t i o n s o f T h e o r e m 2.2 f o r
= ~ $ 1 ( 6 " ( w ; 0 o ) - 0 o )
a n d ~ gi ve n a b o v e . T h e c o n d i t i o n 2) is p r e s e n t e d b y L e m m a 3.3. T h e c o n d i t i o n 1)
f ol l ows f r o m L e m m a 4.1. Co n d i t i o n s 3) a n d 4) a r e b y L e m m a 4.2. L e m m a 2.1 g a v e
5). Th e r e f o r e , f o r A e B k, we h a v e t h e a s y mp t o t i c e x p a n s i o n
O( ~ ) I A( O~ e - l ( 0 * ( w; 0o) - 0o) ) ~ (iDa, o -'~ ~( ~a , 1 ~ - ' ' ' i n / 5 -~~ as e ~ 0
u n i f o r ml y i n A ~ B k, wh e r e
( ]) A, O = I A(f O),
k
~ a , ~ = ~ f l a , I A ( f o ) , . . . .
, = 1
He n c e , b y L e mma s 4.2, a n d 2.2,
E[-qSA, 0] + eE[~/'A, 1] + 9 " "
Asymptotic expansion for small diffusion 299
as e $ 0, u n i f o r ml y i n A e B k. The r e f or e , t he r es t is t o c a l c u l a t e E [ ~ A , i ] ,
i = 0, 1 , . . . . F r o m t he r e g u l a r i t y o f f o a n d i n t e g r a t i o n b y p a r t s f o r mu l a ,
E[Cl)A,~] = E [ G i ( W ) I A ~ = E [ G i ( W ) I A ( f o ) - I
= J E [ G i ( w ) l f o = x ] P f o ( x ) d x
A
f or s o me s mo o t h f u n c t i o n a l G~( w) . The r e f or e , e a c h t e r m is r e p r e s e n t e d b y a n
i n t e g r a t i o n o f a s mo o t h f u n c t i o n . We wi l l o n l y d e t e r mi n e Po a n d p l . Po is t r i vi al .
Co n s i d e r i n g A = {y = (y~); y~ > x i, i = 1 , . . . , k}, x = ( x ~) ~R k, we see
(4.1) = ( - 1 ) k ~ 0~'0~ " " ~i _1~i +1 ' ' " O k E [ f i l H ~ l ( f ~ ) " ' " 6 ~ ( f ~ ) . . . H ~ ( f ~ ) ]
i
= - ~_, a , E [ f ] , S x ( f o ) ] = - ~ ar { E [ f ] I f o = x ] p f o ( X ) } ,
i i
wh e r e H x ( y ) = lrx, ~o)(y) f or x, y e R . Le t
a, = I ' , ~ v ~ ~ v ~
a n d
~ i , , = [ ~ { 6 , V ' o ( V V ' ) + v } ] ~
Th e n , b y L e m m a 4.3 b e l o w we h a v e
i l
g [ m ( ( i ) / 1 ) l f o = x ] = g ~ 7~, t D~dwt l fo = x
l
= ~ E [ i Y i ' t i [ Y ~ l ~ ' Y s l V ~ o o
= ~ ' Tr S S [ 6 V ' o ( V V ' ) + V] ~
l o o
[c3~ { V ' ( V V ' ) + c5~ Vo } ] o ( 0 o ) [ Y t ] 1 g ~ d s d t ( x x ' - I - 1)~
/
= ~ A * j , I ( x J x l - I ~ i l ) ,
j , l
E [ m ( ( j i ) / O ) l f o = x ] = E [ 6 j 6 i a t d w~ = x
0
T
= ~ [. [ 6 j 6 1 V ' o ( V V ' ) + , ~ V o 3 ~ ( O o ) d t x ~
l 0
= 2 B i , j , I X l
1
300 N. Yoshida
a nd
1
E [ n ( ( i ) ( j ) / 1 ) l f o = x ] = E [c~{6~ g ' o ( g g ' ) + 6 j V o } ] ~ = x
= E [a,{6~ v ; ( v w ) + 6 j Vo } ] ~ Y, y 9
( i Y [ 1 V ~ d t i a t d w ~ = x ]
T t
= ~ ~ , [~?~{c~z V' o(VV' )+F)j Vo } ] ~
o o ~
= ~ 2Az, j , f .
l
Hence, t he c ondi t i ona l expect at i ons a ppe a r i ng in t he r i ght h a n d si de of (4.1) a r e
E r f i l f o = x ] = - b~( Oo) + 8 9 o = x ] + 1 E [ U I - I Q I - ~ B ] ~ I I o = x ]
= - b i ( Oo ) + ~ I i J E [ m ( ( j ) / 1 ) l f o = x ]
1
(4.2) - 89 ~, I i J x mx l [ n ( ( j l ) ( m) / O) + n( ( l m) ( j ) / O) + n ( ( j m) ( I ) / O) ]
j , l , m
+ ~', I i 2 x l E [ m( ( j l ) / O) - n ( ( j ) ( 1 ) / 1 ) l f o = x ]
j , l
= - E r~A*~3 S x ' - E 1_
j , l , m ' ' j , l , m 2 I i m B l ' j ' r n x j X l
- b ' ( O o ) - ~ I i " I i ' A * * , .
j , l , m
Then, t he f unct i on p l ( x ) c a n be der i ved f r om (4.1) a nd (4.2). Thi s compl et es t he
pr oof . [ ]
Le mma 4.3 L e t w be an r - di mens i onal Wi e ne r process and let f u n c t i o n s at, bt, ct on
[0, T] be det er mi ni s t i c. L e t Z = ~T ata; dt .
(1) L e t at be R k | and let b, be Rr - val ued. Then,
E b' t dwt a t d wt = x = x'12 - 1 ~ a t b t d t .
0
(2) L e t a, , b, and c, be R k | R' , R ~ | R ~ and R m | R~-val ued, respect i vel y. Then,
[ i ( i ) i I
E bs dws ctdwt at dwt x Tr I2-1 , , - 1 = = at c t bs as Z ( x x ' - Z ) d s d t
o o
He r e Tr s t ands f o r t he trace.
Asymptotic expansion for small diffusion 301
To p r o v e t he opt i ma l i t y of t he ma x i mu m l i kel i hood e s t i ma t or we need t he a s ymp-
t ot i c e xpa ns i on of t he l i ke l i hood r at i o pr ocess. As t he ma x i mu m l i kel i hood es-
t i ma t o r we c a n s how t he n o n - d e g e n e r a c y of t he Mal l i avi n c ova r i a nc e of t he l og
l i kel i hood r at i o a nd o b t a i n t he f ol l owi ng t heor em.
The or e m 4.2 L e t h ~ R k a n d h r O. T h e p r o b a b i l i t y d i s t r i b u t i o n o f t h e l og l i k e l i h o o d
r a t i o l~,h(W; 0o) h a s t h e a s y mp t o t i c e x p a n s i o n
P[ I , , h ( w; Oo ) 6 A ] " ~ p L ( x ) d x + e ~ p L ( x ) d x + ' ' ' , a s e ~, O, A 6 B 1 .
A A
T h e e x p a n s i o n is u n i f o r m i n A ~ B ~. I n p a r t i c u l a r ,
p L( x ) = 0(.~; O, J ) ,
p/ ( x) = [ ~, Ai , j, t h i h J h ; ] J - 3 [ x 3 - Jfc 2 - 3 d x + JZ]q~(x; 0, J )
L . i , j , l ..J
+ 2 . i , j , ~ Bi, j,lhihJhllj_2[fc2_3 J ~ - J ] qS(~; 0, J ) ,
wh e r e J = h ' I ( Oo ) h a n d 2 = x + 89 T h e p r o b a b i l i t y d i s t r i b u t i o n f u n c t i o n o f
l~,h(W; 0o) h a s t h e a s y mp t o t i c e x p a n s i o n
i , j , l
+ ~ I ~ , Bi, j, t h i h J h l ] J - l [ x - J ] } ~ ( x ; O, J ) -t- " " " ,
i , j , l
wh e r e 4~(x; #, a 2) is t h e p r o b a b i l i t y d i s t r i b u t i o n f u n c t i o n o f t h e o n e - d i me n s i o n a l
n o r ma l d i s t r i b u t i o n wi t h me a n # a n d v ar i anc e a 2.
Pr oof . Cons t r uc t i ng an a p p r o p r i a t e s mo o t h t r unc a t i on f unct i onal t he non-
de ge ne r a c y of t he Mal l i avi n c ova r i a nc e is p r o v e d in t he sense of Th e o r e m 2.2 a nd
we ha ve t he c ompos i t e f unct i onal wi t h a t e mpe r e d di s t r i but i on I a , t he a s y mp t o t i c
e xpa ns i on of t he gener al i zed f unct i onal a nd t he e xpa ns i on f or di s t r i but i on
P[l~,h(W; 0o) ~ A] ~ E[~bLA, 0] + eE[~bL, 1] + ' " 9
a s e $ 0 , A ~ B l , wher e L = I a ( L _ c L o [ t c L ~
f 0 ) , ~ ] , 1 The e xpe c t a t i on
( ~ A , O - - J 1 A~, J OJ ~ . . . .
E[~bL, 1] = ~ p L ( x ) d x ,
A
wher e
Le t
p L ( x ) = - - ~ { E [ f L I f L = X] pf ~( X) } .
4, = h' 6 V ~ V V ' ) , +~ V ~ .
302 N. Yoshida
Then, by Le mma 4. 3 we have
[ i d w t i 1
E [ m ( ( i ) / 1 ) l f L = x ] = E ~ [ O { g h V ' o ( V V ' ) + V } ] ~ a t d w t = 2
l 0
T t
= ~ ~ h ' [ S V ' o ( V V ' ) + V]~ [~?{ V' (VV' )+g)~ Vo}]~
O 0
. j - 2 [ - ) ~2 _ _ j ]
~ ' A * hJh 1 1 - 2 l c ' 2 J )
= / ' , i , j , l a ~.~ - - ,
j , l
g [ m ( ( i j ) / O ) l f L = X] = E [ c ~ S j V ' o ( V V ' ) + V ] ~ 8,dw~ = 2
0
T
= ~ h ' [ S V ' o ( V V ' ) + V]t~ [5~5j V ' o ( V V ' ) + V]t~ ' d t J - t 2
0
T
= ~ h ' [ c ~ V ' o ( V V ' ) + 6 ~ 6 j V o ] ~
0
= ~ B i , j , l h l J t X
I
and
E [ n ( ( i ) ( j ) / 1 ) l f L = X] = E [~,{5~V; ( V V ' ) +6 j V o } ] ~ a, dw, =
k_O
T t
= ~ ~ [ e { a i v ; ( v v ' ) + a j Vo } ] O ( 0 o ) Y, Os d s d t J - ~ h 2
O 0
Then, f r om Le mma
E [ f L ] f L = x ] =
i
= ~ 2Ai , j , z h t J - 1 2 .
1
3. 4 we have the condi t i onal expect at i on
1_~ h , h J E [ m ( ( i j ) / O ) l f ~ = x ]
h i E [ m ( ( i ) / 1 ) l f ~ = x ] + 2 i j
21, j h l h J E [ n ( ( i ) ( j ) / 1 ) ] f ~ = x ]
1
2 . ~. hi hJhmn((i j )(m)/ O)
t , j , m
= I ~ Ai , j, t h l h Y h l l j - 2 [ f c 2 - J 2 - J - [
i , j , l
+ 1 F ~ B i j z h i h J h Z l J - l [ 2 - J ] .
2 L , , j , ~ ' ,
Fr o m this equat i on we obt ai n p~( x) , whi ch compl et es the proof.
[]
Asymptotic expansion for small diffusion 303
The st at i st i cal t he or y of t he a s y mp t o t i c efficiency is deepl y r el at ed t o t he
p r o b l e m of t est i ng st at i st i cal hypot hes es . To cal cul at e t he p o we r of t est st at i st i cs
r equi r es t he a s y mp t o t i c e xpa ns i ons of di s t r i but i on f unct i ons unde r c ont i guous
al t er nat i ves. The f ol l owi ng l e mma enabl es us t o t r a ns f or m t he e xpa ns i on of t he
c o mp o s i t e f unct i onal i nt o t hos e unde r c ont i guous al t er nat i ves in t he s pace of t he
gener al i zed Wi e ne r f unct i onal s.
L e mma 4.4 Le t r be t he general i zed Wi ener f unct i onal given in Theor em
2.2. Suppose t hat A~(w) ~ ~ o has t he asympt ot i c expansi on
A ~ ~ A~ + s ~ l ( w) + 8 2 t / ] 2 ( w ) - J r - ' ' " ) i n D ~ as s + O,
wi t h A ~ 1 7 6 e / 5 ~ Then, A~( w) r has t he asympt ot i c
expansi on
A~(w)r ~ A ~ + e~;.,1 + "" 9 ], in D -~176 as e $ O,
uni f orml y in 2 ~ A wi t h A ~ 1 7 6 ~ D -~ det ermi ned by t he f or mal
Tayl or expansi on
(1 + 8 ( / ' / 1 - [ - 8 2 ~ r J 2 J I - " ' ' ) ( ( J T ) ~ , o - F - ~ ] ~ ) . , 1 + " " " ) = ~ 2 , 0 " 4 - 8 ( ] ~ . ~ , 1 J I - ' ' " ,
I n particular,
~
, 0 = ( ~ 2 , 0 ,
~ 1 = ~ , 1 + ~ , o ~ 1 , . . . .
Pr o o f We c a n s how t hi s t e mma by Th e o r e m 2.2 (ii) of Wa t a n a b e [ 17] . [ ]
Le mma 4.5. (1) For Oo ~ 0 and h E R k, t here exi st f unct i onal s (o), s s (0, 1), on
C( [ 0, T] , R d) sat i sf yi ng t he f ol l owi ng condi t i ons f or any compact set K c R k.
( i ) o __< ( o ~ ( x ) < 1 , x ~ C ( E O , T ] , R ~ ) .
(ii) (o~(X~'~176 = 1 - O(s") in D ~176 as s $ 0 uni f orml y in h ~ K f or n = 1, 2 . . . . .
( i i i ) ( o~( X ~' ~176 ) = 1 - O(s") in D ~ as s ~ 0 uni f orml y in h ~ K f or n = 1, 2, . . . .
(iv) For all p > 1,
sup E [ 1 {~2(x,,,o(w}) > 0} exp { ply, h(w; 0o) } ] < oo.
e~(O, 1)
heK
(2) Le t Oo ~ 6) and let K be any compact set o f R k. Then, (o~(X ~'~176 exp{l~,h(W; 0o)}
has t he asympt ot i c expansi on
(o)(X~'~176 0o)} ~ ef~o(1 + s7'1 + Sz~P2 + " ' " )
in D ~176 ar e $ 0 wi t h 7Jl, 7*2 . . . . E D ~176 det ermi ned by t he f or mal Tayl or expansi on
e x p { e f } + e 2f L d- " ' ' } = 1 + sTJl + g2 ~tr-/2 -4- ' ' ' .
Thi s expansi on is uni f orm in h ~ K.
Pr o o f (1) F o r X ~ C( [ 0, T-l, Ra), l et
(oh(X) = O( Q) ( X) - h' I(Oo)h) ,
304 N. Yoshida
wh e r e
T
Q f ( x ) = a - 2 1 [ Vo(X,, Oo + ah) - Vo( X, , 0o) ] ' ( V V ' ) + ( X, )
0
9 [ V o ( X , , Oo + ~h) - Vo(Xt, 0o)] dt.
Us i n g L e m m a 3.2 (1) f o r ~.Y"'0~ - - xO' 0~ we h a v e
s u p P ( s u p , X ~ ' ~ 1 7 6 1 7 6 1 7 6 1 7 6 n)
h EK \ O < t < T
as e $ 0 f o r a n y n ~ N a n d a l > O. Th e n , i t is e a s y t o s h o w (i), (ii) a n d (iii). L e t p > 1.
Th e n ,
E [ l { 4 , ~ ( x ~ , o o ) > o } e x p { p l ~ , h ( w ; O o ) } ] < E e x p p q e - l S ~,Oo eh)
0
_ _ T Z { ",,Z e, Oo
v O ~ X t , 0 0 ) ] '
) -15:/"
9 ( E [ 1 {4):(x"' % > o} e x p { 89 r q ' Q h ( x " ' Oo) } ] )1/q"
< ( e x p { 8 9 + 1] } ) 1/ r
wh e r e q > 1, q' > 1, ~ + ~ = 1 a n d r = p 2 q . Th i s p r o v e s (iv).
(2) L e t G~ = / ~ , h ( w ; 0o). Th e n , f r o m L e m m a 3.4
G~ ~ f ~ d + e f ~ + " "
i n D ~ u n i f o r ml y i n h e K. We h a v e t h e e x p a n s i o n
(G~ - f ~ ) " = e g , , 1 + " ' " + e Z - l g , , , t - 1 + e t r , , ( w , e )
wi t h 9,,,1 . . . . . g , , , ~ - i e D ~ , r , , ( w , e) = O(1) i n D ~ as e $ 0. Us i n g t h e f o r mu l a
1 1
e ~ = 1 + ~ x . + - - " + - - x t - l ( l _ 1)! + R l ( x ) '
wh e r e ] Rl ( x) [ < = = z ! t + l ) l x l l , f o r x G ~ - f o L w e o b t a i n
e Q- f ~ : 1 + e 7 ~ 1 + 9 9 9 + e l - l ~ l - 1 + e r r ( w , e) + R z ( G~ - f x d ) ,
wh e r e r ( w , e) = r , ( w , e). Th e n ,
n = l
II d ? h ( X ~ ' ~ 1 7 6 e G - - ef~o( 1 + ~7~1 + " ' " + d - ~7 ' , - ~) l l ~, o
<11 4 ~ ( x h .,Oo)efo[e~ G~-fo1" _ (1 + eTt l + " ' ' + e l - 1 7 t l - 1 ) ] I[p,O
+ 1 1 ( 4 ) h ( x *, ~176 - - 1)ef~o(1 + e Tq + . . . + , ~ ' - 1 ' . / ' , _ 1 ) I t , , , o
c ~ . ( X ) e f o R , ( G ~ - - f o L ) l l p , o + e'l[ q'~t j ~o- t ~, e) l l p. o
+ I I ( ~ b ) ( X ~ ' ~ 1 7 6 - 1)e/o~(1 + e7~1 + " ' " + e ~ - ~ % - ~ ) l l , , o
<= I] O h ( x " ' ~ 1 7 6 L
- - f o ) I1~,o + o C )
Asymptotic expansion for small diffusion 305
u n i f o r ml y i n h e K. Th e f i r st t e r m i n t he r i g h t h a n d si de is
N d P ~ ( X * ' ~ 1 7 6 - f ~ d ) I l p , o
< = ~ i . I I d k h ( x ~ > ~ 1 7 6 f ~ ) ' I I v , o + ~ l I ! + h ( x ~ ' ~ 1 7 6 - - f o ) L ' I 1 , , 0
1
_-< ~ II l { , , o ( x < , o o ) > o ) e ~ : l l 2 . , o I I I G . --fLo I' 112.,o + o @ ~)
= O ( J )
as e ], 0 b y (1). We c a n s h o w t h a t a s i mi l a r e s t i ma t i o n h o l d s f or H" Np,s-norms,
p > 1, s > 0. Thi s p r o v e s (2). [ ]
Us i n g L e m m a 4. 4 f or
A~: = ( a h ( x * ' ~ 1 7 6 0o)}
(4.3)
--~ A~ + eTJa + e 2 ~ T / 2 - [ - 9 9 9 ) i n D ~176 a s e ,~ O,
wh e r e Ah ~ = exp{ f ~ } , we see
A ~ O ( ~ ) T x ( F ~ ) , . ~ AO[t~h,,~,O - } - ~h, ; ~, t + " " "] i n D - ~ ~ a s e ,~ 0
u n i f o r ml y i n 2 ~ A wi t h qSh, ~, o, 4Sh, z, 1, 9 9 9 ~ D - c ~ d e t e r mi n e d b y t h e f o r ma l T a y l o r
e x p a n s i o n
e x p { e f } + e 2f ~ + ' . "}(45a, o + ecb~,~ + , . . ) = ~h,*,O + e 6 h , a , , + " " .
I n p a r t i c u l a r ,
~ h , ~ , , 0 m _ ~ 2 , 0 ~
~ [ ~ h , 2 , 1 = ( / ) 2 , 1 +flL#~. , O 9
F o r s t a t i s t i c S ( X ) a n d a me a s u r a b l e f u n c t i o n f
E o o + ~ h [ f ( S ( X ) ) ] = ~ f(S(x))P~,oo+~h(dX)
c([o, T],R d )
= C([0flr],a,)f ( S ( x ) ) d ~ h ( x ) P ~ , o o ( d x )
= E~176176176 ( X ) f ( S ( X ) ) 1
Le t K be a n y c o mp a c t set o f R k. Si nce (4.3) is u n i f o r m i n h e K,
P~,oo+~h[F~[X ] ~ A ] = E [ I A ( F ~ [ X ~ ' ~ 1 7 6
h e O0+eh ~ e O0+eh
E[qS~(X ' ) I A ( F I X ' ] ) ]
= E [ O h ( x ~'~176 exp{l~,h(W; 0o) }" I A ( F ~ [ X " ' ~ 1 7 6
E [ 4 ) h ( x ~ ' O ~ O o ) } t p ( ~ ) I A ( U [ X ~ ' ~ 1 7 6
= E [ O h ( x ~ ' ~ 1 7 6 O o ) } O ( ~ ) I A o ( F ~ [ X ~ ' ~ 1 7 6
E [ A ~ + ~ E [ A ~ 1] + . . .
306 N. Yoshida
uni forml y in K for k-di mensi onal regul ar statistic F ~ [ X ] with some t r uncat i on
~p({~) and A e B k, where ~h, a, i are cor r espondi ng t o I A o F ~ for T z o U , i.e.,
~ ) h , A , O ~ ( I ) A , O
~ h , A , 1 = ~ A , I + f L q ~ A , O , " " " ,
where ~bA, 0, ~bA, 1 . . . . e / 5- ~ are coefficients of the expansi on
t P( ~) I A~176176 ~ ~A, O + e~A,~ + " ' "
i n/ 5- ~176 as ~ { O. Fr om this ar gument we can obt ai n the asympt ot i c expansi ons of
di st ri but i ons of the bias correct ed maxi mum l i kel i hood est i mat or and the log
l i kel i hood rat i o under the cont i guous al t ernat i ve wi t hout expandi ng t hem again.
Theor em 4.3 The pr obabi l i t y di st ri but i on o f t he bias cor r ect ed ma x i mu m l i kel i hood
e s t i mat or O* (w; Oo + ~h) under t he cont i guous al t ernat i ve P~, Oo +~h has t he a s y mp t o t i c
e x pans i on
P I ~ * ( w ; O ~ 1 7 6 ]
~ A ~ p~o(y)dy + e ~ p ] ( y ) d y + ' ' ",
A A
a s e $ O , A ~ B k , h ~ R k,
wher e pCo, p~ . . . . are s mo o t h f u n c t i o n s . The e x pans i on is uni f orm in A ~ B k and
h ~ K. I n part i cul ar,
P~o(Y) = qb(y; O, 1-1),
p ~ ( y ) = ~ , i j z " "
Ai, j , ~[ - - y y y - h~yi S + I ' J f + liJh~]O(y; O, 1 - 1 )
i , j , l
+ ~ Bi,j,~ _ ~ y , y J y _ h i y @ ~ + p l y + i U h j O(y; 0,
i , j , l
I - 1 )
- ~ b~( Oo) I j z f 4) ( y ; O, I - 1 ) .
j , l
P r o o f First, we not e t hat
P I O * ( w ; O ~ 1 7 6 ] f o * ( w ; O ~ 1 7 6 ]
~ A = P ~ A + h ,
wh e r e A+ h = { x E Rk ; x - h ~ A} . Let y = x - h . All what to do is to f i ndp~
and p] .
E [ A ~ = ~ A~ d x = ~ O ( Y ; O , I - ~ ) d x ,
A + h A + h
where A ~ = e x p { h ' I x - 89 and 45h,A, o = I A. We see t hat
E [ A ~ L IA+h(f0)] = E [ A ~ L [ f o ] ]
= ~ E [ f L l [ f o = x ] O ( Y ; O , I - i ) d x
A + h
Asymptotic expansion for small diffusion 307
a n d f o r a s mo o t h f u n c t i o n / ~ (x),
k
3" e [ A ~ =
i = 1
By L e m m a 4.3, we h a v e
(4.4)
j /5~ ( x ) d x .
A + h
~ 1 i j l
E [ f ~ l f o x ] y . * ' j ' L - B , ~ z h h x = = Ai , j , l h ( x x - I j l ) +
i , j , 1 i , j , l 2 ' '
1~ i j z.
- ~ Ai, j, l hi hJx I - - )_[. Bi j zh h h
i , j , l 2 i , j , l ' "
O n t h e o t h e r h a n d , f o r B = {z; z ~ > x i, i = 1, . . . , k},
(4.5)
/54 (x) = ( - 1)kOl . . . ~ k ~ E [ A ~ f ~ Oi l B( f o) ]
i
= - ~ O , E [ A ~
i
= - v J [ ~ i { A ~ [f o = x ] P f o ( x ) }
i
Si nc e
= A ~ - ~ [ h ' I ] * A ~ I f o x ] p z o ( X ) .
i
p ~ ( y ) = ~)](y + h) + E [ f L l f o = y + h] q~( y; 0, I - 1 ) ,
(4.4), (4.5), T h e o r e m 4.1 a n d (4.2) l e a d t o
p~l(y) = I ~ Ai , j , l [ - x i x J x I + 2 h l x J x z - hi hJx I +
I i ~x 1]
l i , j , l
- [ E B i ' j ' l [ l l l l i ~ h i h J x l I i l x J - - 1
i,j,1 k - ~ x ' x j x + 2 h x x ~ + + ~hi hJ hl
- ~ b i I ~ j ( x ~ - M) 1 4 ( y ; 0,
I
i , j
)
S u b s t i t u t i n g y + h f o r x we o b t a i n t h e r es ul t . [ ]
F o r t h e l i k e l i h o o d r a t i o s t a t i s t i c we o b t a i n t h e f o l l o wi n g t h e o r e m.
T h e o r e m 4. 4 L e t h e R k and h r O. Th e p r o b a b i l i t y di s t r i but i on o f t he l og l i kel i hood
r at i o le, h(W; 0 0 + eh) has t he a s y mp t o t i c e x p a n s i o n
P[l~,h(W; Oo + eh) ~ A ] ~ S p L ~ ( x ) d x + e S p L C( x ) d x + "" ", a s e .~ O, A ~ B 1 .
A A
308 N . Y o s h i d a
The expansion is uniform in A ~ B 1. I n particular,
p o ~ ( x ) = r o , s ) ,
PLY(x) = I 2 Ai j , ' h i h J h t l J - 3 [ x - 3 + 2 J _ x ; - ( 3 J - J Z ) _ x - 2 J Z ] q ~ ( _ x ; O, J )
Li , j , l _]
+ ~ _ ~ B i j l h i M h J - 2 [ _ x 2 + j x - J ] q ~ ( x ; 0 , J ) ,
Z L ~ , j , ~ . . . .
where x_ = x - 8 9 The probability distribution f unct i on o f I~,h(W; Oo + eh) has the
asymptotic expansion
P [ / , , h ( w ; 0 o + e h ) < x ] - ~b(_x; 0, J ) + e f I ~ A i j , , h i h J h t l J - 2 [ - x_ E - 2Jx_+J _ j 2 ]
LLi , j , l l
+ 1F E B, j , h i h J h l - ] J - l [ - x - J ] } 0 ( x _ ; 0 , J )
2 L ~ , j , z ' ' I -
Proof. W i t h s o m e t r u n c a t i o n f u n c t i o n a l q / ~ ( w ) ~ D | w e h a v e
O#(W)Ia(l~,h(W; 0o)) ~ ~ ) h , A, _~_ ~ L , A , 1 _~_ . . .
i n / ~ - a s e $ O w i t h L L 5- o0
oo ~ h , A , O , C h , A , 1 . . . . ~ I n p a r t i c u l a r ,
L
# h , A , O = l A ( f ~ )
L L
~L,A, 1 = f l al A ( f O) .
B y t h e a b o v e a r g u m e n t , w e o b t a i n t h e e x p a n s i o n
w h e r e
T h e n ,
P(l~,h(w; Oo + eh) ~ A) ~ E[ A~162 ] + ~E[AO~L, A, 1 ] + " ' ' ,
- L
qbh,A, 0 = I A ( f L)
- - L L L
Cl)h,a, 1 = f i OI A ( f o ) + f ~ I a ( f ~ ) .
L L
E [ A h 0 L,A ' O] = E [ mh 0 I A ( f 0 L ) ] = E [ e f o l A ( f o ) l
= SeXr J ) d x = S r J ) d x .
A A
+ 9 . . .
w h e r e
= [Ah f l 3x(fo)]
q l ( x ) - - ~ E o L L
= - O{eXE[f~;lf~d = x ] r 0 , J ) }
= - , ~ { E [ f ~ l f ' d = x ] 4 , ( _ x ; 0 , J ) }
H e n c e , p~dC(x) = qS(x_; 0 , J ) . O n t h e o t h e r h a n d ,
E [ A ~ = ~ q l ( x ) d x ,
A
A s y mp t o t i c e x p a n s i o n f o r s ma l l d i f f u s i o n
= { I 2 A i , j , l h i h J h l l j - 3 ( x 3 + j x _ 2 - 3 J x _ - j 2 )
l _ i , j , l A
+ ~ ~ B~,j,~h~hJh~]J _ - J ) qS(x_;0, J)
t _ i , j , l A
Moreover,
where
Since
E o L L
[ A h f l l a ( f o ) ] = ~ q 2 ( x ) d x ,
A
q2(x) = e ~ E [ Q I f t d = x] ~b(2; 0, J)
= E [ f ~ l f * 6 = x]qS(_x; 0, J)
= { [ i , j ~ , z A i , j , l h i h J h t l j - 3 ( j x _ 2 + j 2 x _ - j 2 )
+ 2 L_i,j,~
we obt ai n p}~(x). []
p}C(X) = q l ( X ) + q 2 ( x ) ,
309
5 Second-order asymptotic efficiency of the maxi mum likelihood estimator
The not i on of second order efficiency of statistical estimators has been i nt roduced
by Fisher, Rao [11], [12], Takeuchi -Akahi ra, Ghos h- Subr amanyam [2] and
other authors. Here, as an example of statistical applications of our expansion
formulas, we adopt Takeuchi -Akahi ra' s criterion by probability of concent rat i on
and show t hat a bias corrected maxi mum likelihood est i mat or is second order
asympt ot i cal l y efficient. For simplicity, we only treat a one dimensional paramet er
0, t hough all is ready for the mul t i paramet er case.
Definition 5.1. An est i mat or T~ is second order asympt ot i cal l y median unbiased
(second order AMU) if for any 0o ~ O and any c > 0
and
lim sup e - l l P ~ , o [ T ~ - O< ~ O] - 89 = 0
~ $ o 0 ~ O: k O - 0ol < ec
lim sup e-a[p~,o[T~ - 0 > 0 ] - 89 = 0 .
~$o 0~O: 10 = 0ol < ~c
Given a second order AMU estimator T~, if
i p __ __ __
l i me - j ~,oo[~-i(T~ 0o)_-< hi Go(h, 0o) e G~( h, Oo)l = 0 ,
~.L0
310 N. Yoshida
t hen Go(h, 0o) + eG1 (h, 0o) is called a second or der asympt ot i c di st ri but i on of T~.
Consi der testing hypot hesi s H+: 0 = 0o + eh against K: 0 = 0o, where h is any
positive number.
Let c, = 89 + ep + q~, where q~ = o(e) is a sequence. Fr om Theor em 4.4, we see
P [ l ~ h ( w ; O o + e h ) < c , ] , = = 8 9 l h 3 [ J -1 - 1] -~gB~, 1, 1 h3} ~b(0; 0, J)
If we t ake
+ q~q~(0; 0, J) + O(~2).
i h 3
p = - A l , l , l h 3 [ J - 1 - 1] + ~B1,1,1
and
q~ = e 3/2 + O( O; O, J ) - l l P~, oo+~h[ e - l ( T~ - - Oo - - eh) <= 0] - 8 9
t hen we see by Neyman- Pear s on' s l emma
P[l~, h(w; 0o) <= c~] > P, , o o [ e - i ( T ~ - 0o) <= h]
for small e. Therefore, by Theor em 4.2, for h > 0,
1
lim inf e- 1 {@(j; O, J ) + e [ A1 , ~, 1 h 3 + ~B1, 1, 1 h3] ~b(J; 0, J)
~+o
- e ~ , 0 o [ ~ - I ( T ~ - 0 o ) < h ] } _>_ 0 .
Similarly, for h < 0 we have
l i msup e - l { ~ ( - J ; 0 , J ) - e [ A l , l , l h 3 + 8 9 l h 3 ] ~ ( J ; O, J )
e$0
- e ~ , 0 o E ~ - ' ( r ~ - 0 o ) _-< h ] } _%< 0 .
In this sense
and
9 (J; O, J ) + e [ A ~ , l , a h 3 + 8 9 O, J)
_ h 3 t
9 ( - - J ; 0, J) e [ Ai , a , ~ + ~ B l , l , l h 3 j d 2 ( J ; O, J )
are called the bounds of second or der distributions. An AMU est i mat or at t ai ni ng
these bounds for any h > 0 and h < 0 is called to be second or der efficient.
The bias correct ed maxi mum l i kel i hood est i mat or O*(w; 0o) with
b(Oo) = - A 1 , 1 , 1 I ( 0 o ) - a
is second or der AMU by Theor em 4.3 and it is seen t hat @(w; 0o) at t ai ns t he
bounds of the second or der distributions. Therefore, the bias correct ed maxi mum
l i kel i hood est i mat or is second or der efficient.
In conclusion, we not e t hat we have al ready got the asympt ot i c expansi ons for
vari ous risk functions, e.g., g [ l O * ( w ; O o ) - 0olP], p >^1 if we take a t emper ed
di st ri but i on T ( x ) = I x iv and not i ce t hat for any p > 1, 0* (w; 0o) has LP-moment s,
see Chap. 3 of Kut oyant s [6].
Acknowledgements. The author is grateful to Professor S. Watanabe for his helpful advice and
also wishes to thank the referee and the associate editor for their valuable comments.
Asymptotic expansion for small diffusion 311
References
1. Akahira, M., Takeuchi, K.: Asymptotic efficiency of statistical estimators: concepts and higher
order asymptotic efficiency. (Lect. Notes Stat., Vol. 7) Berlin Heidelberg New York: Springer
1981
2. Ghosh, J.K., Subramanyam, K.: Second order efficiency of maximum likelihood estimators.
Sankhy~, Ser. A 36, 325-358 (1974)
3. Ikeda, N., Watanabe, S.: An introduction to Malliavin' s calculus. In: It(3, K. (ed.) Stochastic
analysis. Proc. Taniguchi Intern. Syrup. on Stochastic Analysis, Kat ada and Kyot o 1982,
pp. 1-52. Tokyo: Ki nokuni ya/ Nort h-Hol l and 1984
4. Ikeda, N., Watanabe, S.: Stochastic differential equations and diffusion processes, 2nd edn.
Tokyo: Nor t h- Hol l and/ Kodansha 1989
5. Kusuoka, S., Strook, D.W.: Applications of the Malliavin calculus, Part I. In: ItS, K. (ed.)
Stochastic analysis. Proc. Taniguchi Intern. Syrup. on Stochastic Analysis, Kat ada and Kyot o
1982, pp. 271-306. Tokyo: Ki nokuni ya/ Nort h-Hol l and 1984
6. Kutoyants, Yu.A.: Parameter estimation for stochastic processes. Prakasa Rao, B.L.S. (ed.)
Berlin: Heldermann 1984
7. Kutoyants, Yu.A.: Expansion of a maximum likelihood estimate by diffusion powers. Theory
Probab. Appl. 29, 465-477 (1984)
8. Liptser, R.S., Shiryayev, A.N.: Statistics of random processes I. Berlin Heidelberg New York:
Springer 1977
9. Malliavin, P.: Stochastic calculus of variation and hypoelliptic operators. In: It6, K. (ed.)
Proceedings of the International Symposium SDE Kyoto, pp. 195-263. Tokyo: Kinokuniya
t976
t0. Matliavin, P.: k C -hypoellipticity with degeneracy. In: Friedman, A., Pinsky, M. (eds.) Stochas-
tic analysis, pp. 199-214, 327 340. New York: Academic Press 1978
11. Rao, C.R.: Asymptotic efficiency and limiting information. In: Proc. Fourt h Berkeley Symp.
on Math. Statist. Probab. Vol. 1, pp. 531-545. Berkeley and Los Angeles: University of
California Press 1961
12. Rao, C.R.: Criteria of estimation in large samples. Sankhy~ 25, 189 206 (1963)
13. Taniguchi, M.: On the second order asymptotic efficiency of estimators of Gaussian ARMA
processes. Ann. Stat. 11, 157 169 (1983)
14. Taniguchi, M.: Higher order asymptotic theory for time series analysis (Lect. Notes Stat. Vol.
68), Berlin Heidelberg New York: Springer 1991
15. Watanabe, S.: Malliavin' s calculus in terms of generalized Wiener functionals. In: (Lect. Notes
Control Inf. Sci., Vol. 49) Berlin Heidelberg New York: Springer 1983
16. Watanabe, S.: Lectures on stochastic differential equations and Malliavin calculus. Tat a
Institute of Fundamental Research. Berlin Heidelberg New York: Springer 1984
17. Watanabe, S.: Analysis of Wiener functionals (Malliavin calculus) and its applications to heat
kernels. Ann. Probab. 15, 1-39 (1987)
18. Yoshida, N.: Asymptotic expansion for small di ffusi on-an application of Malliavin
Wat anabe theory. Research Memorandum Vol. 383. The Institute of Statistical Mathematics,
1990