ON THE QUADRATIC VARIATION OF SEMI-MARTINGALES by MARC LEMIEUX B . S c , M c G i l l U n i v e r s i t y , 1981 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE i n THE FACULTY OF GRADUATE STUDIES (Department of Mathematics) We accept t h i s thesis as conforming to the required standard THE UNIVERSITY OF BRITISH COLUMBIA July 1983 • Marc Lemieux, 1983 ) I n p r e s e n t i n g t h i s t h e s i s i n p a r t i a l f u l f i l m e n t o f t h e r e q u i r e m e n t s f o r an a d v a n c e d d e g r e e a t t h e U n i v e r s i t y o f B r i t i s h C o l u m b i a , I a g r e e t h a t t h e L i b r a r y s h a l l make i t f r e e l y a v a i l a b l e f o r r e f e r e n c e and s t u d y . I f u r t h e r a g r e e t h a t p e r m i s s i o n f o r e x t e n s i v e c o p y i n g o f t h i s t h e s i s f o r s c h o l a r l y p u r p o s e s may be g r a n t e d by t h e h e a d o f my d e p a r t m e n t o r by h i s o r h e r r e p r e s e n t a t i v e s . I t i s u n d e r s t o o d t h a t c o p y i n g o r p u b l i c a t i o n o f t h i s t h e s i s f o r f i n a n c i a l g a i n s h a l l n o t be a l l o w e d w i t h o u t my w r i t t e n p e r m i s s i o n . D e p a r t m e n t o f The U n i v e r s i t y o f B r i t i s h C o l u m b i a 1956 Main Mall V a n c o u v e r , Canada V6T 1Y3 D a t e 7 DE-6 (3/81) - i i -ABSTRACT Let X be a semi^martingale, Techniques of [1] and E l Karoui ( in [3] and [4]) are used to study the convergence of 2£ times the number of upcrossings of [x, x+e] by X to i t s l o c a l time at x . I f X i s continuous and i f there ex i s t s a bicontinuous ver s ion of i t s l o c a l time process , then of f a s i n g l e n u l l se t , the convergence i s shown to be uniform i n „ (and t ime) . I f X i s such that the sum of the absolute value of i t s jumps over any f i n i t e time i n t e r v a l i s almost sure ly f i n i t e , then, o f f a s i n g l e n u l l se t , the convergence holds at a l l but countably many x . A not ion of genera l ized arc length, i s introduced , i n the s p i r i t of the quadrat ic arc length of [1 ] , and the l a s t r e s u l t above i s c c used to show that <X , X > i s the almost sure arc length of X , a uniform l i m i t recoverable from the geometry of the t r a j e c t o r i e s . - i i i -TABLE OF CONTENTS Page ABSTRACT i i TABLE OF CONTENTS i i i ACKNOWLEDGEMENT iv CHAPTER I Pre l iminary cons iderat ions 1. Presentat ion of r e s u l t s 1 - 3 2. Basic notions 3 - 6 CHAPTER II The upcrossings of a semi-martingale 1. A stopping time representat ion 6 -13 2. Uniform convergence to l o c a l time 13-20 CHAPTER I I I General ized arc length for semi-martingales 1. D e f i n i t i o n s , 21-2. A s p e c i a l case 21-23 3. The jump case , , , 23-37 BIBLIOGRAPHY. 38 - i v -ACKNOWLEDGEMENT Ed Perkins, the supervisor of t h i s work, deserves c r e d i t and thanks for h i s c l e a r understanding of i n f i n i t e s i m a l matters the e f f o r t he invested i n sharing i t with me. - 1 -I - PRELIMINARY CONSIDERATIONS 1. Presentat ion of r e s u l t s For a process X and a p a r t i t i o n Q of [ 0 , t ] , consider 2 the sum S(X,Q) = Z (X - X ) It i s well-known (for example, t €Q i+1 t i i see 5, p . 355 ) that i f X i s a semi-martingale, then S(X, Q^) -> [X,X] i n p r o b a b i l i t y and L''" whenever the Q 's are success ive t n d i v i s i o n s of [ 0 , t ] whose mesh goes to zero as n goes to i n f i n i t y . There i s no almost sure r e s u l t of t h i s k ind for a r b i t r a r y semi-mart ingales , but when B i s Brownian motion, i t i s known that S(B, Q^) -> t a . s . provided the Q n ' s a r e nested p a r t i t i o n s of [ 0 , t ] [2, p. 395]. The convergence of S(X,Q) uniformly over a l l p a r t i t i o n s of small mesh f a i l s : Taylor [7] has shown that sup S(B,Q) = °° a . s . In f a c t , he showed that sup I f (B - B ) i s i n f m i t e i f r Q t . e Q fcl+l t i A * , 2 i s a funct ion such that log log (s) f (s) / s ->• °° as s 0 , where log * ( s ) = max( l , | l o g ( s ) | ) , and i n [6 ] , Munroe has used t h i s r e s u l t to construct a square-integrable mart ingale M and nexted p a r t i t i o n s Q n of [ 0 , t ] such that sup S(M,Qn) = °o . This shows how the quadrat ic v a r i a t i o n approach i s inadequate i n i d e n t i f y i n g [X,X] as a uniform (or almost sure) l i m i t for a r b i t r a r y semi-martingales . There i s , however, another way to look at t h i s problem: i n C l ] , the t r a v e r s a l time for sect ions of path of Brownian motion was computed only us ing information a v a i l a b l e from the geometry of the t r a j e c t o r i e s . The t r a v e r s a l - 2 -time (or quadratic v a r i a t i o n , f o r Brownian motion) was recovered as an almost sure uniform l i m i t from the range of the process, independantly of any time parametrization; more p r e c i s e l y , i t was shown that f o r almost a l l sample paths, t i s the uniform l i m i t over a l l p a r t i t i o n s Q = (x^ : i^Z) of mesh less than e as e goes to zero of: (*) I(x_+i ~x_) x o f crossings of [x±, x _ + _ 3 b v B o v e r l".0,t_) For Brownian motion, t h i s provides a d e f i n i t i o n of a generalized arc length, equal for almost a l l sample paths to the t r a v e r s a l time, independantly of p a r t i c u l a r parametrizations. This thesis i s an extension of t h i s state-space approach to the case of semi-martingales. The main r e s u l t i s theorem ( I I I . 3 . 3 ) ; i t shows that c c c the process <X , X > , where X denotes the continuous martingale part of X , i s the uniform l i m i t , over a l l p a r t i t i o n s of the state-space, of sums of type (*) f o r a broad c l a s s of semi-martingales. Let X be a semi-martingale whose sum of absolute jumps before time t i s almost surely f i n i t e f o r each t . In [ 4 ] , E l Karoui has shown that the l o c a l time L^ _ of X i s , f o r each x , the almost sure l i m i t , as e goes to zero, of e times the number of upcrossings of the i n t e r v a l Dx,x+e] by X up to time t . Theorem ( I I I . 3 . 3 ) complements t h i s r e s u l t by showing that - 3 -the n u l l se t s , which depend on x , can be combined in to a s i n g l e one of f which the convergence holds at a l l but countably many x . The proof of ( I I I .3 .3) combines a g r i d argument of [1] and a stopping time representat ion for the upcrossings due to E l Karoui [43; both are presented i n Chapter I I , where theorem ( I I .2 .4 ) genera l izes a r e s u l t of Perkins [1 , thm. 2 ] : we show that i f a continuous semi-martingale has a bicontinuous v e r s i o n of i t s v e r s i o n of i t s l o c a l time process , then o f f a s i n g l e n u l l se t , the convergence of e times the number of upcrossings of [x, x+e] i s uniform i n x and t . Th i s provides a new g l o b a l c h a r a c t e r i z a t i o n of the l o c a l time associated with these semi-martingales . 2. Basic notions I t i s now time to introduce some conventions, and r e c a l l b r i e f l y the d e f i n i t i o n s and elementary r e s u l t s per t inent to t h i s study. We give ourselves a f i l t e r e d p r o b a b i l i t y space ( f i , F , F t , P ) s a t i s f y i n g the usua l hypotheses on which a l l processes considered i n the next chapters are de f ined . A semi-martingale X i s any adapted process that can be w r i t t e n as the sum of a l o c a l mart ingale M and an adapted process V of bounded v a r i a t i o n on the compact subsets of R + . We w i l l denote the d i s c o n t i n u i t y jumps X g - X g _ by & X s > | d v s | w i l l denote the t o t a l v a r i a t i o n measure induced by V on the Bore l sets , and X w i l l denote the continuous mart ingale part of X . A semi-martingale i s c a l l e d s p e c i a l i f i t admits a representat ion M + V as above where, i n a d d i t i o n , V i s p r e d i c t a b l e ; because of p r e d i c t a b i l i t y , t h i s representation i s unique, and M + V i s c a l l e d the canonical decomposition of X . A semi-martingale i s s p e c i a l i f and only i f the t o t a l v a r i a t i o n of V i s l o c a l l y integrable, and a continuous semi-martingale i s s p e c i a l [ 5]. If X i s a s p e c i a l semi-martingale, we denote i l k k |dV_I) } and we l e t H be the space |X||, = E 1 / k { [ M , M ] k / 2 + ( of a l l X such that ||x|| < « . In [5], P.A. Meyer has defined the l o c a l time process associated to semi-martingales through Tanaka's formula. Namely, l o c a l time i s the continuous, adapted process L which s a t i s f i e s ( 1 . 2 . 1 ) \ L X = ( X t - x ) + - ( X 0 - x ) t - - 1(X _>x)dX - \<L? s s 2 t where \l* = _ ( x -x) 1(X >x) + _ (X -x ) + l(X _<x) s<t S s<t S S and we define LA = L A + Ji^ t t t Meyer has also shown that the l o c a l time process s a t i s f i e s a density of occupation formula: for any p o s i t i v e , Borel measurable function f : R + -> R , f(X )d <X C,X C> = s ' s ( 1 . 2 . 2 ) 0 3R f ( x ) L X d x a.s. This l o c a l time process has been extensively studied. In p a r t i c u l a r , i n [9], Yor has considered the sample path co n t i n u i t y - 5 -properties of the processes L, Z and L ; he has shown that i f a semi-martingale X i s such that the sum of the absolute values of i t s jumps before time t i s almost surely f i n i t e f o r each t , then: (1.2.3) (a) there e x i s t s a v e r s i o n of everywhere cadlag i n (x,t) , (.b) there e x i s t s a version of L everywhere cadlag i n x and continuous i n t , Cc) the jumps of these processes are given by t ~ ^ t = 2 [ l ( X g _ = x ) d V s + I ( A X s ) l ( X s _ = x)] s<t L X - L X = 2 t -t "J 1 (V = X>dVs 0 t = 2[ 0 l ( X s _ = * ) d X s - I (AX s)l(X _ = x ) ] s<t s We w i l l always assume that we are working with these versions whenever conditions permit. Note that the l a s t formula i n (c) implies that a bicontinuous v e r s i o n of L e x i s t s i f and only i f the measure | dV | does not charge the l e v e l sets '{s: X _ = x} , s s f o r any x e R . I t also shows that the stochastic i n t e g r a l t / l ( X _ = x)dX i s the sum of £ AX 1(X _ = x) and a continuous 0 S S s<t S S process of bounded v a r i a t i o n on compacts, of support contained i n {s : X _ = x} . s The proof of (1.2.3) r e l i e s on an estimate f o r a p a r t i c u l a r - 6 -type of stochastic i n t e g r a l s . We record i t below, as i t w i l l be quite h e l p f u l i n handling i n t e g r a l s which a r i s e from the upcrossing representation: (1.2.4) (Yor, [9]): If X e H P , then there e x i s t s a constant c < °° such that: P f 2P | K a < X <b)dX^_|| 2 = E{sup( 1 s s H p t J 0 0 l ( a < X _ < b ) d < X ° , X c > ) P} s s < C (b-a) P||x|| V H-F i n a l l y , we record another r e s u l t of Yor; i t i s an easy consequence of the density of occupation formula and the continuity properties of l o c a l time: (1.2.5) (Yor, [9] ) : Let X be a semi-martingale such that •_ |AX | i s f i n i t e a.s. f o r each t . Then f o r each t , s<t S l i m (h e>0 % t C „ C T X l(x<X_ <x+e)d< X.X1- > = LT" , f o r a l l x , a.s. s s t II - UPCROSSINGS OF A SEMI-MARTINGALE 1. A stopping time representation Throughout the remaining chapters, X denotes a semi-martingale, J(X) denotes the sum of d i s c o n t i n u i t y jumps _ AX and K(X) denotes the increasing process _ |AX | . I f S t s s<t s<t i t i s s p e c i f i e d that K(X) t i s almost surely f i n i t e f o r each t , then we write M + V f o r the canonical decomposition of the continuous semi-martingale X - J : note that i n t h i s case, the representation X = M + V + J i s unique, but i f , i n addition, X i s i n f o r some p , M + V + J i s not, i n general, the canonical decomposition of X . Let I = [a,b] be an i n t e r v a l . We describe the successive entry and e x i t times of I by X a f t e r time s with the following stopping times: S^(s) = i n f ( t > s : X < a) T b ( s ) = i n f ( t > S a ( s ) ; X_ > b) , f o r n > 1 n n t S a(s) = i n f ( t > T b ,(s) : X„ < a) , f o r n > 2 n n—1 t a b As usual, i n f ((f)) = 0 , and we agree to write and a b for S (0) and T (0) r e s p e c t i v e l y . When there i s no possible n n confusion, we w i l l omit the superscripts a and b . We define the number of upcrossings of the i n t e r v a l [a,b] by X between times s and t as: N +(s,t,a,b) = £ 1(T (s) <t) . n>l n S i m i l a r l y , we define the corresponding number of downcrossings as: N"(s,t,a,b) = I 1(S (s) <t) + 1(X >b) , and we set N(s,t,a,b) , n>2 n S the t o t a l number of crossings, as N +(s,t,a,b) + N~(s,t,a,b) . We w i l l modify the notation a l i t t l e b i t and write N"'"(s,t,I) (respectively N~(s,t,I) and N(.s,t,I)) instead of N +(s,t,a,b) (respectively N~(s,t,a,b) and N(s,t,a,b)) when i t i s more - 8 -convenient to do so. We are interes t e d i n i s o l a t i n g the pieces of a sample path that constitute an upcrossing from those that constitute a down-crossing, so we introduce the sets: U +(I) =U + (Ca,b]) = u ((S ,T ]] n n n>:l = { ( c o,s) e 0, x ]R+ ; there e x i s t s an n f o r which S (CJ) < s < T ((D) } n - n and tT(I) = U-(Ca,b]) = uC.CS ,T )) . i n n n>l = { (io,s) e Q, x]R + t there e x i s t s an n f o r which S (d i ) < s < T (co)} n n and we define D +(I) = D + (Ca,b]) ..and D _(I) = D"(Ca,b]) as the complements of IT^CO and U - ( I ) r e s p e c t i v e l y . C l e a r l y , U+(I) , and hence D+(I) , are predictable, while U (I) and D~(I) are optional sets. The next proposition shows how these sets and stopping times given an expression f o r the number of upcrossing of an i n t e r v a l . Although i t s proof can be found i n [43, i t i s reproduced here f o r the sake of completeness, (II.1 . 1 ) PROPOSITION ( E l Karoui): Let X be a semi-martingale, and l e t I = [a ,b3 be an i n t e r v a l . Then: (b-a)N+(0,t,a,b) t = -(X Q-a)+ - 1(D +(I))1(X _>a)dX s s + (X -a)l(D-(I» - I (X - a ) l ( D _ ( I ) ) l ( X _=a<X ) s<t S s s - I (X T -b)l(a<X T _<b<XT ) l ( T n < t ) n 1 n n n - I (X - a ) l ( X <a<b<XT )1(T <t) n>l T n V T n n - I (a-X )1(X_ >a>X„ )1(S <t) -, b b - b n n>l n n n + (b-a) I 1(X _<a<b<X )1(T <t) n>l n n n Proof• The idea i s to express (b-a)l(T n<t) using X T A and n X 0 . By d e f i n i t i o n of the stopping times S and T , we have S A t n n n that X T _^b<XT and X g _^a>Xg for a l l n . Moreover, i f n n n n X < a , then S = 0 (and X = X = X ) , while i f X > a , U J_ J_ 1 then > 0 . From t h i s , i t follows that for n > 2 , (1) (b-a)l(T <t) = (X -X„ )1(T <t) n 1 b n n n - (X T - b ) l ( X T _^b<XT ) l ( T n < t ) n n n - (a-X„ )1(X_ >a>X_ )1(S <t) b b — b n n n n + (a-X Q )1(X„ >a>Xc )l(S^<t<T ) b b - S n n n n n - 10 ,s If n = 1, then one must substract (a - X Q ) 1 ( X 0 < a) from the right-hand side i n order to get a correct expression for (b-a)l(T 1<t) . We note that f o r a l l n , (X T - X s ) l ( V t ) = X T A t - X s A t - ( X t - X s ) K S n . t < T n ) n n n n n i ( ((s ,T n )dx - (x - x . ) i ( r r s r ) ) ) n n b t S n n n and therefore, i n view of (1), by summing (b—a)l(T ^t) over n > 1 , using the d e f i n i t i o n of D'(I) , we get: t (2) ( b-a^CO^.a . b ) = (a-X 0) - l(D+(I))dX s + ( X t - a ) l ( D _ ( I ) ) - I (X - b ) l ( X _<b<X )1(T <t) n>l n n n n - I (a-X )1(X >a>X„ )1(S <t) n>l S n S n " S n n - (a-X 0)l(X 0<a) To f i n i s h the proof, we need to examine some of the terms of (2) more c l o s e l y . By d e f i n i t i o n , D +(I) n (X _ < a} = cf> and so the stochastic i n t e g r a l i s equal to 1(D+(I))1(X _>a)dX . Moreover, s s 0 i f we denote the set D +(I) n.{X _= a} by J(a) , then by (1.2.3), t S the i n t e g r a l l ( J ( a ) ) d X i s equal to _ (X :-a)l (J(a)) + s<t l ( J ( a ) ) d B (a) , where the process B(a) i s continuous, of - 11 -support Included i n {X _ = a} . Now, note that s J(a) = {(u),s) e D +(I) : there e x i s t s an n f o r which s = S (to) , and X (w) = a} n s _ n u {(co.s) e D+(I) , X = a < X (u))} ; s s t h i s shows that f c the sections of J(a) are countable; therefore, the i n t e g r a l l ( J ( a ) ) d B (a) i s zero. F i n a l l y , because the set 0 {((0,s) : s = Tn(uO and X T (u) = a) i s contained i n D"(I) , n _ we have that: I (X - a ) l ( D + ( I ) ) l ( X = a < X ) = £ (X - a ) l ( D - ( I ) ) l ( X _=a<X ) s<t fa s<t s s - I (X - a ) l ( X =a)l(T <t) n>l T n T n ~ n The r e s u l t of these comments i s the following expression f o r the i n t e g r a l of (2): t (3) l ( D + ( I ) ) d X s = l(D+(I))l(X s_>a)dX c J + I (X s-a)l(D-( I))l(X s_=a<X ) s<t - I ( X T - a ) l ( X T _=a)l.CT n<t) n>l n n + I (X -a)l(X„ =a)lCS <t) n>l b n V n 12 V, Now, using simple properties, the two sums of (2) become; (4) I (a-X )1(X >a>X )1(S <t) o b - b n n>l n n n = I (a-X )1(X_ ^ K S ^ t ) n ^ l n n + I (a-X )1(X >a>X )1(S <t) T b b n n>l n n n and (5) I (X T - b ) l ( X T _<b<XT ) l ( T n < t ) n>l n n n = I (X T -b)l(a<X T _<b<XT ) l ( T n < t ) n>l n n n = I (X T - a ) l ( X T _<a<b<XT ) l ( T n < t ) n>l n n n + (a-b) I 1(X T _^a<b<X )1(T <t) n>l n n Noting that Ca-XQ) - (a-X 0)l(X Q<a) = - ( X 0 - a ) + , and su b s t i t u t i n g (3), (4) and (5) back into (.2) f i n i s h e s the proof. QED. Remark: Def i n i t i o n (1,2.1) and Proposition (11.1,1) give r i s e to an expression f o r the differe n c e between (b-a) times the number of upcrossings of the i n t e r v a l I = [a,b] and the l o c a l time at a . Namely, (11,1.2) t l T a (b^a)N+(0,t,a,b) - ± L a = l ( u + ( I ) ) l ( X s _ > a ) d X s 0 + [ ( X t - a ) l ( D - ( I ) ) - ( X t - a ) " r ] + E t(a,b) where (II.1 . 3 ) E t(a,b) = (b-a) _ 1(X T _<a<b<XT ) l ( y t ) n>l n n + C ._ (X - a ) l ( X <a<X ) - _ (X - a ) l ( X _ <a<b<XT )1(T <t) ] + C I (X - a ) l ( X =a<X ) - I (X - a ) l ( D - ( I ) ) l ( X _=a<X )] s<t s s g ^ t s s s + C I (a-X )1(X >a>X ) - I (a-X_ )1(X >a>X )1(S <t)] s<t S S S n>l S n S n " S n n - I (X T -b)l(a<X T _<b<XT ) l ( T n < t ) n>l n n n 2. Uniform convergence to l o c a l time Our aim i s now to show that, f o r c e r t a i n continuous semi-martingales, 2 (b-a)N (0,t,a,b) converges to (as b -I- a) , uniformly i n a and t , o f f a si n g l e n u l l set. This w i l l generalize, f o r these semi-martingales, what i s already known of Brownian motion [ 1 ] . - 1 4 -An easy argument using the d e f i n i t i o n of D~(I) shows that | ( X - a ) l ( D " ( I ) ) - ( X t - a ) + | < (b-a) . Indeed, i f t e D~(I) , then X ^ a and the left-hand side i s zero. It i s also zero i f t | D~(I) and X < a . In the only case that remains, we have a < Xfc < b and t h i s makes the conclusion c l e a r . The following lemma gives conditions under which the Lebesgue-S t i e l t j e s part of the stochastic i n t e g r a l of (II.1.2) converges uniformly over the state-space: (II.2.1) LEMMA: Let X be a semi-martingale. (a) If X i s such that K(X) t i s f i n i t e a.s. f o r each t , then f o r each t , l i m l(x-e<X _<x+e) |d(V+J) | = 0 s s f o r a l l but couiitably many x e TR. , a.s. (b) If X i s continuous, and i f t i s such that x -»• L i s continuous, then t l i m sup e+0 xeR. l(u +([x,x+£ ]))1(X >x)dV | = 0 a.s. s s o Proof of (a): F i x t , and pick CJ out of a n u l l set so that t K(X) and /| dV | are f i n i t e . Then, by dominated convergence, 0 s' t lim e-s-0 l(x-e<X <x+e)|d(V+J) | < s s o The conclusion now follows from the f a c t that i f i(x s_=x)|dv s| + I |AXS|KXS-=X) s^t - 15 -x £ {y : L y ^ L y } u {X ,X _: s<t, AX #)} , then the right-hand t t s s s side of the i n e q u a l i t y i s zero ( r e c a l l (1.2.3)), and t h i s set i s countable, since both and X are cadlag. Proof of (b)' Suppose that t i s such that 1/ i s continuous. Pick di out of a n u l l set so that f IdV I i s f i n i t e . Note that 0 s by d e f i n i t i o n of U +(«) , l(u +([x,x+ej))l(X s>x) < l(x<X g<x+e) , and therefore, by monotonicity, i t i s enough to show that f (x) •> 0 t n uniformly, where f (x) = J l(x<X <x + l/n)|dV | . C l e a r l y , the n Q s s f^'s are decreasing, and by dominated convergence, ^ n ( x ) ^ 0 point-wise. Now, consider g(x) = J 1 (X < x) | dV | . I f y 4- x , o s s then g(y) -> g(x) by dominated convergence, and i f y '+ x , then t g(y) converges to J l ( X <x)|dV | which i s equal to g(x) since o s s |dVg| does not charge the l e v e l sets {X g = x) by co n t i n u i t y of • L . This shows that f (•) i s continuous for each n , since t n f n ( x ) = g(x + l/n) - g(x) . To f i n i s h the proof, we note that for f i x e d co , the maps r n ( " ) have compact support, so that by Dini's theorem, sup f (x) -> 0 . QED. x n Remark: In the p a r t i c u l a r case of a continuous semi-martingale, t part (a) of the lemma says that a.s., |/l(x-e<X <x+e)dV | goes Q s s to zero with e whenever x i s a point of continuity of 1/ . As we have a sharp L P estimate to handle s i m i l a r i n t e g r a l s , i t i s natural to turn to a B o r e l - C a n t e l l i argument to get a uniform bound for the martingale part of the stochastic i n t e g r a l of (II.1.2). F i r s t , as i n [1], we d i s c r e t i z e the state-space by - 16 -defin i n g a c o l l e c t i o n of grids; f o r k e B , l e t R k = Uk 7 : i e 7L , | i | <k8} (II.2.2) LEMMA: Let X be a semi-martingale l o c a l l y i n H 2 Then, f o r any B > 0 , the set t s {max sup xeR, t<B I k 0 l(u + ( [ x , x+k. 6])>1(X _>x)dX^| > k 1 / 2 i.o.} i s n u l l . Note: We may (and w i l l ) assume that the stochastic i n t e g r a l s t Y (x,k~ 6) = l ( u + ( C x , x + k " 6 ] ) ) l ( X _>x)dX° u s s J 0 are defined f o r a l l xe u R, o f f a sing l e n u l l set. k>l k Proof: By hypothesis, there e x i s t s a sequence of stopping times 2 (a : n > 1) such that a f 0 0 and X e H for a l l n . n n tAcr /^c xn j „n, , -o, Denote by X^ , (X C)° and Y^(x,k b) the processes X_ A , t -ft n x£ and Jl(u +([x,x+k ]))1(X n_>x)d(X°) n r e p s e c t i v e l y . Let n 0 B > 0 . Then, f o r any n , i t i s straightforward that: (1) P{max sup lY^x.k" 6)! > k " 1 / 2 i.o'.} xeR^. t<B t < P{max sup|Y^(x,k 6 ) | > k " 1 / 2 i.6 .} + P{a < B} xeR k t<B n Using Chebyschev's i n e q u a l i t y and the fac t that #R^ = 2k , l e t t i n g c be a constant depending only on n whose value may change from l i n e to l i n e , one obtains P{max sup | Y^(x,k" 6)| > k~ 1 / 2} xeRfc t<B - 17 -< ck 1 0E{sup|Y^(x,k - D ) r} t<b ,11, n -6. I 4-, < c k 1 0 E { ( l(u +([x,x+k 6 ] ) ) l ( x n _ > x ) d < ( X C ) n > ) Z} where i n the l a s t l i n e , we used the Burkholder-Davis i n e q u a l i t y . Now use the fa c t that l(u + (Cx,x+k~ 6]))l(X n_>x)<l(x<X n_<x+k~^) s s and L P-estimate (1.2.4) to see that the right-hand side i s majorised -2 by ck . Therefore, by the B o r e l - C a n t e l l i lemma, P{max sup|Y^(x,k ^) | > k i.o.} = 0 for any n . Pick n xeR. t<B k large enough so that > B a.s.; the conclusion now follows (1). QED. (II.2.3) Remark: If X i s continuous, the error term E t(a,b) of (II.1.2) i s zero, and using the argument made i n the beginning of t h i s section, one finds that: t | (b-a)N+(0,t,a,b) < |J l (u+([a,b]))(X s_>a)dX g| + C(X t-a)l(D ([a,b])) - ( X t - a ) + ] l ( u +([a,b]))l(X _>a)dV I + s s Ku + ( C a , b ] ) ) l(X s> a)dM g 0 + (b-a) It follows from t h i s and both previous lemmas that for any 6 > 0 and each t such that 1/ i s continuous, the maximum of - 18 -|k ^N +(0,t,x,x+k ^) - I o v e r the g r i d R^ Is bounded, o f f a —1/2 —6 n u l l set, by 6 + k + k for large enough k . This i s the key fact needed to get uniform convergence of (II.1.2): (II.2.4) THEOREM: Let X be a continuous semi-martingale, and suppose a bicontinuous version of i t s l o c a l time process e x i s t s . Then, with p r o b a b i l i t y one, for any B > 0 , lim sup |eN +(0,t,x,x+e) - y L X | = 0 e-K) xeH t<B Note: To prove t h i s , i t i s enough to show that f o r each t , (*) l i m sup | eN +(0,t,x,x+e) - j L X | = 0 a.s. e-K) xe]R Indeed, given any <S > 0 , because Lj i s uniformly bicontinuous over [0,B] x ]R" (for fixed OJ , i t has compact support i n the space n va r i a b l e ) we can p a r t i t i o n [0,B] in t o u [ t . , t . .^1 , where i = l i i+1 0 = t„ < t„ < ... -, t , < t = B i n such a way that f o r a l l 1 2 n-1 n -1 'i+1 I X X I L - L I < 6 . Then i f t £ [t.. , t ) we use the i t n - j . i t i 1 1 + 1 t r i a n g l e i n e q u a l i t y and the monotonicity of N~*"(0 -,x,x+e) to obtain: | eN +(0,t,x,x+£) - y L X | < | eN+(0,t,x,x+£") - £N +(0,t i,x,x+£) | + |£N +(0, t l,x,x+£) - | L X I + | | L X - | L X | i i < |£N +(0,t 1 + 1,x 1x+£) - y L X I + 2|eN+(0,t.,x,x+e) - y L X | + 5 i+1 i From the l a s t i n e q u a l i t y , the s u f f i c i e n c y of (*) i s apparent. - 19 -Proof of (*): F i x t > 0 , and l e t 0 < 6 < 1 . By remark (II. 2.3), we can select w out of a n u l l set, and N = N(o)) such that the following hold: (1) ( k - l ) ~ 6 - Ck-1)~7 > k" 6 , for a l l k > N (2) sup |X (u))| + 1 < N s<t S (3) max |k 6N +(0,t,x,x+k 6 ) - y L X | < <5 , for a l l k> N x e R k F i x e.<:(N+l)~6 and l e t m > N+1 be such that (m+l)~ 6 < e < m~6 . If |x| > N - l , then L X = N +(0,t,x,x+e) = 0 . On the other hand, i f |x| < N - 1 , then define x^ = max(y e R^ : y < x) f o r a l l k > N . It i s easy to v e r i f y that: (4) x < x < x + e < x , + ( m - l ) ~ m-I m-1 (the l a s t i n e q u a l i t y follows from the choice of m >e (1) and the fac t that x - k 7 < x, .') and (5) x < X m + 2 + (m + 2)"7 < X m + 2 + (m + 2)"7 + (m + 2)"6<x + e (here, the l a s t i n e q u a l i t y uses the fact that (2) y i e l d s (m + 2)"6 + (m + 2)"7 < (m + l ) ~ 6 .) For convenience, set I . = [x , , x - +(m-1) ] m-I m-1 m-1 a n d 42 = Cxm+2 + ( m + 2>" ?» xmf2+(m+2)-7 + (m+2)"6] Then, (4) and (5), along with the monotonicity of I N +(0,t,I) - 20 -and the choice of m imply that (6) (nrf-l)~V(0,t,I _) - y L X m-l z t < eN+^t.x.x+e) - y L X " m ( 0' t' Im+2 ) " 2 L t Next, we note that: sup | (m+l)" 6N +(0,t,I .) - y L X | xelR m _ 1 2 t < (m+l)" 6(m-l) 6 max I (m-l)"V(0,t,Y,Y+(m-l)" 6 -kl ' YeR , Z m-l X ,,,,..-6, -.6 i l T m-l 1 T x I + (m+1) (m-l) sup | _ L f c ~2t ' xelR + C (m+1)" 6(m-l) 6-1] sup | y L X | xelR By the uniform c o n t i n u i t y and boundedness of 1/ , the l a s t two terms on the right-hand side go to zero as m + °° ; therefore, by (3) f o r large enough m , i t i s majorized by 36 ... An i d e n t i c a l argument shows that sup lm 6 N + ( 0 , t , 1 ^ , „ ) - ^ - L X I < 36 for large m . xeJt This, i n view of (6), completes the proof. QED. 1 X I eN (0,t,x,x+e) - y L | -* 0 as xelR C e -> 0 because the differ e n c e between the number of upcrossings and down-crossings of a same i n t e r v a l i s at most one. - 21 -III - GENERALIZED ARC LENGTH FOR SEMI-MARTINGALES 1. D e f i n i t i o n s We now r e c a l l from Cl] the notion of arc length f o r stochastic processes. Throughout t h i s chapter, we w i l l denote by (x^ : i e ZZ) the points of a p a r t i t i o n Q of the r e a l l i n e , and ||Q|| w i l l denote the mesh of Q . We w i l l c a l l a p a r t i t i o n Q regular i f i t s a t i s f i e s x? < x ? M for a l l i , l i m x? = » , and l i m = -» . 1 l + l ' . x , i i -voo i - > - o o We w i l l denote by TT the c o l l e c t i o n of a l l regular p a r t i t i o n s of R , and for any v > 0 , we define i r ( v ) to be the subset of i r c o n s i s t i n g of p a r t i t i o n s of mesh no larger than v . If Q e ' . i r and X i s a process, we define: K^(X,Q) = I ( x J + 1 - x J ) 2 N ( s , t , x ^ x 2 + 1 ) i (III.1.1) DEFINITION: For any a e R and e ' > 0 , l e t P ( a , e ) denote the l a t t i c e {a+ie : ie2Z). We say a process X has arc length L (where L: R + -> R) i f for a l l 0<s<t, l i m sup K t(X,P(a,e)) e+0 a s = L ( t ) - L(s) . 2. A s p e c i a l case It was shown i n [1] that the arc length of Brownian motion i s almost surely L(t) = t ; i n f a c t , i t was shown that |K^(B,Q) - ( t - s ) | converges to zero uniformly over a l l p a r t i t i o n s Q e i ( v ) as v 0 . Note that the uniformity of t h i s convergence f a i l s when p a r t i t i o n s of the time axis are considered [7]. The argument used i n the proof s t i l l holds for the semi-martingales of theorem - 22 -'(II.2.4) and y i e l d s the following: (III.2.1) PROPOSITION: Let X be a continuous semi-martingale whose l o c a l time process i s bicontinuous. Then, f o r a.a. sample paths, for a l l B > 0 , li m sup I K ^ X . Q ) - (<X,X> - <X,X> )| = 0 v+0 Qeir(v) s t s t<B Note: By d e f i n i t i o n , K^(X,Q) i s add i t i v e on path segments up to an error of sup |x (U))|'|[Q|| , and therefore, i t i s s u f f i c i e n t to s<t S assume s = 0 . Moreover, an argument s i m i l a r to the one i n the note preceeding the proof of (II.2.4) uses the uniform co n t i n u i t y of <X,X> on [0,B] and the monotonicity of KQ(X,Q) and shows that .': i t s u f f i c e s to prove: (*) l i m sup |K^(X,Q) - <X,X> | = 0 a.s. for each t v-K) Q£TT(V) Proof of (*): F i x t > 0 . By d e f i n i t i o n , we have that: KQ(X,Q) = (xj - x J ) N ( 0 , t , x ^ x J + 1 ) l [ X Q XQ ) ( x ) d x . Therefore, using the occupation time density formula (1.2.2), i t i s easy to see that: (1) sup | K'(X,Q) - <X,X> ! QeirCv) sup |(x^ + 1-x^)N +(0,t,x^,x^ )1 ( x ) - | L X | d x Qeir(v) Cx?,x? L l) l l + l -- 23 -+ j sup |(xJ+1-«J)N-(0,t,xJ+1)l ( x ) - ± L X | d x E Q £ 7 R ( V ) [ x i ' X i + l ) Now, observe that f or fixed CD , both integrands are of compact support, and that s u p i ^ - x ^ O . t . x J , ^ ) ! «-Kl ^ i ^ i + i ) xeJR x Q , sup Kx^ -xXw.t.xJ.x^ )-!^ 1 I + Qeu (v) 1 X - Q i • I 1 T 1 1 T X I s ? p q Q h L t - 2 L t l x€Lx^,x i + 1) Qeir(v) Theorem (II.2.4) and the uniform c o n t i n u i t y of show that the right-hand side of t h i s l a s t i n e q u a l i t y goes to zero with v . Thus, by dominated convergence, the right-hand side of (1) goes to zero with v , and t h i s completes the proof. QED. 3. The jump case The aim of t h i s section i s to generalize the r e s u l t of proposition (III.2.1) to semi-martingales X such that K(X) f c i s almost surely f i n i t e f o r each t . This extension i s a natural one; the almost sure convergence of K(X,Q) to <X,X> depends on the convergence of 2(b-a)N +(0 ,t ,a to L as a + x and b .4- x , and t h i s , modulo countably many x where |dV | may charge the sets {X _ = x} , hinges on s s the behaviour of the error term E t(a,b) ,. But in proving that - 24 -+ a 2(b-a)N (0,t,a,b) converges to L a.s. as b 4- a i n the jump case [4], E l Karoui showed that E t(a,b) goes to zero a.s. as b 4- a . The next lemma shows that, i n f a c t , E^(a,b) goes to zero as a i x and b 4- x whenever x i s not i n the range of the d i s c o n t i n u i t i e s of the process: (III.3.1) LEMMA: Let X be a semi-martingale such that K(X) f c i s a.s. f i n i t e for each t . Then for each t , with p r o b a b i l i t y one, (a) (El Karoui) lim|E (x,x+e)| = 0 for each x e TR. . £+0 '(b) l i m sup [E t(a,b) = 0 at a l l but countably many e+0 Ca,b]c[ x-e,x+£] x e: TR . Proof of (a): F i x t > 0 , and pick u out of a n u l l set so that K(X) t i s f i n i t e . Consider the following increasing processes: A (x) = I (X - x ) l ( X _<x<X ) S u<s u u u B (x) = T (x-X )1(X >x>X ) s L' u u - u u<s C (x) = y (X - x ) l ( X =x<X ) s u u~ u u<s For each x e TR , A f c(x) , B t(x) and C f c(x) are f i n i t e . Now f i x an x e TR . By ( I I . 1 . 3 ) , the following holds for any e > 0 : E t(x,x+e)| < |A t(x) - I (X T - x ) l ( X T _<x<x+e<XT ) l ( T n < t ) n>l n n n - 25 -+ | B t ( x ) - _ (x-X s )1(X S _>x>Xg ) l ( s n < t ) n>l n n n + |C (x) - I (X -x)l(D ([x,x+«_,))l(X _=x<X ) s<l s s + I (X T-(x+e))l(x<X T _<x+g<XT ) l ( T n < t ) n ^ l n n n + e I 1(X T _<x<x+e<XT HCTSt) n>l n n We w i l l show that each of the terms on the right-hand side of t h i s i n e q u a l i t y goes to zero with e . ( i ) Since every d i s c o n t i n u i t y jump of X from below x to above x+e x + e occurs at some T , using the d e f i n i t i o n of A (x) we n ' 6 s v f i n d that J (X T -x)l(X_ <x<x+e<XT )1(T <t) i;, T T - T n n>l n n n = I (X - x ) l ( X -<x<x+e<X ) = . ^_ s s s 8 " o 1(X >x+e)dA ( x) s s Therefore, |A t(x) - _ - x ) l ( X T _<x<x+e<XT )1(T St)| = n>l n n n 1(X <x+e)dA (x) s s and by dominated convergence and the d e f i n i t i o n of A g(x) , t h i s i n t e g r a l goes to zero as e -> 0 . Moreover, i t i s easy to see that the following i n e q u a l i t i e s hold: I (X T -(x+e))l-(x<X T _<x+e<XT ) l ( T n < t ) n>l n n . n < J (X -(x+e))l(x<X _<x+e<X ) < t; s s s s<t Q t l(x<X -<x+e)dK(X) and e I K X T ^<x+£<X )1(T <t) n>l n" n n < e I 1(X _<x<x+e<X ) = s<t (X^x" ) 1 ( X s ^ + e ) d (A+c> s oo s Again, using dominated convergence shows that the right-hand side of both these i n e q u a l i t i e s goes to zero as t -> 0 . X+£ ( i i ) Using the d e f i n i t i o n of the stopping times , we f i n d that 1(X >x+e)dB (x) = 7 (x-X )1(X _>x+e>x>X ) s s ^ s v s s s<t = I (x-X )1(X >x+e>x>X_ )1(S <t) i t> t> - b n n>i n n n < I (x-X )1(X >x>Xc )1(S £t) n>l S n S n ~ S n n < I (x-X )1(X >x>X ) = B. ( x ) s<t S S s fc Therefore, |B (x) - £ ( X - X _ )1(X Q >x>Xo )1(S <t)I < t n>l Sn S - " S n n n KX s_<x+e)dB s(x) and by d e f i n i t i o n of B (x) , t h i s i n t e g r a l goes to zero as s £ + 0 . ( i i i ) F i n a l l y , using the d e f i n i t i o n of D~(») , one finds that t C (x) - I (X -x)l(D-(Cx,x+eJ))l(X _=x<X ) = s<t l(u-(Cx,x+e]))dC (x) 0 But i f s e u ([x,x+e]) , then X <x + £ , and so t h i s l a s t i n t e g r a l - 27 -t i s majorized by / l ( X <x+e)dC (x) which goes to zero as e -»- 0 o s S This completes the proof of (a). Proof of (b) : F i x t > 0 , and pick to out of a n u l l set so that K(X) t i s f i n i t e . Choose x f' {X ,X : sSt,AX #)} , f i x £ > 0 and l e t a,b be any two r e a l s such that [a,b]c[x - e,x+ e ] . t t By d e f i n i t i o n of A (a) , Jl(X <b)dA (a) < Jl(X-€<X <x+e)dK(X) , s Q s s Q s s and therefore, i t follows from the ca l c u l a t i o n s of ( i ) that: (1) |A t:Ca)- I (X T - a ) l ( X T <a<b<XT ) l ( T n < t ) | < n>l n n n ^ It also follows from these c a l c u l a t i o n s that: t 1(x-e<X s<x+e)dK(X) (2) I -b)l(a<X T _<b<XT ) l ( T n < t ) < n>l n n n l(x - e<X <x+e)dK(X) s s 0 Moreover, using simple properties, one obtains (b-a) I 1(X _<a<b<X ) s<t < (b-a) I 1(X _<x-£<a<b<x+e<X )+ _ (AX )l(x - e<X _<a<b<x+e<X ) s<t S S s<t S S S + I (AX s)l(x-€<X s_^a<b<X s<x+e)+ _ (AX s)l(X s_<x-e<a<b<X g<x+6) and noting that the sum (b-a) _ 1(X _^x-e<x+e<X ) i s majorized s<t t 2 b y / ( T T T ^ T T ) 1 ^ -^x+e)l(X >x+e)d(A+C) (x) shows that: s (3) (b-a) I KX^ _<a<b<XT ) l ( T n < t ) n>l n n - 28 -t t ( x _( x- e )) 1(X sSx+€)d(A+C) (x)+ 0 l(x-e<X <x+e)dK(X) s s + 2 l(x-e<X _<x+e)dK(X) s s Now, i t follows from the d e f i n i t i o n of B g(a) and the c a l c u l a t i o n i n ( i i ) that (4) |B (a) - I (a-X )1(X >a>Xr ) 1 ( S <t) | < n>l b n b-~ s-' n n ~n 1 (x-e<Xg_<x+e)dK(X) And f i n a l l y , the d e f i n i t i o n of C g(x) and ( i i i ) y i e l d : l(x-e<X <x+e)dK(X) s s (5) |C (a)- I (X -a)l(D-([a,b]))lCX =a<X )|< s<t S s s C I n e q u a l i t i e s (1) through (5) show that: l i m ^ sup j E t ( a , b ) | ^ 4 I | AX g| l(X g_=x) + 3 I |AXJI(XC=X) €•^-0 [a,b3=[x-e,x+e] s<t + l i m ^X.(x-£))1(VX+£)d(A+C)s(x> s<t s' s where the right-hand side i s zero by choice of x and dominated convergene- convergence. This completes the proof of the lemma. QED. As t h i s lemma indicates, the convergence of E t(a,b) as a t x and b 4- x f a i l s around points i n the range of d i s -c o n t i n u i t i e s of X ; t h i s does not hinder our chances of generalizing (III.2.1), so long as we are able to show that the convergence of (b-a)N +(0,t,a,b) to L X i s uniform over a l l small i n t e r v a l s - 29 -[a,b] not containing these points. Before proving t h i s g e n e r a l i z a t i o n , we present a lemma which 2 w i l l enable us to remove an H hypothesis on X : (III.3.2) LEMMA: Let X be a semi-martingale such that K(X) t i s a.s. f i n i t e f o r each t . Define the process Y = Arctg(X t) . Then Y i s a semi-martingale l o c a l l y i n H P f o r any X X p > 1 . Let L (X) and L (Y) denote r e s p e c t i v e l y the l o c a l time processes (at x) of X and Y . Then f o r each t , L X(X) = ( l + x 2 ) L ^ r C t s ( x ) ( Y ) f o r a l l x , a.s. 2 Proof: Since Arctg(«) i s a C function, i t follows from Ito's lemma that Y i s a semi-martingale, and moreover, K ( Y ) t i s f i n i t e a.s. for each, t because the d e r i v a t i v e of Arctg(-) i s bounded by 1 . The function Arctg(«) i s i t s e l f bounded by ir/2 , and hence, Y i s l o c a l l y i n H P f o r any p > 1 . Let A(x,e) = Arctg(x+e) - Arctg(x) . Ito's lemma y i e l d s that d<Y C, Y C >g = (l4X 2_)~ 2d<X C,X C> s ; therefore, using the f a c t that c c Arctg(«) i s one-to-one and the f a c t that d<Y , Y > g does not charge the sets {s<t : X g / Xg_} , one obtains that f o r each x e TR and e > 0 , (1+x 2) 2Ce/A(x,e))e 1 t l(x<X <x+e.)d<XC,Xc> s s ^ (A(x,e)) 1 0 t 1 (Arctg (x) <Y g<Arctg (x+e ) ) d<YC, Y C> g J 0 t > (l+(x+ e) 2) 2(£/A(x,e)) e 1 J 0 l(x<X <x+e)d<XC,XC> s s - 30 -Therefore, noting that e/A(x,e) (1+x ) as e -* 0 and using lemma (1.2.5), taking l i m i t s i n the l a s t equation y i e l d s the desired r e s u l t . QED. We are now ready to extend the r e s u l t of (III.2.1) to semi-martingales with jumps. In the following pages, the Lebesgue measure of an i n t e r v a l I i s denoted by | l | . For any 'V > 0 and x e R , we define S(x,v) to be the c o l l e c t i o n of a l l i n t e r v a l s I such that x e I and | l | <v . (III.3.3) THEOREM: Let X be a semi-martingale such that K(X) f c i s a.s. f i n i t e f o r each t . Then with p r o b a b i l i t y one, (a) l i m sup | |l|N +(0,t,I) - j L x | = 0 \H-0 IeS(x,v) t<B f o r any B > 0 , at a l l but countably many x e ]R . (b) l i m sup |Kfc(X,Q) - (<X C,X°> ^ <X°,X C> )| = 0 v+0 Qeir(v) s t s t<B for any B > 0 . c c In p a r t i c u l a r , f o r a.a. sample paths, <X ,X > i s the arc length of X . Note: An argument using the uniform continuity of L on [0,B] and the monotonicity of N +(0,«,I) shows that i n order to prove (a) , i t s u f f i c e s to show that f o r each t , (a') lim sup | |l|N+(0,t,I) - y L X | = 0 v*0 IeS(x,v) C - 31 -at a l l but countably many x e R , a.s. In addition, by lemma (III.3.2) and since the number of upcrossings of any i n t e r v a l Ca,b] by X i s equal to the number of upcrossings of [Arctg(a), Arctg(b)3 by Y=Arctg(X) , we can assume that X 2 i s l o c a l l y i n H . This assumption only serves i n proving (a) ; the proof of (b) uses the r e s u l t of (a) and does not require that X be 2 l o c a l l y i n H any further. R e c a l l from the proof of (III.2.1) . that i n order to prove (b), I t s u f f i c e s to assume that s = 0 and to show that f o r each t : (b') l i m sup |(K^(X,Q) - <XC,XC> | = 0 a.s. QeTT(v) Proof: F i x t > 0 and l e t 0 < 6 < h Define the sets D (co) = {X (cb), X _(.co) : s<t, AX (co) 4 0}' and D„(co) = {x : LX(<o) + LX~(co)} By lemma (II.2.1)(a) and lemma (III.3.1)(b), and, since we can 2 assume that X i s l o c a l l y i n H , by lemma (11.2,2), we can pick t co out of a n u l l set so that l i m /l(x-e<X _<x+e) | d(V+J) |=0 f o r e+0 0 S S' a l l x | D 1 u D 2, lim sup |-E (a,b)| = 0 f o r a l l e->-0 [a,b]cCx-e ,x+e] x \ D and max | J l (u+([x,x+k~ 6])) 1(X _>x)dM |<k~1/'2 for 1 xeR k0 S s large enough k . Pick x | u . By r i g h t - c o n t i n u i t y of l o c a l time and the choice of co above, there e x i s t s e = e (co) , N = N(co) and B = B(co) >T such that the following hold: - 32 -(1) 1(x-e<X g_<x+e)|d(V+J) | < <5 (2) sup |E tCa,b)|<6 [a,b]c[x - e,x+ e ] (3) max | l(u +(Cx,x+k 6 ] ) ) 1 ( X _>x)dM | < k 1 / 2 for a l l k > N X £ R k O (4) I"J-^ t ~ " i ^ t ^ < ^ whenever | y-x | < £ (5) sup |X (o))|+KN (6) s<t t dV I + K(XV + sup L X < B s t t (7) max(N 1 / 2,(1+2N 1)6-1) < 6/B (8) N 6 < e/8 (9) (k-1) 7 - (k-1) 6 > k 6 f o r k > N The f i r s t step i s the set-up of the g r i d approximation: pick —fi v < (N+1) and I e S(x,v) . Then by (8) and the d e f i n i t i o n of S(x,y) , we have that I c [x-£/8, x + £ / 8 ] . Denote the i n t e r v a l I by [a,b] . Now, pick m = m(I) > N + 1 such that (m+1) 6 < | l | < m~6 . I f |x| > N - l , then L X = N+(0,t,I) = 0 and the theorem holds t r i v i a l l y . On the other hand, i f |x| < N - l , def ine: a^ = a^CO = max(y £ : y < a) for a l l k > N , and set Xm-1 = C am-1' a m - l + ( m - 1 ) " ] and Xm+2 " [ am+2 + ( n H- 2 ) _ 7> a m f 2+(m+2)- 7+(m+2)- 6] - 33 -As i n the proof of (II.2.4), i t i s an easy exercise to v e r i f y that m+2 m-1 Therefore, using the monotonicity of N +(0,t,») and the choice of m , one finds that: (10) sup | |l|N+(0,t,I) -IeS(x,v) t < max{ sup |(m+l) 6 N + ( 0 , t , I .) - ^ L X | IeS(x,v) m - 1 2 t' sup |m V(0,t,I° ) - ^ L X|} l£S(x,v) m + 2 2 t To complete the proof of (a'), we w i l l show that the choice of e and N makes the right-hand side of (10) small. We w i l l also prepare the proof of ( V ) by bounding (10) independently of x . Using (6) and the f a c t (which follows from (7)) that |(m+l) 6(m-l)~ 6 - l | < 6/B < 6 for a l l m > N + 1 , one obtains (11) sup | (M+l)"V(0,t,I ) - |-L X| IeS(x,v) m - 1 1 fc IeS( sup | (m-1) V ( 0 , t , I ) - j-L + (1+6) sup \\*^X-\£ \ + ( V B ) | f L X IeS(x,v) Now, i t follows from (8) and the d e f i n i t i o n of a, that k ! a m - l " x l < e / 2 f o r a 1 1 m - N + 1 ' Therefore, by (4), 1 am—1 1 x sup | — L ., ~ " 9 L ^ I < 5 because the sup i s taken over m>N+l IeS(x,v) 1 t 1 t - 34 -Moreover, using (II.1.2) and (8), (12) sup | (m-l) V ( 0 , t , I n ) - ^ L * IeS(x.v) m _ 1 2 t t -6 < N + sup 1 ( a m - l < V " V l + (m-l)6)U(V+J)! IeS(x,v)£ t + sup I l ( u + ( I 1 ) ) 1 ( X _ > am )dM x o / \ J m-l s m-l IeS(x,v) ^ + sup |E (a a m + (m-l) 6 ) | IeS(x,v) Observe that f o r a l l m> N + 1, the: i n t e r v a l s I , are contained m-l i n [x-e, x+e] (t h i s follows from (8) and a c a l c u l a t i o n above). Therefore, using (1), (2), (3) and (7), we see that the right-hand l T x aea D y to . iNocxng mat | equation (11) now shows that: 1 x side of equation (12) i s bounded by 46 . Noting that — L < B , sup | (m+l)" 6N +(0,t,I ) - ^ L X | < H5 l€S(x,v) m 1 t An i d e n t i c a l argument show that: sup |m"V(0,t,I°._) - ^ L X | < 116 IeS(x,v) m + 2 2 t and the conclusion of (a') now follows from (10) and the fac t that the sets and are countable. The proof of (a ) i s complete, To bound the right-hand side of (.12) independently of x , r e c a l l that i n the l a s t l i n e s of the proof of lemma (III.3.1)(b) i t i s shown that sup |E t(a,b)| < 8'K(X) . [a,b]c[ x-e,x+e] - 35 -Therefore, from equations (6) and (12), using the l a s t remark, we get that f o r a l l x , sup ] (m-l)~ 6N +(0,t,I -) - y L X | < cB IeS(x,v) m 1 t where c i s a constant independent of x . Again, an i d e n t i c a l argument shows a s i m i l a r bound for the upcrossings of , and m+z so, i n view of (10), for a l l x , sup | | l['N +(0 ,t ,T) - j L X | IeS(x,v) < c B. To prove (b'), f i r s t note that i t follows from part (a) that: (13) l i m \H-0 3R sup | |l|N(0,t,I) - L X|dx = 0 IeS(x.v) Z because the integrand has compact support, i s bounded independently of x and converges to zero at a l l but countably many x e K. . Consider the interval-valued functions f~(x) = [x?,x?,.,)l - n (x), Q I l + l r Q Q s Cx.,x i + 1) and r e c a l l that from the d e f i n i t i o n of KQ(X,Q) and the density of occupation time formula we have: » |KQ(X,Q) - <X C,X°>:J < sup I |f (x)|N(0,t,f (x)) - L X f d x i Q6TT(V) Now, i f Qeir(v) , then f q ( x ) € 5(x,v) for each x , and therefore, (13) implies that the right-hand side of the above in e q u a l i t y goes to zero with v . This f i n i s h e s the proof of the theorem. QED. The argument of the proof of (a) above can be used to obtain a stronger version of theorem (II.2.4); incorporating i n the proof of (11.24) the set-up of i n t e r v a l s I^,„ c I c I c [x-e,x+e] m+z m-l for each x and using the uniform convergence of the Lebesgue-S t i e l t j e s i n t e g r a l s of (II.2.1) shows that i f X i s a continuous - 36 -semi-martingale whose l o c a l time i s bicontinuous, then a.s., for any B > 0 , lim sup sup | |l|N +(0,t,I) - y L X ] = 0 v-K) xeIR IeS(x,v) t t<B In t h i s new conclusion, i t i s cl e a r that the assumption of b i c o n t i -nuity of l o c a l time i s a necessary one: consider r e f l e c t i n g Brownian motion, l e t L t be i t s l o c a l time process and consider the i n t e r v a l s 1 (x) = Cx-- , x+-] . Then, N +(0,t,I (0)) = 0 for a l l n,t, n n n n and (2/n)N +(0,t,I (0)) - (1/2)L^ doesn't converge to zero, so n t unless l o c a l time i s continuous i n the state-space v a r i a b l e , the I I 4- X I^lN ( 0 , t , I n ) to L f c , along an a r b i t r a r y sequence of i n t e r v a l s I containing the point x f a i l s . S t i l l , f o r r e f l e c t i n g (or skew) Brownian motion, an almost sure point-wise r e s u l t e x i s t s for the convergence of + x 2 e N (0,t,x,x+e) to . Consider a semi-martingale X such that K(X)^_ i s a.s. f i n i t e f o r each t whose l o c a l time process has fixed d i s c o n t i n u i t i e s i n the state-space v a r i a b l e . Then, because the jumps, of L^ . aren't random, (see [8] for the skew Brownian motion case) we can use the point-wise r e s u l t of E l Karoui i n [4] to c o l l e c t n u l l sets where the convergence f a i l s at a l l the (countably many) jump points before carrying on with the proof of (III.3.3)(a). This y i e l d s that f o r such processes, outside a singl e n u l l set, e 1^(0,t,x,x+e) converges to (1/2)L X for a l l x . This i s as far as the techniques used i n t h i s study go, however, and the question of almost sure point-wise convergence - 37 -(when the underlying process i s a continuous semi-martingale) remains open. There i s one l a s t remark to make on theorem (III.3.3). Part c c (b) , besides shedding new l i g h t on the nature of <X ,X > , can be used to i n t e r p r e t some stochastic i n t e g r a l s as uniform l i m i t s of Lebesgue i n t e g r a l s . Let X be such that K(X)^ i s a.s. f i n i t e for each t , and for the sake of argument, l e t g be a continuous function. Then, the dominated convergence argument i n the proof t of (III.3.3)(b) y i e l d s that / g(X )d<XC,Xc> = /g(x)L Xdx = 0 S S R lim sup / g ( x ) ( x ^ + 1 - x 1 ) N ( 0 , t , x ^ , x ^ + 1 ) l Q Q (x)dx . Therefore, v+0 Qen-(v) R. [x^ » x ^ + 1 ) i f f e C^(R) and F' = f , an a p p l i c a t i o n of Ito's lemma gives that / f (X )dX c = F ( X J - F ( X ) - i l i m sup 'Jf' (x) (xj. -x?) 0 S S Z C 1 v+0 Qp(v) K 1 + 1 1 N ( 0 , t , x ^ , x ^ + 1 ) l ,Q Q (x)dx and so, (III.3.3)(b) can be used to rx.,x 1 + 1) recover some stochastic i n t e g r a l s through the geometry of the paths. - 38 -BIBLIOGRAPHY [1] Chacon, R.V. et a l . Generalised arc length for Brownian motion and levy processes, Z.W. Gebiete, Vol. 57 (1981), pp. 197-211. ~-[2] Doob, Stochastic processes, Wiley (1953) [3] E l Karoai, N. Sur l e s mont£es des semi-martingales (CAS continu ), Asterisque, Vol. 52-53 (1977), pp. 63-72. [4] E l Karoui, N. Sur l e s mont£es des semi-martingales (CAS discontinu ), Asterisque, Vol. 52-53 (1977, pp. 73-88. [5] Meyer, P.-A. Un cours sur l e s integrales stochastiques, Lecture notes i n math, Vol. 5111, Springer-Verlag (1976). [6] Monroe, I. The quadratic v a r i a t i o n of martingales: A counter-example, Annals of P r o b a b i l i t y , Vol. 4 (1977), pp. 133-138. [7 3 Taylor, S.J. Exact asymptotic estimates of Brownian path v a r i a t i o n , Duke Math Journal, V o l. 39 (1972), pp. 219-241. [8] Walsh, J. A d i f f u s i o n with a discontinuous l o c a l time, Asterisque, Vol. 52-53 (1977), pp. 37-46. [9] Yor, M. Sur l a con t i n u i t y des temps locaux associ£s A certaines semi-martingales, Asterisque, Vol. 52-53 (1977), pp. 23-36.
- Library Home /
- Search Collections /
- Open Collections /
- Browse Collections /
- UBC Theses and Dissertations /
- On the quadratic variation of semi-martingales
Open Collections
UBC Theses and Dissertations
Featured Collection
UBC Theses and Dissertations
On the quadratic variation of semi-martingales Lemieux, Marc 1983
pdf
Page Metadata
Item Metadata
Title | On the quadratic variation of semi-martingales |
Creator |
Lemieux, Marc |
Date Issued | 1983 |
Description | Let X be a semi-martingale. Techniques of [1] and El Karoui (in [3] and [4]) are used to study the convergence of 2∊ times the number of upcrossings of [x, x+∊] by X to its local time at x. If X is continuous and if there exists a bicontinuous version of its local time process, then off a single null set, the convergence is shown to be uniform in x (and time). If X is such that the sum of the absolute value of its jumps over any finite time interval is almost surely finite, then, off a single null set, the convergence holds at all but countably many x. A notion of generalized arc length, is introduced, in the spirit of the quadratic arc length of [1], and the last result above is used to show that |
Genre |
Thesis/Dissertation |
Type |
Text |
Language | eng |
Date Available | 2010-04-20 |
Provider | Vancouver : University of British Columbia Library |
Rights | For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use. |
DOI | 10.14288/1.0080232 |
URI | http://hdl.handle.net/2429/23964 |
Degree |
Master of Science - MSc |
Program |
Mathematics |
Affiliation |
Science, Faculty of Mathematics, Department of |
Degree Grantor | University of British Columbia |
Campus |
UBCV |
Scholarly Level | Graduate |
AggregatedSourceRepository | DSpace |
Download
- Media
- 831-UBC_1983_A6_7 L45.pdf [ 2.21MB ]
- Metadata
- JSON: 831-1.0080232.json
- JSON-LD: 831-1.0080232-ld.json
- RDF/XML (Pretty): 831-1.0080232-rdf.xml
- RDF/JSON: 831-1.0080232-rdf.json
- Turtle: 831-1.0080232-turtle.txt
- N-Triples: 831-1.0080232-rdf-ntriples.txt
- Original Record: 831-1.0080232-source.json
- Full Text
- 831-1.0080232-fulltext.txt
- Citation
- 831-1.0080232.ris
Full Text
Cite
Citation Scheme:
Usage Statistics
Share
Embed
Customize your widget with the following options, then copy and paste the code below into the HTML
of your page to embed this item in your website.
<div id="ubcOpenCollectionsWidgetDisplay">
<script id="ubcOpenCollectionsWidget"
src="{[{embed.src}]}"
data-item="{[{embed.item}]}"
data-collection="{[{embed.collection}]}"
data-metadata="{[{embed.showMetadata}]}"
data-width="{[{embed.width}]}"
async >
</script>
</div>
Our image viewer uses the IIIF 2.0 standard.
To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0080232/manifest