Variance and Covariance Processes

Similar documents
The shortest path between two truths in the real domain passes through the complex domain. J. Hadamard

On Control Problem Described by Infinite System of First-Order Differential Equations

Lecture-V Stochastic Processes and the Basic Term-Structure Equation 1 Stochastic Processes Any variable whose value changes over time in an uncertain

Deviation probability bounds for fractional martingales and related remarks

7 Wave Equation in Higher Dimensions

General Non-Arbitrage Model. I. Partial Differential Equation for Pricing A. Traded Underlying Security

Extremal problems for t-partite and t-colorable hypergraphs

MEEN 617 Handout #11 MODAL ANALYSIS OF MDOF Systems with VISCOUS DAMPING

336 ERIDANI kfk Lp = sup jf(y) ; f () jj j p p whee he supemum is aken ove all open balls = (a ) inr n, jj is he Lebesgue measue of in R n, () =(), f

An Introduction to Malliavin calculus and its applications

[ ] 0. = (2) = a q dimensional vector of observable instrumental variables that are in the information set m constituents of u

Lecture 28: Convergence of Random Variables and Related Theorems

Sections 3.1 and 3.4 Exponential Functions (Growth and Decay)

arxiv: v1 [math.co] 4 Apr 2019

On The Estimation of Two Missing Values in Randomized Complete Block Designs

Combinatorial Approach to M/M/1 Queues. Using Hypergeometric Functions

4 Sequences of measurable functions

@FMI c Kyung Moon Sa Co.

Today - Lecture 13. Today s lecture continue with rotations, torque, Note that chapters 11, 12, 13 all involve rotations

K. G. Malyutin, T. I. Malyutina, I. I. Kozlova ON SUBHARMONIC FUNCTIONS IN THE HALF-PLANE OF INFINITE ORDER WITH RADIALLY DISTRIBUTED MEASURE

THE MODULAR INEQUALITIES FOR A CLASS OF CONVOLUTION OPERATORS ON MONOTONE FUNCTIONS

Reinforcement learning

A characterization of reciprocal processes via an integration by parts formula on the path space

Properties of the interface of the symbiotic branching model

Example on p. 157

Quantum Algorithms for Matrix Products over Semirings

A STOCHASTIC MODELING FOR THE UNSTABLE FINANCIAL MARKETS

Research Article A Note on Multiplication and Composition Operators in Lorentz Spaces

SOME MORE APPLICATIONS OF THE HAHN-BANACH THEOREM

Hamilton- J acobi Equation: Explicit Formulas In this lecture we try to apply the method of characteristics to the Hamilton-Jacobi equation: u t

Solution to HW 3, Ma 1a Fall 2016

STUDY OF THE STRESS-STRENGTH RELIABILITY AMONG THE PARAMETERS OF GENERALIZED INVERSE WEIBULL DISTRIBUTION

On the local convexity of the implied volatility curve in uncorrelated stochastic volatility models

POSITIVE SOLUTIONS WITH SPECIFIC ASYMPTOTIC BEHAVIOR FOR A POLYHARMONIC PROBLEM ON R n. Abdelwaheb Dhifli

MATH 4330/5330, Fourier Analysis Section 6, Proof of Fourier s Theorem for Pointwise Convergence

A characterization of reciprocal processes via an integration by parts formula on the path space

Representing Knowledge. CS 188: Artificial Intelligence Fall Properties of BNs. Independence? Reachability (the Bayes Ball) Example

r P + '% 2 r v(r) End pressures P 1 (high) and P 2 (low) P 1 , which must be independent of z, so # dz dz = P 2 " P 1 = " #P L L,

Notes for Lecture 17-18

Two Coupled Oscillators / Normal Modes

T L. t=1. Proof of Lemma 1. Using the marginal cost accounting in Equation(4) and standard arguments. t )+Π RB. t )+K 1(Q RB

PHYS PRACTICE EXAM 2

Bernoulli numbers. Francesco Chiatti, Matteo Pintonello. December 5, 2016

Solutions from Chapter 9.1 and 9.2

Math 2142 Exam 1 Review Problems. x 2 + f (0) 3! for the 3rd Taylor polynomial at x = 0. To calculate the various quantities:

GRADIENT ESTIMATES, POINCARÉ INEQUALITIES, DE GIORGI PROPERTY AND THEIR CONSEQUENCES

EXERCISES FOR SECTION 1.5

Degree of Approximation of a Class of Function by (C, 1) (E, q) Means of Fourier Series

CS 188: Artificial Intelligence Fall Probabilistic Models

Representation of Stochastic Process by Means of Stochastic Integrals

A proof of Ito's formula using a di Title formula. Author(s) Fujita, Takahiko; Kawanishi, Yasuhi. Studia scientiarum mathematicarum H Citation

Secure Frameproof Codes Through Biclique Covers

Oscillation of an Euler Cauchy Dynamic Equation S. Huff, G. Olumolode, N. Pennington, and A. Peterson

ME 141. Engineering Mechanics

BMOA estimates and radial growth of B φ functions

k. s k=1 Part of the significance of the Riemann zeta-function stems from Theorem 9.2. If s > 1 then 1 p s

Problem Set #10 Math 471 Real Analysis Assignment: Chapter 8 #2, 3, 6, 8

MODULE 3 FUNCTION OF A RANDOM VARIABLE AND ITS DISTRIBUTION LECTURES PROBABILITY DISTRIBUTION OF A FUNCTION OF A RANDOM VARIABLE

Lecture 18: Kinetics of Phase Growth in a Two-component System: general kinetics analysis based on the dilute-solution approximation

Low-complexity Algorithms for MIMO Multiplexing Systems

Lecture 17: Kinetics of Phase Growth in a Two-component System:

An Automatic Door Sensor Using Image Processing

Statistical inference versus mean field limit for Hawkes processes

arxiv: v1 [math.fa] 20 Dec 2018

KINEMATICS OF RIGID BODIES

The Asymptotic Behavior of Nonoscillatory Solutions of Some Nonlinear Dynamic Equations on Time Scales

dt = C exp (3 ln t 4 ). t 4 W = C exp ( ln(4 t) 3) = C(4 t) 3.

Control Volume Derivation

Computer Propagation Analysis Tools

Homework sheet Exercises done during the lecture of March 12, 2014

On a Fractional Stochastic Landau-Ginzburg Equation

A note on characterization related to distributional properties of random translation, contraction and dilation of generalized order statistics

ENGI 4430 Advanced Calculus for Engineering Faculty of Engineering and Applied Science Problem Set 9 Solutions [Theorems of Gauss and Stokes]

Lecture Notes 2. The Hilbert Space Approach to Time Series

The Production of Polarization

MATHEMATICAL FOUNDATIONS FOR APPROXIMATING PARTICLE BEHAVIOUR AT RADIUS OF THE PLANCK LENGTH

Convergence of the Neumann series in higher norms

Finish reading Chapter 2 of Spivak, rereading earlier sections as necessary. handout and fill in some missing details!

The sudden release of a large amount of energy E into a background fluid of density

Existence Theory of Second Order Random Differential Equations

Stanford University CS259Q: Quantum Computing Handout 8 Luca Trevisan October 18, 2012

Sharif University of Technology - CEDRA By: Professor Ali Meghdari

arxiv: v1 [math.fa] 9 Dec 2018

Order statistics and concentration of l r norms for log-concave vectors

23.5. Half-Range Series. Introduction. Prerequisites. Learning Outcomes

t is a basis for the solution space to this system, then the matrix having these solutions as columns, t x 1 t, x 2 t,... x n t x 2 t...

Ann. Funct. Anal. 2 (2011), no. 2, A nnals of F unctional A nalysis ISSN: (electronic) URL:

11!Hí MATHEMATICS : ERDŐS AND ULAM PROC. N. A. S. of decomposiion, properly speaking) conradics he possibiliy of defining a counably addiive real-valu

156 There are 9 books stacked on a shelf. The thickness of each book is either 1 inch or 2

Finite-Sample Effects on the Standardized Returns of the Tokyo Stock Exchange

f(x) dx with An integral having either an infinite limit of integration or an unbounded integrand is called improper. Here are two examples dx x x 2

Chapter 7: Solving Trig Equations

Backward stochastic dynamics on a filtered probability space

3.1 Random variables

Risk tolerance and optimal portfolio choice

ODEs II, Lecture 1: Homogeneous Linear Systems - I. Mike Raugh 1. March 8, 2004

ON 3-DIMENSIONAL CONTACT METRIC MANIFOLDS

CONTRIBUTION TO IMPULSIVE EQUATIONS

Matrix Versions of Some Refinements of the Arithmetic-Geometric Mean Inequality

Chapter 2. First Order Scalar Equations

Transcription:

Vaiance and Covaiance Pocesses Pakash Balachandan Depamen of Mahemaics Duke Univesiy May 26, 2008 These noes ae based on Due s Sochasic Calculus, Revuz and Yo s Coninuous Maingales and Bownian Moion, Kaazas and Sheve s Bownian Moion and Sochasic Calculus, and Kuo s Inoducion o Sochasic Calculus 1 Moivaion In his secion, we moivae he consucion of vaiance and covaiance pocesses fo coninuous local maingales, which is cucial in he consucion of sochasic inegals w... coninuous local maingales as we shall see. In his secion, unless ohewise specified, we fix a Bownian moion B and a filaion {F } such ha: 1. Fo each, B is F -measuable. 2. Fo and s, he andom vaiable B B s is independen of he σ-field F s. Recall ha fo any Bownian moion, B immediaely implies (2) in he following = whee B is he quadaic vaiaion of B. This Definiion: Define L 2 ad (a, b Ω) o be he space of all sochasic pocesses f(, ω), a b, ω Ω such ha: 1. f(, ω) is adaped o he filaion {F }. 2. b a E f() 2 d = b a E f() 2 d B <. 1

lso ecall when consucing a heoy of inegaion w... a Bownian moion, we begin wih consucing he sochasic inegal fo f L 2 ad (a, b Ω). b a f()db Now, we wan a moe geneal fomalism of inegaing a class of pocesses w... a genealized maingale ha in he case Bownian moion will educe o he above. Definiion: Le G be a igh-coninuous filaion. We define L denoe he collecion of all joinly measuable sochasic pocesses X(, ω) such ha: 1. X is adaped w... G. 2. lmos all sample pahs of X ae lef coninuous. Fuhemoe, we define P o be he smalles σ-field of subses of a, b Ω wih espec o which all he sochasic pocesses in L ae measuable. sochasic pocesses Y (, ω) ha is P measuable is said o be pedicible. The moivaion fo he definiion of a pedicable pocess comes fom he following agumen: If Y is a pedicable pocess, hen almos all is values a ime can be deemined wih ceainy wih he infomaion available sicly befoe ime, since lef coninuiy of he pocess Y implies ha fo almos evey ω Ω and any sequence n as n : lim n Y n (ω) = Y (ω). Now, we have he following heoem which we shall pove a vesion of in he nex secion fo coninuous local mainagles: Theoem 1 (Doob-Meye) Le M, a b be a igh coninuous, squae inegable maingale wih lef hand limis. Then, hee exiss a unique decomposiion (M ) 2 = L +, a b whee L is a igh-coninuous maingale wih lef-hand limis, and is a pedicable, igh coninuous, inceasing pocess such ha a 0 and E < fo all a b. The above heoem ceainly applies o he squae inegable pocess B. 2

Claim 1 In he case M = B in Doob-Meye, = B =. Poof of Claim 1: WLOG, we may ake a = 0 and b = 0. Define P 0 s 0 : = (B ) 2. Then, fo E(B ) 2 F s = E(B B s + B s ) 2 F s = E(B B s ) 2 + 2B s (B B s ) + (B s ) 2 F s = E(B B s ) 2 + 2B s EB B s + (B s ) 2 = s + (B s ) 2 EP F s = E(B ) 2 F s = (B s ) 2 s = P s. Thus, P = (B ) 2 is a maingale, so ha (B ) 2 = P +. Clealy, saisfies all he condiions ha mus saisfy in Doob-Meye, so ha by uniqueness of, = = B. So, anohe way of viewing he inegal w... he maingale M w... he filaion G is he following: Fis, we look fo he unique pocess (guaaneed by Doob-Meye) M such ha L = (M ) 2 M is a maingale. Then, we make he Definiion: Define L 2 ped (a, b M Ω) o be he space of all sochasic pocesses f(, ω), a b, ω Ω such ha: 1. f(, ω) is pedicable w... {G }. 2. b a E f() 2 d M <. Then, we poceed o consuc he inegal b a f()dm fo f L 2 ped (a, b M Ω). I s clea ha in he case M = B and G = F ha he above fomulaion coincides wih he oiginal consucion of he sochasic inegal w... B eviewed a he beginning of his secion. Fo igh coninuous, squae inegable maingales M wih lef hand limis, a leas, his pocess woks. In he case whee M is a coninuous local maingale, we do he same hing. Howeve, i s no immediaely clea: 1. ha we have a vesion of Doob-Meye fo coninuous local maingales. 2. how he consucion of he inegal is affeced by he sopping imes T n ha educe M, if a all. In he nex secion, we deal wih he fis poblem. Then, we poceed o emedy he second. 3

2 Vaiance and Covaiance Pocesses We ake L and P as defined in secion 1. Theoem 2 If X is a coninuous local maingale, hen we define he vaiance pocess X o be he unique coninuous pedicable inceasing pocesses ha has 0 0 and makes X 2 a local maingale. Definiion: If X and Y ae wo coninuous local maingales, we le X, Y = 1 4 ( X + Y X Y ). We call X, Y he covaiance of X and Y. Based on he discussion in he fis secion, i s clea why we e ineesed in vaiance pocesses. I is convenien o define covaiance pocesses since hey ae vey useful and have quie nice popeies, such as: Theoem 3, is a symmeic bilinea fom on he class of coninuous local mainagles. We migh pove i his ime aound. If no, hopefully nex ime. Two quesions I m sill pondeing is 1. Can you un his ino an inne poduc? 2. If so, how you can chaaceize he class of pocesses ha is he compleion of his space? The poof of heoem 2 is long, bu i is insucive o go hough i, since i develops echniques ha will be useful lae. In ode o poceed, ecall ha any pedicable discee ime maingale is consan why?. Thee is a esul analogous o his in coninuous ime, and we use i o pove he uniqueness saemen in heoem 2: Theoem 4 ny coninuous local maingale X ha is pedicable and locally of bounded vaiaion is consan (in ime). Poof: By subacing X 0, WM ha X 0 0. Thus, we wish o show ha X 0 fo all > 0 almos suely. Le V (ω) = sup π Π T π (ω) be he vaiaion of X s (ω) on 0,, whee Π denoes he se of all (finie) paiions of 0,, π = {0 = 0 < 1 < < N = }, and whee fo a given paiion of his so, N T π (ω) = X m (ω) X m 1 (ω). 4

Lemma 1 Fo almos all ω Ω, V (ω) is coninuous Poof of Lemma: Fis, noice ha fo any ω Ω, V (ω) is inceasing: Fo s <, 0, s 0,, so ha any finie paiion π = {0 = 0 < 1 < < N = s} of 0, s gives a finie paiion π = {0 = 1 < < N = s < N+1 = } of 0,. Thus, fo any finie paiion π of 0, s, T π (ω) T π (ω), whee π is a finie paiion of 0,, so ha Since ω was abiay, his is ue fo all ω Ω. T π (ω) T π (ω) sup π Π T π (ω) = V (ω) V s (ω) = sup π Π s T π (ω) sup π Π T π (ω) = V (ω). Thus, o show ha V is coninuous a.s., i suffices o show ha fo almos all ω Ω, V (ω) has no disconinuiies (of he fis kind). Claim 2 Fo any ω Ω, V u (ω) = V s (ω) + V u s (ω) whee V u s (ω) is he vaiaion of X (ω) on s, u. Poof of Claim: Take any wo paiions {s = 0 < 1 < < N = u}, {0 = N < N 1 < < 0 = s}. Then: 0 m= N +1 X m (ω) X m 1 (ω) + N X m (ω) X m 1 (ω) V s (ω) + V u s (ω). Now, he LHS is T π (ω) fo π = {0 = N < < N = u}. Thus, V u (ω) V s (ω) + V u s (ω). Fo he ohe inequaliy, noe ha {0 = N Thus: < < 0 = s < < N = u} is a paiion of 0, u. V u (ω) N m= N +1 X m (ω) X m 1 (ω) = 0 m= N +1 X m (ω) X m 1 (ω) + N X m (ω) X m 1 (ω). Now, fixing one of he paiions on he RHS, we may ake he supemum of he emaining, and hen poceed o ake he supemum of he final em. Thus: V u (ω) V s (ω) + V u s (ω), so ha V u (ω) = V s (ω) + V u s (ω). Now, by hypohesis, X s is of locally bounded vaiaion. So, hee exiss a sequence of sopping imes T n a.s. such ha Xs Tn (ω) is of bounded vaiaion in ime. Le = {ω Ω : T n (ω) }. By definiion, P = 1. Now, le ω be fixed, and suppose ha s V s (ω) has a disconinuiy a. Choosing n lage enough so ha T n (ω) >, hee exiss s 0 < u 0 such ha X s (ω) is of bounded vaiaion on s 0, u 0. 5

Since s V s (ω) has a disconinuiy a, hee exiss ɛ > 0 such ha fo evey δ > 0, u s < δ implies V u (ω) V s (ω) > 3ɛ whee s < < u. By Claim 2 hen, fo evey δ > 0, u s < δ implies V u s (ω) > 3ɛ whee s < < u. Pick δ > 0 so ha if s < δ hen X s X < ɛ (using unifom coninuiy of X s (ω) on s 0, u 0 ). ssuming s n and u n have been defined, pick a paiion of s n, u n no conaining wih mesh less han δ and vaiaion geae han 2ɛ (his is possible since fo evey δ > 0, u s < δ implies V u s (ω) > 3ɛ whee s < < u). Le s n+1 be he lages poin in he paiion less han, and u n+1 be he smalles poin in he paiion lage han. Then u n+1 s n+1 < δ X sn+1 (ω) X un+1 (ω) < ɛ. Thus: N X m (ω) X m 1 (ω) > 2ɛ N, m s n+1,u n+1 X m (ω) X m 1 (ω) > 2ɛ X un+1 (ω) X sn+1 (ω) > ɛ. By omiing he poins s n+1 and u n+1 fom he paiion, we obain a paiion fo s n, u n s n+1, u n+1. Thus, afe aking supemums: V sn,u n s n+1,u n+1 (ω) = V un s n (ω) V u n+1 s n+1 (ω) > ɛ. Thus, V u 0 s 0 (ω) > Mɛ fo abiaily lage inege values M. Thus, i mus be infiniy, conadicing ha X s (ω) has bounded vaiaion on s 0, u 0. Thus, V (ω) mus be coninuous fo evey ω Ω, since ω was abiay. Now, we needed Lemma 1 in ode o guaanee ha he funcions ae sopping imes (why?). Lemma 2 {S n } educe X. Poof of Lemma 2: We poved las ime ha S n (ω) = inf{s : V s (ω) n} If X is a coninuous local maingale, we can always ake he sequence which educes X o be T n = inf{ : X > n} o any ohe sequence T n T n ha has T n as n. Now, suppose ha saisfies n < X = X X 0 V. Then, V > n, so ha { : X > n} { : V n} S n = inf{ : V n} inf{ : X n} = T n. 6

Ceainly, V s n + 1 n V s n so ha { : V s n + 1} { : V s n} S n = inf{{ : V s n} { : V s n + 1} = S n+1. Finally, since V is coninuous a.s., i s clea ha lim n S n (ω) = almos suely. Thus, S n educe X. Now, fix some n > 0. Then, S n implies X n. By Lemma 2, M = X Sn maingale. is a bounded Now, if s < : E(M M s ) 2 F s = EM 2 F s 2M s EM F s + M 2 s = EM 2 F s M 2 s = EM 2 M 2 s F s. (we efe o his elaionship as ohogonaliy of maingale incemens). If 0 = 0 < 1 < < N = is a paiion of 0,, we have: N N EM 2 = E M 2 m M 2 m 1 = E (M m M m 1 ) 2 E ne sup M m M m 1 m V Sn sup M m M m 1 m Taking a sequence of paiions n = {0 = n 0 < n 1 < < n k(n) = } in which he mesh n = sup m n m n m 1 0 coninuiy of sample pahs imply sup m M n m M n m 1 0 a.s. Since sup m M m M m 1 2n, he bounded convegence heoem implies E sup M n m M n m 1 0. m Thus, EM 2 = 0 so ha M = 0 a.s. Le = {ω Ω : M (ω) 0}. Then, since above was abiay, P = 0 fo any, so ha P = 0. Q, 0 Thus, wih pobabiliy 1, M = 0 fo all aional. By coninuiy of sample pahs, we have ha M = 0 wih pobabiliy 1 fo all. Uniqueness in heoem 2: Suppose ha and ae wo coninuous, pedicable, inceasing pocesses ha have 0 = 0 0, and make X2, X 2 local maingales. If T n educe X 2 and T n educe X 2 i s clea ha T n T n educe X 2 (X 2 ) =, so ha is a coninuous local maingale. I s clea ha is pedicable, since each and ae pedicable. 7

Finally, is locally of bounded vaiaion. To see his, ake he sopping imes S n = T n T n. Clealy, T n T n, and he sopped pocesses T n T n T n T n ae of bounded vaiaion fo each ω, being he diffeence of wo inceasing pocesses on he andom ineval 0, T n (ω) T n(ω). Thus, by heoem 4, mus be consan, so ha since 0 = 0 = 0, = 0 fo all. Thus, = fo all. The exisence poof is a lile moe difficul, bu uses some gea analysis. Exisence in heoem 2: We poceed in seps: Sep 1: Poof of exisence in heoem 2 when X is a bounded maingale: (noe ha uniqueness follows fom he pevious agumen) Given a paiion = {0 = 0 < 1 < } wih lim n n =, le k() = sup{k : k < } be he index of he las poin befoe ; noe ha k() is no a andom vaiable, bu a numbe. Define k() Q (X) = (X k X k 1 ) 2 + k=1 ( X X k() ) 2. Lemma 3 If X is a bounded coninuous maingale, hen X 2 Q (X) is a maingale. Poof of Lemma 3: Fis, noice ha k() Q Q ( ) ( ) k(s) 2 2 ( ) ) 2 2 s = Xk X k 1 + X X k() Xk X k 1 (X s X k(s) = k=1 (X k(s)+1 X k(s) ) 2 (X s X k(s) ) 2 + k() k=k(s)+2 k=1 ( Xk X k 1 ) 2 + ( X X k() ) 2. Define u i = i fo k(s) i k() and u k()+1 =. Then, wiing Q = Q s + (Q Q s ): = EX 2 F s Q s (X) E = EX 2 F s Q s (X) E k()+1 i=k(s)+2 k()+1 i=k(s)+2 EX 2 Q (X) F s ( ) 2 Xui X Fs ui 1 E ( 2 ( ) 2 X X k(s)+1 k(s)) Fs +E X s X k(s) Fs Xu 2 i Xu 2 Fs i 1 E X 2 k(s)+1 2X k(s)+1 X k(s) + X 2 k(s) F s +E X s 2 2X s X k(s) + X 2 k(s) F s 8

= Q s (X) + E X 2 k(s)+1 F s E X 2 k(s)+1 F s + 2X s X k(s) X 2 k(s) + Xs 2 2X s X k(s) + X 2 k(s) = X 2 s Q s (X) whee in he fis equaliy, we have used he fac ha Q s (X) is F s measuable, and in he second equaliy, we have used he ohogonaliy of maingale incemens. Lemma 4 Le X be a bounded coninuous maingale. Fix > 0 and le n be a sequence of paiions 0 = n 0 < < n k n = of 0, wih mesh n = sup k n k n k 1 0. Then, Q n (X) conveges o a limi in L 2 (Ω, F, P ). Poof of Lemma 4: Fis, we begin wih some noaion. If and ae wo paiions of 0,, we le denoe he paiion obained by aking all he poins in and. Now, by lemma 3, fo fixed paiions and of 0,, we have ha fo a bounded coninuous maingale X : Y = (X 2 Q ) (X 2 Q ) = Q Q is again a bounded maingale (Since X M fo all 0 implies Q Q KM since he paions and ae fixed). Thus, again by lemma 3: Z = (Y ) 2 Q (Y ) is a maingale wih Z 0 = 0, so ha ( ) 2 EZ = 0 E Q Q = E (Y ) 2 = E Q (Y ). Now, 2a 2 + 2b 2 (a + b) 2 = (a b) 2 0 fo any eal numbes a and b, so ha (a + b) 2 2(a 2 + b 2 ) fo any eal numbes a, b. Thus: Q (Y ) = k=1 k=1 k() ( ) ( ) k() 2 2 Yk Y k 1 + Y Y k() = k=1 ( + Q Q k=1 ) ( ) 2 Q k() Q k() ( ) ( ) 2 Q k Q k Q k 1 Q k 1 k() ) ( ) 2 ( ) ( ) 2 = (Q k Q k 1 Q k Q k 1 + Q Q k() Q Q k() k() ) 2+ ( ) 2+ ( ) 2+ ( ) 2 2 (Q k Q k 1 Q k Q k 1 Q Q ( k() Q Q k() = 2 Q Q ) ( + Q Q ). 9

Puing i all ogehe hen, we have: ( ) 2 E Q ( Q = E Q (Y ) 2 Q Q ) ( + Q Q ). Thus, o show ha {Q n (X)} is Cauchy in L 2 (Ω, F, P ), and hence conveges in his space since i is complee, i is sufficien o show ha + ( 0 E Q Q ) 0. To do his, le {s k } n k=1 = and { j } =. Le s k and j such ha j s k < s k+1 j+1. Then: Q s k+1 Q s k = (X sk+1 X j ) 2 (X sk X j ) 2 = (X sk+1 X sk ) 2 + 2(X sk+1 X sk )(X sk X j ) Q = (X sk+1 X sk )(X sk+1 + X sk 2X j ) (Q ) Q (X) sup(x sk+1 + X sk 2X j(k) ) 2 k whee j(k) = sup{j : j s k }. By he Cauchy-Schwaz inequaliy: ( E Q Q ) E Q (X) 2 1 2 E sup k (X sk+1 + X sk 2X j(k) ) 4 1 2. Since he sample pahs of X ae coninuous almos suely, sup k (X sk+1 + X sk 2X j(k) ) 4 0 almos suely as + 0. Since sup k (X sk+1 + X sk 2X j(k) ) 4 (4M) 4, he bounded convegence heoem implies ha as + 0. Thus, i emains o show ha E Q (X) 2 = E Q = E sup k ( n ) 2 (X sm X sm 1 ) 2 = (X sk+1 + X sk 2X j(k) ) 4 1 2 0 Q (X) 2 is bounded. To do his, noe ha: n n 1 n (X sm X sm 1 ) 4 +2 (X sm X sm 1 ) 2 n n 1 ( ) (X sm X sm 1 ) 4 + 2 (X sm X sm 1 ) 2 Q (X) Q s m (X) n (X) 2 = E (X sm X sm 1 ) +2E 4 n 1 To bound he fis em on he RHS, noe ha X M fo all implies: n n n E (X sm X sm 1 ) 4 (2M) 2 E (X sm X sm 1 ) 2 = 4M 2 E 10 j=m+1 (X sj X sj 1 ) 2 ( ) (X sm X sm 1 ) 2 Q (X) Q s m (X) Xs 2 m Xs 2 m 1 4M 2 E X 2 4M 4

whee in he fis equaliy, we have used ha ohogonaliy of maingale incemens: E(X sm X sm 1 ) 2 F sm 1 = EX 2 s m X 2 s m 1 F sm 1 implies: E(X sm X sm 1 ) 2 = EX 2 s m X 2 s m 1. Fo he second em on he RHS, noe ha (X sm X sm 1 ) 2 F sm. By lemma 3, and ohogonaliy of maingale incemens: So: E E E E Q (X) Q s (X) F s = E X 2 X 2 s F s = E (X X ) 2 F s. (X sm X sm 1 ) 2 ( Q n 1 n 1 E ) ( (X) Q s m (X) F sm = (X sm X sm 1 ) 2 E Q = (X sm X sm 1 ) 2 E (X X sm ) 2 F sm (2M) 2 (X sm X sm 1 ) 2 (X sm X sm 1 ) 2 ( Q ( (X sm X sm 1 ) 2 Q ) (X) Q s m (X) F sm 4M 2 E (X) Q s m (X)) 4M 2 E n 1 n 1 ) (X) Q s m (X) F sm (X sm X sm 1 ) 2 X 2 s m X 2 s m 1 4M 2 E X 2 4M 4. Thus, E Q (X) 2 4M 4 + 2 4M 4 = 12M 4. { Lemma 5 Le { n } be as in Lemma 4. Then, hee exiss a subsequence { nk } such ha Q n k conveges unifomly a.s. on 0,. Poof of Lemma 5: Since { } Q n conveges in L 2 (Ω, F, P ), i is Cauchy in his space. So, choose a { } subsequence such ha fo m n k, Le Q n k By Chebyshev s inequaliy: P k = P E Q m { k = ω Ω : sup sup Q n k+1 Q n k Q n k Q n k+1 2 < 2 k. (ω) Q n k (ω) > 1 } k 2. 1 Q k 2 k 4 nk+1 E Q n k 2 < k4 2 k. } 11

Since he RHS is summable, Boel-Canelli implies ha P lim sup k k = 0. So, fo almos all ω Ω, hee exiss N ω such ha k > N ω implies So, fo m > m > N ω : sup Q n k+1 (ω) Q n k (ω) < 1 k 2. sup (ω) Q n m (ω) Q nm Since he seies k=1 1 k 2 implies hus, fo m, m > max{n, N ω }: { Thus, Q n k m 1 k=m sup Q n k+1 (ω) Q n k (ω) < m 1 1 k 2. k=m conveges, we have ha given ɛ > 0, hee exiss N such ha m, m > N sup Q nm m 1 1 k 2 < ɛ. k=m (ω) Q n m (ω) < ɛ. } conveges unifomly almos suely on 0,. In wha follows, call he limiing funcion in Lemma 5, and define i o be zeo ouside 0,. Now, fo each nk in lemma 5, we can exend i o a paiion n k of 0, + 1, such ha n k 0. Then, fo his sequence of meshes, lemma 4 implies ha Q n k +1 conveges { o } a limi in L2 (Ω, F, P ). Repeaing he pocedue in lemma 5 hen, we can selec a subsequence Q n kj such ha i conveges unifomly almos suely on 0, + 1. Call he limiing funcion +1, and similaly define i o be zeo ouside 0, + 1. I s clea ha fo, Q n kj = Q n k, so ha = +1 fo. Repeaing he pocedue above, we obain a sequence of funcions { +j } j=0 such ha +j is coninuous on 0, + j a.s., and +j = +k fo min{ + j, + k}. So, we can unambiguously define (ω) = lim j +j (ω). If D j = {ω Ω : +j (ω) is no coninuous on 0, + j}, hen clealy D = j=0 D j has measue zeo. Thus, fo ω D c, i s clea ha (ω) will be coninuous, so ha is coninuous a.s. I s clea fom he consucion ha is pedicable. To show ha is inceasing, i is sufficien o show ha each +j is inceasing. To his end, le n be he paiion of 0, + j wih poins a k2 n ( + j) fo 0 k 2 n ; clealy, aking his sequence of paiions doesn ale he above agumens, so ha Q n +j unifomly a.e. on 0, + j. 12

Clealy, n+1 is a efinemen of n and n=1 n is dense in 0, + j. Thus, fo any pai s,, s <, in n=1 n hee exiss n 0 such ha s and belong o n fo n n 0. Thus, Q n s ha +j s Q n fo n n 0, so +j. Since his is ue fo any s <, s, n=1 n, by coninuiy of he pocess, i mus hold eveywhee on 0, + j. Thus, is coninuous, pedicable, and inceasing. ll we need o veify now is ha (X ) 2 is a maingale. Now, fo each j, +j is he limi of pocesses of he fom Q n k suely o +j. Thus, we have convegence in pobabiliy. which convege unifomly almos Similaly, since Q n k was obained as a subsequence of a sequence conveging in L 2 (Ω, F, P ) o say Q, he subsequence Q n k conveges o Q in L 2 (Ω, F, P ), so ha we also have convegence in pobabiliy. Thus, Q = R+j wih pobabiliy 1, so ha Q n k conveges o +j in L 2 (Ω, F, P ). Claim 3 Suppose ha fo each n, Z n is a maingale w... F, and ha fo each, Z n Z in L p (Ω, F, P ) whee p 1. Then, Z is a maingale. Poof of Claim 3: Recall ha since we e woking ove he finie measue space (Ω, F, P ), convegence in L p (Ω, F, P ) implies convegence in L 1 (Ω, F, P ) (why?). 2 Now, he maingale popey implies ha fo s <, EZ n F s = Zs n, so ha fo any F s EZ n F s = Since Zs n Z s in L p (and hence in L 1 ) we have ha lim EZ n n F s = lim n Now, Z n s. Zs n = E EZ n F s EZ F s p = E EZ n Z F s p EE Z n Z p F s = E Z n Z p whee we have used he condiional Jensen inequaliy, and lineaiy of condiional expecaion. Thus, EZ n F s EZ F s in L p (Ω, F, P ), so ha EZ n F s EZ F s in L 1 (Ω, F, P ). Thus: EZ F s = lim EZ n n F s = Z s so ha since F s was abiay, EZ F s = Z s. Z s. Thus, by lemma 3, since (X ) 2 Q n k is a maingale, and Q n k is a maingale. Thus, (X ) 2 is obviously a maingale. +j in L 2 (Ω, F, P ), (X ) 2 +j 13

Sep 2: Poof of exisence in heoem 2 when X is a local maingale Lemma 6 Le X be a bounded maingale, and T be a sopping ime. Then, X T = X T. Poof of Lemma 6: By he consucion in sep 1, M = (X ) 2 X is a maingale. Then, M T = (X T ) 2 X T is a maingale, so ha by uniqueness of he pocess X T, X T = X T. Now, le X be a coninuous local maingale, wih a sequence of sopping imes {T n } ha educe i. WLOG, we may ake he sopping imes o be he canonical imes: T n = inf{ : X > n}. Then, Y n = X Tn 1 Tn>0 is a bounded maingale. By he esuls in sep 1, hee is a unique, coninuous pedicable, inceasing pocess n such ha (Y n ) 2 n is a maingale. By lemma 6, fo T n, n = n+1, so ha we may unambiguously define X = n fo T n. Clealy, X is coninuous, pedicable, and inceasing. By definiion: XT 2 n 1 Tn>0 X Tn is a maingale, so ha (X ) 2 X is a local maingale. We poceed now o pove he analogue above fo he covaiance pocess. In paicula, i is vey useful in compuing X, Y. Theoem 5 Suppose ha X and Y ae coninuous local maingales. X, Y is he unique coninuous pedicable pocess ha is locally of bounded vaiaion, has 0 = 0, and makes X Y a local maingale. Poof of Theoem 5: By definiion: X Y X, Y = 1 (X + Y ) 2 X + Y 4 { (X Y ) 2 } X Y is obviously a coninuous local maingale. To pove uniqueness, noice ha if and ae wo pocesses wih he desied popeies, hen = (X Y ) (X Y ) is a coninuous local maingale ha is of locally bounded vaiaion. Hence, by heoem 4, his mus be idenically zeo, so ha = fo all. 14