This preview shows pages 1–10. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: Statistics 150: Spring 2007 April 24, 2007 01 1 Introduction Definition 1.1. A sequence Y = { Y n : n } is a martingale with respect to the sequence X = { X n : n } if, for all n , 1. E  Y n  < 2. E ( Y n +1  X ,X 1 ,...,X n ) = Y n . Example 1.2 (Simple random walk). Let X i be i.i.d. random variables such that X i = 1 with probability p and X i = 1 with probability q = 1 p . Then S n = X 1 + X 2 + + X n satisfies E  S n  n and E ( S n +1  X 1 ,X 2 ,...,X n ) = S n + ( p q ) , and Y n = S n n ( p q ) defines a martingale with respect to X . 1 Example 1.3 (De Moivres martingale). A simple random walk on the set { , 1 , 2 ,...,N } stops when it first hits either of the absorbing barriers at and at N ; what is the probability that it stops at the barrier ? Write X 1 ,X 2 ,..., for the steps of the walk, and S n for the position after n steps, where S = k . Define Y n = ( q/p ) S n where p = P ( X i = 1) , p + q = 1 , and < p < 1 . 2 We claim that E ( Y n +1  X 1 ,X 2 ,...,X n ) = Y n for all n. If S n equals or N , then the process has stopped by time n and S n +1 = S n and therefore Y n +1 = Y n . If < S n < N , then E ( Y n +1  X 1 ,X 2 ,...,X n ) = E ( ( q/p ) S n + X n +1  X 1 ,X 2 ,...,X n ) = ( q/p ) S n [ p ( q/p ) + q ( q/p ) 1 ] = Y n , By taking expectations we see that E ( Y n +1 ) = E ( Y n ) for all n , and hence E ( Y n ) = E ( Y ) = ( q/p ) k for all n . 3 Let T be the number of steps before the absorption of the particle at either or N then E ( Y T ) = ( q/p ) k . Expanding E ( Y T ) , we have that E ( Y T ) = ( q/p ) p k + ( q/p ) N (1 p k ) where p k = P ( absorbed at  S = k ) . Therefore p k = k N 1 N where = q/p 4 Example 1.4 (Markov chains). Let X be a discretetime Markov chain taking values in the countable state space S with transition matrix P . Suppose that : S R is bounded and harmonic, which is to say that X j S p ij ( j ) = ( i ) for all i S. It is easily seen that Y = { ( X n ): n } is a martingale with respect to X : E ( ( X n +1 )  X 1 ,X 2 ,...,X n ) = E ( ( X n +1 )  X n ) = X j S p X n ,j ( j ) = ( X n ) . 5 Definition 1.5. Given a random variable Z we use the shorthand E [ Z F n ] for E [ Z  X ,X 1 ,...X n ] . We call F = {F , F 1 ,... } a filtration . A sequence of random variables Y = { Y n : n } is said to be adapted to the filtration F if Y n is F nmeasurable for all n , that is, if Y n is a function of X ,X 1 ,...,X n . 6 Definition 1.6. Let F be a filtration of the probability space ( , F , P ) , and let Y be a sequence of random variables which is adapted to F . We can rewrite our previous definition of a martingale by saying that the pair ( Y, F ) = { ( Y n , F n ): n } a martingale if for all n , 1. E  Y n  < 2. E ( Y n +1 F n ) = Y n 7 Definition 1.7. Let F be a filtration of the probability space ( , F , P ) , and let Y be a sequence of random variables which is adapted to...
View
Full
Document
This note was uploaded on 05/07/2008 for the course STAT 150 taught by Professor Evans during the Spring '08 term at University of California, Berkeley.
 Spring '08
 Evans
 Statistics

Click to edit the document details