Page 77 - Elements of Distribution Theory
P. 77

P1: JZP
            052184472Xc02  CUNY148/Severini  May 24, 2005  2:29





                                                     2.6 Martingales                          63

                        additional games will still be c.A formal statement of this result for martingales is given in
                        the following theorem.

                        Theorem 2.10. Let {X 1 , X 2 ,...} be a martingale and let n and m be positive integers. If
                        n < m, then

                                                  E(X m |X 1 ,..., X n ) = X n
                        with probability 1.

                        Proof. Since {X 1 , X 2 ,...} is a martingale,
                                         E[X n+1 |X 1 ,..., X n ] = X n  with probability 1.
                        Note that
                                 E[X n+2 |X 1 ,..., X n ] = E[E(X n+2 |X 1 ,..., X n , X n+1 )|X 1 ,..., X n ]

                                                  = E[X n+1 |X 1 ,..., X n ] = X n
                        with probability 1. Similarly,

                               E(X n+3 |X 1 ,..., X n ) = E[E(X n+3 |X 1 ,..., X n , X n+1 , X n+2 |X 1 ,..., X n ]
                                                = E(X n+2 |X 1 ,..., X n ) = X n
                        with probability 1. Continuing this argument yields the result.

                          The martingale properties of a sequence X 1 , X 2 ,... can also be described in terms of
                        the differences
                                               D n = X n − X n−1 ,  n = 1, 2,...
                        where X 0 = 0. Note that, for each n = 1, 2,..., (X 1 ,..., X n )isa one-to-one function of
                        (D 1 ,..., D n ) since
                                           X m = D 1 + ··· + D m ,  m = 1, 2,..., n.

                        Suppose that {X 1 , X 2 ,...} is a martingale. Then, by Theorem 2.5,
                        E{D n+1 |D 1 ,..., D n }= E{E(D n+1 |X 1 ,..., X n )|D 1 ,..., D n }
                                          = E{E(X n+1 − X n |X 1 ,..., X n )|D 1 ,..., D n }= 0,  n = 1, 2,....
                        A sequence of real-valued random variables D 1 , D 2 ,... satisfying

                                           E{D n+1 |D 1 ,..., D n }= 0,  n = 1, 2,...
                        is said to be a sequence of martingale differences.
                          As noted above, if X 1 , X 2 ,... is a martingale, then X n can be interpreted as the fortune
                        of a gambler after a series of fair games. In the same manner, if D 1 , D 2 ,... is a martingale
                        difference sequence, then D n can be interpreted as the amount won by the gambler on the
                        nth game.

                        Example 2.25 (Gambling systems). Suppose that a gambler plays a series of fair games
                        with outcomes D 1 , D 2 ,... such that, if the gambler places a bet B n on the nth game, her
                        winnings are B n D n .For each n, the bet B n is a function of D 1 ,..., D n−1 , the outcomes of
   72   73   74   75   76   77   78   79   80   81   82