Vous êtes sur la page 1sur 5

A "Fatou Equation" for Randomly Stopped Variables

Author(s): William D. Sudderth


Source: The Annals of Mathematical Statistics, Vol. 42, No. 6 (Dec., 1971), pp. 2143-2146
Published by: Institute of Mathematical Statistics
Stable URL: http://www.jstor.org/stable/2240142
Accessed: 29-05-2018 14:24 UTC

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms

Institute of Mathematical Statistics is collaborating with JSTOR to digitize, preserve and


extend access to The Annals of Mathematical Statistics

This content downloaded from 140.113.213.2 on Tue, 29 May 2018 14:24:47 UTC
All use subject to http://about.jstor.org/terms
The Annals of Mathematical Statistics
1971, Vol. 42, No. 6, 2143-2146

A "FATOU EQUATION" FOR RANDOMLY STOPPED VARIABLES'

BY WILLIAM D. SUDDERTH

University of Minnesota

Let Xn, be a sequence of random variables adapted to an increasing


sequence of a-fields. In this note, convergence properties of EX, are studied
as t -+ oo through the directed set of stopping variables. The analogue of
the inequality in Fatou's Lemma turns out to be an equation, which
strengthens Fatou's Lemma.
These problems arise naturally in the theory of gambling.

Let (Q, F, P) be a probability space, {F,},?, I an increasing sequence of a-fields


contained in F, and {Xn}1, a sequence of random variables such that Xn, is Fn-
measurable for every n. A random variable t is a stopping variable (sv) if its range is
contained in {1, 2, ***, + oo}, P[t < + ao] = 1, and, for every positive integer
n, [t < n] J F,. If t and s are stopping variables, write t < s if t(Q) < s(o) for all
co E Q. With this natural partial ordering, the stopping variables form a directed
set. Some of the convergence properties of the net EX, are considered below. In
particular, Theorem 2 is an analogue of Fatou's Lemma in which an equation
replaces the usual inequality.
In what follows, the letters "s" and "t" always denote stopping variables and the
letters "k" and "n" positive integers.

THEOREM 1. The following inequalities hold whenever all the expectations occurring
in them are well-defined:

(1) E(lim supn , OO Xj) < lim sup, , OO EX,


(I1t) E(lim inf,, 0, Xj) > lim inf, ,OO EX,.
PROOF. It is enough to prove (1). For convenience, let X* = lim sup,, O
Equation (1) is obvious if EX* = - oo. Let us assume now that X* is integrable
and return later to the case when EX* = + oo.
Let s > 0 and let s be a sv. Define

t(w) = inf{n: n ? s(w) and E(X* I Xi, ***, Xj)(C) < X'(W)+81.
By Levy's martingale convergence theorem (29.4, 12]), E(X* I X1, -., Xn) -X*
almost surely as n -+ oo and, hence, P[t < + oo] = 1. Thus t is a sv, t > s, and

EXt =1 f [t=n] E(X* I X1, * Xn)dP-s


= EX*-s, which proves (1).

Received August 10, 1970.


1 This research was supported by the National Science Foundation under NSF Grant GP-9556.
Key words andphrases. Stopping time, Fatou's Lemma, gambling.

2143

This content downloaded from 140.113.213.2 on Tue, 29 May 2018 14:24:47 UTC
All use subject to http://about.jstor.org/terms
2144 WILLIAM D. SUDDERTH

Finally, suppose EX* = + so. Let c be a real number and apply the case just
considered to the random variables min (Xn, c). Then

lim supt,,, EX, > lim supt,, E(min (Xe, c))


> E(min (X*, c)) -+ EX*

as c - + oo. 1
Certain results of Siegmund in [3] are closely related to the ideas of this note. In
particular, the previous theorem seems to follow from his Theorem 3(a) in the
special case that EX,- < + oo for all n.

THEOREM 2. Suppose Z and W are integrable random variables. If Xn < Z for all
n, then

(2) E(lim supnoo Xn) = lim supt,O EX,.


If Xn _ Wfor all n, then

(2') E(lim infn , Xn) = lim inf,c,, EX,.


PROOF. The proof is a simple modification of the proof of Fatou's Lemma
([2], p. 125). Let Wn = supk?fn Xk. Then

W4l X* = lim supn Xn as n s o. Also, Xt < Wn for t > n. Hence


lim supt, 00 EXt < limn {sup,>I n EX,} _ limn0 E Wn EX*
The opposite inequality is true by Theorem 1. 0
When the Xn are dominated above, it is easy to see that lim supt., EX, >
lim supn,0, EXn. Thus, Theorem 2 implies the usual Fatou Lemma.
For uniformly bounded Xn, a version of Theorem 2 has been proved in which
sv' s are not assumed to be measurable. This result has an interpretation for the
Dubins and Savage utility of a measurable strategy and together with a gambling
theorem (3.9.5, [1]) has been used to establish an optimal stopping result (Theorem
5, [5]) parallel to Siegmund's Theorem 4 in [3]. Further applications to gambling
theory are in [6].
The next result is immediate from Theorem 2 and is also easy to prove directly.

COROLLARY. Suppose Xn -+ X a.s. and there is an integrable random variable Z


such that IXnI ? Zfor all n. Then X is integrable and lim,,oc EX, = EX.
Recall that if Xn -+ X a.s. and the family {Xn}ln?1 is uniformly integrable, then
Xis integrable and EXn -* EX. The following example shows that, under the same
hypotheses, it is not necessarily true that EX, -+ EX. It also provides a non-trivial
example (though not the simplest) in which (2) fails to hold.

EXAMPLE. The Xn constructed here will be nonnegative, uniformly integrable,


convergent to zero a.s., and such that

This content downloaded from 140.113.213.2 on Tue, 29 May 2018 14:24:47 UTC
All use subject to http://about.jstor.org/terms
A "FATOU EQUATION" FOR RANDOMLY STOPPED VARIABLES 2145

Let (Q, F, P) be the unit interval with its Borel sets and Lebesgue measure. Let
{Zn}n 1 be independent rv's and assume Zn is uniformly distributed on [0, 1] for
all n. Now define random variables Yi7 for n = 1, 2, - and i = 1, , n by

yn = n if(i-1)/n2 ? Zn < i/n2,


= 0 if not.

Notice that the random vectors (y,n *n,- yn) n = 1, 2, are inde
Choose a sequence of positive integers n1 < n2 < ... such that

j?k(j) 1 -+1 as k -+ oo.


Let X1, X2, be the sequence Y121, Y2"', ..., yni 12
Fn be the a-field generated by X1, , Xn for all n.
The Xn are uniformly integrable since

supn IXn|IdP = o0 ask -oo.


t [|Xnl>nk] nk

Also, if s > 0 and X. = y1nk, then

P[IXnI < e for all n > j] = Hli_ik l[IY1

H> f(1-h -1 ask-*oo.

Hence, Xn - 0 a.s.
Let s be any sv. To prove (3), it suffices to exhibit a sv t such that t > s and
EXt 2 1.
Let w e Q and suppose s(co) = n. Letjn be the firstj such thatj > n and Xj = Y1m
for some m. Define

t(w) = jn+i if Y1m = =Yi -0, Y 1 = m,


=jn+m-1 if ym= = Ymm =0.

Thus X1(o) = max (Y1m(o), , Ymm(o)) if s(o) = n. Also, (Y1m, , Ymm) is


independent of [s = n]. Therefore, if s() =n,

E(X, I s = n)(w) = E(max (Y1m, ,Ymm))


=1.

Hence, EXt = 1.

Acknowledgment. I want to thank Lester Dubins, George Duncan, and Mike


Perlman for their helpful comments, and the referee for a simplification of the
proof of Theorem 1.

This content downloaded from 140.113.213.2 on Tue, 29 May 2018 14:24:47 UTC
All use subject to http://about.jstor.org/terms
2146 WILLIAM D. SUDDERTH

REFERENCES

[1] DUBINS, L. E. and SAVAGE, L. J. (1965). How to Gamble If You Must. McGraw-Hill, New York.
[2] LOEVE, M. (1963). Probability Theory. 3rd ed. Van Nostrand, Princeton.
[3] SIEGMUND, D. 0. (1967). Some problems in the theory of optimal stopping rules. Ann. Math.
Statist. 38 1627-1640.
[4] SUDDERTH, W. D. (1971). On measurable gambling problems. Ann. Math. Statist. 42 260-269.
[5] SUDDERTH, W. D. (1971). A gambling theorem and optimal stopping theory. Ann. Math.
Statist. 42 1697-1705.
[6] SUDDERTH, W. D. (1970). On the Dubins and Savage characterization of optimal strategies.
Univ. of Minnesota, Dept. of Statistics Tech. Report No. 138.

This content downloaded from 140.113.213.2 on Tue, 29 May 2018 14:24:47 UTC
All use subject to http://about.jstor.org/terms

Vous aimerez peut-être aussi