The paper gives necessary and sufficient conditions for an expected-ut
ility-maximizing decision maker to prefer any compound Sigma(i)(n)=1 <
(X)over tilde(i)> of n independent, identically distributed random var
iables over any other such compound Sigma(i)(n)=1 <(Y)over tilde(i)> w
ith E<(Y)over tilde(i)> < E<(X)over tilde(i)>, provided that n is suff
iciently large. A sufficient condition is that absolute risk aversion
go to zero as the decision maker's wealth becomes unboundedly positive
or negative. The analysis is applied to give necessary and sufficient
conditions for the desirability of ''max-expected-log'' policies in m
ultiperiod choice problems with a distant time horizon. (C) 1995 Acade
mic Press, Inc.