In recent years it has been shown empirically that stock returns exhib
it positive or negative autocorrelation, depending on observation freq
uency. In this context of autocorrelated returns the present paper is
the first to derive an explicit analytical solution to the dynamic por
tfolio problem of an individual agent saving for retirement (or other
change of status, like the purchase of a house or starting college). U
sing a normal ARMA(1, 1) process, dynamic programming techniques combi
ned with the use of Stein's Lemma are employed to examine ''dollar-cos
t-averaging'' and ''age effects'' in intertemporal portfolio choice wi
th CARA preferences. We show that with a positive moving average param
eter and positive risk free rates, if first-order serial correlation i
s nonnegative, then the expected value of the optimal risky investment
is increasing over time, while if first-order serial correlation is n
egative this path can be increasing or decreasing over time. Thus a ne
cessary but not sufficient condition to obtain the conventional age ef
fect of increasing conservatism over time is that first-order serial c
orrelation be negative. Further, dollar-cost averaging in the general
sense of gradual entry into the risky asset does not emerge as an opti
mal policy. Simulation results for U.S. data are used to illustrate op
timal portfolio paths.