Forestry investment decisions may be based on the probability distribu
tion of financial return in addition to a point estimate of mean retur
n. This study describes an approach to predicting the present value di
stribution of a plantation investment using actual data on timber pric
e and yield. Changes in stumpage price are modeled with a lognormal di
ffusion process called geometric Brownian motion (GEM), Timber yield i
s modeled with a variant of GEM that includes an age-dependent growth
component. Model parameters are estimated with time-series observation
s of loblolly pine (Pinus taeda L.) price and yield in the southeaster
n United States. Because GEM models have lognormally distributed error
s, present value distributions are skewed with extremely long right-ha
nd tails. The median and quartiles of the distribution provide a bette
r measure of central tendency and spread than do the mean and standard
deviation. A median-maximizing feedback cutting rule does not perform
any better than a median-maximizing fixed rotation age suggesting tha
t no economic gain can be obtained by monitoring timber price and yiel
d under the assumptions of our models, The forecast error, measured by
the distance between quartiles, is about twice the size of the median
present value. System error is the primary cause, and error in the pr
ice process contributes more to the variability in present value than
does error in the yield process, Parameter uncertainty increases forec
ast error 15 to 35%. The large forecast error raises the question of w
hether better predictive models can be built or whether the present va
lue of a plantation investment is inherently uncertain.