A class of univariate fractional ARIMA models with a continuous time p
arameter is developed for the purpose of modeling long-memory time ser
ies. The spectral density of discretely observed data is derived for b
oth point observations (stock variables) and integral observations (fl
ow variables). A frequency domain maximum likelihood method is propose
d for estimating the long-memory parameter and is shown to be consiste
nt and asymptotically normally distributed, and some issues associated
with the computation of the spectral density are explored.