1: \begin{abstract}

2:   The FARIMA models, which have long-range-dependence (LRD), are

3:   widely used in many areas. Through deriving a precise

4:   characterisation of the spectrum, autocovariance function, and

5:   variance time function, we show that this family is very atypical

6:   among LRD processes, being extremely close to the fractional

7:   Gaussian noise in a precise sense. Furthermore, we show that this

8:   closeness property is not robust to additive noise.  We argue that

9:   the use of FARIMA, and more generally fractionally differenced time

10:   series, should be reassessed in some contexts, in particular when

11:   convergence rate under rescaling is important and noise is expected.

12: \end{abstract}

13: