The Hurst exponent
is a scaling exponent used to quantify long-range dependence and roughness in a time series. In the rescaled-range definition, if
is the statistical
range of the cumulative sums of deviations
from the sample mean in a block of length
and
is its sample standard
deviation, then
where
denotes the expectation value of
and the relation is asymptotic as
grows. Brownian motion
has
. In models for which this scaling describes correlations
between increments, a process with
is a persistent process,
while one with
is an antipersistent process. For a self-affine
graph with stationary increments, its fractal
dimension is commonly related by
.
In rescaled-range analysis, is estimated from the slope obtained by least
squares fitting of
against
over a selection of block sizes.
For example, consider the eight observations 2, 4, 3, 6, 8, 7, 9, and 5. Their sample mean is , and the cumulative
sums of their deviations from the sample mean
range from
to
. Consequently,
. The sample standard
deviation is
, so this block contributes
An estimate of
repeats this computation for many blocks at several block lengths and fits the slope
of the resulting logarithms rather than using a single
block.