Computes robust smooth using an average of forward and backward approximate
conditional mean type robust filters.
The robustly smoothed series (component
smo of
the output) will be free of outliers.
univariate time series or a vector.
Missing values are not allowed.
gm
list returned by function
ar.gm applied to
x;
see
acm.flt for details.
OPTIONAL ARGUMENTS:
s0
scale of the nominal Gaussian component of the
additive noise excluding outliers,
the default is the estimated innovations standard deviation
from the
gm list multiplied by 0.001.
s0 must be greater than
0.
iter
number of iterations.
a
first break point for the Hampel two-part psi used in filtering.
b
second break point for the Hampel psi.
psiovw
logical flag: if
TRUE, use psi(t)/t to calculate the weight function w(t);
if
FALSE,
use psi'(t).
savesd
logical flag: if
TRUE, the estimated time varying scale
of smoothing error is returned.
VALUE:
a list with the following components:
ar
vector of length
order (=
length(gm$ar)) containing the
least squares estimate of the autoregression parameters computed
from the smoothed data.
chat
an
order by
order Toeplitz matrix containing
the estimated autocovariance matrix of the smoothed data.
si
estimate of the innovations scale from the smoothed data.
smo
vector or time series containing the smoothed data.
st
vector or time series containing the estimated scale of the smoother error.
This is returned only when
savesd is
TRUE.
REFERENCES:
The chapter "Analyzing Time Series" of the S-PLUS Guide to Statistical and Mathematical Analysis.