filt
of the output) will be free of outliers.
acm.filt(x, gm, s0=0, iter=1, a=2.5, b=5.0, psiovw=T, bottom=T, saves1=T)
ar.gm
function which includes components
ar
containing AR coefficient estimates,
sinnov
containing innovation scale estimates from AR fits of
orders
1
through
order
,
chat
containing an estimate of the
order
by
order
autocovariance matrix,
and
rmu
, the estimated mean of
x
.
Both
ar
and
sinnov
are vectors of length
order
.
TRUE
, use psi(t)/t to calculate the
weight function w(t); if
FALSE
,
use psi'(t).
TRUE
, filtering from the bottom of xhats is performed,
in which case lag p-1 smoothing is performed.
If
FALSE
, filtering is from the top.
The vector xhats contains estimates of x(t), x(t-1),..., x(t-p+1) at time t.
TRUE
, the time varying scale
st
will be
(M(t)[1,1] + v0)^.5
, where
v0 = s0*s0
.
If
FALSE
,
st
will be the square root of the appropriate element of
P(t)
.
P(t)
is the covariance matrix of the error in xhats,
and
M(t)
is a one-step prediction error covariance matrix.
order
(=
length(gm$ar)
) containing estimates of the
autoregression parameters computed via least squares from the filtered data.
order
by
order
Toeplitz
matrix containing an estimate of the autocovariance matrix of the filtered data.
saves1
.
Martin, R. D. and Thomson, D. J. (1982).
Robust resistant spectrum estimates.
Proceedings of the IEEE
70, 1097-1115.
Martin, R. D. (1981).
Robust methods for time series. In
Applied Time Series Analysis II.
D. F. Findley, ed. Academic Press, New York. pp. 683-759.
The chapter "Analyzing Time Series" of the S-PLUS Guide to Statistical and Mathematical Analysis.
gm <- ar.gm(bicoal.tons, 3); acm.filt(bicoal.tons, gm)