Two Filter Robust Smoother

DESCRIPTION:

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.

USAGE:

acm.ave(x, gm, s0=<<see below>>, iter=1, a=2.5, b=5,  
        psiovw=T, savesd=T) 

REQUIRED ARGUMENTS:

x
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.

SEE ALSO:

, , .

EXAMPLES:

gm <- ar.gm(bicoal.tons, 3) 
acm.ave(bicoal.tons, gm)