randomArCoefficients {mixAR}

Random initial values for MixAR estimation

Description

Translations of functions from my Mathematica sources. Not used currently?

Usage

randomArCoefficients(ts, wv, pk, pmax, partempl, sub_size = 10,
                     condthr = 10, nattempt = 10, startfrom = pmax + 1)

randomMarParametersKernel(ts, ww, pk, pmax, partempl, ...)

randomMarResiduals(ts, p, partempl)

tsDesignMatrixExtended(ts, p, ind, partempl)

Arguments

ts

time series.

wv, ww

a vector of weights (?).

pk

the AR order of the requested component.

pmax

the maximal AR order in the model. Needed since it cannot be determined by functions working on a single component.

partempl

parameter template, a list containing one element for each mixture component, see Details.

sub_size

the size of the subsample to use, default is 10.

condthr

threshold for the condition number.

nattempt

if condthr is not reached after nattempt attempts, the function returns the results from the last subset tried.

startfrom

the starting index (in ts) to use for subsampling, default is pmax + 1.

...

arguments to pass on to randomArCoefficients().

p

a vector of non-negative integers, the MixAR order.

ind

a vector of positive integers specifying the indices of the observations to use for the “response” variable.

Details

randomArCoefficients tries small subsamples (not necessarilly contiguous) from the observations in search of a cluster hopefully belonging to one mixture component and estimates the corresponding shift and AR parameters.

randomMarResiduals selects random parameters for each mixture component and returns the corresponding residuals. randomMarParametersKernel is a helper function which does the computation for one component.

tsDesignMatrixExtended forms the extended design matrix corresponding to a subsample. This is used for least square estimation of the parameters.

Author(s)

Georgi N. Boshnakov

See Also

tomarparambyComp