artfima (version 1.3)

bestModels: Best BIC Models

Description

ARIMA(p,0,q), ARFIMA(p,0,q) and ARTFIMA(p,0,q) models are fit for various p=0,1,..., and q=0,1,... and the best models according to the BIC criterion are selected.

Usage

bestModels(z, p = 2, q = 2, nbest = 5, likAlg = c("exact", "Whittle"), 
    use_parallel = TRUE)

Arguments

z

time series data

p

maximum order of AR component

q

maximum order of AR component

nbest

number of models in selection

likAlg

likelihood method to use

use_parallel

By default the R parallel package is used to speed up the computations. The speedup is modest - about 20 to 30 percent less time - see example below. Requires that you have at least 3 cores. Set use_parallel=FALSE if you don't want to use parallel.

Value

• The plausiblity score or the actual AIC/BIC values.

Details

The models are ranked according to the AIC/BIC criterion and the best ones are shown.

See Also

best_glp_models

Examples

#comparing times w/o parallelization
z<-tseg(1000, "BJARMA11")
#comparing times using parallel and not using
startTime <- proc.time()[3]
bestModels(z)
proc.time()[3]-startTime #about 132 sec
startTime <- proc.time()[3]
bestModels(z, use_parallel=FALSE)
proc.time()[3]-startTime #about 167 sec
#Remark: The correct model is ARIMA(1,0,1) but BIC tends to overfit. Usually
#the ARIMA(1,0,1) is in the top 5 and it is usually the most parsimonious 
#model in this group. If we use a stronger penalty such as BICq, q=1/4,
#the ARIMA(1,0,1) would be selected as the top model more often.