gets.larch: General-to-Specific (GETS) Modelling of a heterogeneous...
In gets: General-to-Specific (GETS) Modelling and Indicator Saturation Methods

gets.larch

R Documentation

General-to-Specific (GETS) Modelling of a heterogeneous log-ARCH-X model

Description

The starting model, an object of the 'larch' class (see larch, is referred to as the General Unrestricted Model (GUM). The gets.larch() function undertakes multi-path GETS modelling of the log-variance specification. The diagnostic tests are undertaken on the standardised residuals, and the keep option enables regressors to be excluded from possible removal.

Usage

  ## S3 method for class 'larch'
gets(x, t.pval=0.05, wald.pval=t.pval, do.pet=TRUE, 
    ar.LjungB=NULL, arch.LjungB=NULL, normality.JarqueB=NULL, 
    user.diagnostics=NULL, info.method=c("sc", "aic", "aicc", "hq"),
    gof.function=NULL, gof.method=NULL, keep=c(1), include.gum=FALSE,
    include.1cut=TRUE, include.empty=FALSE, max.paths=NULL, tol=1e-07,
    turbo=FALSE, print.searchinfo=TRUE, plot=NULL, alarm=FALSE, ...)

Arguments

`x`	an object of class 'larch'
`t.pval`	numeric value between 0 and 1. The significance level used for the two-sided regressor significance t-tests
`wald.pval`	numeric value between 0 and 1. The significance level used for the Parsimonious Encompassing Tests (PETs). By default, `wald.pval` is equal to `t.pval`
`do.pet`	logical. If `TRUE` (default), then a Parsimonious Encompassing Test (PET) against the GUM is undertaken at each regressor removal for the joint significance of all the deleted regressors along the current path. If `FALSE`, then a PET is not undertaken at each regressor removal
`ar.LjungB`	`NULL` (default), or a `list` with named items `lag` and `pval`, or a two-element numeric vector where the first element contains the lag and the second the p-value. If `NULL`, then the standardised residuals are not checked for autocorrelation. If `ar.LjungB` is a `list`, then `lag` contains the order of the Ljung and Box (1979) test for serial correlation in the standardised residuals, and `pval` contains the significance level. If `lag=NULL`, then the order used is that of the estimated 'larch' object
`arch.LjungB`	`NULL` (default), or a `list` with named items `lag` and `pval`, or a two-element numeric vector where the first element contains the lag and the second the p-value. If `NULL`, then the standardised residuals are not checked for ARCH (autocorrelation in the squared standardised residuals). If `ar.LjungB` is a `list`, then `lag` contains the order of the test, and `pval` contains the significance level. If `lag=NULL`, then the order used is that of the estimated 'larch' object
`normality.JarqueB`	`NULL` (default) or a numeric value between 0 and 1. If `NULL`, then no test for non-normality is undertaken. If a numeric value between 0 and 1, then the Jarque and Bera (1980) test for non-normality is conducted using a significance level equal to the numeric value
`user.diagnostics`	`NULL` (default) or a `list` with two entries, `name` and `pval`, see the `user.fun` argument in `diagnostics`
`info.method`	character string, "sc" (default), "aic", "aicc" or "hq", which determines the information criterion to be used when selecting among terminal models. See `infocrit` for the details
`gof.function`	`NULL` (default) or a `list`, see `getsFun`. If `NULL`, then `infocrit` is used
`gof.method`	`NULL` (default) or a `character`, see `getsFun`. If `NULL` and `gof.function` is also `NULL`, then the best goodness-of-fit is characterised by a minimum value
`keep`	the regressors to be kept (i.e. excluded from removal) in the specification search. Currently, `keep=c(1)` is obligatory, which excludes the log-variance intercept from removal
`include.gum`	logical. If `TRUE`, the GUM (i.e. the starting model) is included among the terminal models. If `FALSE` (default), the GUM is not included
`include.1cut`	logical. If `TRUE` (default), then the 1-cut model is added to the list of terminal models. If `FALSE`, the 1-cut is not added, unless it is a terminal model in one of the paths
`include.empty`	logical. If `TRUE`, then an empty model is included among the terminal models, if it passes the diagnostic tests. If `FALSE` (default), then the empty model is not included
`max.paths`	`NULL` (default) or an integer equal to or greater than 0. If `NULL`, then there is no limit to the number of paths. If an integer (e.g. 1), then this integer constitutes the maximum number of paths searched (e.g. a single path)
`tol`	numeric value. The tolerance for detecting linear dependencies in the columns of the variance-covariance matrix when computing the Wald-statistic used in the Parsimonious Encompassing Tests (PETs), see the `qr.solve` function
`turbo`	logical. If `TRUE`, then paths are not searched twice (or more) unnecessarily, thus yielding a significant potential for speed-gain. However, the checking of whether the search has arrived at a point it has already been comes with a computational overhead. Accordingly, if `turbo=TRUE`, the total search time might in fact be higher than if `turbo=FALSE`. This is particularly likely to happen if estimation is very fast, say, less than a quarter of a second. Hence the default is `FALSE`
`print.searchinfo`	logical. If `TRUE` (default), then a print is returned whenever simiplification along a new path is started
`plot`	`NULL` or logical. If `TRUE`, then the fitted values and the standardised residuals of the final model are plotted after model selection. If `FALSE`, then they are not plotted. If `NULL` (default), then the value set by `options` determines whether a plot is produced or not
`alarm`	logical. If `TRUE`, then a sound or beep is emitted (in order to alert the user) when the model selection ends, see `alarm`
`...`	additional arguments

Details

See Pretis, Reade and Sucarrat (2018): \Sexpr[results=rd]{tools:::Rd_expr_doi("10.18637/jss.v086.i03")}, and Sucarrat (2020): https://journal.r-project.org/archive/2021/RJ-2021-024/.

The arguments user.diagnostics and gof.function enable the specification of user-defined diagnostics and a user-defined goodness-of-fit function. For the former, see the documentation of diagnostics. For the latter, the principles of the same arguments in getsFun are followed, see its documentation under "Details", and Sucarrat (2020): https://journal.r-project.org/archive/2021/RJ-2021-024/.

Value

A list of class 'larch', see larch, with additional information about the GETS modelling

Author(s)

Genaro Sucarrat, http://www.sucarrat.net/

References

C. Jarque and A. Bera (1980): 'Efficient Tests for Normality, Homoscedasticity and Serial Independence'. Economics Letters 6, pp. 255-259. \Sexpr[results=rd]{tools:::Rd_expr_doi("10.1016/0165-1765(80)90024-5")}

G. Ljung and G. Box (1979): 'On a Measure of Lack of Fit in Time Series Models'. Biometrika 66, pp. 265-270

Felix Pretis, James Reade and Genaro Sucarrat (2018): 'Automated General-to-Specific (GETS) Regression Modeling and Indicator Saturation for Outliers and Structural Breaks'. Journal of Statistical Software 86, Number 3, pp. 1-44. \Sexpr[results=rd]{tools:::Rd_expr_doi("10.18637/jss.v086.i03")}

Genaro Sucarrat (2020): 'User-Specified General-to-Specific and Indicator Saturation Methods'. The R Journal 12:2, pages 388-401. https://journal.r-project.org/archive/2021/RJ-2021-024/

Examples

##Simulate some data:
set.seed(123)
e <- rnorm(40)
x <- matrix(rnorm(4*40), 40, 4)

##estimate a log-ARCH(3) with asymmetry and log(x^2) as regressors:
gum <- larch(e, arch=1:3, asym=1, vxreg=log(x^2))

##GETS modelling of the log-variance:
simple <- gets(gum)

##GETS modelling with intercept and log-ARCH(1) terms
##excluded from removal:
simple <- gets(gum, keep=c(1,2))

##GETS modelling with non-default autocorrelation
##diagnostics settings:
simple <- gets(gum, ar.LjungB=list(pval=0.05))

##GETS modelling with very liberal (40%) significance level:
simple <- gets(gum, t.pval=0.4)

gets documentation built on Sept. 11, 2024, 9:03 p.m.