Internal function that handles excessive indicator retention by re-running
GETS on smaller blocks of indicators using a leave-one-out method. This
can be necessary when following a block search, e.g. for IIS, more indicators
than observations have been identified. Then another block search is carried
out. This can be avoided by choosing tighter significance levels in
t.pval or improving the specification.
ISadditionalblocksearch(mXis, isNames, y, y.n, mX, mXnames, mxkeep,
ISmatrixname, indicator.set, print.searchinfo, tol, LAPACK, userEstArg,
t.pval, gof.method, gofFunArg, max.regs, max.paths, arLjungB, archLjungB,
normality.JarqueB, user.diagnostics, include.gum, include.1cut,
include.empty, turbo, do.pet, ratio.threshold, max.block.size,
wald.pval, additional.block.search)A list containing:
The final selected model.
Updated estimation counter.
Updated function call counter.
Final block structure.
Initial regressor matrix including previously selected indicators.
Character vector with names of the candidate indicator variables.
Response variable, typically a numeric vector or time-series.
Integer indicating the number of observations in the sample.
Matrix of original regressors, excluding indicators.
Character vector of column names corresponding to mX.
Indices of regressors to always retain during selection.
Character string naming the current IS matrix (e.g., "SIS", "TIS").
Matrix of candidate indicators from which subsets are selected.
Logical; if TRUE, outputs progress messages during execution.
Numeric tolerance used in collinearity checks (see qr).
Logical; if TRUE, LAPACK is used for QR decomposition.
List specifying the user-defined estimation function and its arguments.
Significance level for coefficient t-tests in the GETS algorithm.
Character indicating whether to minimize or maximize the goodness-of-fit criterion.
List containing the name and arguments of the GOF function used.
Maximum number of regressions allowed per deletion path.
Maximum number of GETS paths to be explored.
List with order and significance level for the Ljung-Box test on residual autocorrelation.
List with order and significance level for the Ljung-Box test on squared residuals (ARCH).
Significance level for the Jarque-Bera test for normality;
if NULL, test is skipped.
List containing user-specified diagnostic function and p-value thresholds.
Logical; if TRUE, the General Unrestricted Model is included
in terminal models.
Logical; if TRUE, the 1-cut model is included in terminal models.
Logical; if TRUE, the empty model is included in terminal models.
Logical; avoids redundant GETS path evaluations if TRUE.
Logical; if TRUE, performs Parsimonious Encompassing Tests (PETs).
Numeric ratio used to determine block sizes in relation to sample size.
Maximum number of indicators allowed per block during the leave-one-out search.
Significance threshold for Wald tests in the GETS algorithm.
Default is FALSE. If isat() identifies (retains) too many indicators in the block search, a sensible model might not be estimable (e.g. due to a lack of available degrees of freedom). In this case, setting additional.block.search = TRUE carries out an additional block search using a Leave-one-Out method (e.g. for three blocks, indicators are assigned to block 1,2,3,1,2,3,1,2,3). This is an experimental feature.
Moritz Schwarz, https://www.moritzschwarz.org/
ISMatricesLoop, isat