Made change in trycatch error message for HLRTest() to better
describe optimization issues.
Made minor changes to article.R example file.
MSTest 0.1.6
Updated MLE estimation following deprecation of hin>=0
(inequality constraint direction) in slsqr
Changed OLS unbiased estimates of models with k=1 to be consistent
with MLE estimates. Package is for testing more than estimation so
comparison with MLE-based tetsing is prioritized.
Changed use of arma::is_finite(X) to std::isfinite(X) because former
is now deprecated.
MSTest 0.1.5
patch to compile following (Rcpp) Armadillo update (i.e., added
proper namespace scopeas_scalar).
MSTest 0.1.2
Made change to MSARmdl(), MSVARmdl(), and HMmdl(). Specifically,
when msmu or msvar is FALSE, output list copies single regime value for
each k regime. This is needed for simulating the null when either msmu
or msvar is FALSE in LMCLRTest() and MMCLRTest().
Added option to use different number of initial values for
estimating MSMs with observed data vs. for null distribution (see
documentation for LMCLRTest() and MMCLRTest()).
Updated USGNP data set to include 2022 Q3.
MMCLRTest() now has the option to add lower and upper bounds for
autoregressive coefficients and transition probabilities to help reduce
errors from polyroot() when optimizing. MMC_bounds() has been updated to
reflect this.
DLMMCTest() now has the option to add lower and upper bounds for
autoregressive coefficients to help reduce errors from polyroot() when
optimizing. DLMMC_bounds() has been created for this.
Fixed error in bootCV() used by CHPTest(). When NaN occurs, new draw
is used.
Added option to allow user to specify ‘mle_theta_low’ and
‘mle_theta_upp’ which determine the lower and upper bounds for
optimization in HMmdl(), MSARmdl(), and MSVARmdl() when “method=‘MLE’”
is specified.
print.CHPTest() now used two lines to print description.
In HLRTest() user can now define entire grid for transition
probabilities.
added new optional optimization routine for HLRparamSearch()
(nloptr::slsqp() can be used now).
Fixed bug in HLRTest() where grid for sigma options are properly
used now.
Now using nearest_spd() from pracma instead of nearPD() from lmf
package.