coef() now returns a named vector in the same order
and with the same names as vcov()
(e.g. regime1:y1:x); the per-regime tables remain in
summary().
New argument robust = c("none", "hc", "hac") in
sysbreak(), varbreak() and
netbreak() replaces the logical hac;
"hac" is the former hac = TRUE.
"hc" computes the robust SupF tests, coefficient standard
errors and break-date confidence intervals with a
heteroskedasticity-consistent covariance estimate without autocovariance
terms.
The automatic HAC bandwidth is capped at T / (2d), where d is the number of score columns, so that the HAC estimate does not become near singular in short segments.
The internal pseudo-inverse discards only numerically zero singular values (relative tolerance 1e-14 instead of the square root of machine epsilon). Results no longer depend on the units of the variables; the Application 1 estimates and statistics change (the dI-equation coefficients, and the GDP break from 1983:Q1 to 1982:Q4).
print(summary(fit)) has a new layout. The sequential
tests are labelled SEQ(l+1|l), as in Qu and Perron (2007),
also in the row names of summary(fit)$tests.