strucchange_wrapper: Bai-Perron structural break wrapper (strucchange)

In ggchangepoint: Combines Changepoint Analysis with 'ggplot2'

Bai-Perron structural break wrapper (strucchange)

Description

Wraps strucchange::breakpoints() (Zeileis et al., 2002), the dynamic-programming implementation of the Bai and Perron (1998, 2003) multiple structural break estimator. Called with a bare numeric vector it dates mean shifts (y ~ 1); called with a formula and data it dates breaks in arbitrary regression coefficients. Break-date confidence intervals from confint() populate ci_lower/ci_upper and render via autoplot(show_ci = TRUE).

Usage

strucchange_wrapper(
  x,
  data = NULL,
  breaks = NULL,
  h = 0.15,
  conf_level = 0.95,
  ...
)

Arguments

x	A numeric vector (mean-shift mode), or a model formula (regression mode; supply data too).
data	Optional data frame for formula input.
breaks	Maximum number of breaks; when NULL the number is chosen by BIC.
h	Minimal segment size, as a fraction of the sample size (or an integer count). Defaults to 0.15.
conf_level	Confidence level for the break-date intervals. Defaults to 0.95.
...	Additional arguments passed to strucchange::breakpoints().

Value

A ggcpt object with ci_lower/ci_upper columns on the changepoints tibble.

Result size

$fit is the breakpoints object itself, and that object is quadratic in the series length: it keeps RSS.triang, the triangular table of segment residual sums of squares, which is what lets strucchange return the optimal segmentation for any number of breaks without refitting. Measured here, the whole result is about 1.7 MB at n = 200, 5.9 MB at n = 400 and 22.6 MB at n = 800, roughly four times larger each time the series doubles, and that one table outweighs everything else in the fit put together, by a margin that widens as the series grows. A single fit is not a problem; a few hundred of them are, so when running this engine over a panel with cpt_batch() keep what you need (res$changepoints) rather than the whole list of results. No other engine here behaves this way: the median result across the others is under ten times the size of the series it was given.

References

\insertRef

bai2003computationggchangepoint

\insertRef

zeileis2002strucchangeggchangepoint

See Also

Other changepoint engines: bcp_wrapper(), beast_wrapper(), bfast_wrapper(), binsegrcpp_wrapper(), bocpd_wrapper(), capa_wrapper(), cpm_wrapper(), cpop_wrapper(), cpt_wrapper(), decafs_wrapper(), ecp_wrapper(), envcpt_wrapper(), esac_wrapper(), fabisearch_wrapper(), fastcpd_wrapper(), fcov_wrapper(), fmean_wrapper(), fpop_wrapper(), geomcp_wrapper(), hdcov_wrapper(), hdreg_wrapper(), idetect_wrapper(), inspect_wrapper(), kcp_wrapper(), kwc_wrapper(), mcp_wrapper(), mosum_wrapper(), network_wrapper(), not_wrapper(), npmojo_wrapper(), nsp_wrapper(), ocd_wrapper(), pilliat_wrapper(), segmented_wrapper(), smuce_wrapper(), sn_wrapper(), taylor_wrapper(), tguh_wrapper(), trend_wrapper(), var_wrapper(), wbs2_wrapper(), wbs_wrapper(), wbsts_wrapper()

Examples

set.seed(2026)
res <- strucchange_wrapper(c(rnorm(100), rnorm(100, 3)))
res$changepoints

ggchangepoint documentation built on Nov. 5, 2026, 9:06 a.m.