smooth.ES.confint

ES.confint(parm=None, level=0.95, type=None, bootstrap=False, step_size=None, **boot_kwargs)

Confidence intervals for the estimated parameters.

Mirrors R’s confint.adam: standard errors from vcov(), t-interval half-widths (with R’s asymmetric degrees of freedom), then clamping to the admissible region for ETS smoothing parameters (bounds="usual" or "admissible"), multiplicative initial states, and ARIMA AR/MA parameters. With bootstrap=True the intervals are the empirical quantiles of the replicate matrix returned by coefbootstrap() (no clamping / no t-quantile).

Parameters:

  • parm (str or sequence of str, optional) – Subset of names to return.

  • level (float, default=0.95) – Confidence level.

  • bootstrap (bool, default=False) – Switch to empirical-quantile intervals via coefbootstrap().

  • step_size (float, optional) – Finite-difference step forwarded to vcov() for the Fisher-based path. Ignored when bootstrap=True.

  • **boot_kwargs – Forwarded to coefbootstrap() (nsim, size, …).

Returns:

Columns ["S.E.", "<lo>%", "<hi>%"] indexed by coef_names.

Return type:

pandas.DataFrame

Parent Class: ES