MarkovSwitchingARResult.state_space

property MarkovSwitchingARResult.state_space: _MarkovSwitchingStateSpace

The fitted system, re-applicable to data it was not estimated on.

Filtering a new series through this reports what the estimated regimes say about observations the fit never saw – the natural out-of-sample check for a regime model, and one the fitted arrays alone cannot give.

Example

The fitted chain dates a fresh draw from the same process:

>>> import numpy as np
>>> rng = np.random.default_rng(0)
>>> p = np.array([[0.97, 0.03], [0.05, 0.95]])
>>> def draw(n):
...     s = np.zeros(n, dtype=int)
...     for t in range(1, n):
...         s[t] = np.searchsorted(np.cumsum(p[s[t - 1]]), rng.random())
...     return s, np.array([-1.0, 1.5])[s] + 0.5 * rng.standard_normal(n)
>>> _, y = draw(1500)
>>> res = MarkovSwitchingAR(y, order=1, n_regimes=2).fit(seed=0)
>>> abs(res.state_space.loglikelihood(y) - res.llf) < 1e-8
True
>>> s_new, y_new = draw(300)
>>> dated = res.state_space.filter(y_new).filtered_prob.argmax(axis=1)
>>> bool(np.mean(dated == s_new[1:]) > 0.95)
True