MFVARResult.latent_stderr#

property MFVARResult.latent_stderr: NDArray[float64][source]#

Pointwise standard errors of the latent path, (nobs, k).

The diagonal of a smoothed covariance reaches zero from below by a few units in the last place wherever a variable is directly observed, and an unguarded square root would return nan for exactly the entries the model knows best. The clip is not cosmetic.

Example

>>> import numpy as np
>>> A, S = np.array([[[0.5, 0.1], [0.0, 0.4]]]), np.eye(2)
>>> y = np.full((24, 2), np.nan)
>>> rng = np.random.default_rng(0)
>>> y[:, 0], y[2::3, 1] = rng.standard_normal(24), rng.standard_normal(8)
>>> se = MFVAR(A, S, kinds=["high", "flow"], period=3).smooth(y).latent_stderr
>>> se.shape, bool(se[:, 0].max() < 1e-6), bool(se[:, 1].min() > 0.1)
((24, 2), True, True)