SVResult.quasi_state_space

property SVResult.quasi_state_space: _LinearGaussianStateSpace

The Harvey-Ruiz-Shephard linearization on the linear substrate.

Filters log((y - c)**2), not y; its loglikelihood on the linearized estimation sample reproduces llf when the method was "qml".

Example

>>> import numpy as np
>>> rng = np.random.default_rng(0)
>>> y = rng.standard_normal(200)
>>> res = SV(y, mean="zero").fit()
>>> linearized = np.log((y - res.mean) ** 2 + 1e-6)
>>> bool(abs(res.quasi_state_space.loglikelihood(linearized) - res.llf) < 1e-8)
True