MarkovSwitchingAR.update_coefficients#

static MarkovSwitchingAR.update_coefficients(target, lags, smoothed, sigma2, layout)[source]#

M-step coefficient update by responsibility-weighted GLS.

All regimes are solved in one stacked system so that non-switching blocks are estimated jointly across regimes rather than per regime.

Parameters:

  • target (NDArray[double]) – The effective sample.

  • lags (NDArray[double]) – Lagged levels.

  • smoothed (NDArray[double]) – Smoothed regime probabilities.

  • sigma2 (NDArray[double]) – Current per-regime variances.

  • layout (_ParameterLayout) – Column bookkeeping.

Return type:

tuple[NDArray[double], NDArray[double]]

Returns:

A tuple (intercepts, ar_params) of shapes (K,) and (K, p).