Can't set attribute in property class in markov regime switching model

Question

I am trying to set my initial parameters in order to run a markov regime switching model but I always get the following error:

AttributeError: can't set attribute

My code is the following:

from statsmodels.tsa.regime_switching.markov_autoregression import MarkovAutoregression as mark_auto

mod = mark_auto(endog = data.dlgnp, k_regimes = 2, order=1, switching_variance= False, switching_exog= False, switching_trend= False)
mod.k_params
mod.param_names
regression.start_params = np.array([0.4,0.4,1,1])

The source code can be found here but the part specially concerning my problem is the following:

@property
def start_params(self):
    """
    (array) Starting parameters for maximum likelihood estimation.
    """
    # Inherited parameters
    params = markov_switching.MarkovSwitching.start_params.fget(self)

    # OLS for starting parameters
    endog = self.endog.copy()
    if self._k_exog > 0 and self.order > 0:
        exog = np.c_[self.exog, self.exog_ar]
    elif self._k_exog > 0:
        exog = self.exog
    elif self.order > 0:
        exog = self.exog_ar

    if self._k_exog > 0 or self.order > 0:
        beta = np.dot(np.linalg.pinv(exog), endog)
        variance = np.var(endog - np.dot(exog, beta))
    else:
        variance = np.var(endog)

I have also tried np.r_but it didn't help. I am running my code on python 2.7.15 and the strangest thing is that I remember having the code working last time I ran it. Any help would be extremely appreciated.

Answer 1

The start_params property just provides the default starting parameters used when calling the fit function - you don't have to set it yourself.

If you do want to set specific starting parameters, you would do that when calling fit, e.g.:

res = mod.fit(start_params=np.array([0.4,0.4,1,1]))