You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
One reasonably simple improvement that could work quite well arma_order_select_ic checks models of increasing order but it doesn't use the parameters from the smaller model as start_params for the larger model.
one possible implementation
cache the parameters in a dictionary with keys (k_ar, k_ma) and use the first of the following that is available
(k_ar, k_ma - 1)
(k_ar - 1, 0)
(0, k_ma - 1)
handles pure AR models and the initial order selection path of an ARMA model
handles pure MA models.
This should both be more robust and faster, more robust if the initial model AR(1) (or AR(0)?) is easy too estimate without stationarity problems.
The text was updated successfully, but these errors were encountered:
if max_ar or max_ma are large, then there might be problems in the optimization.
There might also be other problems, for example
http://stackoverflow.com/questions/28056601/the-computed-initial-ar-coefficients-are-not-stationary
(I never run examples for this)
One reasonably simple improvement that could work quite well
arma_order_select_ic
checks models of increasing order but it doesn't use the parameters from the smaller model asstart_params
for the larger model.one possible implementation
cache the parameters in a dictionary with keys (k_ar, k_ma) and use the first of the following that is available
(k_ar, k_ma - 1)
(k_ar - 1, 0)
(0, k_ma - 1)
handles pure AR models and the initial order selection path of an ARMA model
handles pure MA models.
This should both be more robust and faster, more robust if the initial model AR(1) (or AR(0)?) is easy too estimate without stationarity problems.
The text was updated successfully, but these errors were encountered: