statsmodels.sandbox.regression.gmm.IVGMM.fititer#

IVGMM.fititer(start, maxiter=2, start_invweights=None, weights_method='cov', wargs=(), optim_method='bfgs', optim_args=None)#

iterative estimation with updating of optimal weighting matrix

stopping criteria are maxiter or change in parameter estimate less than self.epsilon_iter, with default 1e-6.

Parameters:
startndarray

starting value for parameters

maxiterint

maximum number of iterations

start_invweightsarray (nmoms, nmoms)

initial inverse weighting matrix; if None, then the identity matrix is used

weights_method{‘cov’, …}

method to use to estimate the optimal weighting matrix, see calc_weightmatrix for details

wargstuple or dict

required and optional arguments for weights_method, see calc_weightmatrix for details

optim_methodstr, default is ‘bfgs’

numerical optimization method used in fitgmm for each iteration. Currently not all optimizers that are available in LikelihoodModels are connected.

optim_argsdict

keyword arguments for the numerical optimizer.

Returns:
paramsndarray

estimated parameters

weightsndarray

optimal weighting matrix calculated with final parameter estimates