I had a question regarding MaxLik 5.0 Package regarding how the Hessian is calculated when a procedure is provided by the user to calculate the analytical gradient (for the calculation of variance of the estimates). From the documentation (page #11) I understand that, in this case the Hessian is calculated by the outer product of the score (or the first derivative of the observation). I just wanted to know whether this is indeed the case, i.e. GAUSS does not calculate the second derivative of the log-likelihood function (for the calculation of the variance of the estimates) when a procedure is provided to calculate the analytical derivative.
Thank you very much and I look forward to your response.
When there is a user-provided gradient procedure, the iterations computes a Hessian using a numerical gradient procedure with the gradient procedure as a function. That is, it computes a gradient of the gradient.
What is used for the calculations of the covariance matrix of the parameters depends on the setting of the global, _max_CovPar.
Thanks much! It makes sense.