In fact, I like to find the Hessian for a likelihood object to compute the robust variance. It seems that EViews does not provide one.
I try to compute the numerical derivative of gradient vector. Thus, the gradient at b+h (b is the estimated parameter and h is some step size) should be calculated for Hessian. However, the procedure makegrads generates gradients at b, and the view grads also at b once Logl object has been estimated.
LogL: How to get the gradient other than at estimated value
Moderators: EViews Gareth, EViews Jason, EViews Moderator, EViews Matt
-
- EViews Developer
- Posts: 2672
- Joined: Wed Oct 15, 2008 9:17 am
Re: LogL: How to get the gradient other than at estimated value
The gradient view will compute the gradients at the current coefficients if the logl is not estimated, and at the estimated coefficients if it is. Unfortunately, the makegrads proc will only work with an estimated logl.
Obviously, if you have analytic gradients in your logl specification, you can simply use them to set up your numeric seconds. If you don't, then the easiest thing to do is to us your objective functions and compute them directly.
Obviously, if you have analytic gradients in your logl specification, you can simply use them to set up your numeric seconds. If you don't, then the easiest thing to do is to us your objective functions and compute them directly.
Who is online
Users browsing this forum: No registered users and 52 guests