Hi, I wondered how in Gauss numerical derivatives are calculated for example in the context of nonlinear least-squares (NLS) regressions using for example the Newton-Raphson algorithm. Thanks!
The numerical derivatives in GAUSS are computed using the finite-difference method with the option for central, forward or backward differencing. Some of the packages have limited support for complex derivatives as well.
That is what many program languages do. But, where I can get a math representation of the method (the formula) that is implemented in Gauss (for NLS) with the default delta. In some programs delta = 4e-7. Is that the same in Gauss. Can the user change it up or down or it is fixed?
I don't think it is designed to be configurable, but it is really simple to change. Let me know what function you are calling to compute the nonlinear least-squares estimates and we can show you how to control it.
The model is an exponential conditional mean (ECM) model (i.e., exp(x_i'b))
I'm not asking about the details of the function you created to compute the nonlinear equations. I need to know the function that you are passing this function to. For example, are you using