Non-Linear Estimation Options |
The non-linear estimation instructions (BOXJENK, DDV, GARCH, ITERATE, NLLS, NLSYSTEM, LDV, LGT, PRBIT, ESMOOTH, MAXIMIZE, FIND, DLM, and CVMODEL) all share several common options. We describe these below (defaults appear in brackets). Note that the CVCRIT and SUBITERATIONS options are also available on the NLPAR instruction. If you use NLPAR to set values for these options, those values will be the defaults for all subsequent non-linear estimation instructions in that session (you can still override the NLPAR settings by using the options directly on the estimation instructions).
Common Options
TITLE="title for output" [none]
This option allows you to supply your own title to label the resulting output.
PARMSET=PARMSET to use [default internal]
This option selects the parameter set to be estimated. RATS maintains a single unnamed parameter set which is the one used for estimation if you don’t use PARMSET to specify a named set.
ITERATIONS=maximum number of iterations [100]
This sets the maximum number of iterations that the instruction will perform in an attempt to converge to a solution.
TRACE/[NOTRACE]
This displays intermediate results from each iteration.
CVCRIT=convergence limit [0.00001]
Sets convergence limit, used to determine whether or not the process has converged.
SUBITERATIONS=subiteration limit [30]
Limits the number of new coefficient vectors examined once a direction has been chosen. Increase this if you get a warning message indicating that the subiterations limit was exceeded. If you still hit the limit no matter how high it is set, you need to try some different initial values for the parameters, or consider altering the model.
HESSIAN=initial guess for inverse Hessian (METHOD=BFGS only)
This allows you to supply initial values for the inverse Hessian. This option is only available on instructions that offer the BFGS method, and only when using that method. Without it, the estimation will start with a BHHH matrix, but with diagonal elements only.
Copyright © 2026 Thomas A. Doan