README
==========================
This program computes max likelihood estimates using the approximation
techniques described in Durham and Gallant (99). For the CIR model,
the exact MLE is available for comparison.

At each iteration, the optimizer outputs time in seconds, followed by the
parameter values, likelihood, and gradient.


The CIR model is set as the default, but it is easy to substitute other
models. Model definitions should go in a module (see file sde_models.f90
for samples). This module should be included by a 'use' statement in the
main computational modules, 'sde_like' and 'sde_gen'.


The following settings can be used for the estimation procedure (see paper for descriptions):

         Valid values for sampler:
            pedersen
            bridge
            mod
            ecs

         Valid values for subdensity:
            exact
            first_order
            elerian
            kessler
            shoji
            kessler

         Valid random schemes:
          0: iid
          1: antithetic variates
          2: normalized variance
          3: binomial tree  (requires nsim = 2^(naux-1))
          4: draws from binomial tree

         extrap:
          T: uses extrapolation.
          F: no extrapolation


The structure of the program is:
      mle:          main calling program
      sde_like:     computes likelihood
      sde_gen:      generates synthetic data
      sde_models:   model definitions and parameters

The following general purpose modules are used:
      gbd_opt:      optimizer
      gbd_lib:      miscellaneous utilities
      gbd_ran:      random number generators
      gbd_stat:     some statistical routines
      gbd_mat:      matrix package (calls Lapack)
      gbd_util:     assorted utilities based on public domain modules from Numerical Recipes.
      random:       Alan Miller's random number generation module

The following external libraries are required.
   Slatec   : get from www.netlib.org
   Lapack   : get from www.netlib.org
