With this version of the mex file, the overhead cost of mex call is too important.

specially useful when neighborhood_width is used

Added calls to matlab implementation of the perfect foresight solver in homotopic routine (extended path approach).

Changed the test files conformably to the latest commits related to the (stochastic) extended path approach.

Note that, even for small models, there is a (very) substantial gain in using the use_dll option.

Added an option to skip the test on the number of periods over which the perfect foresight models are solved.

Oups! I forgot to add this routine in commit ac54f7f5. This routine provides a matlab implementation of the perfect foresight model solver.

Changed the default value of options_.ep.fp. Test the stability of the solution only for the first period (when the value of periods is increased).

Try first Ferhat's code and if it fails try the matlab's implementation of the the perfect foresight model solver.

Added comments and the possibility to use a matlab implementation of the perfect foresight model solver.

accuracy didn't pass the test in Octave and could generate problems
accross machine or Matlab versions)

Dynare (no function calls it).

*  Removed the  necessity (for  the  user) to  run stoch_simul  bebore
executing the exetended path routine (when options_.init>0).

*  The  value  of  options_.ep.init  defines the  mix  (used  for  the
initialization of  the perfect foresight solver)  between the previous
perfect foresight  solution and  the path obtained  with an  order one
perturbation approach.

* Removed timing related statements.

*  Changed homotopy set-up  for stochastic  extended path:  add future
multivariate innovations one by one.

* Endogeneously increase step_length in the homotopy routine.

* Removed homotopy_2 related code.

It appeared that the Lagrange  multiplier simulated by EP is positive,
but the series simulated by SEP are most of the time negative.

Prevent homotopy routine to enter in the second loop (doi not (re)start the homotopy from weight=0).

for computation of likelihood with presample

diffuse Kalman filter was used before

with correlated measurement errors

(./optimal_policy/nk_ramsey.mod ./ep/linear.mod)