MatELike (version 0.1.4)
John Zedlewski ([email protected])

DESCRIPTION:

MatELike is a Matlab package to estimate Empirical Likelihood (EL)
models by direct maximization. It includes tools to formulate problems
easily and use automatic differentiation to efficiently estimate
models with nonlinear moment conditions. It can use the robust
optimization code Ipopt (which must be installed separately) or the
included ZIPsolver interior-point optimization method to solve the
very large maximization problems that arise in EL.

INSTALLATION:

To install MatELike, unzip the matelike archive and place the
resulting melroot directory wherever you like to store your Matlab
code. Add the subdirectory "melroot/matelike" to your Matlab
path with a command like:

   addpath c:/matlab/melroot/matelike

where c:/matlab is replaced with the base directory in which matelike
is stored. Edit the file matelike/elLoad.m so that the variable
BASEPATH points to the directory where MatELike is installed. Now the
package is installed, but you must load it into Matlab's memory in
each session before you can use it. Load the package with the command:

   elLoad

you should see some output and then "Loaded matElike verson 0.1.4
package successfully". To avoid repeating these steps every time you
run Matlab, you can add those commands to your 'startup.m' file
(search the Matlab help docs for startup.m for information on how to
do this).

Usage:

More documentation coming soon! Individual functions have usage
documentation, so look at the source or type 'help elSetup' and
'help elSolve' in Matlab to view the docs.

For now, look at 'simpleDemoEL.m' (under 'matelike/demos') to see a
trivial usage. (Currently, simpleDemo makes little sense as a model,
but it illustrates the code usage.) testDP.m is a more complicated
dynamic panel data model, based on Imbens 2002 JBES.

The only real steps are:

- write a function that computes the moment conditions for each observation
- call 'elike = elSetup(...)', specifying the number of observations, etc.
- call 'res = elSolve(elike, ...)' to solve the model

For linear two-stage-least-squares models, the 'elLinearIV' function
is much more efficient than using 'elSolve' directly. To use
'elLinearIV' you can skip the 'elSetup' step: just put your
instruments (Z) and structural variables (X) in to matrices and pass
them into 'elLinearIV'. The moment condition is that 'y - X*theta' is
orthogonal to 'Z'.


REQUIREMENTS:

Tested on Matlab 7.4 and 7.6. For large models, a 64-bit platform is
very helpful. The Ipopt code is tested with Ipopt 3.3.4 with the
Matlab mex bridge.

LICENSE:

MatELike includes Intlab, which is copyright by Siegfried Rump @
TUHH. That package's license states that:

"INTLAB is free for private use and for purely academic purposes
provided proper reference is given acknowledging that the software
package INTLAB has been developed by Siegfried M. Rump at Hamburg
University of Technology, Germany. For any other use of INTLAB a
license is required." (See also: http://www.ti3.tu-harburg.de/rump/intlab/)

All other components are copyright 2008 by John Zedlewski and are
distributed under the same license: Free use for academic and/or
personal purposes is permitted as long as this authorship is
acknowledged.