|
|
High-Order Moment GMM Programs
- NOW IN
GAUSS, MATLAB, STATA, and SAS!!!!!!!!!!!!!!!!!
- This page
contains user-friendly versions of the programs that Tim
Erickson and I used in our series of papers on estimating the errors-in-variables
model using high-order moments. (JPE, 2000; Econometric Theory, 2002.)
- They are designed to handle
a mismeasured regressor in a cross section,
panel, or i.i.d. time series. In order for
them to work, the mismeasured regressor must
not be normally distributed. The method appears to have good
finite-sample properties in regressions containing Tobin's q, which by
definition cannot have a normal distribution, since it is bounded below
by zero. The estimators are not just for regressions containing Tobin's
q, though we have only done a few Monte Carlo
experiments along this line.
- I originally wrote the
programs in GAUSS, but I have also translated them into MATLAB. The two
versions give identical answers when applied to the same data set.
- The programs are really easy to use, and they run very quickly: seconds or minutes. You should be
able to use them by changing a few lines in the top of each program.
- I have provided a tutorial
below.
- Please download and read
the accompanying instructions before you attempt to use the programs.
- A number of researchers
have used our model identification test as a "pre-sampling"
test in order to identify samples on which our model is identified. We strongly discourage this practice! It is tantamount
to searching for samples on which the standard errors are low. If the
model is unidentified, you need to find another method.
- I offer no guarantee for
the programs, but they have been debugged thoroughly.
- If you are interested in
programs that handle two mismeasured
regressors, I have a Monte Carlo
program in Gauss and a program than handles data in Matlab.
Neither of these programs is particularly user friendly, but the Matlab program has been used with some success by
folks other than me.
- Zipped files with
the Stata and SAS routines can be found below.
Please
follow the Matlab instructions. I have no idea
how to use Stata and my knowledge of the IML
language in SAS is about 20 years old. These programs produce identical
results on the original data from the Erickson and Whited (2000, JPE)
paper.
Tutorial
Instructions
Identification Test Program in Gauss
One Mismeasured
Regressor Program in Gauss
Identification Test Program in Matlab
One Mismeasured
Regressor Program in Matlab
One Mismeasured
Regressor Program in Stata's MATA language
Identification Test Program in Stata’s MATA language
One Mismeasured
Regressor Program in SAS
Identification Test Program in SAS
|