Motivation

Using the gmm command

Several linear examples
Nonlinear GMM
Summary

GMM Estimation in Stata

Econometrics I

Ricardo Mora
Department of Economics
Universidad Carlos III de Madrid
Master in Industrial Economics and Markets

Ricardo Mora GMM estimation

 Motivation
Using the gmm command
Several linear examples
Nonlinear GMM
Summary

Outline

1 Motivation

2 Using the gmm command

3 Several linear examples

4 Nonlinear GMM

Ricardo Mora GMM estimation

 Motivation
Using the gmm command
Several linear examples
Nonlinear GMM
Summary

Motivation

Ricardo Mora GMM estimation

 Motivation
Using the gmm command
Several linear examples
Nonlinear GMM
Summary

Stata and GMM

Stata can compute the GMM estimators for some linear

models:
1 regression with exogenous instruments using ivregress
(ivreg, ivreg2 for Stata 9)
demand function using 2SLS
ivreg 2sls q demand_shiftrs (p =supply_shiftrs ), vce(robust)
demand function using GMM
ivreg gmm q demand_shiftrs (p =supply_shiftrs )

with heteroskedasticity, the GMM estimator will be more

ecient than the 2SLS estimator
2 xtabond for dynamic panel data

since Stata 11, it is possible to obtain GMM estimates of

non-linear models using the gmm command
Ricardo Mora GMM estimation
 Motivation
Using the gmm command
Several linear examples
Nonlinear GMM
Summary

Using the gmm command

Ricardo Mora GMM estimation

 Motivation
Using the gmm command
Several linear examples
Nonlinear GMM
Summary

Using the gmm command

the command gmm estimates parameters by GMM

you can specify the moment conditions as substitutable
expressions
a substitutable expression in Stata is like any mathematical
expression, except that the parameters of the model are
enclosed in braces {}
alternatively, you may use command program to create a
program that you can use as an argument
we are going to focus on examples using substitutable
expressions

Ricardo Mora GMM estimation

 Motivation
Using the gmm command
Several linear examples
Nonlinear GMM
Summary

The syntax of gmm with instruments

If E [ze (b)] = 0 where z is a q × 1 vector of instrumental

variables and e (b) is a scalar function of the data and the
parameters beta
gmm (e (b)) , instruments(z_list) options

by default, it computes the two-step estimator with identity

matrix in the rst step
use onestep option to get the one-step estimator and igmm to
get the iterative estimator
use vce(robust) to get sandwich standard errors
use winitial(wmtype) and wmatrix(witype) to change
weight-matrix computations
gmm admits if, in, and weight qualiers
Ricardo Mora GMM estimation
 Motivation
Using the gmm command
Several linear examples
Nonlinear GMM
Summary

More general moment conditions (1)

in some applications we cannot write the moment conditions

as the product of a residual and a list of instruments
suppose you have two general moment conditions

E [h1 (b)] = 0
E [h2 (b)] = 0
gmm (h1 (b)) (h2 (b)), igmm

computes the iterative GMM estimator imposing in the sample

the two moment conditions

Ricardo Mora GMM estimation

 Motivation
Using the gmm command
Several linear examples
Nonlinear GMM
Summary

More general moment conditions (2)

instruments may vary with error terms

E [z1 e1 (b)] = 0
E [z2 e2 (b)] = 0
gmm (e1 (b)) (e2 (b)) , instruments(1:z1 ) instruments(2:z2 ) nolog

this computes the twostep GMM estimator without adding

information on the rst step
you can use this syntax to estimate supply and demand
functions simultaneously

Ricardo Mora GMM estimation

 Motivation
Using the gmm command
Several linear examples
Nonlinear GMM
Summary

Several linear examples

Ricardo Mora GMM estimation

 Motivation
Using the gmm command
Several linear examples
Nonlinear GMM
Summary

Linear regresssion

Assume that
depvar = β0 + β1 x 1 + β2 x 2 + v
so that E [v |x 1, x 2] = 0
Then
E [(depvar − (β0 + β1 x 1 + β2 x 2))] = 0
E [x 1 (depvar − (β0 + β1 x 1 + β2 x 2))] = 0
E [x 2 (depvar − (β0 + β1 x 1 + β2 x 2))] = 0
The gmm command:
gmm (depvar-x1*{b1}-x2*{b2}-{b3}), instruments(x1 x2) nolog

equivalently (names of the variables will be displayed in the

output) and simpler to write:
gmm (depvar-{xb:x1 x2}-{b0}), instruments(x1 x2) nolog
Ricardo Mora GMM estimation
 Motivation
Using the gmm command
Several linear examples
Nonlinear GMM
Summary

Estimating OLS with gmm command

. regress mpg gear_ratio turn, r

Linear regression Number of obs = 74

F( 2, 71) = 47.92
Prob > F = 0.0000
R-squared = 0.5483
Root \M\SE = 3.9429

------------------------------------------------------------------------------
| Robust
mpg | Coef. \Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
gear_ratio | 3.032884 1.533061 1.98 0.052 -.023954 6.089721
turn | -.7330502 .1204386 -6.09 0.000 -.9731979 -.4929025
_cons | 41.21801 8.5723 4.81 0.000 24.12533 58.31069
------------------------------------------------------------------------------

gmm (mpg - {b1}*gear_ratio - {b2}*turn - {b0}), instruments(gear_ratio turn) nolog

Final G\M\M criterion Q(b) = 3.48e-32

G\M\M estimation

Number of parameters = 3
Number of moments = 3
Initial weight matrix: Unadjusted Number of obs = 74
G\M\M weight matrix: Robust

------------------------------------------------------------------------------
| Robust
| Coef. \Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
/b1 | 3.032884 1.501664 2.02 0.043 .0896757 5.976092
/b2 | -.7330502 .117972 -6.21 0.000 -.9642711 -.5018293
/b0 | 41.21801 8.396739 4.91 0.000 24.76071 57.67532
------------------------------------------------------------------------------
Instruments for equation 1: gear_ratio turn _cons

Ricardo Mora GMM estimation

 Motivation
Using the gmm command
Several linear examples
Nonlinear GMM
Summary

2SLS and gmm

gmm (depvar-{xb:x1 x2}-{b0}), instruments(z1 z2 z3) onestep

ivregress 2sls mpg gear_ratio (turn = weight length headroom)

Instrumental variables (2\SL\S) regression Number of obs = 74

Wald chi2(2) = 90.94
Prob > chi2 = 0.0000
R-squared = 0.4656
Root \M\SE = 4.2007

------------------------------------------------------------------------------
mpg | Coef. \Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
turn | -1.246426 .2012157 -6.19 0.000 -1.640801 -.8520502
gear_ratio | -.3146499 1.697806 -0.19 0.853 -3.642288 3.012988
_cons | 71.66502 12.3775 5.79 0.000 47.40556 95.92447
------------------------------------------------------------------------------
Instrumented: turn
Instruments: gear_ratio weight length headroom

. gmm (mpg - {b1}*turn - {b2}*gear_ratio - {b0}), instruments(gear_ratio weight length headroom) onestep

\Step 1
Iteration 0: G\M\M criterion Q(b) = 475.42283
Iteration 1: G\M\M criterion Q(b) = .16100633
Iteration 2: G\M\M criterion Q(b) = .16100633

G\M\M estimation

Number of parameters = 3
Number of moments = 5
Initial weight matrix: Unadjusted Number of obs = 74

------------------------------------------------------------------------------
| Robust
| Coef. \Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
/b1 | -1.246426 .1970566 -6.33 0.000 -1.632649 -.8602019
/b2 | -.3146499 1.863079 -0.17 0.866 -3.966217 3.336917
/b0 | 71.66502 12.68722 5.65 0.000 46.79853 96.53151
------------------------------------------------------------------------------
Instruments for equation 1: gear_ratio weight length headroom _cons

Ricardo Mora GMM estimation

 Motivation
Using the gmm command
Several linear examples
Nonlinear GMM
Summary

Linear GMM and gmm

gmm (depvar-{xb:x1 x2}-{b0}), instruments(z1 z2 z3) wmatrix(robust)

. ivregress gmm mpg gear_ratio (turn = weight length headroom)

Instrumental variables (G\M\M) regression Number of obs = 74

Wald chi2(2) = 97.83
Prob > chi2 = 0.0000
R-squared = 0.4769
G\M\M weight matrix: Robust Root \M\SE = 4.1559

------------------------------------------------------------------------------
| Robust
mpg | Coef. \Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
turn | -1.208549 .1882903 -6.42 0.000 -1.577591 -.8395071
gear_ratio | .130328 1.75499 0.07 0.941 -3.30939 3.570046
_cons | 68.89218 12.05955 5.71 0.000 45.25589 92.52847
------------------------------------------------------------------------------
Instrumented: turn
Instruments: gear_ratio weight length headroom

. gmm (mpg - {b1}*turn - {b2}*gear_ratio - {b0}), instruments(gear_ratio weight length headroom) wmatrix(robust) nolog

Final G\M\M criterion Q(b) = .0074119

G\M\M estimation

Number of parameters = 3
Number of moments = 5
Initial weight matrix: Unadjusted Number of obs = 74
G\M\M weight matrix: Robust

------------------------------------------------------------------------------
| Robust
| Coef. \Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
/b1 | -1.208549 .1882903 -6.42 0.000 -1.577591 -.8395071
/b2 | .130328 1.75499 0.07 0.941 -3.30939 3.570046
/b0 | 68.89218 12.05955 5.71 0.000 45.25589 92.52847
------------------------------------------------------------------------------
Instruments for equation 1: gear_ratio weight length headroom _cons

Ricardo Mora GMM estimation

 Motivation
Using the gmm command
Several linear examples
Nonlinear GMM
Summary

Nonlinear GMM

Ricardo Mora GMM estimation

 Motivation
Using the gmm command
Several linear examples
Nonlinear GMM
Summary

Exponential regression with exogenous regressors

exponential regression models are frequently encountered in

applied work
they can be used as alternatives to linear regression models on
log-transformed dependent variables
when the dependent variable represents a discrete count
variable, they are also known as Poisson regression models

E [y |x ] = exp (x β + β0 )

Moment conditions: E [x (y − exp (x β + β0 ))] = 0

this is equivalent to E [x (y − exp (x β ) + γ)] = 0

gmm (depvar-exp({xb:x1 x2})+{b0}), instruments(x1 x2) wmatrix(robust)

Ricardo Mora GMM estimation

 Motivation
Using the gmm command
Several linear examples
Nonlinear GMM
Summary

IV Poisson regression and others

suppose now E [z (y − exp (x β ) + γ)] = 0

gmm (depvar-exp({xb:x1 x2})+{b0}), instruments(z1 z2 z3) wmatrix(robust)

the structure of the moment conditions for some models is too

complicated for the syntax used thus far
you should in these cases use the moment-evaluator program
syntax (see help gmm)

Ricardo Mora GMM estimation

 Motivation
Using the gmm command
Several linear examples
Nonlinear GMM
Summary

Summary

Stata can compute the GMM estimators for some linear

models:
1 regression with exogenous instruments using ivregress
(ivreg,ivreg2 for Stata 9)
2 xtabond for dynamic panel data
since Stata 11, it is possible to obtain GMM estimates of
non-linear models using the gmm command

Ricardo Mora GMM estimation