Part 9: GMM Estimation [ 1/57]

Econometric Analysis of Panel Data

William Greene
Department of Economics
Stern School of Business

Part 9: GMM Estimation [ 2/57]

http://people.stern.nyu.edu/wgreene/CumulantInstruments-Racicot-AE(2014)_46(10).pdf

Part 9: GMM Estimation [ 3/57]

Part 9: GMM Estimation [ 4/57]

The NYU
No Action
Letter

Part 9: GMM Estimation [ 5/57]

Part 9: GMM Estimation [ 6/57]

GMM Estimation for One Equation

1 N
1 N

g
( )= i1z i (yi x
i1z i i
i )
N
N
1 Ni1i2zizi
1 Ni1ei2zi zi
Asy.Var[g
( )]
, estimated with

N
N
N
N

based on 2SLS residuals e.i The GMM estimator then minimizes

1 e z z
1 N

i ) '
i1zi (yi x

N
N
N

N
2
i1 i i i

1 N

z
(y

x
)
i .
N i1 i i

Part 9: GMM Estimation [ 7/57]

GMM for a System of Equations

Simultaneous equations
Labor supply
h h h
hours = f(wage, gh ) h= x
w
wage = f(hours, gw ) w= x

Product market equilibrium

Quantity demanded = f(Price,...)
Price = f(market demand,...)
General format:

y1= x
1 1 1
2
y2= x
...
M
yM= x

Part 9: GMM Estimation [ 8/57]

SUR Model with Endogenous

RHS Variables
SUR System

x G ] 0
y1= x
1 1 1 , E[1 |x 1 ,x 2 ,...
2
y2= x
...
G
yM= x

2 ,...
G ,...

Each equation has a set of Lg Kg instruments, zg

Each equation can be fit by 2SLS, IV, GMM, as before.

Part 9: GMM Estimation [ 9/57]

GMM for the System - Notation

Index: i = 1,...,N for individuals
g = 1,...,G for equations (this would be t=1,...T for a panel)
Data matrices: G rows,
yi1
xi1 0


y
0 xi2
i2

yi
, Xi
...
... ...

yiG
0 0
K1 K2
yi X+
i
i

0

1

... 0

, = 2 , i =
...

... ...

iG
... x
G

... KG columns
...

i1

i2

...

iG

Part 9: GMM Estimation [ 10/57]

Instruments
zi1 0

0 zi2

Zi
... ...

0 0
L1

L2

... 0

... 0
, G rows (1 for each equation)

... ...

... xiG
... LG columns

Such that

E zi1i1

zi1,1i1
0

zi1,2i1
0

for L1 instrumental variables

...
...

i1,L1 i1
0

Same for zi2i2, ...

Part 9: GMM Estimation [ 11/57]

Moment Equations
zi1i1
z
i ] E i2 i2
E[Z
...

ziGiG

0 L1 rows
L rows
0 2
, for observation i
...
...

0 LG rows
Summing over i gives the orthogonality condition,

1 N

i E
i=1Z

zi1i1
0 L1

L
z

0 2
1 N i2 i2

i=1

...
...
N

0 LG
ziGiG

rows
rows
...
rows

Part 9: GMM Estimation [ 12/57]

Estimation-1

ig yig x
ig

For one equation,

= the minimizer of

M
m=1

(1/N)

N
i=1 ig,m

)
(yi xi

gg(g )'gg(g )

Leads to 2SLS
For all equations at the same time
= the minimizer of

G
g=1

g
'g
) g(
g( g

G
g=1

M
m=1

(1/N)

N
i=1 ig,m

(yi xi )

If the gs are all different, still equation by equation 2SLS

Part 9: GMM Estimation [ 13/57]

Estimation-2
Assuming ig are all uncorrelated, equation by equation GMM
1
ig
qg Ni1zig(yig x
N

)
g '

1 e z z 1 N

z
(y

) .

i1 ig
ig
ig g

N

N
2
i1 ig ig ig

For the system,

q = Gg=1qg
Cases to consider:
(1) Coefficient vectors have elements in common or are
restricted
(2) Disturbances are correlated.

Part 9: GMM Estimation [ 14/57]

Estimation-3
Combining GMM criteria

G
g 1

1 N
1 N
1
1 N 2

z
(y

x
)
'

z
z
iz1

ig g
i 1 ig ig ig

N i1 ig ig

N
N

Ni1zi1 (yi1 x
i1 1 ) Ni1i12 zi1zi1

0
N

N
2

z
(y

x
)
0

i 1 i2
i2
i2 2
i 1 i2zi2 zi2

q
'

...
...
...
N

z
(y

x
)
0
0
iG G
i1 iG iG

Ni1zi1 (yi1 x

i1 1 )
N

z
(y

x
)
i2 2
i1 i2 i2

...
N

z
(y

x
)
iG G
i1 iG iG

(y

ig
ig
ig g
)

...
0

...
0

...
...

... Ni1iG
ziGziG

Part 9: GMM Estimation [ 15/57]

Estimation-4
If disturbances are correlated across equations,
i1 1 )
Ni1zi1(yi1 x

z
(y

x
)
1
i1 i2
i2
i2 2
q
'
N

...

z
(y

x
)
i1 iG
iG
iG G

... Ni1i1iGzi1ziG

... Ni1i2iGzi2ziG

...
...

2
... Ni1iG
ziGziG

i1 1 )
Ni1zi1(yi1 x

N
i2 2 )
1 i1zi2 (yi2 x

N
...

iG G )
Ni1ziG (yiG x

Ni1i12 zi1zi1
Ni1i1i2zi1zi2
N
2
Ni1i2
zi2zi2
1 i1i2i1zi2zi1
N2
...
...
Ni1iGi1ziGzi1 Ni1iGi1ziGzi1

Part 9: GMM Estimation [ 16/57]

Estimation-5
If disturbances are correlated across equations,
G
G
(1/ N)Niz1
ig g) W gh
q g1 h1 (1/ N)Ni1zig (yig x

gh = the gh block of the inverse matrix

where W
z z
z z
N
z z
1 i1i2i1zi2zi1
N2
...
...
Ni1iGi1ziGzi1 Ni1iGi1ziGzi1

N 2
i1 i1 i1 i1

N
i1 i1 i2 i1 i2
N 2
i1 i2 i2 i2

... z z

... z z

...
...

N 2
... i1iGziGziG

N
i1 i1 iG i1 iG
N
i1 i2 iG i2 iG

ih

(yih x ih
h )

Part 9: GMM Estimation [ 17/57]

The Panel Data Case

is the same in every equation.
The number of moment equations is T L
if each moment equation is L per period,
E[zit it ] 0,
If every disturbance at time t is also orthogonal
to every set of instruments in every other period, s,
Then
E[zit is ] 0, TL per period, for T periods, or T 2L
E.g., L=10 instruments, T=5 periods, K=5 parameters,
250 moment equations (!) for fitting 5 parameters.

Part 9: GMM Estimation [ 18/57]

Hausman and Taylor FE/RE Model

it
yit x1

x2 it

i
z1

i
z2

it ui

E[ui | x1it ,z1i ] 0

E[ui | x2it ,z2i ] 0 OLS and GLS are inconsistent
Var[ui | x1it , x2it ,z1i ,z2i ] u2
E[it | x1it , x2it ,z1i ,z2i ]=0
Var[it | x1it , x2it ,z1i ,z2i ]=2
Cov[it ,ui | x1it , x2it ,z1i ,z2i ]=0
Var[it ui | x1it , x2it ,z1i ,z2i ]=2 u2
Cov[it ui , is ui | x1it , x2it ,z1i ,z2i ]=u2

Part 9: GMM Estimation [ 19/57]

Useful Result: LSDV is an IV Estimator

y X D
= X w
1
plim
Xw 0, so X is endogenous. Correlated with w because of D.
NT
MDX X* = x's in group mean deviations.
1
1
1
1
X*'w
X*' D+ =
XMDD XMD
X0 XMD

NT
NT
NT
NT
1
1
1

XMD, so plim
X*'w plim
XMD 0
NT
NT
NT
1
1
plim
X*'X plim
X'MDX within groups sums of squares 0.
NT
NT
X* is a valid instrument.
plim b*=plim X* ' X

X
* y =

Part 9: GMM Estimation [ 20/57]

Hausman and Taylor

it
yit x1

x2 it

i
z1

i
z2

it ui

Deviations from group means removes all time invariant variables

yit yi ( x1it - x1i )'1 ( x2it - x2i )'2 it
Implication: 1 , 2 are consistently estimated by LSDV.
( x1it - x1i ) = MDX1 = K1 instrumental variables
( x2it - x2i ) = MDX2 = K2 instrumental variables
z1i
?

= L1 instrumental variables (uncorrelated with u)

= L2 instrumental variables (where do we get them?)

H&T: x1i = (I -MD )X1 = K1 additional instrumental variables. Needs K1 L2.

Part 9: GMM Estimation [ 21/57]

H&Ts FGLS Estimator

(1) LSDV estimates of 1 , 2 , 2
(2) (e* )' = (e1 , e1 ,..., e1),(e2 , e2 ,..., e2 ),...,(eN, eN,..., eN )
(Ni=1Ti observations).
zi1 zi2
z z
i2
Ti rows, repeat invariant variables
Zi * i1
M M
L1 L2 columns

zi1 zi2
zi1 xi1,1

Ti rows, repeat zi1 , time varying xi1,t

z
x
i1,2
i1
Wi

L1 K1 columns
M
M

zi1 xi1,T i

Part 9: GMM Estimation [ 22/57]

H&Ts FGLS Estimator (cont.)

(2 cont.) IV regression of e * on Z* with instruments
Wi consistently

estimates

and

(3) With fixed T, residual variance in (2) estimates u2 2 / T

With unbalanced panel, it estimates u2 2 / T or something
resembling this. (1) provided an estimate of 2 so use the two
to obtain estimates of u2 and 2 . For each group, compute
2
2
2
i 1
/ (
Ti
u)

(4) Transform

[xit1 , xit2 ,zi1 ,zi2 ] to

W*
i = [xit1 , xit2 ,zi1 ,zi2 ] - i[xi1 , xi2 ,zi1 ,zi2 ]
and

yit to yit * = yit - iyi.

Part 9: GMM Estimation [ 23/57]

H&Ts 4 STEP IV Estimator

Instrumental Variables Vi
(x1it - x1i ) = K1 instrumental variables
(x2it - x2i ) = K2 instrumental variables
z1i

= L1 instrumental variables (uncorrelated with u)

x1i

= K1 additional instrumental variables.

Now do 2SLS of y * on W * with instruments V to estimate

all parameters. I.e.,
* W*
)-1W
* y * .
[1 , 2 , 1 , 2 ]=(W

Part 9: GMM Estimation [ 24/57]

Part 9: GMM Estimation [ 25/57]

Part 9: GMM Estimation [ 26/57]

Part 9: GMM Estimation [ 27/57]

Part 9: GMM Estimation [ 28/57]

Part 9: GMM Estimation [ 29/57]

Dynamic (Linear) Panel

Data (DPD) Models

Application
Bias in Conventional Estimation
Development of Consistent Estimators
Efficient GMM Estimators

Part 9: GMM Estimation [ 30/57]

Dynamic Linear Model

Balestra-Nerlove (1966), 36 States, 11 Years
Demand for Natural Gas
Structure
New Demand: G*i,t Gi,t (1 )Gi,t1
Demand Function G*i,t 1 2Pi,t 3Ni,t 4Ni,t 5Yi,t 6 Yi,t i,t
G=gas demand
N = population
P = price
Y = per capita income
Reduced Form
Gi,t 1 2Pi,t 3Ni,t 4Ni,t 5Yi,t 6 Yi,t 7Gi,t1 i i,t

Part 9: GMM Estimation [ 31/57]

A General DPD model

i,t yi,t1 ci i,t
yi,t x
E[i,t | Xi ,ci ] 0
2
E[i,t
| Xi , ci ] 2 , E[i,ti,s | Xi , ci ] 0 if t s.

E[ci | Xi ] g( Xi )
No correlation across individuals

Part 9: GMM Estimation [ 32/57]

OLS and GLS are inconsistent

i,t yi,t1 ci i,t
yi,t x
Cov[yi,t1 ,(ci i,t )]
2c Cov[yi,t2 ,(ci i,t )]
If T were large and -1<<1,
2c
this would approach
1
I mplication : Both OLS and GLS are
inconsistent.

Part 9: GMM Estimation [ 33/57]

LSDV is Inconsistent
[(Steven) Nickell Bias]
yi,t yi ( xi,t xi )'+(yi,t1 yi ) (i,t i )
2 (T 1) T T
Cov[(yi,t1 yi ),(i,t i )] 2
T
(1 )2
Large when T is moderate or small.
Proportional bias for conventional T (5 - 15), is
on the order of 15% - 60% .

Part 9: GMM Estimation [ 34/57]

Anderson Hsiao IV Estimator

Base on first differences
yi,t yi,t1 (xi,t xi,t1)'+(yi,t1 yi,t2 ) (i,t i,t1 )
Instrumental variables
yi,3 yi,2 (xi,3 xi,2 )'+(yi,2 yi,1) (i,3 i,2 )
Can use yi1
yi,4 yi,3 ( xi,4 xi,3 )'+(yi,3 yi,2 ) (i,4 i,3 )
Can use yi2 or (yi,2 yi,1)
And so on.
Levels or lagged differences?
Levels allow you to use more data
Asymptotic variance of the estimator is smaller with levels.

Part 9: GMM Estimation [ 35/57]

Arellano and Bond Estimator - 1

Base on first differences
yi,t yi,t1 ( xi,t xi,t1 )'+(yi,t1 yi,t2 ) (i,t i,t1 )
Instrumental variables
yi,3 yi,2 ( xi,3 xi,2 )'+(yi,2 yi,1) (i,3 i,2 )
Can use yi1
yi,4 yi,3 ( xi,4 xi,3 )'+(yi,3 yi,2 ) (i,4 i,3 )
Can use yi,1 and yi2
yi,5 yi,4 ( xi,5 xi,4 )'+(yi,4 yi,3) (i,5 i,4 )
Can use yi,1 and yi2 and yi,3

Part 9: GMM Estimation [ 36/57]

Arellano and Bond Estimator - 2

More instrumental variables - Predetermined X
yi,3 yi,2 ( xi,3 xi,2 )'+(yi,2 yi,1 ) (i,3 i,2 )
Can use yi1 and xi,1 , xi,2
yi,4 yi,3 ( xi,4 xi,3 )'+(yi,3 yi,2 ) (i,4 i,3 )
Can use yi,1 , yi2 , xi,1 , xi,2 , xi,3
yi,5 yi,4 ( xi,5 xi,4 )'+(yi,4 yi,3 ) (i,5 i,4 )
Can use yi,1 , yi2 , yi,3 , xi,1 , xi,2 , xi,3 , xi,4

Part 9: GMM Estimation [ 37/57]

Arellano and Bond Estimator - 3

Even more instrumental variables - Strictly exogenous X
yi,3 yi,2 ( xi,3 xi,2 )'+(yi,2 yi,1 ) (i,3 i,2 )
Can use yi1 and xi,1 , xi,2 ,..., xi, T (all periods)
yi,4 yi,3 ( xi,4 xi,3 )'+(yi,3 yi,2 ) (i,4 i, 3 )
Can use yi,1 , yi2 , xi,1 , xi,2 ,..., xi, T
yi,5 yi,4 ( xi,5 xi,4 )'+(yi,4 yi,3 ) (i,5 i,4 )
Can use yi,1 , yi2 , yi,3 , xi,1 , xi,2 ,..., xi, T
The number of potential instruments is huge.
These define the rows of Zi . These can be used for
simple instrumental variable estimation.

Part 9: GMM Estimation [ 38/57]

Instrumental Variables
Predetermined variables
, xi,2

yi,1 , xi,1

Zi

...

, xi,2
, xi,3
...
yi,1 , yi,2 , xi,1

...
0

...
...

, xi,2
,...xi, T 1
... yi,1 , yi,2 ,..., yi, T 2 , xi,1

...
0

(T rows)

Strictly Exogenous variables

, xi,2
,...xi, T 1
yi,1 , xi,1

Zi

...
0

...

, xi,2
,...xi, T 1 ...
yi,1 , yi,2 , xi,1

0
...
0

...
...

, xi,2
,...xi, T 1
... yi,1 , yi,2 ,..., yi, T 2 , xi,1

(T rows)

Part 9: GMM Estimation [ 39/57]

Simple IV Estimation

= N XZ N ZZ 1 N ZX

i=1 i i
i=1 i i
i=1 i i

N XZ N ZZ 1 N Zy
i=1 i i
i=1 i i
i=1 i i

This is two stage least squares.

N
N
]=
Est.Asy.Var[
i=1 XiZi i=1ZiZi

N
i=1

ZiXi

(y y )]2
Ni1 tT 3[( yi, t yi, t 1 ) ( xi, t xi, t 1 )'
i, t
i, t 1
Ni1 (Ti 2)

Note that this variance estimator understates the true asymptotic

variance because observations are autocorrelated for one period.
(yi,t yi, t 1 ) ... (i, t i, t 1 ) ... vi, t
Cov[vi,t , vi,t 1 ] [vi,t , vi, t 1 ] 2 (0 for longer lags, and leads)
Use a "White" robust estimator

]= N XZ N Zv

Est.Asy.Var[
i=1 i i vi Zi
i=1 i i

N
i=1

ZiXi

Part 9: GMM Estimation [ 40/57]

Arellano/Bond
First Difference Formulation
yi, t 1 it
yit x
it
Parameters : =
[ , ]
The data

xi3
yi3

xi4
i4

yi
, Xi
...
M

iT i
xiT
K

yi,2 yi,1

yi,3 yi,2
, Ti -2 rows

yi,T yi, T1
1
columns

Part 9: GMM Estimation [ 41/57]

Arellano/Bond - GLS
yi,t yi,t1 ( xi,t xi,t1)'+(yi,t1 yi,t2 ) (i,t i )
i,3 i,2

Cov

2 1 0 ...
1 2 1 ...
0 1 2 ...

i,4 i,3

i,5 i,4

...

i,T i,T 1

...

...

0
0
0 2 i

1 ... 1
... 1 2

Part 9: GMM Estimation [ 42/57]

Arellano/Bond GLS Estimator

Ni=1ZiXi

Ni=1Z iy i

= N XZ N Z Z

i=1 i i i
i=1 i i
N XZ N Z
i iZ i
i=1
i
i
i=1

Z X
= XZ ZZ

X Z Z Z Z y

Part 9: GMM Estimation [ 43/57]

GMM Estimator
yi,t xi,t'+yi,t1 i,t
We make no assumptions about the disturbance. In first differences
yi,t yi,t1 ( xi,t xi,t1 )'+(yi,t1 yi,t2 ) (i,t i,t1)
(1) Two stage least squares

= N XZ N ZZ

i=1 i i
i=1 i i

Zi Xi

N
i=1

N
i=1

XiZi Ni=1ZiZi

1 Ni=1Ziv

(2) Form the weighting matrix for GMM: W

v
Z
i i
i
N2

The criterion for GMM estimation is

1
-1
1 N
q= Ni=1 viZ
W
i=1Zi vi
i

1 Ni=1ZiXi
N

GMM = i=1 XiZi W

N
i=1

1 Ni=1Ziy i
XiZi W

1 Ni=1Zi Xi
N
Est.Asy.Var[
GMM ] i=1 XiZi W

Zi y i

N
i=1

Part 9: GMM Estimation [ 44/57]

Arellano/Bond/Bovers Formulation
Start with H&T
it
yit x1

x2 it

i
z1

i
z2

it ui

Instrumental variables for period t

( x1it - x1i ) = K1 instrumental variables
( x2it - x2i ) = K2 instrumental variables
z1i

= L1 instrumental variables (uncorrelated with u)

x1i

= K1 additional instrumental variables. K1 L2.

Let vit it ui
Let zit [( x1it - x1i )',( x2it - x2i )',z1i , x1']
Then E[zit vit ] 0
We formulate this for the Ti observations in group i.

Part 9: GMM Estimation [ 45/57]

Arellano/Bond/Bovers Formulation
Dynamic Model
it
yit yi,t1+x1

x2 it

i
z1

i
z2

it ui

Parameters : = [, 1
, 2
, 1
, 2']
The data
yi,2
yi,1 x1i2 x2i2 z1i z2 i

y
y
x1
x2
z1
z2
i3
i3
i
i,3
i,2
i
yi
, Xi
, Ti -1 rows

yi,T i
yi,T-1 x1iTi x2iTi z1i z2 i
1 K1
K2
L1 L2 columns

Part 9: GMM Estimation [ 46/57]

Arellano/Bond/Bovers Formulation
it
yit yi,t1 + x1

x2 it

i
z1

i
z2

it ui

Instrumental variables for period t as developed above

Let zit [yi,1 , yi,2 ,...,( x1it - x1i )',( x2it - x2i )',z1i , x1']
Combine H&T treatment with DPD GMM estimator.
Instrumental variable creation is based on group mean
deviation rather than first differences.

Part 9: GMM Estimation [ 47/57]

Arellano/Bond/Bovers Formulation
zit [yi,1 yi ,..., yi,t1 yi ,( x1it - x1i )',( x2it - x2i )',z1i , x1']
Then E[zit vit ] 0
We formulate this for the last Ti -1 observations in group i.
(yi,1 , x1i2 ,x2i2 ,z1i )

(0,0,0)

(0,0,0)

...

(0,0,0)

(0,0,0)

(0,0,0)

...

(0,0,0)

Zi

(yi,1 , yi,2 , x1i3 ,x2i3 ,z1i )

(0,0,0)
(0,0,0)

(0,0,0)

(0,0,0)

(yi,1 , yi,2 , yi,3 , x1i4 ,x2i,4 ,z1i ) ...

(0,0,0)

(0,0,0)

(0,0,0)

(0,0,0)

(0,0,0)

... (yi,1 ,...,yi,T-2 , x1i,(T-1) ,x2i,(T-1) ,z1i )

(0,0,0)

(0,0,0)

(0,0,0)

...

H'i MiD,(T -1)

i

(0,0,0)

(yi,1 ,...,yi,T-1 ,z1, x1i )

1/(Ti 1)

1/(Ti 1)
, where MiD,(T -1) MDi without the last column.
i

...

1/(Ti 1)

These blocks may contain all

previous exogenous variables, or all
exogenous variables for all periods.
This may contain the all periods of data on x1 rather than just the group mean. (Amemiya and
MaCurdy).

Part 9: GMM Estimation [ 48/57]

Arellano/Bond/Bovers Formulation
For unbalanced panels the number of
columns for Zi varies. Given the form of
Zi, the number of columns depends on Ti.
We need all Zi to have the same number
of columns. For matrices with less
columns than the largest one, extra
columns of zeros are added.

Part 9: GMM Estimation [ 49/57]

Arellano/Bond/Bovers Formulation
The covariance matrix defines the model:
i=2 I - Classical (pooled) regression model (no effects)
i=2 I + u2ii' - Random effects model
i= A positive definite TxT
i
i matrix - GR model

Part 9: GMM Estimation [ 50/57]

Arellano/Bond/Bover Estimator

= N X H Z N Z H
H Z

i=1
i
i
i
i=1
i
i
i
i
i

N
N
H Z
i Hi

X
H
Z

Z
i=1
i
i
i
i=1
i
i
i

Two step (GMM) estimation

Zi Hi Xi

Ni=1 Zi Hi yi

N
i=1

= I . Compute residuals v
i yi Xi
(1) Use
i
= 1 N Hv

Then HH
i i i
i1 i iv iHi
N
.
(2) Recompute

]= N X H Z N Z H
H Z
Est.Asy.Var[
i=1
i
i
i
i=1
i
i
i
i
i

Zi Hi Xi

N
i=1

Part 9: GMM Estimation [ 51/57]

GMM Criterion
The GMM criterion which produces this estimator is

H Z
i i Zi Hi
iHZ
q= v
i
i
i
N
i1

N
i=1

i i i
ZHv

Post estimation, use this as 2[DF] to test the overidentifying

restrictions. The degrees of freedom is the total number of
moment conditions (columns in Z) minus the number of
parameters in .

Part 9: GMM Estimation [ 52/57]

Application: Maquiladora

Part 9: GMM Estimation [ 53/57]

Maquiladora

Part 9: GMM Estimation [ 54/57]

Part 9: GMM Estimation [ 55/57]

Side Issue
How does y(t) = 1.220175 y(t-1) - 0.262198 y(t-2) + a behave?
y(t) = 1.220175 y(t-1) + a
is obviously explosive.
1.220175 0.262198
How to tell: A =

1
0

Smallest (possibly complex) root must be greater than 1.0.

Part 9: GMM Estimation [ 56/57]

Postscript

There is no theoretical guidance on the

instrument set
There is no theoretical guidance on the form of
the covariance matrix
There is no theoretical guidance on the
number of lags at any level of the model
There is no theoretical guidance on the form of
the exogeneity and it is not testable.
Results vary wildly with small variations in the
assumptions.

Part 9: GMM Estimation [ 57/57]

Ahn and Schmidt

it
yit yi,t1+x1

x2 it

i
z1

i
z2

it ui

There are (huge numbers of) additional moments.

i,0+ i,0
(1) Initial condition, yi,0 x
E[i,t yi,0 ] 0 implies T more estimating equations
(2) Uncorrelatedness with differences,
E[yis (it i,t1 )] 0, t 2,..., T, s 0,..., T 2 is
T(T-1)/2 conditions
(3) (Nonlinear)
E[iT (it i,t1 )] 0 implies T-2 restrictions.
And so on.
Even moderately sized models embed potentially
thousands of such estimating equations for usually
very small numbers (say 5 or 10) parameters.
How much efficiency can be gained? Is there a cost?