Identification Problem
• By the identification problem we mean whether
numerical estimates of the parameters of a
structural equation can be obtained from the
estimated reduced-form coefficients.
• If this can be done, we say the equation is
identified.
• If this cannot be done, we say the equation is
unidentified or underidentified.
1
Identification Problem
• An identified equation may be either
exactly (or fully or just) identified or
overidentified.
• It is said to be exactly identified if UNIQUE
numerical values of the structural
parameters can be obtained.
• It is said to be overidentified if MORE
THAN ONE numberical value can be
obatained for some of the parameters of
the structural equations. 2
Identification Problem
• The identification problem arises because
different sets of structural coefficients may
be compatible with the same set of data.
• A given reduced-form equation may be
compatible with different structural
equations or different hypotheses
(models).
3
Types of Identification
• Identified - can use the reduced form parameters
to obtain the structural parameters
• Unidentified - not possible to estimate all of the
structural parameters from the reduced form,
however some parameters can be estimated.
• Exactly identified - unique parameter values exist.
• Over-identified - more than one value of the
structural parameters can be obtained from the
reduced form parameters.
4
One way to test identification
• Find reduced form equations and if
Structural parameters > Reduced form parameters UNDER
Structural parameters = Reduced form parameters EXACT
Structural parameters < Reduced form parameters OVER
PROBLEM:
In case of large structural equations, difficult to
construct reduced form equations
5
Underidentification
Demand function : Qtd 0 1 Pt 1t 1 0
Supply function : Qts 0 1 Pt 2t 1 0
Equilibrium condition : Qtd Qts
Setting the two equations equal, we get
0 1 Pt 1t 0 1 Pt 2t or
0 0 2t 1t
Pt 10 1t where 10 and 1t
1 1 1 1
0 1 11t
Qt 20 2t where 20 1 0 and 2t 1 2t
1 1 1 1
6
Underidentification
0 0 1 0 0 1
10 and 20
1 1 1 1
Our structural equations contain 4 parameters ,
but our reduced form parameter equations are
only two. Therefore, we are faced with 2 equations
and 4 unknowns. We do not have enough equations
to solve for the structural parameters in terms of the
reduced form parameters .
Suppose we multiply (1) by λ (0≤λ≤1) and (2) by (1- λ)
λQt = λα0+ λα1Pt+ λμ1t --------------------(3)
(1- λ)Qt =(1- λ)β0+ (1- λ)β1Pt+(1- λ)μ2t ---------(4)
Adding (3) and (4)
Qt ={λ α0 + (1- λ)β0}+{ λ α1 + (1- λ)β1}Pt+ {λμ1t+(1- λ)μ2t}
or Qt = γ0 + γ1Pt + εt -----------------------(5)
The “Bogus” equation (5) is observationally indistinguishable from either (1) or (2) because
7
they involve the regression of Q on P.
Just Identified
Demand : QtD 0 1 Pt 2 I t 1t
Supply : Qts 0 1 Pt 2Wt 2t
Our reduced form equations are
Pt 10 11 I t 12Wt 1t
Qt 20 21 I t 22Wt 2t where
0 0 2 2
10 , 11 , 12
1 1 1 1 1 1
1 0 0 1 2 1 1 2
20 , 21 , 22
1 1 1 1 1 1
8
Just Identified
21
1 , 1 22
11 12
2 11 (1 1 )
2 12 (1 1 )
21
0 20 1 10 20
11 10
0 0 10 (1 1 )
We get unique solutions for all structural parameters
because we have six equations in six unknowns. Each
equation is just identified .
Consider the Bougas equation λ (Equ-1)+(1- λ)(Equ-2)
Qt ={λ α0 + (1- λ)β0}+{ λ α1 + (1- λ)β1}Pt+ λ α2It+ (1- λ)β2Wt + {λμ1t+(1- λ)μ2t}
or Qt = γ0 + γ1Pt + γ2It + γ3Wt + εt
The “Bougas ” will be observationally distinguishable from the
demand function (1) as well as from the supply function (2), because
the demand function doesn’t contain Wt and the supply function 9
doesn’t contain It.
Over-identified
Demand : QtD 0 1Pt 2 I t 3 PSt 1t
Supply : Qts 0 1Pt 2Wt 2 t
Our reduced form equations are
Pt 10 11I t 12Wt 13PSt 1t
Qt 20 21I t 22Wt 23PSt 2 t where
0 0 2 2 3
10 , 11 , 12 , 13
1 1 1 1 1 1 1 1
1 0 0 1 2 1 1 2 3 1
20 , 21 , 22 , 23
1 1 1 1 1 1 1 1
10
Note
Over-identified
α 2β1
-
π 21 α1 -β1
β1 = = =β1
π11 α2
-
α1 -β1
also
α 3β1
-
π 23 α1 -β1
β1 = = =β1
π13 α3
-
α1 -β1
The solution for β1 is not unique. The supply equation
is overidentified. The demand equation is just identified.
We have 8 equations and 7 unknowns. We have too much
information.
11
Conditions for identification
• Order Condition (Necessary)
• Rank Condition (Necessary & Sufficient)
– For an equation to be identified both the conditions
must hold
– If an equation is identified according to ORDER
condition may be unidentified according to RANK
condition and hence equation is unidentified
– RANK condition only distinguish between identified
and unidentified
– ORDER condition distinguish between Exact and over
identified equations
– The order and Rank conditions of identification are
applicable to models which are linear in variables and
parameters 12
Conditions for identification
• Order Condition (Necessary)
• Rank Condition (Necessary & Sufficient)
– For an equation to be identified both the conditions
must hold
– If an equation is identified according to ORDER
condition may be unidentified according to RANK
condition and hence equation is unidentified
– RANK condition only distinguish between identified
and unidentified
– ORDER condition distinguish between Exact and over
identified equations
– The order and Rank conditions of identification are
applicable to models which are linear in variables and
parameters 13
Order Condition for Identification
For an equation to be identified
Definition-1
The number of predetermined variables
excluded from the equation must be greater
than or equal to the number of included
endogenous variables minus 1 (i.e m-1).
Definition-2
The number of variables (Endo. &
Predetermined) excluded from an equation
must be greater than or equal to the number of
endogenous variables in the system minus 141
(i.e M-1).
Order Condition for Identification
M Number of endogenous variables in the system
m Number of endogenous variables in the equation
K Number of predetermined variables in the system
k Number of predetermined variables in the equation
K-k < m-1 unidentified
K-k = m-1 Exactly identified
K-k > m-1 Over identified OR
Excluded variables < M-1 Unidentified
Excluded variables = M-1 Exact Identified
Excluded variables > M-1 Over Identified
15
Qtd P
0 1 t
I u 1t : demand
2 t
Qts
P u : supply
0 1 t 2 t
Endo. P , Q Predetermined I
K=1 (I) M=2 (Q , P)
[Link] k K-k m-1 Identified
1 1 0 2-1=1 Under
2 0 1 2-1=1 Exact
[Link] Excluded M-1 Identified
Variables
1 0 2-1=1 Under
2 1 2-1=1 Exact
16
10 Y 11 X
Y1t 12 2 t 1t 12
X 2 t u 1t : Income
Y2t Y u 2t : Money
20 21 1t
Endo. Y1 , Y2 Predetermined X1 , X2
K=2 (X1 , X2) M=2 (Y1, Y2)
[Link] k K-k m-1 Identified
1 2 0 2-1=1 Under
2 0 2 2-1=1 Over
[Link] Excluded M-1 Identified
Variables
1 0 2-1=1 Under
2 2 2-1=1 Over
17
Rank Condition
In a model containing M equations in M endogen
variables an equation is identified iff one non-
zero determined of order (M-1)x(M-1) can be
constructed from the coefficients of the variables
in other equation of the system
18
Procedure for Rank Condition
1. Write the system in tabular form
2. Strike out the coefficients of the under
consideration equation
3. Strike out the columns corresponding to those
variables that appear under consideration
equation
4. If we can find a non zero determined from the
coefficients of the variables excluded from
under consideration equation but included in
other equations, then the equation under
consideration is identified
19
Example(1)
Q P I
Q
t
d
0
1Pt 2I t u 1t : demand
Qts Pt u 2t : supply
0 1
Endo. P , Q Predetermined I
1 1 α1 α2
By ORDER condition
2 1 β1 0
Demand equation UNDER IDENTIFIED
Supply equation EXACT IDENTIFIED
By RANK condition
2 0
Supply equation is identified by RANK condition
20
Example(2)
Y1i 10 12Y2i 11 X 1i 12 X 2i 1i Income Equation
Y2i 20 21Y1i 2i Money Equation Y1 Y2 X1 X2
Endo.Y1, Y 2 Predetermined X1, X2
1 1 β12 λ11 λ12
By ORDER condition
Income equation UNDER IDENTIFIED
2 β21 1 0 0
Money equation OVER IDENTIFIED
By RANK condition
11 0
Money equation is identified by RANK condition
21
Example(3) Y1 Y2 Y3 X1 X2 X3
Y1 10 11 X 1 13 X 3 U 1
Y2 20 23Y 3 21 X 1 22 X 2 U 2
11 0 0 λ11 0 λ13
Y3 30 33Y 1 31 X 1 33 X 3 U 3
ORDER CONDITION
20 1 β23 λ21 λ22 0
1 Excluded variables=3 > M-1=2 OVER
2 Excluded variables=2 = M-1=2 EXACT
3 β33 0 1 λ31 0 λ33
3 Excluded variables=2 = M-1=2 EXACT
RANK CONDITION
As equation—(3) is unidentified by
1 β 23
Equ-1 =1 0 identified Rank Condition so Equation—(3) is
0 1 unidentified
1 13
Equ-2 =33 13 33 0 identified
33 33
0 0
Equ-3 = 0 unidentified
1 22
22
Key Points About Identification
• Must consider the identification problem before
the estimation problem.
• Can’t use ILS or 2SLS unless an equation is
identified.
• We call an equation that is not identified,
unidentified or underidentified.
• If any equation in the system is unidentified
then the system as a whole is unidentified
• Equations can be overidentified. 2SLS can be
used on overidentified equations.
23
Key Points About Identification
• Must consider the identification problem before
the estimation problem.
• Can’t use ILS or 2SLS unless an equation is
identified.
• We call an equation that is not identified,
unidentified or underidentified.
• If any equation in the system is unidentified
then the system as a whole is unidentified
• Equations can be overidentified. 2SLS can be
used on overidentified equations.
24
