Nature of SEMs and Simultaneous Bias

Denitions of Some Concepts

The Identication Problem
Endogeneity and Estimation of SEMs

Econometrics II
2. Simultaneous Equations Models (SEMs)

Alemu L.(PhD)

Addis Ababa University

April, 2023

Alemu L.(PhD) Econometrics II

 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts
The Identication Problem
Endogeneity and Estimation of SEMs

Table of Contents

1 Nature of SEMs and Simultaneous Bias

2 Denitions of Some Concepts
Endogenous and Exogenous Variables
Structural versus Reduced Form Models
3 The Identication Problem
Necessary and Sucient Conditions for Identication
4 Endogeneity and Estimation of SEMs
Instrumental Variables (IV)
Two-Stage Least Squares (2SLS)

Alemu L.(PhD) Econometrics II

 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts
The Identication Problem
Endogeneity and Estimation of SEMs

References

Verbeek, M. (2008). A Guide to Modern Econometrics (3nd

ed.). John Wiley and Sons Ltd.
Wooldridge (2009). Introductory Econometrics (2nd ed).
Damodar Gujarati, (2001), Basic Econometrics, Mc-Graw Hill,
Tokyo.
Greene, William, H .(2000). Econometric Analysis (Fourth
Edition), Prentice Hall International.
Other Econometrics text books

Alemu L.(PhD) Econometrics II

 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts
The Identication Problem
Endogeneity and Estimation of SEMs

Nature of Simultaneous Equation models

So far, we have been focusing exclusively with the problems and

estimations of a single equation regression models.
Thus, the cause-and-eect relationship between the dependent
(eect) and independent variable (cause) is unidirectional.
But, there are situations where a unidirectional causation is not
meaningful.
This occurs if, for instance, Y (dependent variable) is not only
function of X's (explanatory variables) but also all or some of
the X's are, in turn, determined by Y.
There is, therefore, a two-way ow of inuence between Y and
(some of) the X's which in turn makes the distinction between
dependent and independent variables a little doubtful.
Under such circumstances, we need to consider more than one
regression equations; one for each interdependent variables.

Alemu L.(PhD) Econometrics II

 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts
The Identication Problem
Endogeneity and Estimation of SEMs

Simultaneous Bias

A system describing the joint dependence of variables is called a

simultaneous equations model.
The number of equations in such models is equal to the number of
jointly dependent or endogenous variables involved.
Unlike the single equation models, in simultaneous equation
models, it is not usually possible to estimate a single equation of
the model without taking into account the information provided by
other equation of the system.
If one applies OLS to estimate the parameters of each equation
disregarding other equations, the estimates obtained are not only
biased but also inconsistent; i.e. even if the sample size increases
indenitely, the estimators do not converge to their true values.
The bias arising from application of OLS on the SEM is called
simultaneity bias or simultaneous equation bias.
Alemu L.(PhD) Econometrics II
 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Endogenous and Exogenous Variables
The Identication Problem Structural versus Reduced Form Models
Endogeneity and Estimation of SEMs

Endogenous and Exogenous Variables

Consider the following SEM

yt = β0 + β1 xt + β2 yt−1 + β3 zt + εt
xt = α0 + α1 yt + α2 xt−1 + t
In simultaneous equation models variables are classied as
1 Endogenous: are variables whose values are determined by the
economic model or within the system (yt , xt ).
2 Predetermined: are those variables whose values are
determined outside the model. They can be divided into two
categories.
(a) Current Exogenous Variables (z t )
(b) Lagged exogenous variable (zt−1 ) and lagged endogenous
variable (yt−1 , xt−1 ).

Alemu L.(PhD) Econometrics II

 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Endogenous and Exogenous Variables
The Identication Problem Structural versus Reduced Form Models
Endogeneity and Estimation of SEMs

Structural versus Reduced Form Models

1 A structural model
Describes the complete structure of the relationships among
the economic variables.
Expresses the endogenous variables as a function of
endogenous, exogenous variables and disturbances (random)
variables.
Its parameter measures only the direct eect of each
explanatory variable on the dependent variable.
For instance, a change in consumption aects the investment
indirectly and is not considered in the consumption function.
2 Reduced form of the model (Recursive Model)
Expresses the endogenous variables as a function of
predetermined variables and error term only.
Its parameter measures both the direct and indirect eects of
each explanatory variable on the dependent variable.
Alemu L.(PhD) Econometrics II
 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Endogenous and Exogenous Variables
The Identication Problem Structural versus Reduced Form Models
Endogeneity and Estimation of SEMs

Structural versus Reduced Form Models

Question
Condider the following structural Model
C = β0 + β1 Y + ε
(1)
Y =C +Z

where C is the consumption spending, Z non-consumption

spending and Y is the household income.
1 Identify the endogenous and exogenous variables
2 Derive the reduced model and identify the direct and indirect
eect of income on consumption

Alemu L.(PhD) Econometrics II

 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Endogenous and Exogenous Variables
The Identication Problem Structural versus Reduced Form Models
Endogeneity and Estimation of SEMs

Structural versus Reduced Form Models

1 Endogenous variables: C and Y; Exogenous variable: Z
2 Deriving RFM is just expressing endogenous variables (C and
Y) in terms of exogenous variable (Z).
C = β0 + β1 Y + ε ⇒ C = β0 + β1 (C + Z ) + ε

β0 β1 ε
C= + Z+ (2)
1 − β1 1 − β1 1 − β1
Substituting again (2) into Y = C + Z , we get;
β0 β1 ε
Y = + Z+ (3)
1 − β1 1 − β1 1 − β1
Equation (2) and (3) are called the reduced form of the
structural model of the above.
Alemu L.(PhD) Econometrics II
 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Endogenous and Exogenous Variables
The Identication Problem Structural versus Reduced Form Models
Endogeneity and Estimation of SEMs

Structural versus Reduced Form Models

Parameters of the reduced form measure the total eect

(direct and indirect) of a change in exogenous variables on the
endogenous variable.
For instance, in the above reduced form equation(2), 1−β β1
1
measures the total eect of a unit change in the
non-consumption expenditure on consumption.
This total eect is β1 , the direct eect, times 1−β
β1
1
, the
indirect eect.
Alemu L.(PhD) Econometrics II
 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Necessary and Sucient Conditions for Identication
The Identication Problem
Endogeneity and Estimation of SEMs

The Identication Problem

By the identication problem we mean whether numerical

estimates of the parameters of a structural equation can be
obtained from the estimated reduced-form coecients.
If this can be done, the particular equation is identied.
If this cannot be done, then we say that the equation under
consideration is unidentied, or under identied.
Note that the identication problem is a mathematical (as
opposed to statistical) problem associated with simultaneous
equation systems.
It is concerned with the equation of the possibility or
impossibility of obtaining meaningful estimates of the
structural parameters.

Alemu L.(PhD) Econometrics II

 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Necessary and Sucient Conditions for Identication
The Identication Problem
Endogeneity and Estimation of SEMs

Necessary Conditions for Identication

The Order Condition for Identication

For an equation to be identied the total number of variables
(endogenous and exogenous) excluded from it must be equal
to or greater than the number of endogenous variables in the
model less one.
Given that in a complete model the number of endogenous
variables is equal to the number of equations of the model, the
order condition for identication is sometimes stated in the
following equivalent form.
For an equation to be identied the total number of variables
excluded from it but included in other equations must be at least as
great as the number of equations of the system less one.

Alemu L.(PhD) Econometrics II

 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Necessary and Sucient Conditions for Identication
The Identication Problem
Endogeneity and Estimation of SEMs

Necessary Conditions for Identication

Let:
G = total number of equations (= total number of
endogenous variables)
K= number of total variables in the model (endogenous and
predetermined)
M= number of variables, endogenous and exogenous, included
in a particular equation.
(K − M) ≥ (G − 1)
[Exclude [total number (4)
≥
variable] of equations − 1]

Alemu L.(PhD) Econometrics II

 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Necessary and Sucient Conditions for Identication
The Identication Problem
Endogeneity and Estimation of SEMs

Necessary Conditions for Identication

For example, if a system contains 10 equations with 15

variables, ten endogenous and ve exogenous, an equation
containing 11 variables is not identied, while another
containing 5 variables is identied.
(a) For the rst equation we have: G = 10 K = 15 M = 11
Order condition:
(K − M) ≥ (G − 1)
; the order condition is not satised.
(15 − 11) ≥ (10 − 1)
(b) For the second equation we have: G = 10 K = 15 M = 5
Order condition:
(K − M) ≥ (G − 1)
; that is, the order condition is satised.
(15 − 5) ≥ (10 − 1)
The order condition is necessary for a relation to be identied,
but it is not sucient, that is, it may be fullled in any
particular equation and yet the relation may not be identied.
Alemu L.(PhD) Econometrics II
 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Necessary and Sucient Conditions for Identication
The Identication Problem
Endogeneity and Estimation of SEMs

The Rank Condition for Identication

The rank condition states that, in a system of G equations,

an equation is identied if and only if it is possible to construct
at least one non-zero determinant of order (G-1) from the
coecients of the variables excluded from that particular
equation but contained in the other equations of the model.
Step for rank condition for identication:
1 Write the parameters of all the equations in a separate table
2 Strike out the row of coecients of the equation which is
being examined for identication.
3 Strike out the columns in which a non-zero coecient of the
equation being examined appears.
4 Form determinant(s) of order (G-1) and examine their value.
5 Use the order condition to see whether the equation is exactly
identied or overidentied.

Alemu L.(PhD) Econometrics II

 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Necessary and Sucient Conditions for Identication
The Identication Problem
Endogeneity and Estimation of SEMs

Example:The Rank Condition for Identication

Check the rank conditio for the following SEM
y1 = 3y2 − 3x1 + x2 + ε1
y2 = y3 + x3 + ε2
y3 = y1 − y2 − 2x3 + ε3
1. Write the parameters of all the equations in a separate
table.
−y1 + 3y2 + 0y3 − 3x1 + x2 + 0x3 + ε1 = 0
0y1 − y2 + y3 + 0x1 + 0x2 + x3 + ε2 = 0
y1 − y2 − y3 + 0x1 + 0x2 − 2x3 + ε3 = 0
Ignoring the ε the table of the parameters of the model is:

Alemu L.(PhD) Econometrics II

 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Necessary and Sucient Conditions for Identication
The Identication Problem
Endogeneity and Estimation of SEMs

Example:The Rank Condition for Identication

2. Strike out the row of coecients of the equation.
For example, if we want to examine the identiability of the
second equation of the model we strike out the second row of
the table of coecients.
3. Strike out the columns in which a non-zero coecient.
For example, if we are examining for identication the second
equation of the system, we will strike out the second, third and
the sixth columns of the above table, thus obtaining the
following tables.

Alemu L.(PhD) Econometrics II

 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Necessary and Sucient Conditions for Identication
The Identication Problem
Endogeneity and Estimation of SEMs

Example:The Rank Condition for Identication

4. Form determinant(s) of order (G-1) and examine their value.

If at least one of these determinants is non-zero, the equation
is identied. If all the determinants of order (G-1) are zero, the
equation is underidentied. In the above example of
exploration of the identiability of the second structural
equation, we have three determinants of order
(G − 1) = 3 − 1 = 2. They are:

The symbol ∆ stands for `determinant'. We see that we can

form two non-zero determinants of order (G − 1) = 3 − 1 = 2;
hence the second equation of our system is identied.

Alemu L.(PhD) Econometrics II

 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Necessary and Sucient Conditions for Identication
The Identication Problem
Endogeneity and Estimation of SEMs

Example:The Rank Condition for Identication

5. Use the order condition.

In the case of the second equation we have:
G =3 K =6 M=3
Order condition:
(K − M) ≥ (G − 1)
; the order condition is satised.
(6 − 3) > (3 − 1) ⇒ 3 > 2
Rules for types of identication
1 If (K − M) > (G − 1), the equation is overidentied.
2 If (K − M) = (G − 1), the equation is exactly identied.
3 If (K − M) < (G − 1), the equation is underidentied.
Therefore, the second equation of the model is overidentied.

Alemu L.(PhD) Econometrics II

 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Instrumental Variables (IV)
The Identication Problem Two-Stage Least Squares (2SLS)
Endogeneity and Estimation of SEMs

Endogneity and Instrumental Variables

One of the assumptions of the classical linear regression

models is that the error term (ε) does not correlate with any
of the explanatory variables.
E x 0ε = 0

If this assumption is violated OLS estimates are biased and

inconsistent (endogeneity problem).
Endogeneity arises in one of the following ways:
1 Omitted variables (unobserved heterogeneity)
2 Measurement error in the data
3 Simultaneity
A model can have more than one source of endogeneity.

Alemu L.(PhD) Econometrics II

 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Instrumental Variables (IV)
The Identication Problem Two-Stage Least Squares (2SLS)
Endogeneity and Estimation of SEMs

Endogneity and Instrumental Variables

If an explanatory variable xi is correlated with the error term

for whatever reason, the variable is traditionally called
endogenous explanatory variable.
In strict sense, the term 'endogenous' refers to the case
where there is simultaneity.
That is E (x 0 ε) 6= 0, and hence x is endogenous.
Suppose we estimate a model:
y = β0 + β1 x + ε (5)
where E (x 0 ε) 6= 0 say for a reason that the unobserved
variable is correlated with x .

Alemu L.(PhD) Econometrics II

 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Instrumental Variables (IV)
The Identication Problem Two-Stage Least Squares (2SLS)
Endogeneity and Estimation of SEMs

Instrumental Variables (IV)

If we can nd a good proxy, we would estimate the model

using OLS.
Whenever there is no appropriate proxy variable, consistent
estimation of the parameters β0 and β1 requires a variable (z)
which satises two basic assumptions.
1 z is uncorrelated with ε: cov (z, ε) = 0 [instrument exogeneity ]
2 z is correlated with x : cov (z, x) 6= 0 [instrument relevance]
We call z an instrumental variable for x .
Instrument exogeneity means that z should be uncorrelated
with the other omitted variables (ε).
Instrument relevance means that z must be related to the
endogenous explanatory variable x .

Alemu L.(PhD) Econometrics II

 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Instrumental Variables (IV)
The Identication Problem Two-Stage Least Squares (2SLS)
Endogeneity and Estimation of SEMs

IVs:Testing for Instrument Relevance

We cannot generally hope to test the rst assumption: most

often, we must maintain by appealing to economic behavior.
The second condition which states that z is correlated with x
can be tested given the random sample from the population.

Alemu L.(PhD) Econometrics II

 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Instrumental Variables (IV)
The Identication Problem Two-Stage Least Squares (2SLS)
Endogeneity and Estimation of SEMs

Two-Stage Least Squares (2SLS)

In an equation with one endogenous variable, there may be
more than one exogenous variable that is excluded from
the structural model and may be correlated with the
endogenous variable (can serve as IVs).
The 2SLS is the method of using multiple instruments.
Consider again the linear model with one endogenous and one
exogenous variables:
y1 = β0 + β1 y2 + β2 z1 + ε (6)

Suppose two exogenous variables, z2 and z3 , are excluded from

the model.
The assumptions that the exogenous variables, z2 and z3 , are
excluded from the model and are uncorrelated with the error
term are known as exclusion restrictions.
Alemu L.(PhD) Econometrics II
 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Instrumental Variables (IV)
The Identication Problem Two-Stage Least Squares (2SLS)
Endogeneity and Estimation of SEMs

Two-Stage Least Squares (2SLS)

If the two variables are correlated with y2 , which is the
suspected endogenous variable, we use them as IVs.
Since z1 , z2 and z3 are uncorrelated with the error term, any
linear combination of these variables is also uncorrelated with
the error term and is a valid IV.
To nd the best IV, we choose the linear combination that is
most highly correlated with y2 .
This is given by the reduced form equation for y2 :
y2 = π0 + π1 z1 + π2 z2 + π3 z3 + v2 (7)
where E (v2 ) = 0; E (z1 , v2 ) = 0; E (z2 , v2 ) = 0; E (z3 , v2 ) = 0.
Then, the best IV for y2 is the linear combination of zj :
y2∗ = π0 + π1 z1 + π2 z2 + π3 z3 (8)

Alemu L.(PhD) Econometrics II

 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Instrumental Variables (IV)
The Identication Problem Two-Stage Least Squares (2SLS)
Endogeneity and Estimation of SEMs

Two-Stage Least Squares (2SLS)

Given sample data on zj , we can estimate the reduced form

equation by OLS:
1 We regress y2 on z1 , z2 and z3 and obtain the tted values:
ŷ2 = π̂1 z1 + π̂2 z2 + π̂3 z3 (9)

2 Then ŷ2 can be used as an IV for y2 .

Alemu L.(PhD) Econometrics II

 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Instrumental Variables (IV)
The Identication Problem Two-Stage Least Squares (2SLS)
Endogeneity and Estimation of SEMs

Two-Stage Least Squares (2SLS)

Solving the following three equations for estimating β0 , β1 and

β2 gives the IV estimators.
n  
yi 1 − β̂0 − β̂1 yi 2 − β̂2 zi 1 = 0
X

i=1

n  
zi 1 yi 1 − β̂0 − β̂1 yi 2 − β̂2 zi 1 = 0
X

i=1
n  
ŷi 2 yi 1 − β̂0 − β̂1 yi 2 − β̂2 zi 1 = 0 (10)
X

i=1

Alemu L.(PhD) Econometrics II

 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Instrumental Variables (IV)
The Identication Problem Two-Stage Least Squares (2SLS)
Endogeneity and Estimation of SEMs

Two-Stage Least Squares (2SLS)

With multiple instruments, the IV estimator using the linear

combination such as ŷi 2 is called two stage least squares
(2SLS) estimator.
We can, thus, obtain the 2SLS in two stages:
1 Stage 1: Regress y2 on z1 , z2 and z3 and obtain ŷ2 :

ŷ2 = π̂1 z1 + π̂2 z2 + π̂3 z3 (11)

2 Stage 2: Run OLS regression of y1 on ŷ2 and z1 .

In other words, 2SLS rst purges y2 of its correlation with ε
before doing the OLS regression of y1 on ŷ2 and z1 .

Alemu L.(PhD) Econometrics II

 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Instrumental Variables (IV)
The Identication Problem Two-Stage Least Squares (2SLS)
Endogeneity and Estimation of SEMs

Shortcomings of the IV Estimation

1 Absence of endogeniety: The IV estimates are never
unbiased even in the absence of endogeniety.
2 Weak instrument problem: The 2SLS estimates (IV
estimates in general) would be inconsistent if the
instrument (s) are weak even in large samples.
Consider the simple model:
y = β0 + β1 x + ε (12)

in which x is suspected to be correlated with ε and hence we

intend to use z as an instrument.
After some algebra with the help of plim, you will get
  
σε corr (z, ε)
plimβ̂1 = β1 + (13)
σx corr (z, x)
Alemu L.(PhD) Econometrics II
 Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Instrumental Variables (IV)
The Identication Problem Two-Stage Least Squares (2SLS)
Endogeneity and Estimation of SEMs

Shortcomings of the IV Estimation

In the last equation, if corr (z, ε) 6= 0, the inconsistency of the

IV estimator gets very large when the correlation between x
and z tends to zero [corr (z, x) → 0].
That means, in the case of weak instrument (weak correlation
between x and z ), a small correlation between x and ε can
cause severe inconsistency leading to severe nite sample bias.
In such cases, OLS can even be a choice.
In sum, OLS estimates outweighs the IV estimates when
1 There is no endogeneity (where IV estimates are biased).
2 Instruments for endogenous variables are weak (where IV
estimates are inconsistent).
BOTTOM LINE: It is, thus, really crucial to test for weak
instruments before we apply the IV method.
Alemu L.(PhD) Econometrics II
Nature of SEMs and Simultaneous Bias
Denitions of Some Concepts Instrumental Variables (IV)
The Identication Problem Two-Stage Least Squares (2SLS)
Endogeneity and Estimation of SEMs

Thanks

Alemu L.(PhD) Econometrics II