Chapter Two

Introduction to Simultaneous Equation models

✓ So far, we considered regressions involving single equation models.
❖ In such models, the dependent variable is expressed as a function of one
or more independent variables.
❖ In this chapter, we will consider models involving two or more equations.
❖ These models are known as simultaneous equation models(SEM).
❖ A simultaneous equation is an equation system which has more than
one equation and more than one variables or unknowns.
❖ Usually, the number of equations is equals the number of variables.
❖ Otherwise, it would be difficult to find a unique solution or may not be
able to find a solution at all.
2.1. The Nature of Simultaneous Equation Models
❖ All discussions we have made so far involve single dependent
variable - called most of the time Y - and one or more explanatory
variables, usually called Xs.
❖ In such models the emphasis was on estimating and/or predicting
the average value of the dependent variable(Y) conditional upon
the fixed values of the explanatory variables (X’s).
❖ In such models it is, thus, assumed that Xs cause/determine Y, or Y
is caused/determined by Xs.
❖ For example, if you are given two variables consumption (Y) and
income (X), and asked about the cause-effect relationship, you may
reasonably respond that income determines consumption or X
causes Y.
❖ In any regression modeling, generally an equation is considered to represent
a relationship describing a phenomenon.
❖ Many situations involve a set of relationships which explain the behavior of
certain variables.
❖ For example, in analyzing the market conditions for a particular
commodity, there can be a demand equation and supply equations which
explain the price and quantity of commodity exchanged in the market at
market equilibrium.
❖ So there are two equations to explain the whole phenomenon - one for
demand and another for supply.
❖ In such cases, it is not necessary that all the variables should appear in
all the equations.
❖ So estimation of parameters under this type of situation has those features
that are not present when a model involves only a single relationship.
❖ In particular, when a relationship is a part of a system, then some
explanatory variables are stochastic and are correlated with the
disturbances.
❖ So the basic assumption of a linear regression model that the explanatory
variable and disturbance are uncorrelated or explanatory variables are
fixed is violated and consequently ordinary least squares estimator(OLS)
becomes inconsistent.
Can it always be the case that one variable causes the
other?
❖ We may believe for whatever reason that the dependent variable (Y) is not
only a function of explanatory variables (Xs) but also all or some of the
Xs, in turn, are function of the dependent variable (Y) itself.
❖ To identify such situation, we may reasonably use counterfactual reasoning
that there is two way relationship between two or more variables.
❖ The existence of this two-way flow of influence between Y and the Xs
makes the distinction between dependent and independent variables a
little dubious or doubtful.
❖ Therefore, to understand the multi-flow of influence among the variables,
we need to consider more than one regression equations for each
variable and this is what simultaneous equation models (SEM) deal about.
❖ Similar to the classification of variables as explanatory variable and
study variable in linear regression model, the variables in simultaneous
equation models are classified as endogenous variables and exogenous
variables.
2.1.1. Endogenous variables (Jointly determined variables)
❖ The variables which are explained by the functioning of system and values
of which are determined by the simultaneous interaction of the relations in
the model are endogenous variables or jointly determined variables.
 2.1.2. Exogenous variables (Predetermined variables)
❖ The variables that contribute to provide explanations for the endogenous
variables and values of which are determined from outside the model are
exogenous variables or predetermined variables.
❖ Exogenous variables help is explaining the variations in endogenous
variables.
❖ It is customary to include past values of endogenous variables in the
predetermined group.
❖ Since exogenous variables are predetermined, so they are independent of
disturbance term in the model.
❖ They satisfy those assumptions which explanatory variables satisfy in the
usual regression model.
❖ Exogenous variables influence endogenous variables but are not
themselves influenced by them, however, a variable which is endogenous
for one model can be exogenous variable for the other model.
Note:
❑ In linear regression model, the explanatory variables influence study
variable but not vice versa.
❑ So relationship is one sided.
❖ The classification of variables as endogenous and exogenous is important
because a necessary condition for uniquely estimating all the parameters is
that the number of endogenous variables is equal to the number of
independent equations in the system.
❖ Moreover, the main distinction of predetermined variable in estimation of
parameters is that they are uncorrelated with disturbance term in the
equations in which they appear.
❖ In general, simultaneous equation models have the following form:

𝒀𝒊 = 𝜶𝟎 + 𝜶𝟏 𝑿𝒊 + α𝟐 𝒁𝟏𝒊 +𝝁𝒊
ൠ…………………………….(2.1)
𝑿𝒊 = 𝜷𝟎 + 𝜷𝟏 𝒀𝒊 + β𝟐 𝒁𝟐𝒊 +Ɛ𝒊
❖ Where, 𝒁𝟏𝒊 is a variable which affects 𝒀𝒊 but not 𝑿𝒊 and the variable 𝒁𝟐𝒊
affects only 𝑿𝒊 but not 𝒀𝒊 .
❖ ↋𝒊 and μ𝒊 are the stochastic disturbance terms.
❖ From the first line of equation (2.1) we can notice that 𝒀𝒊 is a function
of 𝑿𝒊 and 𝒁𝟏𝒊 .
❖ On the other hand, in the second equation, 𝑿𝒊 (which was serving as an
independent variable in the first equation) becomes a function of 𝒀𝒊 and
𝒁𝟐𝒊 -which was treated as a dependent variable in the first equation-
❖ So, we can’t identify whether the variables 𝒀𝒊 and 𝑿𝒊 are dependent or
independent in this SEM.
❖ A model constitutes a system of simultaneous equations if all the
relationships involved are needed for determining the value of at least
one of the endogenous variables included in the model.
❖ This implies that at least one of the relationships includes more than one
endogenous variable.
❖ Since 𝒁𝟏𝒊 and 𝒁𝟐𝒊 are observable, they are called observed shifters in
the SEM.
❖ On the other hand μ𝒊 and ↋𝒊 are not observed but affect the two equations
respectively and are, thus, called unobserved shifters.

❖ Important features can be observed from equation (2.1)

1) if the equations are derived from economic theory and have causal
interpretation, they are called structural or behavioral equations.
2) given the values of 𝒁𝟏𝒊 , 𝒁𝟐𝒊 , μ𝒊 and ↋𝒊 , these two equations
determine 𝒀𝒊 and 𝑿𝒊 .
For this reason, 𝒀𝒊 and 𝑿𝒊 are known as endogenous variables in the
SEM.
3) Because 𝒁𝟏𝒊 and 𝒁𝟐𝒊 are determined outside the model we view them as
exogenous variables. From a statistical standpoint, the key assumption
concerning 𝒁𝟏𝒊 and 𝒁𝟐𝒊 is that they are both uncorrelated with error terms.
4) Without including 𝒁𝟏𝒊 and 𝒁𝟐𝒊 in the model, there is no way to identify each
equation.
3.2. Simultaneity bias (Inconsistency of OLS Estimators under
SEM)
❖ In a simultaneous equation model, if an explanatory variable is determined
simultaneously with the dependent variable, it is generally correlated with
the error term and applying OLS will result in biased and inconsistent
estimates.
❖ That is, the least squares estimator of parameters in a structural simultaneous
equation is biased and inconsistent because of the correlation between
the random error and the endogenous variables on the right-hand side
of the equation.
❖ Let’s consider once again equation (2.1) specified previously.
❖ For simplicity, suppress/ignore the constant term in the second equation.
𝒀𝒊 = 𝜶𝟎 + 𝜶𝟏 𝑿𝒊 + α𝟐 𝒁𝟏𝒊 + 𝝁𝒊
ൠ … … … … … … … … … . . (𝟐. 𝟐)
𝑿𝒊 = 𝜷𝟏 𝒀𝒊 + β𝟐 𝒁𝟐𝒊 + ↋𝒊
❖ As previous, equation (2.2) is a simultaneous equation model in which
𝒀𝒊 and 𝑿𝒊 are endogenous variables, and 𝒁𝟏𝒊 and 𝒁𝟐𝒊 are exogenous
variables.
❖ By assumption (which emanates from reasoning) 𝒁𝟏𝒊 and 𝝁𝒊 are
uncorrelated and also 𝒁𝟐𝒊 and ↋𝒊 are uncorrelated.
❖ If we estimate, for example, the first equation alone using OLS, the
estimates for α𝟏 and α𝟐 will be biased and inconsistent.
❖ This is because 𝑿𝒊 and 𝝁𝒊 are correlated which creates endogeniety
problem (the most serious assumption of linear regression model).
This can be shown mathematically as follows
❖ Let’s solve for 𝑿𝒊 in terms of exogenous variables (𝒁𝟏𝒊 and 𝒁𝟐𝒊 ) and
error term.
❖ To do so, substitute the value of 𝒀𝒊 from the first equation in to the
second equation which becomes:
𝑿𝒊 = 𝜷𝟏 (𝜶𝟎 + 𝜶𝟏 𝑿𝒊 + α𝟐 𝒁𝟏𝒊 + 𝝁𝒊 ) + β𝟐 𝒁𝟐𝒊 + ↋𝒊
𝑿𝒊 − 𝜷𝟏 𝜶𝟏 𝑿𝒊 = 𝜷𝟏 𝜶𝟎 + 𝜷𝟏 α𝟐 𝒁𝟏𝒊 + 𝜷𝟏 𝝁𝒊 + β𝟐 𝒁𝟐𝒊 + ↋𝒊
𝜷𝟏 𝜶𝟎 𝜷𝟏 α𝟐 𝒁𝟏𝒊 β𝟐 𝒁𝟐𝒊 𝜷𝟏 𝝁𝒊 + ↋𝒊
𝑿𝒊 = + + +
𝟏 − 𝜷𝟏 𝜶𝟏 𝟏 − 𝜷𝟏 𝜶𝟏 𝟏 − 𝜷𝟏 𝜶𝟏 𝟏 − 𝜷𝟏 𝜶𝟏
𝜷𝟏 𝜶𝟎 𝜷𝟏 α𝟐 β𝟐 𝜷𝟏 𝝁𝒊 +↋𝒊
❖ By denoting = ө0 ; = ө1 ; = ө2 ; and = 𝒘𝒊
𝟏−𝜷𝟏 𝜶𝟏 𝟏−𝜷𝟏 𝜶𝟏 𝟏−𝜷𝟏 𝜶𝟏 𝟏−𝜷𝟏 𝜶𝟏
❖ Then, 𝑿𝒊 can be written as:
𝑿𝒊 = ө𝟎 + ө𝟏 𝒁𝟏𝒊 + ө𝟐 𝒁𝟐𝒊 + 𝒘𝒊 … … … … … … … … … … … … (𝟐. 𝟑)
❖ Recall that 𝒁𝟏𝒊 and 𝒁𝟐𝒊 are assumed to be exogenous.
❖ Therefore, from equation (2.3) 𝑿𝒊 and 𝝁𝒊 are correlated if 𝒘𝒊 and 𝝁𝒊 are
correlated.
❖ And, we can see that 𝒘𝒊 is linear function of 𝝁𝒊 and ↋𝒊 , so it is correlated
with ↋𝒊 .
❖ Note that 𝒘𝒊 and 𝝁𝒊 will be correlated if,
✓ β𝟏 ≠ 0, even if 𝝁𝒊 and ↋𝒊 are uncorrelated, or
✓ ↋𝒊 and 𝝁𝒊 are correlated even if β𝟏 = 0
❑ Because both cases mean 𝑿𝒊 is simultaneously determined with 𝒀𝒊 .

❖ In general, when a variable 𝑿𝒊 is correlated with 𝝁𝒊 because of

simultaneity, we say that OLS suffers from simultaneity bias.
❖ Obtaining the direction of the bias in the coefficients will generally get more
complicated as the explanatory variables in the model increase.
❖ Equation (2.3), expresses 𝑿𝒊 in terms of the exogenous variables and the
error terms.
❖ This is called the reduced form equation for 𝑿𝒊 .
❖ A reduced-form equation is one that expresses an endogenous variable
solely in terms of the exogenous variables and the stochastic
disturbances.
❖ The parameters, ө𝟎 , ө𝟏 and ө𝟐 are called reduced form parameters.
❖ These parameters are nonlinear functions of the structural parameters,
which appear in the structural equation (2.1).
❖ Since 𝝁𝒊 and ↋𝒊 and are each uncorrelated with 𝒁𝟏𝒊 and 𝒁𝟐𝒊 , the reduced
form error term 𝒘𝒊 is also uncorrelated with 𝒁𝟏𝒊 and 𝒁𝟐𝒊 because it is a
linear function of 𝝁𝒊 and ↋𝒊 .
❖ Therefore, we can consistently estimate ө𝟎 , ө𝟏 and ө𝟐 by OLS, something
that is used for two-stage least squares (2SLS) estimation.
2.3. Identification and Estimation of Structural Equations in SEM
❖ Earlier we have said that SEM, unlike linear regression model, cannot
be estimated directly using OLS technique for it will give us biased
and inconsistent estimates.
❖ Rather there are other estimation techniques-like Instrumental Variable
(IV) estimation or two-stage least squares estimation method (2SLS).
❖ However, a SEM has to first pass the criteria called identification.
❖ Hence, we will first discuss the issue of identification condition under
section and then proceed to estimation under section.
3.3.1. Identification (Order and rank conditions) (without
proof)
❖ By identification it is to mean whether numerical estimates of the
parameters of a structural equation can be obtained from the estimated
reduced-form coefficients.
❖ Thus, when we say that an equation is identified, it means we can
estimate the parameters of a structural equation from the estimated
reduced-form coefficients.
❖ Identification is a concern of model formulation, not estimation as
the latter depends up on the empirical data and the form of the model.

❖ Example:
❖ Consider the following SEM involving two-equation system.
𝒀𝟏 = α𝟎 + α𝟏 𝒀𝟐 + α𝟐 𝒁𝟏 + μ
ൠ…………..………………(2.4)
𝒀𝟐 = β𝟎 + β𝟏 𝒀𝟏 + β𝟐 𝒁𝟐 + Ɛ
❖ Where, 𝒀𝟏 and 𝒀𝟐 are endogenous variables (because they are correlated
with the error terms), and 𝒁𝟏 and 𝒁𝟐 are exogenous (because they are
assumed to be independent of, or uncorrelated with the error terms).
❖ For the first equation to be identified, there should be at least one
variable which is excluded from the first equation- but included in the
other equation - and is uncorrelated with the error term of the first
equation.
❖ This is known as exclusion restrictions.
❖ Therefore, since 𝒁𝟐 is excluded from the first equation, and since is
uncorrelated with the error term μ - because it is assumed to be
exogenous-the first equation is identified.
❖ Similarly, for the second equation to be identified there should be at least
one variable which is excluded from the second equation, and is
uncorrelated with the error term (Ɛ) of this equation.
❖ Since 𝒁𝟏 is excluded from the second equation, and since 𝒁𝟏 is assumed
to be exogenous (uncorrelated with the error term, Ɛ ) the second
equation is identified.
Take another SEM which involves three-equation system.
𝒀𝟏 = 𝜶𝟎 + 𝜶𝟏 𝒀𝟐 + 𝜶𝟐 𝒀𝟑 + 𝜶𝟑 𝒁𝟏 + 𝛍
𝒀𝟐 = 𝜷𝟎 + 𝜷𝟏 𝒀𝟏 + 𝜷𝟐 𝒁𝟏 + 𝜷𝟑 𝒁𝟐 + 𝜷𝟒 𝒁𝟑 + Ɛ ቑ … … … … . . . (𝟐. 𝟓)
𝒀𝟑 = 𝜸𝟎 + 𝜸𝟏 𝒀𝟐 + 𝜸𝟐 𝒁𝟏 + 𝜸𝟑 𝒁𝟐 + 𝜸𝟒 𝒁𝟑 + 𝜸𝟓 𝒁𝟒 + 𝐯
❖ Where, 𝒀𝟏 , 𝒀𝟐 , and 𝒀𝟑 are endogenous variables, 𝒁𝟏 , 𝒁𝟐 , 𝒁𝟑 and 𝒁𝟒 and
are exogenous variables. 𝜶𝒊 , β𝒊 and γ𝒊 are estimators of the three
structural equations, respectively.
❖ It is generally difficult to show that an equation is identified in an SEM
with more than two equations, but it is easy to see when certain equations
are not identified.
❖ The first equation is identified (is at least promising) because three
exogenous variables 𝒁𝟐 , 𝒁𝟑 and 𝒁𝟒 are excluded from this equation.
❖ The second equation is identified (is at least promising). Why?
❖ But, the third equation is not identified because no exogenous variables
excluded from this equation.
❖ That, is 𝒁𝟏 (appearing in the first and second equation), 𝒁𝟐 and 𝒁𝟑
(appearing in the second equation) and 𝒁𝟒 all appear in the third
equation itself.
❑ Formally speaking, there are two rules/conditions which must be
fulfilled for an equation to be identified.

1) The order condition for identification

❖ This condition is based on a counting rule of the exogenous and
endogenous variables included and excluded in the SEM.
❖ This condition states that an equation in any SEM satisfies the order
condition for identification if the number of excluded exogenous
variables from the equation is at least as large as the number of right-
hand side endogenous variables in the equation.
❖ Now, identification condition for equation (2.5) can be answered based on
order condition as:
Equations No. of endogenous No. of excluded Identification
No. Variables exogenous variables condition
𝟏𝒔𝒕 2(= 𝒀𝟏 𝐚𝐧𝐝 𝒀𝟐 ) 3(= 𝒁𝟐 , 𝒁𝟑 and 𝒁𝟒 ) (3 ≥ 2) over
identified
𝟐𝒏𝒅 1(=𝒀𝟏 ) 1(=𝒁𝟒 ) (1=1) just
identified
𝟑𝒓𝒅 1(=𝒀𝟐 ) 0 (0 ≤ 1) under
identified

❖ Note from the above that in the first equation the number of excluded
exogenous variables (= 3) is greater than the number of endogenous
variables (= 2).
❖ Such an equation is called over identified.
❖ In the second equation, the number of excluded exogenous variables (= 1) is
exactly equal to the number of endogenous variables (= 1).
❖ In this case, the equation is called just-identified.
❖ In the third equation, the number of excluded exogenous variables (= 0) is
less than the number of endogenous variables (= 1) and the equation is called
unidentified equation.
❖ Under-identified equation means we can’t estimate the parameters of a
structural equation from the estimated reduced-form coefficients.
✓ Order condition is a necessary (but not sufficient) condition for
identification.
❖ For example, from equation (2.5) we have said that the second equation is
identified because of the presence of an excluded variable from this equation.
❖ But, if 𝜸𝟓 = 𝟎, it means 𝒁𝟒 is not correlated with 𝒀𝟏 , 𝒀𝟐 , and 𝒀𝟑 will be
eliminated from the model, so the second equation will be unidentified.
❖ This again illustrates that identification of an equation depends not only
on the presence of an excluded variable but also on the values of the
parameters (which we can never know for sure) in the other equations.
✓ The sufficient condition for identification is called the rank condition.
2) The rank condition for identification
❖ The rank condition states that in a SEM containing G equations any
particular equation is identified if and only if it is possible to construct at
least one non-zero determinant of order (G-1) from the coefficients of the
variables excluded from that particular equation but contained in the other
equations of the model.
❖ Remember from your linear algebra course that, the term rank refers to the
rank of a matrix and is given by the largest-order square matrix (contained
in the given matrix) whose determinant is nonzero.
❖ Alternatively, the rank of a matrix is the largest number of linearly
independent rows or columns of a matrix.
❑ To understand the order and rank conditions, let’s introduce the
following notations:
❖ Let,
M = number of endogenous variables in the model
m = number of endogenous variables in a given equation
K = number of exogenous variables in the model including the
intercept
k = number of exogenous variables in a given equation
I) Order Condition:
✓ In a model of M simultaneous equations in order for an equation to be
identified, it must exclude at least 𝑴 − 𝟏 variables (endogenous as
well as exogenous) appearing in the model.
✓ If it excludes exactly 𝑴 − 𝟏 variables, the equation is just identified.
✓ If it excludes more than 𝑴 − 𝟏 variables, it is over-identified.
✓ In a model of M simultaneous equations, in order for an equation to be
identified, the number of exogenous variables excluded from the
equation must not be less than the number of endogenous variables
included in that equation less 1, that is, 𝑲 − 𝒌 ≥ 𝒎 − 𝟏.
✓ If 𝑲 − 𝒌 = 𝒎 − 𝟏, the equation is just identified, but if 𝑲 − 𝒌 >
𝒎 − 𝟏, it is over identified.
II) Rank condition:
✓ In a model containing M equations in M endogenous variables, an
equation is identified if and only if at least one nonzero determinant of
order (𝑴 − 𝟏)(𝑴 − 𝟏) can be constructed from the coefficients of the
variables (both endogenous and predetermined) excluded from that
particular equation but included in the other equations of the model.
✓ In a model containing simultaneous equations:
1) If 𝑲 − 𝒌 > 𝒎 − 𝟏, and the rank of the A matrix is 𝑴 − 𝟏, the
equation is over identified.
2) If 𝑲 − 𝒌 = 𝒎 − 𝟏, and the rank of the matrix A is 𝑴 − 𝟏,
the equation is exactly identified.
3) If 𝑲 − 𝒌 ≥ 𝒎 − 𝟏, and the rank of the matrix A is less than
𝑴 − 𝟏, the equation is under-identified.
4) If 𝑲 − 𝒌 < 𝒎 − 𝟏, the structural equation is unidentified.
The rank of the A matrix in this case is bound to be less than 𝑴 −
𝟏. (Why?)
❑ Steps of checking rank condition
1) Bring all items of each equation, except the error term, to the left of
the equal sign
2) Put all the endogenous and exogenous variables in a row
3) Put the corresponding coefficients of each variable beneath each
variable
4) Construct a matrix from excluded variables (both exogenous and
endogenous) and check for its rank.
5) Then,
✓ If we can form more than one (M-1) by (M-1) matrix of non-zero
determinant, the matric is said to be over identified.
✓ If we can form exactly one (M-1) by (M-1) matrix of non-zero
determinant, the matric is said to be just identified.
✓ If we can not form at least (M-1) by (M-1) matrix of non-zero
determinant, the matric is said to be under identified.
Illustration
✓ Given the following simultaneous equation model,
𝒀𝟏 = 𝜶𝟎 + 𝜶𝟏 𝒀𝟐 + 𝜶𝟐 𝒀𝟑 + 𝜶𝟑 𝒁𝟏 + 𝛍
𝒀𝟐 = 𝜷𝟎 + 𝜷𝟏 𝒀𝟑 + 𝜷𝟐 𝒁𝟏 + 𝜷𝟑 𝒁𝟐 + Ɛ
𝒀𝟑 = 𝜸𝟎 + 𝜸𝟏 𝒀𝟏 + 𝜸𝟐 𝒁𝟏 + 𝜸𝟑 𝒁𝟐 + 𝐯
𝒀𝟒 = θ𝟎 + θ𝟏 𝒀𝟏 + θ𝟐 𝒀𝟐 + θ𝟑 𝒁𝟑 + 𝒘
✓ To check the order condition for identification, look at the following table.
Equation No. endo. No. excluded Exo. Identification
No. Variables Variables Condition
𝟏𝒔𝒕 2 ( = 𝒀𝟐 and 𝒀𝟑 ) 2 ( = 𝒁𝟑 and 𝒁𝟒 ) 2=2 All are
𝟐𝒏𝒅 1 ( = 𝒀𝟑 ) 1 ( = 𝒁𝟑 ) 1=1 Exactly/Just
Identified
𝟑𝒓𝒅 1 ( = 𝒀𝟏 ) 1 ( = 𝒁𝟑 ) 1=1
𝟒𝒕𝒉 2 ( = 𝒀𝟏 and 𝒀𝟐 ) 2 ( = 𝒁𝟏 and 𝒁𝟐 ) 2=2
❖ Now, Let us recheck with the rank condition.
✓ First, bring all items, except error terms, to the left
𝒀𝟏 − 𝜶𝟎 − 𝜶𝟏 𝒀𝟐 − 𝜶𝟐 𝒀𝟑 − 𝜶𝟑 𝒁𝟏 = 𝛍
𝒀𝟐 − 𝜷𝟎 −𝜷𝟏 𝒀𝟑 − 𝜷𝟐 𝒁𝟏 − 𝜷𝟑 𝒁𝟐 = Ɛ
𝒀𝟑 − 𝜸𝟎 − 𝜸𝟏 𝒀𝟏 − 𝜸𝟐 𝒁𝟏 − 𝜸𝟑 𝒁𝟐 = 𝐯
𝒀𝟒 − θ𝟎 − θ𝟏 𝒀𝟏 − θ𝟐 𝒀𝟐 − θ𝟑 𝒁𝟑 = 𝒘
✓ Second, put all the endogenous and exogenous variables in a row and
put the corresponding coefficients beneath each variable.
Equation No. I 𝒀𝟏 𝒀𝟐 𝒀𝟑 𝒀𝟒 𝐙𝟏 𝐙𝟐 𝐙𝟑
𝟏𝒔𝒕 −𝜶𝟎 1 −𝜶𝟏 −𝜶𝟐 0 − 𝜶𝟑 0 0
𝟐𝒏𝒅 −𝜷𝟎 0 1 −𝜷𝟏 0 − 𝜷𝟐 −𝜷𝟑 0
𝟑𝒓𝒅 −𝜸𝟎 −𝜸𝟏 0 1 0 − 𝜸𝟐 −𝜸𝟑 0
𝟒𝒕𝒉 −θ𝟎 −θ𝟏 − θ𝟐 0 1 0 0 −θ𝟑
✓ Consider the first equation, which excludes variables 𝒀𝟒 , 𝐙𝟐 , and 𝐙𝟑
✓ For this equation to be identified, we must obtain at least one nonzero
determinant of order 3 × 3 from the coefficients of the variables excluded from
this equation but included in other equations.
✓ To obtain the determinant we first obtain the relevant matrix of coefficients of
variables 𝒀𝟒 , 𝐙𝟐 , and 𝐙𝟑 included in the other equations.
✓ In the present case there is only one such matrix, call it A, defined as follows.
𝟎 −𝜷𝟑 𝟎
A = 𝟎 −𝜸𝟑 𝟎
𝟏 𝟎 −𝜽𝟑
✓ If we find the determinant of matrix A, it is equal to zero.
✓ This implies the rank of the matrix is less than 3 and it is not identified.
✓ Therefore, although the order condition shows that the SEM is identified, the
rank condition shows that it is not.
✓ Note: As noted, the rank condition is both a necessary and sufficient condition for
identification.
3.3.2. Indirect Least Squares (ILS), Instrumental
Variable (IV) and Two Stage Least Squares
(2SLS) estimation of structural equations
❖ Once we finished modeling SEM, the next task is estimation.
❖ Yet, estimation problem is rather complex because there are a
variety of estimation techniques with varying statistical
properties.
❖ In an SEM, two approaches may be adopted to estimate the
structural equations, namely:
✓ single-equation methods, also known as limited
information methods, and
✓ system methods, also known as full information
methods.
❖ In the single-equation methods, we estimate each equation in the system
(of simultaneous equations) individually, taking into account any
restrictions placed on that equation (such as exclusion of some
variables) without worrying about the restrictions on the other equations
in the system, hence the name limited information methods.
❖ In the system methods, on the other hand, we estimate all the equations
in the model simultaneously, taking due account of all restrictions on
such equations by the omission or absence of some variables (recall that
for identification such restrictions are essential), hence the name full
information methods.
❖ Although the systems method -such as the full information maximum
likelihood (FIML) method- may be good to preserve the spirit of
simultaneous-equation models, in reality they are not commonly used
for different reasons.
✓ Some of these reasons include:
1) High burden of the computation: for example, estimating 20
equations require incorporating 151 coefficients for US economy in
1955.
2) The systems methods, such as FIML, lead to solutions that are highly
non-linear in the parameters and are, therefore, often difficult to
determine.
3) If there is a specification error (eg., a wrong functional form or
exclusion of relevant variables) in one or more equations of the
system, that error is transmitted to the rest of the system. As a result, the
systems methods become very sensitive to specification errors.
❖ Due to the above problems, therefore, single-equation methods are often
used in practice.
❖ These include: Ordinary least squares (OLS); Indirect least squares
(ILS); and Two-stage least squares (2SLS)
1) Indirect least squares (ILS)
❖ For a just or an exactly identified structural equation, the method of
obtaining the estimates of the structural coefficients from the OLS
estimates of the reduced-form coefficients is known as the method
Indirect Least Squares(ILS), and the estimates obtained are known as the
indirect least squares estimates.
Step 1: Obtain the reduced-form equations; solve for the endogenous
variable in each equation in terms solely of the exogenous variables and the
stochastic error term.
✓ This gives the reduced-form equations.
Step 2: Apply OLS to the reduced-form equations individually.
✓ Since the explanatory variables in these equations are
predetermined/exogenous variables which are uncorrelated with the
stochastic disturbances, this operation is permissible, and the
estimates obtained are consistent.
Step 3: Obtain estimates of the original structural coefficients from the
estimated reduced-form coefficients obtained in Step 2.
✓ If an equation is exactly identified, there is a one-to-one correspondence
between the structural and reduced-form coefficients; that is, one can
derive unique estimates of the former from the latter.
Example:
𝑄𝑡 = 𝛼0 + 𝛼1 𝑃𝑡 + 𝛼2 𝐼𝑡 + 𝜇𝑡
ൠ … … … … … … … … … … … … … … 2.6
𝑄𝑡 = 𝛽0 + 𝛽1 𝑃𝑡 + Ɛ𝑡
✓ Where, 𝑄𝑡 and 𝑃𝑡 are quantity and price which are endogenous, 𝐼𝑡 is
income and is exogenous in the first equation, 𝜇𝑡 is error term of the first
equation and Ɛ𝑡 is error term of the second equation.
✓ Then, given equation (2.6)
1) Determine the identification condition of equation system?
2) Find the estimators of the structural equation using ILS?
Solution:
1) The first equation is demand function (Why?). As a result, the second
equation has to be supply function.
✓ In the first equation, there is no excluded variable; hence the demand
function is under-identified. But, Income,𝑰𝒕 , is excluded from the second
equation; hence supply function is just-identified.
2. Since the second equation is just identified, we can apply ILS and solve
for 𝑸𝒕 and 𝑷𝒕 in this equation as in the following.
✓ At equilibrium, Supply = Demand.
✓ Hence, substitute 𝑸𝒕 from the first equation in to 𝑸𝒕 of the second
equation:
𝜶𝟎 + 𝜶𝟏 𝑷𝒕 + 𝜶𝟐 𝑰𝒕 + 𝛍𝒕 = 𝜷𝟎 + 𝜷𝟏 𝑷𝒕 + Ɛ𝒕
𝜶𝟏 𝑷𝒕 − 𝜷𝟏 𝑷𝒕 = 𝜷𝟎 − 𝜶𝟎 − 𝜶𝟐 𝑰𝒕 + Ɛ𝒕 − 𝛍𝒕
𝑷𝒕 (𝜶𝟏 − 𝜷𝟏 ) = 𝜷𝟎 − 𝜶𝟎 − 𝜶𝟐 𝑰𝒕 + Ɛ𝒕 − 𝛍𝒕
Where 𝜶𝟏 ≠ 𝜷𝟏
 𝜷𝟎 − 𝜶𝟎 −𝜶𝟐 Ɛ 𝒕 − 𝛍𝒕
𝑷𝒕 = + 𝑰𝒕 +
𝜶𝟏 −𝜷𝟏 𝜶𝟏 −𝜷𝟏 𝜶𝟏 −𝜷𝟏
𝑷𝒕 = θ𝟎 + θ𝟏 𝑰𝒕 + 𝒗𝒕 … … … … … … … … … … … … … … … … … … … . 𝟐. 𝟕
𝜷𝟎 −𝜶𝟎 −𝜶𝟐 Ɛ𝒕 −𝛍𝒕
✓ Where, θ𝟎 = ; θ𝟏 = ; and 𝒗𝒕 =
𝜶𝟏 −𝜷𝟏 𝜶𝟏 −𝜷𝟏 𝜶𝟏 −𝜷𝟏
✓ To estimate 𝑸𝒕 , substitute the estimate of 𝑷𝒕 from equation (2.7) in to either
the demand or supply function of equation (2.6):
𝑸𝒕 = 𝜷𝟎 + 𝜷𝟏 𝑷𝒕 + Ɛ𝒕
𝜷𝟎 − 𝜶𝟎 −𝜶𝟐 Ɛ 𝒕 − 𝛍𝒕
𝑸 𝒕 = 𝜷𝟎 + 𝜷𝟏 ( + 𝑰𝒕 + ) + Ɛ𝒕
𝜶𝟏 −𝜷𝟏 𝜶𝟏 −𝜷𝟏 𝜶𝟏 −𝜷𝟏
𝜷𝟎 − 𝜶𝟎 −𝜷𝟏 𝜶𝟐 Ɛ 𝒕 − 𝛍𝒕
𝑸 𝒕 = 𝜷𝟎 + 𝜷𝟏 + 𝑰𝒕 + 𝜷𝟏 + Ɛ𝒕
𝜶𝟏 −𝜷𝟏 𝜶𝟏 −𝜷𝟏 𝜶𝟏 −𝜷𝟏
𝜷𝟎 𝜶𝟏 − 𝜷𝟎 𝜷𝟏 + 𝜷𝟎 𝜷𝟏 − 𝜷𝟏 𝜶𝟎 −𝜷𝟏 𝜶𝟐 𝜷𝟏 Ɛ𝒕 − 𝜷𝟏 𝛍𝒕 + 𝜶𝟏 Ɛ𝒕 + 𝜷𝟏 Ɛ𝒕
𝑸𝒕 = + 𝑰𝒕 +
𝜶𝟏 −𝜷𝟏 𝜶𝟏 −𝜷𝟏 𝜶𝟏 −𝜷𝟏
❖ Simplifying this gives:
𝜷𝟎 𝜶𝟏 − 𝜷𝟏 𝜶𝟎 −𝜷𝟏 𝜶𝟐 𝜶𝟏 Ɛ𝒕 − 𝜷𝟏 𝛍𝒕
𝑸𝒕 = + 𝑰𝒕 +
𝜶𝟏 −𝜷𝟏 𝜶𝟏 −𝜷𝟏 𝜶𝟏 −𝜷𝟏
𝑸𝒕 = γ𝟎 + γ𝟏 𝑰𝒕 + 𝒘𝒕 … … … … … … … … … … … … … … … … . . … 𝟐. 𝟖
𝜷𝟎 𝜶𝟏 −𝜷𝟏 𝜶𝟎 −𝜷𝟏 𝜶𝟐 𝜶𝟏 Ɛ𝒕 −𝜷𝟏 𝛍𝒕
Where, γ𝟎 = ; γ𝟏 = ; and 𝒘𝒕 =
𝜶𝟏 −𝜷𝟏 𝜶𝟏 −𝜷𝟏 𝜶𝟏 −𝜷𝟏
❖ Look once again back to the structural equations (2.6).
❖ It consists of five structural coefficients/parameters; namely
𝜶𝟎 , 𝜶𝟏 , 𝜶𝟐 , 𝜷𝟎, 𝒂𝒏𝒅 𝜷𝟏 .
❖ But, there are only four equations to estimate those structural coefficients,
namely, the four reduced-form coefficients θ𝟎 , θ𝟏 , γ𝟎 and γ𝟏 .
❖ Since the number of coefficients in the structural equations (=5) is greater
than the number of coefficients in the reduced form equations (=4), unique
solution of all the structural coefficients is not possible.
✓ As a result, only the parameters of the supply function can be
identified as (derived from the four reduced form parameters):
𝜸𝟏
𝜷𝟎 = 𝜸𝟎 − 𝜷𝟏 𝜽𝟎 and 𝜷𝟏 = ……………….…….(2.9)
𝜽𝟏
✓ Remember from chapter of econometrics I that estimating equation
(2.7) using OLS results:
σ 𝒑𝒕 𝒊𝒕
෡
𝜽𝟏 = ෡ ഥ ෡ ത
𝟐 and 𝜽𝟎 = 𝑷 − 𝜽𝟏 𝑰
σ 𝒊𝒕
✓ Similarly, estimating equation (2.8) using OLS results:
σ 𝒒𝒕 𝒊𝒕
ෝ𝟏 =
𝜸 ෝ𝟎 = 𝑸
𝟐 and 𝜸
ഥ −𝜸 ෝ𝟏 ത𝑰
σ 𝒊𝒕
✓ Where, the lowercase letters, as usual, denote deviations from mean
value, and, 𝑸ഥ, 𝑷
ഥ and ത𝑰 are the sample mean values of Q and P and I.
✓ Substituting estimates of the reduced-form coefficients in to equation
(2.9) gives the ILS estimators of structural equation parameters for
supply functions:
 ෝ𝟏
𝜸 ෡
𝜽𝟎
෡ ෡ ෡
ෝ 𝟎 − 𝜷𝟏 𝜽 𝟎 = 𝜸
𝜷𝟏 = ෡ and 𝜷𝟎 = 𝜸 ෝ𝟎 − ෡ 𝜸ෝ𝟏 ……..……………(2.10)
𝜽𝟏 𝜽𝟏
✓ But, as far as the demand function is under-identified, there is no unique
way of estimating the parameters, and remains under-identified.

2) Two-Stage Least Squares (2SLS) Method

❖ If we have over-identified equation(s) in a SEM,
✓ OLS will not be appropriate because the existence of endogenous
variable(s) on the right side of an equation will give biased and
inconsistent estimated due to endogeniety problem.
✓ ILS estimation will not be appropriate because it will give more than
one estimate for a single coefficient.
❖ Therefore, other estimation techniques have to be used such as 2SLS.
Illustration
Consider the following SEM:
Income = 𝛼0 + 𝛼1 𝐴𝑠𝑠𝑒𝑡 + 𝛼2 𝐸𝑥𝑝𝑒𝑟𝑖𝑒𝑛𝑐𝑒 + 𝛼3 𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛 + 𝜇𝑖
ൠ ……
𝐴𝑠𝑠𝑒𝑡 = 𝛽0 + 𝛽1 𝐼𝑛𝑐𝑜𝑚𝑒 + Ɛ𝑖
……(2.11)
Where, experience and level of education (both measured in years) are
assumed to be exogenous.
✓ Applying the order condition of identification, we can see that the
income equation is under-identified whereas the asset equation is over-
identified.
❖ If we can find a variable, 𝑍𝑖 , which satisfies the following conditions:
1) highly correlated with the endogenous variable (income in this
case): Instrument relevance
2) But, uncorrelated with error term (Ɛ𝑖 in this case): Instrument
Exogeneity
✓ Then, we can substitute 𝑍𝑖 in place of income in the second equation and
estimate the asset model using OLS directly.
✓ This means since 𝑍𝑖 is correlated with income, it can serve as a “proxy” or
“substitute” for income.
✓ 𝑍𝑖 is then known as instrumental variable (IV) and the estimation is called
instrumental variable estimation.
✓ But, sometimes we may find more than one exogenous variables (like
experience and education in this example) which satisfy the above two
conditions.
✓ As far as these variables are uncorrelated with the error term (s), any
linear combination of these exogenous variables and is also uncorrelated
with the error term(s), and can be used as a valid instrumental variable for
income.
✓ That is, to find the best instrumental variable for income, we choose the
linear combination and that is most highly correlated with income using the
following:
Income = 𝜋0 + 𝜋1 𝐸𝑥𝑝𝑒𝑟𝑖𝑒𝑛𝑐𝑒 + 𝜋3 𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛 + 𝑣 … … … … 2.11
Where, 𝐸 𝑣 = 0, 𝑐𝑜𝑣 𝐸𝑥𝑝𝑒𝑟𝑖𝑒𝑛𝑐𝑒, 𝑣 = 0, and 𝑐𝑜𝑣 𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛, 𝑣 = 0
❑ This is the concept of two-stage least squares (2SLS) estimation.

✓ As the name 2SLS implies estimation of SEM using 2SLS involves two
stages. These are:
Stage 1: Estimate the endogenous variable using all exogenous
variable of the SEM
✓ To get rid of the likely correlation between endogenous variable(income)
and the error term, Ɛ𝑖 , we find the best linear combination among all the
exogenous variables in the whole system, not just that equation.
✓ The linear combination of Experience and Education in equation (2.12),
which we call 𝑖𝑛𝑐𝑜𝑚𝑒 ∗ becomes:
𝑖𝑛𝑐𝑜𝑚𝑒 ∗ = 𝜋0 + 𝜋1 𝐸𝑥𝑝𝑒𝑟𝑖𝑒𝑛𝑐𝑒 + 𝜋2 𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛 … … … … … 2.13
✓ But, since we don’t know the exact value of 𝑖𝑛𝑐𝑜𝑚𝑒 ∗ , we can only
estimate, using OLS, by regressing income on experience and education
as:
෣ = 𝜋ො 0 + 𝜋ො 1 𝐸𝑥𝑝𝑒𝑟𝑖𝑒𝑛𝑐𝑒 + 𝜋ො 2 𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛 … … … … … 2.14
𝐼𝑛𝑐𝑜𝑚𝑒
Or,
𝐼𝑛𝑐𝑜𝑚𝑒 = 𝐼𝑛𝑐𝑜𝑚𝑒෣ + 𝑣ො … … … … … … … . … … … … … … … … … . . (2.15)
✓ Then, conduct a joint significance of variables for equation (2.13) using
F-test. If the variable are found jointly significant (not larger than 5%),
෣ as an IV.
then use the fitted values of income, 𝐼𝑛𝑐𝑜𝑚𝑒,
✓ All the above tasks involve Stage-I.
Stage 2: Substitute the estimated value of the endogenous variable
obtained from Stage I and estimate the over-identified model of
the SEM
✓ Substitute equation (2.15) in the second equation of (2.11) results,

෣ + 𝑣)
𝐴𝑠𝑠𝑒𝑡 = 𝛽0 + 𝛽1 𝐼𝑛𝑐𝑜𝑚𝑒 + 𝜇𝑖 = 𝛽0 + 𝛽1 (𝐼𝑛𝑐𝑜𝑚𝑒 ො + 𝜇𝑖

෣ + 𝑤𝑖 … … … … … . . … … … . … … … (2.16)
𝐴𝑠𝑠𝑒𝑡 = 𝛽0 + 𝛽1 𝐼𝑛𝑐𝑜𝑚𝑒
Where, 𝑤𝑖 = 𝛽1 𝑣ො + 𝜇𝑖
✓ Then, directly estimate the asset model in (2.16) using OLS method.
✓ Equation (2.16) is very similar in appearance, with the second equation of
(2.11) with the only difference being that actual value of income is replaced
෣ using exogenous variables.
by its estimated value, 𝐼𝑛𝑐𝑜𝑚𝑒,
❖ Therefore, the advantage of 2SLS (as in this case) is that the error term,
෣ Why?
Ɛ𝒊 , is correlated with income, but not 𝑰𝒏𝒄𝒐𝒎𝒆.
✓ As a result, OLS estimation of on equation (2.16), will give unbiased
and consistent estimate unlike equation (2.10).
Features of 2SLS
1) It can be applied to an individual equation in the SEM without
directly taking into account any other equation(s) in the system.
✓ For this, the method has been used extensively in practice, for
solving econometric models involving a large number of equations.
2) ILS provides multiple estimates of a parameter in the over
identified equations, but 2SLS provides only one.
3) It is easy to apply because all one needs to know is the total number
of exogenous or variables in the system, without knowing any other
variables in the system.
4. It can also be applied to exactly identified equations and gives
identical estimates with ILS.
5. If the 𝑹𝟐 values in the reduced-form regressions (that is, Stage-1
regressions) are very high, say, in excess of 0.8, the classical OLS
estimates and 2SLS estimates will be very close. Why?
6. In reporting the ILS regression, we did not state the standard errors of
the estimated coefficients.
✓ But we can do this for the 2SLS estimates because the structural
coefficients are directly estimated from the second-stage (OLS)
regressions, though there may be some modification (see 𝒘𝒊 =
ෝ + 𝝁𝒊 under equation (2.16) above).
𝜷𝟏 𝒗
Numerical example: Undertake 2SLS on the following simple Keynesian model

Consumption function: Ct =  0 +  1 Y t + ut 0  1  1 EQ 1

Income identity: Yt = C t + I t (= S t ) EQ 2

Where Yt - National income

Ct - Consumption expenditure
It - Investment expenditure
ut - Stochastic disturbance term

Use the Data Below to demonstrate OLS is biased and inconstant, and 2SLS is
unbiased estimator
Time I C Y
1 2 3 5
2 3 4 7
3 2 2 4
4 4 5 9
5 2 3 5
6 3 4 7
7 2 4 6
First: to understand the difference between OLS and 2SLS, run OLS for C as
function of income (reg C Y on Stata as follow)
. reg C Y

Source SS df MS Number of obs = 7

F(1, 5) = 59.84
Model 5.27360775 1 5.27360775 Prob > F = 0.0006
Residual .440677966 5 .088135593 R-squared = 0.9229
Adj R-squared = 0.9075
Total 5.71428571 6 .952380952 Root MSE = .29688

C Coefficient Std. err. t P>|t| [95% conf. interval]

Y .559322 .0723076 7.74 0.001 .3734495 .7451945

_cons .1355932 .458129 0.30 0.779 -1.042065 1.313251

Conclusion: The above OLS coefficient of Y (=0.559) is upward biased

Now, let’s see the 2SLS estimate for the above same data: Steps for 2SLS
• Run the first-stage regression (Y on I).
• Get the fitted values Ŷ.
• Run the second-stage regression (C on Ŷ).
• Interpret the coefficient on Ŷ (α₁).
• Reflect: Why can’t we just use OLS? What problem does 2SLS solve?
Short cut method of 2SLS in STATA
. ivregress 2sls C ( Y = I )

Instrumental variables 2SLS regression Number of obs = 7

Wald chi2(1) = 52.00
Prob > chi2 = 0.0000
R-squared = 0.9125
Root MSE = .26726

C Coefficient Std. err. z P>|z| [95% conf. interval]

Y .5 .0693375 7.21 0.000 .3641009 .6358991

_cons .5 .4377452 1.14 0.253 -.3579649 1.357965

Endogenous: Y
Exogenous: I

Consultation: 2SLS estimate is a true parameter estimator (0.5)

Discussion Questions
1. Why can't we use OLS directly to estimate equation the above relationship?
2. What makes Y endogenous in this model?
4. How does 2SLS correct for endogeneity?
 Exercise (practice the following to question)
1) Given an Econometric model,
𝒚 = 𝜷𝒙 + Ɛ
a) Show that if x is correlated with Ɛ, the OLS estimate of β, which is β෠
will be biased and inconsistent estimator.
b) If we find a variable Z, correlated with X and uncorrelated with Ɛ, then
Z can be used as an instrument for X. In this case, show that β෠ becomes
unbiased and consistent estimator of β.
2) Consider a two equation model:
𝑦1 = α1 𝑦2 + α2 𝑥 + 𝑒1
𝑦2 = β1 𝑦1 + β2 𝑥 + 𝑒2
a) Identify which variable/s is/are exogenous and endogenous.
b) What will happen to the identification of the equations if β2 = 0 ?
c) Find the reduced-form equations and find the formulas for estimating
the structural coefficients if β2 = 0
 Assignment ONE
Use the attached STATA file (FERTIL2.RAW) and the following information to answer about IV & 2SLS (submit
your results in LOG along with DO file). This question has been taken from Wooldridge textbook.