Chapter 1 SIMULTANEOUS EQUATIONS MODELS

CHAPTER 1
SIMULTANEOUS EQUATIONS MODELS

Abrham Seyoum (PhD)

Assistant Professor of Economics

Yom Graduate College
Addis Ababa, Ethiopia

Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 1

Chapter 1 SIMULTANEOUS EQUATIONS MODELS

1.1 The Nature of Simultaneous Equations models

▪ There are different sources of endogeneity in econometric

estimations such as omitted variables, error in variables, etc.
▪ In this chapter, we discuss one of the important forms of
endogeneity of explanatory variables called simultaneity.
▪ It arises when one or more of the explanatory variables are
jointly determined with the dependent variable, typically
through an equilibrium mechanism.
▪ In this chapter, we study how we estimate simultaneous
equation models (SEMs) as the LS applied to an equation in a
simultaneous system is generally biased and inconsistent.
▪ The most important point to remember in using simultaneous
equations models is that each equation in the system should
have a ceteris paribus, causal interpretation.

Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 2

Chapter 1 SIMULTANEOUS EQUATIONS MODELS

▪ The classic example of an SEM is a supply and demand

equation for some commodity or input to production (such as
labor).
▪ Let’s take a partial equilibrium model of demand and supply to
illustrate SEMs.

At equilibrium,
Where, Y is quantity, P is price,
is weather and is income.

▪ The above two equations are called structural equations.

, , and are structural parameters; and and
are structural error terms. Here, price and quantity are
simultaneously determined.

Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 3

Chapter 1 SIMULTANEOUS EQUATIONS MODELS

▪ In SEMs, we classify variables in to three:

1) Endogenous variables: variables determined within the
model (P and Y in our example);
2) Exogenous variables: variables determined by forces
outside the model and are orthogonal with the error terms;
and
3) Lagged endogenous variables
Note: 2 and 3 are also called predetermined variables.

▪ In this chapter three issues to see: Simultaneous bias,

identification and estimation.

Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 4

Chapter 1 SIMULTANEOUS EQUATIONS MODELS

1.2 Simultaneity Bias

▪ Given the previous partial equilibrium model of demand and

supply as follows:

At equilibrium,

▪ Since P and Y are simultaneously determined, P is correlated

with and . So, LS leads to biased and inconsistent result.
When the endogenous variables P and Y are correlated with the
error terms because of simultaneity, we say that OLS suffers
from simultaneity bias. (Proof!)
Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 5
Chapter 1 SIMULTANEOUS EQUATIONS MODELS

▪ Look at the reduced form using substitution in the above

supply and demand equations:

▪ So, P and Y are determined by both structural error terms in

addition to the exogenous variables.
▪ If we apply OLS we find one estimated parameter for ,
etc as one and OLS is unbiased and efficient.
➢ Can we always find the structural parameters from the
reduced equations?
▪ Re-writing the reduced form equations as follows:

Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 6

Chapter 1 SIMULTANEOUS EQUATIONS MODELS

▪ First, we can calculate for and :

▪ Then, the structural parameters and can be calculated as

follows:

▪ By estimating using OLS, once we find for the estimates of the

parameters in the reduced form equations, we can derive the
structural parameters. This kind of estimation is called Indirect
Least Squares Estimation Method.
▪ This is possible only if the original equations are identified or
over identified but if they are not identified it is not possible to
retrieve the structural parameters.
Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 7
Chapter 1 SIMULTANEOUS EQUATIONS MODELS

▪ Which multi-equation systems are not SEMs?

1) Equations which are derived from the behavior of one
economic agent (decision making units)
▪ Example: Housing expenditure and saving
Housing = f(saving, income, education, age) + random term
Saving = f(housing, income, education, age) + random term

▪ Neither has a ceteris paribus interpretation because housing and

saving are chosen by the same household.

2) Recursive models

▪ A model is called recursive if its structural equations can be

ordered in such a way that the first equation includes only
predetermined variables in the right hand side. The second

Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 8

Chapter 1 SIMULTANEOUS EQUATIONS MODELS

equation contains predetermined variables & the endogenous

variable of the 1st equation in the right hand side & so on. i.e.

and so on.

▪ Example: As an example of recursive system, one may

postulate the following model of wage & price determination.

Price equation:
Wage equation:
Properties:
1. OLS can be applied straight away on each equation to
estimate the parameters.

Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 9

Chapter 1 SIMULTANEOUS EQUATIONS MODELS

2. OLS estimates on the parameters of recursive model are

best & unbiased.
3. There is no independency among the endogenous variable
in recursive model.
4. The same time period disturbances in different equations
are uncorrelated. This is the assumption of zero
contemporaneous correlation.

1.3 Identification

▪ When we estimate a model by OLS, the key identification

condition is that each explanatory variable is uncorrelated with
the error term.
▪ This condition no longer holds, in general, for SEMs.
However, if we have some instrumental variables, we can still

Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 10

Chapter 1 SIMULTANEOUS EQUATIONS MODELS

identify (or consistently estimate) the parameters in an SEM

equation.
▪ Let’s begin illustrating identification conditions using the
normal supply and demand equations.

▪ Identification of demand and supply equations

▪ If z is changing then the supply curve will shift upwards or

downwards; z is a supply shifter. So, the change in z will trace
out the demand curve.
Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 11
Chapter 1 SIMULTANEOUS EQUATIONS MODELS

▪ If x is changing, then the change in x traces out the supply

curve. X is a demand shifter.

▪ If there is a variable(s) in the demand equation which doesn’t

exist in the supply equation then the change in that variable
traces out the supply equation and the supply equation is said
to be identified (over identified). The same works for a variable
(s) in the supply equation which doesn’t exist in the demand
equation.
Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 12
Chapter 1 SIMULTANEOUS EQUATIONS MODELS

Mathematical way for identification

▪ Given a partial equilibrium model of demand and supply;

▪ A new equation is formed by linearly combining the

parameters of the two equations (where :

▪ In a precise form, it can be re-written as:

▪ This equation is statistically the same as the demand equation.

If this is so then we say the demand equation is said to be
unidentified. Statistically, this equation is different from the
Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 13
Chapter 1 SIMULTANEOUS EQUATIONS MODELS

supply equation (since there is no in the supply equation)

hence the supply curve is identified.
✓ If , then the supply equation is exactly the same
as the precise form so that the supply equation will not
be identified.
✓ If , then the demand equation can be identified.

▪ In general, an equation is identified, provided it is not possible

to derive (by taking a linear combination of equations in the
model) an equation of exactly the same form as the equation
being considered.

▪ If an equation is not identified, there is no way in which

unbiased or even consistent estimator of its parameters may be
obtained.

Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 14

Chapter 1 SIMULTANEOUS EQUATIONS MODELS

Order condition for Identification

▪ For a model, to be identified, it has to fulfill both the order and

rank conditions. Order condition is a necessary condition for
identification but not a sufficient condition.
▪ Order condition: an equation is identified if the number of
variables left out of the equation (number of restrictions or
excluded variables) is larger or equal to the number of
equations in the system minus one.
▪ Suppose that M is the number of equations in the system and J
is the number of variables left out of the equation. Then,
If J=M-1, then the equation is exactly identified.
If J>M-1, then the equation is over identified.
If J<M-1, then the equation is not identified.

Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 15

Chapter 1 SIMULTANEOUS EQUATIONS MODELS

Example:

▪ Given the following system of four equations:

(Demand eq. for good 1)
(Supply eq. for good 1)
(Demand eq. for good 2)
(Supply eq. for good 2)
▪ Find if each of the above equations are identified.

Solution
Equation 1: J<M-1 (2<3) not identified (under identified). So, we
can’t estimate the price elasticity of demand.
Equation 2: J=M-1 (3=3) exactly identified. So, we can estimate
the structural parameters.
Equation 3: J<M-1 (2<3) not identified.
Equation 3: J<M-1 (4>3) over identified.
Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 16
Chapter 1 SIMULTANEOUS EQUATIONS MODELS

Rank condition for identification

▪ An equation j in a system is identified iff the matrix has a

rank equal to the number of equations in the model minus one
(M-1). This is rank condition for identification (sufficient
condition).
▪ The matrix contains the coefficients in the other equations of
the variables that are excluded from equation j.
▪ If equation j has j exclusion restriction, the matrix is of
dimension M-1xj.
▪ Rank of can never exceed M-1 (because it has only M-1
rows)

Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 17

Chapter 1 SIMULTANEOUS EQUATIONS MODELS

Example

▪ Using the above example, let’s examine the rank condition for
identification:

(Demand for good 1)

(Supply for good 2)
(Demand for good 2)
(Supply equation for good 2)

Four endogenous variables: and

Two exogenous variables: and

Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 18

Chapter 1 SIMULTANEOUS EQUATIONS MODELS

Solution

▪ First, form the matrix of coefficients associated with all the

exogenous and endogenous variables in the system of
equations:
1
(intercept)
1 -1 0 0
2 -1 0 0
3 0 -1 0
4 0 -1 0 0 0

Equation 1: J<M-1 (2<3) not identified.

Equation 2: J=M-1 (3=3) exactly identified.
Equation 3: J<M-1 (2<3) not identified.
Equation 4: J>M-1 (4>3) over identified.
Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 19
Chapter 1 SIMULTANEOUS EQUATIONS MODELS

▪ For equation 1: , Rank= 2<M-1=3 (did not

pass), and J=2<M-1 (under identified).

▪ For equation 2: , J=3=M-1 (necessary

condition for identification), Rank( =M-1=3 (sufficient

condition for identification).

▪ For equation 3: , J=2<M-1=3 (under

identified), rank( .

Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 20

Chapter 1 SIMULTANEOUS EQUATIONS MODELS

▪ For equation 4: , J=4>M-1=3

(necessary condition: over identified), Rank ( )=3 (sufficient

condition for identification).
▪ Equation 2 and 4 are identified. You can use ILS for equation 2
but not for equation 4 since when you formulate the reduced
form, you will find two formula and hence two values for so
that you can’t have a unique solution.

1.4 Estimation

▪ If an equation is identified we can try to estimate using some

appropriate method. An ILS provides consistent estimates if we
can derive the structural parameters from the reduced form
parameters. However, first look at the identification conditions:
Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 21
Chapter 1 SIMULTANEOUS EQUATIONS MODELS

✓ If exactly identified, ILS and 2SLS are the same

✓ If over identified, ILS is consistent but does not give a
unique estimate. Hence better to use 2SLS or IV and it
is consistent and more efficient.

Estimation Methods

a) Indirect Least Squares

▪ It involves estimating structural parameters by transforming
the reduced form parameters and is already discussed above.

▪ Steps to Indirect Least Squares:

1.Algebraically rewrite the structural equations in reduced
form.
2.Estimate the reduced form equations.

Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 22

Chapter 1 SIMULTANEOUS EQUATIONS MODELS

3.Algebraically calculate formulas for each structural

parameter in terms of the reduced form parameters.
4.Compute the formulas using the reduced form estimates.

b) Instrumental Variable (IV) estimation

i) Arbitrary search and test Method
▪ As the name indicates, this variation of IV method
involves an arbitrary search and use of instrument
variables, which are strongly correlated with the
endogenous variable but not with the error term.
▪ The selection of instruments in this case emanates from
theoretical and empirical expositions. However, this
approach has a weakness in that we may not end up
selecting the best instruments for the endogenous
variables.
▪ Hence, 2SLS method is usually preferred.
Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 23
Chapter 1 SIMULTANEOUS EQUATIONS MODELS

ii) Two Stage Least Squares (2SLS) Method

▪ This method offers an excellent direct estimation
method in the case of exactly or over-identified
equations. Compared to ILS, 2SLS provides unique
estimates of each structural parameter in the over
identified equations.
▪ Two steps:
1) The first stage involves the creation of an
instrument by estimating the reduced form
equations for p (endogenous Variable) to get its
fitted value, .
2) The second stage involves a variant of instrumental
variables estimation. Replace p by in the supply
equation and use OLS for estimation. It is in fact a
special way and less arbitrary way of doing IV
estimation.
Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 24
Chapter 1 SIMULTANEOUS EQUATIONS MODELS

c) Seemingly Unrelated Regression (SUR)

▪ It is also called joint generalized least squares (JGLS) or

Zellner estimation. It is a generalization of OLS for multi-
equation system.
▪ Like OLS, the SUR method assumes that all the regressors
are independent variables, but SUR uses the correlations
among the errors in different equations to improve the
regression estimates.
▪ The SUR method requires an initial OLS regression to
compute residuals. The OLS residuals are used to estimate
the cross-equation covariance matrix.
▪ For SUR to be effective, the models must use different
regressors. SUR produces the same result as OLS unless
the model contains at least one regressor not used in the
other equations.
Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 25
Chapter 1 SIMULTANEOUS EQUATIONS MODELS

d) Three Stages Least Squares Method (3SLS)

▪ It generalizes the 2SLS method to take account of the
correlations between equations in the same way that SUR
generalizes OLS. 3SLS requires 3 steps:
▪ 1st stage regressions to get predicted values for the
endogenous regressors; a two-stage least squares step to get
residuals to estimate the cross-equation correlation matrix;
and the final 3SLS estimation step. 3SLS may produce the
same output as 2SLS (unless the model contains at least
one regressor not used in the other equations) for the same
reason that differentiates SUR from OLS. However, the
application of this principle is more complex when IVs are
used. When all the exogenous variables are used as
instruments, linear combinations of all the exogenous
variables appear in the third stage regressions through
substitution of first stage predicted values.
Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 26
Chapter 1 SIMULTANEOUS EQUATIONS MODELS

e) Limited/Full Information Maximum Likelihood Method

▪ This is beyond the scope of the course and left for more
advanced/specialized courses in econometrics.

Test of Endogeneity/Simultaneity
▪ When simultaneity in the model, OLS is biased and
inconsistent but 2SLS is unbiased and efficient. But if there is
no simultaneity, 2SLS and IV estimates are consistent but not
efficient. Hence, test for endogeneity/simultaneity is necessary
before using 2SLS and/or IV estimation method.
▪ Test for simultaneity is to test whether the regressors are
correlated with the error term or not.
✓ Use Hausman WU tests for endogeneity.
Example of Endogeneity Test

Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 27

Chapter 1 SIMULTANEOUS EQUATIONS MODELS

▪ and are endogenous variables while and are

exogenous variables.
▪ Particularly is suspected to be endogenous hence test of
endogeneity is needed.
▪ Since the demand function is not identified and the supply
function is identified (i.e. over identified as and are IVs
), our interest is to estimate the latter.

Mathematical approach for test of endogeneity

▪ First, find the reduced form equations:

▪ Using OLS, you get:

Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 28

Chapter 1 SIMULTANEOUS EQUATIONS MODELS

▪ Substituting this result in the supply function:

▪ Rename the coefficient of as for analytical purpose. Thus,

▪ By construction, and . Substituting these

equalities in the above equation gives:

▪ Therefore, if there is no simultaneity, , that is, the

coefficient of becomes zero. But, if there is simultaneity,
, that is, the coefficient of is different from zero. So,
test of endogeneity is equivalent to testing the coefficient of
in the structural equation.

Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 29

Chapter 1 SIMULTANEOUS EQUATIONS MODELS

Steps
1. Specify your hypothesis
: No simultaneity i.e. (coefficient of is zero)
: reject i.e.
2. Find the reduced form equation and regress on and
and obtain estimate of .
3. Regress on and estimate of and perform t-test. If
the test suggests the coefficient of is different from zero,
then you reject .

Abrham Seyoum (PhD) Yom Institute of Economic Development (YIED) 30