SIMULTANEOUS EQUATION MODELS

All the regression models considered so far have been single equation regression models in that a single
dependent variable (Y) was expressed as a function of one or more explanatory variables (the Xs’). The
underlying economic theory determined why Y was treated as the dependent variable and the Xs’ the
determining or causal variables. In such models an implicit assumption is that the cause-and-effect
relationship, if any, between Y and the Xs’ is unidirectional: the explanatory variables are the cause,
and the dependent variable is the effect.
However, there are situations in which such a unidirectional relationship between Y and Xs’ cannot be
maintained. It is quite possible that the Xs’ not only affect Y but Y can also affect one or more Xs’. If
that is the case, we have a bilateral , or feedback, relationship between Y and the Xs’. Obviously, if this
is the case, the single equation modelling strategy will not suffice, and in some cases, it may be quite
inappropriate because it may lead to biased (in the statistical sense) results. To take into account the
bilateral relationship between Y and the Xs’, we will therefore need more than one regression equation.
Regression models in which there is more than one equation and in which there are feedback
relationships among variables are known as simultaneous equation regression models.
In contrast to single equation regression models, in simultaneous regression models what is called a
dependent (endogenous) variable in one equation appears as an explanatory variable in another
equation. Thus, feedback relationship between the variables create the simultaneity problem, rendering
OLS inapplicable to estimate the parameters of each equation individually. This is because the
endogenous variable that appears as an explanatory variable in another equation may be correlated with
the stochastic error term of that equation. This violates one of the critical assumption of the OLS method
that the explanatory variable be either fixed, or, non-random, or if random, it be uncorrelated with the
error term. Because of this, if one uses OLS, the estimates then obtained will be biased as well as
inconsistent.
THE NATURE OF SIMULTANEOUS EQUATION MODELS
The best way to proceed is to consider some examples from economics.
Example A: The Keynesian model of income determination
Let C stand for consumption (expenditure) , Y for income, I for investment (expenditure) , I for
investment (expenditure) and S for savings. The simple Keynesian model of income determination
consists of the following two equations:
Consumption function: Ct = B1 + B2 Yt + u t (1)

Income identity : Yt = Ct + I t (2)

where t is the time subscript, u is the stochastic error term,

and
I t = St .

Page 1 of 8
This simple Keynesian model assumes a closed economy (i.e., there is no foreign trade ) and no
government expenditure [recall that the income identity is generally written as:
Yt = Ct + It + G t + NXt ,
where G is the government expenditure and NX is the net export (Export - Import)].
The model assumes that I , the investment expenditure, is determined exogenously, say, by the private
sector.
The consumption function states that consumption expenditure is linearly related to income, the
stochastic error term is added to the function to reflect the fact that in empirical analysis the
relationship between the two is approximate. The national income identity says that the total income
is equal to the sum of consumption expenditure and investment expenditure, the latter is equal to
total savings. The slope coefficient , B2 , in the consumption function is the marginal propensity to

consume (MPC), the amount of extra consumption resulting from an extra dollar (or any appropriate
monetary unit) of income. Keynes assume that MPC is positive but less than one, which is reasonable
because people may save part of their additional income.
Now we can see the feedback, or simultaneous relationship between consumption expenditure and
income. From (1) we see that income effects consumption expenditure, but from (2) we also see that
consumption is a component of income. Thus, consumption expenditure and income are
interdependent. The objective of the analysis is to find out how consumption (expenditure) and
income are determined simultaneously. Thus, consumption and income are jointly dependent
variables. In the language of simultaneous equation model, such jointly dependent variables are
known as endogenous variables. In the simple Keynesian model, investment , I, is not an endogenous
variable, for its value is determined independently; so it is called exogenous or predetermined
variable. In more refined Keynesian models, investment can also be made endogenous.
In general, an endogenous variable is a variable that is an inherent part of the system being studied
and that is determined within the system. An exogenous variable is a variable entering from and
determined from outside the system being studied.
(1) and (2) represent a two-equation model involving two endogenous variables, C and Y. If there are
more endogenous variables, there will be more equations, one for each endogenous variable. Some
equations in a system are structural or behavioural equations and some are identities. A structural
equation depicts the structure or behaviour of a particular sector of the economy. It expresses the
endogenous variables as functions of other endogenous variables , predetermined variables and
disturbances. An identity is a relationship that is true by definition. The coefficients (or parameters) of
the structural equations are known as structural coefficients. In the present example, (1) represents

Page 2 of 8
consumption sector, therefore, it is a structural equation. (2) is an identity which states that total
income is equal to total consumption plus total investment.
EXAMPLE B: Demand and supply model:
The price P of a commodity and quantity Q sold are determined by the intersection of the demand
and supply curves for that commodity. Thus, assuming for simplicity that the demand and supply curve
are linearly related to price and adding the stochastic , or random error, term u 1 and u 2 , we may

write the empirical demand and supply functions as:

d
Demand function: Q t = A1 + A 2 Pt + u1t (3)
s
Supply function : Q t = B1 + B2 Pt + u 2t (4)
d s
Equilibrium condition : Q t = Q t (5)
d s
where Q t is quantity demanded, Q t is quantity supplied, and t is time.

According to economic theory, A2 is expected to be negative and B2 is expected to be positive. (3)

and (4) are both structural equations, the former representing the consumers and latter the suppliers.
The As and Bs are structural coefficients.
Now it is not too difficult to see why there is a simultaneous, or two-way relationship between P and
Q. If for example, u1t in (3) changes because of changes in other variables affecting demand(such as

income, wealth , and tastes), the demand curve will shift upward if u1t is positive and downward if

u1t is negative. As Figure 1.1 shows that a shift in the demand curve changes between P and Q.

Page 3 of 8
 Figure 1.1 : Interdependence of Price and Demand

Similarly a change in u 2t (because of strikes, weather, hurricane) will shift the supply curve again

shifting both P and Q. Therefore, there is a bilateral , or simultaneous relationship between the two
variables, the P and Q variables are then jointly dependent or endogenous variables. This is known as
simultaneity problem.
Besides the simultaneity problem, a simultaneous equation model may have an identification
problem. An identification problem means that we cannot uniquely estimate the values of the
parameters of an equation . If we can estimate the parameters of a particular equation (be it a demand
function or a supply function), we say that the particular equation is exactly identified. If we cannot
estimate the parameters, we say that the equation is under identified. Sometimes it can happen that
there is more than one numerical value for one or more parameters of that equation , in that case we
say that the equation is over identified. Therefore, before one estimates a simultaneous equations
model, one must find out if an equation in such a model is identified.

Page 4 of 8
 RULES FOR IDENTIFICATION: THE ORDER CONDITION OF IDENTIFICATION
To understand the order condition of identification, we introduce the following notations:
m : number of endogenous (or jointly dependent) variables in the model.
k : total number of variables (endogenous and exogenous) excluded from the equation under
consideration.
Then,
1) If k = m-1, the equation is exactly identified.
2) If k > m-1, the equation is over identified.
3) If k < m-1, the equation is under identified.
If an equation is under identified , it is a closed-end case. There is not much one can do, other than
changing the specification of the model (i.e., developing another model).

Estimation of an exactly identified equations

If an equation is exactly identified, we can estimate it by the method of indirect least squares (ILS) ,
ILS is a two-step procedure. In step 1, we apply OLS to the reduced form equations of the model, and
then the original structural coefficients are retrieved from the reduced form coefficients. The reduced
form of a structural equation is the model in which endogenous variables are expressed as a function of
predetermined (or exogenous variables) only.

An Example: Consider the model :

Qdt = A1 + A 2 Pt + A3X t + u1t

Qst = B1 + B2 Pt + u 2t
Qdt = Qst
X = income of the consumer (exogenous)
From the above model:
A1 + A2Pt + A3Xt + u1t = B1 + B2Pt + u 2t
Solving this equation we obtain the following equilibrium value of Pt :

Pt = 1 + 2Xt + v1t (6)

where the reduced form coefficients are:

B1 − A1 −A3 u − u1t
1 = , 2 = , v1t = 2t .
A 2 − B2 A 2 − B2 A 2 − B2
Substituting the equilibrium value of Pt into the preceding demand or supply function, we obtain the

following equilibrium:

Page 5 of 8
Qt = 3 + 4Xt + v2t , (7)

where

A 2 B1 − A1B2 − A 3 B2 A u − B2 u1t
3 = , 2 = , v2t = 2 2t .
A 2 − B2 A 2 − B2 A 2 − B2
Since equations (6) and (7) are both reduced form regressions, OLS can be applied to estimate their
parameters. The question that remains is whether we can uniquely estimate the parameters of the
structural equations form from the reduced form coefficients.
Observe that the demand and supply models contain five structural coefficients
A1, A2 , A3 , B1 and B2 . But we have only four equations to estimate them, i.e., the four reduced
form coefficients – the s. So we cannot obtain unique values of all the four structural coefficients.
We, however, find that B1 and B2 can be uniquely estimated and are given as:

B1 = 3 - B21
and

4
B2 = .
2
Therefore, the supply function is exactly identified. But the demand function is under identified
because there is no unique way of estimating its coefficients, the A coefficients.
ILS estimators are consistent.

Estimation of an over identified equations

The parameters of the over identified equation can be estimated by the method of two-stage least
squares (2 SLS). The basic idea behind 2 SLS is to replace the explanatory variable that is correlated
with the error term of the equation in which that variable appears by a variable that is not so
correlated. Such a variable is called a proxy or instrumental variable. 2 SLS estimators like ILS
estimators are consistent estimators.
An Example: The Method of 2 SLS:
Consider the following model:
Income function: Yt = A1 + A2Mt + A3I t + A4G t + u1t .

Money Supply function: Mt = B1 + B2Yt + u 2t ,

where
Y : income,
M : stock of money,

Page 6 of 8
G: Govt. expenditure on goods and services,
u 1 , u 2 : stochastic disturbance terms.
In this model, the variables I and G are assumed exogenous.
The income function states that income is determined by the money supply, investment expenditure
and Government expenditure. The money supply function states that the stock of the money sipply is
determined on the basis of the level of income. Obviously, we have the simultaneity problem here
because of the feedback between income and the money supply.
Since the income equation is under identified, there is nothing we can do to estimate its parameters.
However, the money function is over identified. So in order to estimate B1 and B2 we shall use the

method of 2 SLS-the method involving two successive applications of OLS. The process is as follows:
Stage 1: To get rid of the likely correlation between income Y and the error term u 2 , first regress

Y on all predetermined variables in the whole model , not just that equation. In the present case, this
means regressing Y on the predetermined variables I (gross private domestic investment) and G
(Government expenditure) as follows:
Yt = 1 + 2It + 3G t + w t , (8)

where w is a stochastic error term.

From the equation above, we obtain
ˆ =
Y ˆ +
ˆ I +ˆ G ,
t 1 2 t 3 t

ˆ s indicate the
where Ŷt is the estimated value of Yt given the values of I and G at time t. 

estimated values of the true s .

We can re-write (8) as
ˆ +w .
Yt = Yt t

ˆ and w are therefore uncorrelated.

Following the OLS theory, Y
Stage 2: The over identified money supply function can now be written as:

M t = B1 + B2 Yt (
ˆ +w +u
t 2t )
ˆ +v ,
= B1 + B2 Y (9)
t t

where v t = u 2t + B2 w t .

Since Ŷ is uncorrelated with v t asymptotically , we can now apply OLS to (9) which will give consistent

estimators of the parameters of the money supply function.

Page 7 of 8
References
Gujarati, D.N, and Porter, D.C.(2010). Essentials of Econometrics, McGraw- Hill Irwin, 4th edition.
Kmenta, J.(1997). Essentials of Econometrics, The University of Michigan Press, 2nd edition.

Page 8 of 8