A Guide to Simulation with EVIEWS1

(Version 7.0)
Afees A. Salisu
1.0

2.0

Preamble

Simulation involves solving for unknown values of the endogenous
variables in a model. A model in Eviews is a set of simultaneous
equations that are used for forecasting and simulation. Unlike other
objects in Eviews that contain system of equations, models do not
contain unknown coefficients to be estimated. Instead, models use
estimated coefficients and projected values of the exogenous variables to
simulate the unknown values for the endogenous variables.
Creating a System of Equations

First, we need to open an Eviews workfile: click File / New / Workfile on
the Toolbar of the Main Window

Contacts: Department of Economics & Centre for Econometric & Allied Research, University of
Ibadan, Ibadan, Nigeria. Email: aa.salisu@mail.ui.edu.ng; aa.salisu@cear.org.ng. Phone:
+2348034711769. Previous versions of this guide using Eviews 3.1 and 4.1 were respectively
written by Mathieu Lequain and Jung Hoon Kim.
1

 A Workfile Range window appears

Choose Workfile structure type, Frequency, Start date and End date
 When you click OK, an Untitled Workfile appears with 2 items: C
and RESID


Load your data into Eviews

We can use the drag & drop method since all the variables are of
interest to us (see my presentation on Getting Started with Eviews 7).
The Eviews workfile is updated to include the new variables as shown
below:

The new variables included are cn - consumption, i - investment, m

money supply, pc personal consumption, ir interest rate, and y
income.

Click Objects in the Workfile toolbar / New Object / System. When you
click OK, an empty System Window should appear. In the dialog that
appears, type in a name for your model (say CEAR)

3.0

SPECIFYING A MODEL
3.1

Before Entering Equations

Eviews is not case sensitive and neither does it distinguish between
subscript and superscript.
In other words, CntS is identical to CNTS in Eviews.
To enter an equation with t lags , we use the sign (-t) after the
variable.
For example, we have:
cn = c(1)+c(2)*cn(-1)+c(3)*y(-1)+c(4)*rir(-1)+c(5)*rir(-2)
We can also use the command d(cn) which is equivalent to
cn-cn(-1).
So cn=d(cn)+cn(-1)
5

3.2

In addition to the behavioural equations, some models may have

identities that allow the model to solve for the endogenous
variables.
Before trying to solve the model, we must make sure we have at
least as many equations as there are endogenous variables to solve
for in the model. We must set up our equations in a way that each
equation has a different endogenous variable as the first series on
the left-hand side.
Entering Equations

As an example, a system of equations in the System Window looks

like the following:

Note the following about the model:

 There are three behavioural equations
 C(1) to C(16) represent the structural parameters. It is more
convenient to number the coefficients serially as demonstrated
above.
6

 PC, I and IR are endogenous variables; Y(-1), PC(-1), IR(-1), and

IR(-2) are predetermined variables while M and Y are exogenous
variables.

The estimation output obtained after applying OLS should appear like
the one shown below:

We have been previously taught how to estimate simultaneousequation models. Nonetheless, to estimate our model with OLS, click
estimate from the System Window / choose OLS estimator / OK

4.0

Model Simulation

This will require using the estimated coefficients above to predict all the
endogenous variables. Also, it allows for the consideration of different
scenario analyses. The following steps are involved:
4.1 Make a Model
From the estimation output, click Proc / Make Model. This shows the
system of equations specified.

To view all the equations, click Proc / Links / Break All Links
Note that the variables on the left hand side of the equations are the
endogenous variables while those on the right hand side are the
explanatory variables.

To view all the equations with their estimated coefficients, Click text
Note that these are the coefficients obtained after applying OLS.

4.2 Solving a Model

Having viewed the estimated equations, the next stage is to solve for the
endogenous variables. To do this, click on the solve button in the model
window button bar.

Let us consider the basic options shown in the dialog box above:
 Simulation type:
 Deterministic: This involves solving the model without simulating
the residuals. In addition, all the estimated coefficients are used in
the simulation at their point estimates likewise all the exogenous
variables are held constant.
 Stochastic: With this option, the equations of the model are solved
and simulated with residuals. Also, the coefficients and exogenous
variables of the model are varied randomly. For stochastic
simulation, the model solution generates a distribution of
10

outcomes for the endogenous variables in every period. We

approximate the distribution by solving the model many times
using different draws for the random components in the model
then calculating statistics over all the different outcomes.
As a preliminary exercise, it will be more appropriate to consider
the two options in order to have an idea of the sensitivity of the
results to various sorts of error.

 Dynamics
 Static solution: This is usually used when dealing with in-sample
simulation where the predicted values of the endogenous variables
in model can be compared with the historical data. Of course, this
will require using the actual values of all the explanatory variables
(both the exogenous and the lagged endogenous variables of the

11

model). The predictive ability of the model can be evaluated based

on the in-sample simulation.
 Dynamic solution: This is the option for Out-of-Sample
Simulation. Under this, we have to forecast all the exogenous
variables. The future paths of the exogenous variables may be
based on the likely policy direction of the economy as to what will
actually happen. Statistically, there are different approaches of
forecasting time series ranging from the Naive, Simple Average,
Simple Moving Average, Weighted Moving Average, Trend
Projection to Exponential Smoothing methods. These methods are
briefly discussed below.
(i)

Naive Method: The forecast for next period (say period T+1)
will be equal to the current periods ( period T) value.
For example:


YT +1 = YT

YT +2 = YT +1

YT +n = YT +n1

(ii) Simple Average Method: The forecast for next period (say
period T+1) will be equal to the average of all past historical
values.
For example:


YT +1

1
=
T

Y
t =1

12

YT + 2

1 T +1
=
Yt

T + 1 t =1

1 T+n1
YT+n =
Yt

T + n 1 t=1

(iii) Simple Moving Average Method: The forecast for next period
(say period T+1) will be equal to the average of a specified
number of the most recent observations, with each observation
receiving the same emphasis (weight).
For example:

YT + YT 1 + YT 2
3
Y + Y + YT 1
= T +1 T
3

YT +1 =

YT + 2

YT +n =

YT +n1 + YT +n2 + YT +n3

3

(iv) Weighted Moving Average Method: The forecast for next period
(say period T+1) will be equal to a weighted average of a specified
number of the most recent observations.
For example:


YT +1 = w1YT + w2YT 1 + w3YT 2

w1 + w2 + w3 = 1;
13

we can assume
Therefore;

w1 = 0.5; w2 = 0.3; w3 = 0.2

YT+1 = 0.5YT + 0.3YT1 + 0.2YT2

YT +2 = 0.5YT +1 + 0.3YT + 0.2YT 1

YT+n = 0.5YT+n1 + 0.3YT+n2 + 0.2YT+n3

(iv) Trend projection method: This involves regressing the variable

under consideration on a time trend.
That is:

Y = a + bTrend

The trend term can also be non-linear depending on the behaviour

of the series being examined.
(v) Exponential smoothing method: The new forecast for next
period (period t) will be calculated as follows:
New forecast = Last periods forecast + (Last periods actual
value Last periods forecast)
Ft = Ft-1 + (At-1 Ft-1)
Or
Ft = At-1 + (1-)Ft-1
Where is a smoothing coefficient whose value is between 0
and 1; Ft denotes forecast at period t and At is the actual at
period t

14

 Solution sample:
This is the simulation period. For the in-sample simulation, the
estimation period is usually the simulation period. However, due
to lagged variables in the model, one must adjust this period so as
to avoid initializing the solution with missing values. In the case of
our model, we set the sample to 1982 2010. Why?
For out-of-sample simulation period however, this rests on the
underlying objective of the model. Generally, it captures n-period
ahead forecasts. For the purpose of our exercise, we set the sample
to 2011 to 2012 (implying a 2-year ahead forecast). Thus, the period
initially indicated in the workfile will now be extended to 2012.
How?

Solver: This is essentially the algorithm used to solve the model.

In the literature there are three approaches of solving a model
period. They are:
 Gauss-Seidel: This is an iterative algorithm, where at each iteration
we solve each equation in the model for the value of its associated
endogenous variable, treating all other endogenous variables as
fixed. The Gauss-Seidel less computationally demanding when
compared to other methods and it performs well on most
econometric models.
 Newton: This algorithm can handle a wider class of problems than
Gauss-Seidel, but requires considerably more working memory
and has a much greater computational cost when applied to large
models.
 Broyden: This is a modification of Newton's method (often referred
to as a quasi-Newton or secant method) and requires more
iterations to converge to a solution than the Newton. However, the
Broyden spends less than the time required to solve the model by
Newton's method.


15

In any case, for a typical macro-econometric model, the Gauss-Seidel

technique is preferred and its application has remained prominent in
macro-econometric literature.

Let us consider a Static-Deterministic Simulation for our exercise:

Deterministic / Static solution / 1982 2010 / Solver / Gauss-Seidel


By clicking OK, the Model solution is displayed as shown below

16



You will observe that the solution is already saved as

CEAR_SOLVE
Check the workfile to view the baseline results for the endogenous
variables. The updated workfile should appear like this:

17


Note that _0 denotes Static-Deterministic Solution and it

distinguishes the baseline from the actual

We can compare the actual with the baseline graphically. To do
this, double click the filename for model solution cear_solve then
click Proc / Make Graph

18

11
10
9
8
7
6
5
4
80

82

84

86

88

90

92

94
I

19

96

98

00

I (Baseline)

02

04

06

08

10

12

40
35
30
25
20
15
10
5
80

82

84

86

88

90

92

94
IR

96

98

00

02

04

06

08

10

12

04

06

08

10

12

IR (Baseline)

13
12
11
10
9
8
7
6
5
80

82

84

86

88

90

92
PC

94

96

98

00

02

PC (Baseline)

If we intend to consider another option for model solution, we

need to create a scenario for ease of comparison. This allows us to
store the results obtained from this new option uniquely while also
retaining the results from the previous options. Scenario 1 is
created by default. Therefore, if we are considering more than one
20

additional option, we have to create relevant scenarios in order to

have a unique label for each of the options.
An Illustration:

Recall, we have previously considered Static-Deterministic
Simulation

One may be interested in comparing the results with StaticStochastic Simulation:

To do this, we are to recall the model solution by double clicking
cear_solve and thereafter, click solve

The model solution is provided below:

For each of the endogenous variables, both the mean and standard
deviation are simulated.

21


The updated workfile is shown below:

22


Note _0m and _0s are for Static-Stochastic Solution.

Let us now consider out-of-sample simulation. Under this, we can
evaluate the impact of different scenarios on the endogenous
variables.

Recall, we have to first of all determine the future paths for all the
exogenous variables.

In the case of our model, we have two exogenous variables namely
Income (y) and Money Supply (m).

We can choose y as our policy variable while we allow m to follow
its natural pattern (Do nothing).
 We can assume a three-year weighted moving average for
Money Supply with the most recent year, 0.5; year prior to
that, 0.3; year prior to that, 0.2:
23

The formula: m=0.5*m(-1)+0.3*m(-2)+0.2*m(-3)

 To do this in Eviews, from the workfile window click genr
and type the formula.
 Note that our out-of-sample period is 2011 to 2012

We can then assume the following scenarios for y:

 Annual Increase in y by 5%
 Annual Decrease in y by 5%
 Allow y to follow its natural path (Do nothing)
Taking each in turn:

To increase y by 5%; we can as well use the genr

The formula: y= 1.05*y(-1)

24



We can now simulate the endogenous variables given the projected

future paths of exogenous variables.
To do this, we need to first create a scenario to distinguish our
results from the previous ones.
We can use the default scenario (scenario 1) for the solution and
therefore simulation results produced under this will be suffixed
by _1
Let us now proceed to the out-of-sample simulation
From the model solution cear_solve,
Click solve /Deterministic / Dynamic solution / 2011 2012 / Solver
/ Gauss-Seidel / OK

25

By viewing the workfile, we can see that the number of objects has
increased to include the new results. Note the variables with _1;
they represent the out-of-sample forecasts for our endogenous
variables.

26


We can follow the same procedure for the second scenario which is to
reduce income by 5%

The formula: y=0.95*y(-1)
 That is, from the workfile window, click genr and type the given
formula

Now we have to create Scenario 2 for the solution obtained here.

Double click cear_solve and then click Scenario / Create New
Scenario / OK
 Thereafter, click solve /Deterministic / Dynamic solution / 2011
2012 / Solver / Gauss-Seidel / OK

27


The updated workfile is shown below with new results suffixed by _2

28


Allowing y to follow its natural path:

Like m, we can also assume a weighted moving average for y given
as: y=0.5*y(-1)+0.3*y(-2)+0.2*y(-3)

Again, from the Model solution cear_solve, click Scenario / Create

New Scenario / OK

This gives Scenario 3 and the simulation results are suffixed by _3

Then, click solve /Deterministic / Dynamic solution / 2011 2012 /
Solver / Gauss-Seidel / OK

29


We can easily tabulate the simulation results and examine how the
endogenous variables respond to the three scenarios.

Period I_1
I_2
I_3
IR_1
IR_2
IR_3
PC_1
PC_2
PC_3
2010 10.01196 10.01196 10.01196 22.50833 22.50833 22.50833 12.30242 12.30242 12.30242
2011 15.98835 5.437221 10.39755 19.71778 25.89938 22.99327 12.86602 12.21942
12.5234
2012
21.5097 0.407454 10.44295 17.53704 27.99791 23.01273 13.65699 11.89575 12.73079

Note the following about the results:

 An increase in y by 5% leads to a higher level of consumption
and investment and a reduction in interest rate
 Conversely, a decrease in y by the same magnitude reduces the
level consumption and investment and increases the rate of
interest
30

 Thus, increasing the level of income appears to have beneficial

effects on the selected macroeconomic variables than reducing
it.
 Also, increasing the level of y is preferred to leaving it to grow
naturally.
 However, if the economy is not buoyant enough as to increase
the level of y by 5%; then, allowing it to follow its natural path is
preferred to reducing it by 5%.

I_1 > I_3 > I_2

IR_1 < IR_3 < IR_2

PC_1 > PC_3 > PC_2
 Participants may consider other plausible scenarios to demonstrate
their understanding of a typical macro-econometric model used for
policy analysis.
 Enjoy!!!!

31