File: WINWORD\ECONMET\PcGive exercises\Lab2.doc
APPLIED ECONOMETRICS: COMPUTER SESSION 2
To complete this exercise, you will need the GiveWin file saved from the previous exercise (that should have the
filename Nic2.in7).
OBJECTIVE
We estimate a parsimonious, restricted regression in which real total consumers' expenditure is explained by real
personal disposable income and real wealth. We examine a little further some of techniques in PcGive for
evaluating the statistical quality of our estimated model. In particular, attention is given to the powerful graphical
procedures in PcGive for testing parameter constancy.
PROCEDURE
Ensure that all estimations are done over the sample period 1970 Q1 to 1988 Q4, except where you are explicitly
asked to use another sample period or where the sample has to be set differently to perform a particular
diagnostic test.
1. Estimate by Ordinary Least Squares (OLS) the restricted dynamic model given by the regression of DLCE on
an intercept, DLYt, DDLYt-1, LYt-1, DLNWt-1, LNWt-1, LCEt-1 and the composite dummy variable, DUM. In
doing this estimation, change the defaults for the output that PcGive produces in the following way. From Model,
select Options. Tick Test summary, and also tick Instability Tests (see manual pages 219, 220 for details),
HCSE standard errors (see page 218 of manual for details on these), and finally tick Information Criteria.
These will now be your defaults for the rest of this session.
The model we estimate can be written as:
LCE t  1   2 LYt   3LYt1   4 LYt1  5LNWt1   6 LNWt1
 7 LCE t1  8 DUM t   t
2. Does the estimated regression have a meaningful economic interpretation, are the signs of the parameter
estimates plausible, and is it statistically well-specified?
3. Report and interpret the static long run equilibrium solution to this model. (Calculate this manually [with a
calculator] from the regression output from Question 1). (Why does the automatic facility in PcGive for obtaining
a static long run solution give us unhelpful results in this case?).
4. Examine the two “forecast” statistics given automatically as part of the OLS regression output. These are
labelled as “Forecast” and “Chow” respectively. How can they be interpreted? (See PcGive manual, Vol 1, pages
46 and 222-223). What conclusions do we draw from these statistics about forecast accuracy and/or parameter
constancy in this case?
RECURSIVE REGRESSION
(Information about recursive estimation, and the graphical output provided by the package, can be found in the
manual in Chapters 6 and 13).
5. From the PcGive menu, select Model then Formulate. Choose the same model as in Step 1. Select Estimate,
and select the same sample period, but this time do not retain 8 observations for prediction. Select Recursive
Least Squares, rather than OLS, and accept the default choice for number of observations for initializations.
1
6. After estimation, go to the PcGive menu, select Test then Recursive Graphics and select variables/tick boxes
so that your options screen looks as follows:
2
3
The graphical output obtained should look as follows:
.75
DLY
.5
.5
0
.25
1975
.2
.1
0
-.1
1980
1985
1990
1975
.5
LNW _1
1980
1985
1990
1980
1985
1990
1985
1990
1985
1990
LCE_1
0
-.5
1975
.02
LY_1
1980
1985
1990
1975
Res1Step
5% crit
1
0
1up CHOW s
.5
-.02
1975
1
.75
.5
.25
1980
5% crit
1985
1990
1975
1
Ndn CHOW s
1980
5% crit
Nup CHOW s
.5
1975
1980
1985
1990
1975
4
1980
Experiment with the Recursive options. Change the critical value, to say 1%. Focus on just
one variable, say LCE(-1), and investigate some of the other recursive options. What is the
effect of changing the number of periods used for recursive initialisation?
5