Forecast Period versus Prediction Period For
the MC1 model we
will use the phrase "forecast period" to refer to the 1997-2002
period. If you
make no changes to the MC1 model and solve it for this period, the
predicted values will be
the values that are in the base dataset. You may, however, deal with
other periods. Much
of the data go back to 1960, and you can examine the data (list, graph,
download, etc.)
beginning with 1960. When solving the model, you should not begin before
1972, since data
for a number of countries do not begin until about 1972. The period that
you choose to
work with will be called the "prediction period," which may, of
course, differ
from what we are calling the forecast period.
The MC1 Model Datasets
The treatment of the MC1 model datasets is similar to the treatment of
the US
model datasets. Everything about the MC1 model is stored in a single
dataset:
1. values on all the endogenous and exogenous variables from 1960:1 through
2002:4,
2. specification information, and
3. coefficient estimates.
The dataset for the January 10, 1997, version of the model is called MCBASE
with
a password of MCBASE. You will name and create your own dataset, using
MCBASE as a
starting point. Say you call your dataset NEW and give it the password of
NEW. When you
first start, NEW and MCBASE are exactly the same, and then NEW gets
modified as you make
changes (change exogenous variables, coefficients, etc.). Once you are
done making
changes, you tell the program to solve the model. The program solves the
model, and after
this solution, the values of the endogenous variables in NEW are the
predicted values. You
can examine (i.e., write to the screen, print to a printer, or download
to a file) the
values in MCBASE and in any datasets you create.
In many cases one is interested in how the values of the endogenous
variables in
NEW compare to the original values in MCBASE. Say that you changed German
government
purchases of goods (GEG) for the forecast period and are interested in
how this change
affected real Japanese GDP (JAY). You can simply tell the program that
you want to compare
JAY in NEW versus MCBASE, and it will show you the two sets of values and
the differences.
(Once you do this a few times you will get the hang of it.) WARNING: A
comparison done
this way only works if you take the beginning of your prediction period
to be no earlier
than the beginning of the forecast period in MCBASE. See Section 2.6 in
Chapter 2 of
The US Model Workbook
for what needs to be done for comparisons within a period for which
historical data exist. This is important: Do not take your prediction
period to begin
earlier than the forecast period until you have read Section 2.6.
The program is flexible as to what you take as your base dataset.
Although the
first time you start the base dataset is MCBASE, after you have created
new ones, you can
use any of these as your base. For example, if you created NEW and now
want to makes
further changes and create NEW1, you simply tell the program that you
want NEW as your
base dataset. Once you have created NEW1, you can either compare the
values in NEW1 with
those in NEW or the values in NEW1 with those in MCBASE. You do this by
simply telling the
program which two datasets to use for the comparison.
If you make changes to your dataset and do not ask the program to
solve it
(which you are allowed to do), the values of the endogenous variables in
the dataset are
not consistent with your changes because they are not the solution
values. Make sure you
remember whether your dataset has been solved. As a general rule, you
might always solve
your datasets immediately after you have made your changes.
Units of the Variables
Although the flow variables in the MC1 Model are at quarterly rates, for
most
purposes it is useful to work with variables at annual rates. For the
most part you will
work with variables at annual rates: the program lists the variables at
annual rates and
your changes are taken to be at annual rates. The only time you need care
about quarterly
rates is if you are downloading the data to your computer for estimation
purposes, since
for estimation purposes the variables should be at quarterly rates. There
is an option in
the program to download the data at quarterly rates.
Years versus Quarters
Since some of the individual country models within the overall MC1 model
are
annual, the MC1 model must be solved in yearly (four-quarter) units.
Thus, when choosing a
period, you need only (and can only) choose years. The solution is,
however, quarterly for
the quarterly countries, and when you examine the output, you will be
allowed to examine
quarterly values for the quarterly countries.
Solution Errors
If you make wild changes to the exogenous variables or coefficients, the
model
may not solve. When the model does not solve, you will get a solution
error message, and
the existing values of the endogenous variables in your dataset will not
be changed. (Your
dataset will still be inconsistent in the sense discussed above.) You
need to be less wild
and try again. As a general rule, as discussed in Chapter 2 of
The US Model Workbook,
you should not try to push the economy into extreme areas.
Macroeconometric models are not
likely to be reliable when variable values are pushed far beyond their
historical ranges. |