The reference for the MCI model is
Macroeconometric Modeling. This contains
the complete discussion and listing of the model. The model is dated
November 11, 2013. See also the
The MCI Model Workbook. You
should look over this workbook before working with the MCI model.
Data from the MCI model can be downloaded online
by going to the output phase when working with the MCI model
and downloading from there.
If you want to download the MCI model for use by EViews, there are two versions
to choose from. The first is due to Matt Cushing.
This is the recommended version. It achieves closer coefficient estimates
to those in the FP program and those on line, and it achieves closer
solutions. In general, it uses the EViews software in a better way.
Version due to Matt Cushing
The file to download is evmatt.zip. The readme.txt
file in the overall zipped file is as follows:
readme.txt
1 The workfile and programs supplied allow users to solve the MC model
in EVIEWS. The key is that for annual countries, certain variables are
defined as averages of future quarterly variables (BEPMP, BEX00$, BEPW$, ..)
Eviews does not handle this specification in the same manner as the FairParke
program. The supplied model AAAAMC omits the equations containing future values
and hence treats those variables as exogenous. The two supplied programs then
include the equations defining those variables as as averages of future
quarterly variables. The program then solves the model iteratively. The
resulting solutions closely match the solutions provided by the MCI model on
the Website.
MC_ITER.WK1 contains the data and the model, AAAAMC. (created with EVIEWS 9)
MC_BASE.prg solves the baseline model.
MC_SCEN.prg allows the user to change exogenous variables and solves for the new scenerio.
If you have an older version of eviews, you may have to create the workfile.
MC_text.txt contains the data in aasci format
Setup.prg loads the data and the equations. (Useful for a new installation.)
AAAAMC.prg loads the model
end of readme.txt
Version due to Ray Fair
For this version there are two zipped files to download:
- EVDATA.ZIP Contains MCEV.DAT.
- EV.ZIP Contains all the other files
listed below.
The individual files after unzipping are:
- MCEV.DAT Data on 19028 variables for 1952:1--2022:4 that can be read
using the file, import, read options in EViews. 19028 series, in rows,
not rectangle.
- MCEVGENR.PRG Generate a few variables that are needed that are not in
MCEV.DAT.
- MCEVEQ.PRG Estimates all but the nonlinear and
trade share equations.
- MCEVNL.PRG Estimates the nonlinear equations.
- MCEVSHR.PRG Estimates the trade share equations.
- MCEVSTOR.PRG Stores the 1689 estimated equations. This file is
not needed for most purposes, so you can ignore it.
- MCEVA.PRG The complete MCI model for solution.
To work with the MCI model in EViews
once you have
downloaded the files, you do the following.
- Enter in the command line: create q 1952.1 2022.4
- Do: file, import, read, ask for file MCEV.DAT, enter 19028 series,
use in rows, and use not rectangle. Hit ok. This will load the data.
- Enter: run mcevgenr
- Enter: run mceveq
- Enter: run mcevnl
- Enter: run mcevshr
- Enter: smpl 2014.1 2021.4 if you want to solve the model over the
forecast horizon (except for 2022)
- Enter: run mceva. Then hit run.
Then go to main workfile and click mceva. Then hit
solve. Click the solver option and set the maximum number of iterations
to about 100 (the default is 5000). Then hit ok. The solver will say it did
not converge, but for all practical purposes it appears to have. Then compare,
say, gdpr with gdpr_0, jaex with
jaex_0, etc. to
see if they are the same aside from small errors. They should be.
The above file MCEVA.PRG is the same as the file
MCEVA.INP for the Fair-Parke program
aside from formatting differences.
See MCI Model in the FP Program.
If you solve the MCI model for the 2014:1--2022:4 period using the
downloading instructions on the FP page, you will get a perfect tracking
solution aside from rounding error. (Compare OUT on the site to your
OUTA after downloading and solving.)
The above Eview instructions load the MCI model in EViews and
estimate the 1689 equations. If you solve the model for the 2014:1-2021:4
period, you will get roughly a perfect tracking solution.
The tracking is not exact because some of the EViews estimates of the
equations for the United States (equations 1 through 30) and not exactly the
FP estimates due to the way EViews treats first stage regressors and
serial correlation. The solution is, however, close enough for all practical
purposes.
You should be aware that there is mixing
of quarterly and annual data in the MC model.
Consider the following lines in MCEVA.PRG for Belgium (BE) and Denmark (DE),
two annual countries.
mceva.append VV1 =BEE
mceva.append BEEN =CC*VV1+CC(-1)*VV1(-1)+CC(-2)*VV1(-2)+CC(-3)*VV1(-3)
mceva.append BEEQ =(BEEN/BEEA)*BEEZ
mceva.append VV2 =BEPX
mceva.append BEPXN =CC*VV2+CC(-1)*VV2(-1)+CC(-2)*VV2(-2)+CC(-3)*VV2(-3)
mceva.append BEPXQ =(BEPXN/BEPXA)*BEPXZ
mceva.append BEPX$Q =(BEE00Z/BEEQ)*BEPXQ
mceva.append VV3 =BEM00$A
mceva.append BEM00$AN =CC*VV3+CC(-1)*VV3(-1)+CC(-2)*VV3(-2)+CC(-3)*VV3(-3)
mceva.append BEM00$AQ =(BEM00$AN/BEM00$AA)*BEM00$AZ
mceva.append VV4 =DEE
mceva.append DEEN =CC*VV4+CC(-1)*VV4(-1)+CC(-2)*VV4(-2)+CC(-3)*VV4(-3)
mceva.append DEEQ =(DEEN/DEEA)*DEEZ
mceva.append VV5 =DEPX
mceva.append DEPXN =CC*VV5+CC(-1)*VV5(-1)+CC(-2)*VV5(-2)+CC(-3)*VV5(-3)
mceva.append DEPXQ =(DEPXN/DEPXA)*DEPXZ
mceva.append DEPX$Q =(DEE00Z/DEEQ)*DEPXQ
mceva.append VV6 =DEM00$A
mceva.append DEM00$AN =CC*VV6+CC(-1)*VV6(-1)+CC(-2)*VV6(-2)+CC(-3)*VV6(-3)
mceva.append DEM00$AQ =(DEM00$AN/DEM00$AA)*DEM00$AZ
mceva.append BEPMP=.25*D1*(BEPMPQ+BEPMPQ(1)+BEPMPQ(2)+BEPMPQ(3))
mceva.append BEX00$=D1*(BEX00$Q+BEX00$Q(1)+BEX00$Q(2)+BEX00$Q(3))
mceva.append BEPW$=.25*D1*(BEPW$Q+BEPW$Q(1)+BEPW$Q(2)+BEPW$Q(3))
mceva.append DEPMP=.25*D1*(DEPMPQ+DEPMPQ(1)+DEPMPQ(2)+DEPMPQ(3))
mceva.append DEX00$=D1*(DEX00$Q+DEX00$Q(1)+DEX00$Q(2)+DEX00$Q(3))
mceva.append DEPW$=.25*D1*(DEPW$Q+DEPW$Q(1)+DEPW$Q(2)+DEPW$Q(3))
CC is a variable that is 1 in quarter 1 and zero in quarters 2, 3, and 4.
D1 is a variable that is 1 in quarter 1 and the missing value in quarters
2, 3, and 4. (1) means a lead value of 1 quarter, (2) two quarters, and
(3) three quarters. The first three lines convert BEE, which is annual,
where the annual value is in the quarter 1 space, to BEEQ, which is
quarterly. Similarly for the others.
The VV variables have a value of 1 in quarter 1 and the
missing value in quarters 2, 3, and 4.
The BEPMP line converts the quarterly values (BEPMPQ) to an annual value
(BEPMP). Similarly for the others.
The bottom line is that for annual variables the values in quarters 2, 3, and
4 are never used. The FP program ignores any calculations that involve
missing values. So for annual variables the values in quarters 2, 3, and 4
are always the missing value (999999.0).
In the EViews data file, MCEV.DAT, the missing values in quarters 2, 3, and
4 have been changed to the value in quarter 1 for each year and annual
variable. This is to avoid overflow and underflow
errors. Unlike FP, EViews computes
values for quarters 2, 3 and 4, and if 999999.0 is used, there can be
overflow or underflow errors. Again, it makes no difference what values are
used as long as there are no overflow or underflow errors since they are
never used. Also note that because of the use of up to three-quarter-ahead
future values, EViews can only solve the model through 2021:4 rather than
2022:4, which is the last quarter of the data. FP does not have this
problem because the model is solved in chunks of four quarters at a time.
|