UPDATING REMI'S CONTROL FORECASTS by woh19525

VIEWS: 28 PAGES: 26

									  UPDATING REMI’S
CONTROL FORECASTS
            September 2003




                     Prepared by

                    Marshall J. Vest
      Director, Economic and Business Research
  Eller College of Business and Public Administration
                  University of Arizona
                    Tucson, Arizona
                    (520) 621-4075
              UPDATING REMI’S CONTROL FORECASTS
                                                       September 2003

 Prepared by:

 Marshall J. Vest
 Director, Economic and Business Research
 Eller College of Business and Public Administration
 University of Arizona
 Tucson, Arizona
 (520) 621-4075

 Peer reviewed by the Arizona Department of Commerce Economic Research Advisory Committee:

 Dan Anderson                                Brian Cary                                   Lisa Danka
 Assistant Executive Director for            Forecast Consultant                          Director, Commerce & Economic
 Institutional Analysis                      Pinnacle West Energy Corporation             Development Commission
 Arizona Board of Regents                                                                 Arizona Department of Commerce

 Kent Ennis                                  Wayne Fox                                    James B. Nelson
 Economic Consultant                         Director, Bureau of Business and             Economic Development Manager
 CH2M Hill                                   Economic Research                            Salt River Project
                                             Northern Arizona University

 William P. Patton, Ph.D.                    Elliott D. Pollack                           Tom Rex
 Director of Economic Development            Elliott D. Pollack & Co.                     Research Manager
 Tucson Electric Power                                                                    Center for Business Research
                                                                                          L. William Seidman Research
 Brad Steen                                  Don Wehbey                                   Institute
 Chief Economist                             Economist                                    W. P. Carey School of Business
 Arizona Department of                       Research Administration                      Arizona State University
 Transportation                              Arizona Department of Economic
                                             Security

 Technical review by:

 David Morf                                  Jack Tomasik
 Economic Consultant                         Regional Development Manager
 Regional Economic Models, Inc.              Maricopa Association of
                                             Governments

 2003 by the Arizona Department of Commerce. This document may be reproduced without restriction provided it is reproduced
accurately, is not used in a misleading context, and the author and the Arizona Department of Commerce are given appropriate
recognition.
This report was prepared for the Arizona Department of Commerce with funding from the Commerce and Economic Development
Commission. Elements of this report may be presented independently elsewhere at the author's discretion. This report will be
available on the Internet for an indefinite length of time at http://www.azcommerce.com/Economic/default.asp. Inquiries should be
directed to the Office of Economic Information and Research, Arizona Department of Commerce, (602) 280-1300.
The Arizona Department of Commerce has made every reasonable effort to assure the accuracy of the information contained herein,
including peer and/or technical review. However, the contents and sources upon which it is based are subject to changes, omissions and
errors and the Arizona Department of Commerce accept no responsibility or liability for inaccuracies that may be present. THIS DOCUMENT
IS PROVIDED FOR INFORMATIONAL PURPOSES ONLY. THE ARIZONA DEPARTMENT OF COMMERCE PRESENTS THE MATERIAL IN THIS
REPORT WITHOUT IT OR ANY OF ITS EMPLOYEES MAKING ANY WARRANTY, EXPRESS OR IMPLIED, INCLUDING THE WARRANTIES OF
MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, OR ASSUMING ANY LEGAL LIABILITY OR RESPONSIBILITY FOR THE
ACCURACY, COMPLETENESS, OR USEFULNESS OF ANY INFORMATION, APPARATUS, PRODUCT, OR PROCESS DISCLOSED, OR
REPRESENTING THAT ITS USE WOULD NOT INFRINGE PRIVATELY OWNED RIGHTS. THE USER ASSUMES THE ENTIRE RISK AS TO THE
ACCURACY AND THE USE OF THIS DOCUMENT AND ANY RELATED OR LINKED DOCUMENTS.
         UPDATING REMI’S CONTROL FORECASTS
                                  September 2003


                               TABLE OF CONTENTS


Foreword                                                  iii

Creating a New National Control                           1

Creating a New Regional Control                           5

Incorporating Local Forecasts Into the Regional Control   9

Observations and Recommendations                          12

Appendix A, Questions related to the Arizona Forecast     14

Appendix B, Comments from Peer Review Process             18




                                          ii
FOREWORD

Arizona’s economic challenges are extremely complex and policy makers generally do not have
the time or training to appropriately analyze them. To fill the void, the Office of Economic
Information and Research (EIR) within the Arizona Department of Commerce (ADOC) focuses
on producing objective, high quality, peer-reviewed analyses to support the development of
public policy.

One of the tools in EIR’s arsenal is the REMI model, a robust policy simulation program. The
Arizona Department of Commerce owns the primary license on behalf of state government, and
a number of “secondary” licensees have joined the ranks of REMI users since 2001– the Joint
Legislative Budget Committee (JLBC), the Arizona Department of Housing, the University of
Arizona, and Arizona State University. The Maricopa Association of Governments also uses the
program.

Although widely used throughout the United States and other countries for policy analysis
purposes, the REMI model has not been used extensively in Arizona. Therefore, ADOC felt it
appropriate to acquaint the State’s economist community with REMI’s “black box” and develop a
generally accepted understanding of the use of the model before actually employing it for public
policy purposes. In April 2003, twenty-five public, private and university economists came
together at the first Arizona REMI user’s conference, hosted by Salt River Project. It was the
first step in developing a REMI knowledge base in Arizona.

This report represents the second step. As one of Arizona’s leading economic forecasters,
Marshall Vest was given the task of “test driving” REMI by updating the model’s national and
regional forecasts. His work further informs the model’s value for Arizona, primarily as a policy
analysis tool. Additionally, the baseline forecasts established by Vest provide the methodology
to reflect current economic conditions for those using the ADOC model.

Future projects will build on this base as ADOC begins to implement REMI in policy analyses.
EIR’s collaborative, peer-reviewed research process will be fundamental in these efforts to
ensure consistent and quality information for policy makers at all levels.




Gilbert Jimenez
Director
Arizona Department of Commerce




                                                  iii
             UPDATING REMI’S CONTROL FORECASTS


Creating a New National Control
REMI allows users to produce a new national control forecast. Providing a more
current forecast that takes into account the recent recession and slow growth
over the 2001-03 time frame is the motivation to do so. We used the most recent
long-term forecast from Global Insight (GI) -- their 25-year standard forecast from
February 2003.
REMI allows three different types of adjustments: macroeconomic value,
employment update and policy variables.
Macroeconomic variables consist of (a maximum) 22 major GDP components, 10
components of personal income and a personal consumption price index. A
subset of these may be updated if desired, and the model will calculate
consistent values for the remaining variables. We updated the maximum.
Employment update allows the user to change the forecast for all 53 sectors.
The major challenge here is that most national (and regional models) use BLS
employment categories from the “ES-202” program. These data are based on
unemployment insurance reports filed by employers each quarter, and do not
include self-employed, domestic, unpaid family workers, farm workers, or the
military. These data are reported monthly as part of the Current Employment
Statistics (or CES program) and are referred to as “nonfarm” employment.
By contrast, REMI uses BEA’s “full and part-time employment” -- data that is built
off of ES-202 data, but includes self-employed, domestic, unpaid family workers,
farm workers, and the military. These extra jobs are spread across industries.
The data are of annual frequency and are reported with a lag of two or more
years.
The result is two very different measures of employment, both in total and in
industry detail1. For example, as shown in the workbook “EMP VARS GI FEB03
25yr.xls” on the RemiVars sheet, the number of employees nationwide in Real
Estate in the year 2000 was over 4.5 million according to REMI, compared to
only 1.5 million in the Global Insight database. This is due to the fact that a large
number in the real estate industry are self employed. Another example is
construction: 9.6 million versus 6.7 million. As a result, using forecasts of
nonfarm industry detail from Global Insight or other forecast provider to update
REMI’s full and part time concept is problematic. Additionally, Global Insight
does not forecast 20 of the 53 employment categories; for these categories,
values from the REMI control would need to be used or otherwise constructed.




1
    See peer review comment from Dan Anderson, Appendix B for usage of BEA versus BLS data.
                                                                    UPDATING REMI’S CONTROL FORECAST



Finally, policy variables, numbering in the hundreds might be used to make fine-
tuning changes. Included are such items as shares of industry output, for each
of 53 sectors; components of disposable income by industry; consumer
spending, government and investment spending shares; employment shares;
productivity; wages, prices and profits for 53 sectors each; and market shares by
industry. It is not clear how information from Global Insight would be used to
change any of these policy variables. In any case, it would not be a
straightforward process, and it was evident from the beginning that policy
variables would not be used in the generation of a new control.
Another complicating factor is that these three components (or any two, for that
matter) may not be done all at the same time – they must be done in sequence.
For example, a modeler may first incorporate macroeconomic values, solve and
save the results. Then in a second solve that builds upon the first, employment
may be changed in similar fashion, and those results saved.
With each run of the model, values that were incorporated in the prior solution
will change, however. That’s because REMI is a simultaneous model, and REMI
offers no feature to exogenize variables, calculate implicit add factors or to
perform goal searching. The user therefore winds up chasing the desired target
– i.e., fix macro variables, then run employment, then fix macro again, then fix
employment, etc.
This complicating factor, along with the fact that the employment measures vary
greatly by definition, led us to decide on an approach of updating macroeconomic
values only.
Workbook “Macro VARS GI FEB03 25yr.xls” contains a series of spreadsheets
that document this process. The first sheet lays out the macro values that a
REMI user may change, and identifies matching measures from Global Insight,
along with GI’s long-term forecast from February 2003. Sheet 2 contains GI
variable documentation and sheet 3 contains fixes that were incorporated into
sheet 1. A command file that creates the data in sheet 1 was developed to
facilitate future updates (RemiMacroList.cmd).
Additional sheets are used to compare forecasts for 40-some measures from the
REMI Standard Control and the new national control. We first checked to see
that the GI values that were incorporated went in correctly. Sheet 6 shows that
REMI did not accept the new values exactly, but in percentage terms the results
were generally within a percent or two (sheet 7). (Note: we first formed the new
control and saved it. Then we opened the new one and ran a simulation with no
changes. It’s necessary to do this to “see” the values.) We were perplexed as to
why the values weren’t exact, but concluded that having to solve the new control
(which iterates 30 times, even with no changes), probably accounted for the
differences.




Economic and Business Research, Eller College of Business and Public Administration         2
The University of Arizona
                                                                    UPDATING REMI’S CONTROL FORECAST



REMI economists suggested that we perform the macro variables update
process a second time, building upon the results of the first solution2. This would
allow the model to iterate from a closer starting point, and converge more closely
to the Global Insight values as entered. The results of this simulation were saved
as REMI file “National Control GI Feb03 Run2,” and are summarized in the
above-referenced workbook “Macro VARS GI FEB03 25yr.xls.”
The second simulation produced mixed results -- some measures were closer to
the entered values -- and others were not. The results failed to make the case
that running the macro variables a second time improved convergence. In either
case, however, the model results were not far from the entered values. In
producing a new regional control (described later) we decided to use the first
simulation.
Even more interesting is a comparison of REMI’s standard control with our new
one (the first run). As shown in sheet 8, these vary widely from measure to
measure and over time, with the new control lower by a few percentage points in
the early years (Global Insight contains the recent recession and period of slow
growth that has persisted since, while the REMI standard appears to be more of
a trend forecast that uses 2000 as a jump off point). In the latter years of the
forecast, however, the new control zooms ahead by large margins. For
computers & furniture and medical care, the new amounts are more than double.
Personal income and private nonfarm employment are both 45% higher by the
year 2035.
Although employment is much higher in the new control, population is identical in
both scenarios, i.e., population is exogenous in the national model. According to
REMI economists, a different “build” of the model does allow population to
change – a desirable result for a long-term model3. If the nation were to run
short of labor, surely immigration quotas would be raised thereby allowing the
nation to import labor.
Updating to a new national control will make a huge difference in the regional
results, as suggested in the following graphs and summary table.




2
    See Appendix A, Reply #1.
3
    See Appendix A, Reply #8.

Economic and Business Research, Eller College of Business and Public Administration         3
The University of Arizona
                                                                                                        UPDATING REMI’S CONTROL FORECAST




                                         U.S. Private Nonfarm Employment (Mil)

                                                                                   New GI                      REMI

           300000
           250000
           200000

           150000
           100000
             50000
                     0
                           2001
                                   2003
                                           2005
                                                   2007
                                                           2009
                                                                   2011
                                                                           2013
                                                                                   2015
                                                                                          2017
                                                                                                 2019
                                                                                                        2021
                                                                                                               2023
                                                                                                                      2025
                                                                                                                             2027
                                                                                                                                    2029
                                                                                                                                           2031
                                                                                                                                                  2033
                                                                                                                                                         2035
                                                    U.S. Personal Income ($ Bil)

                                                                                  New GI                       REMI

           70000
           60000
           50000
           40000
           30000
           20000
           10000
                 0
                         2001
                                  2003
                                          2005
                                                  2007
                                                          2009
                                                                  2011
                                                                          2013
                                                                                  2015
                                                                                          2017
                                                                                                 2019
                                                                                                        2021
                                                                                                               2023
                                                                                                                      2025
                                                                                                                             2027
                                                                                                                                    2029
                                                                                                                                           2031
                                                                                                                                                  2033
                                                                                                                                                         2035




Economic and Business Research, Eller College of Business and Public Administration                                                                             4
The University of Arizona
                                                                      UPDATING REMI’S CONTROL FORECAST



Control Comparisons, Selected Measures


                                                           2001          2010         2020         2030
U.S. Population (000s)
 REMI Standard National Control                            285005         309228       336268       365222
 National Control GI Feb 03                                285005         309228       336268       365222

U.S. Total Employment (000s)
 REMI Standard National Control                            167723         179240       186730       196182
 National Control GI Feb 03                                167249         182088       212832       251323

U.S. Personal Income ($bil nominal)
 REMI Standard National Control                               8718          13601       21294        33365
 National Control GI Feb 03                                   8594          13519       24394        44602

PCE-Price Index (Fixed 96$)
 REMI Standard National Control                                 110             133          167      209
 National Control GI Feb 03                                     111             134          178      245




Creating a New Regional Control
The first step is to simply create a new regional control using the new national
control. Workbook “New REG Control created with Nat Control GI Feb 03.xls”
contains a series of spreadsheets documenting this process.
The REMI model (or “build”) used by the Arizona Department of Commerce is a
three-area model: Maricopa County, Pima County and Balance of Arizona.
Summary measures for each of the three areas as well as Arizona (all regions)
are presented in sheets 1-4. Four corresponding sheets follow those containing
the REMI Standard Reg Control. Sheet 9 contains graphs comparing the two.
Sheet 11 contains tabular comparisons, which are reproduced below.




Economic and Business Research, Eller College of Business and Public Administration                          5
The University of Arizona
                                                                      UPDATING REMI’S CONTROL FORECAST



Control Comparisons, Selected Measures


                                                            2001         2010         2020         2030
AZ Population (000s)
 REMI Standard Reg Control                                    5332           6626        7885         8965
 Reg Control using GI Feb 03                                  5332           6584        7852         9091

AZ Personal Income ($bil)
 REMI Standard Reg Control                                      138             232          379          603
 Reg Control using GI Feb 03                                    136             229          435          831

AZ Per Capita Personal Income
 REMI Standard Reg Control                                   25938          35070       48046        67224
 Reg Control using GI Feb 03                                 25547          34763       55404        91421

AZ Total Employment (000s)
 REMI Standard Reg Control                                    2882           3295        3571         3820
 Reg Control using GI Feb 03                                  2872           3308        4051         5000

AZ Labor Force (000s)
 REMI Standard Reg Control                                    2560           3075        3458         3771
 Reg Control using GI Feb 03                                  2559           3050        3443         3840

AZ Employment to Population Ratio
 REMI Standard Reg Control                                   0.541           0.497      0.453        0.426
 Reg Control using GI Feb 03                                 0.539           0.502      0.516        0.550


Both REMI Standard and the New control show Arizona’s 2030 population near 9
million, and differ by only 130,000. But, like national control comparisons,
differences are large and varied for other measures. By 2030, employment is
nearly 31% higher (5 mil versus 3.8 mil) and personal income is 38% higher.
Employment to population ratios are radically different. In REMI, the ratio falls
throughout the forecast period -- from near .55 today to .42 in 2030. This is a
radical departure from the trend seen since 1960 when this ratio stood near .25.
Historically, this ratio has grown steadily except for recession periods when it
tends to level off for a year or two. The New scenario shows a decline through
2010 to near .5 before increasing again to finish near .58 in 2035.
Neither scenario seems particularly believable. EBR forecasts (and Global
Insight’s for the nation) show a continuing upward trend in the ratio, with a
slowing, or flattening, as the population ages. EBR’s forecast, with the ratio of
nonfarm jobs to population near .5, winds up closer to the New scenario.




Economic and Business Research, Eller College of Business and Public Administration                             6
The University of Arizona
                                                                                                     UPDATING REMI’S CONTROL FORECAST



REMI economists suggested updating regional employment to see if that would
address the fall in employment to population ratio4. (It did – more later).



                                          AZ Employment to Population Ratio

                                                                             New                      REMI

             0.6

           0.55

             0.5

           0.45

             0.4
                   2001
                            2003
                                     2005
                                            2007
                                                   2009
                                                           2011
                                                                  2013
                                                                         2015
                                                                                2017
                                                                                           2019
                                                                                                    2021
                                                                                                            2023
                                                                                                                    2025
                                                                                                                            2027
                                                                                                                                   2029
                                                                                                                                          2031
                                                                                                                                                 2033
                                                                                                                                                        2035
Per capita personal income also varies significantly in the two scenarios. In the
new control, PCPI in 2030 exceeds $91,000 compared to only $67,000 in REMI.
EBR’s forecast shows $85,000.



                                                 AZ Per Capita Personal Income

                                                                                New                        REMI

           140000
           120000
           100000
             80000
             60000
             40000
             20000
                   0
                          2001
                                   2003
                                          2005
                                                 2007
                                                        2009
                                                               2011
                                                                      2013
                                                                             2015
                                                                                    2017
                                                                                             2019
                                                                                                     2021
                                                                                                             2023
                                                                                                                     2025
                                                                                                                            2027
                                                                                                                                   2029
                                                                                                                                          2031
                                                                                                                                                 2033
                                                                                                                                                        2035




4
    See Appendix A, Reply #3.

Economic and Business Research, Eller College of Business and Public Administration                                                                            7
The University of Arizona
                                                                    UPDATING REMI’S CONTROL FORECAST



The ratio of Arizona PCPI to US PCPI also reveals interesting results. In both
scenarios, AZ continues to lose ground, falling to .75 in the New scenario. This
continues the downward trend that began in the mid 1980s for this “standard of
living” indicator. This is a “better” result than contained in EBR’s last long-term
update, in which this measure falls to .6.



                         Per Capita Personal Income, AZ ratio to US

                                                       New            REMI

               1
           0.95
            0.9
           0.85
            0.8
           0.75
            0.7
             60

                    65

                         70

                              75

                                    80

                                          85

                                                90

                                                     95

                                                           00

                                                                 05

                                                                      10

                                                                             15

                                                                                  20

                                                                                       25

                                                                                            30

                                                                                                 35
           19

                   19

                        19

                             19

                                  19

                                        19

                                              19

                                                   19

                                                         20

                                                               20

                                                                    20

                                                                          20

                                                                                20

                                                                                      20

                                                                                           20

                                                                                                20
Finally, there is an inconsistency between the labor force and total employment
in the two scenarios. First, total employment is greater than the labor force
throughout the forecast, starting in the base year. We normally think of the labor
force as being the sum of total employment and total unemployment. The
second concern is that employment grows much faster than the labor force in the
New scenario, so that by the year 2030, the labor force is 3.8 million, compared
to employment of 5 million. This unlikely result can happen only if multiple job
holding becomes widespread (or if Arizona imports a large number of commuting
workers from out of state -- but there are no large population centers close to
Arizona’s border)5.
With the labor force caveat, the New regional control based on Global Insight’s
forecast of February 2003 appears to be a significant improvement over the
regional control provided by REMI.




5
    See Appendix A, Reply #5.

Economic and Business Research, Eller College of Business and Public Administration                   8
The University of Arizona
                                                                    UPDATING REMI’S CONTROL FORECAST




Incorporating Local Forecasts Into the Regional Control
To this point we have made no attempt to incorporate local forecasts into the
regional control. We have simply prepared a regional control using the new
national control.
Similar to the process to update the national control, REMI allows users to
change regional employment and policy variables. As with the national control,
employment categories and definitions are problematic: REMI’s 53 categories
contain the more-inclusive BEA concept while EBR’s models are based on
nonfarm measures collected and reported by BLS/DES.
As shown in workbook “PHX Mesa Emp Vars.xls”, sheet 1, EBR’s models
contain employment forecasts for only 13 of the 53 required industries for
Maricopa County. To fill in all the cells (REMI requires that no row be blank), it is
necessary for the EBR forecasts to be allocated to component industries as done
in sheet 3. This can be accomplished using the proportion that each represents
in the base year 2000. In doing this, the cells are filled, but it assumes that each
sub component will grow at the same rate as the aggregate – not a very
satisfying result. For example, railroads, trucking, local/interurban, air
transportation and the “all other” categories would grow at the same rate as the
forecasted aggregate for transportation.
EBR does not forecast three of the required components: Federal Military, Farm
and Agriculture/Forestry/Fishery Services. For these, the REMI standard control
forecasts are used.
There also is a challenge related to geography. Forecasts prepared by EBR
are for the Phoenix-Mesa metro area (Maricopa plus Pinal), Metro Tucson, and
Arizona as a whole. All three are stand-alone models, and are not designed to
produce meaningful forecasts for the residual (i.e., the balance of the state).
Equally challenging is the mismatch for Pinal county. In EBR’s system, Pinal is
added in with Maricopa, but in REMI, Pinal is part of the Balance. Therefore, two
of the three regions do not match in composition.
Fortunately, REMI handles employment updates in a fashion that overcomes
these challenges. REMI does not change the history for any of these measures.
Rather, it applies the calculated growth rates from the values provided by the
user to the REMI base year values6. Therefore, REMI will apply to Maricopa
County the growth rates from EBR’s forecast for the PHX-Mesa metro area.
Since Maricopa County dominates the PHX-Mesa metro area, this is quite
reasonable.
Another fix is required. EBR’s forecasts end in 2027. The growth rate from 2026
to 2027 was used to extend the forecasts through 2035. (Unfortunately, REMI
was unable to run the forecast beyond 2027 because of what we believe to be

6
    See Appendix A, Reply #7.

Economic and Business Research, Eller College of Business and Public Administration         9
The University of Arizona
                                                                    UPDATING REMI’S CONTROL FORECAST



the zero employment in tobacco manufacturing in the three regions. Although we
were able to change the zeros to .01 for the years 2001-27, we were unable to
do so for 2028-35, and were therefore forced to shorten the forecast horizon.
This appeared to be an abnormality in the program that we couldn’t work
around.)
We prepared a new regional control “Reg Control GI Feb03 with employment
update” that incorporates local forecasts from EBR. This control uses the
“National Control GI Feb 03” and its corresponding regional control, “Reg Control
using GI Feb 03.” In this run, we used the employment update features for
regions. Workbooks “PHX Mesa Emp Vars.xls”, “Pima Emp Vars.xls”, and
“Balance Employment Vars.xls” contain the employment forecasts that were
provided to the model.
Incorporating the local forecast produces significantly higher results. As shown
in the following table, population is 8.7 million by 2020, or 850,000 higher than in
Run 1. Total employment crosses the 5 million mark by 2020 -- an entire decade
before Run 1. Per capita income approaches $59,000 in 2020 – about 73.8% of
the national figure. And, the employment to population ratio winds up close to
.58.
The employment to population ratio increases over the forecast period, a much
more acceptable result than in Run 1.
Compared to the EBR forecasts that served as the basis for the employment
updates process, Run 2 matches quite closely. For example, the year 2020
population projections differ by only 50,000. Run 2 produces a marginally higher
personal income, as well as per capita income.




Economic and Business Research, Eller College of Business and Public Administration        10
The University of Arizona
                                                                     UPDATING REMI’S CONTROL FORECAST



Control Comparisons, Selected Measures


                                                            2001        2010          2020      2030

AZ Population (000s)
 REMI Standard Reg Control                                    5332           6626       7885      8965
 Reg Control using GI Feb 03                                  5332           6584       7852      9091
 With Employment Updates                                      5328           6843       8718       N/C
 EBR model results                                            5320           6744       8666       N/C

AZ Personal Income ($bil)
 REMI Standard Reg Control                                     138            232        379       603
 Reg Control using GI Feb 03                                   136            229        435       831
 With Employment Updates                                       136            250        513       N/C
 EBR model results                                             135            227        449       N/C

AZ Per Capita Personal Income
 REMI Standard Reg Control                                  25938          35070       48046     67224
 Reg Control using GI Feb 03                                25547          34763       55404     91421
 With Employment Updates                                    25474          36588       58895       N/C
 EBR model results                                          25329          33671       51855       N/C

AZ Total Employment (000s)
 REMI Standard Reg Control                                    2882           3295       3571      3820
 Reg Control using GI Feb 03                                  2872           3308       4051      5000
 With Employment Updates                                      2854           3717       5045       N/C
 EBR model results - nonfarm only                             2266           2976       4142       N/C

AZ Labor Force (000s)
 REMI Standard Reg Control                                    2560           3075       3458      3771
 Reg Control using GI Feb 03                                  2559           3050       3443      3840
 With Employment Updates                                      2556           3239       3972       N/C
 EBR model results                                             N/A            N/A        N/A       N/A

AZ Employment to Population Ratio
 REMI Standard Reg Control                                   0.541          0.497       0.453     0.426
 Reg Control using GI Feb 03                                 0.539          0.502       0.516     0.550
 With Employment Updates                                     0.536          0.543       0.579       N/C
 EBR model results - nonfarm only                            0.430          0.440       0.480       N/C

Note: Control with Employment Updates runs only through 2027.




Economic and Business Research, Eller College of Business and Public Administration                    11
The University of Arizona
                                                                    UPDATING REMI’S CONTROL FORECAST




Observations and Recommendations
The process of updating the forecast controls makes it clear once again that
REMI is a model intended for policy analysis. Even though it contains a forecast,
it is not designed for forecasting7.
The choice of BEA’s employment measures, which are reported only annually
and with significant lag, seriously hamper any efforts to use the model in a
forecasting environment -- especially for short-term forecasting. BEA’s
employment does make sense for a policy model, however, where the more
comprehensive measure is needed for such things as transportation/trip
generation analysis.
We feel that the REMI model is appropriate and useful for analyzing a wide range
of projects in a policy analysis setting. Earlier versions of the REMI model
(based on Regional Purchase Coefficients, or RPC’s) have been widely used in
many states. The newest version, based on New Economic Geography
concepts, will require further investigation to ascertain it dynamic properties.
The New national control appears to better reflect the recent recession, and it
does not grow in the outer years nearly as slowly as the REMI control. Running
a second control that feeds the same macro changes on top of the first run did
not improve convergence to the entered values. Even so, the new control is an
improvement over the REMI-supplied forecast. We do not see value and
therefore did not pursue using the employment update or policy variables
features in creating a new national control. We recommend using the “National
Control GI Feb 03” for future analysis.
The new regional control required two runs: first, to incorporate the new national
control, and then a second run to incorporate regional forecasts using
employment updates. The second run, “Reg Control GI Feb03 with employment
update” appears to be reasonable and is recommended to replace the standard
REMI regional control.
It is not clear if the “DOC build” contains the right geography. The way Pinal
county is developing, it makes more sense for it to be either included separately
or aggregated with Maricopa to reflect the metro definition, rather than lumped
together with the Balance. If it were added as a fourth area, it would
unfortunately complicate any statewide analysis even further, as the researcher
would then need to fashion changes for four areas rather than three.
Ideally, for analysis of statewide issues (such as alternative tax proposals), an
Arizona model is the best build. The analyst doesn’t need results for individual


7
  See Appendix B, Comments from Pat Patton and Kent Ennis. Jack Tomasik agrees that REMI
“is not suited for short-term forecasts” but argues that it is useful (and “clearly superior” to cohort-
survival models) for creation of long-term projections. Pat Patton believes that REMI should not
be used to prepare projections beyond 10-15 years.

Economic and Business Research, Eller College of Business and Public Administration                  12
The University of Arizona
                                                                    UPDATING REMI’S CONTROL FORECAST



counties (and having a model with all 15 counties simply makes the analyst’s job
15 times more complicated).
Ideally, DOC would have an Arizona-only build for statewide research projects
(or an Arizona and rest of US build which would enhance usage for tourism and
retirement-related research). An additional 15-county model could be used for
those projects that require sub-state analysis8.
Suggestions from peer reviewers concerning documentation and “Black Box” are
valid, but REMI has gone to extreme lengths to document their model. A lengthy
list of references, ranging from books to referred journal articles to technical
documentation is available from REMI. Suggestions to examine the sensitivity of
REMI by shocking the model are a logical next step but deemed to be beyond
the current scope of work.
Finally, we documented the control updating process as best we could and
automated as much as possible. Although that should save effort next time
around, it’s also possible that both Global Insight and REMI will have new
employment variables based on NAICS rather than SIC, and that likely will
require significant effort once again.




8
    See Appendix B, Dan Anderson’s comments regarding a 15 county model.

Economic and Business Research, Eller College of Business and Public Administration        13
The University of Arizona
                                                                    UPDATING REMI’S CONTROL FORECAST




                                            Appendix A
          QUESTIONS RELATED TO THE ARIZONA FORECAST
The following questions were posed by Marshall Vest and others in the process
of updating the regional and national forecasts. The replies are from David Morf,
Economic Consultant, REMI. This exchange is included to help further the
reader’s understanding of Mr. Vest’s approach and the model’s response.

Question: Why did we not get the Global Insight macro values that we input into
the new national control? (They were generally within a couple of percent. You
recommended running it again using the new control.)

Reply: The model interacts with new macro national demand vectors to
converge on endogenous results for a new national control file. By saving the
new national control file, then re-inputting the initial macro demand vectors to the
saved control file, the second national control results will more closely
approximate the input macro vector values. This process can be repeated until
the approximation is satisfactory.


Question: What is the source for the REMI national control? What is the last
actual? Why hasn't it been updated recently?

Reply: BEA, BLS, CBP, CDC, Census, and other institutions provide the core
national, state, and county data. A detailed sources table is included in the
Model Documentation, Version 5.x. There is approximately a 2-year lag between
the last history year and the time frame when BEA and BLS release their data to
our model-building process. However, due to the NAICS conversion process and
the model software upgrade effort to implement New Economic Geography, the
current model being delivered in 2003 continues to apply the last SIC database,
which is 2000. This also has resulted in our using the 2-year near-term business
cycle control forecast from the University of Michigan’s RSQE estimated in March
2002, rather than in 2003.

Later in 2003, we plan to release a final SIC database for 2000, using the 2-year
RSQE business cycle estimated as of the then-most-recent prior quarter. This
will restore timeliness to the near-term business cycle forecast estimate applied
to the model-building process. In 2004, we plan to deliver the new NAICS
database for 2001, using the 2-year RSQE business cycle estimated as of the
then-most-recent prior quarter. Depending on the availability of data from BEA
and BLS, either later in 2004 we plan to deliver NAICS data for 2001 and 2002,



Economic and Business Research, Eller College of Business and Public Administration        14
The University of Arizona
                                                                    UPDATING REMI’S CONTROL FORECAST



or in 2005 we plan to deliver NAICS data for 2001, 2002, and 2003. Either way,
this will restore timeliness to our 2-year lag cycle for last history year.


Question: Why does the employment to population ratio decline from the get-
go? Baby-boomers don't reach retirement until 2010 or so...

Reply: As noted below in the reply to Q5, the model maintains a relationship
between jobs and the labor force over time so that jobs do not become
unreasonable in relationship to the labor force, even though jobs may exceed
labor force slightly due to the definitional difference between jobs in the
numerator and labor force (people) in the denominator. If the population
changes, or if the participation rates change, then in either case labor force will
follow suit, and jobs will be conformed. Therefore, the population aging process,
and the cohort participation rate behavior controlling the labor force, combine to
affect labor force and employment levels. The process will accelerate after 2010,
but already the US aging process is at work, and it is cumulative.

It will be useful if you ascertain if or how Global Insight controls the relationship
between employment and labor force. More importantly, it will be useful to see if
or how Global Insight accounts for current and future population aging behavior
and related cohort participation rate behavior. If Global Insight is not accounting
fully for behaviors and relationships among employment, labor force, and
population, it may be over-estimating its labor force and employment levels,
resulting in higher employment-to-population ratios than would appear
defensible.


Question: Why does total employment exceed total labor force? By definition,
labor force is the sum of employment plus unemployment.

Reply: BEA, BLS, and the underlying ES-202 employment data reflect jobs, not
people. However, the labor force reflects people. Therefore, multiple job-holding
can result in a numeric artifact whereby “employment” seems to exceed the labor
force. The relationship over time clearly remains valid as a longitudinal
estimation parameter.


Question: Why does total employment grow so much faster than labor force?
Shouldn't these two concepts be linked?

Reply: As discussed above in the reply to Q3, all state and county models, and
multi-region US models, control the relationship between employment and labor
force. Additionally, their population is endogenous, and they apply endogenous

Economic and Business Research, Eller College of Business and Public Administration        15
The University of Arizona
                                                                    UPDATING REMI’S CONTROL FORECAST



cohort participation rates to derive the labor force. However, the single-region
US model included in Policy Insight does not control the relationship between
employment and labor force, and does not apply New Economic Geography; its
population is exogenous; and it uses exogenous cohort participation rates to
derive the labor force. Thus, if the macro demand vectors or employment being
applied to build a new national control reflect growth compared to the model’s
standard national forecast, then employment growth in the new national control
will outstrip labor force growth, and this will be passed to any new regional
control created from the new national control.


Question: How does REMI deal with the unemployment rate?

Reply: The model does not estimate an unemployment rate. It estimates
percentage changes in a benchmark unemployment rate established for about
1970. For years subsequent to 1970, an increase (decrease) in employment
relative to labor force will be reflected as an unemployment percentage decrease
(increase) relative to the benchmark rate.


Question: In preparing the new regional control, if we input forecasts for
Phoenix-Mesa (Maricopa plus Pinal) for Maricopa, what happens? Does this turn
Maricopa into metro PHX? (How does the model use the numbers we provide?)

Reply: Employment data being input to adjust a regional or national forecast is
assessed for the growth rate implied by the new data. That growth rate is
applied to the model’s last history year employment by sector to create new
forecast employment. It is sufficient that the numbers selected for usage as input
for employment calibration be defensible; e.g., the employment numbers for
calibrating a new regional forecast do not necessarily need to correspond with
the region in the model.


Question: Population was exogenous when we were making a new national
control. Over the long term, one would expect immigration to increase if labor
grew scarce. You mentioned that in a different build, U.S. population is
endogenous. Could you comment on this?

Reply: The single-region US model included in Policy Insight reflects an
exogenous national population and zero national economic migration. However,
state and county models reflect endogenous regional population and
endogenous regional economic migration. A multi-region model built with state
or county specifications will behave endogenously, including population, even if
the model’s regions collectively account for the entire US. However, the net

Economic and Business Research, Eller College of Business and Public Administration        16
The University of Arizona
                                                                    UPDATING REMI’S CONTROL FORECAST



economic migration effect among regions in any multi-region model automatically
is scaled in relation to the whole US. Therefore, the net economic migration for
the US as a whole in a multi-region model covering the US will be approximately
zero when summed across the economic migrations in the individual regions in
the model. It is recommended that a US-wide multi-region model be applied to
assess effects on a regional basis, including the region for “rest-of-US,” and to
assess effects on sub-national regions due to shocks to the “rest-of-US” region or
due to shocks used to build a new national control, but be applied cautiously to
evaluate effects on a national (“all-regions”) perspective.




Economic and Business Research, Eller College of Business and Public Administration        17
The University of Arizona
                                                                    UPDATING REMI’S CONTROL FORECAST




                                            Appendix B
                 COMMENTS FROM PEER REVIEW PROCESS

The Arizona Department of Commerce uses a process of peer and/or technical
review to ensure accurate and objective research. The following comments were
received as a result of the peer and technical review of this report and are
included to further the reader’s understanding of the discussion surrounding the
REMI model in Arizona’s economic community.


Jack Tomasik, Maricopa Association of Governments:
The main comment I have on Marshall's report is to expand on his observation
that REMI is a model not designed for forecasting, but for policy analysis. I agree
that REMI is not suited for short-term forecasts. However, REMI is a rich
structural model, and there are many policy variables that can be used by the
analyst to create long-term projections.
1. As an example, for Maricopa County it has been possible to account for critical
mass through industry-specific commodity access indices.
2. As another example, for Maricopa County it has been possible to also
account for the effects of increasing transportation congestion with future growth.
3. For a third example, it has been possible to account for new residential
development in Pinal County, using development project-specific data there to
estimate future net residency adjustment of personal income in both Maricopa
and Pinal Counties, thus accounting for commuter-based population growth in
Pinal.
4. As a fourth example, it is also possible in REMI to exogenously estimate
retirement and foreign immigration, so that future studies of these topics could
provide better long-term population projections.
To me, REMI is clearly superior than merely a cohort-survival model, that
requires exogenous assumptions of net in-migration, for long term population
projections. I believe that REMI's best use besides policy analysis is for long
term projections that develop a range of scenarios (high-medium-low). Another
benefit of REMI beyond policy analysis is to make county estimates for a number
of economic and demographic variables upon release of year-end employment
estimates by DES.




Economic and Business Research, Eller College of Business and Public Administration        18
The University of Arizona
                                                                    UPDATING REMI’S CONTROL FORECAST



Dan Anderson, Arizona Board of Regents
I had a chance to read Marshall's write-up on the REMI model and offer the
following.
I agree that REMI would be useful as a policy impact tool, and less as a
forecasting model. It wasn't constructed to do forecasting. A particular problem
is their use of BEA and not BLS data. You have to forecast up to two years
ahead just to get to the current time period this way. And do to the simultaneous
nature of REMI, and the lack of perfect accuracy in the estimates, you'll never
match current figures. All credibility is lost before you even start.
I don't think that having 15 models, one for each county, is at all workable. We
got pressed a lot be folks who wanted models for places like Greenlee and La
Paz counties. But they are so influenced by factors that just can't be modeled, it
isn't a useful path to pursue. And if you think a simultaneous system of
Phoenix/Mesa, Tucson, and the balance of state is hard to manage, see what
you get with 15 counties. I don't think its 15 times tougher than a statewide
model, I think its 225 (15 x 15) times tougher.
I wonder about the construction of some on the equations in REMI. If they use a
lot of lagged dependent variables, and the economy is on the decline, as it was in
2000, it can pull everything down. Marshall might try running some simulations
that begin in the mid 1990s or so to see what effect happens there. REMI may
be sensitive to the period selected.
I also wonder how sensitive the model is on both the upside and the downside.
Maybe he could shock it with good and bad news and see if it stabilizes at
reasonable levels. Or take a shock you know of, like the loss of lots of high-tech
jobs, and see how closely it comes to what was observed.
Marshall identified a problem that has been observed by others, the closing gap
between data reported from employers in terms of jobs and data reported from
the household survey in terms of employment. That gap has been closing
throughout the 1990s, so its not surprising that a database that ends in 2000 and
forecasts forward from there produces some counterintuitive results. A lot of
issues contribute to this problem. In my opinion, a lot of it has to do with
collection problems in the 1990 Census and the decision not to statistically adjust
the Census. For a lot of reasons, people work actively not to be counted in the
Census and unfortunately it is used in a lot of the statistical systems. If you're
interested in a fuller explanation of all the issues, I would refer you to Tom
Nardone at the Bureau of Labor Statistics.


William P. Patton, Ph.D., Tucson Electric Power Company
I have reviewed Marshall's paper, "Updating REMI Control Forecasts". The
comparisons between the control forecasts for REMI and Global Insight reveal

Economic and Business Research, Eller College of Business and Public Administration        19
The University of Arizona
                                                                    UPDATING REMI’S CONTROL FORECAST



some interesting differences between competitive general equilibrium models
(REMI) and econometric models (GI).
In general, competitive equilibrium models and/or input/output models contain
detailed structural equations that relate to the interrelationships among all of the
major industries in the economy. These types of models are extremely well
suited for use in policy analysis since they can track the impact of changes in the
final demand for one industry on other related industries. The strength of this
approach is that it enforces an internally consistent response among all
industries to a change in a policy variable.
In addition, a competitive equilibrium model yields extremely detailed information
on the impact of policy changes on individual industries.
There are several drawbacks to the use of competitive equilibrium models. First,
they require detailed data in order to update and sometimes rely on data that has
a considerable lead-time. Second, these models tend to be somewhat static in
nature. For example input/output matrices are based on a technology base
which may change over time. Finally, the competitive equilibrium models show
the impact of changes in policy or macroeconomic variables after the economy
has had a chance to reach a complete economic equilibrium. This equilibrium
rarely occurs in practice due to the fact that other economic shocks are
constantly affecting the economy.
Econometric models tend to be less detailed than competitive equilibrium
models. In addition, they do not necessarily maintain internal consistency among
various industrial sectors. However, they are easily and quickly updated. More
importantly, in a dynamic sense, they statistically measure the actual response of
industries to changes in the economic environment. Econometric models tend to
be more accurate as predictive models of what is expected to occur in the future,
based on past experience.
Based on Marshall's analysis there are two things that need to be investigated
further. First, why doesn't the model always converge to changes in
macroeconomic values? Second, and more importantly, there appears to be an
internal inconsistency between the growth in the number of jobs and the growth
in the labor force. REMI's explanation about multiple job holders is quite
unconvincing.
Marshall's analysis showed that there is an increasing gap over time between the
Global Insight and REMI control forecasts for major economic variables such as
Private Nonfarm Employment, Inflation and Personal Income. To some extent
this would be expected given the nature of the two models. I agree with
Marshall's conclusion that, "even though it (REMI) contains a forecast, it is not
designed for forecasting." Indeed, if REMI is to be used as a control forecast
model, it really should not be used past 10 to 15 years in the future, due to its
extreme diversion from the results of a predictive econometric model.


Economic and Business Research, Eller College of Business and Public Administration        20
The University of Arizona
                                                                    UPDATING REMI’S CONTROL FORECAST



REMI should definitely be used to evaluate the impacts of policy choices and to
understand the detailed industrial linkages of policy decisions. However, its use
for long term forecasting should be augmented by information from a reputable
econometric forecasting model.



Kent Ennis, Economic Consultant, CH2M Hill
Marshall has done an excellent job detailing and documenting the capabilities
and shortcomings of the REMI model. Even without attending the REMI
demonstration, I have long been aware (after going to several REMI seminars)
that it's no forecasting tool but for policy simulation work. Even so, Marshall's
comments raise further questions.
1) Does the lack of compatibility with and convergence to the "standard" data
sources (personal income, employment) create an always temporary issue that
can be satisfactorily resolved, or can we live with these differences, and their
implications? Do we yet have a confidence range for the accuracy of the model's
output?
2) The model seems complicated, which raises at least three issues:
First, Marshall loves models and he is probably the best, most credible,
practicing regional modeler in the region. Fine, but can and should others be able
to do REMI model work as well? Can others use/manipulate the model? While it
is potentially very dangerous for inexperienced users to play around with these
models, I personally think it's also a problem if only 1 person can do it.
While I guess that REMI's code is open, the model seems like a "black box". The
assumptions about what's "going on inside" have to be made crystal clear to
interested laymen.
3) If we are satisfied with the credibility of the model (a big if, but I am in no
position to say), then it's important that the results and range of outcomes of the
model's output (as well as the assumptions and its inner workings) can be put
into terms that reasonably intelligent people can understand it.
In other words, I can easily see a scenario where: "Marshall V., the only one who
knows how to run this model, came up with these outputs for policy choice
scenario X. In truth, there are data compatibility issues and we are checking with
REMIT to determine what's going on inside the model, but here are the results."
This is a gross oversimplification to be sure, but we need to do everything to
avoid the image of "our one guy with a black box says X."
A better scenario is that we clearly lay out the assumptions, how the model
achieves its outcomes, and the limitations of the data and the model's
methodology. Hopefully, others will be interested and capable of using the REMI



Economic and Business Research, Eller College of Business and Public Administration        21
The University of Arizona
                                                                    UPDATING REMI’S CONTROL FORECAST

model as well. The REMI model can be a very valuable ADDITIONAL tool for
policy analysis.




Economic and Business Research, Eller College of Business and Public Administration        22
The University of Arizona

								
To top