close

Se connecter

Se connecter avec OpenID

BEA Regional Price Parities

IntégréTéléchargement
MAJOR NEW FEATURES IN PI+ V1.7
Regional Economic Models, Inc.
Data Modifications and New Policy
Variables
Sherri Lawrence
Senior Vice President
Regional Economic Models, Inc.
BEA Regional Price Parities



Real Personal Income for States and Metropolitan Areas, 20082012
RPP’s measure geographic differences in the price levels of
consumption goods and services relative to the national
average.
Develop procedure for normalizing REMI’s estimate of state
PCE’s with RPP’s for 2008-2012. Apply 2008 adjustment back to
1990, apply state adjustments to each county within a state.
BEA Regional Price Parities, continued
Regional Price Parities
REMI relative to BEA
1.15
1.1
1.05
1
0.95
0.9
0.85
0.8
2008
2009
2010
2011
2012
ALABAMA
ALASKA
ARIZONA
ARKANSAS
CALIFORNIA
COLORADO
CONNECTICUT
DELAWARE
DISTRICT OF COLUMBIA
FLORIDA
GEORGIA
HAWAII
IDAHO
ILLINOIS
INDIANA
IOWA
KANSAS
KENTUCKY
LOUISIANA
MAINE
MARYLAND
MASSACHUSETTS
MICHIGAN
MINNESOTA
MISSISSIPPI
MISSOURI
MONTANA
NEBRASKA
NEVADA
NEW HAMPSHIRE
NEW JERSEY
NEW MEXICO
NEW YORK
NORTH CAROLINA
NORTH DAKOTA
OHIO
OKLAHOMA
OREGON
PENNSYLVANIA
RHODE ISLAND
SOUTH CAROLINA
SOUTH DAKOTA
TENNESSEE
TEXAS
UTAH
VERMONT
VIRGINIA
WASHINGTON
WEST VIRGINIA
WISCONSIN
WYOMING
Expanded Property Income Components
• The model now incorporates historical State Property Income
data from the BEA to expand Property Income into its
components of Personal Dividend Income, Personal
Interest Income, and Rental Income of Persons. The
state shares are applied to applicable County Property Income
in order to estimate the component shares at the county level.
• Each component is now forecasted independently (with the
same equation previously used but with a potentially different
national trend).
• Policy variables have been added for each component.
Expanded Transfer Payment Components
• The model now incorporates historical County Personal
Current Transfer Receipts based on 21 detailed
components.
• Each component is now forecasted independently (with the
same equation previously used but now based on four
major types of transfer payment, and with a potentially
different national trend).
• Policy variables have been added for each component.
Expanded Transfer Payments, continued
Current transfer receipts of individuals from governments
Retirement and disability insurance benefits
Social Security benefits
Other retirement and disability insurance benefits
Medical benefits
Medicare benefits
Public assistance medical care benefits
Military medical insurance benefits
Income maintenance benefits
Supplemental security income (SSI) benefits
Earned Income Tax Credit (EITC)
Supplemental Nutrition Assistance Program (SNAP)
Other income maintenance benefits
Unemployment insurance compensation
State unemployment insurance compensation
Other unemployment insurance compensation
Veterans benefits
Education and training assistance
Other transfer receipts of individuals from governments
Current transfer receipts of nonprofit institutions
Current transfer receipts of individuals from businesses
Historical Data Modifications
• County nonresidential investment and equipment
investment is now estimated based on the county’s
employment weighted by capital use share of the
nation instead of the county’s construction industry
employment share of the nation.
Historical Data Modifications, continued
• County residential investment is now based on the
county’s disposable income share of the nation instead
of the county’s construction industry employment
share of the nation.
Historical Data Modifications, continued
• County output by industry is now estimated based on
the county’s earnings share of the nation instead of the
county’s compensation share of the nation when the
earnings share is greater than the compensation share.
Historical Data Modifications, continued
• County residence-adjusted employment is calculated
based on estimated commuter data when not
inconsistent with the flow of residence adjusted income
reported by the BEA.
New Productivity Policy Variables
Alternative labor productivity and labor access policy
variables will have an immediate effect on market shares,
instead of a lagged effect.
• An increase in the current Labor Productivity policy variable leads
to an immediate short-term loss of jobs because the factor that
converts from output to employment (labor productivity) is changed
by 100% of the user-supplied policy variable value, but the cost of
production response is lagged, resulting in only 20% of the
improved labor productivity benefiting market shares in the first
year, 40% in the second year, etc.
Productivity Policy Variables, continued
• The assumption is that it takes time for markets to respond to
changes in business conditions, such as the cost of production. This
response is appropriate for some scenarios, such as labor
productivity changes due to new technology implementation, or
production equipment, but may not be appropriate for others.
• Overall the short-term results are very different, but the longerterm impacts are essentially the same.
Labor Productivity Policy Variables
Chart 1: 1% Increase in Labor Productivity, All Industries, All Regions – RLABPV (current default)
Chart 2: 1% Increase in Labor Productivity, All Industries, All Regions – RLABPV_Immediate (alternative)
Labor Access Policy Variables
Chart 3: 1% Increase in Labor Access Index, All Industries, All Regions – FLPRMPV (current default)
Chart 4: 1% Increase in Labor Access Index, All Industries, All Regions – FLPRMPV_Immediate (alternative)
Occupational Training Policy Variables
Chart 5: +100 Jobs Occupational Training, All Occupations, All Regions – OTRPV (current default)
Chart 6: +100 Jobs Occupational Training, All Occupations, All Regions – OTRPV_Immediate (alternative)
Data Unsuppression
and Automatic Parallelization
Hoksung Yau, PhD
Senior Analyst
Regional Economic Models, Inc.
DATA UNSUPPRESSION
Overview

BEA Historical Data
 PI+
input
 All US states and counties at 2digit industry level

Data Suppressed on Purpose
 Compensation
 Employment
 Personal
Income
Goals and Method

Constraints
 Industrial
consistency
 Regional consistency

Smoothness Across Years
 Compensation/Employment/Personal
 Compensation

Income
Rate/Earnings Rate
Unsuppression Process
 Estimate
suppressed numbers
 Run models to adjust these numbers, so that the constraints are
satisfied while keeping all concepts’ trend lines as smooth as possible.
Major Changes Since V1.7

Improved unsuppression model
 More
accurate initial estimation
 Better parameters
 More concise structure



Implemented efficient algorithms for fast computing
Parallelized the most time-consuming computation
Automated unsuppression process
Data Quality Improvement
PI+ v1.6
PI+ v1.7
Efficiency Enhancement
The short processing time enables us to achieve the best solution by testing as many different
unsuppression strategies as possible.
PI+ AUTOMATIC PARALLELIZATION
Automatic Parallelization



Key parts of the 64-bit version of the model have been
converted into multi-threaded code in order to utilize multiple
processors (cores/threads) simultaneously in a shared-memory
multiprocessor computer.
The maximum number of threads will be detected
automatically upon model use.
The larger the model in terms of sectors and regions, the
larger the savings in run time.
Automatic Parallelization, continued
Single CPU vs Parallelization
Model
Sector
Region
Single CPU (sec)
Parallelized (sec)
GAP
Uwin12.23
23
29
34
30
12%
Fwin12.23
23
67
127
83
35%
SCwin12.70
70
4
17
11
35%
Wwin12.70
70
23
220
79
64%
Mwin12.70
70
83
1,934
564
71%
NEBwin12.70
70
93
2,200
608
72%
AFwin12.160
160
4
65
40
38%
Wwin12.160
160
23
538
162
70%
NGCwin12.160
160
65
4,099
1,110
73%
Overall Average
Processors: Intel® Core™i7-2600 CPU @ 3.4GHz (4 cores, 8 threads)
52%
FORECAST ASSUMPTIONS
Jerry Hayes
Senior Analyst / Project Management
Regional Economic Models, Inc.
Forecast Assumptions

Users may create and store own assumption sets.




Macroeconomic Update, Employment Update, Population Update
National demographic assumptions
Users can easily switch forecasts assumptions and generate
new forecast.
Assumptions are available across workbooks.
Forecast Assumptions
NATIONAL DEMOGRAPHIC
FORECAST
Jerry Hayes
Senior Analyst / Project Management
Regional Economic Models, Inc.
National Demographic Forecast



The national demographic forecast may now be adjusted.
Changes to national demographic variables will flow through
to the regional model.
However, changes to the national demographic forecast will not
lead to endogenous changes in the economic forecast.
National Demographic Adjustments


Users can create new demographic assumptions and reuse
them with other forecasts.
New demographic policy variables can be used for simulations.
National Demographic Adjustments

Birth Rates

Survival Rates

International Migration

Participation Rates
National Demographic Assumptions


Users may select from predefined “assumption” datasets, or
create their own, and generate a new national control forecast.
Alternative migration assumptions are included :



Low
Middle
High
National Demographic Assumptions
Double Precision, Year Selection
Dialog and Internationalization
Gabor Lukacs
Senior Software Engineer
Regional Economic Models, Inc.
Legacies of Development Environment and
64-bit Upgrade





Memory space (Model outside of crowded 32 bit area)
Larger models (All MSA-s ~ 380 regions)
Speedier calculations (Parallelization)
Double Precision (Results for smaller changes)
Internationalization (Alternative language display)
Double Precision


The REMI model now works with Double Precision which use
approximately 16 significant digits when performing
calculations (instead of 8 significant digits it used before).
There is also currently a very slight difference in the baseline
forecast generated with our 32-bit version of the application
vs. the 64-bit version. This difference is reduced by using
double precision calculations.
Double Precision, continued

Previously, when a small change was introduced in a relatively
large region, the impact results may have been “noisy”
(wavering up and down year to year without apparent reason).
Year Selection Dialog

Preferences can be
stored and a default
can be set with a
new dialog, to allow
for personalization.
Other Preferences
Further areas we plan to personalize the interface:
•
•
•
•
Grouping regions, industries or commodities,
Managing unit selection, precision, printer preferences,
Storing color scheme and ribbon theme selections,
Making these transportable between models.
Future “Internationalization” Option


The model application has been modified to more easily allow
for international compatibility (alternative language display for
the user interface and data inputs and outputs).
Localization (translating the user interface and resources into
different languages) will not be done by REMI, but could be
provided to REMI for inclusion in an Internationalized version
of the application.
Sample Spreader - English
Sample Spreader - Hungarian
REMI Research
Equation Re-estimation
Wei Kang, PhD
Regional Economist
Regional Economic Models, Inc.
State and Local Government Spending Equation
State and Local Government Spending

State and local government demand equation (original)



=


∗





=


∗











∗


∗ 
∗


∗ 

Where
SG/LG = State/local government expenditures in chained 2005$
R = calibration factor for state/local government expenditures
GDP = gross domestic product in chained 2005$
N = population
β = GDP elasticity of state government expenditures
 = GDP elasticity of local government expenditures
k = state
t = time
u = U.S.
State and Local Government Spending

State and local government demand equation (new)


=


=


∗
 
__
∗
 
__
∗
∗




∗ 
∗ 
where
__
_ ,
1 −  ∙ _ +  ∙ __−1 ,
=


_
= 





 = the speed of adjustment of the moving average.
  = 0;
ℎ.
State and Local Government Spending

Data: Panel data 1997-2011
 GDP:
BEA
 State and local government spending: Government
expenditure from Census Bureau
 Population: BEA

Methodology: Fixed effects model
State and Local Government Spending

GDP elasticity of government spending (GRP)
New Estimates
Original Model Parameters
State government spending (β)
0.377
0.904
Local government spending (Υ)
0.498
0.798

State fixed effects calibration factor
New Estimates
State
government
spending
Local
government
spending
Original Model Parameters
Highest
Lowest
Highest
Lowest
State
Alaska
Nevada
Alaska
Texas
Calibration
factor
2.448
0.732
1.760
0.690
State
D.C.
Hawaii
New York
Hawaii
Calibration
factor
2.275
0.418
1.442
0.394
State and Local Government Spending

PI+ simulation results:
 The
state government spending responds less then the
local government spending to a given change in GDP
per capita.
 The use of the moving average term of GDP leads to a
slower response in government spending due to
changes in GDP per capita.
Housing Price Equation
Housing Price Equation
 =
1




− 1 + 2
−1
−1
−1


−1
−1
where
PH = relative housing price;
RYD = real disposable income;
N = population;
u represents the U.S.;
t represents the time period;
1 = income elasticity of housing price;
2 = population elasticity of housing price.
+ 1 × −1
Housing Price Equation

Data: State-level annual panel data 1990-2013
Housing price index: Federal Housing Finance Agency HPI
quarterly data (All-transaction indexes)
 Income (BEA) and population (BEA) data: PI+ V1.6


Methodology:


Pooled regression with panel data
Overall demand response:
New Estimates
Original Model Parameters
Income Elasticity
0.462
0.211
Population Elasticity
0.648
0.548
Housing Price Equation

Estimating State/MSA price elasticity


−1=×
−1

where
HP = the Freddie Mac Home Price Index;
BP = the number of building permits for single-family homes;
U = the number of housing units (1 unit detached);
t represents the time period;
b is the parameter to be estimated.


Methodology: time series OLS regressions
Data:



House price index: Freddie Mac house price index
Building permits: Census Building Permits Survey
Housing units: American Community Survey 1-year estimates
Housing Price Equation

Estimating State/MSA scaling factors
 − 
+

  =

where
 = estimated state/MSA parameters;
 = the mean of ′ ;
 = the standard deviation of ′ .

Regional parameters:   × 1 and   × 2
Housing Price Equation
State scaling factors
Housing Price Equation
County-level scaling factors
Housing Price Equation

PI+ simulation results
 The
larger new coefficients lead to larger responses in
relative housing price.
 The original housing price equation followed
population changes more closing due to the relatively
higher population elasticity coefficient. The new
equation is more balanced between response to
changes in population and income since the
coefficients are now closer in magnitude.
Consumption Equation
Consumption Equation
(2). Age
Composition Effect
(1).
Calibration
Effect

,
=




×
7

7



%,
× ,


%,
× ,
(3). Regional Effect
×


   (2
(4). Marginal
Income
Effect
×








(5). RegionSpecific
Marginal Price
Effect

×

,


,


,


,

(6). U.S. (7). Local
Forecast Population
Effect


,
×  × 

Consumption Equation

Estimating income and price elasticities by commodity



∆
=∙∆
+ ∙∆
+ 



where
 = consumption expenditure on the commodity for time period t;
 = population;
 = real disposable income for time period t;
 = price index of the commodity for time period t;
 = average price index of all commodities for time period t;
 and  are the parameters (elasticities) to be estimated;
 is the error term.
Consumption Equation

Data: BEA National Income and Product Accounts
1990-2012
 Disposable
Personal Income
 Price Indexes for Personal Consumption Expenditures
by Type of Product
 Personal Consumption Expenditure by Type of Product

Methodology: OLS

 < 1, 
Classification:
 > 1, 
Consumption Equation

Panel Data Approach:
,

∆
=  ∙ ∆
+  ∙ ∆


,

∆
=  ∙ ∆
+  ∙ ∆


,
+ ,

,
+ ,

L stands for luxuries and N stands for necessities.

Normalization:
 ×  +  ×  = 1
 ×  +  ×  = 1
where
 =
  
;

 =
  

Consumption Equation
New Estimates
Income Elasticity
Price Elasticity
Original Model Parameters
Income Elasticity Price Elasticity
Necessities
0.614
-0.657
0.49
-0.66
Luxuries
1.729
-1.649
2.24
-1.82
Consumption Equation

PI+ simulation results
 The
consumption responses are similar between the
two versions, due to similar coefficients.
 For a given change in population or output, the price
index changes more in the new V1.7 model, which has
a stronger effect on the consumption responses.
REMI RESEARCH
Auteur
Document
Catégorie
Uncategorized
Affichages
6
Taille du fichier
3 007 KB
Étiquettes
1/--Pages
signaler