Constructing Interregional CommodityIndustry Accounts for US States

1
Constructing Interregional Commodity-Industry
Accounts for US States

• Randall W. Jackson, Director
• Yasuhide Okuyama, Research Associate
• Walter Schwarm, Postdoctoral Research Fellow
• Regional Research Institute, West Virginia
  University
• Morgantown WV 26508
• Presentation to the Bureau of Economic Analysis
• Washington DC, December 4, 2003

2
Objectives

• Implement an efficient procedure for estimating
  an Interregional Social Accounting Matrix (ISAM)
  for the US
• Make effective use of the US Bureau of
  Transportation Statistics Commodity Flow Survey
  Data

3
Motivation

• General inaccessibility of ISAMs for the US
• Single-region IO and SAMs relatively accessible
• ISAMs needed for own sake and for use in other
  modeling frameworks
• CGE
• Economic drivers for other models, e.g.,
  transportation demand models
• Lack of generally accepted method in US context

4
Foundation - IMPLAN data

• Readily available at reasonable cost
• Comprehensive and consistent economic and
  geographic coverage
• Flexible regional definitions
• IMPLAN software can be used to generate
  single-region SAMs as modeling foundations

5
IMPLAN data issues for ISAMs

• Single region SAMS are generated independently
• Sample Problem SAMs for eastern and western US
  would not necessarily sum to US SAM totals.
• Consistency problems increase with greater
  numbers of regions
• Domestic imports and exports are not distributed
  across other subnational regions
• IMPLAN Classification schemes are not completely
  consistent with SNA, and no Land sector

6
This project

• Uses an export distribution estimation method
• Integrates the steps necessary to
• Generate the interregional trade flow portions of
  the US ISAM, while
• insuring the consistency of both the
  single-Region SAM accounts and the system as a
  whole

7
Procedure

• Define the model regions (50 states plus
  Washington, DC)
• Define sectoral classifications and detail (54
  commodities and industries, IMPLAN detail on
  factors and institutions)
• Generate single-region social accounting matrices
  IMPLAN data are in a format consistent with the
  GAMs CGE software (18 files/SAM)
• Estimate interregional trade distributions by
  commodity using BTS data
• Apportion aggregate interregional commodity flow
  values to regions, by industry and by institution
• Adjust the accounts to insure the integrity of
  the intra-regional and system-wide accounts

8
Methodological Overview
Obtain and Transform IMPLAN output
Calibrate flow value equations distance and
population elasticities (?, ?)
Apportion institution and industry-based domestic
commodity exports from each region using
parameterized flow equation
Re-compute total domestic imports and total
domestic exports
Set model foreign exports equal to actual exports
Force domestic import totals to equal
domestic export totals by redistributing regional
foreign exports
Adjust industry and institution foreign exports
ensuring that the sum is consistent with the
national model (RAS)
Reorder intraregional partitions and create the
I-SAM
9
Preliminary Steps

• Defining regions and sectoring schemes
• 50 regions in our model
• Industry/commodity sectors detail (54 sectors)
  consistent with commodity flow data
• IMPLAN factor and institutional classification
  and detail retained
• Single-region SAMs
• IMPLAN data are in a format consistent with the
  GAMs CGE software (18 files/SAM)

10
Data Crosswalk A Sample
11
SAM Sectors
12
Single-region SAM
13
ISAM format

• 3-region format shown next generalizes to
  n-regions

14
Interregional SAM Configuration
15
Export distributions

• BTS Commodity Flow Survey
• Severe disclosure problems
• BTS provides summary data
• Value of shipment by distance ranges
• Need to generalize the distance-volume
  relationships embedded in the BTS data, to smooth
  out observed irregularities in O-D shipments data

16
Necessary Assumptions

• Intra-regional IMPLAN SAM data are correct
• Allows a focus solely on interregional trade
• Generalized functions apply equally across
  regions
• Interregional exports distributions are fixed for
  each commodity, while export levels vary with
  regional production activity
• Domestic export (interregional flow) and import
  distributions are identical across industries and
  institutions (as in MRIO model)

17
Final Assumption Elaborated

• Domestic export and import distributions are
  identical across industries and institutions (as
  in MRIO model)
• Each industry or institution imports/exports the
  same proportion of a commodity used/produced
• The proportions will differ for each commodity

18
Primary tasks for 50 states US ISAM generation

• Trade flow estimation
• Account reconciliation

19
Regional Definition

• Regions exhaust the entire US
• This step provides control totals such that the
  sums of domestic imports by commodity across all
  regions must equal sums of domestic exports

20
Exports Distribution

• Because supply is given by single-region SAMs,
  the estimating equation need only be a function
  of
• transportation costs (interregional distances)
  and
• region-specific commodity demand (also available
  from IMPLAN data)

21
Export Estimation

• For each commodity i, let the predicted value of
  the flow from region m to region n be computed as

22
Variables
is a weight reflecting region ns demand for
imports of commodity i,
is the distance separating region m from region n,
is total domestic commodity i exports from region
m
ideally, are actual (observed) shipments derived
from observed values
are elasticities on distance and commodity demand
23

• However, because of the gaps in the BTS data,
  they are not used directly
• Instead, we generate synthetic observed flows

24
(No Transcript)
25
(No Transcript)
26
Generalized function

• Estimate the commodity-specific distance decay
  functions with double log regression
  specification
• Use regression parameters to create synthetic
  observed flows for interregional distances via
  optimization
• Calibrate l and b

27
Distance Decay Function
where fmn commodity flow from region m to
region n dmn distance between regions m and
n bi commodity-specific coefficients Here
, all coefficients ( ) were significant at
p0.759 This
step captures only the distance decay relationship
28
Regression-generated Commodity Flow
where Regression-generated
(synthetically observed) commodity flow from
region m to region n dmn Interregional
distance s Size of buffer around
interregional point-to-point distances Xr
Domestic export share
29
Two-Step Optimization

• Calibrate the buffer
• Step 1 Find optimal buffer width, s, by
  minimizing absolute difference between
  regression-generated import demand and IMPLAN
  import demand
• where
• Regression-generated total import
  demand for region n
• IM IMPLAN domestic import demand (IM)

30
Two-Step Optimization

• Step 2 Calibrate l and b by minimizing the
  absolute percentage error between logit-predicted
  and regression-generated flows
• where
• Predicted flow of commodity i from region m
  to region n
• Regression-generated commodity flow from
  region m to region
  n

31
Logit function

• The final l and b parameters, which represent the
  relative importance of distance and demand in
  trade of this commodity, are then inserted into
  the logit function to generate the final export
  distributions

32
Computer Application

• Once the parameters have been generated, a
  program written in C implements the remainder
  of the algorithm
• Collects and organizes all input
• Reconciles the sum of regional foreign exports
  with known total from national model (RAS)

33
Reconciliation

• Because the single-region SAMs are generated
  independently and sequentially, the result is
  unlikely to be balance with national control
  totals.
• Thus, an additional step is implemented prior to
  regional and sectoral import and export
  apportionment

34
RAS (bi-proportional adjustment) for each
commodity

• Given
• National total foreign exports (FE) from national
  SAM
• Regional domestic and foreign export estimates
  from single region models
• We construct an nx2 table n regions, domestic
  exports (DEr) and foreign exports (FEr)
• Foreign export column margin FE
• Domestic exports column margin S(DEr FEr)
  FE
• Row margins DEr FEr for all r
• Adjust bi-proportionally
• Manual inspection and adjustment may be required
  for some commodities

35
Sector specific interregional commodity flows

• Assumption The interregional export
  distributions will apply equally to both
  industries and institutions
• Interregional export distributions (rows) are
  then unstandardized by IMPLAN-generated exports,
  and normalized by column sums to generate import
  distributions, which are again applied to both
  industrial and institutional imports

36
Final reconciliation

• For each commodity, the sum of regional exports
  for each region pair must equal the sum of
  regional imports for reverse pair
• Domestic and foreign imports are adjusted to
  ensure the consistency

37
Resulting framework
38
Industry and Commodity Accounts

• Industry Row Total regional industrial output
  (make)
• Industry Column Total regional industry input
  (use)
• Commodity Row Total regional commodity supply
  (disposition)
• Commodity Column Total regional commodity supply

39
Factor and Institution Accounts

• Factor Row Total factor receipts (payments to
  factors) of production
• Factor Column Total factor payments to
  institutions (and trade)
• Institutions Row Total institutions receipts
  (payments to institutions)
• Institutions Column Total regional institutions
  expenditures

40
Summary

• Procedure uses but overcomes the deficiencies of
  the Commodity Flow Survey data
• Implemented via two optimization procedures
• Generalized distance decay function
• Logit form of export distribution function
• Export/import adjustment insures single-region
  and system-wide accounting consistency

41
Extensions-Overview

• Establish a validation framework
• Assess model performance
• Modification/refinements
• Buffer width
• Intangibles
• Updating
• Re-estimate
• Assess model performance

42
Extensions Validation Framework

• Estimates of regional trade generated by our
  method could be validated/assessed using
• Federal Highway Administrations Freight Analysis
  Framework Database(FAFD)
• Question of evaluating the amount of unique
  informational content within the database as it
  also uses the CFS. However this is enriched with
  other data sources including Survey, Reebies
  TRANSEARCH, and SP/DRI Regional estimates.
• Selected (available) interstate estimates from
  BTS
  

43
Extensions Buffer widths

• Current buffer widths (s) are estimated in the
  same way for all states and for all distances
• May underestimate regional trade in the small
  states, and overestimate interstate trade in the
  largest states
• Can be modified using spatial/geographic
  information

44
Modifying the buffer (s)

• Should address relative sizes of states
• Should address relative location (i.e., border
  states vs. central states)
• Should be formulated so as to fit within the
  existing optimization procedures

45
Extensions- Intangibles

• Intangibles like services and energy are handled
  in a non-unique way as a function of a states
  trade in tangible commodities.
• Some sectors, such as Accounting, Finance,
  Software, may not be subject to the same kinds
  of friction of distance effects as are the
  physical commodities
• Differing trade profiles may also apply to some
  commodities such as electricity.

46
Extensions- Intangibles

• Proxy solution is potentially found in a distance
  sensitivity(elasticity) matrix applied to the
  current estimates.
• Based on the relative concentration of firms
• Many firms ? High local bias
• Ex Hair salons, Million in US.
• Few firms ? Low local bias
• Ex Finance Houses, less than 1000.
• Incomplete solution
• Industry concentration skews result
• Availability of data necessary to estimate such a
  matrix of sufficient detail is unclear

47
Extensions- Intangibles

• Complete solution requires the development and
  evaluation of estimates for non-commodity
  interregional flows
• Review and devise the methods for distributing
  non-tangible commodity flows (such as Services
  sector)
• Survey the interregional trade patterns of
  non-tangible commodities
• Evaluate the methods for estimation
• Integrate and adjust with the tangible commodity
  part, and complete I-SAMs

48
Extensions-Updating

• Interest in updating to maintain current
  estimates.
• Investigate and devise the updating methods for
  I-SAMs
• Non-Survey and/or hybrid update methods
• Linkage with econometric models
• Integration of interregional models (tangible and
  non-tangible commodities)
• Provides motivation for greater automation within
  the spreadsheet portion, yielding a more user
  friendly final package