WIOD Conference: Industry-Level Analysis of Globalization and its Consequences

Vienna, 26 – 28, May, 2010

Dynamic Inter-Regional Econometric IO Modelling

Kurt KratenaGerhard StreicherMichael Wueger

• Past:

– National level: MultiMACI-IV, PROMETEUS

– Regional level (9 provinces): MultiREG

• Present:

– „Local“ model (99 districts) ETMOS

• Planned/Ongoing:

– New national model DEIO – „Dynamic Econometric Input-Output Model“

– From this, derivation of „family of regional models“

• (essentially) identical model structure

– Sub-Austrian level:

• 9 provinces

• 99 districts

– Supra-Austrian level:

• 27 EU members

WO

StS

K

N

V T

B

Modelling Activities – Past and Present

• Data Base:

– Supply-Use Tables;

– 2-digit NACE(2003) and CPA-level;

– Utilization of complete information provided by EUROSTAT (or national statistical offices);

• Modelling Structure:

– Quantity and price model

– Dynamic models of private consumption and production

• Intertemporal optimisation of households with durables and liquidity constraints

• Cost function and factor demand with short term fixed input (K) in a ‚dynamic duality‘ model

– Endogenous commodity structures:

• AIDS for private consumption

– Armington (CES) function for import demand

• „frame shifting“ in intermediate demand

General Characteristics of national Model DEIO

• Idea: use essentially identical model structure at different geographical levels

• One new element in model structure:

– trade matrix (inter-regional linkages)

• Data: decreasing data availability with increasing geographical detail

– Some information on 9 province-level;

– Very limited information on 35 NUTS3-level;

– Almost no information on level of districts

– When no information at regional level:

• „simple“ breakdown of last available regional level

– Trade matrix: increasing reliance on transport statistics

From national to regional

• Example: sectoral output , value added, employment

– National level: Full national account-information

– 9 Province level: Regional analysis of primary statistics at the firm level (materials inputs, production statistics)

– 99 districts: breakdown of province values using employment data

• Assumption: within one province, sectors show identical stucture in all districts

• Example: private consumption

– Regionalisation based on official consumer survey

– Utilization of information concerning demographic composition (age structure, educational characteristics)

• Here, we do not employ the assumption of identical structures below the province level

From national to regional

• Private Consumption:

– Commuting:

• Regional re-distribution of regional value added (wages/profits) to regional income

– Shopping:

• Regional re-distribution of place of demand for consumption goods

– Tourism:

• Regional re-distribution of leisure expenditure

– Data Base: official statistics (Commuters, Overnight Stays) and various official and commercial sources (Shopping Surveys)

• Inter-regional Trade:

– Balancing of regional demand and regional supply

Interregional Linkages

• Trade Matrix:

– Trade survey (MultiREG, 2001) among firms with tradeable products and wholesalers; feasibility of new trade survey……

• Result: trade matrix at level of 9 provinces

– Update using transport data;

– Transport data also used for further disaggregation to district level

– Not without problems (classifications – NSTR vs. CPA; modal split; statistical problems,….)

– But: We have access to transport data of quite high quality:

• „Verkehrsprognose 2025+“ – a projection of future traffic trends;

• Collaboration of traffic scientists and economists;

• One result: a very detailed description of current (2002, 2005) transport flows, involving a thorough overhaul of official transport statistics

Interregional Trade

• Consistent trade matrix derived by balance of goods:

– for each good, regional demand equals regional supply => balancing algorithm

• Transport matrix as starting point for trade-matrix• row and column sums: from regional and national make – use matrices

Interregional Trade

abroad region 1 region 2 region 3 region 4 region 5 region 6 region 7 region 8 region 9

abroadimported exports

=national imports

region 1region 2region 3region 4region 5region 6region 7region 8region 9

=

national exports total regional use (intermediary + final)

plac

e of

pro

duct

ion

place of consumption

=

=inter-regional tradeforeign exports

foreign imports

regional produc-

tion

• Implementation in modelling framework

– Ongoing work without definitive solution

– Status q uo:

• Changes in regional flows based on relative prices

• Similar to modelling of imports:

– Armington (CES) functions

• Ensure consistency within model framework

– New approaches arising from WIOD-environment ?!

• Issue: „endogenous arrival“ of sectors in regions where they had no previous representation

• Of minor concern in 9-province-model

• But of major concern in model of 99 districts

Interregional Trade

Example: Export boom in Vienna

• Increase in Viennese (900) exports of electrical equipment (CPA31)

Example: Price shock in Eisenstadt

• All output prices -5% in Eisenstadt (101)

• Future work:

– Alternative specifications/modelling approaches

– Gravity specification

• Exploratory co-operation with TU institute – link of ETMOS with systems dynamics model of traffic flows

– Test/implementation of New Economic Geography aspects

– Separate treatment of trade in intermediates and consumption goods

Interregional Trade

• Regional data and quality:

– National: full information from National Accounts

– Province: disaggregation (to a sizable extent) based on primary statistics

– District: breakdown based on employment; more information on CP and CG

• Econometrics and estimation:

– National: all coefficients econometrically derived from appropriate equations

– Province: some coefficients based on provincial data; some (majority?) derived from national elasticities with calibration to provincial data;

– District: all equations calibrated, based on provincial elasiticities

• Applications:

– National: base run (projections or forcasts), simulation runs

– Provincial: limited forecasting abilities (in conjunction with national model); mainly simulation applications

– District: VERY limited forecasting ability; simulation applications (transport and trade; location of sectors?)

Family of regional models – Appraisal and Applications

Econometric Modelling Philosophy

Philosophy of a dynamic EIO model: an alternative to static CGE models– Dynamic macro-modelling (DSGE): optimising agents with numerous institutional

frictions: durables, liquidity constraints, technology lock-in, adjustment costs for investment

– Keynesian closure with explicit modelling of labour markets: matching frictions, limited mobility and institutions (wage bargaining)

– Calibration based on econometric estimation with panel data sets (EU 27)

– Trade modelling beyond the Armington assumption

Modelling Block: Consumption

Private Consumption: intertemporal optimisation– Chah, et.al. (1995): household maximes the expected value E0 of utility U in t = 0,

chosing levels of K (durables), C (nondurables) with given financial wealth A:

measures the part of durables that can be financed and lies between 0 and 1 (close to unity)

– Lagrangean function of the maximization problem & first order conditions

– Relationship: marginal utility of C(UC)/marginal utility of K (UK) in t

– Relationship: [Uc(t+1) - Uc(t)] / [UC(t) - UK(t)]

0

0,,,1

ttt

t

AKCKCUEMax

11 11 ttDtttt KKpCYArA

0 tDt KpA

Modelling Block: Consumption

Private Consumption: intertemporal optimisation– Chah, et.al. (1995): solution yields explicit equation for C that can be estimated

econometrically:

– Long-run (cointegrating) equation with error term Z

– Short-run equation with interest rate rt and random shock .

Private Consumption: stock of K (durables)– Stock adjustment (household level) between bounds s and S (Eberly, 1994 and

Caballero, 1993)

– Approximated by stock adjustment with aggregate data:

111210 loglog tttttt ZrC

tt

KtDttt ZrRpKconstC 11logloglog/log

Modelling Block: Production

Cost function and factor demand with fixed K: ‘dynamic duality‘

– Cost functions with variable inputs xv and quasi fixed inputs xk: Shephard‘s lemma (demand for xv) and envelope condition (shadow price of xk), survey article by Galeotii (1996)

– With or without explicit adjustment costs for xk

– Adjustment of actual xk to optimal xk* in the spirit of Jorgenson‘s flexible accelerator

model

– Pindyck and Rotemberg (1983): Translog cost function with 2 quasi fixed inputs, L and K, and explicit adjustment costs estimation of Euler equations

• L, E, Mm, Md cost function with quasi fixed K and without explicit adjustment costs and including technical change

– Factor demand with autonomous (bias), embodied and induced (Jin and Jorgenson,2008) technical change

– (Implicit) investment demand as adjustment towards xk * with expectation formation mechanism

– Mm is the link to trade modelling

Conclusions and Future Research

• Trade matrix as a general consistent framework of modelling.

• Consistency between: (i) Trade and IO/National Accounts Data (no problem in WIOD), (ii) Trade and Transport Data (monetary and physical)

• Econometric estimation at an adequate regional level with recent calibration (combining EIO and CGE philosophies)

• Dynamic New Keynesian modelling in consumption and production: intertemporal optimisation with frictions, institutions, etc. (durable goods, liquidity constraints, adjustment costs for factor adjustment…)

• Trade modelling beyond the Armington assumption:– Differentiating final (consumption) imports and intermediate imports– Testing economic geography assumptions (increasing returns) vs.

Importance of transport costs– Testing new new trade theory (?)