Spatial data analysis, multiregional modeling and macroeconomic growth

1

• Spatial data analysis, multiregional modeling and
  macroeconomic growth
•  
•   by
• Attila Varga
• Center for Research in Economic Policy (GKK)
• and
• Department of Economics
• University of Pécs, Hungary

2
Introduction

• A-spatial mainstream growth theory
• K, L and A only? How about their spatial
  arrangements?
• Why should we care about space?
• - Transport cost (evident, but can be
  integrated)
• - Spatial externalities (requires a different
  approach)
• Policy relevance (EU)

3
Outline

• Introduction
• Technological progress, spatial structure and
  macroeconomic growth An empirical modeling
  framework
• Geographical growth studies - methodological
  issues
• Dependence in space Spatial data analysis in
  knowledge spillover research
• Spatial macro modeling Integrating macro and
  regional levels
• Endogenizing spatial structure
• Summary

4
Technological progress, spatial structure and
macroeconomic growth

• Complex issue treated in three separate fields of
  economics
• A. EG Endogenous economic growth models
  endogenized technological change in growth theory
  (Romer 1986, 1990, Lucas 1986, Aghion and Howitt
  1998)
• in Romer (1990)
• for-profit private RD
• knowledge spillovers and growth
• rate of technical change equals rate of
  per-capita growth on the steady state
• Simplistic explanation of technological progress,
  no geography

5
Technological progress, spatial structure and
macroeconomic growth
B. IS Systems of innovationliterature
innovation is an interactive process among actors
of the system (Lundval 1992, Nelson
1993) actors of the IS - innovating firms -
suppliers, buyers - industrial research
laboratories - public (university) research
institutes - business services - institutions
level of innovation depends on - the
knowledge accumulated in the system - the
interactions (knowledge flows) among the
actors - codified, non-codified (tacit)
knowledge and the potential significance of
spatial proximity - does not say anything about
geography and growth
6
Technological progress, spatial structure and
macroeconomic growth
C. NEG New economic geography models
endogenized spatial economic structure in a
general equilibrium model (Krugman 1991, Fujita,
Krugman and Venables 1999, Fujita and Thisse
2002) - spatially extended Dixit-Stiglitz
framework - increasing returns, monopolistic
competition - spatial structure depends on some
parameter conditions that determine the
equilibrium level of centrifugal and centripetal
forces - cumulative causation - C-P model by
Krugman still the point of departure - models
quickly become complex simulations if analytical
solutions are not accessible - Technological
change not explained (not even included until
very recently), the study of its relation to
growth is a recent phenomenon
7
Technological progress, spatial structure and
macroeconomic growth

• Theoretical integration endogenous growth and
  new economic geography (Baldwin and Forslid 2000,
  Fujita and Thisse 2002, Baldwin et al. 2003)
• EG, IS, NEG methodological problems in
  THEORETICAL integration (dramatically diverging
  initial assumptions, different theoretical
  structures, research methodologies)
• EMPIRICAL integration very few work (Ciccone and
  Hall 1996, Varga and Schalk 2004, Acs and Varga
  2004)

8
Technological progress, spatial structure and
macroeconomic growth an empirical modeling
framework

• Starting points
• Technological change is a collective process that
  depends on accumulated knowledge and interactions
  (IS)
• Technological change is the simple most important
  determinant of economic growth (EG)
• Codified and tacit knowledge different channels
  of spillovers (the geography of innovation
  literature)
• Centripetal and centrifugal forces shape
  geographical structure via cumulative processes
  (NEG)
• The resulting geographic structure is a
  determinant of the rate of growth (NEG)

9
Technological progress, spatial structure and
macroeconomic growth an empirical modeling
framework

• Y AKaLß (EG)
• The Romer (1990) equation as in Jones (1995)
• dA ? HA? Af,
• - HA the number of researchers
  (person-embodied, codifiable/tacit knowledge
  component of knowledge production)
• - A the total stock of technological knowledge
  (codified knowledge component of knowledge
  production)
• - dA the change in technological knowledge
• - ? the research productivity parameter
  (0lt?lt1)
• f codified knowledge spillovers parameter
• - reflects spillovers with unlimited spatial
  accessibility
• ? the research spillovers parameter
• - reflects localized knowledge spillover effects
• - regional and urban economics and the new
  economic geography suggest ? increases with
  geographic concentration of economic activities

10
Technological progress, spatial structure and
macroeconomic growth an empirical modeling
framework

• Eq.1 Regional knowledge production
• Kr K (RDr, URDr, Zr)
• Eq.2 Agglomeration effect RD spillovers
• ?Kr/?RDr f (RDr, URDr, Zr)
• Eq.3 RD location
• dRDr R(?Kr/?RDr)
• Eq.4 Geography and ?
• ? ? (GSTR(HA))
• Eq.5 dA ? HA? Af
• Eq.6 dy/y H(dA, ZN)

11
Empirical research on geography, technology and
growth 1986-2004
1986-2004 253 papers on the geography of
knowledge spillovers journal articles 175 books,
book chapters, working papers 78

12
Geographical growth studies - methodological
issues
13
Geographical growth studies - methodological
issues

• I. Dependence in space Spatial data analysis in
  knowledge spillover research
• II. Spatial macro modeling Integrating macro and
  regional levels
• III. Endogenizing spatial structure

14
I. Dependence in space Spatial data analysis in
knowledge spillover research
The spatial distribution of US innovations, 1982
15
I. Dependence in space Spatial data analysis in
knowledge spillover research

• Tendency of innovation to cluster in space
• Clustering is a consequence of dependence among
  spatial units
• Spatial dependence makes traditional econometric
  techniques no longer appropriate (Anselin 1988,
  2001)
• Spatial data analysis
• Exploratory spatial data analysis (ESDA)
• Spatial econometrics

16
I. Dependence in space Spatial data analysis in
knowledge spillover research

• ESDA global and local measures of spatial
  dependence
• Global measures general form
• G Si,j wij cij
• Local measures
• Moran Scatterplot
• Local Moran

17
Moran Scatterplot
18
Local Moran statistics
19
I. Dependence in space Spatial data analysis in
knowledge spillover research

• Spatial econometrics models with high intuitive
  value to study spatial knowledge spillovers
• Basis innovation equation in a form of a
  classical linear regression
• y Xb e
• where y innovation output x inputs to
  innovation
• Modeling geographical spillovers two main
  issues (Anselin 2003)
• A. their spatial extent (local or global)
• B. direct or indirect modeling

20
I. Dependence in space Spatial data analysis in
knowledge spillover research

• Modeling the spatial extent of spillovers
• A.1. global autocorrelation modelling
• e lWe u I - lW-1 u
• A.2. local autocorrelation modelling
• e I gW u

21
I. Dependence in space Spatial data analysis in
knowledge spillover research

• Direct or indirect modelling the most commonly
  used solutions
• B.1. Direct modelling (the spatial lag model)
• y (I - rW)-1 Xb ( I-lW)-1 u rWy Xb u
• B.2. Indirect modelling (the spatial error
  model)
• y Xb ( I-lW)-1 u

22
The facts spatial econometrics in empirical
innovation research
23
Spatial econometrics Facts, needs and
opportunities

• Urgent need for extending the toolbox
• spatial logit, probit, Tobit, Poisson, panel
• User-friendly softwares with support
• New intermediate level textbook with applications
  

24
II. Spatial macro modeling Integrating macro and
regional levels

• Q how to integrate eqs (1) to (3) (regional
  level) with eqs (5) and (6) by eq (7)
  empirically?
• An example the EcoRET model (Schalk and Varga
  2004, Varga and Schalk 2004)

25
EcoRET The main characteristics

• macroEconometric model with Regionally
  Endogenized Technological change
• General features (cost minimization vintage
  capital production function technology and
  labor/capital demand, output goods markets
  final demand)
• Geography and technology development the
  conceptual basis
• - New economic geography
• - Endogenizing technological change in
  endogenous economic growth models (Romer 1986,
  1990, Lucas 1986, Aghion and Howitt 1998)
• - The geography of knowledge spillovers (Jaffe,
  Trajtenberg and Henderson 1993, Audretsch and
  Feldman 1996, Anselin, Varga and Acs 1997)

26
EcoRET The modeling framework

• Structure of EcoRET four blocks
• The supply side block (labor market, production,
  productivity, investment, employment and
  unemployment, production costs, inflation)
•  
• The demand side block (behavioral relationship of
  private households, consumption, and other
  components of final demand (government
  consumption, foreign trade etc.) in real and
  nominal terms and their deflators)
•  
• The income distribution block (determining
  private and government income - labor and
  property income, profits - and the transfers of
  income between private households and the
  government - taxes, social security and other
  transfers)
•  
• The Total Factor Productivity (TFP) block
  (modeling changes in regional level TFP as a
  function of certain knowledge-related variables
  as well as CSF measures such as promotion of
  physical infrastructure and human capital)
• EcoRET consists of 106 variables, 32 of them are
  explained by behavioral or technical
  relationships, 16 variables are exogenous while
  the remainder of the endogenous variables is
  explained by definitional identities

27
EcoRET Data and estimation

• Various Hungarian (Hungarian Central Statistical
  Office, Hungarian Patent Office) and
  international (OECD, IMF) data sources
• For the period of 1990 - 2000
• Units of observation
• - country (macromodel)
• - counties (technology model)
• Parameters
• - estimation/calibration (macromodel)
• - pooled estimation (technology model)

28
EcoRET The regional TFP block

• The estimated regional model of technological
  change
• TFPGR a0 a1KNAT a2RD a3 KIMP a4INFRAINV
  a5HUMCAPINV e,
• TFPGR the annual rate of growth of Total Factor
  Productivity (TFP),
• KNAT domestically available technological
  knowledge accessible with no geographical
  restrictions (measured by stock of patents),
• RD private and public regional RD,
• KIMP imported technologies (measured by FDI),
• INFRAINV investment in physical infrastructure,
• HUMCAPINV investment in human capital,
• region i and time t
• a1 estimates domestic knowledge spillover effects
• a2 estimates localized (regional) knowledge
  spillover effects
• a3 estimates international knowledge spillover
  effects

29
EcoRET Linking the TFP block to the rest of
EcoRET in policy simulations

• Problem
• - Macro blocks time series estimation
• - TFP block time-space data
• Literature agglomeration and technological
  change (Feldman 1994, Fujita and Thisse 2002,
  Varga 2000)
• Solution weighted averaged county TFP growth
  rates (Excellent historical forecast of national
  level TFP!)
• The linkage
• TFP TFP-1e?eDNTFPGR

30
EcoRET Simulated effect of the geography of CSF
support on the national growth rate

• The ratios of the growth effects of concentrating
  CSF resources in
• leading areas (LEAD/LAG)
• lagging areas (LAG/EQUAL)
• equal distribution (LEAD/EQUAL)

31
III. Endogenizing spatial structure

• Q How to endogenize and integrate equation (3),
  the RD location equation, i.e., the long run
  spatial effects?
• A promising solution is to integrate Spatial
  Computable Equilibrium (SCGE) models (to
  endogenize RD distribution) with
  macroeconometric models to simulate the
  macroeconomic growth effects.

32
Summary

• An empirical modeling framework is presented
• Methodological reasons for a relative negligence
  of the spatial aspects of macroeconomic growth
  are reviewed
• Challenges in spatial data analysis
• Difficulties in integrating regional and macro
  levels
• Complications in endogenizing spatial structure
  in empirical macroeconomic growth models