Transcript Integrating the geography of innovation to policy modeling

Integrating the geography of innovation to
policy modeling
by
Attila Varga
Department of Economics and Regional Studies
and
Center for Research in Economic Policy (GKK)
Faculty of Business and Economics
University of Pécs, Hungary
III. Integrating agglomeration effects to
development policy modeling
• Knowledge-based development policies (R&D promotion,
infrastructure investments, education support etc.)
• Modeling the effect of geography on policy effectiveness - three
steps:
1. modeling static agglomeration effects generated by the spatial
distribution of the instruments
2. modeling dynamic agglomeration effects of policy intervention:
“cumulative causation” – induced technological change
3. modeling the resulting macroeconomic effects
• In most of the current policy analysis models: no geography
incorporated
III. A key issue in development policy modelling: integrating
the spatial dimension of technological change
• The GMR Hungary model:
- integrates all the above three aspects
- developed for ex-ante CSF intervention analysis for the Hungarian
government (planning period 2007-13)
- result of on international collaboration with German, Dutch and
Japanese institutes
- both macro and regional aspects are estimated
IV. Outline of the GMR model
• CSF instruments targeting technology
development:
– Infrastructure investments
– Education/training support
– R&D promotion
IV. Outline of the GMR model
IV. Outline of the GMR model
• GMR consists of three sub-models:
- the TFP sub-model (static agglomeration
effects)
- the spatial computable general equilibrium
(SCGE) sub-model (dynamic agglomeartion
effects)
- a complete macroeconomic model (the effects
of geography on macroeconomic variables)
The function of the TFP sub-model
• To generate STATIC TFP changes as a
result of CSF interventions (direct shortrun CSF-effect)
• NOT for forecasting but for impact analysis
Main characteristics of the TFP
sub-model
• TFP equation:
- estimates the effects of geographically
differently located knowledge sources
(local, national, international)
- estimates the effects of CSF-instruments
(infra, edu)
• Time-space data
The TFP equation
The estimated regional model of technological change
TFPGR = α0 + α1KNAT + α2RD+ α3 KIMP + α4INFRAINV + α5HUMCAPINV + ε,
TFPGR: the annual rate of growth of Total Factor Productivity (TFP),
KNAT: domestically available technological knowledge accessible
geographical restrictions (measured by stock of patents),
RD: private and public regional R&D,
KIMP: imported technologies (measured by FDI),
INFRAINV: investment in physical infrastructure,
HUMCAPINV: investment in human capital,
region i and time t
α1 estimates domestic knowledge effects
α2 estimates localized (regional) knowledge effects
α3 estimates international knowledge effects
with no
Table 1: Pooled FGLS estimation results for TFP growth rates (TFPGR) and for 20
Hungarian counties, 1996 – 2003
Final
C
Model 1
Model 2
Model 3
Model 4
Model 5
Model 6
Model
-2.5434
-2.4740
-2.4797
-2.4965
-2.2423
-1.8243
-1.0389
(0.2989)
(0.2910)
(0.2919)
(0.2735)
(0.2728)
(0.2372)
(0.3408)
TFPGR(-2)
-0.2587
(0.0749)
KNAT (-2)
0.0002
0.0002
0.0002
(2.68E-05) (2.59E-05) (2.60E-05)
KIMP (-3)
0.0002
0.0002
0.0002
8.84E-5
(2.45E-05) (2.44E-05) (2.10E-05) (3.04E-05)
0.1582
0.1526
0.1455
0.0892
0.1219
0.0826
(0.0449)
(0.0456)
(0.043)
(0.0430)
(0.0393)
(0.0392)
3.79E-06
1.46E-06
1.56E-06
2.11E-06
RD (-2)
1.29E-06
(1.77E-06)
d(INFRA(-1))
(9.60E-07) (1.34E-06) (9.41E-07) (8.44E-07)
d(HUMRES(-2))
6.95E-06
4.74E-06
5.63E-06
(2.84E-06) (2.47E-06) (2.41E-06)
DUM99
-0.0601
-0.0610
(0.0081)
(0.0080)
Weighted Statistics
R2-adj
0.31
0.37
0.37
0.42
0.42
0.59
0.62
54.02
35.71
23.83
31.15
18.44
29.27
28.36
Prob (F-statistic)
0.00
0.00
0.00
0.00
0.00
0.00
0.00
Durbin-Watson stat
1.90
2.06
2.07
2.02
1.68
2.22
2.42
N
120
120
120
120
100
100
100
Unweighted Statistics
R2-adj
0.14
0.19
0.20
0.21
0.23
0.35
0.42
21.3***
16.18***
18.55***
14.79***
1.25
21.3***
19.23***
20.64***
18.12***
3.78*
F-statistic
ML Spatial error
Neighb
ML Spatial lag
Neighb
Note: estimated standard errors are in parentheses; Neighb is first order neighborhood standardized weights
matrix; *** is significance at 0.01, ** is significance at 0.05, * is significance at 0.1.
1,0400
1,0200
1,0000
0,9800
0,9600
0,9400
0,9200
0,9000
0,8800
0,8600
1999
2000
TFP level as in GMR observ
2001
2002
TFP level as in GMR forecasted
Figure 1: Observed and predicted levels of national TFP
2003
The function of the SCGE submodel
• To generate DYNAMIC TFP changes that
incorporate the effects of agglomeration
externalities on labor-capital migration (induced
long-run CSF effect)
• Agglomeration effects depends on:
- centripetal forces: local knowledge (TFP)
- centrifugal forces: transport cost, congestion
• To calculate the spatial distribution of L, I, Y, w
by sectors for the period of simulation
The SCGE sub-model
• Adaptation of RAEM-Light (Koike, Thissen
2005)
• C-D production function, cost
minimization, utility maximization,
interregional trade, migration
• Equilibrium:
- short run (regional equilibrium)
- long run (interregional equilibrium)
Main characteristics of the SCGE
sub-model
• NOT for historical forecasting
• The aim: to study the spatial effects of
shocks (CSF intervention)
• Without interventions: it represents full
spatial equilibrium - regional and
interregional (no migration)
• Shock: interrupts the state of equilibrium,
the model describes the gradual process
towards full spatial equilibrium
The function of the MACRO submodel
• Based on dynamic TFP values: the
resulting effects on macro variables
The characteristics of the
MACRO sub-model
• Complete macro model (supply, demand, income
distribution) – the EcoRET model (Schalk, Varga
2004)
• C-D production technology, cost minimization
• Supply and demand side effects of CSF
• A-spatial model
• Describes the effects of exogenous technological
change
• Baseline: TFP growth without CSF interventions
• Policy simulations: describe the effects of CSFinduced TFP changes on macro variables
Regional and national level short run and long run
effects of TFP changes induced by TFP-related CSF
interventions
1. Intervention in any region increases regional TFP level in the mth sector
(static agglomeration effect)
2. Short run effect:
- price of the good decreases
- decreasing demand for both L and K (assuming output
unchanged)
- increasing regional and interregional demand for the good that
increases demand for L and K
- increased regional demand increases utility levels of consumers
in the region
3. Long run effects: increasing utility levels induces labor migration into the
region followed by capital migration
- resulting in a further increase in TFP (dynamic agglomeration
effect)
- and finally a changed spatial economic structure
4. Macroeconomic variables reflect the long run equilibrium TFP level
Regional and national level short run and long run effects of TFP
changes induced by TFP-related CSF interventions
Effects on spatial
economic structure
Macroeconomic effects
7
6
SCGE
sub-model
(regional model)
MACRO
sub-model (demand,
supply, income
distribution)
5
Long run effects
Short run effects
TFP
sub-model
(regional model)
4
3
2
1
Economic policy instruments:
infrastructure, R&D and
education
Does geography matter in public policy?
Allocation of CSF support in Mill. 1995 HUF
400 000
Expenditures in Mill. HUF
350 000
300 000
250 000
200 000
150 000
100 000
50 000
0
2007
2008
2009
2010
2011
2012
2013
2014
2015
Year
Infrastructure
Education
R&D
Investment
Demand side only
Core-periphery structure of Hungarian counties with
respect to Gross Value Added per employee
Co re-p eriiphe ry s tru cture of H ung ary
Co re
Perip hery
The effects of policy scenarios
on the GDP growth rate
2,00
1,50
1,00
0,50
0,00
2007
2008
2009
2010
2011
2012
2013
2014
-0,50
Core
Periphery
Equal
2015
2016
2017
The policy effects on convergence measured by
standard deviation of regional value added
2,50
2,00
1,50
1,00
0,50
0,00
2007
2008
2009
2010
2011
Core
2012
2013
Periphery
2014
Equal
2015
2016
2017
Measuring the cost of growth promotion
εc,g = [(σRGVA, scen - σRGVA, bline)/ σRGVA, bline]/[(GDP scen - GDP bline)/ GDPbline];
where
εc,g is the elasticity of the change in the standard deviation of regional GVA relative to the
baseline with respect to the change in GDP relative to the baseline,
σRGVA, scen and σRGVA,
baseline,
bline
are standard deviations of regional GVA in the scenario and the
GDP scen and GDP bline are GDP at the national level in the scenario and the baseline.
Elasticity of the standard deviation
of regional GVA with respect to GDP (relative to baseline)
0,120
0,100
0,080
0,060
0,040
0,020
0,000
2007
2008
2009
2010
2011
Core
2012
2013
Periphery
2014
Equal
2015
2016
2017
Conluding remarks
• Growth and the geography of innovation:
theoretical versus empirical integration
• Geographic effects in policy modelling: the
GMR model
• Results show that agglomeration effects are
important factors in macroeconomic
performance and neglecting them in
development policy analyses could result in
misleading expectations as to how a
particular mixture of policies affect the
economy.