Infrastructure

Infrastructure investment
and growth
Luis Servén
The World Bank
IMF, November 2010
Background
How important is infrastructure for economic growth?
•
•
Old question – even in Adam Smith’s Wealth of Nations
Empirically revived after Aschauer (1989) – who found
huge rates of return on public capital in the U.S.
For policy-making, the key link is that between
infrastructure spending and growth.
What do we know about it empirically?
2
Background
Two distinct logical steps involved:
(1) From spending to infrastructure services – the cost of
acquiring (and operating) infrastructure assets
(2) From infrastructure assets to growth – the productivity
of infrastructure assets (or their services)
The bulk of empirical macro research has focused on (2).
But for optimal spending / provision decisions we need to
know more about (1) too.
3
Infrastructure and growth
Start with (2): from assets to output / growth
Two common empirical approaches:
– Growth regressions augmented with infrastructure measures
– Infrastructure as another input in aggregate production function
(or its dual, the cost function)
On balance, majority of studies – especially recent ones on
developing economies – find significant positive effects
Some methodological caveats with many studies – reverse
causality, heterogeneity, non-stationarity...
A key issue is the measurement of infrastructure assets
– Physical indicators (e.g., road km)
– Monetary indicators (investment flows or their cumulative totals)
4
Infrastructure and growth
Measures of infrastructure spending can be very poor
proxies for the quantity / quality of assets
•
Poor project selection; procurement; corruption (Tanzi-Davoodi
1997; Pritchett 2000; Keefer-Knack 2007) – more on this later
Empirical studies, by reported finding (Straub 2007)
Negative
None
Positive
5
4.1.1 Infrastructure Stock and Economic Growth
7
CHN
6
TWN
BWA KOR
Growth in real GDP per capita (%)
5
THA
MYS
4
LKA
-4
MLT
HKG
CYP
OMN
SGP
IRL ARE
JPN
ROM
PRT
HUNESP
TUN
GRC
LUX
IDN
3
IND
GHA
NOR
PAN
PAK
ISL
DOM
RUS
EGY
AUT
TTO
POL
ITA
ISR
BEL
FIN
MAR
LVA
FRA
CZE
BLR
USA
BRA
TUR
SVN
DEU
CAN
CHL
AUS
GBR
DNK
NLD
SYR
SWE
2
COL
MEX SAU
BHR
IRN
SVK
MRT
EST
ETH
TZA
PHLPRY
ECU ZAFCRI
CHE
NZL
NPL
URY
BGRLBY
BGD MWI
SLV
1DZA
GTM
PERARG
COG
CIV
BFA
CMR
BEN
GAB
RWA
GMB HND YEM
JAM LTU
UGA
BOL
VEN
SDN
GNB
KAZ
NGA
BDI
KEN ZMB 0
TCDTGO
JOR
GIN
HRV
ZWE
HTI
SEN
-3
-2CAF SLE
-1 AGO NIC 0
1
2
3
4
KGZ
NER
-1
MDG
QAT
YSR
UKR
-2
ZAR
y = 0.4812x + 1.6724
IRQ
-3
R2 = 0.1701
-4
Aggregate Index of Infrastructure Stocks, IK1 (in logs)
6
4.1.2 Infrastructure Quality and Economic Growth
7
Growth in real GDP per capita (%)
BWA6
HTI
-4
-3
CHN
KOR
TWN
MLT
HKG
5
CYP
OMN
THA
MYS
SGP
4
JPN IRLARE
ROM
LKA
PRT
HUN
GRCESP NOR LUX
TUNIDN
IND
3
GHA
PAN
PAK
ISL
DOMTTO
EGY
RUS
AUT
POL
ISR
ITAFRA
FIN
BEL
MAR
LVA
CZE
BLR
BRA CHL
USA
TUR
SVN
DEU
CAN
AUS
GBR
NLD
SWE DNK
SYRCOL
2
MEX
SAU
BHR
IRN
MRT
EST
ETH
TZA
PHL
CRI
ZAF
CHE
ECU
PRY NZL SVK
NPL MWI
URY
DZA
BGR
BGD
LBY
SLV
1
GTM
COG
PERARG
CIV
CMR
BFA
BEN
HND
GAB
YEM
RWA
GMB
JAM
UGA
LTU
BOL
VEN KAZ
SDNBDI
GNB
NGA
ZMB
KENTCD
0
GIN
HRV JOR
TGO
ZWE
SEN
CAF
-2 NIC SLE
-1AGO
0
1
2
KGZ
NER
-1
MDG
QAT
YSRUKR
-2
ZAR
y = 0.6312x + 1.8891
IRQ
-3
R2 = 0.2054
3
-4
Aggregate Index of Infrastructure Quality, IQ (in logs)
7
Infrastructure and growth
Example of growth approach: Calderón-Servén 2009
Empirical growth framework with physical measures of infrastructure
Synthetic index of telecom, transport and power assets – plus
(noisy!) measures of quality of assets
Large panel dataset
Results:
– Infrastructure quantity and quality have robust growth effect –
and economically significant.
– Not much evidence of heterogeneity (in log-log terms)
• Across developing regions
• Landlocked vs other countries
• Related to infrastructure endowment (i.e., non-linearities)
Hence the marginal contribution of infrastructure development to
growth is higher wherever quantity / quality are lower.
8
Figure 2
Growth changes across regions due to infrastructure development
(2001-5 vs.1991-5 averages)
3.0
Changes in growth per capita (%)
2.5
2.0
1.5
1.0
0.5
0.0
-0.5
-1.0
World
Western
Europe
Infrastructure capital -- Brookings 2010
East Asia
Infrastructure Stocks
South Asia
Infrastructure Quality
Middle East & Sub-Saharan
North Africa
Africa
9
Infrastructure and growth
How does the growth contribution of infrastructure
development vary across countries? Closer look
Calderón, Moral and Servén (2010): Cobb-Douglas production function
approach -- physical and human capital; infrastructure
•
Focus on the contribution of infrastructure to labor productivity
(GDP per worker)
•
synthetic infrastructure index (as before)
•
88 industrial and developing countries,1960-2000 (> 3,500 obs)
•
Allow for heterogeneity of infrastructure contribution – both
generic and along specific dimensions
–
–
Empirical framework allows intercepts, error variances and short-run
dynamics to differ freely across countries.
Imposes (testable) cross-country homogeneity of long-run
coefficients.
10
Infrastructure and growth
Main results
•
•
Infrastructure elasticity in range .07 to .10 – and robust.
Elasticities of other inputs (physical and human capital) in line
with literature (around 0.35, 0.10 respectively)
No evidence of (general) cross-country parameter heterogeneity
•
–
•
Accords with cross-country stability of factor shares (Gollin 2002)
But country-specific estimates are noisy (especially in LICs), so
tests may have low power. Test for specific forms of heterogeneity
–
–
–
–
By income level: infrastructure elasticity could differ in rich and poor
countries
By level of infrastructure endowment: nonlinear effects of
infrastructure (network effects?)
By population size: economies of scale / congestion effects.
By quality of policy framework: high / low distortions
Only this test comes close to 10% significance
11
5
Long-run coefficient
KEN
NGA
BEN LSO
SEN
NPL
GHA
GMB
MLI
UGA
ETH
MWI
MOZ
MDG
BFA
0
TTO
TUN
SGP
ROM
JPN
COL
CMR
PER
BRA
JOR
ESP
NZL
URY KOR
CHN IDN THA EGY DZA
IND
CHL
PRT ISR
CHE
CAN
FRA USA
SYR
ARGGRC
DNK
CIV
GTM
ZAF MYS
MUS
DOM MEX
CPV
SWE
ISL
TUR
BOL PHL
AUS
GINZWE
MAR
PRY PAN
NLDBLX
GBR
FIN
ITAIRL
LKA
NOR
ECU
IRN
HND
PAK
NIC
RWA TGO
VENCRI
JAM
SLV
TTO
SGP
ROM
JPN
COL
CMR
PER
KEN
JORBRA
KOR
ESP
NGA
EGY
NZL
URY
CHN
IDN THA
LSO
GRC
BEN
DZA MEX
MYS
CHL
PRT
CHE
CAN
FRA SWE
MDG
SYR
ARG
DNK
CIV IND
GTM
ZAF
ISR
DOMMUS
ZWE
BFA NPL
TUR
BOL
AUSISL
GIN SEN
PHL
MARCPV
BLXUSA
PRY
PAN
NLD
GBR
FIN
GHA
ITA
GMB LKA
IRL
MLI MWI
ETHUGA
NOR
MOZ
ECU
IRN
HND
PAK
RWA
0
5
SLV
TUN
VEN CRI
AUT
GAB
NIC
TGO
GAB
AUT
JAM
TZA
-5
-5
TZA
NER
NER
7
8
9
10
Logarithm of GDP per worker
-8
11
0
ISL CPV
PAK
TUN
NIC
SGP
ROM COL
TGO RWA
JPN
CMR PER
KEN
BRA
JOR
ESP
KOR
EGY NGAIDN
NZL
THA
LSOURY
GRC
BEN
DZA
CHL
MYS
PRT
MEX
CHE
CAN
FRA
MDG
SYR
ARG
DNK
CIV
GTM
ISR
ZAF
MUS
DOM
USA
SWE
SEN
ZWE
BFA
NPL
TUR
BOL
AUS
GIN
MAR
PHL
BLX
PRY
PAN FIN
NLD
GBR
GHA
ITA
GMB
IRL
MLI
UGA
ETH
MWI LKA
NOR
MOZ
IRN
HND ECU
VEN
CRI
GAB JAM
SLV
CHN
IND
TTO
PAK TUN
NIC
SGP
RWA JPN
ROM
TGO
COL
PER CMR
KEN
BRA JOR
KOR
ESP
NGA
EGY
NZL
URY
CHN
IDN
THA
LSO
GRC
IND
BEN
DZA
CHL
MYSPRT
MEX
CHE
CAN
FRA
ARG
MDGSEN SYR
DNK
GTM
ISR
ZAF
DOM
USA
ZWE
CPV
SWE
BFA
ISL
NPL
TUR
BOL
AUS
GINMUS
PHL
MAR
BLX
PRY
PAN
NLD
GBR
FIN
GHA
ITAMLI
GMB
IRL
UGA
LKA
MWI ETH MOZ
ECU HND NOR
IRN
0
TTO
Long-run coefficient
5
SLV
-2
Output elasticity of infrastructure vs. Infrastructure stock
5
Output elasticity of infrastructure vs. per capita GDP
-6
-4
Infrastructure index per worker
VENCRI
AUT JAM
AUT
CIV
GAB
TZA
-5
-5
TZA
NER
NER
12
14
16
18
Logarithm of population
20
22
Output elasticity of infrastructure vs. population
0
100
200
Price of investment
300
400
12
Output elasticity of infrastructure vs. distortion index
Infrastructure and growth
Further homogeneity tests
Break sample into high / low along the relevant dimension – and test for
equality of mean infrastructure elasticity.
Per Capita Per Capita Infrastructure
Total
Distortions
Income (A) Income (B) Endowment Population
High
Low
p-value
0.054
0.059
0.985
0.044
0.062
0.94
0.059
0.055
0.988
-0.016
0.131
0.576
-0.156
0.271
0.102
Overall, no strong evidence that the elasticity varies across countries.
The implication is that the marginal product of infrastructure declines as
the infrastructure / output ratio rises.
13
From spending to assets
•
•
All this is only about the benefit side of infrastructure –
what about the cost?
Cost-benefit comparison is needed to assess the extent
of under-provision of infrastructure – and whether it is
greater than that of other inputs (e.g., human capital)
– Limited evidence on this (e.g., Canning and Pedroni 2008: no
generalized infrastructure shortage across countries / sectors)
– Calderón and Servén (2010): big growth impact of infrastructure
catch-up in Africa – but massive cost involved: 10% to 15% of
GDP for (half) catching-up, even ignoring O&M.
– Loayza (2010): big growth impact in Egypt – but only if much of
the cost is financed by spending cuts elsewhere
14
Sub-Saharan Africa: cost of infrastructure catch-up
15
From spending to assets
Is spending a good proxy for infrastructure development?
– Spending is often very inefficient: bad projects, waste…
– Weak link between spending and assets / services (Pritchett
2000): it is mediated by government technical capacity, fiscal
institutions, budgetary practices, governance...
– Big scope for corruption and political clientelism
• Tanzi and Davoodi 1997; Keefer-Knack 2007: weak governance
and corruption raise measured ‘public investment’ – in reality, much
is rent extraction rather than asset acquisition.
• Incentives for investment over O&M: bigger room for corruption,
political favors and photo-ops – so build new roads rather than
maintain old ones.
16
From spending to assets
Better understanding of costs of asset acquisition, and their
determinants, is needed for policy decisions.
– Without it, a ‘big push’ to infrastructure may lead to massive
waste
But this requires much more complete and detailed data on
infrastructure spending and performance
– Often we cannot even establish the facts – especially on O&M,
but in many countries also on sector-wise investment spending
– Bad data also hampers accountability / transparency
Poor data arguably is one of the biggest obstacles to better
diagnostic of the problems and design of solutions
17
From spending to assets
New database under construction – investment by
infrastructure sector in developing countries
• Builds on earlier data collection for Latin America
(Calderón and Servén 2004, 2010)
• Disaggregation into power, water/sanitation, roads, rail,
telecom – when feasible
• Disaggregation into public and private investment
• Time coverage from 1980
Preview of work in progress.
Infrastructure investment and governance
4
PHL
2
IDN
TUR
6
TUR
JOR
JOR
BOL
IND
CHL
PHL
IDN
IND
CHL
COL
COL
BRA
ARG
BRA
ARG
PER
MEX
MEX
0
0
PER
EGY
4
BOL
EGY
2
6
Cumulative investment (% of GDP) ('85-)
8
8
(cumulative investment, percent of 1985 GDP)
0
.2
.4
.6
Bureaucracy (Ave. of the 1980s)
.8
Infrastructure investment vs quality of bureaucracy
0
.2
.4
Corruption (Ave. of the 1980s)
.6
Infrastructure investment vs control of corruption
• Weaker governance is associated with higher investment / GDP ratios (like in Tanzi-Davoodi
1997 and Keefer-Knack 2007)
.8
From spending to assets
How big is the cost impact of governance?
Look at the average unit cost of infrastructure assets
T
c
I
t 0
t
KT  K 0
Note that c should be expected to vary across
infrastructure sectors.
Construct sector-specific c – with K respectively given by
– road km
– power generation capacity (Gw)
– telephone lines
From spending to assets
Across countries, c may vary with per capita GDP (wages,
RER) and other factors
• Add also climatic / geographic / demographic controls in regressions
Some data limitations:
• No information on O&M spending
• No information on asset quality (except % roads paved)
• Small country sample (only 17 countries for the moment…)
No retrospective info on fiscal / budgetary institutions, so
look instead at the correlation of unit cost with broader
governance-related variables:
• (Control of) corruption [higher is better]
• Quality of bureaucracy [project selection, procurement…]
1
1
Unit cost of infrastructure assets and control of
corruption (controlling for GDP per capita)
PHL
Unit costs for ml ('85-)
.5
.5
IDN
ARG
EGY
PER
Power
TUR
MNG
MEX
IDN
COL
Phone
lines
IND
VNM
IND
-.5
PAK
ARG BRA
0
ZAF
THA
-.5
0
BOL
PER
JOR
BRA
JOR
CHL
COL
PHL
CHL
BOL
-1
-1
EGY
MEX
-.2
0
Corruption (Ave. of the 1980s)
.2
.4
-.4
3
-.4
TUR
-.2
0
Corruption (Ave. of the 1980s)
EGY
IDN
1
JOR
TUR
Roads
0
PHL
ARG
IND
-1
BOL
MEX
PER
COL
BRA
-2
Unit costs for roads ('85-)
2
CHL
MNG
-.4
-.2
0
Corruption (Ave. of the 1980s)
.2
.4
.2
.4
1
1
Unit cost of infrastructure assets and quality of
bureaucracy (controlling for GDP per capita)
PHL
.5
EGY
BOL
PER
VNM
TUR
Power
ZAF
THA
JOR
BRA
ARG
MEX MNG
IDN
COL
IND
IND
-.5
PAK
PER
0
COL
CHL
BOL
Phone
lines
-.5
0
BRA
JOR
CHL
Unit costs for ml ('85-)
.5
ARG
PHL
-1
EGY
-1
TUR
MEX
-.2
0
.2
Bureaucracy (Ave. of the 1980s)
-.4
.4
3
-.4
-.2
0
.2
Bureaucracy (Ave. of the 1980s)
2
CHL
EGY
IDN
1
JOR
TUR
0
ARG
IND
BOL
PER
-1
Roads
PHL
MEX
COL
BRA
-2
Unit costs for roads ('85-)
Unit costs for EGC ('85-)
IDN
MNG
-.4
-.2
0
.2
Bureaucracy (Ave. of the 1980s)
.4
.4
From spending to assets
Weaker governance is associated with higher asset costs
in all sectors examined.
Corruption and low bureaucratic quality drive a wedge
between investment spending and actual infrastructure
development.
Regression estimates suggest that the magnitude of the
wedge is considerable
– 1 sd improvement in corruption raises by 25% the volume of
assets acquired with a given amount of investment
– 1 sd improvement in bureaucratic quality raises it by 18%
Conclusions
Extensive research on the growth benefits of infrastructure
– Overall, solid evidence that infrastructure quantity and quality
help growth
– Also evidence of a positive impact on equity – hence scope for a
double poverty-reducing effect
– Effects are bigger where infrastructure endowments are lower –
so potentially a big payoff from infrastructure catchup for LICs
But much less attention has been paid to the costs
– More analysis of returns vs costs of infrastructure assets is
needed – with more focus on specific sectors in specific
countries
– Poor data on spending (especially O&M) is one of the biggest
obstacles to better understanding of costs
25
Conclusions
Poor governance drives a wedge between infrastructure
spending and actual infrastructure development
– Low government capacity, weak fiscal governance and
corruption inflate asset costs, discourage O&M, and reduce the
efficiency of spending
– This has big effects on the cost of infrastructure development
Raising the efficiency of spending is a key priority
– Research needs to identify institutional mechanisms and
incentives that favor sound spending decisions
Developing stronger project selection and evaluation capacity
Assessing if / how different budgetary institutions, checks and
balances can help
– Otherwise, more spending may yield a lot of waste and not much
more infrastructure development.
26
End