Transcript Expenditure flows and asset stocks

Fiscal Policy, Infrastructure Expenditure and
Growth: Sharing experience from Latin America
Luis Serven,
Research Manager, DEC
Workshop: Modeling Fiscal Policy, Public Expenditure
and Growth linkages, June 14-15, 2006
Background


Work focused on Latin America
Motivation: perception that fiscal adjustment has led to a
steep decline in productive public expenditure –
specifically infrastructure.
 Private sector entry insufficient to offset the decline,
hence potentially adverse growth effects.
 Concern with myopic fiscal rules targeting liquidity and
ignoring intertemporal dimension of solvency
 Research program including
 Analytical papers
 Empirical cross-country work
 Country studies (Brazil, Colombia…)
Selective summary
1.
2.
3.
4.
Assembling infrastructure spending data
From expenditure flows to asset stocks
The output contribution of infrastructure
Country simulation models
Infrastructure expenditure


Very limited data availability: only Central Gov
expenditure from GFS – unreliable because of (i)
decentralization and (ii) PEs.
Collect data from national sources for 7
countries (ARG, BRA, BOL, CHL, COL, MEX,
PER), 1980-2000.



4 different sectors: roads, railways (if applicable),
power, telecommunications.
public and private investment
No luck with O&M
Infrastructure expenditure
 A casualty of fiscal adjustment
Latin America
Infrastructure expenditure
Brazil
Primary deficit and public infrastructure investment
(% GDP)
2.00%
3.50%
1.00%
3.00%
0.00%
1995
1996
1997
1998
1999
2000
2001
2002
2003
2.50%
-1.00%
2.00%
-2.00%
1.50%
-3.00%
-4.00%
Primary deficit (left scale)
1.00%
Infrastructure investment (right scale)
-5.00%
0.50%
Infrastructure expenditure
Latin America
Total (public +private) investment in Infrastructure
(weighted average of 7 countries, percent of GDP)
Expenditure flows and asset stocks
• What matters is the availability / quality of services, not
(necessarily) the volume of expenditure.
• Their trends may be very different -- due to changing
efficiency, corruption, waste…(Pritchett, Tanzi etc)
Expenditure flows and asset stocks
Brazil: the power sector
Investment
Capacity change
Expenditure flows and asset stocks
• Easterly and Servén (2003): regression approach
relating the trajectory of infrastructure asset stocks to
investment flows.
• Simple-minded ARDL models (to allow time-to-build)
A( L) ln Kt  B( L)( I t / Kt 1 )   t
• Kt = {power generation capacity in Gw; Km of roads
(+railways); # of phone lines}
• It = total investment (also private and public separately)
• Panel data for 8 countries / 20 years; fixed effects (+time
effects)
Expenditure flows and asset stocks
• Results: can explain very well telecom (R2 = .7-.8), fairly
well roads / rail (R2 = .4-.6), power not so well (R2 = .2-.3).
• For roads, some evidence that private investment makes
a bigger contribution to stocks (= lower unit cost)
• Other regressions relating measures of service quality
(e.g., phone faults, power losses…) to private sector
participation – with mixed results: positive for telecom,
negative for power…
• Heterogeneity of assets potentially a big problem here
(e.g., private sector only does “easy” roads)
Expenditure flows and asset stocks
• Ferreira (2005) does similar exercises for Brazil. He gets
very close tracking for telecom, fair for power, poor for
roads.
• Unresolved major question: how does the stock / flow
link depend on governance and fiscal institutions – or
other country features ?
One way to address this would be through countryspecific (rather than pooled) estimates -- e.g., a RCM.
But that requires broader cross-country data
coverage...
The output contribution
Standard Cobb-Douglas specification imposing CRS:
yit  lit  ai  bt   (kit  lit )   (hit  lit )   ( zit  lit )   it
Panel data for 90+ countries, 1960s-1990s
h = years of secondary schooling (other measures also)
z = physical infrastructure measures (km of roads, phone
lines, power generation capacity Gw)
Endogeneity a major problem. 2 empirical approaches:
-- GMM (large N asymptotics: Arellano-Bond etc) with
internal instruments as well as demographic instruments
-- Panel time-series approach (large N and T: Mark-Sul,
Philips-Moon) w/ unrestricted cross-country heterogeneity
The output contribution
Table 3.4
Alternative GMM estimates
(Dependent variable: log GDP per worker)
Model specification
Instruments
Physical capital
Secondary schooling
Electricity generating capacity
Roads
Main phone lines
Wald test of joint significance (p-value)
Sargan test (p-value)
1st-order autocorrelation (p-value)
2nd-order autocorrelation (p-value)
Number of observations
Number of countries
1
2
3
4
Differences
Levels t-2
Differences
Levels t-3
Differences
Demographics
System
Levels +diffs
0.363
0.361
0.351
0.222
(10.832)
(11.034)
(7.903)
(7.867)
0.148
0.169
0.159
0.222
(3.361)
(3.815)
(3.443)
(5.520)
0.112
0.123
0.177
0.109
(1.809)
(2.148)
(2.468)
(2.970)
0.119
0.117
0.105
-0.005
(2.197)
(2.'195)
(1.241)
(0.084)
0.151
0.140
0.138
0.147
(3.634)
(3.236)
(3.168)
(6.164)
0.000
0.319
0.111
0.793
3232
101
0.000
0.312
0.106
0.794
3232
101
0.000
0.141
0.143
0.888
3232
101
0.000
0.002
0.555
0.778
3232
101
Note: All variables are measured per worker and (except for schooling) expressed in logs. Heteroskedasticity-consistent T-statistics in brackets.
The output contribution
Table 4. Infrastructure-augmented production function
Panel DOLS estimates controlling for common factors
Lag selection subject to a maximum of 1 lead and 1 lag.
Lag selection criterion
Physical capital
Years of Secondary Schooling
Electricity Generation Capacity
Roads
Main Telephone Lines
Cross-section independence test (p-value)
Parameter homogeneity test (p-value)
Observations / countries
Imposed
(1,1)
0.299
0.025
0.036
0.019
0.082
0.018
0.043
0.025
0.028
0.015
SBC
**
*
**
*
*
0.282
0.023
0.051
0.020
0.066
0.019
0.047
0.024
0.049
0.016
PIC
**
**
**
**
**
0.276
0.023
0.054
0.019
0.074
0.018
0.060
0.022
0.046
0.015
0.010
0.000
0.385
0.429
0.240
0.365
3382 / 89
3382 / 89
3382 / 89
**
**
**
**
**
Notes:
The dependent variable is real GDP. All variables are expressed in logs and relative to the labor force.
Standard errors under each coefficient computed using the QSPW method of Andrews and Monahan (1992)
(**) significant at the 5% level; (*) significant at the 10% level
The output contribution
Significant contributions of infrastructure in both cases – but
smaller coefficients (as much as 50%) in the panel timeseries approach.
Tests of country-specific vs pooled estimates do not show
much evidence of cross-country parameter heterogeneity –
in general, null of homogeneity cannot be rejected…
Simulation models
Focus on assessing alternative fiscal strategies / fiscal rules
2 applications: Brazil & Colombia
•
Brazil (Ferreira and Gonçalves 2005): main concern is
the interplay of fiscal rules, investment and growth.
•
Colombia (Suescún 2005): upcoming fiscal correction to
accommodate pension deficit.
Models of intertemporally-optimizing infinitely-lived agents,
with endogenous labor supply.
…but model details differ.
Brazil
model
Brazil model
Questions:
(1) Would it make sense to raise public infrastructure
investment ?
(2) If so, what form of financing should be used ? Cuts in
other expenditure, taxes or debt ?
Main conclusion from simulations: the best strategy for
growth and welfare is an investment increase financed by
reducing public consumption.
Colombia model
•
•
•
•
•
•
Three reproducible factors: infrastructure capital,
business capital and human capital
Infrastructure capital can be supplied by the government
or the private sector (imperfect substitutes)
Market-determined user fees on infrastructure services
(infrastructure expansion may cost revenues ! [Uruguay])
Public consumption does not yield utility
Learning-by-doing human capital externality
Solvency assured by tax rates that depend on public debt
Colombia model
Question: should the adjustment be financed via tax hikes or
infrastructure investment cuts ?
Simulations show investment cuts yield lower growth in long
run (about 1% less per annum ) – i.e., the cost of lower
public capital outweighs the gain from lower real interest
rates allowed by reduced indebtedness.
The less easily substitutable public infrastructure, the deeper
the slowdown.
End
When can public investment pay for itself ?
The marginal impact on net worth of a deficit-financed
rise in the public investment ratio:


m
pK
r
Y
K

d nw   Y 
       m  pK r   
di
  K 

Tax ratio
User charge on public services
Unit maintenance cost
Purchase cost of capital (including corruption, waste etc…)
Marginal cost of borrowing
Marginal product of public capital
When can public investment pay for itself ?
Perotti (2004): In most industrial countries, no –
marginal return is too low.
But marginal return should be higher with lower stocks.
Ferreira (2005): In Brazil, yes – depending on slope of
cost of borrowing.
Estache (2005): In many African countries, yes.