Professional Documents
Culture Documents
Evidence from a Natural Experiment in Peru
Stanislao Maldonado
UC Berkeley
ARE Development Workshop
ARE D l tW k h
April 30, 2010
Research Question
Research Question
• Specific research question:
S ifi h i
How corruption (approached by the demand of bribes by public
officials) is affected by changes in local government revenues?
officials) is affected by changes in local government revenues?
H0: Increase of local resources Low Corruption
• Hard to assess this empirically (omitted bias, measurement error
and reverse causality).
Resource windfall and corruption 2
Economic models
Economic models
– Becker and Stigler (1974):
• Corruption=f(wage, probability of being audited)
• Prediction: Higher salaries help to deter corruption
g p p
• Prediction: More resources may lead to more corruption (moral
y p (
hazard) but also increases number of challengers.
Resource windfall and corruption 3
Data
• Annual repeated cross‐section of household survey (The Encuesta
Nacional de Hogares ‐ENAHO): 2002‐2006.
o Includes a detailed module on payment of bribes by households.
• Annual data about local revenues and transfers from central to local
g
governments (Ministry of Finance): 1998‐2006 .
( y )
• Annual data on mineral production and prices (Ministry of Energy
and Mines): 1998 2008
and Mines): 1998‐2008.
Resource windfall and corruption 4
Empirical strategy
Empirical strategy
• Exploit an interaction between a fiscal rule (Canon Law) and a
positive shock in international prices of the mineral resources:
o Cross‐sectional variation:
Various minerals across districts (districts with and without
l d (d h d h
minerals/districts with different minerals).
o Time variation:
Time variation:
Movement of international prices of different minerals over time.
• Identification:
Compare the bribery behavior of public servants from mineral‐rich
and non mineral‐rich local governments, before and after the rise of
transfers
Resource windfall and corruption 5
Preview of results
Preview of results
• Overall, a negative effect of transfers on corruption with stronger
results for mineral producers
o 1.5‐1.8 percentage point reduction in the probability of being asked to pay a
bribe to a local public official in all districts benefited with Mining Canon
bribe to a local public official in all districts benefited with Mining Canon
transfers.
o 2.0‐2.8 percentage points in the probability of being asked to pay a bribe to a
local public official in mineral producer districts
local public official in mineral producer districts.
• Most benefited areas show an increase in corruption
o 4.3‐4.7 percentage points for the coefficient of the triple interaction in the DD.
• IV results consistent with DD results for a subset of districts.
Resource windfall and corruption 6
Take home message
Take home message
• Size of the increase in transfers matters!
• Implications for the design of intergovernmental transfers schemes
in developing countries
Resource windfall and corruption 7
Outline
1 Motivation
1.
2. Background
3. Research design
4. Empirical model and results
5. Robustness
6. Conclusions
Resource windfall and corruption 8
Motivation
• Corruption is a critical issue in the developing world, but little it is
known about its causes and consequences
• Current literature: subjective measures + macro‐level data
– Omitted bias: E.g. cultural and institutional differences across
countries (Fisman and Miguel 2007)
– Measurement error: corruption measures based on opinion leaders
surveys (Mauro 1995 and 93.7% of the current literature)
– Reverse causality: feedback effects between corruption and economic
performance
Resource windfall and corruption 9
Motivation
• IIssues of external validity in most of recent research based on
f t l lidit i t f t hb d
objective data
p p
• This paper:
– Controls for cultural and institutional differences by studying a country
case.
– Uses high quality data on a relevant dimension of corruption: payment
of bribes to public officials by households.
– Exploits
Exploits an exogenous variation in economic conditions (measured by
an exogenous variation in economic conditions (measured by
local government revenues) to study how corruption (measured by the
demand of bribes by public officials) is affected by changes in local
economic conditions.
Resource windfall and corruption 10
Background
• Some basic characteristics
o Highly centralized country: 97% taxes collected by central government
(Polastri and Rojas 2007)
o Local government highly dependent from central government transfers:
Transfers from central government represent 57% of total local government
revenues
• Intergovernmental transfers system in Peru
o Universal versus targeted transfers
o Universal: FONCOMUN (56% of IGT) and Glass of Milk (10% of IGT)
Universal: FONCOMUN (56% of IGT) and Glass of Milk (10% of IGT)
o Targeted: Canon (mining, oil, hydro power, fishing, forest and gas), Royalties, Camisea
(FOCAM), etc.
o Canon
Canon represents 91% of targeted transfers (Mining canon: 79% of canon
represents 91% of targeted transfers (Mining canon: 79% of canon
transfers)
Resource windfall and corruption 11
Background
• High concentration
Hi h t ti of Mining Canon transfers in some areas
f Mi i C t f i
Resource windfall and corruption 12
Research design
Research design
• Exploit an interaction between a fiscal rule (Canon Law) and a
positive shock in international prices of the mineral resources:
o Cross‐sectional variation:
Various minerals across districts (districts with and without
minerals/districts with different minerals).
o Time variation:
Movement of international prices of different minerals over time.
• H
Huge increase of transfers to mining districts starting 2004 due to
i f f i i di i i 2004 d
extraordinarily high prices of mineral resources
Resource windfall and corruption 13
Research design
Increase of the
Increase of the
Increase Increase
Shock fiscal revenues
of value of of income
prices of rich‐mineral
exports tax
governments
Identification:
Compare the bribing behavior of public servants from mineral‐rich and
non mineral‐rich local governments, before and after the rise of
transfers due to movements in mineral prices.
Resource windfall and corruption 14
Research design
Evolution of international prices of mineral resources: Cupper
300
Prrice of Cupper (ctvs US$/lb)
100 0 200
Resource windfall and corruption 15
Research design
Evolution of international prices of mineral resources: Zinc
140 120
Price of Zinc (cttvs US$/lb)
80 100
1
P
60
40
Resource windfall and corruption 16
Research design
Evolution of international prices of mineral resources: Lead
120 100
Price of Lead (ctvs US$/lb)
60 80
P
40
20
Resource windfall and corruption 17
Research design
Evolution of total exports and mineral exports
3000020000
s of US$
Millions
10000 0
Resource windfall and corruption 18
Research design
Evolution of transfers to mineral‐rich regional and local governments
4000
s)
s of Nuevos Soles (1996 prices
00 2000 3000
Millions
100 0
Resource windfall and corruption 19
Mining canon transfers allocation in 2006 Research design
Resource windfall and corruption 20
Theoretical background and testable
h
hypothesis
h i
• Theoretical models of corruption are often based in simple
demand‐supply framework (See Shleifer and Vishny 1993).
• In
In the Becker and Stigler model, the decision of a public official to
the ecker and Stigler model, the decision of a public official to
become corrupt depends on her wage and the probability of being
audited.
• The testable hypothesis is:
An exogenous increase in municipalities’ revenues which translates
i t hi h
into higher wages (or in general, better labor conditions) for public
( i l b tt l b diti ) f bli
officials will lead –ceteris paribus‐ to a reduction in the supply of
corruption opportunities.
Resource windfall and corruption 21
Potential Channels
Economic factors
Higher wages
Higher wages
Increase Lower corruption
in local
in local More effective
More effective
revenues audit technologies
Political factors
More political Shorter political Higher corruption
competition Horizons
Resource windfall and corruption 22
Econometric specification:
Diff
Differences‐in‐differences
i diff
• The DD empirical specification (Bertrand et al 2004):
yijt = α j + λt + β (Canon jt .HighPt ) + X ijt' δ + ε ijt
Where:
yijt : Outcome of interest for household i in district j.
Canon jjt : Treated districts (Mining Canon beneficiaries /Mineral
Producers).
d )
HighPt : Dummy =1 for high mineral prices period.
X ijt : Household and district level characteristics in period t.
: Household and district level characteristics in period t.
• β recovers causal effect of interest under standard DD
assumptions.
p
Resource windfall and corruption 23
Econometric specification
• The extended DD empirical specification (heterogeneous
effects):
effects)
• δ 3 recovers causal effect for the most benefited areas.
Resource windfall and corruption 24
Empirical Results: DD for Canon recipients
Table II: Impact of Mining Canon Transfers in the Probability of a Bribery Episode in Local
Governments
Difference in Differences Estimates
(1) (2) (3) (4) (5)
Dependent variable: 1=If bribery episode in the municipal
Treatment (1= Canon receiver after
-0.010 -0.015** -0.015** -0.013* -0.014*
increase of prices)
(0.007) (0.007) (0.007) (0.007) (0.008)
M ineral producer 0.022
(0.029)
Constant 0.065*** 0.433*** 0.430*** 0.395** 0.400**
(0.005) (0.165) (0.165) (0.166) (0.166)
Transfer controls No Yes Yes Yes Yes
District Fixed Effects Yes Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes Yes
Urbanization control No No Yes Yes Yes
Household level controls No No No Yes Yes
M ean dependent variable 0
0.03
03
Observations 23,662 22,580 22,580 22,484 22,484
R-Squared 0.011 0.012 0.012 0.013 0.014
Note: * significant at 10%; ** significant at 5%; *** significant at 1%. Standard errors clustered at the district
level.
Resource windfall and corruption 25
Empirical results: DD Canon receivers
Table II: Impact of Mining Canon Transfers in the Probability of a Bribery Episode in Local
Governments
Difference in Differences Estimates
(1) (2) (3) (4) (5)
Dependent variable: 1=If bribery episode in the municipal
Treatment (1= Canon receiver after
-0.010 -0.015** -0.015** -0.013* -0.014*
increase of prices)
(0.007) (0.007) (0.007) (0.007) (0.008)
M ineral producer 0.022
(0.029)
Constant 0.065*** 0.433*** 0.430*** 0.395** 0.400**
(0.005) (0.165) (0.165) (0.166) (0.166)
Transfer controls No Yes Yes Yes Yes
District Fixed Effects Yes Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes Yes
Urbanization control No No Yes Yes Yes
Household level controls No No No Yes Yes
M ean dependent variable 0
0.03
03
Observations 23,662 22,580 22,580 22,484 22,484
R-Squared 0.011 0.012 0.012 0.013 0.014
Note: * significant at 10%; ** significant at 5%; *** significant at 1%. Standard errors clustered at the district
level.
Resource windfall and corruption 26
Empirical results: DD Canon receivers
Table II: Impact of Mining Canon Transfers in the Probability of a Bribery Episode in Local
Governments
Difference in Differences Estimates
(1) (2) (3) (4) (5)
Dependent variable: 1=If bribery episode in the municipal
Treatment (1= Canon receiver after
-0.010 -0.015** -0.015** -0.013* -0.014*
increase of prices)
(0.007) (0.007) (0.007) (0.007) (0.008)
M ineral producer 0.022
(0.029)
Constant 0.065*** 0.433*** 0.430*** 0.395** 0.400**
(0.005) (0.165) (0.165) (0.166) (0.166)
Transfer controls No Yes Yes Yes Yes
District Fixed Effects Yes Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes Yes
Urbanization control No No Yes Yes Yes
Household level controls No No No Yes Yes
M ean dependent variable 0
0.03
03
Observations 23,662 22,580 22,580 22,484 22,484
R-Squared 0.011 0.012 0.012 0.013 0.014
Note: * significant at 10%; ** significant at 5%; *** significant at 1%. Standard errors clustered at the district
level.
Resource windfall and corruption 27
Empirical Results: DD for mineral producers
Table III: Impact of Mining Canon Transfers in the Probability of a Bribery Episode in Local Governments (Producer Districts)
Difference in Differences Estimates
(1) (2) (3) (4) (5) (6) (7) (8)
Dependent variable: 1=If bribery episode in the municipal government
Treatment (1(1= Producer district after
-0.020* -0.020* -0.020* -0.020* -0.027** -0.028** -0.027** -0.027**
increase of prices)
(0.012) (0.011) (0.011) (0.011) (0.012) (0.011) (0.011) (0.011)
Resource windfall and corruption 28
Empirical results: DD mineral producers
Table III: Impact of Mining Canon Transfers in the Probability of a Bribery Episode in Local Governments (Producer Districts)
Difference in Differences Estimates
(1) (2) (3) (4) (5) (6) (7) (8)
Dependent variable: 1=If bribery episode in the municipal government
Treatment (1(1= Producer district after
-0.020* -0.020* -0.020* -0.020* -0.027** -0.028** -0.027** -0.027**
increase of prices)
(0.012) (0.011) (0.011) (0.011) (0.012) (0.011) (0.011) (0.011)
Resource windfall and corruption 29
Empirical results: DD mineral producers
Table III: Impact of Mining Canon Transfers in the Probability of a Bribery Episode in Local Governments (Producer Districts)
Difference in Differences Estimates
(1) (2) (3) (4) (5) (6) (7) (8)
Dependent variable: 1=If bribery episode in the municipal government
Treatment (1(1= Producer district after
-0.020* -0.020* -0.020* -0.020* -0.027** -0.028** -0.027** -0.027**
increase of prices)
(0.012) (0.011) (0.011) (0.011) (0.012) (0.011) (0.011) (0.011)
Resource windfall and corruption 30
Summary of findings DD analysis
Summary of findings DD analysis
• Overall, a negative effect of transfers shock on corruption:
o 1.5‐1.8 percentage point reduction in the probability of being asked to pay a
bribe to a local public official in all districts benefited with Mining Canon
transfers.
transfers
• Effects are stronger in mineral producer districts.
o 2.0‐2.8 percentage points in the probability of being asked to pay a bribe to a
local public official in mineral producer districts.
o Heterogeneous effects: Most benefited areas show an increase in the demand
g
of bribes (4.3‐4.7 percentage points more).
Resource windfall and corruption 31
Econometric specification:
I
Instrumental variables
l i bl
• The empirical IV specification:
Where:
yijt : outcome of interest for individual/household i in district j.
α j : district fixed‐effects.
λt : time fixed‐effects.
R jt : measure of revenues allocated to the district j in period t.
X ijt : individual/household and district level characteristics in period t.
: individual/household and district level characteristics in period t
log C jt * MostBen j
• IV approach: is instrumented with
R jt
Resource windfall and corruption 32
Econometric specification
• Validity of exclusion restriction:
o Mining canon transfers only affect the corruption measure through its effect
on local revenues.
o The change of prices affected basically fiscal revenues and not production
levels.
o Mining
Mining lacks of linkages with other sectors of the economy and only employs
lacks of linkages with other sectors of the economy and only employs
1% of the labor force.
Resource windfall and corruption 33
Econometric specification
Evolution of volume and prices of mineral exports
Resource windfall and corruption 34
Empirical Results (IV: First Stage)
Table IV: The Impact of Mining Canon Transfers on Local Revenues
First S tage
(1) (2) (3) (4)
Log (Total Revenues)
Log (M ining
0.334*** 0.333*** 0.334*** 0.334***
Canon)*M ostBenefitedArea
(0.050) (0.050) (0.050) (0.049)
F-value
F value 109 85
109.85 103 03
103.03 70 09
70.09 53 04
53.04
Transfers controls Yes Yes Yes Yes
District Fixed Effects Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes
Urbanization control No Yes Yes Yes
Household level controls No No Yes Yes
Household's head controls No No No Yes
M ean dependent variable
Observations 22546 22546 22450 22425
R-Squared 0.696 0.696 0.697 0.190
Note: * significant at 10%; ** significant at 5%; *** significant at 1%. Standard errors clustered at the
district.
Resource windfall and corruption 35
Empirical results: IV
Log (M ining
0.334*** 0.333*** 0.334*** 0.334***
Canon)*M ostBenefitedArea
(0.050) (0.050) (0.050) (0.049)
F-value
F value 109 85
109.85 103 03
103.03 70 09
70.09 53 04
53.04
Transfers controls Yes Yes Yes Yes
District Fixed Effects Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes
Urbanization control No Yes Yes Yes
Household level controls No No Yes Yes
Household's head controls No No No Yes
M ean dependent variable
Observations 22546 22546 22450 22425
R-Squared 0.696 0.696 0.697 0.190
Note: * significant at 10%; ** significant at 5%; *** significant at 1%. Standard errors clustered at the
district.
Resource windfall and corruption 36
Empirical results: IV
Log (M ining
0.334*** 0.333*** 0.334*** 0.334***
Canon)*M ostBenefitedArea
(0.050) (0.050) (0.050) (0.049)
F-value
F value 109 85
109.85 103 03
103.03 70 09
70.09 53 04
53.04
Transfers controls Yes Yes Yes Yes
District Fixed Effects Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes
Urbanization control No Yes Yes Yes
Household level controls No No Yes Yes
Household's head controls No No No Yes
M ean dependent variable
Observations 22546 22546 22450 22425
R-Squared 0.696 0.696 0.697 0.190
Note: * significant at 10%; ** significant at 5%; *** significant at 1%. Standard errors clustered at the
district.
Resource windfall and corruption 37
Empirical Results (IV and Reduced Form)
Table V: The Impact of Revenues on Corruption
IV-2S LS Reduced Form
(1) (2) (3) (4) (5) (6) (7) (8)
Bribery episode in the municipal government Bribery episode in the municipal government
Log (Total Revenues) 0.024*** 0.024*** 0.024*** 0.024***
(0.009) (0.009) (0.009) (0.009)
Log (M ining
0.008*** 0.008*** 0.008*** 0.008***
Canon)*M ostBenefitedArea
((0.003)) ((0.003)) ((0.003)) ((0.003))
Constant 0.382** 0.378** 0.350** 0.360**
(0.167) (0.167) (0.167) (0.168)
Transfer controls Yes Yes Yes Yes Yes Yes Yes Yes
District Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes
Urbanization control No Yes Yes Yes No Yes Yes Yes
Household level controls No No Yes Yes No No Yes Yes
Household's head controls No No No Yes No No No Yes
p
M ean dependent variable 0.03
Observations 22546 22546 22450 22425 22580 22580 22484 22459
R-Squared 0.011 0.011 0.013 0.015 0.012 0.012 0.013 0.015
Note: * significant at 10%; ** significant at 5%; *** significant at 1%. Standard errors clustered at the district.
Resource windfall and corruption 38
Empirical results: IV
Resource windfall and corruption 39
Empirical results: IV
Resource windfall and corruption 40
Summary of findings IV analysis
Summary of findings IV analysis
• Strong first stage:
o Elasticity revenue‐Mining canon transfers: 0.325.
o FF‐values
values first stage is really high: 53
first stage is really high: 53‐109
109 (no weak instrument issues).
(no weak instrument issues).
• Results consistent with DD.
o A
A unit change in the log of local revenues is associated with 2.4 percentage
it h i th l fl l i i t d ith 2 4 t
points increase in the probability of being asked to pay a bribe to a local public
official.
o Large effect: 80% increase in the average probability of being asked to pay a
bribe.
Resource windfall and corruption 41
Robustness
• Placebo tests:
o Basic idea: only public officials working for the treated municipalities should
be affected by the treatment
o No effect on public officials whose wages are decided by central government
(police, judiciary, teachers, etc) working in the treated areas should be
expected.
• Validity of exclusion restriction:
o Mining canon transfers should affect corruption only through its effect on local
g p y g
revenues
o No direct effect of Mining canon transfers on incomes should be expected.
Resource windfall and corruption 42
Placebo test: Judiciary workers
Table VI: Placebo Test
Impact of Mining Canon Transfers in the Probability of a Bribery Episode in the Judiciary (Producer Districts)
Difference in Differences Estimates
(1) (2) (3) (4) (5) (6) (7) (8)
Dependent variable: 1=If bribery episode in the Judiciary
Treatment (1(1= Producer district after
-0.010 -0.005 -0.004 -0.007 -0.018 -0.007 -0.006 -0.010
increase of prices)
(0.053) (0.054) (0.055) (0.055) (0.060) (0.062) (0.063) (0.063)
• Significant negative effect of revenues on bribery‐based corruption
in Peru
– After the increase of prices of mineral resources, the predicted probability of
being asked to pay a bribe by local public officials in districts with access to
mining canon transfers reduces by 1.5‐1.8 percentage points (52‐62%
reduction).
– This effect is larger in producer districts (a reduction of 2.7 percentage points).
This effect is larger in producer districts (a reduction of 2.7 percentage points).
• However, when analyzing differential effects in those areas most
benefited from the positive shock of mineral prices, a positive effect
on corruption with an increase in the former predicted probability
on corruption with an increase in the former predicted probability
of 0.043 is found.
• IV results consistent with DD approach.
IV results consistent with DD approach
Resource windfall and corruption 52
Conclusion
– An increase of one unit in the log of local revenues increases the
probability of being required to pay a bribe by 2.4 percentage points.
– This effect is large and represents an increase of 80% in the average
probability of being asked to pay a bribe
probability of being asked to pay a bribe.
• Results suggest that the transfers have differential effects
depending on the magnitude of the shock
depending on the magnitude of the shock.
Resource windfall and corruption 53
Next steps
Next steps
• Still a work in progress.
• Explore potential causal channels using alternative datasets.
• Study with more detail the most benefited areas.
St d ith d t il th tb fit d
• Explore political outcomes, political beliefs, social capital, etc
(ongoing research projects)
(ongoing research projects)
Resource windfall and corruption 54
Q
Questions about bribes in the ENAHO survey
y
Resource windfall and corruption 55
Mining Canon (Law 27506, 2001)
Mining Canon (Law 27506, 2001)
• States that the 50% of income tax paid by mining companies will be
allocated to the regional and local governments located in the area
where the minerals are extracted
• This amount is distributed between:
o The regional government (20%)
o The municipality of the district (10%)
o The municipalities located in the province (25%)
o The municipalities located in the region (40%)
The municipalities located in the region (40%)
o In addition, a 5% is allocated to the public universities of the region
Resource windfall and corruption 56
Table 1. Current revenue of local governments (In percent of GDP)
Resource windfall and corruption 57
Table 2. Transfers to local governments, 2001‐08 (In thousands of new soles)
Resource windfall and corruption 58
Mining canon transfers allocation in 2006
CAJAMARCA
ANCASH
TACNA
MOQUEGUA
PUNO
AREQUIPA
CUSCO
LA LIBERTAD
PASCO
LIMA
ICA
JUNIN
APURIMAC
HUANCAVELICA
AYACUCHO
HUANUCO
SAN MARTIN
MADRE DE DIOS
PIURA
AMAZONAS
LAMBAYEQUE
CALLAO
UCAYALI
TUMBES
LORETO
Source: MEF
Resource windfall and corruption 59
Table 1.Distribution of the mining canon
(1) Ministerial Resolution N 261‐202‐EF/15, priority is given to rural population by weighting the rural district
for two (2) and the urban population by one (1).
(2) Established by the Canon Law (Law No. 27,506), according to population density (Habitante/Km2)
(3) A
(3) According to Population and Poverty linked to basic infrastructural needs.
di t P l ti dP t li k d t b i i f t t l d
60