The Political Resource Curse Redux

Hanyuan Jiang

In the study of the Political Resource Curse (Brollo et al.,2013), the authors identified a new chan-
nel to investigate whether the windfalls of resources are unambiguously beneficial to society, both
with theory and empirical evidence. This paper revisits the framework with a new dataset. Specif-
ically, we implemented a regression discontinuity design and difference-in-difference specification.
arXiv:2405.19897v1 [econ.EM] 30 May 2024

1 Introduction
Unlike the commonly accepted “Dutch disease” hypothesis, which reveals how a windfall of re-
sources can reduce income via the market mechanism, or the literature about the conflict between
different interest groups, the focus of our study is adverse effects on the functioning of political in-
stitutions:how such windfall exacerbates the political agency problem and deteriorates the quality
of political candidates.

According to the extended political agency model introduced in ”The Political Resource Curse”(the
study), an incumbent competes for reelection against a set of heterogeneous challengers. When
faced with an increase in non-tax government revenues (windfall), the incumbent faces a trade-
off between grabbing benefits for himself versus pleasing the voters to increase the probability of
reelection. One result is the moral hazard, as the electoral punishment of corruption diminishes
with budget size, and this induces more frequent misbehaving (corruption). Another result is the
selection effect, which will cause a decline in the average ability of the pool of political candidates.
This is a by-product of the first result and of the assumption that these benefits are more valuable
for political candidates with lower ability.

To test this theory, we use data from a sample of Brazilian municipalities. One unique aspect of
the Brazilian federal transfer system, known as the Fundo de Participação dos Municipios (FPM),
is that transfers to municipalities change exogenously and discontinuously at given population
thresholds, with all municipalities in the same state and a given population bracket receiving the
same transfers. The key empirical evidence is that municipalities just above and below these
thresholds are comparable in all aspects except for the amount of transfers they receive. Under
this setting, we implement a fuzzy RD design with population discontinuities as the instrument
to study the effects of a discrete change in transfers between these municipalities. To complement
the RD design, we also conduct a difference-in-difference estimation.

2 Empirical Framework
2.1 Federal Transfers to Municipal Governments
According to the study, FPM is the single most important source of municipal revenues (40%),
consisting of automatic federal transfers established by the Federal Constitution of Brazil. Our
study focuses on FPM, both for its importance and its allocation depending on a discontinuous
population size which is crucial for identification strategy.

According to the FPM allocation mechanism, municipalities are divided into population brackets
that determine the coefficients used to share total state resources earmarked for the FPM, with
smaller population brackets corresponding to lower coefficients. Each state receives a different

1
 Table 1: FPM Coefficients

Population interval FPM coefficient Actual Transfers Theoretical Transfers

Below 10,189 0.6 19.68 18.15
10,189–13,584 0.8 25.19 23.90
13,585–16,980 1 31.25 30.51
16,981–23,772 1.2 37.22 36.86
23,773–30,564 1.4 43.58 42.86
30,565–37,356 1.6 49.69 49.32
37,357–44,148 1.8 55.68 55.24
44,149–50,940 2 62.61 63.45
Total 33.14 32.27
Source: The Political Resource Curse (Brollo et al., 2013)

share of the total resources earmarked for FPM, two municipalities in the same population bracket
receive identical transfers only if they are located in the same state.

In our model, τ denotes the actual transfers, τ̂ denotes the theoretical transfers. Table 1 shows the
FPM population threshold and the associated FPM coefficients, as well as descriptive statistics,
by population intervals, on the actual and theoretical FPM transfers. Figure 1 depicts the actual
(left panel) and theoretical (right panel) FPM transfers against the population estimates. In each
figure, FPM transfers are averaged over cells of 100 inhabitants, plus the smoothed average of
transfers (solid line) calculated separately in each interval from one threshold to the next. It is
clear that both figures show ”sharp” discontinuities at the FPM thresholds. At the same time,
theoretical transfers show some within-bracket variability because of the different shares received
by each state, and this variability increases with population size.

Overall, the visual evidence provided by the figure justifies the necessity for the regression discon-
tinuity design which we will focus on later. The differences between the actual and theoretical
transfers suggest the presence of non-compliance or manipulation, vindicating the use of fuzzy RD.

Table 2: Outcome Measures

Broad Narrow
Broad Narrow Years of Incumbent
Population Fraction Fraction College
Corruption Corruption schooling reelection
Amount Amount

Below 10,189 0.79 0.37 5.69 2.17 0.40 11.52 0.31

10,189–13,584 0.79 0.50 5.59 1.96 0.39 11.69 0.23
13,585–16,980 0.77 0.44 4.12 1.63 0.42 11.75 0.29
16,981–23,772 0.82 0.55 5.79 2.68 0.43 11.70 0.26
23,773–30,564 0.76 0.49 5.89 2.16 0.50 12.31 0.25
30,565–37,356 0.74 0.43 5.47 2.02 0.55 12.56 0.34
37,357–44,148 0.77 0.39 5.20 1.96 0.50 12.54 0.38
44,149–50,940 0.74 0.52 2.15 1.00 0.59 12.89 0.26
Total 0.78 0.46 5.34 2.09 0.44 11.86 0.28
Source: BNPT small

2
 Source: BNPT small
Figure 1: Actual(left) and Theoretical(right) Transfers

3 Measurement Framework
We will test the FPM discontinuities and the randomness to evaluate the theoretical hypotheses
in the study following a fuzzy RD and DD setting, then we will briefly touch on the electoral
punishment of corruption.

3.1 Instrumental Variables and Regression-Discontinuity Designs

As discussed, the empirical framework suggests a clear link between the variation in population
size and the treatment assignment, which creates an ideal setup for fuzzy RD design.

Namely, FPM Transfers are allocated to municipalities based on population thresholds, with mu-
nicipalities above each threshold receiving higher theoretical transfers than those below. However,
the actual transfers received do not always perfectly align with the theoretical transfers determined
by the allocation rule.

In a sharp RD design, treatment assignment would be a deterministic function of the population

size. In contrast, the fuzzy design accommodates the fact that treatment assignment depends
on the population size (running variable) in a stochastic manner.This fuzzy nature is evident in
Figure 1, which reveals cases of misassignment around the cutoffs, where some municipalities near
a threshold receive transfers inconsistent with their theoretical entitlement.Despite that, the prob-
ability of receiving higher or lower transfers(propensity score) still exhibits discontinuities at the
population thresholds, it allow us to identify the causal effect of transfers on outcomes of interest,
including corruption levels and political indicators.

Based on the study, we use a potential outcome notation to capture both the outcome of interest
(yi ), including corruption, politicians’ education, or reelection, and actual transfers depend on
theoretical transfers plus other stochastic elements. yi (τ̂ ) and τi (τ̂ ) are the potential values of the
outcome and of actual transfers, expressed as a function of theoretical transfers.

First stage:
τi = g(Pi ) + ατ τ̂ + δt + γp + µi (1)

Reduced Form:
yi = g(Pi ) + αy τ̂i + δt + γp + ηi (2)

g(Pi ): high-order polynomial in Pi δt : time fixed effects

γp : state fixed effects
µi , ηi : error terms clustered at the municipality level
ατ : intention-to-treat (IIT) effect of theoretical transfers on actual transfers

3
αy : IIT effect of theoretical transfers on the outcome

As we all know, “fuzzy RD is IV”(Angrist and Pischke, 2009). Fuzzy RD allows us to estimate
the local average treatment effect (LATE). The LATE represents the causal effect of transfers for
complier municipalities, those whose treatment status is influenced by the assignment rule. With
the instrument we could recreate a randomized experiment’s conditions.To evaluate the validity of
this IV strategy, we need to assess whether the following key assumptions are satisfied:

1. Relevance: The instrument (theoretical transfers) must be strongly correlated with the en-
dogenous variable (actual transfers). The result of first-stage regressions presented in (Ta-
ble3) show a highly significant and positive relationship between theoretical and actual trans-
fers. The F-statistics for the excluded instrument are well above the conventional threshold
of 10, indicating a strong first stage and mitigating concerns about weak instrument bias.
2. Exclusion restriction: The instrument should affect the outcome variables (i.e., corruption,
political selection) only through its impact on the endogenous variable (actual transfers). In
other words, there should be no direct effect of theoretical transfers on the outcomes, and no
omitted variables that are correlated with both the instrument and the outcomes. The FPM
allocation rule, which is based on population thresholds, is unlikely to directly influence
corruption or political selection, as it is determined by a formula and not by political or
economic factors. However, there may be concerns about potential violations of the exclusion
restriction if, under exceptional circumstances, the population thresholds coincide with other
policy changes that could affect the outcomes.
3. Monotonicity: The instrument should affect the endogenous variable in the same direction
for all units (municipalities). In the context of the paper, this means that an increase in
theoretical transfers should not lead to a decrease in actual transfers for any municipality. It
seems plausible given the institutional context and the strong positive correlation between
theoretical and actual transfers.
4. Independence: The instrument should be as good as randomly assigned, meaning that it
should not be correlated with potential outcomes or unobserved factors that could influence
the outcomes. In the RD design, this assumption is satisfied if municipalities just above and
below the population thresholds are similar in all respects except for the amount of transfers
they receive.
5. In the original study, comprehensive tests are implemented, the result suggests that the run-
ning variable of our fuzzy RD does not show any evidence of manipulation, so that we can
safely use it as a (local) source of exogenous variation.

Under the continuity assumption discussed in the study, we could write the second stage to identify
the causal effect of FPM transfers on the outcome:

Yi = g(Pi ) + βYi Ti + δt + γp + ϵi (3)

ϵi : error terms clustered at the municipality level

The causal effects we identify by (3) are local in a twofold meaning. First, because of the RD
setup, they only refer to municipalities around the thresholds. Second, because of the IV setup,
they only refer to compliers: municipalities that received larger transfers because of the FPM
revenue-sharing mechanism. The external validity of our exercise is enhanced by the presence of
multiple thresholds. Yet, the identification on compliers leaves aside a subpopulation that might
be of interest on its own: the always-takers, that is, municipalities receiving larger transfers irre-
spective of their position above or below the cutoffs.

4
Note that we control the two-time-period by including time dummies in all specifications and
clustering the standard errors at the municipality level, so we could ignore time subscripts in the
specification in the RD design

We will also delve a little bit deeper into the population estimate (Pi ) term. Recent research by
Imbens et al.(2017) highlighted the potential pitfalls of High-order polynomials, i.e. they can lead
to noisy estimates, sensitivity to the degree of the polynomial, or poor coverage of confidence inter-
vals. However, under these potential concerns regarding high-order polynomials, our results turn
out to be quite robust, so it’s not a big issue. At the same time, alternative methods, including
Local regression could induce over-complex issue, so these models are not covered in the discussion.

3.2 Difference in Difference

To complement the regression discontinuity (RD) analysis, we propose conducting a difference-in-
differences (DD) estimation using the panel structure of the data. The DD approach compares the
changes in outcomes between municipalities that receive additional transfers (treatment group)
and those that do not (control group), before and after the treatment occurs.

The DD model specification is as follows:

Yi,t = α + β1 T reati + β2 P ostt + δDD (T reati × P ostt ) + ϵi,t (4)

where:
• Yi,t is the outcome variable for municipality i at time t
• T reati is a dummy variable indicating whether municipality i belongs to the treatment group
(i.e., receives additional transfers) = 1 if a municipality’s actual FPM transfer(τi,2001 ) exceeds
its theoretical entitlement (τi,2001
ˆ ) in the first mayoral term; otherwise = 0
• P ostt is a dummy variable indicating the post-treatment period, with 0 representing the first
mayoral term, and 1 marking the second.
• (T reati × P ostt ) is the interaction term between the treatment and post-treatment dummies

• δDD is the coefficient of interest, capturing the causal effect of the treatment on the outcome
δDD =(Ytreated,2005 − Ytreated,2001 )-(Yutreated,2005 − Yuntreated,2001 )
• α is a constant term
• ϵi,t is the error term

For T reati , we control for city-specific effect by controlling for for city fixed effects γi , and time
fixed effects ϕt :
Yi,t = α + δDD (Treati × Postt ) + ϕt + γi + ϵi,t (5)
Lastly, we replace the interaction term (T reati × P ostt ) by a continuous variable Windfalli,t ,
measuring the fraction of municipalities receiving excess political windfalls(FPM):

Yi,t = α + δDD Windfalli,t + ϕt + γi + ϵi,t (6)

Where:
• Windfalli,t is defined as [(τi − τ̂ ) × Postt ], representing the excess of FPM received over the
theoretical entitlement at mayoral term.
In the context of the model, we define the treatment group as municipalities that experience an
increase in transfers above a certain threshold (e.g., the median value) from one electoral term to
the next. The control group consists of municipalities that do not receive such an increase. The
DD coefficient δDD captures the average treatment effect on the treated (ATT).

5
 Table 3: The Effect of Federal Transfers on Political Corruption

(a) Reduced-Form Effects: FPM Transfers and (b) IV Estimates: Corruption

Corruption Measures Measures

FPM Transfers Narrow Fraction Amount Narrow Fraction

Overall effect 0.637*** 0.160** Overall effect 0.245**
(0.035) (0.080) (0.121)
Thresholds 1-3 0.577*** 0.193*** Thresholds 1-3 0.322***
(0.048) (0.072) (0.122)
Thresholds 4-7 0.651*** 0.156* Thresholds 4-7 0.239*
(0.039) (0.095) (0.137)

• *** p<0.01, ** p<0.05, * p<0.1

4 Estimation Results
4.1 Regression Discontinuity
The Effect of Federal Transfers on Political Corruption

We start by evaluating the effect of federal transfers on political corruption. The estimation result
of the RD design shows significant effects of fiscal windfalls on the frequency of corruption episodes,
which is consistent with the theory.

Table 3a estimates the Reduced-Form Effects, namely equation (1) and (2). While Table 3b inves-
tigate the baseline IV regressions, which is equation (3).

We focused on the Narrow Fraction Amount, as it captures the proportion of the audited budget
that is associated with serious corruption violations. Therefore we are zeroing in on the most
egregious forms of corruption, which are more likely to reflect intentional wrongdoing rather than
mere administrative irregularities. This provides a clearer conceptual link between fiscal windfalls
and rent-seeking behavior.

In both the reduced-form and IV estimates presented in Table 3, the coefficients for the Narrow
Fraction Amount are consistently positive and statistically significant, both for the overall effect
and for the pooled thresholds 1-3 and 4-7. This suggests that the relationship between federal
transfers and serious corruption is robust. The IV estimates in panel (b) of Table 3 indicate that
a one-standard-deviation increase in actual FPM transfers more than doubles the fraction of the
audited budget involved in serious corruption violations. Figure 2 provides graphical proof of rel-
evant and statistically significant discontinuities in the corruption measures induced by the FPM
policy.

As a conclusion, the estimation results of the RD design confirms that additional revenue triggered
by FPM thresholds significantly boost corruption in the municipality.

The Effect of Federal Transfers on Political Selection

Following the same logic, we investigate the effect of transfers on the quality of political opponents
and on the incumbent’s reelection. Unlike the corruption results, the coefficients here are generally
small and statistically insignificant, especially for the overall sample. The reduced-form estimates
in Table 4a show no significant relationship between theoretical transfers and opponents’ education
or reelection outcomes, except for a marginally significant positive effect on the fraction of oppo-
nents with a college degree in the pooled thresholds 4-7. The IV estimates in Table 4b tell a similar
story: there is no robust evidence that actual transfers affect opponents’ education or incumbent
reelection, although the point estimate for the college variable is again marginally significant for
thresholds 4-7. As noted, the sample size in our analysis is smaller than in the original paper,
which reduces statistical power and makes it harder to detect significant effects, especially for the

6
 Source: BNPT small
Figure 2: ITT Discontinuities: Corruption Measures

Table 4: The Effect of Federal Transfers on Political Corruption

(a) Reduced-Form Effects: FPM Transfers, (b) IV Estimates: Opponents’

Opponents’ Education, and Election Outcome Education and Election Outcome
FPM Transfers College Years of Schooling Incumbent Reelection College Years of Schooling Incumbent Reelection
Overall effect 0.009 -0.009 0.005
Overall effect 0.637*** 0.006 -0.006 0.003 (0.006) (0.042) (0.008)
(0.035) (0.027) (0.005) (0.005) Thresholds 1-3 0.007 -0.041 -0.004
(0.008) (0.061) (0.010)
Thresholds 1-3 0.577*** 0.004 -0.029 -0.003 Thresholds 4-7 0.011* 0.020 0.004
(0.048) (0.039) (0.006) (0.006) (0.006) (0.043) (0.008)
Thresholds 4-7 0.651*** 0.007* 0.016 0.003
(0.039) (0.016) (0.003) (0.003)

• *** p<0.01, ** p<0.05, * p<0.1

heterogeneity analysis across different threshold groups.

4.2 Difference in Difference

The fixed-effects model exploits within-municipality variation over time to estimate the effect of
the continuous treatment variable windfall on the proportion of the audited budget associated with
broad corruption violations (fraction broad). The estimated coefficient on windfall is -0.6714153,
which is statistically significant at the 5% level (p-value = 0.016). This suggests that, on average,
a one-unit increase in the excess transfers received by a municipality (i.e., the difference between
actual and theoretical FPM transfers) leads to a 0.6714153 percentage point decrease in the pro-
portion of the audited budget involved in broad corruption violations. The low between R-squared
(0.0017) suggests that the treatment variable has limited explanatory power for the cross-sectional
variation in corruption levels across municipalities.

It is worth noting that the negative coefficient on windfall is somewhat counterintuitive, as it

suggests that municipalities receiving more transfers than their theoretical entitlement experience
lower levels of corruption. This result could potentially be explained by several factors:

1. Measurement error: If the broad corruption measure is subject to significant measurement

error, the estimated coefficient may be biased towards zero or even have the wrong sign.

7
 Table 5: Difference-in-Differences Estimation Results

(1)
broad corruption
windfall −0.671∗∗
(0.279)

Constant 5.228∗∗∗
(0.0486)
Observations 1097
Adjusted R2 0.011
Standard errors in parentheses
Clustered standard errors at the municipality level in parentheses.
∗ p < 0.1, ∗∗ p < 0.05, ∗∗∗ p < 0.01

2. Omitted variable bias: Despite the inclusion of municipality fixed effects, there may be time-
varying unobserved factors that are correlated with both the treatment and the outcome
variables, leading to biased estimates.
3. Reverse causality: It is possible that municipalities with lower levels of corruption are more
likely to receive excess transfers, rather than the other way around. This could happen if
higher-level governments allocate additional resources to municipalities that are perceived as
less corrupt or more efficient.

4.3 The Electoral Punishment of Corruption

In order to investigate whether the electoral punishment for corruption decreases with larger fed-
eral transfers, the study compare the electoral outcomes of mayors who were audited before the
election to those who were audited after the election, and examine how this difference varies with
the level of federal transfers.

The key specification is a triple difference (DDD) model:

Ei = β1 (T̂i · beforei · yi ) + β2 (beforei · yi ) + β3 (T̂i · beforei )

+ β4 (T̂i · yi ) + α1 T̂i + α2 yi + α3 beforei
+ g(Pi ) + g(Pi ) · beforei · yi + g(Pi ) · beforei + g(Pi ) · yi
+ δt + γp + ψi

Utilising this specification, we could briefly touch on the impact of FPM transfers on the punish-
ment of corruption and we focus on incumbent runs for reelection. As shown in Table 6, The key
variables of interest are:

1. The interaction between the ”before” dummy (indicating whether the audit was released
before the election) and the corruption measures. This captures the effect of disclosing
corruption on the probability of running for reelection.
2. The ”super” interaction between the ”before” dummy, the corruption measures, and the
theoretical FPM transfers. This captures how the effect of disclosing corruption on the
probability of running for reelection varies with the level of transfers.

The results in Panel A show that the disclosure of corruption before the election has a negative
effect on the probability of running for reelection, as indicated by the negative coefficients on
the interaction terms ”Before × corruption” across all corruption measures. The logic behind this
analysis is that if larger transfers reduce the electoral punishment for corruption, incumbent mayors
in municipalities with higher transfers should be more likely to run for reelection even when their
corrupt activities are exposed before the election. The positive coefficients on the triple interaction
terms, while not always significant, provide some support for this hypothesis.

8
 Table 6: Impact of FPM Transfers on the Punishment of Corruptionn

Broad Narrow
Broad Narrow
Incumbent runs for reelection corruption corruption
corruption corruption
Amount Amount

Before × corruption × FPM -0.009 0.166 0.007 0.015

(0.125) (0.112) (0.004) (0.015)
Before × corruption -2.471*** -1.482** -0.050* -0.106
(0.747) (0.588) (0.029) (0.093)
Before × FPM 0.122 -0.011 0.083 0.082
(0.116) (0.090) (0.071) (0.069)
Observations 789 789 740 740
R-squared 0.082 0.088 0.077 0.076
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1

5 Conclusion
This paper revisits a ”political resource curse” framework proposed by Brollo et al.. So far, we
have focused on two fundamental mechanisms in political institutions: the effects of additional
resources on political corruption and on the incentives to participate in politics. In light of these
theoretical results, we have investigated a specific Brazilian institution that provides an ideal
quasi-experimental setting, and we found considerable support for the implications of the theory.

6 Appendix
6.1 The Anti-Corruption Program and Measurements of Corruption
The Brazilian anti-corruption Program initiated in 2003 randomly selects municipalities to be au-
dited by the Controladoria Geral da União (CGU), an independent body that assesses the use of
federal transfers at the local level. The audit reports generated by the CGU form the basis for the
study’s corruption indicators. These reports contain information on the total amount of federal
transfers audited. Because of the random selection process, the program provides a valuable source
of information on budget irregularities and corruption episodes in municipal governments.

We are mainly concerned about how an exogenous increase in FPM transfers around the popu-
lation thresholds affects corruption in the use of all sources of municipal revenues. Since FPM
transfers account for the largest fraction of municipal revenues(40%), and 70% of FPM trans-
fers are unrestricted, we can safely test how benefits grabbed by politicians(corruptions) react to
a change in overall budget size τ (caused by the windfall). According to the theory discussed
in the study, as FPM transfers increase, municipal governments feel less restrained in pleasing
the voters and results in increasing corruption of all kinds, and not just corruption concerning the
FPM transfers. We follow the classification of narrows and broad corruption described in the study.

To proceed, we refer to “BNPT small,” consisting of 1,157 observations, as the random sample
for which we have information on the corruption measures. Table 2 reports descriptive statistics
on these variables by population thresholds. Broad and Narrow corruption are dummy variables
determining the type of corruption. Fraction Amounts measure the amount of corruption.

6.2 Measuring the Quality of Politicians

We measure the Quality of Politicians with reference to education(average features of the pool of
candidates in each municipal election included in our sample). As showed in Table 2, College is the
fraction of opponents holding a college degree. Years of schooling measures the opponent’s aver-
age years of schooling. Incumbent reelection is the probability of the incumbent being reappointed.

9
7 Reference
Brollo, Fernanda, Tommaso Nannicini, Roberto Perotti, and Guido Tabellini. “The Political Re-
source Curse.” The American Economic Review 103, no. 5 (2013): 1759–96.
http://www.jstor.org/stable/42920629.

Gelman, Andrew, and Guido Imbens. “Why High-Order Polynomials Should Not Be Used in
Regression Discontinuity Designs.” Journal of Business & Economic Statistics 37, no. 3 (2019):
447–56. doi:10.1080/07350015.2017.1366909.

Angrist, Joshua D., and Jörn-Steffen Pischke. Mostly Harmless Econometrics: An Empiricist’s
Companion. Princeton University Press, 2009. https://doi.org/10.2307/j.ctvcm4j72.

10