Do Natural Resources In‡uence Who Comes to Power, and How?
An Analysis Oil Price Shocks, Elections and Con‡ict in Colombia
Maria Carreri
Oeindrila Dube
December 2014
Abstract
Do natural resources impair institutional outcomes? We examine this question
with two innovations. First, existing work studies how natural resources in‡uence the
behavior of leaders in power. In contrast, we study how they in‡uence who comes to
power. Second, past work has relied primarily on cross-country approaches. We instead
employ a within-country approach, tracing mechanisms using mico-data. In particular,
we assess how oil price shocks a¤ect local democracy in Colombia, a country mired in
civil con‡ict. We show that when the price of oil rises internationally, legislators a¢ liated with right-wing paramilitary groups win elections di¤erentially in oil producing
municipalities. These e¤ects are larger in con‡ict-ridden locations, where armed groups
are best positioned to intervene in local elections. Consistent with such intervention,
electoral competition also falls with positive oil price shocks. Our …ndings highlight
how natural resources undermine democracy by distorting elections, and demonstrate
that con‡ict leaves the political sector more vulnerable to the resource curse.
We thank Stephen Haber, Pablo Querubin, Sendhil Mullainathan, Shanker Satyanath and David
Stasavaage for useful discussions and suggestions. We are grateful to Juan Vargas for sharing data on
the party a¢ liation of legislators, and to Pablo Querubin for sharing data on mayoral elections.
1
Introduction
Do natural resources lead countries to develop faster? Paradoxically, many have argued
that they instead hinder development (Sachs and Warner 1995 and 2001; Sala-i-Martin
and Subramanian 2013; Gelb 1988). A key reason behind this hindrance may lie in how
natural resources a¤ect politicians’incentives (Ross 1999; Robinson et al. 2006; Caselli and
Cunningham 2009; Mehlum et al. 2006). For example, they may lower accountability by
easing taxation (Madhavy 1970; Huntington 1991; and Morrison 2007), increasing repression
(Caselli and Tesei 2011) or allowing leaders to buy o¤ the opposition (Acemoglu et al. 2004).
Alternatively, they may exacerbate redistributive demands, leading the elite to maintain
autocracy (Boix 2003).
Given these possible distortions in the political sector, many empirical papers have tried
to assess how natural resources a¤ect democratic development. These studies have relied
primarily on cross-country variation and found a range of mixed results. These include: negative e¤ects (Barro 1999; Ross 2001; Jensen and Wantchekon 2004; Ramsay 2006; Aslaksen
2008; Tsui 2011; Brückner Ciccone and Tesei 2011); heterogenous e¤ects (Dunning 2008;
Caselli and Tesei 2011; Ross 2012; and Andersen and Ross 2012); and insigni…cant e¤ects
(Herb 2005; Alexeev and Conrad 2009; Haber and Menaldo 2011; Wacziarg 2011).1
Our paper has two innovations. First, existing work examines how natural resources a¤ect
politicians’behavior once in power.2 However, our work examines how they a¤ect who comes
to power. This is important because the desire to control windfall revenues may motivate
in‡uential groups to constrict electoral participation in a bid to alter election outcomes.
In addition, these groups may turn to violence in the …ght for power. To investigate these
issues, we discern how natural resources a¤ect electoral competition and the type of candidate
elected to o¢ ce. We additionally examine the role of violence.
Our second innovation builds on the observation that it is di¢ cult to identify causal
relationships between natural resources and democracy using cross-country data. It may
be inappropriate to pool countries together given their heterogeneity (Haber and Menaldo
2011). Moreover, there are data limitations to exploring causal mechanisms across nations.
To circumvent these concerns, we conduct a within-country analysis. We trace mechanisms
using micro data, and address key identi…cation challenges in the cross-country literature.
In particular, we examine how oil price shocks in‡uence local elections in Colombia. We
focus on electoral competition and politician selection, as measured by the political a¢ liation
1
Past work has also shown that nontax revenue, including oil revenue, exerts postive e¤ects on regime
stability (Morrison 2009).
2
An exception is Caselli and Cunnginham (2009) who theorize that greater rents will incentivize more
candidates to run for o¢ ce, increasing electoral competition.
1
of winning candidates. Our analysis harnesses panel data on the election of mayors and
local councils in over 1,000 municipalities, covering the 1997-2007 period. Our empirical
strategy assesses whether movements in the international oil price, which are exogenous to
Colombia’s production, in‡uence elections di¤erentially in more oil-dependent municipalities.
Oil dependence is measured on the basis of per capita oil production in each municipality
prior to the start of our sample period.
Colombia is the ideal context for examining this topic. Its long internal con‡ict has
bred many illegal armed groups that seek to control rents from natural resources and are
politically motivated. During our study period, left-wing guerrillas fought the state, and
right-wing paramilitary groups tried to counter these insurgents. Both types of groups employed politically-targeted violence to achieve their goals. But the paramilitaries additionally
engaged in direct electoral intervention. For example, evidence suggests that they channeled
funds to politicians in exchange for favorable policies (Acemoglu et al. 2013).
Our results show that rising oil value alters the political equilibrium: a rise in the price
of oil leads to the di¤erential election of legislators a¢ liated with right-wing paramilitary
groups in more oil-dependent areas. This results in a larger number of pro-paramilitary
mayors and pro-paramilitary local councilors. In addition, we …nd impacts on electoral
competition: positive oil price shocks reduce the number of candidates running for o¢ ce,
and increase the vote margin with which mayors are elected.
Does violence play a role in these distortions? We uncover two pieces of evidence
suggesting it does. First, price shocks exert signi…cant e¤ects on the presence of armed
paramilitaries in oil-rich areas. This is consistent with an account in which illicit groups try
to control territories ‡ush with resource rents, and intervene in elections to achieve control.
Second, the e¤ects are stronger in the locations that experienced higher levels of con‡ict
prior to our sample period. Since armed groups have better infrastructure for intervening
politically in con‡ict areas, our evidence suggests that violence is in fact a part of the story
— something prior work on natural resources and democracy has ignored.
In addition, we …nd little evidence of alternate channels that emphasize the actions
of those already in o¢ ce. Although oil price shocks signi…cantly increase total municipal
revenue, they do not decrease tax revenue, or increase spending in areas such as municipal
employment. Thus we do not observe taxation and spending patterns that are typically
predicted to lower accountability. We also do not detect signi…cant impacts of price shocks on
party-level incumbency, which suggests a limited role of local-level entrenchment.3 Overall,
our results suggest that competition and who comes to power are important dimensions of
3
Individual level incumbency is already restricted by law as mayors are not allowed to for immediate
re-election.
2
the resource curse.
Our within-country analysis has the disadvantage that it draws on data from one institutional context. But this very feature o¤ers at least three advantages for causal identi…cation
relative to cross-country analysis in this area. First, cross-national measures of democracy
such as Polity and Freedom House rely on subjective assessments of political freedoms, which
may not be directly comparable across countries. In contrast, we employ objective measures
of participation in elections, which are clearly comparable across sub-national units. Second,
if natural resources are correlated with other unmeasured characteristics that in‡uence institutional outcomes, these omitted variables may bias the estimates. Since characteristics are
more constant across municipalities relative to countries, municipal-level analysis minimizes
risk of potential omitted variables bias. Finally, our di¤erence-in-di¤erences strategy uses
two sources of variation. We use cross-sectional variation based on the extent of oil production in the municipality. In particular, we measure municipal oil dependence as the value of
oil produced in the beginning of the sample period, scaled by total municipal revenue in the
beginning of the sample period. We also use variation over time in the international price of
oil. Importantly, this oil price is exogenous to the production of any individual country such
as Colombia, which holds a small share of the world oil market. In contrast, this assumption
is implausible for large producers in a cross-country sample.
Our paper is complementary with a handful of other within-country studies of the political
resource curse. Three examine e¤ects of revenue in Brazil. Caselli and Michaels (2013)
show that municipal governments receiving oil revenue spend more but do not provide more
public goods. Monteiro and Ferraz (2012) …nd that mayors increase public employment
and enjoy a short run incumbency advantage in response to oil transfers. Brollo et al.
(2013) also show that general …scal transfers allocated by population thresholds in‡uence
mayoral re-election and corruption in Brazil. Outside of Brazil, Goldberg et. al (2008)
suggests that resource dependence in U.S. states boosts incumbent governor’s vote shares.
The biggest di¤erence between our work and these papers is that they focus on the behavior
of leaders already in o¢ ce, while we focus on who comes to o¢ ce, demonstrating an impact
on party a¢ liation.4 In the remainder of the paper, we describe Colombia’s institutions, oil
production and internal con‡ict; detail the empirical strategy and data; present the results;
and conclude.
4
Monteiro and Ferraz (2012) look at competition, but using a cross-sectional speci…cation which di¤ers
from their strategy in examining other outcomes. They and Brollo et. al (2013) also examine candidate
characteristics such as education, but neither paper examines impacts on party a¢ liation. Similarly, Goldberg
et al. (2008) examines competition but not the party a¢ liation of elected legislators.
3
2
Democracy and Local Government in Colombia
Colombia has held national elections have occurred with regularity since its independence in
1819, with the exception of two periods of military intervention. One occurred during the
mid-nineteenth century. Another occurred in response to La Violencia, a violent confrontation between the two traditional political parties, the Liberals and Conservatives, which took
place over 1948-1958. Although Colombia’s electoral regime is functional, and basic rights
to vote and run for o¢ ce are guaranteed by its Constitution, scholars have qualitatively
noted that the ongoing con‡ict and electoral manipulation by armed groups threatens to
limit these freedoms, particularly in rural areas (Avilés 2006; Bejarano and Pizarro 2002).
In this paper, our goal is to identify how an important natural resource, and its interaction
with the ongoing con‡ict, shape Colombian democracy dynamics.
The direct popular election of local o¢ cials, including mayors, governors and local councils was introduced in 1988. Since then, local elections have taken place at regular predetermined intervals of either two, three or four years. Our analysis focuses on elections of
mayoral and local councils, as these positions vary at the municipality level, and we aim to
identify how municipal oil dependence in‡uences local electoral outcomes.5 In our sample
period, these elections were held in 1997, 2000, 2003 and 2007.
Notably, mayors are not allowed to run for immediate re-election (Dávila 2009), which
minimizes the extent of incumbency in this context.6 Mayors are also considered far more
powerful as compared to the local councils. Councils largely assist mayors in various activities
such as planning public works projects. Yet, their functions are limited,7 and there has even
been recent debate about whether these entities should be abolished.8 The size of the local
council varies based on municipal population, and each municipality typically elects between
seven and 21 councilors.9 The elections for council positions are municipality-wide, and
candidates are elected through a list system.
3
The Oil Sector
Colombia became a net exporter of oil in the mid-1980s, with the discovery of 2 billion barrels
of high-quality oil at the Cusiana and Cupiagua …elds (ANH). Oil is in fact the country’s
5
Governors are department level positions. Over 1,000 Colombian municipalities are aggregated into 33
departments.
6
Non-consecutive re-election is also relatively uncommon, occurring in only 5 percent of the elections
events in our sample.
7
http://www.citymayors.com/mayors/colombian-mayors.html.
8
http://www.eltiempo.com/archivo/documento/MAM-1305716
9
The exception is the capital city of Bogotá, which elects 45 councilors.
4
largest export, though Colombian oil exports constitute a small fraction of the world oil
market.
Over the period of our analysis, municipalities received revenue from oil production based
on the following allocation process. Foreign oil companies operating in Colombia were required to pay the government royalties amounting to 50 percent of their oil export values.
An explicit revenue sharing agreement divided these royalties across the central, departmental and municipal governments.10 The amount given to each municipality was proportional
to its production level. As of 1996, royalty revenues from oil and other natural resources,
which are termed "regalias", have been categorized separately under the …scal accounts. As
we detail below, revenue from natural resources such as oil has played an important role in
the Colombian con‡ict.
4
The Colombian Con‡ict
The Colombian civil war has its roots in La Violencia. The power sharing agreement that
ended this confrontation was perceived to exclude representation of the rural poor and other
political forces. It thus contributed to the formation of revolutionary leftist guerrilla groups,
which launched a communist insurgency during the 1960s. The 1980s saw the rise of a
…rst generation of right-wing paramilitary groups, formed by drug lords and the rural elite
in response to guerrilla extortion and violence. The con‡ict remained low intensity during
this decade when it e¤ectively served as a Cold War proxy, but escalated sharply during
the 1990s for a number of di¤erent reasons, including the growth of paramilitary groups in
their second incarnation. These were illegal armed groups for the period we analyze: they
were not formally a¢ liated with the government, and received no o¢ cial state support in
their formation. Thus, for our sample period, the con‡ict can be characterized as threesided, with the government military, the guerrillas and the paramilitary groups …ghting one
another. Typically, however, the state and the paramilitary groups have been uno¢ cially
allied in …ghting the guerrillas.
4.1
The Guerrillas
The insurgency today is led by the largest and oldest of the guerrilla groups, the Armed Revolutionary Forces of Colombia (FARC) and the National Liberation Army (ELN), which
10
The government places 80 percent of the oil royalties into an Oil Stabilization Fund. As codi…ed in
Law 141, of the remaining amount, 32 percent goes to the central government, 47.5 percent goes to the
department, and 12.5 percent goes to the municipality.
5
were both formed in 1964. These groups had 16,000-20,000 and 4,000-6,000 combatants respectively during our sample period.11 Both groups …ght with the stated aim of overthrowing
the government, and also claim to represent the rural poor by supporting policies such as
land redistribution. Despite the stated ideological motivation, the guerrillas are perceived to
be economically motivated, and pro…t from their involvement in the con‡ict (Richani 1997).
A number of other revolutionary movements such as M-19 (the Movement of April 19) and
Quintín Lame also joined the insurgency during the 1980s. However, most of these other
groups demobilized and formed political parties during the early 1990s. For example, the
M-19 movement formed the M-19 Democratic Alliance political party. In contrast, FARC
and ELN continued as insurgents.
4.2
The Paramilitaries
Independent paramilitary groups emerged separately across di¤erent regions of Colombia
during the late 1980s and early 1990s, but the organizations shared the same underlying view
of violently opposing the guerrillas and left-leaning politicians a¢ liated with these insurgents.
The disparate paramilitary groups started becoming a consolidated force in 1994, with the
launch and expansion of the Peasant Self-Defense Forces of Córdoba and Urabá (ACCU),
a paramilitary group formed by brothers Fidel, Carlos and Vicente Castaño, whose father
had been killed by the FARC. The ACCU began crafting regional alliances, and in 1997
an umbrella organization called the United Self-Defense Forces of Colombia (AUC) was
launched under the leadership of Carlos Castaño. The start of the AUC has been dubbed
the initiation of a second generation of paramilitarism. At peak strength in 2003, the AUC
had approximately 15; 000 members. The paramilitary consolidation marked the beginning
of more direct paramilitary intervention in electoral politics. It also accords with a sharp
upturn in civilian killings, which was an explicit strategy employed by these groups.
In 2003, President Álvaro Uribe began negotiating the demobilization of the paramilitaries. The AUC declared a partial cease-…re, and most paramilitary groups signed an
agreement to disarm in 2005, in exchange for de facto amnesty and lenient sentencing. The
demobilization process generated much controversy around whether the paramilitary units
actually disarmed. Human rights groups have documented continued paramilitary violence
and the emergence of new paramilitary groups since that time.12
11
Both organizations have seen their membership fall in the post-2005 period.
See “Evaluation of the Paramilitary Demobilization in Colombia,"
2006 conference proceedings from Corte a
Impunidad - Colombia en la Mira de la Corte Penal Internacional.
http://www.kolko.de/downloads/evaluation_demobilization.pdf.
12
6
4.3
Armed Group Financing and Predation
Both the guerrillas and paramilitaries have sought territorial dominance via warfare and
targeted political killings. Both groups also rely heavily on the cocaine trade for …nancing
purposes. Correspondingly, increases in the cocaine production have been shown to increase
violence in rural Colombia (Angrist and Kugler 2008; Mejía and Restrepo 2014).
Natural resources also play a central role in the con‡ict: revenues generated by these
resources act as a prize, increasing the value of controlling the territories where they are
located. As shown by Dube and Vargas (2013), exposure to commodity price shocks is a
key determinant of violent attacks by armed groups, with di¤erent actors specializing in
the predation over di¤erent natural resources. Paramilitary groups are documented to be
particularly strong in the oil region. Audits show that oil and gas royalties are often missing
from municipal co¤ers where they exert in‡uence (Human Rights Watch 2005). Revenue
predation takes places as armed groups extort resources under threat of force, for example,
by kidnapping and assassinating mayors (El Tiempo 2007). This form of budgetary predation
became especially important after a major decentralization in 1991 transferred more …scal
resources to local governments (Sanchez and Palau 2006).13
The guerrillas are also known to target infrastructure in oil regions by blowing up oil
pipelines. This represents another way in which the presence of oil and con‡ict are connected
in this context, though these guerrilla attacks don’t appear to be connected to changes in
the value of oil (Dube and Vargas 2013).
5
Paramilitary Intervention in Electoral Politics
Paramilitary groups have sought to in‡uence political outcomes through di¤erent methods
during di¤erent phases. In the mid-1980s, the …rst generation paramilitaries violently targeted candidates from the Unión Patriótica (UP) (Dudley 2004) –a political party formed
by the FARC that successfully …elded candidates in both national and local elections. By
1988, the UP reported that over 500 of its members had been assassinated, including four
Congressman and the party’s presidential candidate. The paramilitaries also targeted other
extreme left groups such as the left-wing M-19 Democratic Alliance Party, whose leader was
assassinated on the orders of Carlos Castaño (El Espectador, 2012).
Evidence suggests that when the AUC formed in 1997, the second generation paramilitary
groups made a strategic decision to in‡uence electoral outcomes. As a part of this strategy,
they formed explicit pacts with politicians to support particular candidates. For example,
13
As such, decentralization of …scal resources to the municipal level has been shown to increase con‡ict
(Chacón 2014).
7
the Pacto de Ralito called for a “refounding of the country” and was signed by prominent
paramilitary leaders and more than 50 politicians including senators, MPs, mayors and
local councilors (Lopez and Sevillano 2008). The large number of links between paramilitary
groups and politicians were revealed by the media after 2006, in what came to be termed
the para-politics scandal. Much of the qualitative documentation on these links stems from
the con…scation of a laptop belonging to the paramilitary leader Rodrigo Tovar Pupo, also
known as “Jorge 40”(Semana 2006).14
The evidence indicates that paramilitary organizations used many strategies to achieve
their political ends. This included providing illegal …nancing to favored politicians, or aiding
their campaigns by eliminating opponents via assassinations (Lopez 2010). As an example,
Jorge 40’s computer revealed a recording of Carlos Maria Garcia Davila, a fellow paramilitary member, coordinating with politicians on important electoral campaign in the Caribe
Coast (Pedraza Saravia and Olaya 2011). Paramilitary groups also engaged in a number of
electorally fraudulent practices such as vote buying, ballot stu¢ ng and the use of voter IDs
among the deceased (Valencia 2007).
Yet another strategy was to coerce voters by threatening violence or carrying out massacres. These e¤orts were aimed at both obtaining votes for particular candidates, and
preventing people from voting (BBC, 2002). One example of this type of voter suppression
comes from the elections in La Jagua de Iberico in the department of César.15
Who were the candidates supported by the paramilitary groups? Acemoglu et al. (2013)
posit that the small new political parties that emerged in Colombia during the 2000s were
actually those associated with paramilitary groups. These authors show that where paramilitaries were present, vote shares for these third parties increased. Fergusson et al. (2013)
also present evidence that greater media exposure did not necessarily improve accountability
— politicians exposed in the para politics scandal prior to elections appeared to receive the
same vote share as politicians exposed after.
6
Mechanisms
There are several di¤erent channels that can in‡uence how natural resources a¤ect institutional outcomes. Natural resources can a¤ect the process through which leaders win o¢ ce,
and thus, determine who gains power. Alternatively, they may in‡uence the actions of those
who already hold o¢ ce. Our paper focuses on the former leader selection channel. Thus, we
14
http://www.semana.com/on-line/articulo/el-computador-jorge-40-puede-inicio-nuevo-proceso8000/81379-3
15
‘Un Abrebocas de estas Elecciones" by Cristina Velez in Votebien.com, February 2006.
8
begin this section by describing the pathway through which oil prices can in‡uence leader
selection, and specify the empirical predictions that should hold if this mechanism is at play.
Then, we describe other theorized channels that have focused on the actions of o¢ ce holders,
and specify additional empirical predictions that should hold if these other mechanisms are
at play.
6.1
Who Comes to Power
How do oil price shocks in‡uence the electoral process through which leaders come to o¢ ce?
When the price of oil increases, more resources are up for grabs in oil-producing areas. In
contexts such as Colombia, where explicit agreements allocate resource revenue across various
levels of government, greater revenue will accrue within the co¤ers of oil-rich municipalities.
The presence of these stealable resources increases the value of controlling oil areas,
particularly when the oil price rises. The bid to control valuable territory, in turn, will boost
armed group’s incentives to intervene in elections as means of achieving political control.
Armed groups face a strong incentive to install favored candidates in local governments,
since local o¢ cials are key to accessing municipal revenue. Within Colombian municipal
governments, mayors, for example, decide on the allocation of public contracts. Moreover,
they can serve as direct conduits for diversion of public funds. In short, revenues may be
siphoned by armed groups through either extortion or collusion with government o¢ cials.
Armed groups can accomplish their goal of manipulating electoral outcomes through several strategies. First, they can …nance and support politicians from parties that are aligned
with their political agenda or intimidate and threaten politicians from non-aligned parties.
These forces will reduce the degree of electoral competition by reducing individuals’ willingness to run for elected o¢ ce. Backing favored candidates within the aligned party may
reduce competition within the favored party, while threatening potential contenders from
other parties may also reduce competition across parties. Together, these e¤ects should
correspond to a reduction in the overall number of candidates running for o¢ ce.
Second, armed groups can also manipulate the electorate to gather votes for their favored
candidates. They can either intimidate voters to keep them away from the polls, or buy their
votes.16 Both the reduction in candidates and manipulation of voters will in turn, in‡uence
the political a¢ liation of elected legislators. All of these e¤ects should be strongest in
municipalities that produce more oil, which is implicit in our predictions as positive oil price
increases are by de…nition, larger in these areas.
16
In fact, both strategies have been documented in Colombia. The impact on voter turnout is therefore
theoretically ambiguous in this empirical context.
9
But, what types of candidates can we expect to observe getting selected into o¢ ce? In
the Colombian context, both the guerrillas and paramilitaries have vied for political control
of valuable territory. However, the paramilitaries in particular have operated disproportionately in oil areas and used explicit strategies to intervene in elections. This generates
two additional testable implications. If electoral intervention is the work of paramilitary
groups, then positive oil price shocks should also increase the set of o¢ cials aligned with
pro-paramilitary parties. Moreover, it should also directly increase the presence of paramilitary groups in oil producing areas.
Finally, the decision to intervene in elections forcefully may interact with the presence of
armed con‡ict. For example, if violent groups already have a base of operations in locations
where con‡ict has taken place, this may lower the cost of electoral intervention in these
places. This account implies a …fth testable implication — the impact of oil price shocks on
election outcomes should be larger in the locations that have previously experienced con‡ict.
6.2
Holding Onto Power
In contrast to the above account of leader selection, natural resources may also a¤ect institutional outcomes through the actions of those who are already hold power. A number of
theories posit that the presence of these resources will entrench leaders. As Boix (2013) has
theorized, the very nature of these assets may serve as a disincentive for transitioning into
democracy. Because resources are …xed factors, they di¤er from other types of capital, in the
there is little threat of exit. This feature may can lead to excess demands for redistribution,
limiting leaders’ incentives to allow democratic politics. Other accounts of entrenchment
have emphasized that o¢ ce holders can spend resources repressing their opponents (Caselli
and Tesei 2011) or buying o¤ the opposition (Acemoglu et al. 2004)
These accounts of entrenchment imply that leaders who are already elected to o¢ ce,
will in turn, remain in o¢ ce. There are two ways in which this entrenchment can matter
for the mechanism we have laid out. First, it can simply be another dimension of the
resource curse. Resources can a¤ect who comes to power, and promote their persistence in
o¢ ce. Thus, oil price shocks may both lead to the election of pro-paramilitary mayors and
promote their subsequent re-election. However, it can also serve as an alternative pathway
for reducing political competition and promoting political selection. In particular, if proparamilitary politicians have already installed themselves oil-rich municipalities at the outset
of our analysis, and rising oil value …nances their e¤orts to retain o¢ ce, this incumbency
advantage itself can promote the re-election of pro-paramilitary politicians.
Although incumbency is limited in the Colombian context since mayors cannot run for
10
immediate re-election, political persistence can manifest in party-level incumbency. Thus, if
incumbency plays a key mediating role, we should observe that oil price shocks lead to the
re-election of mayors who won in the previous election. To the extent that oil value leads to
the disproportionate election of pro-paramilitary mayors, these e¤ects should be larger for
mayors from pro-paramilitary parties.
A number of theoretical accounts also specify that leaders will take di¤erent actions
while in o¢ ce in the bid to retain o¢ ce. Under the canonical rentier mechanism, resource
booms allow o¢ cials to buy o¤ political support through light taxation and increased spending, particularly on patronage (Madhavy 1970; Huntington 1991). Patronage e¤ects play
a prominent role in the work of Robinson, Torvik and Verdier (2006), who theoretically
demonstrate that the desire to stay in o¢ ce leads politicians to ine¢ ciently expand public
sector employment during resource booms. These spending and taxation patterns, in turn,
are held to lower accountability as citizens become politically disengaged and make only
weak demands for representation.
As with the incumbency accounts, the rentier mechanism can both represent another
resource curse outcome in our empirical context, or represent an alternate mechanism giving
rise to political selection. In other words, oil price shocks may lead politicians to alter
taxes and increase spending on municipal level elections. At the same time, these spending
patterns may lead to fewer candidates willing to run for o¢ ce as citizens become disengaged.
If the rentier mechanism plays an important role, then we should observe increased lower
tax revenue in municipal co¤ers as municipal governments lower tax rates in other areas. In
addition, we should observe signi…cant increases in total spending, particularly on municipal
employment.
6.3
Predictions
In summary, if leader selection is important then positive oil price shocks should lower electoral competition, boost the election of pro-paramilitary legislators, and promote greater
paramilitary presence in oil areas. To empirically assess impacts on pro-paramilitary legislators, we draw on both mayoral and local council elections. In particular, we examine if the
mayor elected to o¢ ce was from a pro-paramilitary party, and the share of the local council
that came from pro-paramilitary parties. However, the mayoral elections are better suited
for studying electoral competition. First, the vote margin between the leading candidate
and runners-up can be de…ned for these elections. Second, the interpretation of number of
candidates running for o¢ ce is cleaner since council elections occur through a list system.
If leader retrenchment plays an important role, then oil price shocks should lead to a
11
greater tendency of parties that won o¢ ce last period to win o¢ ce again. Finally, if rentier
type e¤ects are at play, then we should observe lower tax revenue, and increased municipal
spending, particularly on public sector employment.
7
Empirical Strategy
We use a di¤erence-in-di¤erences empirical strategy to test our predictions linking oil prices
to elections. In particular, we assess whether changes in the international oil price exert
di¤erential impacts on institutional outcomes among municipalities that produce more oil.
Our sample encompasses the 1997-2007 period.
The cross-sectional variation in our empirical strategy is based on the amount of oil
produced in each municipality in 1993. Speci…cally, our measure of oil dependence is the
value of oil produced in per capita terms, in 1993.17 57 municipalities produced oil during
that year. Figure 1 shows the quartiles of oil production across municipal locations.
This variable circumvents endogeneity concerns for a number of reasons. It re‡ects the
spatial distribution of oil reserves and precedes the start of the sample period. Thus, it does
not re‡ect potentially endogenous oil discovery or extraction undertaken in response to electoral outcomes. It also precedes the 1997 election by several years, which minimizes concerns
that it re‡ects extraction decisions made with the aim of in‡uencing political outcomes in
the direct run-up to the …rst election in the sample. Finally, the measure precedes paramilitary consolidation which started with the expansion of ACCU in 1994, and culminated in
the formation of the AUC in 1997 (as detailed in the section on the Colombian con‡ict).
The time variation in this empirical strategy is the international price of oil. Importantly,
this price is exogenous to Colombia’s production, as the country holds less than one percent
of the world oil market.
The estimating equation that represents our empirical strategy is:
yjrt =
j
+
t
+
rt
+(Oil jr
OilPrice t ) + Xjrt + "jrt
(1)
where yjrt are elections-related outcomes in municipality j, region r and year t;18 j are
municipality …xed e¤ects; t are year …xed e¤ects; and Xjrt are time-varying controls which
always include the natural log of population. Oil jr is the oil dependence of municipality j
and region r during 1993; OilPrice t is the natural log of the international price of oil in year
t.
captures the di¤erential e¤ect of the oil price on political outcomes in municipalities
17
This is de…ned as the product of barrels of oil produced in each municipality in 1993, multiplied by the
per barrel international oil price in 1993, scaled by the municipal population in millions 1993.
18
The four major geographical regions of Colombia are Andean, Caribbean, Southeastern and Paci…c.
12
producing more oil. Note that the constituent terms do not appear in equation (1) since municipality …xed e¤ects absorb the municipality level Oil jr variable while year e¤ects absorb
the annual level OilPrice t variable.19
r t are linear time trends in the four major regions. These account for potential omitted
variables since natural resources may be concentrated in particular regions, and institutional
outcomes may vary across regions based on factors such as varying economic growth rates
or geographic shifts in the presence of armed groups. For example, oil is concentrated in
the Southeastern region, and armed group presence is held to have increased there in the
latter part of our sample period, when the government seized control of the Demobilized
Zone (DMZ), pushing the FARC eastward toward Venezuela.20
We estimate equation (1) using OLS. We also account for potentially endogenous boundary changes over this period. In particular, new municipalities were carved out of pre-existing
units. We therefore aggregate municipalities to their boundaries in 1988, a year preceding
our sample period. We use a boundary mapping developed by Dube and Vargas (2013). In
all speci…cations, standard errors are clustered at this original municipality level, to control
for serial correlation over time within a municipality.
8
Data
In order to estimate the impact of oil price shocks on electoral outcomes, we utilize data on
oil production, prices and a number of elections related outcomes.
Data on mayoral and local council elections in 1997, 2000, 2003 and 2007 come from the
Colombian national elections council, the Registraduría Nacional del Estado Civil.21 The
standard electoral data source typically excludes elections that did not occur on the o¢ cial
election day. We also avoid using these irregular elections since the timing of re-scheduled
elections may be endogenous. In examining the political competition channel, we use the
number of candidates running for o¢ ce in mayoral elections, and also calculate the margin
of victory. This is de…ned as the di¤erence in the vote share between the elected candidate
and the runner-up in each race.
To gauge impacts on electoral outcomes, we de…ne whether municipal o¢ cials elected to
o¢ ce are a¢ liated with the paramilitaries. These measures are based on an original event19
We examine the e¤ect of prices in levels versus growth since a growth speci…cation may lead to an excess
focus on short-run e¤ects by capturing only year-to-year changes.
20
The DMZ comprises …ve municipalities in Southern Colombia that the FARC were allowed to administer
over 1999-2002. This was a concession by the government of President Andrés Pastrana as a part of peace
negotiations.
21
http://www.registraduria.gov.co/
13
based dataset collected by Fergusson et al. (2013). This data records all news events from
El Tiempo, Colombia’s leading newspaper, in which politicians are accused of collaborating
with paramilitaries over 1997-2010. If a legislator is charged with collaboration, his party
is designated as a pro-paramilitary party for the duration of our sample. This measure is
combined with data from the Registraduría to code the share of the local council that comes
from pro-paramilitary parties, and whether the mayor is from a pro-paramilitary party.
Our data on municipal spending and revenue, including on natural resource regalias span
1997-2005 and come from the National Planning Department (NPD). We also use data on
paramilitary and guerrilla activity, which are available from 1997-2005. We de…ne a measure
of whether the paramilitaries, FARC or ELN are active in a municipality in a given year
based on data from the Center for Study of Economic Development (CEDE). This dataset
originates from the Observatory of Human Rights of the Vice-Presidency of Colombia and is
constructed on the basis of reports from the Administrative Department of Security (DAS),
the Colombian security agency. Our measure of paramilitary and guerilla activity captures
any aggressive event coded by CEDE as involving either a paramilitary or a guerilla group.
The range of activities include arson, attacks on private property, kidnappings, blocking
transport routes, injuring members of the armed forces, or carrying out political homicides.
Our con‡ict data is from Dube and Vargas (2013) and originates from the Con‡ict Analysis Resource Center (CERAC). This data covers war-related episodes in over 950 Colombian
municipalities over 1988-2005. It is event-based, drawing from 25 major newspapers, and
oral reports on political violence from a network of Catholic priests operating in rural areas.
We use three measures of con‡ict: number of paramilitary attacks, guerrilla attacks and
total clashes between the armed groups and the armed groups and the state. These are used
to construct an indicator of a high con‡ict sample, which equals one if any these variables
exceed their mean prior to our sample period, over 1988-1992.
In terms of our independent variables, our measure of oil production comes from the
Ministry of Mines and Energy (MME).22 It is de…ned as the average daily production of
barrels of crude oil, in hundreds of thousands of barrels, in each municipality in 1993. The
international price of crude oil is obtained from the International Financial Statistics (IFS)
and is measured in thousands of 2012 pesos per barrel.
In addition, we use data on coca production in 1994 from the Dirección Nacional de
Estupefacientes (DNE), and obtain data on municipal population from Departamento Administrativo Nacional de Estadística (DANE).23 Table 1 presents the descriptive statistics
of the key variables in the analysis.
22
23
http://www.minminas.gov.co/
https://www.dane.gov.co/
14
Table 2 examines the cross-sectional characteristics of the oil-dependent and non oildependent municipalities, by comparing the mean of these variables that produced any oil in
1993 versus those that did not.24 The table shows that these two types of municipalities do
not di¤er signi…cantly on socioeconomic characteristics at the outset of the sample period,
including an o¢ cial poverty measure of unmet basis needs, and several measures of state
presence from 1997. For example, several variables capturing the degree of service delivery and law enforcement, including the number of schools, hospitals, tax o¢ ces and police
stations do not di¤er vary across these regions.
However, the table also indicates that oil locations are larger as measured by population,
and di¤er on geographic dimensions, as captured by height. Though the indicator of coca
cultivation is not signi…cantly di¤erent at the 10% level, the mean di¤erence of 4% is
substantial considering the overall mean of 7% in our sample. If electoral dynamics di¤er
in the coca belt, highly populated areas, or in geographic regions where oil is located, this
could our estimates if the dynamics also correlate with changes in the oil price. To account
for this possibility, all of our speci…cations include time-varying controls for log population.
We also present speci…cations where we control for elevation and the 1994 coca indicator
interacted with the price of oil. We refer to these as our additional controls.
9
9.1
Results
Oil Price Shocks and Pro-Paramilitary Legislators
We examine the impact of oil price shocks on institutional outcomes by estimating equation
(1) :We start by gauging whether oil price shocks lead to the election of pro-paramilitary
legislators in Table 3. We draw on both mayoral and local council elections. The …rst
two columns examine the election of pro-paramilitary mayors, while the latter two columns
examine the share of the local council from pro-paramilitary parties.25
The results indicate that price rises lead to signi…cant, di¤erential increases in the election of legislators from pro-paramilitary parties among more oil dependent municipalities.
These results hold with and without our additional controls.26 The coe¢ cients also imply
substantial e¤ects. We can evaluate the magnitude by comparing how a given increase in
the price of oil di¤erentially in‡uences the outcome of an oil producing municipality (which
24
The picture looks the same if we instead compare municipaltieis above and below mean oil dependence
in 1993.
25
The pro-paramilitary council share variable is unavailable for 2007. Thus our analysis of council outcomes
include the 1997 to 2003 elections.
26
Even the inclusion of linear time trends based on whether municipalities produce oil leave the coe¢ cients
unchanged, and the e¤ect remains signi…cant for the mayoral elections, for which we have a larger sample.
15
has average oil dependence of 1.01), with a non-oil producing municipality. The coe¢ cient
of .081 in column (2) tells us that a 10% increase in the price of oil increases the likelihood
of a pro-paramilitary mayor by .008 more in the average oil municipality. This represents a
5.7% increase above the mean of this outcome. The price of oil rose increased dramatically,
by 1.3 in log terms, or 130%, over 1997-2007. This large price would translate into a .1
higher likelihood (or 74% greater chance) of getting at least one pro-paramilitary mayor over
the course of four elections that occurred in our sample period. Scaling this e¤ect by four
suggests a per election e¤ect of 18.5%.
The e¤ects on the pro-paramilitary council share are also substantial, though smaller in
magnitude. The coe¢ cient in column (4) suggests that a 10% price hike would di¤erentially
increase the share of pro-para councilors by .004, or by 2.9% above its mean. The e¤ect
corresponding to a 130% price increase would imply a 38% di¤erential increase in the share
of legislators who are pro-paramilitary over the duration of the sample period. Scaling by
four implies a per election boost of 9.5%.
These results provide support for the empirical prediction on political selection. The
higher election of pro-paramilitary legislators in response to oil price shocks is consistent
with the idea that paramilitary groups intervene in more oil dependent areas owing to their
rising value. The larger e¤ect for the mayoral elections may arise because mayors are more
powerful relative to local councils. This power di¤erential suggests that it would be strategic
for paramilitary groups to focus more of their intervention e¤orts on capturing mayorships.
9.2
The Role of Con‡ict
If the impacts on the election of pro-paramilitary legislators re‡ect intervention by armed
groups, then we should expect these e¤ects to be stronger in locations that have experienced
high violence. These are the places where armed groups should have a better developed
infrastructure for carrying out forceful interventions.
To test this prediction, we de…ne a high con‡ict sample based on whether paramilitary
attacks, guerrilla attacks or total clashes exceeded their average levels in the years prior to
our sample period. We include the years between 1988, the …rst year for which con‡ict data
is available, and 1992, the year preceding our 1993 oil production measure. 37 percent of
our municipalities falls into this high con‡ict sample.27
In dividing the sample based on this variable, 38 of the oil producing municipalities fall
27
Specifying this time period enables us to capture the dramatic rise in violence in the …rst half of the
1990s. For example, average total clashes increased 3-fold from .19 to .57 between 1988 and 1992. The
pattern of results also look the same if we measure con‡ict for other time periods such as the …rst three
years through 1990.
16
within the high con‡ict sample, while 19 lie within the lower con‡ict sample. Thus, though
mean oil dependence is higher in the high con‡ict sample, importantly, there is meaningful
variation in oil presence across municipalities in both groups.
Table 4 presents the results examining e¤ects in these two samples. The results indicate
that the impact of the oil price shock is larger and more precisely estimated in the higher
con‡ict sample, for both the pro-paramilitary mayor and council share outcomes. The
mean oil dependence is 1.18 in the con‡ict sample and .7 in the non-con‡ict sample, so these
coe¢ cients have to be evaluated at di¤erent means. For example, the estimate in column
(6) suggests that a 10% price increase would result in a 6% rise above the .16 mean of the
pro-paramilitary mayor variable in the high con‡ict sample. In contrast, the estimate in (5)
would imply a corresponding e¤ect of 2.4% in the low con‡ict sample above the mean of
.14, but this e¤ect is statistically insigni…cant. The implied e¤ects for the pro-paramilitary
share variables in columns (7) and (8) also suggests that the e¤ect would be nearly two times
as large in the high con‡ict sample.
9.3
Oil Price Shocks and Political Competition
To further explore if our main e¤ects on the election of pro-paramilitary legislators re‡ect
intervention by armed groups, we next analyze the competitiveness of local elections. To
do so, we draw on rich, elections data, which track the number of legislators running for
elections at the municipal level, and allow us to calculate the vote margin of the winner in
the mayoral race.28
Table 5 presents these results. All speci…cations include our preferred controls.29 The
…rst column focuses on the total number of candidates. The coe¢ cient of -.239 tells us
that oil price shocks lead to a signi…cant decrease in this outcome. On average, there are
approximately 3.8 candidates running for mayoral elections in a given municipality. The
estimate suggests that the 130% increase in prices over this period would lead to a di¤erential
decline of .3 in the mean oil dependent municipality. This constitutes a 8% di¤erential
reduction in the course of four elections, or a 2% per election e¤ect.
In the next four columns we decompose this e¤ect by looking at candidates a¢ liated with
candidates a¢ liated with the extreme political parties, as well as the non-extreme middle.
In column (2), we examine candidates from the radical left. These extreme left parties are
those that have historically been associated with the guerrillas such as the UP, M19, the
28
As discussed earlier, we focus on the mayoral elections for this anlysis since the vote margin of the winner
can be de…ned, and because the council elections occur via list systems, which makes it di¢ cult to interpret
results around the number of candidates running for o¢ ce.
29
Table A.1 presents the corresponding table without the inclusion of controls.
17
Revolutionary Independent Labour Movement, or the Quintin Lame Movement. In column
(3), we look at all remaining candidates. The results clearly show that there is no e¤ect on
candidates from the radical left, and the fall in candidates re‡ects the fall in the number of
remaining other candidates. In columns (4) and (5) we further disaggregate the grouping in
(3) into the candidates from the right-wing pro-paramilitary groups, and all other candidates
who are neither a¢ liated with the extreme left or right. These coe¢ cients tell us that the
e¤ect of -.271 in (3) is split almost equally between these two sub-groups, with associated
coe¢ cients of -.116 and -.155 for the pro-para and non extreme groups, respectively.
These results suggest three points. First, if paramilitary groups are carrying out electoral intervention, there is little evidence that it targets the radical left. This likely re‡ects
a rational calculus – areas where radical left mayors stand a chance of winning are likely
to be costly to convert into a pro-paramilitary win. Thus, there may be little bene…t from
electorally targeting areas where these candidates are …elded. Second, there seems to be
nearly equitable reductions in within-party and across-party competition. There are fewer
candidates from pro-paramilitary parties, which is consistent with paramilitary groups backing a particular favored candidate. There are also fewer candidates from the center of the
distribution, which is consistent with paramilitary groups targeting groups who are closer
competitors.
Finally, in column (6), we examine the margin of victory with which mayors are winning.
The oil price shock appears to widen the vote margin substantially. A 10% increase in the
price of oil boosts it by 1.4% more in the mean oil municipality, as compared to a non-oil
municipality. The 130% increase in price, in turn, would correspond to a 18.5% e¤ect over
the duration of our sample period.
9.4
Revenue and the Presence of Armed Groups
Do the e¤ects on electoral competition and outcomes re‡ect the work of armed groups seeking
political control in revenue rich areas? To answer this question, we gauge whether oil price
shocks in‡uence regalias revenue and total revenue within municipal accounts. We also
determine whether these price shocks di¤erentially increase the likelihood that paramilitary
and guerrilla groups are active in these municipalities.
Table 6 presents these estimates. The coe¢ cients in column (1)-(2) imply substantial
e¤ects on regalias revenue, which translate into detectable increases in total municipal
revenue. In particular, they indicate that a 10% increase in the price of oil would boost
regalias revenue by 2% more and total revenue by .69% more in the average oil producing
municipality. The 130% oil price increase over this period in turn would imply cumulative
18
e¤ects of 26% and 9% respectively.
To examine if the increase in this oil revenue increases the activity of armed groups,
columns (2) and (3) examine the impact on indicators of whether any activity is undertaken
by the paramilitary, FARC and ELN. The results show a clear pattern: the oil price shock
induced di¤erential impacts on paramilitary presence, without a¤ecting the outcome for
either of the two guerrilla groups.30
Overall, these results point to the importance of political selection and reduced competition as important dimensions of the resource curse. In particular, they accord with the
idea that greater revenue accruing in oil producing municipalities from higher oil value leads
paramilitary groups to intervene in local elections, reducing electoral competition in these
locations. This manifests both as increased detectable presence of paramilitary groups in
the oil region, and ultimately in the election of more pro-paramilitary mayors and local
councilors.
9.5
Alternative Accounts
Finally, in Table 7, we consider and rule out a number of alternative channels that could both
give rise to reduced competition and represent additional resource curse outcomes. These
mechanisms largely correspond to actions undertaken by politicians who already hold power.
First, we consider whether oil price shocks lead to greater re-election of the party in
power among mayoral elections.31 Column (1) shows that this is not the case, on average.
Column (2) considers if there are di¤erential incumbency advantages in municipalities where
the previous mayor was pro-paramilitary. However, the three way interaction with the oil
price shock variable also shows insigni…cant e¤ects. Next we consider whether the rentier
mechanism is at play. This is important because an alternative account would posit that
lower taxation and higher spending or patronage may reduce political engagement, and serve
as an alternate path for individual’s willingness to participate in elections. Columns (3)-(5)
examine impacts on tax revenue, total spending and spending on personnel in the municipal
governments. We …nd no signi…cant e¤ects on any of these variables. The coe¢ cient on
30
This …nding also casts doubt on an alternative account of why right-wing paramilitary groups rise in
response to the oil shock. This alternative account stems from the idea that there may be electoral backlash
against the guerrillas for their role in blowing up oil pipelines. If this destruction reduces the appeal of leftleaning candidates, this might reduce the left parties’incentives for …elding potential candidates, ultimately
reducing their representation on the councils. However, if the oil price shock actually led to more guerrilla
activities in the oil region, we should observe a corresponding increase in the guerrilla presence variable,
which we do not (in Table 6). Moreover, we also do not observe decreases in the number of extreme left
candidates running for o¢ ce (Table 5).
31
This restricts our elections sample to the 2000-2007 period. However, we verify that our main e¤ect
holds with the same magnitude and remains statistically signi…cant amongst this subset of elections.
19
total spending is positive, but the e¤ect is not statistically distinguishable from zero. The
coe¢ cient on the personnel spending is negative, which casts doubt on the importance of
patronage networks. Table A.2 also veri…es that these non-e¤ects are not dependent on the
inclusion of our core control variables. Generally, we …nd little evidence of the incumbency
and rentier mechanisms.
10
Conclusion
This paper has examined how natural resource dependence in‡uences institutions using
a within-country approach. While much of the past literature has focused on incumbent
behavior, and the actions of those in power, we ask whether resource reliance can in‡uence
the electoral process, and determine who comes to power.
Exploiting oil price shocks, we …nd that increases in oil value lead to the increased
election of legislators a¢ liated with right-wing pro-paramilitary parties in more oil dependent
municipalities. We document that these e¤ects are larger in con‡ict locations, and correspond
to the increased presence of armed groups in oil rich regions. We also …nd that these price
shocks reduce electoral competition, as re‡ected in fewer candidates running for o¢ ce and
higher vote margins of winners.
Our results suggest that where armed groups predate on natural resources, and are positioned to intervene in elections forcefully, resource rents will undermine institutions as these
groups act to constrict political participation. As such, the combined e¤ect of con‡ict and
the presence of natural resources may prove particularly inimical to local democracy.
References
[1] Acemoglu, Daron, James Robinson and Rafael J. Santos-Villagran. 2013, "The
Monopoly of Violence: Evidence from Colombia." Journal of the European Economic
Association 11(s1): 5-44.
[2] Acemoglu, Daron, James Robinson and Thierry Verdier. 2004, “Kleptocracy and divideand-rule: a model of personal rule.” Journal of the European Economic Association
2(2-3): 162-192.
[3] Alexeev, Michael and Robert Conrad. 2009, “The Elusive Curse of Oil.”The Review of
Economics and Statistics 91(3): 586-598.
20
[4] Andersen, Jørgen J. and Michael L. Ross. 2014, “The Big Oil Change: A closer look at
the Haber-Menaldo analysis.”’Comparative Political Studies November 2014.
[5] Angrist, Joshua D., and Adriana Kugler. 2008, “Rural Windfall or a New Resource
Curse? Coca, Income and Civil Con‡ict in Colombia.” The Review of Economics and
Statistics. 90(2): 191–215.
[6] ANH, Agencia Nacional de Hidrocarburos, Colombia. http://www.anh.gov.co/
portalregionalizacion/Paginas/antecedentes-historicos.aspx, last accessed:
July 4, 2014.
[7] Aslaksen, Silje. 2010, “Oil and democracy: More than a cross-country correlation?.”
Journal of Peace Research 47(4): 421-431.
[8] Ávila Martinez, Ariel F. 2010, “Injerencia Politica de los Grupos Armados Ilegales.”in
Claudia Lopez H. (ed.) Y Refundaron la Patria: de como ma…osos y politicos recon…guraron el Estado Colombiano, Corporación Nuevo Arco Iris, Random House Mondadori,
Bogotá.
[9] Avilés, William. 2006, “Paramilitarism and Colombia’s Low-Intensity Democracy,”
Journal of Latin American Studies 38: 380.
[10] Barro, Robert J. 1999, “Determinants of Democracy.” Journal of Political Economy
107(6): 158-183.
[11] BBC. 2002, “Intimidación amenaza elecciones.”May 26. http://news.bbc.co.uk/hi/
spanish/latinamerica/newsid2008000/2008716.stm, last accessed: July 3, 2014.
[12] Bejarano, Ana María and Eduardo Pizarro, “From ‘Restricted’ to ‘Besieged: The
Changing Nature of the Limits to Democracy in Colombia,”(University of Notre Dame,
Helen Kellogg Institute for International Studies, Working Paper no. 296, April 2002):
5.
[13] Boix, Carles. 2003, Democracy and Redistribution. Cambridge University Press. Cambridge, UK.
[14] Brollo, Fernanda, Tomasso Nannicini, Roberto Perotti and Guido Tabellini. "The Political Resource Curse." American Economic Review. 103(5): 1759-1796.
[15] Brückner, Markus, Antonio Ciccone and Andrea Tesei. 2011, “Oil Price Shocks, Income
and Democracy.”Review of Economics and Statistics 94(2): 389-399.
21
[16] Caselli, Fancesco and Tom Cunningham. 2009, “Leader behaviour and the natural resource curse.”Oxford Economic Papers 61: 628-650.
[17] Caselli, Francesco and Andrea Tesei. 2011, “Resource Windfalls, Political Regimes, and
Political Stability.”Working Paper, Serie Documentos Cede, 2013-14.
[18] Caselli, Francesco and Guy Michaels. 2013, “Do Oil Windfalls Improve Living Standards? Evidence from Brazil.” American Economic Journal: Applied Economics 5(1):
208-238.
[19] Chacon, Mario L. 2013, “In the Line of Fire: Political Violence and Decentralization in
Colombia.”Working Paper.
[20] Citymayors.com. 2012, “Colombian mayors and local government.” August 12. http:
//www.citymayors.com/mayors/colombian-mayors.html, last accessed: July 3, 2014.
[21] Dávila, Julio D. 2009, “Being a mayor: the view from four Colombian cities.”Environment and Urbanization 21(1): 37-57.
[22] Dube, Oeindrila and Juan Vargas. 2013, “Commodity Price Shocks and Civil Con‡ict:
Evidence from Colombia.”Review of Economic Studies 80(4): 1384-1421.
[23] Dudley, Steven S. 2004. Walking Ghosts: Murder and Guerrilla Politics in Colombia.
New York: Routledge.
[24] Dunning, Thad. 2008, Crude Democracy: Natural Resource Wealth and Political
Regimes New York: Cambridge University Press.
[25] El Espectador. 2012, “El crimen de Carlos Pizarro Leongómez.” October
19th. http://www.elespectador.com/especiales/el-crimen-de-carlos-pizarro-leongomezarticulo-381843, last accessed: July 4, 2013.
[26] El Tiempo. 2007, “Mancuso admitió responsabilidad en asesinato de alcaldes para
extender dominio de paras.” January 16th. http://www.eltiempo.com/archivo/
documento/CMS-3402097, last accessed: July 28, 2012.
[27] El Tiempo. 2000. “Dónde Están los Concejos?.” September 2nd. http://www.
eltiempo.com/archivo/documento/MAM-1305716, last accessed: July 3, 2014.
[28] Fergusson, Leopoldo, Juan F. Vargas and Mauricio A. Vela. 2013, “Sunlight Disinfects?
Free Media in Weak Democracies.”Working Paper.
22
[29] Gelb, Alan H. 1988, Windfall Gains: Blessing or Curse?. Oxford University Press, New
York.
[30] Goldberg, Ellis, Erik Wibbels, and Eric Mvukiyehe. 2008. "Lessons from Strange Cases.
Democracy, Development and the Resource Curse in the U.S. States." Comparative
Political Studies 477-514.
[31] Golosov, Grigorii. 2010. "The E¤ective Number of Parties: A New Approach," Party
Politics 16: 171-192.
[32] Haber, Stephen and Victor Menaldo. 2011, “Do Natural Resources Fuel Authoritarianism? A Reappraisal of the Resource Curse.”American Political Science Review 105(1):
1-26.
[33] Herb, Michael. 2005, “No Representation without Taxation? Rents, Development, and
Democracy.”Comparative Politics 37: 297- 317.
[34] Human Rights Watch. 2005, Smoke and Mirrors: Colombia’s Demobilization of Paramilitary Groups. New York: Human Rights Watch.
[35] Huntington, Samuel. 1991, The Third Wave: Democratization in the Late Twentieth
Century. Norman: University of Oklahoma Press.
[36] Laakso, Markku and Rein Taagepera. 1979, “‘E¤ective Number of Parties: A Measure
with Application to West Europe.”, Comparative Political Studies 12: 3-27.
[37] Jensen, Nathan and Leonard Wantchekon. 2004, “Resource Wealth and Political
Regimes in Africa.”Comparative Political Studies 37(7): 816-841.
[38] López, Claudia. 2010, Y refundaron la patria. . . De cómo ma…osos y políticos recon…guraron el Estado colombiano. Bogotá: Random House Mondadori.
[39] López, Claudia and Oscar Sevillano. 2008, “Balance Politico de la Parapolitica.” Corporacion Nuevo Arco Iris, Bogota.
[40] Mahdavy, Hossein. 1970, “Patterns ad Problems of Economic Development in Rentier
States: The Case of Iran”, in Cook, M.A. (ed.) Studies in the Economic History of the
Middle-East, Oxford University Press.
[41] Mehlum, Halvor, Karl Moene and Ragnar Torvik. 2006, “Cursed by Resources or Institutions?.”The World Economy 29(8): 1117-1131.
23
[42] Mejía, Daniel and Pascual Restrepo. 2014, “Bushes and Bullets: Illegal Cocaine Markets
and Violence in Colombia.”Documento CEDE No. 2013-53.
[43] Monteiro, Joana and Claudio Ferraz. 2012, “Does Oil Make Leaders More Accountable?
Evidence from Brazil’s o¤shore oil boom.”Working Paper.
[44] Morrison, Kevin. 2007. "Natural Resources, aid and democratization: A best-case scenario", Public Choice 131: 365-386.
[45] Morrison, Kevin. 2009. "Oil, Nontax Revenue, and the Redistributional Foundations of
Regime Stability," International Organization 63: 107-38.
[46] Pedraza Saravia, Hernán and Ángela Olaya. 2011, “Regalías, Cooperativas y Finanzas
Paramilitares en la Costa Caribe”, in Romero Vidal, Mauricio (ed.), La Economía de los
Paramilitares: redes de corrupción, nogocios y política, Corporación Nuevo Arco Iris,
Bogota.
[47] Pérez Salazar, Bernardo (2011), “Historias de la Captura de Rentas Públicas en los
Llanos Orientales,”in Mauricio Romero Vidal (ed.), La Economía de los Paramilitares:
redes de corrupción, negocios y política, Corporación Nuevo Arco Iris, Bogotá.
[48] Ramsay, Kristopher W. 2011, “Revisiting the Resource Curse: Natural Disasters, the
Price of Oil, and Democracy.”International Organization 65: 507-529.
[49] Richani, Nazih. 1997, “The Political Economy of Violence: The War-System in Colombia.”Journal of Interamerican Studies and World A¤airs 39(2): 37-81.
[50] Robinson, James A., Ragnar Torvik and Thierry Verdier. 2006, “Political foundations
of the resource curse.”Journal of Development Economics 79: 447-468.
[51] Ross, Michael L. 1999, “The Political Economy of the Resource Curse,”World Politics
51: 297-322.
[52] Ross, Michael L. 2001, “Does Oil Hinder Democracy?.”World Politics 53(3): 325-361.
[53] Ross, Michael L. 2012, The Oil Curse: How Petroleum Wealth Shapes the Development
of Nations, Princeton: Princeton University Press.
[54] Sala-i-Martin, Xavier X. and Arvind Subramanian, 2013, “Addressing the Natural Resource Curse: An Illustration from Nigeria.”Journal of African Economies, Centre for
the Study of African Economies (CSAE) 22(4): 570-615.
24
[55] Sachs, Je¤rey D. and Andrew M. Warner. 1995. revised 1997, 1999. “Natural resource
abundance and economic growth.” National Bureau of Economic Research Working
Paper No. 5398, Cambridge, MA.
[56] Sachs, Je¤rey D. and Andrew M. Warner. 2001, “The curse of natural resources.” European Economic Review, 45(46): 827-838.
[57] Sanchez, Fabio, and María Palau. 2006,“Con‡ict, Decentralization and Local Governance in Colombia, 1974-2004.”CEDE Working Paper No. 2006-20.
[58] Tsui, Kevin K.. 2010, “More Oil, less Democracy: Evidence from Worldwide Crude Oil
Discoveries.”The Economic Journal 121: 89-115.
[59] Valencia, León. 2007, “Los Caminos de la Allianza entre Los Paramilitaries y los Políticos.” in Romero Mauricio (ed.) Para Política: La Ruta de la Expansión Paramilitar y
los Acuerdos Políticos, Corporación Nuevo Arco Iris and Intermedio Editores, Bogotá.
[60] Wacziarg, Romain A. 2012, “The First Law of Petropolitics.”Economica 79: 641-657.
25
Oil Production
Missing
No production
1st quartile of production
2nd quartile of production
3rd quartile of production
4th quartile of production
Figure 1: Oil Production in 1993
Table 1
Summary Statistics of Key Variables
Panel-level variables
Pro-para mayor
Pro-para council share
Number of candidates
Number of radical left candidates
Number of non radical left candidates
Number of pro-para candidates
Number of non radical left / non pro-para candidates
Margin of victory
Paramilitary activity
FARC activity
ELN activity
Log regalias revenue (millions of pesos)
Log total revenue (millions of pesos)
Log tax revenue (millions of pesos)
Log total spending (millions of pesos)
Log spending personnel (millions of pesos)
Party re-elected
Log population (millions), 1997-2007
Population (millions), 1997-2007
Cross-sectional variables
Cultivated coca, indicator, 1994
Oil dependence, 1993
Annual-level variables
Log oil price, thousands of 2012 pesos/barrel
Obs.
Mean
Std. Dev.
Min.
Max.
3415
2814
3415
3415
3415
3415
3415
3358
8442
8442
8442
3165
7756
7742
7873
7777
2489
10,318
10318
0.143
.126
3.819
0.129
3.688
1.670
2.018
.170
.440
.379
.186
4.710
8.595
5.992
8.599
6.363
.323
-4.233
0.041
0.350
.201
1.984
0.348
1.900
1.397
1.451
.155
.496
.485
.389
2.735
.953
1.745
.953
1.040
.468
1.099
0.242
0
0
1
1
1
0
1
0
0
22
2
22
13
12
0
0
0
0
-2.259
4.768
-1.465
5.887
-.929
0
-7.264
0.001
1
1
1
1
11.026
15.580
14.666
15.706
13.039
1
1.953
7.050
1023
938
0.070
.014
0.575
.089
0
0
9.081
1.166
24
10.883
0.651
10.064
12.138
Table 2
Charactersitics of Oil and Non-Oil Municipalities
Variable
Unsatisfied Basic Needs 1993 (poverty)
Secondary education 1993
Police stations
Judicial officers
Post offices
Rural development banks
Schools
Hospitals
Tax Office
Length of primary rivers
Land inequality (gini)
Located in Demilitarized Zone (indicator)
Log population 1997
Elevation
Cultivated coca, indicator, 1994
Mean in Non-Oil Dependent Areas
53.108
0.45
37.807
81.265
41.04
46.711
283.97
21.063
17.757
272,554.29
0.687
0.004
-4.299
1,271.46
0.048
Mean in Oil Dependent Areas
55.217
0.472
47.563
104.843
43.369
45.895
430.737
24.889
19.924
349070.615
0.696
0
-3.75
446.459
0.088
Difference
2.109
0.022
9.756
23.578
2.329
-0.816
146.767
3.826
2.167
76,516.330
0.009
-0.004
0.549***
-825.003***
0.040
Table 3
Oil Price Shocks and the Election of Pro-Paramilitay Legislators
Oil dependence x log oil price
Observations
Number of municipalities
Election Years
Additional Controls
(1)
(2)
(3)
(4)
Pro-para
mayor
Pro-para
mayor
Pro-para
council share
Pro-para
council share
0.081***
[0.022]
0.081***
[0.022]
0.039**
[0.019]
0.036**
[0.018]
3,659
959
1997-2007
N
3,659
959
1997-2007
Y
3,069
1,023
1997-2003
N
3,066
1,022
1997-2003
Y
Notes. Standard errors clustered at the municipality level are shown in parentheses. Variables not shown in all specifications
include municipality and year fixed effects, linear trends by region and log of population. Additional controls include the
interaction of elevation and municipalities cultivating coca in 1994 with the price of oil. *** is significant at the 1% level, **
is significant at the 5% level, * is significant at the 10% level.
Table 4
Oil Price Shocks, Conflict and the Election of Pro-Paramilitary Legislators
VARIABLES
Oil dependence x log oil price
Observations
Number of municipalities
Election Years
Above Mean Conflict?
Additional Controls
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Pro-para
mayor
Pro-para
mayor
Pro-para
council share
Pro-para
council share
Pro-para
mayor
Pro-para
mayor
Pro-para
council share
Pro-para
council share
0.050
[0.079]
0.082***
[0.020]
0.032
[0.037]
0.038*
[0.020]
0.047
[0.079]
0.083***
[0.021]
0.026
[0.037]
0.032*
[0.019]
2,290
599
1997-2007
N
N
1,369
360
1997-2007
Y
N
1,941
647
1997-2003
N
N
1,128
376
1997-2003
Y
N
2,290
599
1997-2007
N
Y
1,369
360
1997-2007
Y
Y
1,938
646
1997-2003
N
Y
1,128
376
1997-2003
Y
Y
Notes. Standard errors clustered at the municipality level are shown in parentheses. Variables not shown in all specifications include municipality and year fixed effects, linear trends by region and
log of population. Additional controls include the interaction of elevation and municipalities cultivating coca in 1994 with the price of oil. Above mean conflict denotes the sample that experienced
above mean paramilitary attacks, guerrrilla attacks or clashes among armed actors over 1988-1992. *** is significant at the 1% level, ** is significant at the 5% level, * is significant at the 10%
level.
Table 5
Oil Price Shocks and Political Competition
(1)
(2)
(3)
(4)
Number of
pro-para
candidates
(5)
Number of
non radical
left / non propara
candidates
Number of
candidates
Number of
radical left
candidates
Number of
non radical
left candidates
Oil dependence x log oil price
-0.239***
[0.084]
0.0011
[0.034]
Observations
Number of municipalities
Election Years
Additional Controls
3,712
967
1997-2007
Y
3,712
967
1997-2007
Y
(6)
Margin of
victory
-0.271***
[0.103]
-0.116
[0.082]
-0.155
[0.117]
0.024***
[0.009]
3,712
967
1997-2007
Y
3,712
967
1997-2007
Y
3,712
967
1997-2007
Y
3,597
957
1997-2007
Y
Notes. Standard errors clustered at the municipality level are shown in parentheses. Variables not shown in all specifications include municipality and year fixed
effects, linear trends by region and log of population. Additional controls include the interaction of elevation and municipalities cultivating coca in 1994 with the price
of oil. *** is significant at the 1% level, ** is significant at the 5% level, * is significant at the 10% level.
Table 6
Oil Price Shocks, Revenue and the Presence of Armed Groups
Oil dependence x log oil price
Observations
Number of municipalities
Years in sample
Additional Controls
(1)
Log
regalias
revenue
(2)
Log
total
revenues
(3)
(4)
(5)
(7)
Log
total
revenues
(8)
(9)
(10)
ELN
activity
(6)
Log
regalias
revenue
Paramilitary
activity
FARC
activity
Paramilitary
activity
FARC
activity
ELN
activity
0.220***
[0.066]
0.068**
[0.030]
0.026*
[0.014]
-0.021
[0.015]
-0.023
[0.015]
0.201***
[0.063]
0.068**
[0.030]
0.024*
[0.013]
-0.018
[0.015]
-0.022
[0.015]
3,427
8,234
9,063
9,063
9,063
3,427
8,234
9,063
9,063
9,063
683
1,007
1,007
1,007
1,007
683
1,007
1,007
1,007
1,007
1997-2005 1997-2005 1997-2005 1997-2005 1997-2005 1997-2005 1997-2005 1997-2005 1997-2005 1997-2005
N
N
N
N
N
Y
Y
Y
Y
Y
Notes. Standard errors clustered at the municipality level are shown in parentheses. Variables not shown in all specifications include municipality and year fixed effects, linear trends by region and
log of population. Additional controls include the interaction of elevation and municipalities cultivating coca in 1994 with the price of oil. *** is significant at the 1% level, ** is significant at the
5% level, * is significant at the 10% level.
Table 7
Alternative Accounts: Incumbency and Rentier Effects
Oil dependence x log oil price
(1)
(2)
(3)
(4)
(5)
Party
re-elected
Party
re-elected
Log tax
revenue
Log total
spending
Log personnel
spending
-0.015
[0.037]
0.003
[0.067]
-0.002
[0.005]
0.027
[0.025]
-
0.044
[0.029]
-
-0.020
[0.034]
-
2,662
905
2000-2007
Y
2,662
905
2000-2007
Y
8,219
1,006
1997-2005
Y
8,397
1,007
1997-2005
Y
8,291
1,007
1997-2005
Y
Oil dependence x price x last mayor pro-para
Observations
Number of municipalities
Sample period
Additional Controls
Notes. Standard errors clustered at the municipality level are shown in parentheses. Variables not shown in all specifications include municipality and year
fixed effects, linear trends by region and log of population. Columns 1-2 also include interactions of the last mayor pro-para indicator with the price of oil.
Additional controls include the interaction of elevation and municipalities cultivating coca in 1994 with the price of oil. *** is significant at the 1% level,
** is significant at the 5% level, * is significant at the 10% level.
Table A.1
Oil Price Shocks and Political Competition (Without Additional Controls)
Oil dependence x log oil price
Observations
Number of municipalities
Election Years
Additional Controls
(1)
(2)
(3)
(4)
Number of
pro-para
candidates
(5)
Number of
non-radical
left and non
pro-para
candidates
Number of
candidates
Number of
radical left
candidates
Number of
non-radical
left candidates
-0.223**
[0.092]
0.0004
[0.034]
3,712
967
1997-2007
N
3,712
967
1997-2007
N
(6)
Margin of
victory
-0.254**
[0.110]
-0.124
[0.081]
-0.129
[0.123]
0.023**
[0.009]
3,712
967
1997-2007
N
3,712
967
1997-2007
N
3,712
967
1997-2007
N
3,597
957
1997-2007
N
Notes. Standard errors clustered at the municipality level are shown in parentheses. Variables not shown in all specifications include municipality and year fixed effects,
linear trends by region and log of population. Additional controls include the interaction of elevation and municipalities cultivating coca in 1994 with the price of oil. *** is
significant at the 1% level, ** is significant at the 5% level, * is significant at the 10% level.
Table A.2
Alternative Accounts: Incumbency and Rentier Effects (Without Additional Controls)
Oil dependence x log oil price
Oil dependence x price x last mayor pro-para
Observations
Number of municipalities
Sample period
Additional Controls
(1)
(2)
(3)
(4)
Party
re-elected
Party
re-elected
Log tax
revenue
Log total
spending
(5)
Log
personnel
spending
-0.023
[0.037]
-
-0.007
[0.068]
-0.002
[0.005]
0.024
[0.025]
-
0.044
[0.029]
-
-0.024
[0.034]
-
2,662
2,662
8,219
8,397
8,291
905
905
1,006
1,007
1,007
2000-2007 2000-2007 1997-2005 1997-2005 1997-2005
N
N
N
N
N
Notes. Standard errors clustered at the municipality level are shown in parentheses. Variables not shown in all specifications include
municipality and year fixed effects, linear trends by region and log of population. Columns 1-2 also include interactions of the last mayor propara indicator with the price of oil. Additional controls include the interaction of elevation and municipalities cultivating coca in 1994 with the
price of oil. *** is significant at the 1% level, ** is significant at the 5% level, * is significant at the 10% level.