65
PEERING THROUGH THE
HAZE: DETECTING CARTEL
SIGNS IN BRAZIL’S LPG
RESALE MARKET
1
Enxergando através da neblina: detectando
sinais de cartel no mercado de revenda de GLP
do Brasil
Rosangela Aparecida Soares Fernandes
2
Universidade Federal de Ouro Preto (UFOP) – Mariana/MG, Brasil
Leonardo Bispo de Jesus Júnior
3
Universidade Federal da Bahia (UFBA) – Salvador/BA, Brasil
Victor Henrique Lana Pinto
4
Universidade Federal de Viçosa (UFV) – Viçosa/MG, Brasil
STRUCTURED ABSTRACT
Objective: the aim of this article was to detect economic signs of cartel in the resale of LPG in the
Brazilian cities of Campina Grande-PB, Caruaru-PE, Jaboatão dos Guararapes-PE, and Recife-PE; the
first market condemned by the Administrative Council for Economic Defense (Cade).
Method: we used a filter suggested in Bolotova, Connor and Miller (2008) that consists of estimating
an ARCH model and its generalization, a Generalized Autoregressive Conditional Heteroskedasticity
(GARCH), to test the hypothesis that, during the cartel period, the average price is higher and its
variance lower.
Conclusions: we showed that the average LPG price increased and the variance decreased during
the cartel period in all cities we analyzed. We also corroborate the eciency of the method, as the
hypothesis of economic signs of cartel was not rejected for Campina Grande-PB.
Keywords: economic signs of cartel; economic filters; LPG resale market; collusion.
1 Editor responsável: Prof. Dr. Victor Oliveira Fernandes, Conselho Administrativo de Defesa Econômica (Cade), Brasília,
DF, Brasil. Lattes: http://lattes.cnpq.br/5250274768971874. ORCID: https://orcid.org/0000-0001-5431-4142.
Recebido em: 24/08/2024 Aceito em: 11/10/2024 Publicado em: 11/12/2024
2 Doutorado em Economia Aplicada pela Universidade Federal de Viçosa.
3 Doutorado em Ciências Econômicas pela Universidade Estadual de Campinas.
4 Doutorado em Economia Aplicada pela Universidade Federal de Viçosa.
3
66
FERNANDES, Rosangela Aparecida Soares; JESUS JÚNIOR, Leonardo Bispo de; PINTO, Victor
Henrique Lana. Peering through the haze: detecting cartel signs in Brazil's LPG resale market.
Revista de Defesa da Concorrência, Brasília, v. 12, n. 2, p. 65-84, 2024.
https://doi.org/10.52896/rdc.v12i2.1400
RESUMO ESTRUTURADO
Objetivo: o objetivo deste artigo foi detectar indícios econômicos de cartel na revenda de GLP nas
cidades brasileiras de Campina Grande-PB, Caruaru-PE, Jaboatão dos Guararapes-PE e Recife-PE; o
primeiro mercado condenado pelo Conselho Administrativo de Defesa Econômica (Cade).
Método: utilizamos um filtro sugerido em Bolotova, Connor e Miller (2008) que consiste em estimar um
modelo ARCH e sua generalização, a heterocedasticidade condicional autorregressiva generalizada
(GARCH), para testar a hipótese de que, durante o período de cartel, o preço médio é maior e sua
variância menor.
Conclusões: mostramos que o preço médio do GLP aumentou e a variância diminuiu durante o
período de cartel em todas as cidades analisadas. Também corroboramos a eficiência do método,
pois a hipótese de indícios econômicos de cartel não foi rejeitada para Campina Grande-PB.
Palavras-chave: indícios econômicos de cartel; filtros econômicos; mercado de revenda de GLP; conluio.
JEL codes: L4; L40; L41.
Summary: 1. Introduction; 2. LPG resale: structure, conduct, and summary of
the AP; 3. Theoretical considerations on economic filters for cartel detection;
4. Methodology; 5. Results; 6. Concluding remarks; References; Appendix.
1 INTRODUCTION
In several countries, the antitrust law presumes that the cartel is harmful to consumers, as
it promotes an increase in prices, reduces the supply of products on the market, their quality and
competition between companies. For this reason, cartels are considered illegal acts per se by several
jurisdictions (Malan; Resende, 2022). For Hovenkamp (2011), the overchange, the dierence between
the price of a product practiced in a cartelized environment and the one that should be charged if
sold in a competitive market, is the main eect of the cartel. The results of empirical research carried
out by Connor (2007), for example, showed that cartels generate, on average, an overchange of 25%.
Due to the social welfare losses generated by this anti-competitive conduct, the detection of
cartels represents one of the most important tasks for competition authorities around the world, in
addition to being one of their greatest challenges. As highlighted by Abrantes-Metz, Bajari and Murphy
(2010), despite the absolute success of leniency programs, some cartels still remain undetected.
Members of the collusion, fearing detection, oten take steps to conceal their actions. Thus, due to
the diculties faced, antitrust agencies have sought alternative approaches to identify cartelized
conducts. According to Abrantes-Metz and Bajari (2010), one of the alternative tools used is the
economic filter for cartel detection.
For Abrantes-Metz and Bajari (2010), economic cartel filters seek to identify atypical or unlikely
patterns in markets in competitive equilibrium. Abrantes-Metz, Bajari and Murphy (2010) point out
that the use of modern statistical methodologies enables the discovery of conspiracies based on the
anomalous patterns they generate in the data. Harrington Jr. (2008) adds that the application of filters
67
makes it possible to detect suspected collusive behavior, although it does not provide conclusive
evidence of this conduct. In summary, they produce indirect proof, based on the use of economic
data, to verify possible occurrences of cartels, in order to channel the action of the control agencies
to possible cases of such an illicit act.
In this context, in recent years, there has been an expansion of studies that include the
application of statistical methods, known in the literature as economic cartel filters, to detect the
pattern of price behavior that prevails in markets where collusive anticompetitive conduct is practiced.
According to Connor (2005), since the mid-1990s, several empirical studies have been motivated by
discoveries of international cartels and sanctions established by antitrust agencies around the world.
Bolotova, Connor and Miller (2008) highlighted that, in the past decades, researchers in the area of
industrial organization have used available data and new econometric techniques to study cartels
and test hypotheses based on theoretical literature.
Given the importance of economic cartel filters as an alternative tool for detecting cartels, the
aim of this article was to spot the economic signs of cartel in the resale of Liquefied Petroleum Gas
(LPG) in Campina Grande-PB, (a market recently condemned by Cade
5
-
6
), and Caruaru-PE, Jaboatão dos
Guararapes-PE, and Recife-PE, (previously in the Administrative Proceeding (AP) 08700.003067/2009-
67
7
). Therefore, we conduct an ex-post evaluation of the filter suggested by Bolotova, Connor, and
Miller (2008), thus verifying its eciency, based on its application for the resale of LPG in Campina
Grande-PB, a market recently convicted of forming a cartel by the Brazilian antitrust agency. In other
words, we seek to ascertain whether the results obtained are not subject to rejection of the cartel
hypothesis, when the practice was detected by Cade.
In sum, we believe that the results found in this study show a behavior pattern in LPG resale
prices consistent with what prevails in cartelized markets. In other words, we expect that the hypothesis
raised in Bolotova, Connor and Miller (2008) that the average price is higher and its variance lower,
during the cartel period, is not rejected. The acceptance of this hypothesis, for the results of Campina
Grande-PB, makes it possible to corroborate the eciency of the application of this methodology
for the resale of LPG, as it shows that the detection of economic signs of cartel for this market is not
subject to type II error (false negative). Abrantes-Metz and Bajari (2010) observe that, for a robust
economic assessment of cartel economic evidence, the tests must have strong statistical power and
minimize the number of false positives and negatives.
This study makes two significant contributions to the existing literature. Firstly, it employs
an economic cartel detection filter, which has been little explored through this methodology in the
Brazilian context. Secondly, the study’s findings on a market already condemned by the Brazilian
antitrust agency have the potential to validate the economic cartel filter proposed by Bolotova, Connor,
and Miller (2008) in the Brazilian LPG market.
5 All Cade’s cases mentioned in this article can be consulted at the following link: https://tinyurl.com/y7obr4z5.
6 The next section presents a historical summary regarding the Administrative Proceeding (AP) 08700.003067/2009-67.
7 Initially, in the Administrative Proceeding 08700.003067/2009-67, the investigation into the formation of cartels, in
distribution and resale, in cities of the Ipojuca area was mentioned. This area comprises 408 municipalities although the price
survey carried out by the National Petroleum Agency (ANP) involves only 31 municipalities. In addition, of these, 6 of them have
at least 15 resale stations in operation: Caruaru-PE, Jaboatão dos Guararapes-PE, Recife-PE, Olinda-PE, Campina Grande-PB,
and João Pessoa-PB. In this article, a priori, the analyses would be carried out for these 6 municipalities, however, Olinda-PE
and João Pessoa-PB were excluded because the previous volatility test of the LPG price series showed that both municipalities
did not have ARCH eect in these markets.
68
FERNANDES, Rosangela Aparecida Soares; JESUS JÚNIOR, Leonardo Bispo de; PINTO, Victor
Henrique Lana. Peering through the haze: detecting cartel signs in Brazil's LPG resale market.
Revista de Defesa da Concorrência, Brasília, v. 12, n. 2, p. 65-84, 2024.
https://doi.org/10.52896/rdc.v12i2.1400
This study is divided into five sections, in addition to this introduction. In the second, a
characterization of the structure and conduct of the LPG resale is described and a historical summary
of the case involving the AP 08700.003067/2009-67 is shown. In the third, a brief theoretical approach
on cartel economic filters is presented. In the fourth, the methodology is exposed. In the fith, the
results and discussion are presented and, in the sixth section, our concluding remarks.
2 LPG RESALE: STRUCTURE, CONDUCT, AND SUMMARY OF THE AP
The LPG is highly representative in the consumption basket of the Brazilian population, as
it is the most used source of energy in cooking food. The northeastern region of Brazil, where the
municipalities of Campina Grande-PB, Caruaru-PE, Jaboatão dos Guararapes-PE, and Recife-PE are
located, represents the second largest LPG consumer market in the country, with 25% of national
consumption, second only to the southeast region, which consumed 43% of LPG in 2020 (Sindigás,
2021). In 2020, the national demand for this product was 13.68 million m3, with 80.6% of this volume
destined for domestic consumption (Empresa de Pesquisa Energética - EPE, 2021). In view of this, the
practice of abusive LPG prices denotes a topic of considerable importance for society and object of
monitoring by the antitrust authorities, due to the negative impacts generated on the social well-
being. Colomer and Vernersbach (2022) point out that the final price of the LPG is the sum of the
production/import value, taxes and distribution and resale margins. Therefore, it reflects to some
extent the structure and competition dynamics of each stage of the Brazilian LPG industry.
In this section, we present a brief characterization of the LPG resale in Brazil, highlighting the
structure and conduct of companies operating in this market segment. Next, a historical summary of
the AP 08700.003067/2009-67, which involves the investigation into the cartel case, object of analysis
of this study. Given that the focus of this article is the reseller market, which supplies homes, the
bottled LPG with a capacity of 13 kilograms, that is, 13 kg cylinders or P-13 is considered the relevant
product market, as defined by Cade. Cylinders with a capacity of P20 to P90 are part of the bulk
market, which serves consumers who demand greater volumes of gas, such as commercial clients,
small and medium-sized companies and, above all, the food, metallurgical and steel industries. In
this case, it is the distributors that sell and deliver the LPG directly to the final consumer.
The LPG production chain is long and complex, made up of producers, importers, distributors
and resellers. The last stage, the focus of this article, is responsible for the acquisition, storage,
transport, and commercialization of the product in transportable containers. Resellers can be linked
to a single distributor, using its brand, or independent, purchasing the product from more than one
distributor. In this case, the distributors’ brands cannot be displayed (Lodi; Bicalho, 2022).
The market structure of the LPG resale in Brazil is comprised of thousands of companies,
most of which are small. According to Sindigás (2021), in 2020, there were 61,610 thousand authorized
LPG dealers in the country, supplying 91% of Brazilian families. However, despite the fact that the
Brazilian reseller market is atomized, for analyses of concentration acts and anticompetitive conducts
involving the resale of LPG, Cade delimits the municipality as a relevant geographic market. For this
reason, the number of resellers decreases considerably when compared to the national market. More
recently, in the northeast region of the country, there exist nearly 12,707 authorized LPG resale points.
Specifically, in the states of Paraíba and Pernambuco, the number of resellers reaches 1,080 and 1,604,
69
respectively. In Campina Grande-PB, for instance, a relevant geographic market convicted of forming
a cartel, there are 86 registered resellers (ANP, 2023).
The resale of LPG has characteristics that favor the adoption of cartel behavior, as it oers
a homogeneous and essential product, in which there are no close substitutes, the cost structure
is similar, there are regulatory barriers to entry, active action by unions and associations, and price
advertising. In view of this, even in municipalities where there is a large number of dealerships
operating, cartels may form as agents can make stable agreements, with the participation of unions
or distributors that implement eective coordination, monitoring and punishment mechanisms.
Specifically, it appears in the documents of the AP 08700.003067/2009-67, that dealerships
in the city of Campina Grande-PB and surroundings adopted cartel-like behavior with the support
of employees of distributors and the Union of Fuel and Petroleum Derivatives Dealers of the interior
of Paraíba – Sindirev. According to the information reported in Technical Note no. 66, because of the
complexity of the LPG commercialization logistics chain, with its resale composed of a high number
of agents and points of sale, a sophisticated monitoring and retaliation system was adopted against
LPG resellers or subordinate employees who disrespected the agreements of price fixing and division
of the reseller market.
Regarding the AP 08700.003067/2009-67, the case began in 2009, with a representation from
the ANP and, later, investigations were initiated by the Federal Police and the Public Prosecution
Service of several states of the Northeast region. The Federal Police, in partnership with the
Secretariat of Economic Law of the Ministry of Justice (SDE/MJ) and the Public Prosecutor’s Oce of
the state of Paraíba, launched the “Operação Chama Azul” or Blue Flame Operation in March 2010,
when several search, seizure and temporary arrest warrants in several states. Ater the filing of the
criminal action, before the Justice of Paraíba, Cade obtained the sharing of the evidence related to
the alleged anticompetitive practices, which included telephone interceptions and documents seized
in the oces of the investigated companies.
According to the findings of the investigation, the accused organized themselves to restrict
competition through price fixing, division of product distribution and resale markets. Cade concluded
that the dealers exchanged commercially sensitive information to promote the artificial regulation
of the LPG market and facilitate the maintenance of the alleged cartels. The evidence presented
in the lawsuit revealed that the cartel reached the resale of LPG in Paraíba and, also, reached the
distribution market of several Northeastern states. In this context, the companies Nacional Gás Butano
Distribuidora, Revendedora de Gás da Paraíba, and Frazão Distribuidora de Gás were convicted, in
August 2022, of forming a cartel in the LPG distribution and resale market.
3 THEORETICAL CONSIDERATIONS ON ECONOMIC FILTERS FOR CARTEL
DETECTION
This section seeks to present a general theoretical view of economic cartel filters used
to identify the behavior pattern of companies that operate in markets where this anticompetitive
conduct is likely to occur. As the literature on the subject is dense, especially the international one,
the focus in this section is to lay out a brief theoretical foundation associated with the empirical
model used in this article, as a tool for detecting economic signs of cartel.
70
FERNANDES, Rosangela Aparecida Soares; JESUS JÚNIOR, Leonardo Bispo de; PINTO, Victor
Henrique Lana. Peering through the haze: detecting cartel signs in Brazil's LPG resale market.
Revista de Defesa da Concorrência, Brasília, v. 12, n. 2, p. 65-84, 2024.
https://doi.org/10.52896/rdc.v12i2.1400
Cartels are considered one of the most harmful anticompetitive conducts to society, as there
are no eciency gains associated with this practice. Furthermore, the secret nature of the agreements
makes their detection and investigation a major challenge. According to Cuiabano et al. (2014), due
to the diculties faced by antitrust authorities in obtaining clear proof regarding a cartel, economic
filters can oer additional evidence for the existence of a possible conspiracy. In addition, the use of
filtering techniques can help antitrust authorities to initiate investigations in markets where there are
still no formal complaints.
In recent years, there has been significant progress in the development of tools capable
of detecting economic signs of cartel in sectors where this conduct is likely to occur. Silveira et al.
(2021) highlighted that the literature on economic cartel filters has been growing considerably and
the study of Bolotova, Connor and Miller (2008) represents one of the most notable contributions to
the literature. One of the advantages of the model used by these authors is that it only demands the
price series of the product commercialized in the stage of the productive chain investigated. Thus, it
makes it possible to carry out the analysis in the absence of information regarding costs, for example,
relying solely on the prices charged by distributors.
In their study, the authors used a price volatility model, with the introduction of a dummy
variable, during the cartel period, aiming to detect the eect of the conspiracy, known or hypothetical,
on the prices of citric acid and lysine markets in the United States, from 1990 to 1997, and from 1990
to 1996, respectively. The authors sought to test the hypothesis that average prices, during the cartel
period, are comparatively higher than in the pre- and post-conspiracy periods and that the price
variance, during the collusion, is lower than in other periods. Their results confirmed the hypothesis
of higher price and lower variance, during the conspiracy period, for the lysine market. However, it
was not possible to corroborate a lower variance for the citric acid market.
Another notable contribution to the literature, which is based on a hypothesis similar to
that adopted in Bolotova, Connor and Miller (2008), with a dierent technique, however, is the article
developed by Abrantes-Metz et al. (2006). These authors investigated economic evidence of cartel at
279 gas stations in Louisville, United States, from 1996 to 2002. To do so, they used a filter to detect
economic evidence of cartels, based on the variation coecient of prices over time, defined by the
ratio between the standard deviation and the average price. Their results suggested that the variation
in price was comparatively smaller during the period of the cartel, and, ater the discovery of the cartel,
excluding the transition from collusion to non-collusion, there was an increase in price variance.
Harrington Jr. and Chen (2006) also assume that when companies are involved in a cartel,
average prices increase and their variance decreases. Thus, the behavior of cartel participants is
divided into two stages. In the first, there is a gradual increase in prices, that does not follow the
oscillations in costs, maintaining the growth trend. This increase must appear to be a natural increase
and therefore allow the justification that the increase in prices occurred due to intrinsic market
conditions. In the second stage, prices are stabilized, as, ater reaching the collusive equilibrium
point, firms readjust them proportionally to cost variations. Nevertheless, prices fluctuate less than
costs when compared to the competitive market.
There are two types of economic filters for detecting cartel evidence, structural and behavioral.
The former seeks to identify markets that have characteristics considered conducive to collusive
behavior. In this case, cartels probably occur in sectors where the number of companies is relatively
71
small, the products oered are homogeneous and the demand is more stable. The behavioral
approach, however, aims to verify the means by which companies coordinate and the result of such
a coordination (Harrington Jr., 2008). According to the Organization for Economic Cooperation and
Development (OECD) (2013), behavioral filters attempt to identify patterns of unusual and unexplained
behavior that may indicate that a cartel is active. Aaltio (2019) stresses that neither approach provides
a definitive answer on the cartelized conduct of companies, but the behavioral one is more satisfactory
in terms of causality.
Although there exist several methods to empirically investigate cartels, behavioral economic
filters, based on price dynamics, allow us to obtain consistent indications regarding this conduct. Price
is an accessible variable, because of its interaction with the consumer and is frequently revealed so
that negotiations take place. In the short term, it is sensitive to agents’ behavioral changes. Moreover,
for cartel members, it is dicult to determine quantity as a cooperation aim, since, unless the company
is willing to provide its sales and production figures to cartel participants, this information is dicult
to access. In this context, price is the control variable within the cartel and, consequently, an element
conducive to filtering (Harrington Jr., 2008).
Regarding the identification of collusion, Harrington Jr. (2008) points out that behavioral
economic filters can be applied using the screening and verification techniques. In the first, the filter
evaluates the possible conspiracy before it is detected while in the second, it analyzes the economic
conspiracy during its existence, that is, cases confirmed by the antitrust agency.
In short, the theoretical and empirical findings in the literature allow us to try and spot
dierent pricing patterns associated with the existence of economic signs of a cartel
8
. However, as
emphasized in Froeb et al. (2015), in economics there is still no universal methodology that makes it
possible to detect it. On the other hand, Harrington Jr. (2006) points out that economic analyses can
play a significant role in detecting cartels, stimulating political discussions and academic research.
4 METHODOLOGY
To achieve the aim of this study, firstly, we estimated the ARCH model for a cartel case in the
resale of LPG, recently condemned by Cade, referring to the AP 08700.003067/2009-67. Our findings
enabled us to determine the eectiveness of the economic filter in detecting indicators of cartel
behavior. Ater that, we applied the methodology to other relevant geographic markets that were
previously mentioned in the AP. In all cases, we intended to verify whether the average LPG prices
practiced in those municipalities were higher and their variance lower during the cartel period, thus
suggesting economic signs of a cartel.
8 The following features represent price markers for detecting cartels in the literature: i) Very low price variances
are indicative of a stable cartel; ii) High and inconstant variances over time would be associated with cartels with punitive
measures; iii) Non-linear price variations to cost adjustments, in particular, the drop in upstream prices not being passed on in
the same proportion or with the same timeliness as the increase in costs; iv) Pricing parameters in relation to costs are dierent
in collusive and competitive markets, which can result in relatively higher profit margins in cartel sectors (Carrijo, 2019).
72
FERNANDES, Rosangela Aparecida Soares; JESUS JÚNIOR, Leonardo Bispo de; PINTO, Victor
Henrique Lana. Peering through the haze: detecting cartel signs in Brazil's LPG resale market.
Revista de Defesa da Concorrência, Brasília, v. 12, n. 2, p. 65-84, 2024.
https://doi.org/10.52896/rdc.v12i2.1400
a Price volatility model according to the approach in Bolotova, Connor and
Miller (2008)
There are various filtering methodologies in the literature that allow for the detection of
economic signs of a cartel. However, in the context of this study, we apply the filter introduced by
Bolotova, Connor, and Miller (2008) as it is adequate to test whether there was an increase in the
average price of LPG and a decrease in its variance during the cartel period, as stated in the AP
08700.003067/2009-67. Additionally, this method is suitable because it only requires the price series
of the product commercialized in the investigated production chain stage. Therefore, the method
proposed by Bolotova, Connor, and Miller (2008) enables us to conduct an investigation in the absence
of information on costs, such as the prices charged by distributors. In this section, we present the
price volatility model proposed in Bolotova, Connor, and Miller (2008).
The methodology applied by these authors consists of estimating an ARCH model and its
generalization, a Generalized Autoregressive Conditional Heteroskedasticity (GARCH), to test the
hypothesis that, during the cartel period, the average price is higher and its variance lower.
Formally, the ARCH model allows us to model the conditional variance of the series,
simultaneously, with the mean (Hamilton, 1994). Assuming that the conditional variance is not constant,
we estimated an AR(m) process using the square of the estimated residual, as shown in equation (1):
(1)
where v
t
is a new “white noise”, with E(v
t
)=0 and E(v
t
, v
s
)=λ
2
, for all t = s, and 0 otherwise.
Equation (1) is an ARCH process of order m, denoted by u
t
~ARCH(m). Thus, the linear projection
of the squared error of the prediction of Y
t
, from the previous q squared predicted errors, is defined by:
(2)
The GARCH model, proposed by Bollerslev (1996), denotes an extension of the ARCH model
and allows the use of a larger lag structure, aiming to capture impacts that possibly occurred in the
past. Furthermore, it can be used to describe volatility with fewer parameters than ARCH. Formally,
the GARCH variance equation is represented by:
(3)
where u
t
=√h
t
and v
t
is independent and identically distributed (iid) with zero mean and unitary
variance. The sucient condition of stationarity (regularity) requires k > 0, δi ≥ 0 for all i ≤ r, α j ≥ 0 for
all j ≤ m and ∑
r
t
δ
t
+∑
m
t=1
α
t
≤1. The calculation of the sequence of conditional variances {ht} for t = 1...T
requires its pre-sample values. Typically, they are calculated as a sample mean of the squared predicted
residuals for each pre-sample observation in the T-sequence (Bollerslev, 1996; Hamilton, 1994).
Bolotova, Connor and Miller (2008) used a specification of the volatility models, ARCH and
GARCH, which includes a cartel dummy variable in the mean and variance equations, as well as a
73
variable of an interaction between the lagged price and the cartel dummy. The sum of the coecient
of the cartel dummy variable of the average equation and the coecient of the interaction variable
quantifies the increase in the LPG resale price during the period in which the dealers maintained the
cartelized conduct.
In this article, we estimated an ARCH model, according to the econometric specification used
by Bolotova, Connor and Miller (2008). The cartel dummy variable equaled 1, for the cartel period from
January 2008 to March 2010, and zero otherwise.
For the mean equation, we expected the coecient of the cartel dummy variable to show a
positive and statistically significant coecient, suggesting that, during the cartel period, there was an
increase in the price of LPG in the municipalities analyzed. On the other hand, in the variance equation,
we expected a negative and statistically significant coecient, showing a decrease in price variability.
b Data
In this article, weekly data on average LPG prices
9
practiced in resale in the cities of Campina
Grande-PB, Caruaru-PE, Jaboatão dos Guararapes-PE, and Recife-PE, from May 2004 to May 2020 were
used. These data come from the database of the ANP - Price Survey website
The full period of analysis was chosen due to data availability. The time cut for the occurrence
of the cartel, from January 2008 to March 2010, was defined according to the information cited in the
documents of the AP 08700.003067/2009-67. The price volatility models as well as other tests were
estimated in Eviews 10.
5 RESULTS
The detection of cartels has been one of the most relevant and challenging tasks for antitrust
agencies around the world. Even with the advancement and success of leniency programs, some cartels
still go undetected. According to Galindo (2018), the illegality of cartels makes them highly confidential,
which makes it dicult to prove their existence. Due to the diculties faced, economic cartel filters
can be used as an alternative tool for detecting economic evidence of this anticompetitive conduct.
The limitations faced by antitrust agencies to spot and punish cartels also reflects, to a
certain extent, a research problem in the field of Industrial Economics. In view of this, researches
with applications for the detection of economic signs of a cartel have been gaining relevance in the
international and national literature. Silveira et al. (2021) observe that, in recent decades, problems
to detect cartels have led to a new relevant field of research.
In this context, this study aimed to try and identify the economic signs of cartel in the resale
of LPG in Campina Grande-PB, a market recently condemned by Cade, and in Caruaru-PE, Jaboatão dos
Guararapes-PE, and Recife-PE. These cities were mentioned previously in the AP 08700.003067/2009-67,
which was instituted to investigate the resale and distribution of LPG in the northeast region of Brazil.
9 A limitation of this research relates to the use of municipal data on average weekly LPG resale prices, released by
the ANP, which provide information that may include LPG prices practiced by resellers that, eventually, did not participate in
the collusion.
74
FERNANDES, Rosangela Aparecida Soares; JESUS JÚNIOR, Leonardo Bispo de; PINTO, Victor
Henrique Lana. Peering through the haze: detecting cartel signs in Brazil's LPG resale market.
Revista de Defesa da Concorrência, Brasília, v. 12, n. 2, p. 65-84, 2024.
https://doi.org/10.52896/rdc.v12i2.1400
In Brazil, the distribution and resale of LPG have been the object of investigations into the
formation of cartels by competition antitrust agencies, because of the number of complaints about
suspected practice of such a conduct. In recent years, there has been a significant increase in LPG
prices in several municipalities in the country. Specifically, with regard to the relevant geographic
markets under analysis, the upward trend in LPG resale prices from May 2004 to May 2020 can be
observed in graphs 1A, 3A, 5A and 7A (see appendix). It is noteworthy that the increase in the price of
this product is a topic of particular importance for society because of the implications that abusive
pricing practices have on social well-being. For this reason, it is important to analyze the behavior
pattern of the prices practiced in these markets to verify whether they are consistent with the one
that prevails in cartelized settings.
Finally, in the next section, we present the econometric results obtained by estimating the
price volatility model, which allowed testing the hypothesis raised in Bolotova, Connor and Miller
(2008). This hypothesis verifies whether the average LPG resale prices in the municipalities we analyze
increased and if their variances decreased during the cartel period. Additionally, the estimated model
also allows us to obtain a quantification of the increase in prices, in the markets where this pattern
of behavior was observed.
a Detection of economic signs of cartel based on the approach of Bolotova,
Connor and Miller (2008)
This section presents the results of applying the ARCH price volatility model with the
introduction of the cartel dummy variable in the mean and variance equations, as well as the insertion
of the interaction variable, the LPG lagged price and the cartel dummy variable, in the LPG reseller
market in the cities of Campina Grande-PB, Caruaru-PE, Jaboatão dos Guararapes-PE, and Recife-PE.
First, the Augmented Dickey-Fuller (ADF) unit root test was applied and the results showed
that the series of LPG resale prices, in the first dierence, were stationary, (see Table 1A in appendix).
To verify whether these series were likely to be modeled through the price volatility model, we
performed the ARCH test. The results suggested rejection of the null hypothesis; therefore, the series
presented statistical evidence of conditional variance (see Table 2A in appendix). To obtain consistent
estimates, we used the first dierence of the LPG resale price series.
Table 1 reports the results of the ARCH price volatility model estimated for the resale of LPG
in Campina Grande-PB, Caruaru-PE, Jaboatão dos Guararapes-PE, and Recife-PE from May 2004 to
March 2020.
75
Table 1 – Results of the ARCH price volatility model estimated for the mean and variance of resale
LPG prices in Campina Grande-PB, Caruaru-PE, Jaboatão dos Guararapes-PE, and Recife-PE
Campina Grande-PB Caruaru-PE Jaboatão dos
Guararapes-PE
Recife-PE
Mean equation
Cartel dummy 3.03648*** 4.91922*** 0.82085*** 0.63771**
Standard error 0.40114 1.10969 0.26873 0.30056
PR(-1)* Cartel dummy -0.08783*** -0.14648*** -0.02265*** -0.01714*
Standard error 0.01233 0.03416 0.00811 0.01001
AR(1) -0.31032*** 0.02001NS -0.59329*** -0.53279***
Standard error 0.04915 0.05996 0.02690 0.02530
Variance equation
Intercept 0.26119*** 0.94499*** 0.45413*** 0.62508***
Standard error 0.00245 0.00818 0.01168 0.00729
Residual (-1)2 0.29197*** -0.00605*** 0.90399*** 0.64957***
Standard error 0.03836 0.00001 0.08020 0.05749
Cartel dummy -0.12861*** -0.63224*** -0.11005*** -0.35540***
Standard error 0.01249 0.02048 0.03585 0.02387
Note: ***, **, and * indicates statistical significance at the 1%, 5%, and 10% level, respectively. NS denotes
statistically not dierent from zero.
Source: Research results.
The results showed that most estimated coecients were statistically significant and showed
signs consistent with the assumptions of the model of Bolotova, Connor and Miller (2008).
The cartel dummy variable, inserted in the average equations of the municipalities we
analyzed, was positive and statistically significant, suggesting that, from January 2008 to March 2010,
the average LPG resale price, during the cartel period, was relatively higher than than the average
of LPG prices, in the period in which it did not occur. The sum of the coecient of the cartel dummy
variable and the coecient of the interaction variable reflects the increase in the LPG price during the
cartel period. Therefore, increases in LPG resale prices were R$ 2.95 in Campina Grande-PB; R$ 4.77 in
Caruaru-PE; R$ 0.82 in Jaboatão dos Guararapes-PE, and R$ 0.62 in Recife-PE.
Regarding the variance equations, the estimated coecients for the cartel dummy were all
negative and statistically dierent from zero, showing that, during the cartel period, there was less
volatility in LPG prices. In other words, there was a greater alignment of LPG resale prices in all the
municipalities we evaluated.
Our results allow us to observe that during the cartel period the average LPG resale prices, in
the analyzed municipalities, increased and their variances decreased. Thus, the application of this filter
made it possible to obtain estimates of the behavior pattern of LPG resale prices, in these municipalities,
consistent with that prevailing in markets where stable cartels exist. Specifically, the results obtained
for Campina Grande-PB corroborate the eciency of applying the filter for analysis of LPG resale prices,
as this municipality has been recently condemned by Cade. In other words, the application of the filter
76
FERNANDES, Rosangela Aparecida Soares; JESUS JÚNIOR, Leonardo Bispo de; PINTO, Victor
Henrique Lana. Peering through the haze: detecting cartel signs in Brazil's LPG resale market.
Revista de Defesa da Concorrência, Brasília, v. 12, n. 2, p. 65-84, 2024.
https://doi.org/10.52896/rdc.v12i2.1400
to detect economic signs of a cartel was not subject to type II error (false negative).
According to the information reported in the AP 08700.003067/2009-67, the resellers Frazão
Distribuidora, Gás do Brasil, and Gás da Paraíba were condemned in the LPG resale market, as there
was evidence that the price increases practiced in the resale of LPG were due to the formation of
a cartel. Also, in the records of this AP, there are strong indications that the Sindirev coordinated
periodic meetings to exert influence over LPG resellers in Campina Grande-PB and region. The
LPG distributors also aected the adoption of uniform commercial conduct between competitors,
facilitating price fixing agreements and the exchange of commercially sensitive information. In
addition, the distributors were responsible for taking action against resellers who deviated from the
price alignment conduct, and for monitoring the behavior of other market agents (Cade, 2022).
In sum, the behavioral filter, used in this article, showed that there are economic signs of a
cartel in the cities of Campina Grande-PB, Caruaru-PE, Jaboatão dos Guararapes-PE, and Recife-PE.
The application of this method, for a market condemned by Cade, corroborates the model of Bolotova,
Connor and Miller (2008) for the analysis of economic signs of cartel in the resale of LPG. Moreover,
the filter detected a pattern of price behavior consistent with that prevailing in stable cartels, in cities
that were not condemned, but mentioned, in the aforementioned AP.
We expect that the results of this study can encourage and guide future analyses of this
nature, for other relevant geographic markets as well as for dierent segments of the petroleum
derivatives production chain, with similar characteristics to those of the LPG market. Also, our results
emphasize that the Brazilian antitrust agency remains attentive to the conduct of LPG resellers.
6 CONCLUDING REMARKS
The social welfare losses resulting from cartel behavior make detecting cartels one of the
most important tasks for antitrust agencies. Nonetheless, many times, it has also been one of its
biggest challenges. Consequently, because of the diculties faced, economic cartel filters emerge as
alternative tools to investigate the behavior pattern of companies that may be involved in a cartel.
In view of the pertinence of using filters for the investigation of economic evidence of cartel,
this article aimed to detect economic signs of cartel in the resale of LPG in Campina Grande-PB,
Caruaru-PE, Jaboatão dos Guararapes-PE, and Recife-PE, the first market recently condemned by Cade.
To achieve the aim of this study, we applied an ARCH price volatility model with a cartel
dummy variable in the mean and variance equations, as well as the interaction variable, the LPG
lagged price and the cartel dummy, in the LPG reseller market of the cities under analysis.
Our findings indicated that the coecients for the cartel dummy variable included in the
mean equations for all municipalities were both positive and statistically significant. This indicates
that from January 2008 to March 2010 the average LPG resale price was relatively higher in the cartel
period than during the non-cartel period. Regarding the increase in prices, during the cartel period,
we found that Caruaru-PE had the highest increase in average LPG prices, R$ 4.77; followed by Campina
Grande-PB, R$2.95; Jaboatão dos Guararapes-PE, R$0.82; and Recife-PE, R$ 0.62.
Our analysis of the variance equations revealed that the estimated coecients for the cartel
dummy variable were consistently negative and statistically significant. This suggests that during the
77
cartel period, there was less variability in LPG prices, resulting in greater alignment of resale prices
across all municipalities included in our study.
Therefore, we conclude from the application of the filter, that there are economic signs of
cartel in the analyzed cities. Our application of this method to a market that has been sanctioned by
Cade confirmed the economic indicators of cartel behavior in the LPG resale sector. This suggests that
the use of the filter to detect such behavior in this market is unlikely to result in type II error (false
negative). Furthermore, our analysis revealed a consistent pattern of pricing behavior in cities that
were not subject to sanctions but were included in the AP.
In short, given the harmful eects associated with cartel-like conduct, we expect that the
Brazilian antitrust agency remains attentive to the behavior of LPG resellers. We note that even in
cities where the number of participants is suciently high, the formation of a cartel can be guaranteed
by eective coordination, monitoring, and punishing mechanisms for misconduct. In addition, unions
and distributors can act to influence the formation and maintenance of the cartel.
Moreover, the findings of this study could serve as a basis for future research on similar
segments of the petroleum derivatives production chain in other relevant geographic markets,
building upon the characteristics observed in the LPG market.
REFERENCES
AALTIO, Aapo. Competition or collusion? Screening the Finnish road maintenance Market. 2019. Master’s
thesis (MSc in Finance and Economics) - Hanken School of Economics, Helsinki, 2019. Available at:
https://tinyurl.com/276pvwkr. Retrieved: May 2, 2023.
ABRANTES-METZ, Rosa M.; BAJARI, Patrick. Screen for Conspiracies and their Multiple Applications.
Competition Policy International, [s. l.], v. 6, n. 2, p. 129-144, 2010. Available at: https://tinyurl.com/
ynwfg75n. Retrieved: May 2, 2023.
ABRANTES-METZ, Rosa M.; BAJARI, Patrick; MURPHY, Joe. Antitrust Screening: Making Compliance
Programs Robust. SSRN, [s. l.], 2010. Available at: https://dx.doi.org/10.2139/ssrn.1648948. Retrieved:
May 2, 2023.
ABRANTES-METZ, Rosa M.; FROEB, Luke M.; GEWEKE, John; TAYLOR, Christopher T. A variance screen for
collusion. International Journal of Industrial Organization, [s. l], v. 24, n. 3, p. 467-486, 2006. Available
at: https://doi.org/10.1016/j.ijindorg.2005.10.003. Retrieved: Jul 20, 2023.
AGÊNCIA NACIONAL DE PETRÓLEO, GÁS NATURAL E BIOCOMBUSTÍVEIS (ANP). Painel dinâmico: Processo
cadastral de revendas de combustíveis e de GLP. Agência Nacional do Petróleo, Gás Natural e
Biocombustíveis: Rio de Janeiro, 2023. Available at: https://tinyurl.com/2d67eod7. Retrieved: Jul 21, 2023.
BOLLERSLEV, Tim. Generalize Autoregressive Conditional Heteroskedasticity
Journal of Econometrics, [s. l.], v. 31, n. 3, p. 307-327, 1986. Available at: https://doi.org/10.1016/0304-
4076(86)90063-1. Retrieved: Dec 9, 2022.
BOLOTOVA, Yuliya; CONNOR, John M.; MILLER, Douglas J. The impact of collusion on price behavior:
empirical results from two recent cases. International Journal of Industrial Organization, [s. l.], v. 26, n.
6, p. 1290-1307, 2008. Available at: https://doi.org/10.1016/j.ijindorg.2007.12.008. Retrieved: Jul 12, 2023.
78
FERNANDES, Rosangela Aparecida Soares; JESUS JÚNIOR, Leonardo Bispo de; PINTO, Victor
Henrique Lana. Peering through the haze: detecting cartel signs in Brazil's LPG resale market.
Revista de Defesa da Concorrência, Brasília, v. 12, n. 2, p. 65-84, 2024.
https://doi.org/10.52896/rdc.v12i2.1400
CARRIJO, Gustavo Dantas. Análise estatística dos preços de combustíveis para auxílio na detecção
de cartéis no setor de distribuição. 2019. Thesis (MBA in Governance and Regulation Control in
Infrastructure) - Escola Nacional de Administração Pública, Brasília, 2019. Available at: https://tinyurl.
com/237wg2l7. Retrieved: Jun 30, 2023.
CONNOR, John M. Collusion and price dispersion. Applied Economics Letters, [s. l.], v. 12, n. 6, p. 335-
338, 2005. Available at: https://doi.org/10.1080/13504850500068061. Retrieved: Jun 30, 2023.
CONNOR, John M. Price-Fixing Overcharges: Legal and Economic Evidence. Research in Law and
Economics, [s. l.], v. 22, p. 59-153, 2007. Available at: http://dx.doi.org/10.1016/S0193-5895(06)22004-9.
Retrieved: May 4, 2023.
CONSELHO ADMINISTRATIVO DE DEFESA ECONÔMICA (Cade). Cade condena cartel no mercado de
distribuição e revenda de gás de cozinha no Nordeste. Conselho Administrativo de Defesa Econômica,
Brasília, 19 out. 2022. Available at: https://tinyurl.com/28s4ymky. Retrieved on: Feb 7, 2023.
CUIABANO, Simone; LEANDRO, Tainá; OLIVEIRA, Glauco; BOGOSSIAN, Paula. Filtrando cartéis: a
contribuição da literatura econômica na identificação de comportamentos colusivos. Revista de
Defesa da Concorrência, Brasília, v. 2, n. 2, p. 43-63, 2014. Available at: https://tinyurl.com/2ctnnyza.
Retrieved: Dec 22, 2023.
COLOMER, Marcelo; VERNERSBACH, Aldren. Comportamento dos preços do GLP no Brasil: uma análise
nacional e regional. In: PINTO JUNIOR, Helder; COLOMER, Marcelo (ed.). Mercados de combustíveis e
GLP, questões de regulação setorial e de concorrência. Rio de Janeiro: Sindigás. 2022. p. 185-216.
EMPRESA DE PESQUISA ENERGÉTICA (EPE). Balanço Energético Nacional 2021. Rio de Janeiro: Empresa
de Pesquisa Energética, 2021. Available at: https://tinyurl.com/27kfz8uz. Retrieved: Feb 7, 2023.
FROEB, Luke; SIBLEY, David; DOANE, Michael; PINTO, Brijesh. Screening for collusion as a problem
of inference. In: BLAIR, Roger; SOKOL, Daniel (ed.). The Oxford Handbook of International Antitrust
Economics II. Oxford: Oxford University Press, 2015. p. 523-553.
GALINDO, Carmen García. Essays on cartels and competition policy. 2018. Doctoral dissertation (DSc in
Economics) - European University Institute, Florence, 2018. Availble at: https://tinyurl.com/2cpozo3a.
Retrieved: Feb 8, 2023.
HAMILTON, James. Time Series Analysis. New Jersey: Princeton University Press. 1994.
HARRINGTON JR., Joseph E. Behavioral screening and the detection of cartels. In: EU COMPETITION
LAW AND POLICY WORKSHOP, 2006, Florence. Proceedings […]. Florence: European University Institute,
2006. Available at: https://tinyurl.com/25otyfcg. Retrieved on: Aug 23, 2024.
HARRINGTON JR., Joseph E. Detecting cartels. In: BUCCIROSSI, Paolo (ed.). Handbook in Antitrust
Economics. Cambridge: MIT Press. 2008. p. 213-258.
HARRINGTON JR., Joseph E.; CHEN, Joe. Cartel pricing dynamics with cost variability and endogenous
buyer detection. International Journal of Industrial Organization, [s. l.], v. 24, n. 6, p. 1185-1212, 2006.
Available at: http://dx.doi.org/10.1016/j.ijindorg.2006.04.012. Retrieved on: Jan 25, 2023.
HOVENKAMP, Herbert J. Quantification of Harm in Private Antitrust Actions in the United States.
Philadelphia: University of Pennsylvania, 2011. Available at: https://tinyurl.com/2de9ljhn. Retrieved
on: Jan 17, 2023.
79
LODI, Carlos Felipe; BICALHO, Lucia Navegantes. Padrão de concorrência e estruturas dos mercados
regionais de GLP. In: PINTO JUNIOR, Helder; COLOMER, Marcelo (org.). Mercados de combustíveis e GLP,
questões de regulação setorial e de concorrência. Rio de Janeiro: Sindigás, 2022. p. 114-126. Available
at: https://tinyurl.com/2492b9gy. Retrieved on: Jan 17, 2023.
MALAN, Fabiane Fernandes Hanones; RESENDE, Guilherme Mendes. Estimação de sobrepreço em
cartéis: o caso do cartel de combustíveis na região metropolitana de Belo Horizonte-MG: Documento
de Trabalho nº 007/2022. Brasília: Conselho Administrativo de Defesa Econômica, 2022. Available at:
https://tinyurl.com/28jwna6z. Retrieved on: Jan 16, 2023.
ORGANISATION FOR ECONOMIC CO-OPERATION AND DEVELOPMENT (OECD). Ex-ocio cartel
investigations and the use of screens to detect cartels. Paris: Organisation for Economic Co-operation
and Development, 2013. (Series Roundtables on Competition Policy, 147). Available at: https://tinyurl.
com/2aqdhbjk. Retrieved on: Feb 10, 2023.
SILVEIRA, Douglas; VASCONCELOS, Silvinha; BOGOSSIAN, Paula; NETO, Joaquim. Cartel Screening in the
Brazilian Fuel Retail Market. Economia, Niterói, v. 22, n. 1, p. 53-70, 2021. Available at: https://doi.
org/10.1016/j.econ.2021.01.001. Retrieved on: Jan 10, 2023.
SINDIGÁS. Panorama do Setor de GLP em Movimento. Rio de Janeiro: Sindigás, 2021. Available at:
https://tinyurl.com/2ct64pv9. Retrieved on: Aug 22, 2023.
APPENDIX
Campina Grande - PB
Graph 1A – LPG price
80
FERNANDES, Rosangela Aparecida Soares; JESUS JÚNIOR, Leonardo Bispo de; PINTO, Victor
Henrique Lana. Peering through the haze: detecting cartel signs in Brazil's LPG resale market.
Revista de Defesa da Concorrência, Brasília, v. 12, n. 2, p. 65-84, 2024.
https://doi.org/10.52896/rdc.v12i2.1400
Graph 2A – First dierence of LPG price
Caruaru – PE
Graph 3A – LPG price
81
Graph 4A – First dierence of LPG price
Jaboatão dos Guararapes – PE
Graph 5A – LPG price
82
FERNANDES, Rosangela Aparecida Soares; JESUS JÚNIOR, Leonardo Bispo de; PINTO, Victor
Henrique Lana. Peering through the haze: detecting cartel signs in Brazil's LPG resale market.
Revista de Defesa da Concorrência, Brasília, v. 12, n. 2, p. 65-84, 2024.
https://doi.org/10.52896/rdc.v12i2.1400
Graph 6A – First dierence of LPG price
Recife - PE
Graph 7A – LPG price
83
Graph 8A – First dierence of LPG price
Source: Research results.
Table 1A – Unit root test (ADF) in the series of LPG prices charged by resellers in Campina Grande-
PB, Caruaru-PE, Jaboatão dos Guararapes-PE, and Recife-PE
Variable Test statistic Probability
PRCampina Grande -1.832898 0.6879
PRCampina Grande -18.49327 0.0000
PRCaruaru -1.789599 0.7092
PRCaruaru -24.75071 0.0000
PRJaboatão dos Guararapes -2.481758 0.3372
PRJaboatão dos Guararapes -32.75392 0.0000
PRRecife -2.468093 0.3441
PRRecife -37.04887 0.0000
Source: Research results.
84
FERNANDES, Rosangela Aparecida Soares; JESUS JÚNIOR, Leonardo Bispo de; PINTO, Victor
Henrique Lana. Peering through the haze: detecting cartel signs in Brazil's LPG resale market.
Revista de Defesa da Concorrência, Brasília, v. 12, n. 2, p. 65-84, 2024.
https://doi.org/10.52896/rdc.v12i2.1400
Table 2A – Heteroskedasticity test (ARCH eect) on the first dierence of the LPG price series
practiced by resellers in Campina Grande-PB, Caruaru-PE, Jaboatão dos Guararapes-PE, and Recife-
PE
Variable Test statistic Probability
PRCampina Grande 5.270288 0.0053
PRCaruaru 8.934143 0.0001
PRJaboatão dos Guararapes 3.677579 0.0257
PRRecife 2.382070 0.0930
Source: Research results.
Table 3A – Serial correlation test of the residuals on the first dierence of the LPG price series
practiced by resellers in Campina Grande-PB, Caruaru-PE, Jaboatão dos Guararapes-PE, and Recife-
PE
Variable Test statistic Probability
PRCampina Grande 16.1710 0.0000
PRCaruaru 16.9331 0.0000
PRJaboatão dos Guararapes 9.2465 0.0001
PRRecife 28.0926 0.0000
Source: Research results.
