Ex post-merger evaluation: evidence from the Brazilian stock exchange

Avaliação ex post de fusões: evidências da bolsa de valores brasileira

Abstracts

ABSTRACT  This paper aims to analyse the (anti)competitive effects of a merger reviewed by the Brazilian competition Authority (CADE): a transaction between BM&FBOVESPA S.A. (BVMF) and CETIP S.A., originating B3 S.A. in 2017. The deal involved the industries of over-the-counter trade and stock, commodities, and futures exchanges. By applying the difference-in-differences and augmented synthetic control methods, this paper reports an ex-post evaluation of the BVMF-CETIP merger. The results revealed a reduction in the average trading fee of B3 in the stock exchange market after the merger. Moreover, tests of robustness revealed similar results; in addition, some estimates showed a coefficient for the merger effects statistically equal to zero. Therefore, we found no adverse competitive effects (relating to a rise in trading fees) resulting from the BVMF-CETIP merger. Considering the literature on the topic, this study seems to have filled an academic gap by conducting the empirical analysis of a merger in the stock market to assess its effect.

KEYWORDS:
Competition policy; Quasi-experiments; Stock exchanges; Ex post evaluation; Merger

RESUMO  Este artigo busca analisar os efeitos (anti)competitivos de um ato de concentração julgado pelo CADE: uma operação que resultou da união das atividades da BM&FBOVESPA S.A. e CETIP S.A, dando origem à B3 S.A., em 2017. O negócio envolveu os segmentos de administração de mercados organizados de balcão, bolsa de valores e bolsa de mercadorias e futuros. Utilizando os métodos de diferenças em diferenças e controle sintético aumentado, o estudo conduz uma avaliação ex post da fusão entre BVMF-CETIP. Os resultados revelaram decréscimo na tarifa média de negociação praticada pela B3 após a referida operação. Ademais, os testes de robustez mostraram resultados similares e, em algumas estimações, o coeficiente que demonstra o efeito da fusão foi estatisticamente igual a zero. Portanto, não foram verificados efeitos concorrenciais adversos (relativos a aumentos no valor das tarifas de negociação) em decorrência do ato de concentração. Considerando a literatura sobre o tema, este estudo preenche uma lacuna acadêmica ao conduzir a análise empírica de uma fusão no mercado de bolsa de valores a fim de avaliar o seu efeito.

PALAVRAS-CHAVE:
Defesa da concorrência; Quase-experimentos; Bolsas de valores; Avaliação ex post; Fusão

1. INTRODUCTION

Evidence-based assessment has been increasingly used in picking and formulating powerful and efficient public policies, and the ex-post assessment of these policies has become a valuable tool towards this goal. Since the same applies to competition policies, the OECD recommends antitrust authorities analyse the results of their interventions (Organisation for Economic Co-operation and Development, 2016) – including merger reviews, which, in Brazil, are conducted ex-ante by the Brazilian competition Authority (the Administrative Council for Economic Defense or CADE).

Against this backdrop, this study focuses on the (anti)competitive effects of a merger reviewed by CADE: a transaction between BM&FBOVESPA S.A. (BVMF) and CETIP S.A., originating B3 S.A.¹, on 22 March 2017. The deal involved the industries of over-the-counter trade and stock, commodities, and futures exchanges.

By applying the difference-in-differences and synthetic control (in its Ridge augmented form) methods, this paper aims to analyse the ex-post impact of the BVMF-CETIP merger. Based on potential competition concerns, CADE cleared the deal subject to a merger control agreement that intended to, amongst other goals, facilitate competitors’ access to the post-trade services of the newly merged company – especially central depository services, regarded as an essential facility.

Additionally, we created a database of quarterly data covering from 2012 to 2021 with fees levied by the Brazilian stock exchange, affected by the decision of the CADE. The database also includes data from a comparison group composed of stock exchanges from around the world.

Aguzzoni et al. (2016) claim that retrospective merger analyses are relevant to validate the simulation models used in the ex-ante review of a merger. In addition, Farrell, Pautler and Vita (2009) stress that retrospective analyses should not choose mergers at random but, rather, pick cases “at the margin”, i.e., that could produce anticompetitive outcomes but were not challenged. Analysing these specific cases is particularly relevant as it is impossible to assess the effects of a representative sample of all transactions because antitrust authorities will block those believed to be anticompetitive before their consummation (Ashenfelter; Hosken; Weinberg, 2009).

The BVMF-CETIP merger fulfils these requirements. The commissioners who heard the merger at CADE cleared it, albeit diverging about the imposed remedies². Furthermore, throughout the proceedings, there was continuing preoccupation with the effects of the deal on potential competition in the exchange sector: not a trivial concern for a sector that has undergone substantial transformations worldwide over the last few decades, with exchanges turning into for-profit businesses, numerous entering platforms, and a large number of mergers between them (Cantillone; Yin, 2011). Moreover, the transaction at issue involved two monopolists (or quasi-monopolists) in their respective markets, thus rendering it impossible to employ traditional models to gauge merger effects, making an ex-post assessment even more important.

Hence, by selecting and retroactively examining a suitable merger (as the one at issue), we believe that we can add significant academic knowledge of finance and industrial organisation. In other words, this paper seeks to contribute to the academy and society alike by combining theory and practice to cover a gap in similar research into stock market mergers.

Next, we describe the organization of the paper. The next section brings a review of the literature used for this study, covering relevant economic matters on stock markets. In turn, the third section focuses on research methodology and is divided into three subsections: some considerations on the deal at stake and functioning of the stock market; the difference-in-differences (DID) and synthetic control techniques as employed herein; and a detailed description of the collected data. Finally, the fourth section examines and discusses the results and robustness analysis, while our conclusions are introduced in the fifth section.

2. LITERATURE REVIEW

Nevo and Whinston (2010) argue that structural analyses and estimates of treatment effects are complementary, not replaceable, and may be applied to different situations. The authors claim that retrospective estimates of treatment effects may be used to predict the impact of other mergers, although limited to the particularities of the earlier merger. Nevertheless, structural estimation also imposes some limitations since it relies on certain instrumental variables and the models cannot measure the effects of long-term investments, research and development, and new entrants. For these reasons, retrospective assessments may be valuable in validating structural models that simulate mergers.

In an essay reviewing research conducted by the Federal Trade Commission (FTC) and others into the ex-post assessment of the competitive effect of mergers, Farrell, Pautler and Vita (2009) found the simplest approach to be surveying market players affected by the merger and investigating its impact on their prices and product output. This method, however, draws criticism for the subjective nature of its resulting evidence and the lack of rigour in estimating what would have happened had the merger not taken place. For this reason, we searched for methods that could allow for more accurate prices and counterfactuals, in the same manner as other studies that employed the DID approach for a retrospective examination of the mergers in the hospital sector.

Ashenfelter, Hosken and Weinberg (2009) also analysed the extant literature on the matter and stressed the importance of ex-post-merger assessment in founding antitrust authorities’ future decisions. Kwoka Junior (2013) also conduct a meta-analysis of existing economic literature on merger outcomes, focusing on price effects post-merger in the United States using a database of sixty merger retrospectives. The evidence suggests that “stronger policy measures – outright opposition or structural remedies instead of conduct/conditions approaches – may be warranted in cases where they are not presently employed” (Kwoka Junior, 2013, p. 644).

Meanwhile, Ashenfelter and Hosken (2008) employed a DID estimation to examine five mergers in the United States deemed complex by the antitrust authority³. Their study used two control groups and private label products believed to share multiple features with those of the affected brands (albeit being distant substitutes). Their results revealed a price rise in four out of the five analysed mergers. Avoiding expressing criticism over the enforcer’s decisions, the authors posited that a comprehensive assessment of the appropriate level of merger monitoring was beyond the scope of their study.

Liebersohn (2024) used a difference-in-differences strategy to assess bank antitrust rules concerning mandatory divestitureand found evidence that antitrust enforcement in bank mergers increases deposit rates and mortgage lending, but has no effect on small business credit due to relationship-based lending frictions. In turn, Ma et al. (2023) also used a quasi-experimental approach to conduct both analytical modeling and empirical analyses to examine whether domestic mergers can improve the competitiveness of airlines in the international market. By employing a DID framework, estimations based on the merger of China Eastern and Shanghai Airlines found evidence of enhanced international competitiveness through cost-saving and service quality.

For their part, Aguzzoni et al. (2018) assessed the effects of two mobile telecommunication mergers on prices through an approach that is similar to the one adopted in this paper: a DID estimation comparing price evolution across markets located in different countries. Yet another similarity is the complexity of the pricing in the mobile telecommunication market, which requires pre-paid tariffs, post-paid tariffs, and subsidies to be taken into consideration.

Research using a similar methodology has also been carried out in Brazil. Severino, Resende and Bispo (2019) investigated the effects of a merger between Sadia and Perdigão, two companies operating in the frozen food industry and found no price rise in the analysed products.

Severino, Resende and Lima (2021) adopted DID estimators to examine two mergers in the Brazilian airline sector: the Azul-Trip and the Gol-Webjet. Both transactions showed no anticompetitive effects, thus corroborating the decision of the Brazilian authority on these cases. Meanwhile, when assessing the Gol-Webjet case through the potential competition’s perspective (by focusing on specific airports linking cities where the two companies operated, but not through the same routes), Franke and Resende (2024) found evidence of price rises. These findings are particularly relevant since the BVMF-CETIP merger also involved a matter of potential competition.

To the best of our knowledge, there are no reports of ex-post analyses of mergers between companies operating in the stock and over-the-counter markets that ascertain the impact on price levels through treatment effect methods. Nonetheless, there is substantial literature on the debate over the existence of trade-offs between market fragmentation (with or without competition)⁴ and liquidity. For instance, Cantillone and Yin (2011) contextualise this debate, clarifying that some people argue in favour of monopolies of trading infrastructure providers due to their economies of scale and network externalities, which bring about more liquidity (and, hence, consumer welfare for players of that market), a higher volume of transactions, and lower transaction costs. At the same time, competition leads to pressure for lower trading fees. In this context, Pagano (1989) believes volume and liquidity are related to the level of fragmentation/concentration of markets. Markets with fewer transactions (thus combined with low asset liquidity) can only make large transactions by charging traders high prices. Similarly, several other studies analyse the welfare implications of fragmentation in capital markets (Cespa; Vives, 2018; Bernales et al., 2018; Babus; Parlatore, 2021; Chao; Yao; Ye, 2019; Santos; Scheinkman, 2001; Bernstein et al., 2018).

In Brazil, the Securities and Exchange Commission (CVM) ordered a study to assess the outcomes of a hypothetical change in the competitive structure of the trade and post-trade service marketsby comparing it to a regulatory approach that included preserving the status quo, facilitating market entry, and monitoring trade and post-trade fees (Oxera, 2012). Drawing on international experiences and surveys on the cost characteristics of the main players in the country, the study estimated both the explicit and implicit outcomes of a drop in trade and post-trade prices, amongst other factors, in counterfactual scenarios where competition increased in the trade and post-trade markets only. The findings are favourable to investors, showing a potential price reduction for trade and post-trade services. However, a significant rise in regulatory costs was likely to occur, which could be passed on to investors.

Considering the above-mentioned literature, this study aims to fill an academic gap through the empirical analysis of a merger in the stock market, specifically focusing on assessing its impact on trading fees. Therefore, we expect to contribute to research on the scope of capital market.

3. METHOD AND DATA

3.1. CONSIDERATIONS ON THE ANALYSED MERGER

In order to clarify some of our methodological choices, we should first consider some specificities of the BVMF-CETIP merger that will help us understand the markets involved in this transaction and its most sensitive points from the perspective of CADE (Conselho Administrativo de Defesa Econômica, 2016). The organised trading structures of the Brazilian capital market are the stock, commodities, and futures exchanges (which the BVMF monopolises) and the over-the-counter market, where CETIP held a dominant position (and a monopoly of some types of securities). Unlike its over-the-counter counterpart, exchange trading has high liquidity, standardised contracts, low flexibility, high infrastructure costs, and a mandatory central intermediary (not allowing for bilateral trading).

Exchange and over-the-counter trading encompass three different vertical levels of services: a) trading, b) clearing and settlement, and c) trade registration (for over-the-counter trading) and central securities depository (CSD). The latter level refers to post-trade services. Clearing means ascertaining the net position of each market participant (credits minus debits), whereas settlement is the processing of the transaction and discharging obligations. Generally, CSD services, mandatory in exchanges, as well as for some types of OTC assets, consist of safekeeping and controlling the ownership of securities and activities that affect them under a fiduciary ownership arrangement. Finally, trade registration is a simpler modality of control that involves no fiduciary transfer.

In its analysis of the applicants’ horizontal integration, the Brazilian enforcer found their operations only overlapped in markets where OTC trade was not intermediated by a central counterparty (CCP), although they did not effectively compete since the BVMF only had a tiny share of these markets⁵. Nevertheless, in assessing the likelihood of an exercise of market power, CADE identified regulatory barriers and substantial economies of scale and scope in the applicants’ industries (Conselho Administrativo de Defesa Econômica, 2016). Moreover, the authority deemed the verticalization of the two firms in all industries where they operated could constitute a significant entry barrier, especially as to CSD services, which take on the characteristics of an essential facility. The Brazilian authority also examined competitors’ attempts to enter the stock exchange industry, which have failed due to their struggle to access the post-trade services of BVMF (Conselho Administrativo de Defesa Econômica, 2016).

These issues raised concerns that the monopolists could hinder entry by refusing to provide post-trading, implementing an unaffordable pricing policy, or squeezing margins in the industry where the entrant operated. The Brazilian enforcer received notification of this merger on 28 June 2016 and conditionally cleared it on 22 March 2017, subject to a merger control agreement intended to facilitate competitors' access to the post-trade services of BVMF. The agreement obliged the BVMF to offer CSD services under fair, transparent, and non-discriminatory conditions, which could be contested by interested parties via an arbitration system. It also introduced corporate governance mechanisms, such as having a minimum number of the applicants’ clients in the board of directors of BVMF, as well as requiring a special quorum for pricing deliberations, which allowed interested parties to veto changes with a minority of votes.

Hence, the antitrust authority focused on addressing entry barriers, particularly those related to the supply of post-trading to entrant firms. The authority also noticed entry attempts to the stock exchange market, with competitors struggling to access the post-trade services of BVMF. Since the stock market is where the applicants would face competitive pressure with successful entries, we focused on the effects of the BVMF-CETIP merger on this market.

3.2. DIFFERENCE-IN-DIFFERENCES

Amongst the tools used in ex-post-merger assessments, it is worth highlighting natural experiments (quasi-experiments), in which researchers look at the outcomes of interventions in a non-randomised group (the treatment group) and compare them with those of a group without intervention (the control group). In face of data for the two groups before and after the intervention, researchers usually adopt a method called difference-in-differences (Meyer, 1995).

The method is especially useful for comparing the difference in the applicants’ price levels before and after an intervention with the price levels of untreated players, as seen in the literature review above. This paper analysed the industry of stock exchange trade, therefore not covering commodities and futures exchanges, nor the post-trade level.

We considered that the intervention began in the second quarter of 2017 (2Q2017) since the merger was announced on 22 March 2017 and our price database uses quarterly data. The control group employed data published by stock exchanges that operate in financial centres unaffected by the merger at issue. Prices were determined by the average trading fee levied on the transactions that occurred in that quarter, as detailed in the following subsection. Hence, we estimate Equation 1 below:

$$T_{it}=\alpha +{\beta }_1{X}_{it}+{\delta }_{1}B{3}_t+{\delta }_2{T}_i+{\delta }_3{T}_{i}B{3}_t+{\epsilon }_{it}$$ (1)

Where T_it is a dependent variable that takes the value of the average fee for each examined exchange (i) in that quarter (t); X_it is the matrix of control covariates⁶; B3_t is a dummy variable that takes the value of 1 for the Brazilian stock exchange (the BVMF or the B3, depending on the analysed period) and 0 for the exchanges used as control group; T_i is a dummy variable that is assigned the value of 1 after CADE decides on the merger (2Q2017) and 0 before the decision; ε_it is the error term. δ₃ represents the parameter of interest that informs the merger effects on prices (see the next subsection for further details on this variable). Regressions were calculated with robust standard errors to circumvent heteroscedasticity issues.

3.3. SYNTHETIC CONTROL

The Synthetic Control Method (SCM) is another treatment-effect approach commonly used in ex-post-merger assessments. This method builds a control unit as a weighted combination of untreated (“donor”) units, selecting weights such that the synthetic unit closely replicates the trajectory of the pre-treatment outcomes of the treated unit. This approach is particularly suitable when only a few or even a single unit is exposed to intervention. In its canonical formulation (Abadie; Gardeazabal, 2003; Abadie; Diamond; Hainmueller, 2010), the method creates a vector of non‑negative weights that sum up to one and seeks to minimize the discrepancy between the treated and untreated units. The training data used to obtain the weights vector precede the intervention.

In the standard notation, if we have J + 1 units observed over T periods, then, unit j = 1 is exposed to the treatment from period T_{0 + 1} onward, and the remaining units j = 2, …, J + 1 form the donor pool. For each unit j and period t, the outcome Y_jt is observed. The causal effect of interest for the treated unit, for t > T₀, is given by Equation 2, as follows:

$\widehat{\tau }=Y_{1t}^I-Y_{1t}^N$ (2)

where $Y_{1t}^I$ is the realized outcome under intervention, and $Y_{1t}^N$ is the untreated counterfactual, which is unobserved. The synthetic control estimate $Y_{1t}^N$ is given by a convex combination of the donors, as formalized in Equation 3:

$Y_{1t}^N={{\displaystyle \sum }}_{j=2}^{J+1}{\omega }_j{Y}_{jt},{\omega }_j\ge \mathrm{0,}{{\displaystyle \sum }}_{j=2}^{J+1}{\omega }_j=1$ (3)

The weights $W$=($\omega$₂, …, _$\omega$j+1) are chosen such that the pre‑treatment characteristics of the treated unit are reproduced by the synthetic unit. Let X1 denote the vector of predictors for the treated unit and X0 the corresponding matrix for the donors. We seek aWthat minimizes the distance betweenX₁andX₀W, given by Equation 4 (Abadie; Diamond; Hainmueller, 2010).

$$\left|\left|{\text{X}}_{\text{1}}{\text{- X}}_{\text{0}}\text{W}\right|\right|\text{V =}\sqrt{{\text{(X}}_{\text{1}}{\text{-X}}_{\text{0}}{\text{W)'V(X}}_{\text{1}}{\text{-X}}_{\text{0}}\text{W)}}$$ (4)

where V is a diagonal matrix of weights that determine the relative importance of each predictor when measuring the discrepancy between X₁ and $X_{0}W$.

Abadie (2021) mention several extensions to the original SCM, including the Ridge variant of the Augmented Synthetic Control Method (ASCM), proposed by Ben‑Michael, Feller and Rothstein (2021). The standard synthetic control approach recommends achieving a good fit between the treated unit and its synthetic counterpart. To assist in situations where the pre‑treatment fit is poor, the Ridge ASCM applies a bias correction to the weight vector obtained from the SCM by means of a Ridge regression. Negative weights and weights summing to more than one are allowed (unlike the restrictions in Equation 3) and the degree of extrapolation is governed by the hyper‑parameter 𝜆_ridge. When the classical synthetic control already exhibits a good pre‑treatment fit, the ASCM draws near the conventional SCM estimator. Moreover, by allowing more flexibility in constructing the synthetic control, ASCM can also help mitigate concerns arising from violations of the parallel trends assumption (Ben‑Michael; Feller; Rothstein, 2021).

3.4. DATABASE DESCRIPTION

The most challenging aspect of adopting treatment effect techniques in retrospective merger analyses is the lack of price information available. There are many problems involved in simply adopting the nominal trading fees usually available on exchanges websites: they only provide static data, not time series; fee models are often intricate and include criteria that vary with time and across financial centres, hampering comparisons with other exchanges or the same exchange over time; and discount policies are equally complicated, failing to be captured by comparisons of nominal values.

All these factors point to a need for an alternative manner of measuring actually charged prices. To do so, one can measure “the unit cost for the trading and post-trading services according to the revenues (divided by the number or value of transactions) of the service providers” (Oxera, 2012). Measuring prices with the revenues earned via a specific activity is not unprecedented in ex-post-merger analyses that adopt the DID method, as reported by Ashenfelter and Hosken (2008, p. 16), who use the weekly revenue and sold units to calculate average prices.

In our study, we created a database from quarterly published financial statements and investor presentations published every quarter by stock exchanges on their investor relations websites. Data from exchange trade revenue per quarter is divided by the number of transactions made in the same quarter (which is available on the World Federation of Exchanges website) to determine the average trade rate of that stock exchange.

This procedure has the additional benefit of avoiding foreign exchange issues that could arise from comparing the prices of exchanges from multiple countries. Indeed, using a parameter expressed by a currency would require conversions for comparison purposes. In addition to implying additional data manipulation, this would bring about issues inherent to currency fluctuations, which could change prices due to many factors that cannot be controlled in regressions (e.g., imbalances in the balance of payments, government decisions on the national economic policy, regional geopolitical matters, among others).

The method for data collection employed in this study requires that exchanges publish their financial results detailing the revenue earned specifically from trade in the stock market (shares). This is not always the case since companies often operate in several markets and do not detail the source of their revenues clearly. Sometimes the available data refers to a very short period, rendering a comparative analysis impossible, or the financial information is displayed on an aggregated basis for periods longer than a quarter. From the thirty exchanges considered in this study, including the BVMF-B3, we could collect data from four companies to use as a control group: the Mexican Stock Exchange (BMV), the Warsaw Stock Exchange (GPW), the Japan Exchange Group (JPX), and the Intercontinental Exchange (ICE, which operates the New York Stock Exchange, or NYSE)⁷.

Since the period covered by each exchange varied, our time series started in 2008 for the BVMF-B3, 2011 for the BMV, 2010 for the GPW, and 2014 for the JPX and ICE. For all of them, the time series covered up to the last quarter of 2021. Figure 1 shows the average fee from 2012 (when available) to 2021, the period used for our main estimations.

Figure 1
– Exchange trading fees (2012-2021). Note: the Brazilian stock exchange (BVMF-B3), the Mexican Stock Exchange (BMV), the Warsaw Stock Exchange (GPW), the Japan Exchange Group (JPX), and the Intercontinental Exchange (ICE, which operates the New York Stock Exchange, or NYSE). Source: elaborated by the authors.

To obtain the quarterly GDP per capita in US dollars, we used the quarterly exchange rate and GDP data from the OECD database and population data from the World Bank⁸. The average number of companies listed in each exchange and the transactions carried out in the quarter were collected from the database of the World Federation of Exchanges. Below, Table 1 shows descriptive statistics for the average fee and control variables from 2012 (when available) to 2021, the period considered for our main estimations.

Table 1
– Descriptive statistics for exchange trading fees

	Pre-intervention period		Post-intervention period
Average fee
BVMF-B3	0.0104%	(0.0022%)	0.0068%	(0.0014%)
BMV	0.0399%	(0.0055%)	0.0509%	(0.0075%)
GPW	0.0504%	(0.0029%)	0.0461%	(0.0017%)
ICE	0.0019%	(0.0001%)	0.0015%	(0.0002%)
JPX	0.0065%	(0.0004%)	0.0052%	(0.0008%)
N. of transactions
BVMF-B3	52,702,655.71	(8,486,042.20)	136,834,133.16	(64,133,081.82)
BMV	10,666,891.90	(4,316,662.63)	20,414,712.11	(3,048,711.69)
GPW	3,931,170.00	(787,137.64)	6,749,188.42	(2,508,403.66)
ICE	424,223,570.00	(55,700,040.20)	570,752,014.74	(182,914,022.93)
JPX	190,314,872.31	(27,524,579.94)	221,342,895.26	(23,352,873.28)
GDP per capita
BVMF-B3	2,715.89	(445.58)	2,103.24	(294.39)
BMV	2,495.56	(215.21)	2,394.28	(196.56)
GPW	3,297.49	(189.78)	3,982.88	(314.06)
ICE	14,190.89	(349.83)	16,078.10	(850.40)
JPX	9,423.31	(564.71)	9,951.55	(261.37)
Listed firms
BVMF-B3	361.83	(6.76)	344.56	(16.15)
BMV	139.08	(3.96)	145.37	(1.36)
GPW	882.83	(29.40)	840.93	(32.65)
ICE	2,397.41	(55.57)	2,267.46	(85.78)
JPX	3,487.49	(41.62)	3,679.42	(71.09)

• Note: “average fee” refers to the average trading fee by quarter. “GDP per capita”, “listed firms”, and “N. of transactions” are control variables for, respectively, the GDP per capita by quarter in US dollars, the number of transactions by quarter, and the average number of listed companies by quarter. Standard deviation in parentheses.
• Source: elaborated by the authors.

4. RESULTS

4.1. DIFFERENCE-IN-DIFFERENCE RESULTS

It is worth highlighting once again the relevance of having a comparable control group when using the DID method. There are some common characteristics. The BMV, GPW, and JPX work in similar competitive environments to the Brazilian stock exchange, operating in a scenario of domestic monopoly or quasi-monopoly⁹. They are also similar in terms of vertical integration, which CADE deemed relevant in this sector, characterised by entry barriers, as they operate in the trade and post-trade stock exchange industries like the BVMF-B3.

The same cannot be said about ICE, the firm that runs the New York Stock Exchange, as the local stock exchange industry has different regulatory framework and level of fragmentation from the Brazilian exchange. The company faces a fiercer competition scenario (as described even in its investor reports), with more domestic rivals and exposure to international competitors. Finally, it also differs from the BVMF-B3 and other exchanges in its level of vertical integration since ICE does not offer the same post-trade services in the US stock exchange market (namely, settlement and CSD). That being said, the differences above call for the ICE removal from the group of exchanges that composed the control group in the main estimates.

Furthermore, to improve comparability, we used observable characteristics as controls in the DID regressions. As in other empirical studies that examined the stock market (such as Bernstein; Lerner; Dev, 2018), we controlled for macroeconomic characteristics and those specific to the many exchanges analysed: the variation of the GDP per capita by quarter in US dollars, the number of transactions by quarter, and the average number of listed companies by quarter¹⁰. The Table 2 below shows estimates from 2012 to 2021, excluding data from the US exchange (robust standard errors in parentheses).

Table 2
– Estimates for 2012 to 2021, excluding ICE-NYSE

	(1)	(2)	(3)	(4)
VARIABLES	Average fee	Average fee	Average fee	Average fee
Treatment/control	-0.02566***	-0.03583***	-0.03579***	-0.03049***
	(0.002)	(0.002)	(0.002)	(0.002)
Pre-/post-treatment	-0.00198	0.00395***	0.00404***	0.00532***
	(0.004)	(0.001)	(0.001)	(0.001)
DID impact	-0.00157	-0.01110***	-0.01140***	-0.00457**
	(0.004)	(0.002)	(0.002)	(0.002)
GDPPC		-0.00001***	-0.00001***	-0.00001***
		(0.000)	(0.000)	(0.000)
Listed firms			0.00000	0.00001***
			(0.000)	(0.000)
N. of transactions				-0.00000***
				(0.000)
Constant	0.03603***	0.06223***	0.06286***	0.06149***
	(0.002)	(0.002)	(0.003)	(0.002)
Observations	152	152	152	152
R2	0.329	0.911	0.911	0.936

• Note: “average fee” is the dependent variable, expressed in percentage terms. “Treatment/control” is a dummy variable that takes the value of 1 for the Brazilian stock exchange and 0 for the exchanges used as the control group. “Pre-/post treatment” is a dummy variable that is assigned the value of 1 after CADE decides on the merger (2Q2017) and 0 before the decision. “DID impact” is a variable given by the interaction between the variables of groups and periods (the DID), and its coefficient informs the impact of the merger on trading fees. “PIBPC”, “listed firms”, and “N. of transactions” are control variables for, respectively, the GDP per capita by quarter in US dollars, the number of transactions by quarter, and the average number of listed companies by quarter. Robust standard errors in parentheses.
• ^***
  p<0.01, and ** p<0.05.
• Source: elaborated by the authors.

As above-mentioned, by employing the DID approach and using all the controls, the parameter of interest reveals a decrease of 0.00457% in the average fee of B3. We found a statistical significance of 5% for these values. Hence, we found no adverse competitive effects (rise in trading fees) resulting from the BVMF-CETIP merger.

In order to provide the results outlined above with robustness, additional estimations were carried out based on different groups of exchanges and different periods. Table A1 of the Appendix A lists the results found for the parameter of interest (δ₃ of Equation 1 above). For example, the Table 3 below shows estimates from 2012 to 2021, excluding data from the US and the Mexican exchanges.

Table 3
– Estimates for 2012 to 2021, excluding ICE-NYSE and Mexican (BVM)

	(1)	(2)	(3)	(4)
VARIABLES	Average fee	Average fee	Average fee	Average fee
Treatment/control	-0.02325***	-0.04323***	-0.04654***	-0.04503***
	(0.004)	(0.001)	(0.001)	(0.001)
Pre-/post-treatment	-0.00797	0.00110	-0.00193***	-0.00157***
	(0.005)	(0.001)	(0.001)	(0.001)
DID impact	0.00443	-0.00883***	-0.00206**	-0.00140
	(0.005)	(0.002)	(0.001)	(0.001)
GDP per capita		-0.00001***	-0.00000	-0.00000
		(0.000)	(0.000)	(0.000)
Listed firms			-0.00001***	-0.00001***
			(0.000)	(0.000)
N. of transactions				-0.00000***
				(0.000)
Constant	0.03362***	0.07215***	0.06309***	0.06278***
	(0.004)	(0.001)	(0.001)	(0.001)
Observations	112	112	112	112
R2	0.279	0.970	0.987	0.988

• Note: “average fee” is the dependent variable, expressed in percentage terms. “Treatment/control” is a dummy variable that takes the value of 1 for the Brazilian stock exchange and 0 for the exchanges used as the control group. “Pre-/post treatment” is a dummy variable that is assigned the value of 1 after CADE decides on the merger (2Q2017) and 0 before the decision. “DID impact” is a variable given by the interaction between the variables of groups and periods (the DID), and its coefficient informs the impact of the merger on trading fees. “PIBPC”, “listed firms”, and “N. of transactions” are control variables for, respectively, the GDP per capita by quarter in US dollars, the number of transactions by quarter, and the average number of listed companies by quarter. Robust standard errors in parentheses.
• ^***
  p<0.01 and ** p<0.05.
• Source: elaborated by the authors;

The robustness checks include an additional set of estimations where the post-treatment period extends only up to the first quarter of 2020, aiming to control for potential distortions arising from the COVID-19 pandemic. The outcomes shown in Table A1 in the Appendix A indicate a negative effect of the merger on the fees charged by the Brazilian stock exchange. All these estimates corroborate the previous results by providing no evidence that the merger resulted in adverse competitive effects regarding trading fees¹¹.

As usual for difference-in-difference frameworks, an event-study approach was employed to assess the parallel trends assumption, using BVMF/B3 dummies for the twenty quarters preceding the merger. A joint F-test performed on these pre-treatment coefficients rejected the null hypothesis of no differences in trends before the merger. Nevertheless, recent research cautions regarding the interpretation of pre-trends tests should be considered: (1) parallel trends verified before treatment do not guarantee validity after treatment; (2) failure to reject the null hypothesis of parallel trends does not necessarily confirm their absence, given potential low statistical power; (3) conditioning the analysis on “passing” a pre-trends test may introduce a selection bias, commonly referred to as pre-test bias; and (4) even when significant pre-trends indicate that the parallel trends assumption likely does not hold exactly, “researchers may still wish to learn something about the treatment effect of interest.” (Roth et al., 2023, pp. 2233-2234). The last point is particularly relevant here, given the scarcity of literature addressing mergers in stock market industries and the limitations imposed by data. Furthermore, the results presented thus far will also be complemented with an alternative methodological approach, detailed in subsection 3.3, to address issues arising from violations of the parallel trends assumption.

The price drop observed in the DID results, which occurred regardless of the scenario of domestic monopoly, could have been caused by a stronger international competitive pressure in the stock exchange market – as illustrated by the growing number of Brazilian firms pursuing an IPO abroad – potentially being an object for further research. Yet another possible explanation for these results, more linked to the particularities of the BVMF-CETIP merger, is the existing efficiencies produced by the deal. These are especially difficult to measure, considering that the merger review by CADE kept most of the information on the topic confidential (Conselho Administrativo de Defesa Econômica, 2016, pp. 79-82). Finally, they could be a result of the merger remedies imposed by the antitrust authority, such as the corporate governance mechanisms mentioned in subsection 3.1.

4.2. SYNTHETIC CONTROL RESULTS

Estimates based on the synthetic control method were performed to further corroborate the findings presented earlier, namely, the absence of evidence suggesting adverse effects on competition due to the merger. By employing the Ridge Augmented Synthetic Control Method (ASCM)¹², we derived a synthetic control unit composed of the following weighted stock exchanges: 0.8280 (BMV), –0.5360 (GPW), –0.1220 (ICE), and 0.8290 (JPX).

In the main ASCM estimation, the predictors included the same control covariates employed in the differences-in-differences specification (expressed in natural logarithms) alongside average trading fees¹³. Table 4 below shows that the characteristics of the synthetic unit are very close to those of the treated unit.

Table 4
– Mean values of predictors before treatment

VARIABLES	BVMF/B3	Synthetic	Sample mean
GDP per capita (ln)	7.9930	8.6100	8.6980
Listed firms (ln)	5.9050	6.2260	6.9040
N. of transactions (ln)	17.7030	18.3290	17.4740
Average trading fees	0.0090	0.0060	0.0230

• Note: “average fee” refers to the average trading fee by quarter. “GDP per capita (ln)”, “listed firms (ln)”, and “N. of transactions (ln)” are variables in natural logarithm form for, respectively, the GDP per capita by quarter in US dollars, the number of transactions by quarter, and the average number of listed companies by quarter. Statistics were calculated from the balanced panel database period before treatment (therefore, beginning in 2014).
• Source: elaborated by the authors.

Below, Figure 2 graphically represents the estimate. Although the average fees of the synthetic unit are volatile, their proximity to the Brazilian exchange in the pre‑treatment period is clear, as is the divergence of both series after the merger. As a result, an average treatment effect on the treated (ATT) of – 0.0147% was observed for trading fees of B3 – a magnitude more than nine times the corresponding standard error of 0.0016 –, thus consistent with the previous findings.

Figure 2
– Average trading fees for the Brazilian exchange and its synthetic counterpart. Note: visual output of the Ridge ASCM, whose variable of interest is the average trading fee charged by the exchanges. Source: elaborated by the authors.

Placebo tests were conducted for robustness purposes. First, a permutation test, as proposed by Abadie, Diamond and Hainmueller (2010), was performed by assigning the treatment separately to each control unit; the difference in outcome between the treated unit and its synthetic counterpart was collected in each iteration for comparative analysis. Figure 3 displays the resulting graph, where the black line depicts the difference between the Brazilian exchange and its synthetic unit, while the grey lines depict the gaps for each placebo. As shown, BVMF/B3 exhibits the largest gap after the intervention compared with the placebos, thus indicating the reliability of the results.

Figure 3
– Gap in average trading fees. Note: test comparing the gap in average trading fees in BVMF/B3 with placebo gaps in the other four control exchanges. Source: elaborated by the authors.

In addition to using placebo-treated units, an estimation with a placebo treatment date was carried out, anticipating by two years the actual date of the merge – a back‑dating procedure analogous to that proposed by Abadie, Diamond and Hainmueller (2015). Figure A1 in the Appendix A shows the results of this test. In this scenario, the average fees of BVMF/B3 closely follow the trajectory of their synthetic until approximately the true intervention date, after which a clear divergence emerges. As reported by Abadie (2021), such an outcome, where the back-dating method is blind to the true intervention date, strengthens the causal interpretation of the original ASCM estimation results.

Finally, in order to demonstrate that these results are robust to the economic shocks caused by the COVID-19 pandemic, an additional estimate was performed restricting the database to the first quarter of 2020. The outcome, shown in Figure A2 in the Appendix A, is consistent with the previous findings.

5. CONCLUSION

Ex-post analyses are a valuable tool in formulating evidence-based public policies, including competition policies. This study focused on the competitive effects of a merger reviewed by CADE in March 2017, the transaction between BM&FBOVESPA S.A. and CETIP S.A. that originated the B3 S.A.

The research was restricted to the stock exchange trade market (hence excluding post-trading), the industry in which previous entry attempts occurred. In addition, we focused on investigating the competitive effects of the merger; thus, our goal was not to conduct a broad comparative analysis of the transaction fees charged by the financial centres used as parameters in this paper.

To this end, we employed the difference-in-differences and augmented synthetic control methods, intending to assess how the deal affected the fees in those markets. This may indicate the economic welfare in those markets and, as a consequence, the impact of the adopted competition policy. We created a database of quarterly data covering from 2012 to 2021 with the charges levied on users of the Brazilian stock market, which was affected by the decision of CADE. The database also includes data from a comparison group composed of stock exchanges from around the world.

The results revealed a reduction in the average trading fee of B3 after the merger. Moreover, tests of robustness revealed similar results and, in some estimates, the coefficient for the merger effects was statistically equal to zero. Therefore, we found no adverse competitive effects (relating to a rise in trading fees) as a consequence of the BVMF-CETIP merger.

In view of the literature on the topic, our study seems to have filled an academic gap by conducting the empirical analysis of a merger in the stock market to assess its impact. Furthermore, the results may contribute to society as a whole, considering the importance of ex-post assessments for the formulation of evidence-based public policies.

The observed price drop that occurred despite the scenario of a domestic monopoly has a number of possible explanations. In addition to issues more related to the particularities of the merger, this effect could have resulted from an increased global competitive pressure in the stock exchange market, as illustrated by the growing number of Brazilian firms pursuing an IPO abroad. Exploring this competitive international dimension represents a promising avenue for further research.

APPENDIX A – Robustness tests

Table A1 – Results (Robustness tests in DID)

Estimates A.1 – All exchanges and all time periods
DID impact	-0.01156**		-0.02322***	-0.02256***	-0.01999***
	(0.005)		(0.003)	(0.003)	(0.003)
Observations	212		212	212	212
R2	0.109		0.746	0.801	0.805
Estimates A.2 – All exchanges from 2012 to 2021
DID impact	0.00005		-0.00954***	-0.00845***	-0.00731***
	(0.004)		(0.002)	(0.001)	(0.001)
Observations	184		184	184	184
R2	0.145		0.861	0.926	0.927
Estimates A.3 – All time periods excluding the US exchange (ICE-NYSE)
DID impact	-0.01367***		-0.02468***	-0.02566***	-0.01156***
	(0.005)		(0.003)	(0.004)	(0.004)
Observations	180		180	180	180
R2	0.228		0.747	0.748	0.807
Estimates A.4 – All time periods excluding the US (ICE-NYSE) and the Mexican (BMV) exchanges
DID impact	-0.00641		-0.02289***	-0.01940***	-0.00943**
	(0.006)		(0.003)	(0.004)	(0.004)
Observations	136		136	136	136
R2	0.213		0.778	0.784	0.812
Estimates A.5 – Excluding the US exchange (ICE-NYSE) from 2012 to 2021
DID impact	-0.00157		-0.01110***	-0.01140***	-0.00457**
	(0.004)		(0.002)	(0.002)	(0.002)
Observations	152		152	152	152
R2	0.329		0.911	0.911	0.936
Estimates A.6 – Excluding the US (ICE-NYSE) and the Mexican (BMV) exchanges from 2012 to 2021
DID impact	0.00443		-0.00883***	-0.00206**	-0.00140
	(0.005)		(0.002)	(0.001)	(0.001)
Observations	112		112	112	112
R2	0.279		0.970	0.987	0.988
Estimates A.7 – Excluding the US (ICE-NYSE) and Japanese (JPX) exchanges for all time periods
DID impact	-0.02029***		-0.02025***	-0.02413***	-0.01165**
	(0.003)		(0.004)	(0.005)	(0.005)
Observations	148		148	148	148
R2	0.659		0.659	0.676	0.723
Estimates A.8 – Excluding the US (ICE-NYSE) and Japanese (JPX) exchanges from 2012 to 2021
DID impact	-0.00687***		-0.00619***	-0.00995***	-0.00811***
	(0.002)		(0.002)	(0.002)	(0.002)
#Observations	120		120	120	120
R2	0.919		0.920	0.929	0.931
Estimates A.9 – All time periods with common data availability only for the US exchange (ICE-NYSE)
DID impact	-0.00173***		-0.00112**	-0.00095*	-0.00069
	(0.000)		(0.000)	(0.001)	(0.001)
#Observations	64		64	64	64
R2	0.935		0.937	0.937	0.948
Estimates A.10 – All time periods with common data availability only for the Japanese exchange (JPX)
DID impact	-0.00080*		-0.00002	-0.00164***	-0.00108**
	(0.000)		(0.000)	(0.000)	(0.000)
#Observations	64		64	64	64
R2	0.668		0.709	0.753	0.806
Estimates A.11 – All time periods with common data availability only to the Mexican exchange (BMV)
DID impact	-0.01739***	-0.01286***	-0.00935***	-0.00634*	-0.00327
	(0.003)	(0.003)	(0.003)	(0.004)	(0.004)
#Observations	88	88	88	88	88
R2	0.847	0.863	0.868	0.870	0.871
Estimates A.12 – All time periods with common data availability only to the Polish exchange (GPW)
DID impact	-0.00640**		0.00365	0.00290	0.00473
	(0.003)		(0.004)	(0.004)	(0.004)
#Observations	96		96	96	96
R2	0.814		0.828	0.830	0.835
Estimates A.13 – Only the Mexican exchange (BMV) from 2012 to 2021
DID impact	-0.01447***	-0.01342***	-0.01254***	-0.01073***	-0.00802***
	(0.002)	(0.002)	(0.002)	(0.003)	(0.003)
#Observations	80	80	80	80	80
R2	0.938	0.939	0.939	0.940	0.942
Estimates A.14 – Only the Polish exchange (GPW) from 2012 to 2021
DID impact	0.00072		0.00352***	0.00354***	0.00384***
	(0.001)		(0.001)	(0.001)	(0.001)
Observations	80		80	80	80
R2	0.988		0.990	0.990	0.990
Estimates A.15 – Excluding the US exchange (ICE-NYSE) from 2012 to the first quarter of 2020
DID impact	-0.01339**		-0.02304***	-0.02484***	-0.01142***
	(0.005)		(0.003)	(0.004)	(0.004)
Observations	152		152	152	152
R2	0.203		0.702	0.707	0.794

Note: each set of estimates contains the parameter of interest expressed as a percentage (δ₃ in Equation 1) for four regressions: in the first column, from left to right, the estimate is performed excluding the controlling variables; in the second, it is controlled by GDP per capita (US dollars); in the third, it includes the number of companies listed on each stock exchange; in the fourth, in addition to the previous controls, the number of transactions by quarter is inserted. To the set of estimates made individually with the Mexican exchange, in the regressions A.11 and A.13, a fifth column was added to the right showing the result added with a dummy variable to control the entry into operation of the competitor BIVA in the Mexican market. Statistical significance: *** p<0.01, ** p<0.05, and * p<0.1. Robust standard errors in parentheses. Source: elaborated by the authors.

Figure A1 – Backdating placebo test in synthetic control

Note: even by using a placebo intervention date (two years before the actual treatment), the average fees of BVMF/B3 follow a trajectory similar to that of its synthetic control counterpart until around the actual intervention date, after which the divergence increases. Source: elaborated by the authors.

Figure A2 – Pandemic robustness check

Note: average treatment effect on the treated (ATT) of -0.0148%, and standard error of0.0021. Source: elaborated by the authors.

• 1
  Considering the complexity of this transaction and the competitive risks incurred, the Administrative Council for Economic Defense, by a majority decision, cleared the merger subject to remedies.
• 2
  As afore-mentioned, the transaction was conditioned on a merger control agreement. In a dissent opinion, the rapporteur of the case also voted for approving the deal with conditions.
• 3
  The mergers are Pennzoil’s acquisition of Quaker State, Proctor and Gamble’s acquisition of Tambrands, General Mills’ acquisition of Chex Cereal Brands, the consolidation of Guinness and Grand Metropolitan, and Aurora’s (Mrs. Butterworth) acquisition of Log Cabin.
• 4
  For example, fragmentation may occur even in the absence of competition if the same business group owns more than one stock exchange operating in the same relevant market.
• 5
  There were also overlaps in the registration of Brazilian Agribusiness Letters of Credit (LCA) and investment fund shares, in which the BVMF had a relevant market share. However, they constitute a small portion of the revenue of the companies and, hence, were not a source of concern for the antitrust authority (Conselho Administrativo de Defesa Econômica, 2016).
• 6
  Namely, quarterly GDP per capita in US dollars, the average number of companies listed in each stock exchange by quarter, and the number of transactions carried out in the same quarter.
• 7
  It is worth noting that ICE publishes its stock exchange revenue aggregated with that from its clearing activities.
• 8
  Since there were no population data available for 2021 at the time, we repeated the figure of 2020.
• 9
  This holds true even for the Mexican BMV, which started to compete with the Institutional Stock Exchange (BIVA) from July 2018. Despite its entry in the Mexican market as the second largest firm, the BMV continues to dominate the industry by a wide margin, as seen in statistics published in BIVA website (Bolsa Institucional de Valores, 2024), which indicates no effective rivalry between the two companies.
• 10
  Although the robustness analysis indicated that the inclusion of these controls did not significantly alter the main results, it is worth noting that “listed firms” and “number of transactions” could be regarded as “bad controls” (Angrist; Pischke, 2009), since these variables might be directly influenced by the trading fee levels – higher fees could make the exchange less attractive to investors and firms.
• 11
  The estimates containing comparisons with different exchanges were based on an unbalanced panel database, since data availability was not identical for all time periods (Wing, Simon and Bello-Gomez, 2018). However, regressions were also carried out for the periods in which there was common availability of data for all exchanges (from the year 2014, therefore), achieving similar results.
• 12
  An initial estimation using the canonical SCM did not yield a satisfactory match between the average fees of the synthetic unit and those of the Brazilian exchange in the pre-merger period. This issue arose primarily from the limited size of the donor pool combined with restrictive weighting constraints of SCM (non-negative weights summing exactly to one), which led to only two units receiving positive weights, while the remaining two received zero weights. To overcome this limitation, our study employs the Augmented Synthetic Control Method (ASCM) proposed by Ben-Michael, Feller and Rothstein (2021).
• 13
  Additional estimates were performed using the control covariates in levels rather than logarithms, yielding similar trajectories for the variable of interest in the synthetic unit and comparable treatment effects. The main ASC estimate reported here retains the logarithmic form because it produced a synthetic unit whose values were even closer to those of the treated unit, despite the final outcomes in the variable of interest being very similar across all specifications.
• Funding:
  No funding was received for conducting this study.
• Data availability:
  The data that support the findings of this study are available from the corresponding author upon reasonable request.
• JEL classification:
  D04, G29, L40.

References

• ABADIE, A. Using synthetic controls: feasibility, data requirements, and methodological aspects. Journal of Economic Literature, Nashville, v. 59, n. 2, p. 391-425, 2021. DOI: http://doi.org/10.1257/jel.20191450
  » http://doi.org/10.1257/jel.20191450
• ABADIE, A.; DIAMOND, A.; HAINMUELLER, J. Synthetic control methods for comparative case studies: estimating the effect of California’s Tobacco Control Program. Journal of the American Statistical Association, New York, v. 105, n. 490, p. 493-505, 2010. DOI: http://doi.org/10.1198/jasa.2009.ap08746
  » http://doi.org/10.1198/jasa.2009.ap08746
• ABADIE, A.; DIAMOND, A.; HAINMUELLER, J. Comparative politics and the synthetic control method. American Journal of Political Science, Austin, v. 59, n. 2, p. 495-510, 2015. DOI: http://doi.org/10.1111/ajps.12116
  » http://doi.org/10.1111/ajps.12116
• ABADIE, A.; GARDEAZABAL, J. The economic costs of conflict: a case study of the Basque Country. The American Economic Review, Cambridge, v. 93, n. 1, p. 113-132, 2003. DOI: http://doi.org/10.1257/000282803321455188
  » http://doi.org/10.1257/000282803321455188
• AGUZZONI, L. et al Ex post Merger Evaluation in the U.K. Retail Market for Books. The Journal of Industrial Economics, Hauppauge, v. 64, n. 1, p. 170-200, 2016. DOI: http://doi.org/10.1111/joie.12099
  » http://doi.org/10.1111/joie.12099
• AGUZZONI, L. et al Ex-post analysis of mobile telecom mergers: the case of Austria and The Netherlands. De Economist, Netherlands, v. 166, n. 1, p. 63-87, 2018. DOI: http://doi.org/10.1007/s10645-017-9308-5
  » http://doi.org/10.1007/s10645-017-9308-5
• ANGRIST, J. D.; PISCHKE, J. Mostly harmless econometrics: an empiricist’s companion. Princeton: Princeton University Press, 2009. DOI: http://doi.org/10.1515/9781400829828
  » http://doi.org/10.1515/9781400829828
• ASHENFELTER, O.; HOSKEN, D. The effect of mergers on consumer prices: evidence from five selected case studies. Cambridge: National Bureau of Economic Research NBER, 2008. (Working Paper, 13859).
• ASHENFELTER, O.; HOSKEN, D.; WEINBERG, M. Generating evidence to guide merger enforcement?. Cambridge: National Bureau of Economic Research NBER, 2009. (Working Paper, 14798).
• BABUS, A.; PARLATORE, C. A. Strategic fragmented markets Cambridge: National Bureau of Economic Research NBER, 2021. Working Paper 8729.
• BEN-MICHAEL, E.; FELLER, A.; ROTHSTEIN, J. The augmented synthetic control method. Journal of the American Statistical Association, London, v. 116, n. 536, p. 1789-1803, 2021. DOI: http://doi.org/10.1080/01621459.2021.1929245
  » http://doi.org/10.1080/01621459.2021.1929245
• BERNALES, A. et al A tale of one exchange and two order books: effects of fragmentation in the absence of competition. Frankfurt am Main: Leibniz Institute for Financial Research, 2018. (Working Paper, 34).
• BERNSTEIN, S.; LERNER, J.; DEV, A. The creation and evolution of entrepreneurial public markets Cambridge: National Bureau of Economic Research NBER, 2018. http://doi.org/10.3386/w25414
  » http://doi.org/10.3386/w25414
• BOLSA INSTITUCIONAL DE VALORES – BIVA. Histórico 2024. Available from: https://www.biva.mx/informacion_de_mercado/estadisticas/historico Access in: 21 Aug. 2024.
  » https://www.biva.mx/informacion_de_mercado/estadisticas/historico
• CANTILLONE, E.; YIN, P. Competition between exchanges: a research agenda. International Journal of Industrial Organization, Amsterdam, v. 29, n. 3, p. 329-336, 2011. DOI: http://doi.org/10.1016/j.ijindorg.2010.12.001
  » http://doi.org/10.1016/j.ijindorg.2010.12.001
• CESPA, G.; VIVES, X. Exchange competition, entry, and welfare Munich: CESifo, 2018. (Working Paper, 7432).
• CHAO, Y.; YAO, C.; YE, M. Why discrete price fragments U.S. stock exchanges and disperses their fee structures. Review of Financial Studies, New York, v. 32, n. 3, p. 1068-1101, 2019. DOI: http://doi.org/10.1093/rfs/hhy073
  » http://doi.org/10.1093/rfs/hhy073
• CONSELHO ADMINISTRATIVO DE DEFESA ECONÔMICA – CADE; SUPERINTENDÊNCIA-GERAL. Parecer Técnico n.º 25/CGAA2/SGA1/SG/CADE Brasília: CADE, 2016.
• FARRELL, J.; PAUTLER, P.; VITA, M. Economics at the FTC: retrospective merger analysis with a focus on hospitals. Review of Industrial Organization, Dordrecht, v. 35, n. 369, p. 364-385, 2009. DOI: http://doi.org/10.1007/s11151-009-9231-2
  » http://doi.org/10.1007/s11151-009-9231-2
• FRANKE, F. D.; RESENDE, G. M. Efeitos da concorrência potencial: o caso do ato de concentração Gol-Webjet. Estudos Econômicos, São Paulo, v. 54, n. 3, p. 1-38, 2024. DOI: http://doi.org/10.1590/1980-53575435ffgr
  » http://doi.org/10.1590/1980-53575435ffgr
• KWOKA JUNIOR, J. E. Does merger control work? A retrospective on U.S. Enforcement Actions and Merger Outcomes. Antitrust Law Journal, Chicago, v. 78, p. 619-650. 2013. DOI: http://doi.org/10.2139/ssrn.1954849
  » http://doi.org/10.2139/ssrn.1954849
• LIEBERSOHN, J. How does competition affect retail banking? Quasi-experimental evidence from bank mergers. Journal of Financial Economics, Hauppauge, v. 154, p. 103797, 2024. DOI: http://doi.org/10.1016/j.jfineco.2024.103797
  » http://doi.org/10.1016/j.jfineco.2024.103797
• MA, W. et al Do domestic mergers improve airlines’ international competitiveness? Theoretical and empirical evidence from China. Transportation Research Part E: Logistics and Transportation Review, Cham, v. 176, p. 103211, 2023. DOI: http://doi.org/10.1016/j.tre.2023.103211
  » http://doi.org/10.1016/j.tre.2023.103211
• MEYER, B. D. Natural and quasi-experiments in economics. Journal of Business & Economic Statistics, London, v. 13, n. 2, p. 151-161, 1995. DOI: http://doi.org/10.1080/07350015.1995.10524589
  » http://doi.org/10.1080/07350015.1995.10524589
• NEVO, A.; WHINSTON, M. Taking the dogma out of econometrics: structural modeling and credible inference. The Journal of Economic Perspectives, Pittsburgh, v. 24, n. 2, p. 69-82, 2010. DOI: http://doi.org/10.1257/jep.24.2.69
  » http://doi.org/10.1257/jep.24.2.69
• ORGANISATION FOR ECONOMIC CO-OPERATION AND DEVELOPMENT – OECD. Reference guide on ex-post evaluation of competition agencies’ enforcement decisions Paris: OECD, 2016. Available from: http:// www.oecd.org/daf/competition/Ref-guide-expost-evaluation-2016web.pdf. Access in: 21 Aug. 2024.
• OXERA. What would be the costs and benefits of changing the competitive structure of the market for trading and post-trading services in Brazil? CVM, 2012. Available from: https://conteudo.cvm.gov.br/export/sites/cvm/menu/acesso_informacao/serieshistoricas/estudos/anexos/Oxera-report-18062012.pdf Access in: 21 Aug. 2024.
  » https://conteudo.cvm.gov.br/export/sites/cvm/menu/acesso_informacao/serieshistoricas/estudos/anexos/Oxera-report-18062012.pdf
• PAGANO, M. Trading volume and asset liquidity. The Quarterly Journal of Economics, Cambridge, v. 104, n. 2, p. 255-274, 1989. http://doi.org/10.2307/2937847
  » http://doi.org/10.2307/2937847
• ROTH, J. et al What’s trending in difference-in-differences? A synthesis of the recent econometrics literature. Journal of Econometrics, Amsterdam, v. 235, n. 2, p. 2218-2244, 2023. DOI: http://doi.org/10.1016/j.jeconom.2023.03.008
  » http://doi.org/10.1016/j.jeconom.2023.03.008
• SANTOS, T.; SCHEINKMAN, J. A. Competition among exchanges. The Quarterly Journal of Economics, Cambridge, v. 116, n. 3, p. 1027-1061, 2001. DOI: http://doi.org/10.1162/00335530152466304
  » http://doi.org/10.1162/00335530152466304
• SEVERINO, L. S. M.; RESENDE, G. M.; BISPO, F. Avaliação ex post de ato de concentração: o caso Sadia-Perdigão. Brasília: Ministério da Justiça e Segurança Pública, 2019. (Documento de Trabalho, 0003/2019). Available from: https://cdn.cade.gov.br/Portal/centrais-de-conteudo/publicacoes/estudos-economicos/documentos-de-trabalho/2019/documento-de-trabalho-n03-2019-avaliacao-ex-post-de-ato-de-concentracao-o-caso-sadia-perdigao.pdf Access in: 21 Aug. 2024.
  » https://cdn.cade.gov.br/Portal/centrais-de-conteudo/publicacoes/estudos-economicos/documentos-de-trabalho/2019/documento-de-trabalho-n03-2019-avaliacao-ex-post-de-ato-de-concentracao-o-caso-sadia-perdigao.pdf
• SEVERINO, L. S. M.; RESENDE, G. M.; LIMA, R. C. A. Ex post mergers evaluation: evidence from the Brazilian airline industry. Brasília: Conselho Administrativo de Defesa Econômica, 2021. (Documento de Trabalho, 0003/2021). Available from: http://cdn.cade.gov.br/Portal/centrais-de-conteudo/publicacoes/estudos-economicos/documentos-de-trabalho/2021/Documento-de-Trabalho_Ex-post-mergers-evaluation-Evidence-from-the-Brazilian-airline-industry-versao-final.pdf Access in: 21 Aug. 2024.
  » http://cdn.cade.gov.br/Portal/centrais-de-conteudo/publicacoes/estudos-economicos/documentos-de-trabalho/2021/Documento-de-Trabalho_Ex-post-mergers-evaluation-Evidence-from-the-Brazilian-airline-industry-versao-final.pdf
• WING, C.; SIMON, K.; BELLO-GOMEZ, R. A. Designing difference in difference studies: best practices for public health policy research. Annual Review of Public Health, Palo Alto, v. 39, n. 1, p. 453-469, 2018. DOI: http://doi.org/10.1146/annurev-publhealth-040617-013507. PMid:29328877.

Edited by

• Editor in charge:
  Marta Castilho: https://orcid.org/0000-0002-1483-4597, Carolina GS Dias: https://orcid.org/0000-0002-0478-8777

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Publication Dates

• Publication in this collection
  21 Nov 2025
• Date of issue
  2025

History

• Received
  21 Aug 2024
• Accepted
  27 Aug 2025