The ex-post evaluation of the GDF/Suez merger on gas prices in the Belgian ZEE hub

(1)

Master

's Thesis

The

ex-post evaluation of the GDF/Suez merger on gas prices in

the

Belgian ZEE hub.

Adrianna Wrona (10661689) University of Amsterdam

MSc Economics

Specialisation: Markets and Regulation Supervisor

Dr. Jo Seldeslachts University of Amsterdam

July 27, 2018

Abstract

This paper presents a retrospective study on the effects of the merger between Gaz de France and Suez on day-ahead wholesale gas prices charged in the Belgian Zeebrugge (ZEE) hub. The European Commission cleared the merger, with remedies, in November 2006 but it came into force only in June 2008, when the remedies got implemented. The occurrence of 2008 constitutes the main event of interest studied in this paper. The ex-post analysis carried out in this paper is based on Difference-in-Difference methodology with the ZEE hub acting as the treatment group. The main goal is to assess the effect of the merger on prices charged in the hub, but the econometric analysis also serves as a tool to evaluate the efficacy of the remedies that came as a part of merger clearance deal. If successfully implemented, the remedies were expected to remove barriers to entry in the Belgian gas market and facilitate access to the hub through the ownership unbundling, that in turn, should have created more competition in the market and caused downward pricing pressure. The results of this study indicate that the prices charged in the ZEE hub declined after the merger and, therefore, the remedies were not only successful in mitigating the anticompetitive effects of the merger, but they might have also created competition in the market.

Keywords: Mergers; Ex-post evaluation; Energy market; and Hub prices

(2)

Statement of Originality

This document is written by Student Adrianna Wrona who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Appendix 1

Appendix 2

Appendix 3

Appendix 3 - A1 Appendix 3 - A2 Appendix 3 - A3 Appendix 3 - A4 31 3 32 33 33 34 35 36

(4)

List of Abbreviations

EC

EEX

EU

GDF

GPL

ICE

NBP

NCG

PEG

PEG-N

PEG-S

PSV

TSO

TTF

ZEE

European Commission

European Energy Exchange

European Union

Gaz de France

Gaspool Balancing Services hub

Intercontinental Exchange

National Balancing Point

Netconnect Germany

Point d'Echange de Gaz

Point d'Echange de Gaz North

Point d'Echange de Gaz South

Punto di Scambio Virtuale

Transmission System Operator

Title Transfer Facility

(5)

1 Introduction

Up until this day, merger control institutions continue being perceived as ineffective in prevent-ing anticompetitive outcomes. According to some, merger control practices are too lenient and allow for anticompetitive concentrations to take place (Newbery, 2007). Others believe that intervention in the market may obliterate synergistic efficiencies and technological progress (Aktas, Bodt, & Roll, 2007), consequently, diminishing consumers welfare. Divided opinions give rise to an on-going debate regarding the efficacy of competition policy as such, and its role in the European Union (EU) energy sector that is the focus of this study.1 The importance of the debate is further enhanced by a sharp increase in a number of merger cases that have taken place in the European energy market since the year 2000 (The European Commission, 2015). The literature provides multiple arguments for and against the intervention of competition au-thorities in the marketplace. Most of it is theoretical, and proves that the ex-post analysis of the effects of merger control on competition, is both important and timely. Ex-post evaluations are carried out to inspect the effect of an enforcement decision on the market based on the evidence available at the time the decision was made. Such analysis, however, must be per-formed some time after the decision has been made to benefit from the additional information on how the market of interest has developed. The majority of retrospective studies focus on the effect of an intervention on prices (however, other variables of interest are also sometimes chosen) and employ econometric techniques to quantify the sign and magnitude of those changes. The findings from the econometric analysis are later used to ascertain if there is a causal relationship between an intervention and the changes observed in the market (OECD, 2016). Retrospective studies are both essential in assessing the impact of the past decisions and they help to improve decision making for future cases (Kovacic, 2005).

The decisions to clear (possibly under some conditions) or block the merger are made by com-petition authorities (e.g. national competition authority or the European Commission). First, each planned merger needs to be notified at the relevant competition authority. After such notification, the competition authority commences to Phase 1, in which a case is screened to assess whether the concentration would raise competition concerns in the relevant market. If so, the case is moved to Phase 2. If not, the merger is cleared (possibly with remedies). In the latter phase the thorough investigation of the merger, and its impact on competition, is conducted and a final decision on whether to clear (possibly with remedies) or block the merger is made (Motta, 2004). Nowadays, the European Commission (EC) oftentimes uses remedies as a surpassing instrument to merger prohibitions, and designs them to eliminate or resolve the anticompetitive effects that the merger may result in (Duso, Gugler, & Yurtoglu, 2011). Remedies are composed of a set of directives that the merging parties must commit to fulfill

1

Some claim that the European Commission uses its merger policy decisions as a tool to shape the development of the EU energy market. According to that statement, an increase in a number of European cross border merger proposals creates windows of opportunities for the European Commission (EC) to deliver on the EU goal to liberalise the European energy market (Pakalkaite, 2014). It is also in line with the European Commission view that the EU liberalisation directives fail to be successful.

(6)

in order for the merger, that would have been otherwise blocked, to be cleared. Therefore, the primary aim of remedies is to tackle the issue of market conduct without simultaneously eliminating efficiencies that the merger could potentially bring about(OECD, 2012).

Remedies are also used as a tool to prevent anticompetitive outcomes in the energy market (although some argue that they are effectively designed to foster the EU energy market liberali-sation and unbundling of network and supply activities) (Argentesi, Banal-Esta˜nol, Seldeslachts, & Andrews, 2017). The European energy sector faces a number of EU directives that aim at achieving market liberalisation. The effectiveness of those directives is, however, limited due to political sentiments that differ in each of the Member States. These result in difficulties with full implementation of the energy package requirements by some countries. The problem that further impedes the effectiveness of the directives is the difficulty in assuring third party access to transportation capacities, and favouritism of national vertically integrated businesses by the Member States (Harrison & Mordaunt, 2012). As the result of combined failings in both areas - regulation and implementation, the mergers in the energy sector often resulted in negative effects as compared to the environment where the barriers to entry and the issue of expansion have been eradicated. Therefore, the EC may require merging parties to propose remedies that could solve above-mentioned regulatory/implementation gaps and alleviate potential anticom-petitive effects of mergers, simultaneously promoting energy market liberalisation (Argentesi et al., 2017). As such, remedies play a crucial role in the merger review process, and thus need to be carefully designed and thoroughly evaluated. Ex-post evaluations offer a method to examine whether the remedies were successful and delivered the intended result (OECD, 2012).

In spite of the vast economic importance of ex-post analyses, not enough econometric studies have yet been carried out to test the efficacy of the recent merger control guidelines. Therefore, it is indeed important to test whether remedies involved in merger control decisions can lead to the protection and restoration of effective competition in the market (Duso et al., 2011). It is often due to complicated structures ruling the energy market and lack of data that would enable quantitative examination. This paper, therefore, provides an econometric analysis of the merger in the energy sector and aims to measure its direct effects on the energy market. In particular, day-ahead prices of gas charged in this market. The case study covered in this research is the merger between Gaz de France (GDF, France) and Suez (Suez, France) perceived as one of the most important mergers in the energy sector since the turn of the century. The EC cleared the merger in November 2006 but under the condition that the parties would implement previously agreed remedies. The ECs decision was followed by a period of preparation and renegotiation that resulted in the merger taking effect in June 2008. At that time, the merging parties created a new company (NewCo), now known as Engie, that became one of the largest energy compa-nies in the world and continues affecting multiple stages of the supply chain in Belgium and France. The effects of the merger are tested for trading on the Belgian Zeebrugge gas hub (ZEE, Belgium) in which the impact of the policy enforcement is possible to be isolated and quantified (Argentesi et al., 2017). The remedies that are claimed to have the greatest positive effect on promoting competition on the energy market are those related to removing entry barriers to

(7)

the hub, and in the case of this research they include the ownership unbundling of network and supply activities. If implemented successfully, they would have attained an increase in volumes traded and a decrease in prices charged in the hub (Miriello & Polo, 2015).

This paper builds on the study conducted by Argentesi et al. (2017). In their paper, researchers study the effect of the merger between the GDF and Suez and its associated remedies on prices charged in the ZEE hub. The methodology used in the aforementioned evaluation is known as Difference-in-Difference, and it is widely used in the retrospective studies. The ex-post evalua-tion of the merger boils down to comparing the post-merger outcome - the prices observed in the ZEE hub after the merger between GDF and Suez came into force - to the scenario in which the merger has not taken place, known as a counterfactual (Neven & Zenger, 2008). The premise is to compare the treatment group (the group that underwent an intervention) to the control group (the group that shows the same characteristics as a treatment group except the inter-vention). Given those criteria, the treatment group was represented by the Zeebrugge gas hub (ZEE), while the Dutch Title Transfer Facility (TTF, The Netherlands) served as the control. The outcome of the empirical evaluation carried out by (Argentesi et al., 2017) proved remedies to be successful in preventing the merger0s anticompetitive effects. The remedies could have even positively contributed to creating competition in the Belgian gas trading market. Those conclusions were made based on the body of evidence that shows lower prices of gas charged in the ZEE hub post-merger and improved access to the hub.

The main goal of this paper is to assess whether a similar outcome would have been concluded if different counterfactuals were used. In order to do so, data on different Continental European hubs are employed and alternative controls are created. As available data allows for testing four alternative hubs, four new regressions are run in this study. The hubs of interest include two German hubs Gaspool (GPL, Germany) and NetConnect Germany (NCG, Germany), French Point d0Echange de Gaz - North (PEG-N, France), and Italian Punto di Scambio Virtuale (PSV, Italy). An overview of the European gas market is offered in Figure 1.

The main questions addressed in this paper are: What is the direct effect of the merger between Gaz de France and Suez, and its associated remedies, on prices charged in the Belgian energy market using other European hubs as counterfactual scenarios? Were remedies successful in preventing an anticompetitive outcome?

(8)

Figure 1: An overview of the West-European gas markets with the hubs and gas exchanges active in each of them.

Indeed, a body of theoretical work on the issue of mergers in the energy market already ex-ists but there is not enough empirical work conducted on the topic. As a consequence, in the process of a merger review, competition authorities make decisions regarding notified mergers based only on particular ex-ante assumptions. In order to test whether the assumptions made by competition authorities and the predictions that followed were sound, it would be worthwhile to compare theory to empirical evidence. In doing so, and in the attempt to answer the research questions, an ex-post evaluation of the merger between Gaz de France and Suez is performed. The methodology employed is Difference-in-Difference. The Belgian ZEE serves as a treatment group and the remaining four European hubs take on the role of counterfactuals. The analysis of the result obtained is later presented to the reader, and the efficacy of remedies in maintaining competition in the energy market is judged. Conclusions of this ex-post analysis are used to answer the research questions and they aim to provide guidance for improving future resolutions. The structure of this paper is as follows. In section 2, the summary of the European gas market is provided, and the background information on the merger between GDF and Suez is given. In this section the description of all five hubs regarded in this research is also offered. There-fore, the reader has a chance to familiarise oneself with each of the gas markets of interest but the main focus is paid to the Belgian gas market. In section 3 the econometric analysis is presented. The section starts with the description of the data used in this study, and then it provides an explanation of an econometric model used. Next, the analysis and interpretation of the econometric results is presented to the reader. Finally, in section 4, conclusions about the impact of the GDF/Suez merger with remedies on the gas prices charged in Belgian ZEE

Source: Argiro Roinioti. (2014). The outlook for natural a gas trading hub in se europe. Institute of Energy for South-East Europe.

(9)

hub are made. Those conclusions are also used to answer the research questions and they aim to provide guidance for improving future resolutions.

2 Background Information

This section provides an overview of the gas market in Europe and, in particular, focuses on the merger between the GDF and Suez and its effects on the Belgian gas market. The role of gas hubs is crucial for the functioning of the gas market as they play a role of liquidity instruments in gas exchange. Their main purpose is to facilitate gas trade between buyers and sellers, which enables beneficial short-term arrangements in terms of volumes supplied or sold, to all market participants (The European Commission, 2006). It is therefore important to understand the history of the European gas market and the interdependencies between the European gas hubs. This section is organised as follows. Section 2.1, provides a short summary of the European gas market and introduces hubs that are looked into in this research, particularly focusing on the Belgian ZEE. Later, in section 2.2, the merger between GDF/Suez is described and explained in the context of the Belgian wholesale gas market, here represented by the ZEE hub. Then, in section 2.3 the description of remaining hubs considered in this research is offered.

2.1 Summary of the European Gas Market

This section provides an overview of the evolution in the North West European gas market that, in fact, began in Belgium. As Heather (2015) informs, in 2000 Belgian Zeebrugge hub became the first operational hub in Continental Europe. It followed an example of the British National Balancing Point (NBP) that has been active in that market ever since the late 1990s.2 _{Ever since}

that time, the rapid growth and development has been noted with new hubs entering the market. In as early as 2003, the Dutch TTF and the Italian PSV joined and were closely succeeded by French PEGs in 2004. The landscape continued to develop up until 2009 when German Gaspool became established in the market and the situation remained that way throughout the entire period investigated in this research, namely from June 18th 2007 until the end of 2011 (Heather, 2015). The evolution history of the European gas hubs is presented in Figure 2.

2_{At the time period under investigation, the NBP hub was the most active trade hub in the market. Nowadays,}

the status has changed to the benefit of the Dutch TTF, which was mostly caused by the major expansion in traded volumes exercised by the latter. In spite of scrupulous data on British NBP hub being widely available, it is excluded from the analysis performed in this research because it has not proved to be a suitable counterfactual.

(10)

After a period of intensified market entry, the time of progress of 2006 took place. In the years of 2006-2007, the gas market experienced substantial recovery after a chapter of low confidence in it that was triggered by an event of near collapse in the Eastern Gas Market in 2002. By the year 2011, most of the gas hubs managed to recover from the crisis of 2002 and even the economic turmoil of 2008 did not have a detrimental effect on the volumes traded in the market (Heather, 2012). 3

Figure 2: Evolution history of the gas hubs in Europe.

Although all of the hubs serve a similar purpose, which is the facilitation of trade, they also dif-fer in some aspects. Due to difdif-ferent statuses and stages of development, the hubs are assigned to one of three categories: trading hubs, transit hubs and transition hubs (Heather, 2012). In his paper, Heather (2012) defines European hubs and classifies them in the following manner. The trading hubsare characterised by a high degree of maturity and are based on virtual trad-ing points. A high degree of maturity implies the hub0s transparency, low barriers to entry and allows the hub to be perceived as a reliable market.4 _{This status is also associated with}

a high degree of a hub0s liquidity. Transit hubs (ZEE hub being one of them), on the other hand, are physical points whose primary role is to facilitate the transit of large volumes of gas for onward transportation. However, it is also possible to trade gas there. The remaining European hubs (GPL, NCG, PEG-N, PSV) are classified as transition hubs. They primarily adopted a virtual trading strategy but have not yet managed to achieve the level of maturity comparable to the trading hubs. In fact, in some cases, doubt arises about whether they will ever be able to become more than only national markets. Based on the classification presented above, it should not come as a surprise that, at the time of the merger, TTF and ZEE showed the highest liquidity levels in continental Europe, yet they both underperformed relative to the British NBP. The ICIS Traceability Index, which measures the bid/offer spreads is an important measure in liquidity assessment. The high scores inform about the high degree of competition in the market. As market participants engage in transaction, the competition that they drive decreases the spread between the bid prices and ask prices, along all given contracts. This, in turn, encourages more trade, making the hub a transparent and reliable market. The opposite is true for less liquid hubs. Low levels of liquidity discourage transactions, increase bid/ask spread and, consequently, impede trade. At the third quarter of 2008 the British NBP achieved a score of 19/20 in ICIS making it the number one hub in terms of liquidity in North West Europe. Continental TTF, ZEE and NCG followed with 12 out of 20 points. The French PEG

3

They, in fact, increased almost to previous highs. Prices, on the other hand, have been negatively affected in the times of recession.

4

(11)

got 8/20 and the German GPL obtained 5/20. The Italian PSV appeared at the very bottom of the ranking with a score 0/20 (Heather, 2015). The status of the latter could have been severely impacted by the political situation in the country that has been outlined previously. Moreover, there are two possible forms in which hubs can be established, physical or virtual. The ZEE hub in Belgium is a physical hub located on the Zeebrugge beach in Belgium. Virtual hubs, on the other hand, dont have a specific location and are referred to as national points (Heather, 2015). The Belgian transmission network is an integrated network that is used for both domestic transmission and international transit (Argentesi et al., 2017).5 Belgium is per-ceived as a vital transit country and the Zeebrugge hub is claimed to be the one of the most significant gas hubs in the European Union (Argiro Roinioti, 2014). Its capacity exceeds 250% of the gas consumed in Belgium and constitutes 10% of the gas consumed in Western Europe. The geographic positioning and the physical nature of the hub are ideal to enable gas flows to and from many other European countries (Heather, 2012). The same physical nature of the hub makes it fall behind other continental hubs with regards to volumes traded.6 Moreover, GDF and Suez were active on all levels of Belgian gas sector and were important users of the ZEE hub (Argentesi et al., 2017).

It would, indeed, be interesting to look into the effects of the merger on the volumes traded in ZEE but it is beyond the scope of this research. As for now, however, attention is paid towards the impact of the GDF/Suez merger on the wholesale prices observed in ZEE hub. The particularities of the merger are described below.

2.2 Summary of the Merger between Gaz de France and Suez

On May 10th, 2006 the European Commission got notified about the planned merger between the Gaz de France group and Suez group by the means of exchange of shares, which would result in the GDF absorbing Suez. Given the structure of the market, this notification immediately raised competition concerns and doubts about the compatibility of the concentration with the common market and the operation of the European Economic Area (EEA) agreement (The Eu-ropean Commission, 2006) . The primary area of concern was the Belgian gas market. At the time of notification, Suez group enjoyed a 57 per cent stake in Distrigas (natural gas wholesale and supply) and the same amount of shares in Fluxys (natural as transmission system operator). The Belgian high-pressure transmission network is operated as a monopoly, which is owned by

5_{Integrated network is a network in which the pipelines that are used for international gas transit are also}

used for domestic gas transmission.

6_{To tackle this problem, a Belgian regulator pushed towards the strategy of market coupling with the}

neigh-boring countries. In fact, since December 2008, the ZEE started to operate a joint venture with the French TSO, GRTgaz, using the electronic platform Capsquere. This very fact increases the degree of interconnection between the European hubs, especially the French hub, and is aligned with the European strategy of market liberation. It does, on the other hand, complicate the attempt to separate the effects of particular events on performance of specific hubs as the hubs become increasingly interdependent.

(12)

Fluxys, the company responsible for its management, development and maintenance. Also at that time, GDF was Suez0s second largest competitor both in the market for gas wholesale and supply (Argentesi et al., 2017). Both GDF and Suez were, therefore, active in the Belgian gas market, with GDF focusing its activities around the gas chain and gas/energy services while Suez was mostly present in the utility industry and services.

The most apparent concern of the notified merger was the horizontal one, as the concentration would cause clear overlaps of merging parties0 activities. Nevertheless, the vertical effects were also taken into consideration in the Commission0s final decision. It is due to the fact that the Belgian market is also characterized by the vertical relationship in transmission and distribu-tion networks. Those are owned by the legal monopolies and, therefore, the impact of the concentration on the ownership and the access to the various infrastructures came into play and must have been evaluated. The competition assessment run by European Commission (2006) pointed out the strengthening of the Distrigas0s dominant position in the Belgian market as a result of the merger. That is due to the fact that at that time, Distrigas was an incumbent operator in Belgium with GDF being its competitor. The concentration between GDF and Suez could, therefore, cause an incorporation of GDF0s Belgian activities by Distrigas, and remove the competitive pressure between the two. That would, in turn, increase barriers to entry in the already difficult to enter market. In response to the European Commission0s reservations, GDF and Suez proposed remedies that included the divestitures of the Suez group0s holdings in Distrigas to the third party and the ownership unbundling of network and supply activities.

7 _{Moreover, in order to eliminate the issue of entry barriers caused by the control of Zeebrugge}

market access held by Distrigas, the merging parties offered the transfer of marketing rights to the transmission network to Fluxys. The investigated remedies proposed by the parties, thus, aimed at facilitating market entry for the third parties and a simultaneous liquidity boost in the ZEE hub (which at the time of the merger suffered liquidity problems). The package of remedies offered by GDF and Suez met the Commission0s approval and, as a consequence, the merger between the parties got cleared in November 2006. This is referred to as Event 1 in the paper of Argentesi et al. (2017). However, rigid contracts and other operational challenges pre-vented the merger from taking effect up until July 2008 when remedies got implemented. The aforementioned study even calls this Event 2. The history of events is summarized in Figure 3.

7_{The report of European Commission on the case lists all of the commitments presented by the parties. The}

commitments submitted by the parties on 13 October 2006 are made up of five main parts: i) divestiture to a third party of the Suez groups holding in Distrigaz; ii) divestiture to a third party of GDFs holding (via Segebel) in SPE; iii) restructuring of the activities of Fluxys s.a. and relinquishing of all control over the company765; iv) a series of additional measures relating to the gas infrastructures in Belgium and France; v) divestiture to a third party of Cofathec Coriance and the district heating networks operated by Cofathec Services

(13)

Figure 3: Individual and overall events related to the GDF/Suez merger.

The second event constitutes a major time frame specification in the econometric analysis of an impact of GDF/Suez merger on the prices of gas in the ZEE hub, in this research. This is mostly due to a lack of data on prices charged by other hubs in Continental Europe prior to June 18th 2007. It means that the effect of the first event on gas prices cannot be tested but the effect of the Event 2 can, indeed, be looked into.

2.3 Description of West European Gas Hubs

The NBP was the largest West European hub active at the time of the merger but because of the high degree of interconnection between the NBP and ZEE, the NBP cannot serve as a suitable control group and, therefore, will be disregarded from this research. The NBP hub is physically linked to the ZEE through the Interconnector UK pipeline, which causes a high degree of interconnection between the hubs and classifies them as players in the same common market. Interconnection between the hubs is problematic due to the fact that the possibility of a major effect affecting one hub and not affecting another hub cannot be ruled out (Argentesi et al., 2017). Another large and mature European hub is the Dutch TTF. It is the hub that was originally used as a counterfactual in the research of Argentesi et al. (2017). The reason behind the choice is that, at the time of the merger between GDF and Suez, the two hubs were claimed to belong to a separate market (The European Commission, 2015). The study also concluded a negative impact of the GDF/Suez merger on the wholesale prices charged in the

13

Note: The Event 1 represents the time when the GDF/Suez merger got cleared by the European Commission and the Event 2 is the time when the merger came into effect and remedies got implemented. June 2007, is the beginning of the time frame because of lack of data on prices charged in West European gas hubs prior to that date.

(14)

ZEE hub while using TTF as a counterfactual scenario. For more details the reader referred to the paper of Argentesi et al. (2017).

The hubs that will serve as counterfactuals in this study include German GPL and NCG, French PEG-N and Italian PSV. As indicated in section 2.1, all of those hubs belong to the category of transit hubs but they also differ in several aspects. Each of the hubs is described in this section, in the context of the gas market within which it operates. Firstly, the German market is outlined. It is the market characterized by two players, namely NetConnect Germany and Gaspool. The favorable geographical location (almost in the heart of Europe) coupled with advanced infrastructure, has catalysed Germany0s process of becoming the second largest gas market in Continental Europe. Both GPL and NCG have high and low calorific gas networks, which they balance individually (Lohmann, 2009). On the other hand, the gas pipeline system in Germany differs from the usual European practice such that each of the hubs is run by 6 TSOs.8 GPL is a physical hub that is mainly used by traders to balance their storage portfolios against the German virtual trading point, NCG. That explains the differences in the ICIS Trad-ability Index as presented earlier in this paper. While the activity of GPL has never covered a large geographical area, at the time relevant for this research, GPL was considered the most promising hub in the North West Europe (Heather, 2015). It is mostly due to the ever-soring volumes traded. However, NCG was always larger and more liquid than GPL.9

As for the French market, it is dominated by the two PEGs -PEG-N and PEG-S whose area of influence is divided into north and south respectively. Both hubs have been relatively stable since the time they started to operate but only the PEG-N is considered in this research. At the time of the merger, PRG-N enjoyed a continued increase in the volumes traded but the progress was relatively slow compared to other European markets. In January 2009 the mar-ket that previously consisted of five separate zones became concentrated into three zones with PEG-N being the most active trader. Although the French PEG-N is responsible for most of the gas trades in the country, its status is still considered as poor in terms of liquidity and the stage of development. High calorific contracts remain the most popular there but low calorific contracts are also traded in this market. The development that proved beneficial to the French market is a Capsquere trading platform that facilitated trade between ZEE hub and PEG-N. It managed to slightly boost the volumes traded but the level of the hubs liquidity continues to be rather low. This technological advancement might have also positively impacted the degree of interconnection between the Belgian and French hub. Additionally, French political attitude to trade differs from the practices of the hubs described before. It is a part of cultural practice to favour national companies (0flag bearing0), which may pose some sort of barrier to the commercial development of the hubs (Heather, 2015).

8_{The company(ies) responsible for a gas pipeline system and its safe operation. Some countries have only one}

TSO while others have multiple TSOs. Germany is characterized by 6 TSOs active in each of the hubs. Other countries investigated in this paper usually have one or two TSOs. Those TSOs are: Belgium, Fluxys; France, GRTgaz/TIGF; Italy, Edison Stoccaggio/Snam Rete Gas.

9_{The case of NCG is somewhat similar to the Dutch TTF. Even the liquidity levels of both hubs were the}

same at the time of the merger. That fact coupled with the close geographic proximity between the hubs could make, most likely, TTF serve as a price benchmark for other hubs that were capable of following it.

(15)

The last hub looked into in this study is Italian PSV. It is a hub operating on a virtual model with the Italian Trading Code for gas being an almost identical equivalent to the British prede-cessor. The storage is designed to be open to traders and all gas that enters the system passes through the PVS. The entry capacity, however, did not prove to be as flexible as the one of the British counterparts. As a result, only marginal volumes of gas that enter the market were actually traded in the PVS hub. Trading difficulties experienced in Italy opened a political de-bate that alleged the Italian government of supporting tacit behavior exercised by the National Incumbent. Accusations were further intensified by the lack of an English version of the rules governing the third party access to the hub. The external traders saw it as an intentional action aiming at increasing the barriers to trade. The above, coupled with several other shortcomings led to the outcome, which allowed the market to continue operating, but in an inefficient way. That malfunctioning was later reflected by the high prices of gas (Heather, 2015). On the other hand, the introduction of new balancing regimes in 2012 resulted in some positive changes in the Italian market and PSV experienced a rapid spike in volumes traded within the short period of three months. This event, however, falls behind the time frame investigated in this research. Additional developments have also been proposed, and they have the potential to increase the competitiveness of the Italian PVS hub and to further boost its development.

3 Empirical Analysis

This section deals with the empirical evaluation of the decision to clear the GDF/Suez merger with remedies and its impact on the wholesale day-ahead prices of gas charged in the Belgian ZEE hub. Within this section, the reader is provided with a description of the data used to an-swer the research questions, and an explanation of the methodology implemented to do so. The econometric method used is the Difference-in-Difference (DiD) econometric approach. There are no other hubs active in the Belgian energy market, so the ZEE hub needs to be compared to the hubs from other countries. As stressed by Argentesi et al. (2017), however, the comparison of different countries poses several challenges in the context of DiD. Those will be addressed in detail later in this section. As explained before, the hubs under investigation are two German hubs - GPL and NCG, French Hub Exchange - PEG-N, and Italian PSV. The period inves-tigated is the interval between 18th of June 2007 and 31st of December 2011 (second event identified by Argentesi et al. (2017).

This section section is organised as follows. Section 3.1 treats the topic of data used in the research. Next, in section 3.2 the econometric model used in this study is explained and section 3.3 presents the empirical results of the research and offers their analysis. Finally, the empirical part is concluded by the robustness check of the main results obtained in this ex-post evaluation, and it is presented in part 3.4.

(16)

3.1 Data

The Difference in Difference econometric method is used to estimate the effect of the merger between GDF and Suez on the prices of gas in the Zeebrugge hub. The panel data for the Belgian hub (ZEE) and other European hubs used for the analysis is accessed through Platts. The dependent variable, in this case, is a daily transaction price for day-ahead wholesale natural gas traded during working days. An explicit data on daily prices and other supply and demand shifters used in this research is available for both the Belgian hub and the other European hubs considered. The sample period used in this research falls between June 18th 2007 and December 2011 with data points for each working day i.e. Monday to Friday, excluding public holidays. Moreover, the control variables are included in the analysis to account for demand and supply shifts. Those controls include daily temperature, price of oil, coal, and power for Belgium and other countries under investigation. Temperatures are taken either from national climatology registries or from the air force, depending on the availability of the former. As for the prices of oil and coal, the standard European benchmarks are used, also as published by Plaats. The electricity prices are obtained by taking the average of the base and peak load experienced in each of the national power generating facilities. Additionally, the dummy variables for each day of the week, month and year are included to control for varying demand in different seasons. The descriptive summary of the data together with the sources and links is presented in Appendix 1. Before moving to the empirical model employed in this research, the historical pricing patterns across all hubs of interest are looked into. In Figure 4, one can observe a similar direction of price changes across the entire sample. The pricing levels, nevertheless, varied between the hubs with Italian PSV charging highest prices throughout the entire period. The price discrepancy and the upward pricing trend was the most prominent there in the period starting in early 2011 as compared to other North West European hubs but there were also other periods when the changes in price levels at Italian PSV follow an opposite direction to the remaining hubs. One such occasion could be observed in late 2009. Overall, however the prices charged by each of the hubs move in the same direction which is in line with the common trend assumption,that underlies the Difference-in-Difference strategy. What the common trend assumption implies is that both treated and non-treated groups follow the same pattern in the variable of interest. So, any deviation in that trend that is observed for the treated group can be directly attributed to the treatment itself, as opposed to the differences in the characteristic of treated and con-trol groups (Lechner et al., 2011). The only problematic hub, from the point of view of this research, may be the Italian hub whose observed prices, at times, deviate from the common trend. Nevertheless, as most of the time the price trend for all the hubs involved is similar, they will all undergo further econometric evaluation.

(17)

Figure 4: Historical prices charged in the European gas market.

Moreover, it is clear from the graph that the pricing behavior complies with the aforemen-tioned description of the gas market history. Prices charged keep on rising from mid-2007 to later sharply drop in mid-2008 and start to pick up again around mid-2009 for all hubs involved.

3.2 Econometric Model

The major problem at hand is assessing the effects of the merger between GDF and Suez on the wholesale day-ahead prices of gas charged in the Belgian ZEE hub. However, the remedies imposed on the merging parties by the European Commission also played an important role on the decision to clear the merger. Their goal was to reduce barriers to entry, and through unbundling the ownership of infrastructure and transmission networks, facilitate third party access to the hub. Those remedies, if successfully implemented, should have created competi-tion on the Belgian gas market that would have been reflected in downward pricing pressure. In order to test whether the desired effects were, indeed, achieved the Difference-in-Difference econometric approach is used in this analysis. The DiD is popularly used in policy evaluation exercises because it is based on a simple and intuitive setting of comparing two groups in for two distinct time periods (Weinberg, 2007). This is also the basic set up that will be used in

(18)

this study.10 The premise is to compare the treatment group (ZEE, affected by the merger) to the control group (other European hubs that are not affected by the treatment but, otherwise, reveal similar characteristics as a treatment group). In simple words, the methodology aims at finding what happened to the control group that underwent the treatment of a merger as opposed to what would have happened to it in the absence of this merger. The absence of treatment is exactly what the control group represents. Intuitively, the DiD methodology is based on two main assumptions. Firstly, the control group must be comparable to the treated group, which hinges on the assumption of a so called, common trend that both groups should follow. The second condition that must be present is conditional independence, which in simple terms means that the choice of the treatment and control group should be random, therefore, not correlated with the possible outcomes (Rubin, 1974). For this reason the treated and con-trol hubs should not be interconnected. After fulfilling the above-mentioned conditions, the regression can be run.

The specification used in this study treats prices as a dependent variable. Moreover, a set of demand and supply side variables is included in the regression together with the merger specific dummies. Lastly, in order to remove the common factors that affected the countries of interest, the double differentiation is applied. The above strategy allows for the identification of the effect of the GDF/Suez merger on prices charged in the ZEE hub. The estimated model takes on the following form:

As represented by the equation, the panel data will be used in this analysis. The dependent variable pit informs about prices that were charged in a given hub i at a given time t. Next,

the variable treati represents the treated group, so it is equal to one for Belgium, and is equal

to zero for each of the control groups. Moreover, dummy posti is equal to one for all prices

charged post merger i.e. after June 2008 and it is equal to zero otherwise. The main variable of interest is the interaction term treati*posti. It measures the effect of a merger (treatment)

on the prices that were charged in the treated group (ZEE hub) post-merger. Additionally, the supply side variables cost factors are included. Those are, the daily prices of powerit charged

in each of the power generators closest to the hub of interest, i. The prices of oiltand coaltare

European benchmark prices, so they have the same values for each of the hubs (this explains the lack of a hub specific subscript, i ).

10

The paper of Argentesi et at. (2017) uses a more complicated setup but it couldnt be applied to this research due to data being unavailable.

(19)

Lastly, the demand side shifters are added, and those include daily temperatures (tempit) in

each country, as well as the dummies for the day of the week (Monday-Friday), month and year. There are, of course, potential problems with the DiD methodology that may undermine the validity of the results. The primary econometric problem that users of the DiD may encounter is the sample selection bias (Heckman et al., 2013). This type of bias arises when the control group that the treated group is compared to, differs from the treated group in aspects other than the pure absence of treatment. That is against the rules for assessing the impact of a treatment and in order to make valuable inferences about the intervention, the very intervention should be the only difference between the two groups. The reality is always more complicated, and such precise similarities will be hard to find but one should always try obeying this rule while choosing a control group. The second problem that can occur while dealing with the assess-ment of treatassess-ment effects is the problem of omitted variable bias (Dehejia & Wahba, 1999). The difference between this issue and the first problem mentioned in this section is that the former concerns the sample included (or omitted) from the study as such, while the latter treats the presence of the omitted variables within the chosen sample. The omitted variable bias can be spotted while obtaining counterintuitive or illogical results. It informs the researcher about the existence of some unobserved or uncontrolled factors that must be included in the regression in order to make unbiased and valid inferences on the effect of the intervention. Moreover, panel data is often associated with the presence of serial correlation between the observations. In order to account for it, and the heteroscedasticity of the error term, the error term estimation is performed using Newey-West standard errors. The main specification used in this research allows for the autocorrelation up to seven lags, which is up to seven days. Nevertheless, several other specifications are designed to check the robustness of the results. Those will be described in section 3.4.

3.3 Main Results

In this section the main results of the econometric analysis will be shown. For each counterfac-tual, it includes the presentation and analysis of the results that fall into two main frameworks. The first one measures the effect of a merger from the exact date of its implementation, which is from June 2008. The second one excludes the time window of the three months surrounding the event. The former is usually done in order to allow for the anticipatory effect or the delays that the event of a merger might have caused. In order to test how long the time window should be, to offer the most precise estimation, three regressions with skipped time windows are run. The remaining two, however, are treated as a robustness check. The specification design is inspired by the paper of Argentesi et al. (2017), and following researchers0 methodology, additional time windows of one month and six months are omitted in robustness check regressions. Moreover, in order to eliminate the problem of autocorrelation between the observations, that is often present in time series data, the Newely-West standard errors are estimated and they allow for

(20)

data autocorrelation of up to seven lags.

This section is therefore organised as follows. First, the results of the baseline regressions are presented and analysed. However, due to the relatively repetitive nature of the analysis, baseline results for each counterfactual are jointly presented in Table 1. Table 1, therefore, offers the combined outcome from the regressions run for each of the potential counterfactuals separately (represented in separate columns). Next, reasons behind the robustness check are explained and their outcomes are presented and described. Finally, the short conclusions from the analysis are offered to the reader.

3.3.1 The individual effect of the merger

Table 1 presents the outcome of the regressions that measure the direct effects of the merger between GDF and Suez on gas prices in the Zeebrugge hub after the merger got implemented with remedies. Namely, after June 2008. The prices charged in the Belgian hub were tested against control groups as described earlier in the research. The baseline analysis for those hubs is presented in columns 1, 3, 5, and 7 respectively. For the purposes of this paper, it is referred to as Full time. Additionally, a similar regression, albeit with a 3-month time window surround-ing the merger dropped, is run on all of the aforementioned hubs. Those results are therefore presented in the remaining columns,2, 4, 6, 8, for each of the hubs respectively. This is typically done to check the validity of the previously obtained results and it is often performed in the retrospective studies to account for the anticipatory effects or delays in reaction to the event at hand. The main variable of interest is T reat ∗ post2. It is the interaction term that represents

the effect of the merger in the treated group, which is Belgian ZEE, on the prices of gas in the ZEE hub. Table 1 is presented below for interested readers to get familiarised with it.

(21)

Table 1.

Price effects of the merger (w

ith remedies) betw

een Gaz de France and Suez compared to other West European hubs.

No te : T he de pe nde nt v ar ia bl e is t he da ily g as p ri ce a t the hu b. In a ll sp ec if ic at ion s, da y, m on th, a nd ye ar du m m ie s ar e in cl ude d. T he re su lts are p re se nte d fo r ea ch of t he i nve st igat ed hu bs an d ar e re pr es en te d by G PL: G as pool ; N CG : N et Con ne ct G er m an y; P EG -N: Th e Po int d ’Ec ha ng e de Ga z, an d PS V : Th e Punt o di S ca m bi o Vi rt ua le . Ne w ey -We st s ta nd ar d er ro rs a re r ep or te d in th e pa re nt he se s. T he s ym bo ls *** , ** , * re pre se nt si gn if ic an ce a t 1 % , 5 % a nd 1 0% le ve ls re sp ec ti ve ly .

GP L NC G PE G -N PS V (1 ) (2 ) (3 ) (4 ) (5 ) (6 ) (7 ) (8 ) VA R IA B LES Fu ll ti m e 3-mt h w in do w Fu ll ti m e 3-mt h w in do w Fu ll ti me 3-mt h w in do w Fu ll ti m e 3-mt h w in do w Tr ea t 0. 693* ** 0. 851* ** 0. 258* ** 0. 320* ** 3. 664* ** 4. 048* ** -2. 515* ** -2. 453* ** (0 .1 09 ) (0 .1 03 ) (0 .0 58 8) (0 .0 58 8) (0 .4 84 ) (0 .5 25 ) (0 .2 03 ) (0 .2 15 ) Po st2 0. 957* ** 1. 236* ** 0. 988* ** 1. 163* ** 1. 855* ** 2.4 93 ** * 3. 124* ** 3. 559* ** (0 .1 62 ) (0 .1 64 ) (0 .1 47 ) (0 .1 72 ) (0 .5 88 ) (0 .6 63 ) (0 .4 34 ) (0 .4 97 ) Tr ea t* Po st2 -1. 085* ** -1. 264* ** -0. 806* ** -0. 878* ** -0. 175 -0. 478 -3. 820* ** -3. 954* ** (0 .1 19 ) (0 .1 13 ) (0 .0 77 4) (0 .0 77 9) (0 .6 01 ) (0 .6 41 ) (0 .2 97 ) (0 .3 15 ) Te m p sq ua re d 0. 000479 0. 000301 0. 000425 0. 000453 -0. 00170 -0. 00252 -0. 00627* ** -0. 00616* ** (0 .0 00 40 8) (0 .0 00 39 8) (0 .0 00 39 8) (0 .0 00 40 8) (0 .0 02 58 ) (0 .0 02 72 ) (0 .0 01 51 ) (0 .0 01 55 ) Te m p -0. 0198* * -0. 0182* * -0. 0206* -0. 0205* -0. 0466 -0. 0322 0. 110* ** 0. 107* * (0 .0 09 05 ) (0 .0 09 04 ) (0 .0 10 5) (0 .0 10 6) (0 .0 57 8) (0 .0 59 0) (0 .0 42 1) (0 .0422) Oi l 0. 972* ** 0. 977* ** 0. 974* ** 0. 980* ** 0. 674* ** 0. 690* ** 0. 817* ** 0. 824* ** (0 .0 07 61 ) (0 .0 07 11 ) (0 .0 09 50 ) (0 .0 09 07 ) (0 .0 41 0) (0 .0 42 0) (0 .0 37 3) (0 .0 37 8) Coal -0. 0453* ** -0. 0398* ** -0. 0487* * -0. 0463* * -0. 0954 -0. 0898 -0. 362 *** -0. 353* ** (0 .0 13 7) (0 .0 13 2) (0 .0 19 2) (0 .0 19 0) (0 .0 65 7) (0 .0 66 4) (0 .0 50 6) (0 .0 51 1) Po w er -0. 00102 -0. 00218 7. 79e -05 -0. 000847 -0. 00811* * -0. 00856* * 0. 00234 0. 00148 (0 .0 01 45 ) (0 .0 01 49 ) (0 .0 01 23 ) (0 .0 01 22 ) (0 .0 03 81 ) (0 .0 04 15 ) (0 .0 01 82 ) (0 .0 01 60 ) Con st an t -0. 349* -0. 424* * 0. 317 0. 254 7. 786* ** 7. 297* ** 5. 235* ** 5. 227* ** (0 .1 85 ) (0 .1 90 ) (0 .1 95 ) (0 .1 98 ) (1 .0 61 ) (1 .0 61 ) (0 .7 50 ) (0 .7 83 ) Ob se rv at io ns 2, 292 2, 166 2, 292 2, 166 2, 292 2, 166 2, 289 2, 163 R -sq ua re d 0. 902 0. 902 0. 991 0. 991 0. 678 0. 669 0. 921 0. 921 21

(22)

The outcome of the regression suggests that the event of June 2018 had a negative impact on gas prices charged in the Belgian hub relative to all control groups of interest and under both specifications. All coefficients of T reat ∗ post2 are characterised by negative coefficients and six

of them are statistically significant at one percent significance level. The sign of the coefficients, thus, suggests that the GDF/Suez merger had a negative effect on prices charged in the Belgian hub as compared to the two German hubs and the Italian hub. As for the French hub, the impact on prices is indeed negative but not statistically significant, so no major conclusions can be made in this case.

In column 1, where the German GPL is a counterfactual, the impact of the merger on the post-merger price charged in Belgium (T reat ∗ post2) has a value of -1.085 on average. It is

much larger than the value of T reat ∗ post2of -0.351 obtained by Argentesi et al. (2017) in their

original analysis.11 However, the effect is larger than in the case of another German hub NCG. In the case of the latter, the coefficient has a value of 0.806 and is presented in column 3 in Table 1. The above indicates a smaller magnitude of impact that the merger between GDF and Suez had on the prices charged in the ZEE hub while using NCG as a counterfactual compared to GPL. Nevertheless, the decision to clear the merger with remedies has proved to cause negative pricing pressure in both cases. The difference in the value of the coefficients may be mostly due to the higher liquidity of NCG compared to the liquidity of GPL. A high degree of liquidity is said to be associated with greater competition on the market and lower average prices are charged (Heather, 2012). This theory could also explain the relatively low negative value of the coefficient for TTF hub, which at the time of the research was the second most liquid hub in Europe and, nowadays, it is claimed to be a European liquidity leader (Heather, 2012). As for the specification with the three-month period dropped, the results confirm the validity of the original regression. The direction of the coefficients is as expected but the magnitude of its effect is slightly larger for the latter specification. The reason for the difference can indeed be the fact that other hubs implemented some anticipatory measures and they adjusted prices downwards before the event of June 2008. Especially that the merger was already cleared in November 2006. Now, focusing on the two German hubs, other variables included in the regression are anal-ysed.Now, focusing on the two German hubs, other variables included in the regression are analysed. For both German hubs, only values for squared temperatures and power prices are not statistically significant. The rest of the variables show, at least, some degree of statistical significance, which means that they are all responsible for explaining gas prices in the Belgian hub. The variable Treat is positive in all four regressions (columns 1-4), which means that the prices charged by the treated group, ZEE hub, were on average higher from the prices charged by German GPL and NCG. In the regression 1, the price difference is 0.693 and statistically significant. The results for NCG are presented in column 3 and the value of Treat is 0.258, so it’s lower than in the former case. This, again, may be connected to the greater maturity and liquidity of NetConnect Germany. Liquid, transparent, and mature hubs are more likely to

11

(23)

serve as price benchmarks. The smaller coefficient on Treat for the latter hub could potentially be explained by that phenomenon. Moreover, the smaller value of coefficient on T reat ∗ post2

could also be explained this way. The more established the control hub, the less of the price differences should be there between the counterfactual and treatment group. Additionally, the more of the price benchmark the control hub is for the treated hub, the less the merger should impact the prices post-merger due to already relatively competitive prices being charged in the hub of interest. As for the variable P ost2, it is positive and significant in all instances and

its value, again, increases with the length of the days dropped from the analysis. Its positive coefficient informs about the average increase in price levels after the event of the merger for both treatment and control group. The remaining control variables have expected signs and are, generally, in line with the results obtained by Argentesi et al. (2017). Finally, the values of R-squared are high and have values of 0.902 and 0.991 for Gaspool and NetConnect Germany respectively. It indicates that the econometric model used in this paper accounts for a fair share of the variation in data for the Belgian and two German hubs.

The overall conclusion form the analysis carried out in this section is that the merger between the GDP and Suez with remedies that included divestitures of Distrigas, Distrigas & Co, and partial divestiture in Fluxys − had a negative impact on the prices of gas in the ZEE hub relative to both the control groups, Gaspool and NetConnect Germany. The effect is, therefore, similar (but of a different magnitude) to the outcome obtained by Argentesi et al. (2017) when researched, took the Dutch TTF as a control. The preliminary conclusion would, therefore, be that the European Commissions decision to clear the merger not only managed to prevent the anticompetitive effects that the merger could have caused but it might have also been respon-sible for creating more competition in the market.

The results of a baseline regression with French PEG-N as a counterfactual are presented in columns 5 and 6 of Table 1. In the case of this French hub, the value of the main coeffi-cient of interest, T reat ∗ post2, is indeed, negative (-0.175 and -0.478, for the first and second

specification respectively) but this time, not statistically significant. When trying to find an explanation, one should try to understand the impact of remaining variables included in the regression on prices charged in a Belgian ZEE hub. As for the coefficient on Treat, while PEG-N is a counterfactual, the value is larger than in the results obtained for hubs analysed before. The value of 3.664 suggests that the prices charged in the Belgian hub were significantly higher from the French prices at the time period under investigation. The value is even larger when the 3-month time window is dropped from the analysis. Moreover, the coefficient on P ost2

in-dicates that the prices post-merger were also, on average, higher than in the pre-merger period. Again, the magnitude of the effect increases with dropping data around the event of matter. In the case of France, temperature does not seem to play a significant role in pricing strategies but the remaining supply and demand variables follow the same pattern as in the econometric analysis of German market. One of the main problems that might have occurred in the case of French hub is the issue of omitted variable bias. In fact, the value of R-squared is 0.678, which informs about an over 30 percent of the variance in price not being explained by the regressors

(24)

used in the model. Moreover, other market characteristics of the hub might have affected the output. Firstly, the French hub is much less liquid than the previously analysed hubs. Its level of maturity is also much lower than in the case of the hubs analysed before, as well as the Belgian ZEE. Additionally, French political attitude to trade differs from practices of the hubs described before. The practice of favouring national companies could both pose some sort of barriers to the commercial development of the French hub and impact pricing behavior in the hub. Finally, the introduction of the trading platform, Capsquare, might have probably lead to some degree of interconnection between the Belgian ZEE and French PEG-N. There are, therefore, several reasons why the results of the regression are not statistically significant and the recommendation for the future research would be to account for them in order to obtain more valuable outcome.

The last hub investigated in this research is the Italian PSV. The results of the baseline regres-sion with the Italian PSV as a counterfactual are presented in column 7 and the specification with three months dropped as presented in column 8. From columns 7 and 8 it appears that virtually all coefficients are statistically significant and T reat ∗ post2 is one of them. It has a

relatively large coefficient of -3.820 that becomes even larger when the 3-month time period is dropped. It indicates a large change in the pricing scheme that occurred in the Zeebrugge hub after the implementation of the merger. Just as expected, the values on the variable Treat are negative and statistically significant in both specifications. This is clearly visible from the graph on the price trends presented in the data section, and it proves that the prices charged by ZEE were lower than those charged by PSV. All the signs of the remaining explanatory variables are as expected. The only variable that differs from the expected direction of the coefficient is the Temp. The direction of the coefficient is positive, which loosely means that the increase in temperature has a positive effect on the prices of gas. That is a counterin-tuitive outcome. As explained in the section treating the technicalities of the model used in this study, the counterfactual results often signal the omitted variable bias. At the time of the merger it was the least liquid hub out of the hubs tested. Not only did it show higher price levels than the remaining continental hubs but it also experienced some political problems that contributed to its already low degree of transparency. The price levels could have been driven by factors other than the market forces and the supply and demand characteristics considered in this research. Furthermore, the outcome may have been caused by the imperfect data on historical daily temperatures in Italy. Firstly, there is no publicly available data on historical daily temperatures observed in Italy. The data points used in this research were obtained from the registry of the temperatures created by the international air force. Then, the country daily averages were taken from a group of location-based observations. The above methodology is considered the best way of obtaining needed observations but, of course, the values may not be a perfect reflection of the reality. Moreover, Italy is a country spread across a long geo-graphical location. Some parts of the country are much warmer than others, so the average temperature may not be particularly informative and useful from the standpoint of this research.

(25)

Apart form the shortcomings explained above, the rest of the variables included in the econometric model show a great degree of statistical significance and the values of Treat*post inform about the large negative effect of the merger on prices charged in Belgian ZEE. Moreover, the R-squared statistics, that has a value of 0.921, informs that the model used in this study ex-plains a large part of the variance in data.

Overall, the output of the baseline regressions presented in Table 1, indicates that the merger of GDF and Suez with remedies in the form of divestitures of Distrigas, Distrigas & Co, and partial divestiture in Fluxys had a negative, and statistically significant, impact on the prices of gas in ZEE hub relative to the three out of four counterfactuals used in this research. The results suggest that the decision of the European Commission to clear the merger, and the remedies that were subsequently implemented, were successful in preventing the anticompeti-tive outcomes on the Belgian gas market. The merger and the remedies could have, in fact, be responsible for creating competition in that market.

The next section aims to check the robustness of the output obtained from the baseline analysis, and examines the validity of the conclusions made in this part of the paper.

3.3.2 The robustness check

Every econometric analysis that is based on assumption faces potential estimation issues that need to be addressed. So, the robustness checks should be performed in order to overcome them. The main idea behind the robustness check is to carry out the same analysis but under different specifications. Later, depending on whether the choice of the specification significantly impacted the results or not, the original results can be accepted as valid or disregarded as invalid. In this research, several methods of checking robustness of the results are used. The choice was inspired by the study performed by Argentesi et al. (2017). The reason this choice was made is that the methods used by the aforementioned researchers are widely regarded as the most relevant in retrospective studies, and it is also desirable to keep all five analyses performed in this paper and the paper mentioned above in the same general format (to make it readable and easy to compare). The usual issue that occurs while dealing with high frequency data is the problem of high auto-correlation across observations, which results in the high degree of auto-correlation in the error term. In order to tackle this issue, four alternative specifications were used. They included an estimation of the regression with Newey-West standard errors but this time including 1-lag specification. It differs from the original regression as the original model allowed for the autocorrelation up to 7-lags. Another way to account for the autocorrelation in standard errors is to bootstrap them. The idea behind bootstrapping is to mimic the process of random sampling in an infinite sample and later replace it with the original value. This methodology is a widely used tool to measure the accuracy of sample estimates, and therefore it is also used as a robustness check in this paper. Another issue that often occurs in the retro-spective evaluation studies is the problem of defining before/after periods. It is mostly due to particular anticipatory or delayed effects that the event at hand i s exposed to. In order to deal with this problem, the particular time period i s excluded from the analysis.

(26)

In this econometric analysis, two specifications are designed to deal with the aforementioned issue. They boil down to dropping the one and six-month time windows surrounding the event of the merger from the analysis. Finally, the specification using the logs is identified and tested. This robustness check yields coefficients that should be straightforward and easy to interpret for a reader as they take on a form of elasticity.

The robustness check of the results presented in column 1 of Table 1, confirms the validity of the original analysis carried out in this paper for the case with the German GPL acting as a counterfactual. The results of this robustness check are presented in Table A1 in Appendix 3. The main variable of interest, T reat ∗ post2, is negative and significant in all specifications. The

value is also similar across specifications. The only regression that results in a much smaller value of the interaction term is the specification using log methodology. The outcome may call for a cautious interpretation of the results, as the estimate for the effect of the merger on post-merger prices charged in the ZEE hub is smaller here than it is in other specifications. Notwithstanding, the very fact that the sign is still negative informs about the drop in prices as a consequence of a merger. All the remaining variables follow the pattern that is similar to the main regression. The coefficient on variable P ost2 is again positive across all of the

robustness check regressions. The demand and supply side shifters also follow the same canon and their significance level is mostly alike. The only variable that differs across specifications is the temperature squared. In some cases it is significant, at the 5% significance level, while in other cases it does not appear to have a significant effect on the price values.

The overall conclusion from the robustness check performed for the case of Gaspool is that the methodology used in this paper appears robust enough to conclude that the merger be-tween the GDF and Suez resulted in lower prices of gas traded in the Belgian ZEE hub relative to the German Gaspool. Therefore, the deductions from the baseline analysis should be correct. As for the robustness check of the results obtained for NCG, the outcome is presented in Table A2 available in Appendix 3. The robustness check yields results that are highly in line with those obtained from the original baseline analysis that allows for the correlation up to 7 lags. The direction of the coefficient on T reat ∗ P ost2 is negative and significant for all specifications.

The values of those coefficients, for regressions presented in columns 3, and 4, are almost iden-tical to the original values from baseline analysis given in column 1. The only difference lies in the values for standard errors. While dropping one month and six months time windows surrounding the event, the magnitude of the effect of the event on the price drop has increased but, also, slightly in both cases. The only coefficient that is substantially different from the expected value is the coefficient of T reat ∗ P ost2 presented in column 2. The aforementioned

coefficient is, in fact, negative and therefore in line with expectations, but its lower value calls for a cautious interpretation of the results as it suggests that the impact of the merger on the prices charged in the Belgian ZEE may be somewhat smaller.

(27)

group, offer some interesting findings. The outcomes are presented in the Table A3 in Appendix 3. As one can see from the Table A3, the results from the bootstrap presented in column 4 and the results from Newey-West standard errors with 1 lag autocorrelation, bring about the same outcomes as the outcomes of the original analysis. The level of significance varies slightly between the specifications but it does not undermine the validity of the originally obtained results. While a 1-month time period is dropped from the analysis, the results yield different outcomes but they are similar to the original ones. While dropping a 6-month window, however, the results change significantly. The coefficient on T reat ∗ P ost2that was initially insignificant became significant

(at the 5% significance level) in the latter specification. The value of that coefficient also indicates that the merger caused an average drop in the prices charged in Zeebrugge hub by 0.881, which is similar to the outcome obtained while using German NCG as a counterfactual. Moreover, the logarithmic specification yields statistically significant results for Treat ∗ Post₂that is negative and significant at 5% significance level. The value of the coefficient is, nevertheless, much smaller than in the regression with time window dropped. Overall, it is hard to make an unambiguous conclusion about the effect of the merger on prices charged in ZEE hub while treating the French hub as a counterfactual. It may be due to the fact that the differences between the Belgian ZEE and French PEG-N are too significant or that the level of interconnection between the hubs is too significant to use the French hub as a counterfactual.

The last robustness check is that of the Italian PSV. Table A4 (Appendix 3) gathers results of the robustness check regressions for this hub. The robustness check regressions are closely in line with the baseline results. The lesson is that the conclusions made while analysis baseline regressions should be confirmed.

4 Conclusions

This case study aimed at evaluating the effect of the merger between Gaz de France and Suez on day-ahead wholesale prices of gas charged in the Belgian Zeebrugge hub. The econometric approach used in this paper was Difference-in-Difference methodology with the Belgian ZEE acting as a control group, and two German hubs Gaspool and NCG, a French PEG-N, and Italian PSV serving as counterfactuals.

The results of this research show that the decision of European Commission to clear the merger between Gaz de France and Suez has negatively impacted prices charged in the Belgian ZEE hub as compared to other European hubs under investigation.12 Therefore, the merger clear-ance, and the remedies implemented, as a consequence of it, could have had a positive impact on competition in the ZEE hub. The above conclusions are made based on the body of

econo-12_{The only instance in which the valuable conclusions about the effects of the GDF/Suez merger on prices}

charged in ZEE hub cannot be made is the case when French PEG-N serves as a control group. In this case the coefficient on the variable T reat ∗ post2 is indeed negative which would indicate a negative impact on prices but

it is not statistically significant.