Does central clearing of over the counter derivatives reduce risk? Effect
of European Market Infrastructure Regulation on clearing participants in
Europe for the period of 2009 to 2015
MSc Thesis Name: Anastasia Rusanova Student number: 10598057 Supervisor: Simas Kucinskas Date: 01.07.2018 Programme: MSc Finance, specialisation: Quantitative Finance Amsterdam Business School, University of Amsterdam
Statement of Originality This document is written by Student Anastasia Rusanova 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.
Abstract This paper investigates the effect of central clearing on clearing participants, using the European Market Infrastructure Regulation to conduct an event study. In total 153 European firms for the period of 2009 to 2015 have been analysed, with 72 clearing members and clients as a treatment group and 81 comparable control group firms. The main finding is a reduction of systematic risk proxied by rolling market betas of the clearing participants of estimated 20%. No significant effect on uncertainty, represented by return rolling volatility has been found. Results are robust to rolling window specifications and sovereign CDS bond spreads as control variables in contrast to dummies. These findings advocate for a possible standardization and further improvement of central clearing counterparty risk.
Table of Contents 1. Introduction 5 2. Literature review 7 2.1. CCP Institutional background: 7 2.2. EMIR Regulation 9 2.3. Related literature 10 3. Methodology 14 4. Data 18 5. Results 23 6. Robustness tests 28 7. Conclusion and Discussion 33 Bibliography 36 Appendix 41
1. Introduction In the wake of the 2007-8 financial crisis the regulators strived to improve the health of the financial system. Under attention came the over the counter (OTC) derivatives market, since concerns have been expressed about its opaqueness, lack of regulation and potential harbouring of risks (Gugerell, 2010). Moreover, in January 2010, the chairman of the Commodity Futures Trading Commission claimed that over the counter derivatives were part of the reason for the 2007-8 financial crisis (Meyer, 2010). Therefore, a European Market Infrastructure Regulation (EMIR) has been developed and adopted on July 4th 2012, under which all standardized OTCs must be settled through a central clearing counterparty (CCP). A CCP interposes itself between parties and becomes a seller to every buyer and a buyer to every seller. With this, it was hoped that the OTC market would become more transparent and secure. However, there has been a debate about the impact of central clearing among academic scholars as well as policy makers. After the collapse of Lehman Brothers its derivative positions were successfully and swiftly unwound by CCPs, with many believing that mitigated adverse outcomes (Gugerell, 2010). Furthermore, it is widely argued that settlements through CCPs decrease the counterparty risk of default (Acharya, Engle, Figlewski, Lynch, and Subrahmanyam, 2009). However, concerns have been raised that risk will not be eliminated, but shifted to CCPs instead, making them concentrated points of systemic risk (Singh, 2011; Menkveld, 2014). Moreover, according to the Bank for International Settlements (2012) at the end of 2011, the OTC market had a notional principal of $648 trillion. As Loon and Zhong (2014, p. 2) note, OTC central settlement would have far reaching consequences. Nevertheless, with the debate ongoing, there has been little empirical evidence on the performance of CCP structures, which this thesis will strive to contribute to.
Hence, the research question is: Does mandatory settlement of over the counter derivatives through a central clearing counterparty under European Market Infrastructure Regulation decrease risk of parties previously involved in bilateral settlement in Europe in the period from 2009 to 2015? Previous literature has mainly examined the derivatives market, analysing how derivative key characteristics behave before and after central clearing and during crises times. However, no definite consensus has been reached. In line with Jones and Perignon (2013), who study the clearing members and losses on CCP accounts, this paper studies the clearing members directly. I employ risk measures used by Sarin and Summers (2016), who estimate the impact of post crisis regulation on US as well as international firms. In particular, share price volatility and market betas are estimated. The adoption of EMIR regulation is used to conduct an event study. Share prices of 71 firms involved in clearing settlement has been collected. To ensure that the research design centres on the impact of central clearing a difference in difference method is implemented, where 81 additional European firms of similar size not involved in central clearing are included in the analysis. To have the balance of data before and after regulation, the period from 2009 to 2015 is studied. Robustness tests using sovereign bond CDS spreads are conducted. The remainder of this paper is organized as follows: section two contains the literature review, with an overview of CCP institutional background, EMIR regulation and the related literature. Section three and four contain the methodology and collected data. Section five and six present results and robustness tests. Finally, section seven concludes with a discussion.
2. Literature review 2.1. CCP Institutional background: I focus on analysing the risk inherent in CCPs, in order to do that it is important to understand how they operate. I start with a brief description of the CCP mechanism. Under EMIR all standardized derivatives must be settled through a Central Clearing Party affiliated with a trading platform, i.e. Bloomberg, Reuter, SwapsMonitor (Jones & Perignon, 2013, p. 4). A relationship with a CCP is formed through two legal mechanisms - either "novation" or "open offer", under which the CCP becomes a principal to both parties that agree on a contract on the trading platform (Steigerwald, 2013, p. 2). It is important to note that CCPs' risk profile is different from that of a bank or an insurer. Whereas banks and insurers run unmatched books for the purpose of making a profit, CCPs' positions are always matched, except for the case of a member default. CCPs' primary role and obligation to its members is to act as a regulator, supervisor and loss allocator for the markets that they clear. Therefore, unlike banks and insurance companies, CCPs are exposed only to default risks of their members. To use CCP services, a company must either posses a clearing membership or a be client of a clearing member (CM). In case of a default of one of the clearing members, losses must be allocated promptly, as otherwise the CCP is exposed to the market due to an unmatched positions book and faces risk of default. Hence, CCPs have a set of measures termed "waterfall loss sharing" which they can consecutively employ to cover the losses. First of all, initial margins are posted by clearing members on accounts held with a CCP for the purpose of covering open and possible future exposures. Initial margins can take a form of either securities or cash. According to Jones and Perignon (2013, p. 5), the most prevalent system for margin calculation is the Standard Portfolio Analysis of Risk (SPAN)
margining system. This system considers 16 potential one day changes (i.e. unchanged, Up / Down 33%, Up/Down 67% etc.) in the key parameters of the underlying securities such as price, volatility and time to expiration, which are defined by clearing houses (CME SPAN, 2010). For each scenario, losses and gains are calculated. The total member's margin then is given as the largest daily loss a portfolio class might incur at a statistical confidence level of 95-99%, with an adjustment for diversification benefits (CFTC, 2001, p. 4). Moreover, at the end of each day realized exposure gains and losses are calculated and if the posted margins fall below mandatory levels extra variation margin might be required. As Jones and Perington (2013, p. 3) state, clearing members' and their clients' margin accounts are held separately, and whereas member margins are based on the net exposures, client margins are calculated on a gross basis, with no offsetting between positions. This, as is noted (Jones & Perington 2013, p. 4), contributes to the protection of clients' funds. Secondly, CCPs hold default funds (DF) which are designed to cover losses in extreme scenarios, such as members' defaults (CFTC, 2001, p. 4). The fund consists of clearing members' capital, where each member contributes in proportion to its risk. When a member defaults, its own contributed capital is used first (De Genaro, 2016, p. 2). Generally, CCPs set the size of DF and IM to cover extreme stress scenarios. Moreover, CCPs dedicate part of its own capital in tranches at various stages of the loss waterfall, as most CCPs are run for profit, this aligns CCPs incentives with that of other members and helps reduce potential CCP moral hazard issues, such as insufficient monitoring and improper risk measures. There exists a number of other measures a CCP can implement during member default. Such measures include the CCP's ability to request more default fund capital from clearing members, partial or full termination of contracts and forced allocation of
unmatched positions to a non-defaulting member. As Bank of England noted in 2011 (p. 53), not all loss waterfall structures are formally and clearly defined and under some measures loss allocation is to the discretion of CCP. Nevertheless, CCPs generally adopt either a survivor pay or defaulter pay models (De Genaro, 2016, p. 13). Under the former, initial margins are minimized, with its advocates arguing that this helps facilitate clearing members' involvement in the post-default procedures. Under the latter, margins on the contrary, are maximized, with the main argument being decreased moral hazard of clearing members. 2.2. EMIR Regulation In this section I will briefly outline the European Market Infrastructure Regulation and how it helps facilitate the research. There are three main ways in which EMIR regulation influences the financial institutions (ESMA, 2012). Firstly, derivative trades must be reported within 2 working days of the trading day to an authorised trade repository, which is then monitored by a European regulatory body (ESMA, 2012, p. 11). Secondly, OTC derivative contracts must be settled through a Central Clearing Counter Party. Under the regulation, entities are divided into financial (FC) and non financial counterparties (NFC) (Mariottini & Camara, 2012). Mandatory CCP settlement applies to FCs (i.e. banks, insurance companies etc.) and NFCs that exceed specified position thresholds, termed NFC+. Thirdly, FCs and NFCs+ must implement enhanced risk management practices associated with settlement of derivatives, with two major ones being: 1) periodic portfolio reconciliation, where both parties compare the records and key details of each trade, including market-to-market valuation; 2) agreeing on the procedure for dispute resolutions between parties relating to minor discrepancies on portfolio valuation and collateral exchange (ESMA, 2012). As the reporting and enhanced risk management practices are related to the
operational aspects of entities and do not appear to be significantly connected to the nature of entities' risk profiles, I expect the main effect on economy risk of EMIR regulation to be the settlement of OTC derivatives through CCPs. From the 2.1. Institutional background section above it is clear that settlement through a CCP changes the riskiness of the trades in question, therefore affecting the riskiness of entities engaged in those trades. In the following section I will discuss theoretical predictions and empirical findings related to settlement through CCPs and will develop my hypothesis. 2.3. Related literature One area of theoretical literature on OTC markets suggests that settlements through CCPs might make the market more transparent and lead to reduced moral hazard (Cecchetti, Gyntelberg & Hollanders, 2009; Duffie & Lubke, 2010). This has been modelled by Acharya and Bisin (2014). They explain (2014, p. 2), that when markets are opaque, counterparty risk externalities arise with entities taking on additional risk which increases their probability of default. However, with introduction of CCP settlement and transparency of members' and clients' positions, those externalities are eliminated, as entities take into account each others' exposures and impose risk premiums. Nevertheless, CCPs make this information public only through a centralized registry, as Acharya and Bisin (2014, p. 26) note, this is might not be enough. It could be plausible that companies might take advantage of that information, although it may require software or labour investments. Overall, proponents of CCP settlement (Acharya, Engle, Figlewski, Lynch & Subrahmanyam, 2009; Duffie & Zhu, 2011) argue that with increased control, appropriate risk measures against default and position netting efficiencies, economy risk will go down.
On the other hand, opponents of CCP settlements argue that with capital contribution and loss mutualisation between members, CCP structures become concentrated nodes of systemic risk that could exacerbate extreme times (Singh, 2011; Menkveld, 2014). Additional concerns have been expressed about the CCPs' design. Menkvel (2014) and Cruz Lopez, Harris, Hurlin and Perignon (2012) point out that CCPs do not implement monitoring of trades on a single factor in their capital requirements. Due to factor trading, clearing members' losses become correlated in adverse times, which is not reflected in the margin measuring system SPAN (Mekveld, 2014, p. 6). Empirical evidence for the clustering of CCP member losses has been presented by Jones and Perignon (2013), who study Chicago Mercantile Exchange CCP's members in the period from 1999 to 2001. Moreover, theoretical literature also suggests that CCPs might encourage members' moral hazard (Pirrong, 2011; Koeppl, Monnet & Temzelides, 2012; Biais, Heider & Hoerova, 2012). As Biais et al.'s (2012) model shows, clearing members and their clients, feeling insured, might be discouraged from performing risk due diligence on their counterparties. This can possibly counterbalance the benefits of OTC market transparency described above, as members would not use the provided information. Previous empirical work on the effects of CCPs is, by the words of Boissel, Derrien, Ords and Thesmar (2017, p. 3), relatively nascent. Among the studies, is one by Loon and Zhong (2016), who look at the impact of the Dodd-Frank act on trading costs and liquidity in the OTC index CDS market in the United States during the year 2013. Their main findings are that trades cleared by CCPs exhibit lower transaction costs, as calculated by the relative effective bid ask spread, and liquidity improvement, taken as the Amihud and Roll measures. This, as noted (2016, p. 647), could be attributed to higher post-trade transparency, which, in my opinion, could balance the moral hazard argument and support
the transparency claims. Similarly to their paper, this thesis will take the EMIR regulation an event for the analysis. Another one of their studies (Loon & Zhong, 2014), inspects effects of central clearing on the CDS spreads for North American firms between 2009 and 2011. They find, that as central clearing begins, the relationship between CDS spreads and dealer credit risk weakens. The lowering of default correlation suggests, that investors are reassured about CCP's ability to insulate the counterparties from each other's default in the CDS market (2014, p. 22). This comes in opposition against Singh's (2011) concerns about CCPs increasing systemic risks. However, as is noted in the Global Risk report of the World Economic Forum (WEC, 2015), the various crises occurring during the period of 2007-2015 have revealed how weak the global economy is and shifted the markets' perspectives. Therefore, in my opinion, another study with more recent information is needed to confirm Loon and Zhong's (2014) findings. In Europe, two studies examine the repurchase (repo) market (Mancini, Ranaldo & Wrampelmeyer, 2015; Boissel et al. 2017). Mancini et al. (2015) study the trades cleared by CCPs on the repo interbank market in the period from 2006 to 2013. Their analysis focuses on the relationship between repo characteristics (i.e. spreads, maturities and haircuts) and a systemic stress index, which is based on volatilities from 11 separate markets. They find that during crisis times this relationship is insignificant and conclude that CCPs act as market shock absorbers (2015, p. 29). However, Boissel et al. (2017) show that in the period from 2008 to 2012, European CCP cleared repo trades respond to sovereign risk as priced by the CDS spread. They explain, that the sovereign bonds are common collateral for repo trades, with the lender receiving sovereign bonds in the event of default. Moreover, CCP cleared trades are not less sensitive than bilateral ones (2017, p. 2). These findings signify that CCPs were unable to isolate the sovereign risk of default (2017, p. 18). As they explain, this could
be because CCPs are perceived as 'too big to fail' institutions, with governments acting as lenders of last resort. However, reliance of too big to fail institutions on the governments is well documented in the literature and does not allow to conclude whether CCP structures increase risk. From the above, there is not enough direct evidence on whether the CCP structures reduce or increase risk in the economy, which this thesis will strive to fill. In contrast to the repo studies described above, I will take the approach of Jones and Perignon (2013) and directly examine CCP clearing members and clients. In line with Boissel et al., I will incorporate government bond CDS spreads in robustness tests. Overall, the literature does not provide a clear conclusion on whether settlements through CCPs increase risk. In my opinion, Menkveld's (2014) crowded trades argument is a strong one, and, furthermore, loss mutualisation increases interconnectedness, which with a large enough shock could cause significant harm. In bilateral trading, exposure to one counterparty could be diversified, however, with CCPs, counterparty risk becomes more systemic in nature. My hypothesis, therefore, is that settlements through CCPs increase the counterparties' risk. As will be described in the methodology section below, due to research limitations, I will analyse the market perspective, utilizing the information inherent in members' and clients' share prices. Hence, I expect the following outcome: Research Hypothesis: According to the market, mandatory settlement through a central clearing counterparty increases risk of parties previously involved in bilateral settlement.
3. Methodology In this section I will outline the methodology that will be used to conduct the analysis. As mentioned in the literature review section earlier, in order to have an indication of the risk inherent in CCPs, I will follow Jones and Perignon (2013) and analyse the clearing members directly. Primarily, the share prices will be used for the analysis. The rationale is the following, as EMIR regulation made mandatory CCP settlement of OTC derivatives, it is logical to expect that the average settlement through CCPs increased. As EMIR was adopted in July 2012 and came into force in August 2012, to form a balanced dataset, the sample will span a period from 2009 to 2015. In order to make sure that I focus on firms affected by the regulation, and therefore, firms with increase in settlements through CCPs, the sample will contain either CCP members or clients of the members. To isolate the EMIR effect, a difference in difference model will be used, where comparable non-CCP member/client firms will furthermore be included in the analysis as a control group. Two measures of risk as suggested by the literature will be used and analysed, that is, CAPM betas and stock return volatility. These measures are in line with Sarin and Summers (2016), who analyse the riskiness of banks in the United States as an evaluation of post 2007-8 financial crisis regulation. As cited by them (2016, p. 6), Bulow and Klemperer (2013, 2015) and Haldane (2013) state that regulatory measures are poor historic predictors for financial institution failures. They further explain, that, by contrast, financial indicators on debt and equity have shown early signs of underlying economic problems. To compute the CAPM betas, the following regression (1) will be run with robust standard errors and firm clustering:
(1) 𝑅",$− 𝑟',$ = 𝛼",$ + 𝛽",$ 𝑅-,$ − 𝑟',$ + 𝜀",$
Where 𝑅",$ is the stock return of company 𝑖 at time 𝑡, 𝑟',$ is the risk free rate at time 𝑡, with 𝑅",$ − 𝑟',$ representing the excess company return. 𝑅-,$ is the market return at time 𝑡, and 𝛽",$ is the coefficient of company 𝑖 at time 𝑡 on the excess market return 𝑅-,$ − 𝑟',$ . 𝛼",$ and 𝜀",$ are correspondingly the constant and error terms of company 𝑖 at time 𝑡. Sarin and Summers (2016) estimate the betas on an annual basis, in comparison, I will compute rolling betas, this choice will allow to gain additional insight into beta dynamics over the period. Windows of 100, 50 and 200 observations will be used to evaluate the sensitivity of results. 𝛽",$ (beta) is therefore, the first measure of risk. Cederburg and O'Doherty (2016, p. 3) note that the beta is commonly used to gauge the level systematic risk in stock returns. Moreover, Jagannathan and Want (1996) explain that the rolling beta has greater explanatory power over the static beta, and provide evidence that structural parameters of stocks change over time. The resulting betas are averaged for each date, with equal weights between companies and analysed further, where the below regression (2) will be run with robust standard errors and clustering on the firm level: (2) 𝛽",$ = 𝛾2,$𝑅𝑒𝑔𝑢𝑙𝑎𝑡𝑖𝑜𝑛 + 𝛾:,$𝑅𝑒𝑔𝑢𝑙𝑎𝑡𝑖𝑜𝑛×𝐶𝑜𝑛𝑡𝑟𝑜𝑙 + 𝛾=,$𝐶𝑜𝑛𝑡𝑟𝑜𝑙 + 𝛾>,$𝐶𝑟𝑖𝑠𝑖𝑠2+ 𝛾@,$𝐶𝑟𝑖𝑠𝑖𝑠:+ 𝜉",$ From above, 𝑅𝑒𝑔𝑢𝑙𝑎𝑡𝑖𝑜𝑛 is the dummy variable that is equal to 0 if the data is before the effective date of the regulation and to 1 if it is after, the exact date taken is 2nd of July 2012, the effective date, when market expect the companies to have already started implementing measures of EMIR compliance. I expect its coefficient, 𝛾2,$ to reflect a general
trend in risk of the European economy during the analysed period. It could be expected that this coefficient will be positive and significant. As mentioned previously the WEC Global Risk report (WEC, 2015), points out that past crises have shown the weakness of the global economy. As the 2007-8 financial crisis and the European sovereign debt crisis happened majorly before the cut off date of the regulation, hence, I hypothesise a higher perceived risk by the market after the cut off. 𝐶𝑜𝑛𝑡𝑟𝑜𝑙 is a dummy equal to 1 if the company is a CCP-member or a client of a member and 0 it belongs to a control group. 𝐶𝑟𝑖𝑠𝑖𝑠2,: are the dummies controlling for the financial and debt crisis, with 𝜉",$ being the error term. 𝐶𝑟𝑖𝑠𝑖𝑠2 is equal to 1 if the data is in the year 2009 and 0 otherwise. This dummy will allow to separate the after effects of the 2007-8 financial crisis from the rest of the results. As noted by IMF in the 2010 World Outlook Report, 2009 was a still year of financial instability, with uneven and gradual recovery, with a more stable economy towards the end of the year. As a result, I expect the coefficient on 𝐶𝑟𝑖𝑠𝑖𝑠2 to be significantly positive, with high betas and hence exposure to the market during that time. 𝐶𝑟𝑖𝑠𝑖𝑠: is equal to 1 for the years 2011 and 2012 and 0 otherwise. These years were chosen as according to et al. (2015, p. 10) 2011-2 were years of a substantial rise in risk as proxied by the repo spreads due to the European sovereign debt crisis. Similarly to 𝐶𝑟𝑖𝑠𝑖𝑠2, I expect the coefficient on 𝐶𝑟𝑖𝑠𝑖𝑠: to be significantly positive, perhaps with a larger economic effect. Further analysis with Europe sovereign bond CDS spreads will conducted as a robustness check. The 𝑅𝑒𝑔𝑢𝑙𝑎𝑡𝑖𝑜𝑛×𝐶𝑜𝑛𝑡𝑟𝑜𝑙 is the dummy interaction term. The coefficient of interest therefore, is 𝛾:,$, which could be interpreted as the average change in betas between the treatment and control groups after the implementation of the regulation. A positive significant coefficient would indicate an increase in betas, signalling increase in exposure of
the company to the overall market, and hence an increase in risk. The statistical hypotheses therefore are: H0: 𝛾:,$ = 0 H1: 𝛾:,$ ≠ 0 If the coefficient is statistically significant from 0, then the null hypothesis can be rejected. The sign of the coefficient will determine whether the research hypothesis of increased risk can be rejected. The second risk measure will be volatility, taken as the standard deviation of stock returns. This is in also line with Sarin and Summers (2016), however as with the betas, a rolling basis of 100, 50 and 200 days will be used for comparison. After computing the volatility, the following regression (3), similar to regression (2) will be run, with robust standard errors at firm level clustering: (3) 𝜎",$ = 𝜃2,$𝑅𝑒𝑔𝑢𝑙𝑎𝑡𝑖𝑜𝑛 + 𝜃:,$𝑅𝑒𝑔𝑢𝑙𝑎𝑡𝑖𝑜𝑛×𝐶𝑜𝑛𝑡𝑟𝑜𝑙 + 𝜃=,$𝐶𝑜𝑛𝑡𝑟𝑜𝑙 + 𝜃>,$𝐶𝑟𝑖𝑠𝑖𝑠2 + 𝜃@,$𝐶𝑟𝑖𝑠𝑖𝑠: + 𝜈",$ Here, the dummies 𝑅𝑒𝑔𝑢𝑎𝑙𝑡𝑖𝑜𝑛, 𝐶𝑜𝑛𝑡𝑟𝑜𝑙, 𝐶𝑟𝑖𝑠𝑖𝑠2 and 𝐶𝑟𝑖𝑠𝑖𝑠: have the same meaning as in regression (2), with 𝜃2..@,$ being the corresponding coefficients and 𝜈",$ the error term. As with regression (2), robustness tests will be conducted using CDS spreads. For coefficients 𝜃2,$, 𝜃>,$ and 𝜃@,$ my expectation is the same as for coefficients 𝛾2,$, 𝛾>,$ and 𝛾@,$ in
regression (2). In regression (3), the coefficient of interest is the coefficient on the
significant and negative number would indicate a decrease in volatility. For all regressions firm fixed effects with robust standard errors will be done. Statistical hypotheses: H0: 𝜃:,$= 0 H1: 𝜃:,$≠ 0 4. Data In this section the sample data will be described. For the control group, lists of CCP members and clients are easily accessible and were downloaded from the websites of 17 CCPs, authorized to operate in the European Union under EMIR. Often companies are members / clients in more than one CCP, with the large majority being privately owned. Due to this and data availability for the period 2009 to 2015, the treatment group sub-sample contains 72 firms. All of the firms are big institutions, with majority being banks and insurances, hence for the control group companies with highest capitalization across Europe part of EUROPE STOXX Large index, representing the market were selected. Due to data availability and removal of similar companies as in the treatment group, the control group consists of 81 firms. The full sample is 153 firms, with a complete firm list is included in Appendix 1 in Table 2. STOXX Europe 600 was chosen as a proxy for the market return. This index represents large, middle and small capitalization companies across Europe. Stock prices for the companies and the index level were downloaded from Datastream for 2009 to 2015 with a daily frequency. Next, returns were
manually computed. In order to remove any outliers that might come with Datastream, the company and market returns were winsorized from both tails with percentile cut-offs of 2.5 and 97.5. All statistic manipulations were done statistical software STATA. The risk-free rate is chosen as the 10 year German government bond yield, which is in line with Boissel et al. (2017), who state that German bonds are regarded as the safest in the European Union. Hence, the daily yield was downloaded from Datastream. The volatility is computed as the daily rolling sample standard deviation per company with a window of 100 days. The first 100 return observations for each company then have no corresponding volatilities. Data summary statistics are presented in Table 1, displayed for the pre regulation period in Panel A and and post regulation in Panel B, returns are further separated per treatment and control groups. Henceforth, I will refer to the computed rolling standard deviation of returns as 'volatility' and static return deviations as 'standard deviations'. Table 1 shows volatilities for the window of 100 observations. Complete statistics of volatility for the 50 and 200 observations can be found in Appendix in Table 3. The data is strongly balanced, with the total number of observations for each variable for the whole period approximately equal to 268,000. From Table 1, it can be seen that treatment and control group returns are close to 0. The treatment group returns fluctuate more than control group during both periods, with their volatilities being correspondingly 0.0243 and 0.0197 pre and 0.0191 and 0.0158 post regulation. As for the market, its average return appears to be the same and close to 0 0.0006 during two periods. Post regulation the market experienced a reduction in standard deviation as well, equal to 0.0133 before and 0.0097 after. This suggests that the difference in difference method is justified and necessary, as there appears to be a general trend of decreasing volatility. It is worth noting that both
treatment and control groups' returns overall fluctuate noticeably more than the market by a minimum of 0.013 standard deviation points. Table 1. Summary Statistics for the key variables of the main results This table contains the summary statistics for the entire sample period of 6 years for the data. (2009 to 2015). The statistics are broken down into Panels A and B: pre regulation and post regulation subsamples. Returns and return volatility are displayed for the 72 treatment firms (CCP members or their clients, who are involved in settlement) and 81 control firms (comparable market representative firms comprising the EUROPE STOXX Large index). Return volatility is given for the window of 100 trading days. Risk free return is the 10 year German bond yield. Market return is the STOXX EUROPE 600. All the data is daily frequency.
Panel A: Pre regulation No of Obs. Mean Median Std. Dev. Min. Max.
Treatment group return 63,360 0.0002 0 0.0292 -0.3559 0.5035 Control group return 70,936 0.0005 0.0002 0.0221 -0.2010 0.2711 Risk free return 134,640 0.0290 0.0300 0.0065 0.0146 0.0396 Market return 134,640 0.0003 0.0006 0.0133 -0.0498 0.0715 Treatment group volatility 56,232 0.0243 0.0217 0.0131 0.0001 0.1053 Control group volatility 62,917 0.0197 0.0182 0.0076 0.0071 0.0799 Panel B: Post regulation Treatment group return 62,784 0.0004 0 0.0225 -0.3056 0.4172 Control group return 70,632 0.0007 0.0005 0.0163 -0.1566 0.3394 Risk free return 133,416 0.0224 0.0224 0.0038 0.0139 0.0301 Market return 133,416 0.0005 0.0006 0.0097 -0.0533 0.0420 Treatment group volatility 62,784 0.0191 0.0169 0.0119 0.0003 0.1287 Control group volatility 70,632 0.0158 0.0151 0.0047 0.0066 0.0493 Figure 1 presents the 100 trading days rolling volatility plotted for the whole period for the two sub samples: treatment and control. As was noted, the treatment group's volatility is historically higher, with a sharper spike during the time of the European sovereign debt crisis. The graph shows that the two lines have similar dynamics. In the beginning of the period, the volatility was elevated, with the two lines starting out around 0.041 and 0.031 and further decreasing, with moderate elevation during 2010. Apart from the additional
mentioned increase during the debt crisis, volatilities appear to be between 0.01 and 0.024, with a slight increase at the end of the period for both the control and treatment groups. Figure 1. Rolling volatility, window 100 observations, plotted for the whole period. This figure displays rolling volatility, computed as the rolling sample standard deviation of firm returns with a window of 100 observations. The volatility is separated by treatment and control groups. The treatment group contains firms that are either CCP members or their settlement clients in Europe with a total of 72, whereas the control group includes 81 other market representative companies. Figure 2, Panel A presents volatility as computed for 200 observations and equivalently broken down by the two samples. 200 observations allow for better long-term trend comparison and would be more in line with analysis conducted by Sarin and Summers (2016) who compare the yearly (260 observations) standard deviations and betas to asses the change in risk. The same trends are noticeable as in Figure 1, moreover, the slight volatility increase in the right end is more pronounced. This might indicate to a preliminary backing of the hypothesis. Volatility as computed for a window of 50 observations is presented in Figure 2, Panel B, the dynamics are the same as in Figure 1, although with
peaks and lows more pronounced. Judging from this, a window of 100 observations appears to balance relatively well between short-term and long-term trends. For all graphs, equal weights are used between firms. Figure 2. Rolling volatility, windows 200 and 50 observations, plotted for the whole period. This figure presents rolling volatility, taken as the rolling sample standard deviation of firm returns with windows of 200 and 50 trading days, shown in Panel A and Panel B respectively. Equal weights in sub samples are used. The trend is shown separately for the two sub samples. The treatment group contains firms that are either CCP members or their settlement clients in Europe with a total of 72, whereas the control group includes 81 other market representative companies. Panel A Panel B
5. Results First, regression (1) was run on the full sample with a window of 100 observations. The resulting betas are plotted in Figure 4 below separately for the control and treatment sub-samples. There appear to be similarities between the betas and volatility in Figure 1. Firstly, as with the volatility in Figure 1, the two lines appear to trace each other, however they start diverging in the middle of the year 2014. In the beginning of the period, treatment group's betas are higher that control group's, starting at 1.25 and 1 respectively and both changing to around 0.95 - 1 in 2010. The figure shows that years 2011 and 2012 are periods of high exposures to the market, with betas varying around 1.15. Consequently, there is a sharp decrease and a subsequent sharp increase in the year 2013, betas further fluctuate between 0.9 and 1.1, finally diverging in the middle of 2014. In the second half of 2015, treatment group betas are noticeably lower, fluctuating around 0.6, with control group betas averaging around 1. Unlike volatilities in Figure 1, betas imply a different dynamic of risk after the regulation. Whereas volatilities show a slight increase in risk, with treatment group riskier throughout the period, betas show a decrease in treatment group's risk in Figure 3, with control group betas unchanged on average after the regulation. This is in contrast to Sarin and Summers (2016), who find that the betas of the largest US financial institutions post financial crisis regulation period of 2014-15 increased relative to the as pre crisis years of 2004-5. Moreover, they find a visibly larger increase in volatilities. Figure 3. Rolling betas, window 100 observations, plotted for the whole period. This figure contains computed rolling betas with a window of 100 observations, denoted as 𝛽",$ in regression (1). The betas are plotted for the whole period of analysis and shown separately by treatment and control groups. Equal weights in sub samples are used.
Figure 4. Rolling betas, window 200 observations, plotted for the whole period. This figure contains computed rolling betas with a window of 200 observations, denoted as 𝛽",$ in regression (1). The betas are plotted for the whole period of analysis and shown separately by treatment and control groups. Equal weights in sub samples are used.
Figure 5. Rolling betas, window 50 observations, plotted for the whole period. This figure contains computed rolling betas with a window of 50 observations, denoted as 𝛽",$ in regression (1). The betas are plotted for the whole period of analysis and shown separately by treatment and control groups. For comparison, betas computed for 200 and 50 observations are presented in Figures 4 and 5 respectively. In comparison to volatility, the betas appear to be more volatile, which could be seen when comparing Figures 5 and 2 (panel B) with windows of 50 observations. In Figure 5, there is a lot of noise, with long-term trends hardly visible. It occurs, that a window of 200 for betas in more optimal, as it displays long-term trends clearly without masking short-term prominent dynamics. The results of regressions with betas and volatility as the dependent variable (regressions (2) and (3)) are summarized in Table 4. From the below, it appears that a window of 100 observations is the best fitting specification for the volatility regressions, with highest R squared of 0.1278. For beta regressions, however, the 200 window specification explains more data. Overall, a higher percentage of the volatility data than betas data is explained, with the highest R squared for betas equal to 0.0221 respectively.
Table 4 provides the answer to the research question. The coefficients on the Regulation and Control interaction variable are negative in all specifications, hence there is not enough statistical evidence to support the research hypothesis of increased risk of clearing members and clients due to central clearing. By contrast, the negative coefficients suggest a reduction in risk. From Table 4, with betas as the depended variable, the 100 windows regression coefficients for the regulation and control dummies interaction term is -0.0977, for the 200 window -0.1371, with the 50 window coefficient in between. For the volatility regression, the interaction variable coefficients range from -0.007 to -0.0015, with 50 window lowest and 200 window highest. The coefficients, however, are significant for beta regressions only. The significance is at relatively a strong level, at 5% for window of 100 and 1% for windows of 50 and 200 observations. The interpretation of such coefficients are: for the rolling window of 100 observations, after the regulation came into effect the exposure to the market, proxied by betas, decreased on average by 0.0977 points relative to the control group. The economic effect of this is important as well, as for a window of 200 observations, from Figure 4, the minimum and maximum levels of treatment group betas have an interval of around 0.5, which would make the coefficient of 0.0977 approximately equal to 20% of that. This is considering that the coefficients in other window specifications are more negative by around 0.04. The interaction dummy coefficients of the volatility regression are also negative, however, insignificant. The economic scale is less too, as for the 100 observation window regression the decrease in risk in treatment group relative to control in post regulation period is 0.0013, which is 9% of the standard deviation of pre regulation treatment group volatility from Table 1. These findings are in contrast to Sarin and Summers, who, as
mentioned previously, find both volatilities and betas of large financial institutions in the US throughout the period of 2004 to 2015 have increased. However, by contrast, this thesis's focus is at the institutions involved in central clearing. More comparable are the results of Loon Zhong (2014, p. 22), who find that for North American firms during 2009 - 2011 central clearing of trades weakens the relationship between the CDS spread and dealer credit risk. This, as they note, suggests lower systemic risk. For the European sample of firms in this thesis, systemic risk, proxied by the betas, also goes down by an average (across specifications) of 0.1239 points after the regulation and relative to the control group. The Control dummy coefficient, equal to 1 for the treatment sub sample and 0 for the control, suggests that on average, the treatment group is more volatile, with coefficients positive and significant, which was also clearly seen in Figure 2, panel A. This could be due to sample selection. From Table 2 in Appendix, most of the firms in the treatment group are financial institutions, whereas a significant part of the control group is non-financial. Higher volatility of the treatment firms could suggest that they are surrounded by larger uncertainty about their true riskiness. However, there are no significant differences between the sub-samples in exposure to the market, measured by betas, with coefficients of the betas all insignificant. There is no consistency in the sign of the coefficients, the 100 observations specification coefficients are negative, and the 50 and 200 window regressions' are positive. Hence, the control group has comparable betas to the treatment group. The Regulation coefficient is partly in line with expectations, the coefficients for all specifications are significant. They show that compared to the pre regulation period, the betas have increased post regulation, with volatility, by contrast, going down. It was expected that the market perceived the economy and its institutions in general more
unstable, which the betas, proxy for systemic risk, could signify. Under this reasoning, it could be stipulated that the uncertainty about firm's riskiness, proxied by volatility, has gone down after the regulation. Table 4. Market Betas and Return Volatility: Estimates for different window specifications This table presents results of regressions (2) and (3) with rolling betas and volatility (computed as the rolling standard deviation) as the dependent variables. Both variables are used as proxies for assessing the risk of European firms involved in CCP settlement. Two variables are run on a rolling basis with robust standard errors clustered on firm level with windows of 100, 50 and 200 observations. The table reports the corresponding coefficients with p-values in the brackets below. Regulation is a dummy variable that takes a value of 1 before 2nd of July 2012 and 0 otherwise. Control is a dummy equal to 1 if a firm belong to a treatment group (involved in CCP settlement) and 0 if it belongs to a control group. Regulation x Control is an interaction term of the two dummies. Crisis 1 is a dummy with value 1 for the year of 2009 and 0 otherwise, crisis 2 is another dummy equal to 1 for the years 2011 and 201 and 0 otherwise. The regressions use daily data from 2009 to 2015, inclusively. *, **, and *** indicate significance at 10%, 5% and 1% correspondingly. Dependent variable: Betas Dependent variable: Volatility (1) (2) (3) (4) (5) (6) w: 100 w: 50 w: 200 w: 100 w: 50 w: 200 Regulation 0.0536** 0.0728*** 0.0585*** -0.0014*** -0.0012*** -0.002*** (0.016) (0.002) (0.005) (0.001) (0.003) (0.000) Regulation x Control -0.0977** -0.1369*** -0.1371*** -0.0013 -0.0015 -0.0007 (0.046) (0.001) (0.000) (0.155) (0.116) (0.405) Control -0.0645 0.0114 0.0178 0.0045*** 0.0045*** 0.0043*** (0.552) (0.887) (0.822) (0.000) (0.000) (0.001) Crisis 1 0.0179 0.0685*** 0.0455* 0.0081*** 0.0087*** 0.0091*** (0.469) (0.006) (0.056) (0.000) (0.000) (0.000) Crisis 2 0.110*** 0.1229*** 0.0851*** 0.0028*** 0.0029*** 0.0023*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Constant 0.9665*** 0.9441*** 0.9712*** 0.0168*** 0.0165*** 0.0176*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) N 252,565 262,860 247,204 252,565 262,860 247,204 R-sq 0.0144 0.0129 0.0221 0.1278 0.1261 0.1219 6. Robustness tests In this section robustness tests using the Europe sovereign bond credit default swap (CDS) spreads as the measures of risk will be conducted. The 10-year sovereign bond CDS
spreads as controls for crises are in line with Mancini et al. (2015, p. 16), who study the period from 2006 and 2013 and include the difference in 10 year CDS sovereign spreads between Italy and Germany and Spain and Germany as controls for risk in their robustness tests. Moreover, in their analysis for the period of 2008 to 2012 Boissel et al. (2017) used 5 year sovereign CDS spreads as measures of risk. I include the CDS spreads of the countries mostly affected by the European debt crisis (i.e. Greece, Portugal, Ireland, Spain and Cyprus) and the countries of the treatment group from Table 2 in the Appendix, in total 17 countries. Hence, the following rolling regressions are run for windows of 100, 50 and 200 observations: (4) 𝛽",$ = ∝2,$ 𝑅𝑒𝑔𝑢𝑙𝑎𝑡𝑖𝑜𝑛 + ∝:,$ 𝑅𝑒𝑔𝑢𝑙𝑎𝑡𝑖𝑜𝑛×𝐶𝑜𝑛𝑡𝑟𝑜𝑙 + ∝=,$ 𝐶𝑜𝑛𝑡𝑟𝑜𝑙 + ∝>,$𝐺𝑟𝑒𝑒𝑐𝑒 +∝@,$𝐶𝑦𝑝𝑟𝑢𝑠 +∝L,$ 𝐼𝑡𝑎𝑙𝑦 +∝N,$𝑃𝑜𝑟𝑡𝑢𝑔𝑎𝑙 +∝P,$𝐼𝑟𝑒𝑙𝑎𝑛𝑑 +∝R,$ 𝑆𝑝𝑎𝑖𝑛 + +∝2T,$ 𝐺𝑒𝑟𝑚𝑎𝑛𝑦 +∝22,$ 𝑈𝐾 +∝2:,$ 𝐹𝑟𝑎𝑛𝑐𝑒 +∝2=,$ 𝑆𝑤𝑒𝑑𝑒𝑛 +∝2>,$ 𝐷𝑒𝑛𝑚𝑎𝑟𝑘 + ∝2@,$ 𝐻𝑢𝑛𝑔𝑎𝑟𝑦 +∝2L,$ 𝑃𝑜𝑙𝑎𝑛𝑑 +∝2N,$ 𝐵𝑒𝑙𝑔𝑢𝑖𝑚 +∝2P,$ 𝑁𝑒𝑡ℎ𝑒𝑟𝑙𝑎𝑛𝑑𝑠 +∝2R,$𝐴𝑢𝑠𝑡𝑟𝑖𝑎 + +∝:T,$ 𝑆𝑤𝑖𝑡𝑧𝑒𝑟𝑙𝑎𝑛𝑑 + 𝜏",$ (5) 𝜎",$ = 𝜑2,$𝑅𝑒𝑔𝑢𝑙𝑎𝑡𝑖𝑜𝑛 + 𝜑:,$𝑅𝑒𝑔𝑢𝑙𝑎𝑡𝑖𝑜𝑛×𝐶𝑜𝑛𝑡𝑟𝑜𝑙 + 𝜑=,$𝐶𝑜𝑛𝑡𝑟𝑜𝑙 + 𝜑>,$𝐺𝑟𝑒𝑒𝑐𝑒 + 𝜑@,$𝐶𝑦𝑝𝑟𝑢𝑠 + 𝜑L,$𝐼𝑡𝑎𝑙𝑦 + 𝜑N,$𝑃𝑜𝑟𝑡𝑢𝑔𝑎𝑙 + 𝜑P,$𝐼𝑟𝑒𝑙𝑎𝑛𝑑 + 𝜑R,$𝑆𝑝𝑎𝑖𝑛 + +𝜑2T,$𝐺𝑒𝑟𝑚𝑎𝑛𝑦 + 𝜑22,$𝑈𝐾 + 𝜑2:,$𝐹𝑟𝑎𝑛𝑐𝑒 + 𝜑2=,$𝑆𝑤𝑒𝑑𝑒𝑛 + 𝜑2>,$𝐷𝑒𝑛𝑚𝑎𝑟𝑘 + 𝜑2@,$𝐻𝑢𝑛𝑔𝑎𝑟𝑦 + 𝜑2L,$𝑃𝑜𝑙𝑎𝑛𝑑 + 𝜑2N,$𝐵𝑒𝑙𝑔𝑢𝑖𝑚 + 𝜑2P,$𝑁𝑒𝑡ℎ𝑒𝑟𝑙𝑎𝑛𝑑𝑠 + 𝜑2R,$𝐴𝑢𝑠𝑡𝑟𝑖𝑎 + +𝜑:T,$𝑆𝑤𝑖𝑡𝑧𝑒𝑟𝑙𝑎𝑛𝑑 + 𝜍",$ From the above, 𝑅𝑒𝑔𝑢𝑙𝑎𝑡𝑖𝑜𝑛, 𝑅𝑒𝑔𝑢𝑙𝑎𝑡𝑖𝑜𝑛×𝐶𝑜𝑛𝑡𝑟𝑜𝑙 and 𝐶𝑜𝑛𝑡𝑟𝑜𝑙 have the same meaning and values as in regression (2) and (3), with 𝜏",$ and 𝜍",$ being the error terms. ∝2...:T,$ and
𝜑2...:T,$ are coefficients. CDS spreads were downloaded from Datastream, to isolate the scale effects, relative CDS changes were computed through the formula: ((value at t) - (value at t+1))/(value at t). The resulting variables were winsorized from both tails with percentile cut-offs of 2.5 and 97.5. Table 5 contains the summary statistics. Table 5. Sovereign bond CDS spread summary statistics. This table contains the relative changes in 10 year CDS sovereign bonds' spread summary statistics for the entire period of analysis, 2009 to 2015. The changes are computed as ((value at t) - (value at t+1))/(value at t). The data is shown by pre regulation and post regulation periods, with the regulation date being 2nd of July 2012. In total CDS spreads for 17 countries is shows, with 5 being the most affected countries during the Europe sovereign debt crisis and 12 countries for the firms in the treatment group. The changes in CDS spreads come as a measure of risk in the economy and are used in the robustness tests. Pre regulation Post regulation
Country Mean Std. Dev. Min. Max. Mean Std. Dev. Min. Max.
Austria 0.0019 0.0580 -0.3397 0.4546 -0.0007 0.0334 -0.1919 0.2320 Belgium 0.0020 0.0506 -0.2306 0.2274 -0.0006 0.0384 -0.1770 0.2132 Cyprus 0.0035 0.0421 -0.2917 0.4163 -0.0016 0.0335 -0.2617 0.3595 Denmark 0.0008 0.0414 -0.2333 0.3320 0.0003 0.0562 -0.2727 0.4176 France 0.0019 0.0481 -0.2089 0.2046 -0.0004 0.0320 -0.1870 0.1580 Greece 0.0026 0.0363 -0.2175 0.2567 -0.0003 0.0572 -0.5474 0.6202 Germany 0.0018 0.0567 -0.2903 0.4693 -0.0002 0.0422 -0.2076 0.2424 Hungary 0.0009 0.0351 -0.2053 0.3028 -0.0008 0.0144 -0.0925 0.0782 Ireland 0.0020 0.0473 -0.3085 0.2852 -0.0018 0.0251 -0.2150 0.2142 Italy 0.0023 0.0553 -0.3716 0.2611 -0.0005 0.0276 -0.1229 0.1866 Netherlands 0.0020 0.0450 -0.4842 0.4031 -0.0001 0.0475 -0.3522 0.5437 Poland 0.0010 0.0401 -0.2703 0.2587 -0.0006 0.0142 -0.0859 0.1588 Portugal 0.0036 0.0587 -0.4781 0.3571 -0.0006 0.0275 -0.1470 0.1516 Spain 0.0025 0.0555 -0.3229 0.3379 -0.0005 0.0282 -0.1170 0.1954 Sweden 0.0007 0.0486 -0.2527 0.2331 0.0037 0.1099 -0.3831 0.7060 Switzerland 0.0004 0.0446 -0.2094 0.3298 -0.0002 0.0139 -0.2076 0.1137 UK 0.0003 0.0406 -0.2015 0.3247 -0.0003 0.0359 -0.1842 0.2263 N 134,640 133,416 From the above, it can be noted that countries with the highest means for CDS changes are the ones who suffered the most from the European debt crisis, i.e. Cyprus, Greece, Italy,
Portugal and Spain, with the means of 0.0035, 0.0026, 0.0023, 0.0036 and 0.0025 correspondingly. Sweden, Switzerland and the UK appear to be the least affected, with the means close to 0. The standard deviations of all countries range between 0.0363 and 0.580. It is noticeable that post regulation the means turn negative and are close to zero, with the exception of Denmark and Sweden (0.0003 and 0.0037 respectively). This might indicate a reduction in economy-wide risk. The results of regressions (4) and (5) are shown in Table 6. Table 6. Market Betas and Volatility: robustness tests with sovereign bond CDS as controls This table contains the results of regressions (4) and (5) with betas and volatility (computed as the standard deviation of returns) as the dependent variable. The regressions were run on a rolling basis, with the results for different window specifications included. The standard errors are robust clustered on firm level. The coefficients are shown with corresponding p-values in the brackets below. Regulation is a dummy variable that takes a value of 1 before 2nd of July 2012 and 0 otherwise. Control is a dummy equal to 1 if a firm belong to a treatment group (involved in CCP settlement) and 0 if it belongs to a control group. Regulation x Control is an interaction term of the two dummies. The countries' names represent changes in sovereign 10 year bond CDS. The regressions use daily data from 2009 to 2015, inclusively. *, **, and *** indicate significance at 10%, 5% and 1% correspondingly. Betas Volatility
window: 100 window: 50 window: 200 window: 100 window: 50 window: 200
Regulation 0.0140 0.0176 0.0200 -0.0039*** -0.0043*** -0.0036*** (0.555) (0.462) (0.362) (0.000) (0.000) (0.000) Regulation xControl -0.0977** -0.1368*** -0.137*** -0.0013 -0.0015 -0.0007 (0.046) (0.001) (0.000) (0.153) (0.114) (0.405) Control -0.0645 0.0113 0.0177 0.0045*** 0.0045*** 0.0043*** (0.552) (0.888) (0.823) (0.000) (0.000) (0.001) Greece 0.0066 0.0230 -0.0159 -0.0025*** -0.0044*** -0.0005** (0.642) (0.135) (0.107) (0.000) (0.000) (0.026) Cyprus 0.0054 0.0050 -0.0214* -0.0052*** -0.0071*** -0.0022*** (0.697) (0.810) (0.067) (0.000) (0.000) (0.000) Italy 0.0010 0.0551** 0.0250 -0.0011*** -0.0031*** -0.0021*** (0.965) (0.049) (0.22) (0.000) (0.000) (0.000) Portugal -0.0300 -0.0613** -0.0838*** -0.0021*** -0.0039*** -0.0011*** (0.103) (0.026) (0.000) (0.000) (0.000) (0.001) Ireland -0.1074*** -0.2065*** -0.0445* -0.0029*** -0.0019*** -0.0030*** (0.000) (0.000) (0.07) (0.000) (0.000) (0.000) Spain 0.0907*** 0.0320 0.1160*** 0.0058*** 0.0082*** 0.0063*** (0.003) (0.324) (0.000) (0.000) (0.000) (0.000) Germany 0.0014 -0.0327*** -0.0112** -0.0017*** -0.0029*** -0.0007*** (0.830) (0.000) (0.013) (0.000) (0.000) (0.000) UK -0.0057 -0.0340** 0.0056 -0.0010*** -0.0001 0.0024***
(0.533) (0.011) (0.341) (0.000) (0.696) (0.000) France 0.0225** 0.0114 0.0085 -0.0025*** -0.0026*** 0.0001 (0.023) (0.494) (0.305) (0.000) (0.000) (0.464) Sweden -0.0020 0.0264*** -0.0085** 0.0002*** 0.0001 -0.0003*** (0.677) (0.000) (0.031) (0.001) (0.155) (0.000) Denmark -0.0276*** -0.0234* -0.0434*** -0.0025*** -0.0037*** -0.0023*** (0.003) (0.082) (0.000) (0.000) (0.000) (0.000) Hungary 0.1294*** 0.2922*** 0.0271 0.0051*** 0.0075*** 0.0045*** (0.000) (0.000) (0.126) (0.000) (0.000) (0.000) Poland -0.1320*** -0.2878*** -0.0992*** -0.0051*** -0.0034*** -0.0061*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Belgium -0.0117 0.0274* -0.0491*** -0.0031*** -0.0042*** -0.0031*** (0.430) (0.094) (0.000) (0.000) (0.000) (0.000) Netherlands 0.0026 0.0040 -0.0241*** -0.0001 0.0028*** -0.0009*** (0.700) (0.717) (0.001) (0.277) (0.000) (0.000) Austria 0.0393*** 0.0497*** 0.0272*** 0.0015*** 0.0003** 0.0009*** (0.000) (0.000) (0.000) (0.000) (0.024) (0.000) Switzerland 0.0317* 0.0906*** 0.0409*** 0.0022*** -0.0053*** 0.0037*** (0.084) (0.000) (0.004) (0.000) (0.000) (0.000) Constant 1.0218 1.0163*** 1.0217*** 0.0198*** 0.0200*** 0.0196*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) N 252,565 262,860 247,204 252,565 262,860 247,204 R-sq 0.0139 0.0071 0.0099 0.0885 0.0832 0.092 The results are consistent with main results in Table 4. As with Table 4, coefficients do not change much throughout window specifications, however, this time, the 100 window presents the highest R squared for both dependent variables. Using the CDS spreads as controls, however, does not help explain more data, with R squared decreasing by around 0.06 and 0.04 for betas and volatility regressions respectively. As with the main results, volatility is better explained by the variables than betas. It is noticeable, that for dummies 𝑅𝑒𝑔𝑢𝑙𝑎𝑡𝑖𝑜𝑛×𝐶𝑜𝑛𝑡𝑟𝑜𝑙 and 𝐶𝑜𝑛𝑡𝑟𝑜𝑙 the coefficients, their signs and significance stay on the same level. Therefore, the finding of a reduction in risk in the treatment group, relative to the control group with dummies as controls is robust. The regulation dummy coefficients are no longer significant with betas as the depended variable. They are, however still positive, ranging from 0.014 to 0.02. Nevertheless, volatility regression regulation dummy coefficients are significant and in line with previous results. Although they are more
negative, averaging around -0.0040 across the window specifications. As was previously mentioned, this is in contrast to the results of Sarin and Summers, who find that post regulation volatility has risen. It seems that most negative influence on betas, among the countries, exhibited Spain, France, Hungary and Austria, with their coefficients ranging from 0.09 to 0.1294, with the highest being Hungary. This could be interpreted as: due to a point rise in CDS changes, the betas increase by 0.09 to 0.1294. As mentioned previously, a 0.09 point coefficient (Spain) comprises around 20% of the interval between the lowest and highest values for betas, window 200 observations. From Table 6, for betas regressions there are also coefficients that are negative and significant, i.e. for Ireland, Denmark and Poland. Their coefficients are approximately in the same range as the negative ones, however, are harder to interpret, as one would expect an increase in CDS spreads to increase risk measures. It is harder to interpret volatility regression coefficients, as all coefficients, except for the Netherlands appear to be significant. These coefficients are on average 10 times smaller than the ones for the betas and are mainly negative. Among the positive coefficients are Spain, Sweden, Austria and Switzerland. 7. Conclusion and Discussion I have analysed the change in risk of firms as a result of implementation of European Market Infrastructure Regulation, which makes central clearing of OTC derivatives mandatory. The motivation comes from the ongoing debate about the effects of central clearing and inconclusive empirical evidence. Further motivation stems from the size of OTC derivatives market, with EMIR estimated to have wide consequences. Two risk measures, share return volatility and CAPM betas were analysed on a rolling basis for a total sample of
153 European firms for the period of 2009 to 2015. The sample consisted of 72 treatment firms, which were subjected to EMIR regulation and 81 other firms representing the market. The main finding is that risk in the treatment group, relative to the control group decreases, as per both risk measures. However, the reduction of the systemic part of the risk, represented by betas is more substantial, comprising approximately 20% of the interval between the highest and lowest treatment group beta values as calculated for a window of 200 observations. The reduction in volatility is insignificant with the economic effect being that of 9% of volatility's pre regulation standard deviation. These findings are robust to using the sovereign CDS spreads as controls as opposed to crisis period dummies. Research using the same methodology of Sarin and Summers has concluded that in the US for the period of 2004 to 2015 these risk measure have, in contrast, increased. The main difference, however, is that this thesis uses a difference in difference analysis and focuses on institutions involved in central clearing. Overall, I corroborate Loon and Zhong's (2014) findings of CCPs not generating systemic risk, on the contrary, my findings suggest the they decrease it. This could be due to CCPs promoting implementation of better risk measures, to the extent that members might not want to pay higher margins and contributions to the default fund due to their own perceived riskiness. As the literature review has shown, CCPs are not subject to strict regulation standards of the waterfall loss structures. However, if CCPs decrease systemic risk, then supervisory bodies should conduct deeper studies about the effect of each CCP measure. As a consequence, possibly more standardized risk controlling approach among CCPs should be enforced, as markets might have a clearer expectation and certainty about central clearing effects and the distribution of losses in adverse times.
This paper presents support for the arguments of proponents of central clearing. In addition, with lower systemic risk after the regulation, Menkveld's (2014) argument about unsuitable risk measures of Central Clearing Counterparties, that do not take the crowding of trades into account does not have credibility. However, the finding that volatility does not go down, if taken from a perspective that volatility represents largely uncertainty, is in contrast to Loon and Zhongs' (2016) findings of increased transparency in the US. Therefore, perhaps Acharya and Bisin's (2014) concerns about a centralized registry of trades not being enough carry weigh and more transparency enhancement measures, such as weekly distributions of reports to clearing members about each member's positions, should be implemented. The limitations are contained in the data, as the share prices were used, therefore, all results come as the markets' expectations. Further limitation could lie in the selection of the control group for the analysis, even though those firms are comparable in size and CAPM betas to the treatment group, there is a possibility of them using central clearing services elsewhere in the world. Additionally, the control group's proportion of financial companies is less than treatments', therefore, financial industry trends, other than the effect of central clearing, might not have been fully isolated. Other limitations are that of the simplicity of the risk measures used and the uncertain quality of the data provided by Datastream. Suggestions for future research include conduction of the analysis with a larger data span. Furthermore, it could be worthwhile to look at how CCP member's share prices behave in extreme times, i.e. conduct the same analysis on the left tail of return distributions. Other suggestions would include investigating the effect of OTC derivative volumes, where risk measures could be correlated with OTC volumes to have additional indication of EMIR's risk consequences.
Bibliography Acharya, V., & Bisin, A. (2014). Counterparty risk externality: Centralized versus over-the- counter markets. Journal of Economic Theory, 149, 153-182. Acharya, V. V., Engle, R., Figlewski, S., Lynch, A., & Subrahmanyam, M. (2009). Centralized clearing for credit derivatives. Restoring financial stability: How to repair a failed system, 251, 251. Bank for International Settlements (2012). Statistical release: OTC derivatives statistics at end-December 2011. Retrieved from https://www.bis.org/publ/otc_hy1205.pdf Bank of England (2011). Financial Stability Report. December. Biais, B., Heider, F., & Hoerova, M. (2012). Clearing, counterparty risk, and aggregate risk. IMF Economic Review, 60(2), 193-222. Boissel, C., Derrien, F., Ors, E., & Thesmar, D. (2017). Systemic risk in clearing houses: Evidence from the European repo market. Journal of Financial Economics, 125(3), 511-536.
Cecchetti, S. G., Gyntelberg, J., & Hollanders, M. (2009). Central counterparties for over-the- counter derivatives. Cederburg, S., & O'Doherty, M. S. (2016). Does it pay to bet against beta? on the conditional performance of the beta anomaly. The Journal of finance, 71(2), 737-774. CME SPAN (2010). Standard Portfolio Analysis of Risk. Report. Retrieved from https://www.cmegroup.com/clearing/files/span-methodology.pdf Commodity Futures Trading Commission. (2001). Review of Standard Portfolio Analysis of Risk (“SPAN”) Margin System. Report of the Division of Trading and Markets. CPSS, I. (2011). Principles for Financial Market Infrastructures. Consultative Report, March. Cruz Lopez, J., Harris, J. H., Hurlin, C., & Pérignon, C. (2012). Co-margin: A system to enhance financial stability. De Genaro, A. (2016). Systematic multi-period stress scenarios with an application to CCP risk management. Journal of Banking & Finance, 67, 119-134. Duffie, D., Li, A., & Lubke, T. (2010). Policy perspectives on OTC derivatives market infrastructure.
Duffie, D., & Zhu, H. (2011). Does a central clearing counterparty reduce counterparty risk?. The Review of Asset Pricing Studies, 1(1), 74-95. ECB. (2013). Annual Report 2013. European Central Bank. Retrieved from https://www.ecb.europa.eu/pub/pdf/annrep/ar2013en.pdf?cd2b3304656cefbe0f9a 287f15923a2f ESMA. (2012). Questions and Answers. Implementation of the Regulation (EU) No 648/2012 on OTC derivatives, central counterparties and trade repositories (EMIR). European Securities and Markets Authority (ESMA). Retrieved from https://www.esma.europa.eu/sites/default/files/library/esma70-1861941480- 52_qa_on_emir_implementation.pdf Gugerell, G. (2010). Why OTC derivatives must be cleared. Financial Times. Retrieved from Hess, B., Mariottini, C., & Camara, C. (2012). Regulation (EU) No 650/2012 of the European Parliament and of the Council of 4 July 2012 on jurisdiction, applicable law, recognition and enforcement of decisions and acceptance and enforcement of authentic instruments in matters of succession and on the creation of a European Certificate of Succession: Note. Retrieved from https://eur-lex.europa.eu/legal- content/EN/TXT/?uri=CELEX:02012R0648-20160112 IMF. (2010). Sovereigns, Funding, and Systemic Liquidity; October 2010. IMF Global Financial Stability Report (GFSR). Retrieved from
https://www.imf.org/en/Publications/GFSR/Issues/2016/12/31/Sovereigns-Funding- and-Systemic-Liquidity Jagannathan, R., & Wang, Z. (1996). The conditional CAPM and the cross-section of expected returns. The Journal of finance, 51(1), 3-53. Jones, R. A., & Pérignon, C. (2013). Derivatives clearing, default risk, and insurance. Journal of Risk and Insurance, 80(2), 373-400. Koeppl, T., Monnet, C., & Temzelides, T. (2012). Optimal clearing arrangements for financial trades. Journal of Financial Economics, 103(1), 189-203. Loon, Y. C., & Zhong, Z. K. (2014). The impact of central clearing on counterparty risk, liquidity, and trading: Evidence from the credit default swap market. Journal of Financial Economics, 112(1), 91-115. Loon, Y. C., & Zhong, Z. K. (2016). Does Dodd-Frank affect OTC transaction costs and liquidity? Evidence from real-time CDS trade reports. Journal of Financial Economics, 119(3), 645-672. Mancini, L., Ranaldo, A., & Wrampelmeyer, J. (2015). The euro interbank repo market. The Review of Financial Studies, 29(7), 1747-1779.
Menkveld, A. J. (2014). Crowded trades: an overlooked systemic risk for central clearing counterparties (No. 14-065/IV/DSF75). Tinbergen Institute Discussion Paper. Meyer, G. (2010). CFTC head blames OTC derivatives for crisis. Financial Times. Retrieved from https://www.ft.com/content/3be62c7a-fae8-11de-94d8-00144feab49a Pirrong, C. (2011). The economics of clearing in derivatives markets: Netting, asymmetric information, and the sharing of default risks through a central counterparty. Working Paper (University of Houston). Sarin, N., & Summers, L. H. (2016). Have big banks gotten safer?. Brookings Institution. Singh, M. M. (2011). Making OTC derivatives safe: A fresh look (No. 11-66). International Monetary Fund. Steigerwald, R. S. (2013). Central Counterparty Clearing. Understanding Derivatives: Markets and Infrastructure. Federal Reserve Bank of Chicago, Chicago. Retrieved from https://www. chicagofed.org/publications/understandingderivatives/index. WEC. (2015). Global Risks 2015. World Economic Forum. Retrieved from: http://reports.weforum.org/global-risks-2015/part-1-global-risks-2015/introduction/
Appendix Table 2. List of companies in the full sample List of companies in the full sample categorized by countries. Panel A contains 81 companies in the control group, whereas Panel B contains the 72 firms for the treatment group. Panel A: Control Group France Ireland AIR LIQUIDE CRH ATOS KERRY GRP AXA RYANAIR BOUYGUES Italy CARREFOUR ASSICURAZIONI GENERALI DANONE ATLANTIA ENGIE ENEL ESSILOR INTERNATIONAL ENI HERMES INT FIAT CHRYSLER AUTO KERING LUXOTTICA L'OREAL STMICROELECTRONICS LEGRAND ARCELORMITTAL MICHELIN Netherlands ORANGE AHOLD DELHAIZE PERNOD RICARD AKZO NOBEL PEUGEOT ASML HLDG PUBLICIS GRP HEINEKEN RENAULT KONINKLIJKE DSM RWE PHILIPS SAFRAN RELX NV SAINT GOBAIN UNILEVER NV SANOFI WOLTERS KLUWER SCHNEIDER ELECTRIC Germany THALES ADIDAS TOTAL BASF VALEO BAYER VEOLIA ENVIRONNEMENT BMW VINCI CONTINENTAL Spain DAIMLER AMADEUS IT GROUP DEUTSCHE BOERSE BCO BILBAO VIZCAYA ARGENTARIA DEUTSCHE POST CAIXABANK DEUTSCHE TELEKOM FERROVIAL DEUTSCHE WOHNEN IBERDROLA E.ON Industria de Diseno Textil SA FRESENIUS REPSOL FRESENIUS MEDICAL CARE TELEFONICA HEIDELBERGCEMENT Finland HENKEL PREF KONE B INFINEON TECHNOLOGIES NOKIA MERCK
SAMPO SAP UPM KYMMENE THYSSENKRUPP WIRECARD Belgium ANHEUSER-BUSCH INBEV Panel B: Treatment group Austria Netherlands BKS BANK AEGON ERSTE GROUP BANK ING GROEP RAIFFEISEN BANK INTL. Royal Dutch Shell Belgium Poland KBC GROUP BANK POLSKA KASA OPIEKI Denmark HANDLOWY DANSKE BANK Portugal SYDBANK BANCO COMR.PORTUGUES France EDP ENERGIAS DE PORTUGAL AIRBUS ROYAL BANK OF SCTL.GP. ABC ARBITRAGE Spain BNP PARIBAS AHORRO CORPORACION CAI DINERO BOURSE DIRECT BANCO DE SABADELL CREDIT AGRICOLE BANCO SANTANDER NATIXIS BBV.ARGENTARIA SOCIETE GENERALE ENDESA Germany GAS NATURAL SDG ALLIANZ Sweden BAADER BANK NORDEA BANK COMDIRECT BANK SVENSKA HANDBKN. COMMERZBANK SWEDBANK DEUTSCHE BANK Switzerland EUWAX BANQUE CANTON.VE. IKB DEUTSCHE INDSTRBK. BERNER KANTONALBANK LANG & SCHWARZ WERTPAH. CREDIT SUISSE GROUP N
MERCEDES-BENZ BANK INVESCO ZUWACHS EFG INTERNATIONAL N
MWB FAIRTRADE (XET) WERTPAH. LUZERNER KANTONALBANK SIEMENS SCHWEIZERISCHE NAT.BK. STADTWERKE HANNOVER GSH. ST GALLER KANTONALBANK VARENGOLD BANK UBS GROUP VOLKSWAGEN VONTOBEL HOLDING Greece UK AAREAL BANK AVIVA BANK OF PIRAEUS BARCLAYS NATIONAL BK.OF GREECE BP Italy HSBC BANCA FINNAT EURAMERICA INTERACTIVE BROKERS GP. BANCA PROFILO INVESTEC CREDITO EMILIANO LLOYDS BANKING GROUP
INTESA SANPAOLO STANDARD CHARTERED MEDIOBANCA BC.FIN Hungary UNICREDIT MOL MAGYAR OLAJ-ES GAZIPARI OTP BANK Table 3. Return rolling volatility computed for different window specifications Summary firm's return volatility statistics computed as the sample standard deviation of stock returns on a rolling basis. This table presents volatility for three windows: 50, 100 and 200 trading days. Treatment group and Control group statistics are presented separately and broken down by pre and post regulation, with 2nd of July 2012 as the cut off. Pre regulation window = 100 Number of
observations Mean Median Std. Dev. Min. Max.
Treatment group 56,232 0.0243 0.0217 0.0131 0.0001 0.1053 Control group 62,917 0.0197 0.0182 0.0076 0.0071 0.0799 window = 50 Treatment group 59,832 0.0245 0.0212 0.0146 0.0001 0.1340 Control group 66,967 0.0199 0.0179 0.0087 0.0056 0.0910 window = 200 Treatment group 49,032 0.0239 0.0223 0.0114 0.0001 0.0823 Control group 54,817 0.0196 0.0184 0.0064 0.0080 0.0621 Post regulation window = 100 Treatment group 62,784 0.0191 0.0169 0.0119 0.0003 0.1287 Control group 70,632 0.0158 0.0151 0.0047 0.0066 0.0493 window = 50 Treatment group 62,784 0.0188 0.0164 0.0124 0.0002 0.1370 Control group 70,632 0.0157 0.0149 0.0051 0.0053 0.0560 window = 200 Treatment group 62,784 0.0196 0.0170 0.0116 0.0003 0.1066 Control group 70,632 0.0160 0.0152 0.0045 0.0074 0.0439
