Screening for manipulation and collusion after interbank rate reforms

(1)

Screening for manipulation and collusion after interbank rate reforms

Nuria Boot

Student number: 6041213

Supervisors: prof. dr. M.P. Schinkel and dr. M.J.G. Bun

MSc in Economics Thesis, Industrial Organization, Regulation and Competition Policy track

(2)

2 Statement of Originality

This document is written by Nuria Boot who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is 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)

3 Abstract

Rates such as Euribor and Libor have been making the news since it came to light that they had been manipulated. These rates serve as a reference for hundreds of trillions of dollars in derivatives and other financial products. By trading in these financial products panel banks have an interest to manipulate them. Several initiatives have been put in place to reform the rates, aiming mainly at using as much transaction data as possible to reduce the scope for manipulation. This thesis identifies two main ways in which ways panel banks could still manipulate or collude after the reforms, and proposes a screen that could be used to flag for manipulation or collusion. First, banks’ portfolio exposure positions to the rates have a much larger volume than the subset of eligible transactions that banks can use as a basis for their quote submissions on. Banks could engage in eligible transactions at rates that benefit their overall trading exposure position. Second, without actually manipulating the rate and by communicating with other panel banks, banks could have an idea of what the future rate will be before it becomes known to the rest of the market. In this window of time, banks could engage in transactions that increase their own profits. This type of manipulation is also known as front running. Data requirements for the screen are listed and since this data it is not

publicly available, it would have to be requested by the regulator. In this thesis, data is generated according to a theoretical model of quote submission to illustrate the screen.

However, one of the main challenges is that very little is known about what the required input data looks like in practice.

(4)

4

Interbank rates such as the Euribor (Euro Interbank Offered Rate) and the Libor (London Interbank Offered Rate) have been making the news since it came to light that they had been manipulated. Every trading day, these rates are calculated as a trimmed average of quotes given by banks in a panel. Banks in the Libor panel are asked to quote at which rate they would be able to borrow funds. The question is slightly different for banks in the Euribor panel as banks are asked for a “market” rate for a “prime bank”, not the rate at which they could borrow themselves. Hundreds of trillions of dollars in derivatives and other financial products are tied to these rates.

Panel banks have a portfolio exposure position to the rates, as they trade in all kinds of financial products that use these interbank rates as a reference. As a result, they have a

financial interest to manipulate them. It came to light that banks had manipulated the rate by giving quotes that were biased in the direction of their trading positions as well as by giving lower quotes to appear a more trustworthy counterparty1. This thesis focuses on manipulation based on trading exposure positions, which is evidenced by many messages that were

published during investigations from regulators and competition authorities. For example a message from a New York trader to a British submitter on 31 May 2006 indicates a

preference for a high rate due to the portfolio trading position (Farrell, 2012):

"We have another big fixing tom[orrow] and with the market move I was hoping we could set the 1M and 3M Libors as high as possible."

Banks have not only been found to have manipulated their own quotes, but they also colluded with other banks in the panel and exchanged sensitive information on trading positions and quotes.

The European Commission (2013) conducted antitrust investigations under Art. 101 and fined several banks for participating in cartels in the interest rate derivatives industry. The Commission considered it a cartel as the collusive exchanges of information (on trading positions, pricing strategies and quote submissions) led to different benchmark rates than would have been the case under normal competitive market conditions. This affected the

1_{In the Euribor panel, bansk are asked to quote what they think is the market rate for a prime bank, while in the}

Libor panel, banks are asked at what rate they perceive to be able to obtain funding at the London interbank money market themselves. Therefore, especially in the Libor panel, a bank with a high submission could be perceived as a less trustworthy counterparty by others in the market.

(6)

6 markets for mortgages, loans and other financial products. Three cases were investigated, two regarding collusion in the Libor rate-setting process and one regarding the Euribor.

A 2012 regulatory review (Wheatley, 2012) found that the Libor was not fit for purpose and several reforms have been suggested and implemented since. In 2014, the ICE Benchmark Administration took over the administration of the rate from the British Bankers’ Association. Aiming to restore trust in the rate, the new administrator aims at basing the rate on actual transactions “to the greatest extent possible” (ICE Benchmark Administration, 2015). The European Money Markets Institute (EMMI), the administrator of Euribor, plans on

introducing an entirely transactions based rate in the future (EMMI, 2015).

Although the proposed reforms reduce the scope for manipulation and collusion, this should still remain a concern given large financial interests at stake. For example, internal documents from Deutsche Bank showed that on 30 September 2008 the banks could gain or lose €68 million for each one-hundred of a percentage point change in Euribor and Libor (Eaglesham, 2013). This thesis identifies in which ways panel banks could still manipulate or collude after the reforms, and proposes a screen that could be used to flag for manipulation and collusion.

Part of the reformed rate-setting mechanism still relies on banks’ expert judgment and not on actual transactions. Previous research has addressed the detection of expert judgment manipulation. The first type of manipulation for which a screen is proposed is eligible

transaction rigging. , Bank’s exposure positions to rates such as Euribor and Libor consists of a much larger volume of transactions than those that are eligible to base its quote on.

Therefore, banks could manipulate the rate by choosing its eligible transactions at rates that favor the overall exposure position. The second type of manipulation is front running. Banks could exchange information on future quotes with other panel banks before the rates become publicly available. This way, banks could have an idea on what the future rate is going to be and use this information to create a more beneficial exposure position to the rate,.

This thesis proceeds as follows. Chapter 2 discusses interbank rates, the proposed reforms and provides an overview of recent literature on interbank rate manipulation. Chapter 3 illustrates how the rates could still be manipulated under the proposed reforms, introduces the manipulation hypotheses and data needed requirements. In Chapter 4 the screen is simulated using generated data and the challenges of screening for manipulation in practice are discussed. Chapter 5 discusses the challenges of implementing the screen in practice and how banks could circumvent it. Finally, chapter 6 concludes and points out directions for further research.

(7)

7 2. Interbank rates: design, literature and reforms

This chapter provides an overview of Euribor and Libor, the most often used interbank rates. Section 1 gives some background on these interbank rates, why they were introduced and how they were designed. Section 2 summarizes previous research that has been done on interbank rate manipulation and collusion. Section 3 discusses why screening for manipulation and collusion in interbank rates is particularly challenging. Finally, section 4 provides a more detailed overview of the recent reforms that have been suggested to make the rate

manipulation proof in the future.

2.1 Introduction on interbank rates 2.1.1. Libor2

The London Interbank Offered rate (Libor) was introduced in 1986 to help banks set interest rates on large corporate loans. It gradually became more important when banks incorporated this rate in financial contracts and started to trade these (Wheatley, 2012). The rate is used to price many financial products and contracts such as interest rate swaps, mortgages, student loans and is used on many futures and option exchanges. It is the most used interbank rate and about $350 trillion in financial contracts are estimated to be tied to it (ICE Benchmark

Administration, 2016). However, the unsecured interbank lending market has become significantly smaller during the financial crisis and has not restored to pre-crisis levels.3

The Libor is meant to reflect the rate at which banks can borrow unsecured funds from other banks. On every trading day, the Libor is computed for five currencies and seven

maturities (ranging from overnight to 12 months).4 The panel varies over currencies and the panels currently consist of 11 to 18 banks.5 Bank of America is for example only part of the USD Libor panel, while Santander is only in the GBP and EUR Libor panels. Banks can become part of the panel if their market activity, reputation and expertise in the currency is deemed sufficient.

Every trading day, the banks in the panel are surveyed by the ICE Benchmark Administration, the administrator of the rate. They submit quotes answering the following question:

2_{See also: https://www.theice.com/iba/libor.}

3_{Gensler (2012) and Kuo et al. (2012) list a couple of reasons for this. First, the perceived risk of counterparty}

banks to default has increased. Second, regulatory capital requirements require banks to hold more cash and have reduced activity in the market. The introduction of liquidity coverage ratios has changed the interbank funding market and had lead banks to an increased use of longer maturities and secured funding sources.

4_{The number of maturities was reduced from 15 to 7 in June 2013 and the number of currencies from 10 to 5 in}

March 2013. 5

(8)

8 “At what rate could you borrow funds, were you to do so by asking for and then accepting interbank offers in a reasonable market size just prior to 11 am London time?”.

So the rate is not (yet) based on actual transactions and banks may not have borrowed reasonable loan amounts in all currencies and maturities that day. For each currency and maturity, the quotes are ordered and the highest and lowest quartiles are excluded. The remaining middle 50% of the quotes are averaged. At noon, the ICE Benchmark Administration publishes the rates.

2.1.2 Euribor6

The Euro Interbank Offered rate (Euribor) is used in a similar way as the Libor as a reference rate for euro denominated instruments. The Euribor was first published in 1999 and is

administered by the European Money Market Institute. The Financial Stability Board

estimated a notional volume of outstanding contracts indexed to Euribor of over 180€ trillion (Financial Stability Board, 2014b).The definition used by panel banks to submit quotes is different from the Libor definition:

“Euribor is the rate at which euro interbank term deposits are being offered within the EU and EFTA countries by one prime bank to another at 11.00 a.m. Brussels time”.

As opposed to banks in the Libor panel, banks in the Euribor panel are not asked about the rate at which they could borrow funds themselves. They are asked about the “market rate” or a rate at which one prime bank could lend to another.

The Euribor is computed as the average of the quotes after trimming the top 15% and the bottom 15% of the quotes. The middle 70% of quotes are then averaged to compute rates for 8 maturities (ranging from one week to 12 months).7 While the panel currently comprises 22 banks, it has comprised 44 banks in the past. According to a Wall Street Journal article this decline can be attributed to the fact that the scandals have led banks to perceive their participation in the setting of the rates as a potential liability (Enrich, 2013).

6_{See also: http://www.emmi-benchmarks.eu/euribor-org/about-euribor.html.}

7_{In November 2013 the number of maturities was reduced from 15 to 8, leaving only the most heavily traded}

(9)

9 2.1.3 Vulnerability to manipulation and collusion prior to the reforms

These interbank rates are vulnerable to manipulation and collusion for a number of reasons. First of all, before the reforms that are currently taking place, the rates were not based on actual transaction data and the rates could even be published without any relevant transactions took place. Experimental evidence shows that hypothetical questions about for example “willingness to pay” typically do not lead to accurate answers (Harrison and Rutstrӧm, 2008). Moreover, there are two main ways in which banks have financial interests in manipulating the rates. First, banks may have an interest to submit lower quotes to signal their financial health as a bank with a high quote may be perceived as a more risky counterparty by other banks. This holds especially for the Libor panels, where banks are specifically asked to provide the rate at which they could borrow themselves. Second, banks can manipulate the rate by submitting quotes in that are in line with their trading exposure positions in order to obtain higher profits. Banks have large financial interests at stake from trading in all kinds of financial products that are tied to the rates. Regulators’ investigations have revealed extensive evidence of communication between traders showing awareness of the potential to increase profits by manipulating the rate. Moreover, from internal documents from Deutsche Bank it followed that on 30 September 2008 the banks could gain or lose €68 million for each one-hundred of a percentage point change in Euribor and Libor (Eaglesham, 2013).

Trimming the quartiles (or 15% tails for Euribor) makes the rates more robust to manipulation from small numbers of banks than a rate based on a simple untrimmed average. However, each quote is relevant for the ordering of quotes and therefore affects the rate to some extent. For larger effects on the rate an agreement between more than 25% (Libor) or 15% (Euribor) of the panel banks is needed.8 Until early 2014 another aspect about the design of the Libor facilitated collusion: banks’ individual submissions were published on a daily basis, which is still the case for Euribor.9 This facilitated both tacit and explicit collusion as it gave banks the opportunity to monitor each other’s behavior and if necessary punish the banks which had deviated from the agreement. Moreover, as economic literature on cartel states, both tacit and explicit collusion are easier with a smaller number of banks.10 Before the reduction in currencies, the smallest Libor panel used to comprise 6 banks. After the scandal

8

A group with a size of at least 25% of the banks plus one (or 15% of the panel plus one) would be needed to have a larger effect on the rate, where the quotes falling in the tails would be discarded and the ones left would have a significant effect on the average.

9

As of February 2014 these individual submissions are published with a delay of three months and only the Libor is published directly after being calculated.

10

(10)

10 many of the larger panels have become smaller and the administrators of both Euribor and Libor struggle to get more banks on board (Enrich, 2013 and Albanese, 2015).

2.2 Previous research on interbank rate manipulation, collusion and screen design This thesis is connected to theoretical and empirical literature on interbank rate manipulation as well as to the general literature on empirical screens for manipulation and collusion. A Wall Street Journal article by Mollenkamp and Whitehouse (2008) was the first to provide evidence of possible manipulation of the Libor. The authors created an alternative benchmark and indicated that banks’ Libor quotes may have been artificially low to avoid signaling their own deteriorating credit quality. Abrantes-Metz et al. (2012) extend the Wall Street Journal and analyze individual quotes, compare them to CDS spreads and market capitalization data and compare the Libor to other short-term borrowing rates. They find some anomalies, but the evidence is inconsistent with a material manipulation of the US Dollar Libor. Other studies in which possible manipulation is flagged by comparing Libor to other measures of bank

borrowing include Kuo, Skeie and Vickery (2012) and Snider and Youle (2010). Kuo, Skeie and Vickery (2012) compared the Libor quotes to bank bids in the Federal Reserve Term Auction Facility and found that in times of financial turmoil Libor quotes were significantly lower than comparable rates. However, the authors indicate several reasons why these comparable rates could differ from the Libor and can therefore not conclude that the rate has been manipulated. Snider and Youle (2010) show that bank quotes are difficult to rationalize with other measures of bank borrowing costs and predict that if manipulated, quotes would bunch around the 25% and 75% quartiles, for which they find empirical evidence.

A number of studies have used Benford’s second digit (SD) distribution to flag for possible manipulation of the Libor (Abrantes-Metz et al. (2011) and Rausch et al. (2013)). The idea is that digits in many numerical datasets occur in certain frequency distribution. The law has been shown to apply to a wide variety of data sets, including financial ones.

Departure from this law could therefore be used to flag for possible manipulation. Rausch et al. (2013) also investigated the Euribor and showed it performed even worse than the Libor.

Research on the effect of manipulation on the final quotes and possible profits has, among others, been done by: Gandhi et al. (2015), Monticini and Thornton (2013) and Eisl et al. (2013). Monticini and Thornton (2013) study whether manipulation materially affected the Libor rates. They find evidence that suggests the Libor declined due to the manipulation during the financial crisis but that it returned to pre-underreporting levels. Gandhi et al. (2013) find suggestive evidence for a relation between the bank’s Libor exposure and their

(11)

11 monthly average submission. They estimate that by manipulating panel banks could have increased their market value of the panel banks by $22.76 billion. Eisl et al. (2013) quantified how much influence individual banks can have on the final rate and compare it to different designs of the rate (including a rate based on the median of all quotes and the average). They show that manipulation by one bank could result in an average change of 0.48 basis points (3 month USD Libor) or 0.17 basis points (3 month Euribor), while three colluding banks would be able to affect the rate by 1.61 basis points (3 month USD Libor) or 0.53 basis points (3 month Euribor). The authors recommend a rate based on the median of all quotes as it is harder to manipulate with sufficient banks in the panel.

Empirical studies from Snider and Youle (2012) and Gandhi et al. (2015) provide suggestive evidence that banks in the Libor panel manipulated their quotes such that they would be in line with their exposure positions to the rate. Snider and Youle (2012) develop tests for manipulation based on a theoretical model for quote submission. Gandhi et al. (2015) test hypotheses of manipulation and attempt to quantify the potential gains from

manipulation. They find evidence mainly for portfolio-based manipulation of Libor and also find that after the Libor scandal broke out, evidence for manipulation disappeared.

Another strand of literature this thesis is connected to is the literature on empirical screens for collusion. Collusion is difficult to detect because it exists in many forms. Even cartels in “normal markets”, where prices are raised above competitive levels, are difficult to detect because often cost data, that would be needed to estimate the competitive price level, is not available. Abrantes-Metz and Bajari (2009) give a comprehensive overview of screens and their applications. Different types of screens include: screens based on market shares, screens based on mathematical laws (such as Benford’s law), screens based on price and cost information and screens based on improbable events or control groups. Abrantes-Metz et al. (2006) propose a screen based on the coefficient of variation. They found that after a collapse of a cartel in the seafood processing industry the price level decreased by 16%, while the standard deviation increased by 263%. They apply the screen to the retail gasoline industry in Louisville, but fail to find similar results. Bolotova et al. (2008) use extensions of ARCH and GARCH models to examine the impact of the lysine and citric acid cartels on price level and variance. They find that the cartels increased prices by 25 cents per pound and 9 cents per pound respectively. During the lysine cartel the variance of prices was higher than in more competitive periods, while the opposite result was found for the citric acid cartel. Abrantes-Metz et al. (2012) and Fouquau and Spieser (2015) included the possibility of collusion in their analysis of Libor submissions. Abrantes-Metz et al. (2012) mainly analyzed the

(12)

12 aggregate Libor time series to detect anomalies suggesting both individual and coordinated manipulation. Fouquau and Spieser analyzed individual banks’ Libor submissions and found breaks in the data corresponding to significant economic events. Moreover, they use threshold regression models and a hierarchical clustering method to identify banks that have similar quoting strategies.

Theoretical models that shed light on manipulation incentives of banks have been introduced by Chen (2013), Snider and Youle (2012) and Diehl (2013). Snider and Youle investigated the Libor in a theoretical model that predicts quote clustering of the banks’ rates around the pivotal 25% and 75% quartiles (due to the trimming of the rate), a test with data on Libor quote submissions confirms this result. In earlier work (Boot, 2015) I derived bank’s optimal quoting behavior under individual manipulation as well as under collusion. In a cartel, banks submit quotes in the direction of the average exposure position in the cartel, rather than in the direction of their own exposure position and banks are willing to submit more extreme quotes, as under collusion they have a larger effect on the rate. I found that a cartel could be profitable under this mechanism in the benchmark model, as well as in the two extensions. The first extension was motivated by the European Commission’s (2013)

evidence which suggested that not all banks were part of the cartel at all times, and consists of allowing for banks to become “active cartel” members whenever their interests were in line with other cartel members. In the second extension , colluding panel banks can use their knowledge on the future rate to create a more favorable trading exposure position before the rate becomes effective.

2.3 Challenges of screening for manipulation and collusion in interbank rates Collusion in interbank rates is different than the “classic” cartel in which firms agree to charge a higher price in order to obtain more profits. In interbank rates, banks can desire either high or low rates, depending on whether they are net lenders or net borrowers.

Moreover, banks in the panel can have opposite interests on certain trading days but similar interests on other days. Therefore, in the of collusion the group of conspiring banks may vary over time due to changing interests. Investigations by regulators and antitrust authorities also show that banks did not attempt to manipulate the rate on every trading day. Banks may have little at stake on some days, while on other days they could have big fixings and would have more incentives to manipulate the rate. This occasional nature of the manipulation could significantly increase the difficulty of detection.

(13)

13 Collusion could also lead to higher profits if it takes place in the form of an information exchange and banks use this information to create more beneficial trading positions, also known as front running. With this type of collusion, the actual rate would not be manipulated and information on profits and trading positions of banks would be needed to detect it. Collusion can take many forms and may not be beneficial for each of the banks on every trading day. Banks could be interested to engage in it if it leads to higher profits on the longer term.

Several papers have investigated manipulation or collusion empirically in Euribor and Libor. However, evidence is mixed and highly dependent on manipulation hypotheses and the assumptions made in these models. In particular, determining what the rate should have been absent manipulation on a certain day is difficult, and one of the main drivers of results. Abrantes-Metz et al. (2012) find some anomalous quotes, but no evidence for material manipulation of the Libor. The authors acknowledge that better proxies of short-term borrowing costs may help explain some of the statistical behavior they find. On the other hand, Snider and Youle (2012) find that manipulation was widespread and may have persisted until more recently, which is inconsistent with publicly available information on

investigations. In earlier work, Snider and Youle (2010) show that bank’s Libor submissions are difficult to rationalize with other observable bank borrowing costs measures, including with same bank’s Libor submission in another currency panel.

2.4 Interbank rate reforms

Concerns about the integrity of interbank rates date back to at least 2008, when a Wall Street Journal article was published (Mollenkamp and Whitehouse, 2008). After the scandal came to light, several reforms have been put in place. Among others, significant steps have been taken in order to improve the governance, auditing and oversight of both rates. The next two

sections focus on the reforms regarding the design of the rates and the submission requirements.

2.4.1 Libor

Following the 2012 Wheatley Review, the Libor has been evolving. After 18 months of consultation with regulators, central banks and market participants, the ICE Benchmark Administration has published two position papers and a roadmap (ICE Benchmark Administration, 2014, 2015 & 2016).

(14)

14 In the course of 2016, measures will be updated and implemented, including a uniform

submission methodology, a clear and robust Libor definition and fully transaction-based submissions where possible. The ICE Benchmark Administration plans on using a “waterfall” process to determine the rate, meaning it starts with transaction as the basis of the rate, but relies more on human estimates (or “expert judgment”) when the volume of trading falls. In particular, the method consists of three levels (ICE Benchmark Administration, 2015):

 The volume weighted average price of the bank’s eligible transactions11

 Data derived from transactions (historical, interpolation, extrapolation or parallel shift)12

 Expert judgment, appropriately framed13

Increasing banks’ volume of transactions on which to base their submissions is one of the main goals of the administrator. Interbank transaction volumes dropped during the crisis and have not restored since. The rate was therefore based on a small volume of transactions and depended largely on the “expert judgment” of bank employees. By expanding the eligible transactions for the rate beyond interbank lending the administrator can switch to a more transaction-based rate. Eligible transactions will now also include short-term funding coming from central banks, corporations, non-banks financial institutions and other counterparties.

A recurring topic in the consultation process was the view that banks should only provide the administrator with data of transactions, who would then calculate and publish the rate on a daily basis. This would reduce the need for subjective decisions and expert

judgment. In the course of 2016, the ICE Benchmark Administration plans on completing a feasibility study of further evolving Libor to a centralized calculation using a robust

algorithm.

Another debated aspect is on the use of rates such as Libor for financial products that are very different from unsecured interbank transactions. Due to the differences between the unsecured bank funding transactions and the many derivative transactions, the Market

11_{i.e. transactions that took place between the publishing of the previous rate and 11.00 a.m. on the submission}

day. More recent transactions receive a higher weight. 12

Historical transactions of a maximum of 3 to 10 days (depending on the maturity) can be used and would be adjusted by the change of a correlated rate (e.g. OIS, Central Bank rates). Interpolation will be limited to determining the 2 – 6 month maturities, by using the transacted rates from adjacent maturities. Parallel shift means that when a maturity has no transactions and only one neighboring maturity has a transaction-based rate, a bank can parallel shift rates based on the day-on-day change in value of the neighboring maturity’s rate.

13

The IBA has published a list with allowable inputs to base the judgment on. Allowable inputs include related market instruments, market observations (observed third party transactions and broker quotes) and funding transactions which are not eligible under Level 1 or 2 (either because of the counterparty or for exceeding the historical rolling date). Framing needs to be accompanied by documentation of the rationale and supporting evidence.

(15)

15 Participants Group is of the opinion that there should be a to the greatest extent possible transaction-based “IBOR” plus an alternative reference rate for other financial transactions (Financial Stability Board, 2014b).14

2.4.2 Euribor

Similar to the administrator of Libor, the administrator of Euribor (the European Money Markets Institute) plans on widening the range of eligible transactions for the calculation of the rate (EMMI, 2015). Unsecured cash deposits attracted from corporations, financial institutions and other deposit-taking corporations, as well as short-term securities, will be eligible. For Euribor, panel banks are asked to select transactions on day t, for their submission on day t+1.15 In its revised code of conduct (EMMI, 2016), a wider number of eligible transactions are listed that have to be used as a basis for the submissions. While some transactions have to be given more weight than others, Euribor panel banks seem to have a bit more freedom when determining the methodology for submission compared to those in the Libor panel. Like in the case of the Libor, there is still room for expert judgment, which has to be appropriately framed. The data reported to the administrator on a daily basis is the

submitted quote, but also the methodology and all input data is to be retained.

As indicated in its consultative position paper (EMMI, 2015), EMMI is exploring the option of switching to an entirely transaction-based Euribor in late 2016. In cases where the panel bank has no eligible transactions, EMMI plans on using a gap-filling technique, using the most recent volume-weighted average rate within a given number of days.16 If a bank still does not have eligible transactions, it is excluded from the calculation of the index for that trading day. The resulting set of submitted quotes is then ordered and the trimmed group average (the average of the middle 4 or 5 rates, depending on whether the number of quotes is even or odd) then determines the rate.

14_{In particular, they believe that most derivative products, that are intended solely to manage duration risk,}

should generally be encouraged to start using reference rates based on near risk free rates, while for example transactions with corporates and syndicated loans do benefit from using rates that contain bank credit risk.

15_{There is no minimum transaction volume.}

16_{4 trading days will be used for most maturities. Only in the case of the 12-month Libor the approach extends}

(16)

16 3. Manipulation and collusion after the reforms

The proposed reforms for Euribor and Libor aim at improving the integrity of the rate in order to make it more robust to manipulation. One of the main proposed reforms is to make the rates transaction-based to the largest extent possible. Although these reforms are a major step in the right direction, there are still ways in which banks could manipulate the rate. Given the large financial interests at stake, this should remain a concern.

Part of the rate-setting mechanism still relies on banks’ expert judgment, in particular during times when there are few transactions. Therefore, manipulation can still take place in a similar ways as prior to the reforms. Before to the reforms no clear guidelines were in place regarding how banks should choose their quotes, leaving more room to manipulate the rate. Although the switch to the current, more transaction-based rate has reduced the scope for manipulation, it is still especially vulnerable to manipulation when fewer transactions take place. The feasibility of switching to an entirely transaction-based Euribor and Libor is under investigation. In the case of Libor, Level 3 (expert judgment based) quotes will used when no transaction data is available. These quotes will have to be accompanied by a description of the rationale and evidence of the used inputs, which leaves room for banks to choose inputs according to their own trading interests. In the case of Euribor, expert judgment can always be used, even when there are eligible transactions. Expert judgment has to be framed

appropriately and (with a justification) banks can also disregard certain, normally eligible, transactions when determining their quote. However, manipulation is easier during times with fewer transactions. Typically this means there is less evidence of what the rate should be and therefore proving manipulation is more difficult.

Essentially, manipulation through expert judgment would happen in a similar way as manipulation prior to the reforms. Manipulating banks submit quotes that are in line with their own trading exposure positions. If a bank is a net lender it would benefit from a higher rate, while if it’s a new borrower it would prefer a lower rate. However, banks face “detection costs” and would try to submit quotes that are not too extreme to prevent suspicions from regulators. After the scandal and the reforms it is likely that banks have become more cautious as authorities have become more vigilant and submission guidelines are clearer and therefore detection costs have increased. An empirical test for Libor has been designed by Snider and Youle (2012) who study portfolio exposure position driven manipulation. They predict that quotes will bunch around the 25% and 75% quartiles, as the quotes are trimmed when calculating the rate and therefore submitting quotes that are more extreme than those at the 25% and 75% quartiles is unnecessarily risky. Several other papers aim at detecting

(17)

17 manipulation of the rate, but results are mixed as there are no obvious proxies for Libor or Euribor and therefore it is hard to determine what the rate should have been absent

manipulation.17

In earlier work (Boot, 2015) the optimal quote for a cartel was found to be:

(1)

Where is a cost term for manipulating the , is the average exposure position of all banks in the panel and c is the true borrowing cost (which banks are supposed to quote). Therefore banks in cartel would submit the same quotes, which are driven by the portfolio exposure position interests of the cartel as a whole. Abrantes-Metz et al. (2006) designed a variance screen for collusion. The idea behind the screen is that a conspiracy not only increases the price, but also reduces the variance of prices. Although cartels can operate in very different ways across industries, this screen could also be useful for interbank rates as colluding banks are predicted to submit the same quotes, while different quotes are submitted by banks that only take their own interests into account. While the effect of a cartel on the mean rate is ambiguous, the prediction for the variance is that it is lower under collusion. This is also consistent with Figure 1 (Abrantes-Metz et al., 2012) where the cross-sectional coefficient of variation18 is shown for the 1 month USD Libor. In the figure the coefficients of variation are much higher from 8/2/2007 onwards, which suggests there could have been collusion before that date.

17_{See among others Gandhi et al. (2015), Fouquau and Spieser (2015), Abrantes-Metz et al (2012) and Kuo et al.}

(2012).

18_{In the case of Euribor and Libor the coefficient of variation is given by the standard deviation divided by the}

(18)

18

Figure 1. Libor 1 month cross-sectional coefficient of variation for banks’ quotes. Source: Abrantes-Metz et al. (2012)

This chapter proceeds with two main ways in which banks could still manipulate the rate or collude after the reforms. Section 1 discusses the manipulation of eligible transactions, while section 2 focuses on manipulation based on the exchange of information. Both sections outline the hypotheses and data requirements to screen for suspicious behavior.

3.1 Eligible transaction rigging

The first way in which banks could manipulate the rate after the reforms is by manipulation of the transactions that are eligible for calculation of the Euribor and the Libor. Only a small part of the transactions that form a bank’s exposure position to Euribor and Libor have to be used to justify banks’ submissions, while the other part cannot be used. As can be seen in Tables 1a and 1b, by far the largest volume of the Euribor and Libor market footprints can be attributed to over-the-counter derivatives, trading between two parties without supervision of an

exchange, which are not standard eligible transactions for both rates.19 Banks are asked to base their rate on a number of different transactions with banks and corporations which have a much smaller volume. Banks in the Libor panel can use transactions in which they receive

19_{According to the Libor Roadmap (ICE Benchmark Administration, 2016) these transactions can be used as a}

(19)

19 funding from a range of wholesale market counterparties.20 In the case of Euribor an interbank transaction is defined as a cash deposit between two credit institutions maturing by one year from inception, transactions that most closely accord with this definition should be given the highest weight (EMMI, 2016). There is a wide range of examples of relevant market data that can be used and banks should identify the methodology themselves. Moreover, banks can also use executable and non-executable quote data, but actual transactions should be given higher priority.

Asset class Outstanding

volume ($BN)

% EURIBOR-related

Most common tenors

Loans Syndicated loans21 535 90% 3m and 6m

Corporate loans (bilateral)

4322 60% 3m and 6m

SME loans 1518 60% 1m and 3m

CRE/Commercial mortgages22

- 60% 1m and 3m

Retail mortgages 5073 28% 3m, 6m and 12m

Consumer loans 800 Low 1m

Other Loans to Households 1082 Low 1m Bonds Floating/Variable Rate Notes 2645 70% 3m

Covered Bonds 2557 23% 3m and 6m

Securization RMBS 952 100% 3m

CMBS 107 100% 3m

ABS 197 91% 1m

CDO 165 78% 3m and 6m

OTC Derivatives IR Swaps 137,553 High 1m, 3m and 6m

FRAs 25,559 High 1m, 3m and 6m

IR Options 24,249 High 1m, 3m and 6m

X-currency swaps 9731 High 1m, 3m and 6m

ETD Derivatives IR Options 12,439 100% 3m

IR Futures 4905 100% 3m

Deposits Retail deposits 8102 Low

Corporate deposits 2336 Medium 1m and 3m

SME deposits Medium 1m and 3m

Table 1a. Euribor Market Footprint overview. Source: adapted from Financial Stability Board (2014a).

20

Transactions with the following counterparties are eligible: banks, central banks, corporations (for maturities >35 days), government entities, multilateral development banks, non-banks financial institutions, sovereign wealth funds and supranational corporations (ICE Benchmark Administration, 2016).

21_{Significant overlap exists between syndicated loans and corporate business loans.}

22

(20)

20

Asset class Overall volume

($BN)

% LIBOR-related

Most common tenors

Loans Syndicated loans23 ~3400 97% 1m, 3m and 6m

Corporate loans (bilateral) 1650 30-50% 1m and 3m Noncorporate business loans 1252 30-50% 1m and 3m CRE/Commercial mortgages 3583 30-50% 1m and 3m Retail mortgages 9608 15% 6m Other Loans to Households 2080 Low 1m and 3m Bonds Floating/Variable Rate Notes 1470 84% 1m and 3m Securization RMBS ~7500 24% 1m and 3m CMBS ~636 4% 1m and 3m ABS ~1400 37% 1m and 3m CLO ~300 71% 1m and 3m

OTC Derivatives IR Swaps 106,681 65% 1m, 3m and 6m

FRAs 29,044 65% 1m, 3m and 6m

IR Options 12,950 65% 1m and 3m

X-currency swaps 22,471 65% 1m, 3m and 6m

ETD Derivatives IR Options 20,600 98% 3m

IR Futures 12,297 82% 3m

Deposits Retail deposits 7110 Low 1m and 3m

Corporate business deposits 948 TBC 1m and 3m Noncorporate business deposits 908 TBC 1m and 3m

Table 1b. Libor Market Footprint overview. Source: adapted from Financial Stability Board (2014a).

The large gap in volume between transactions that are eligible as a basis for the quote submissions and those that are not but have a large impact on a bank’s portfolio exposure position, may give incentives for manipulation. In particular, banks could manipulate the transactions that are eligible as a basis for the quote submissions, engaging in transactions at rates that are beneficial to their overall exposure position. For example, if a bank would have a large overall net lending position, it would benefit from a higher rate. A bank could then engage in eligible transactions (for example loans) at higher rates, which it can then use to justify a higher quote submission. Even though these “manipulated” transactions themselves may not benefit the bank, the gains from the significantly higher volume of other types of transactions (that are not used for the calculation of Euribor and Libor) could be much higher. The rate-setting process for Euribor and Libor is illustrated in Figures 2a and 2b. A

manipulating bank would look at its exposure position to the rate (that becomes effective at t+2) at the start of the input transaction window. It would then engage in transactions that are

23

(21)

21 eligible and beneficial for this trading exposure position during the input transaction window. In the case of Libor, banks submit quotes immediately after the end of the input transaction window, while Euribor’s input transaction window is the trading day and quotes are submitted by the end of the morning of the next trading day.

Figure 2a. Eligible transaction rigging timeline Libor

Figure 2b. Eligible transaction rigging timeline Euribor

Two sources of information are needed when screening for eligible transaction manipulation. First, the overall exposure position of a bank at the beginning of the input transaction window is needed. This exposure position consists of all transactions, including those that are not eligible as a basis for the quotes, that took place in the past and affect the bank’s portfolio exposure position towards the rate at time t+2. Second, the eligible transactions for

calculation of the rate are needed to determine the resulting rate for each of the banks. Banks that rig eligible transactions end up having a quote that favors the exposure position to the rate at time t+2. Banks manipulating their eligible transactions will submit quotes higher than the panel average whenever they have a positive trading exposure position.

Hypothesis 1. Banks that manipulate eligible transactions, have eligible transactions that are more strongly correlated with their initial exposure positions than other banks in the panel.

This hypothesis can be tested by looking at the correlation between a bank’s exposure position and its quote compared to other banks in the panel and testing for breaks in this

(22)

22 relationship. This hypothesis has to be tested for each bank in the panel and a resulting graph would be similar to the one in Figure 1, where a sudden increase in the correlation could indicate eligible transaction rigging.

Banks’ eligible input transactions, needed to verify whether the submitted quote has not been manipulated, are currently not provided on a daily basis. However, Euribor panel banks should be able to provide these upon request (EMMI, 2016). More importantly, banks’ total portfolio exposure positions to the rates and all transactions determining these positions are not publicly available. In order to screen for eligible transaction rigging this data should have to be requested by the regulator.

3.2 Front running

Second, even without manipulating quotes, banks could illegally exchange information and use this to their benefit. In particular, by exchanging information on each other’s submissions banks could obtain information on the future rate and accordingly create more beneficial trading positions, also known as front running. From previous work (Boot, 2015) it followed that by adapting exposure positions in the direction of the rate before it becomes effective, banks could benefit even more from being in a cartel. However, this was assuming that banks not only exchanged information, but also colluded when determining their quotes. This way, banks were willing to give more extreme quotes, knowing that they could also adapt their exposure position such that they would benefit more from the rate, even if they didn’t have the same interests as the cartel as a whole. In the current scenario it is assumed that banks do not manipulate their eligible transactions and therefore the rate itself is not manipulated.

In the case of Libor, the rate fixed at time t (using transactions from after publication of the previous rate until the morning of the fixing), is published later that morning and becomes effective at time t+2. Banks in the panel could have an advantage by exchanging information on their (future) submissions for time t, thereby getting a better idea of what the final rate (the trimmed average) would become, and use this information to create a better trading exposure position to the rate before it becomes publicly available at 11.45 that

morning. For example, if after consulting the panel it follows that the Libor on day t+2 (when the rate becomes effective) is going to be high, a bank would want to have a positive (net lending) trading position towards the rate. A bank could then engage in transactions with banks or corporations or trade derivatives in order to create a more favorable exposure position for that day. In the case of Euribor, panel banks submit their quote, based on the transactions of the previous trading day, before 10.45am, the rate is then calculated and

(23)

23 published on the same day. Figures 3a and 3b illustrate the rate-setting timelines for Euribor and Libor. In principle, the window of manipulation for Euribor transactions is larger as banks already have all eligible transactions for calculation of the rate by the end of the previous trading day. However, the manipulation window for both rates could be wider than the one shown in Figures 3a and 3b, as banks may already have some idea of what the rate is going to be before the end of the eligible input transaction window. Banks could exchange this information earlier, or even commit to not engaging in eligible transactions in the last part of the eligible transaction window, in order to maximize their profits from front running.

Figure 3a. Libor rate-setting timeline

Figure 3b. Euribor rate-setting timeline

When screening for manipulation based on information exchange two sources of information have to be looked at on a daily basis. First, the eligible input transactions for Euribor or Libor are needed. These transactions are the ones that serve as a basis for banks’ quote submissions. Second, all transactions that take place between when the window of eligible transactions ends and when the rate is published are needed. This second group of transactions contains all the transactions that affect a bank’s exposure position. With these transactions it can be determined how the exposure position (towards for example the Euribor on value day t+2) of a bank has changed during the manipulation window. Banks that often adapt their exposure

(24)

24 positions in the direction of the rate may have participated in a cartel in the form of an illegal exchange of information. Also periods in which most banks adapted their exposure in the direction of the rate could indicate collusion. Therefore the information exchange collusion hypothesis is:

Hypothesis 2. In periods of collusion, the correlation between the rate and the change in exposure position during the manipulation window is higher than in other periods.

In particular, this could be tested by assessing whether there are breaks in the correlation between the rate and the transactions on the day before it becomes public. This hypothesis has to be tested for each bank in the panel and a resulting graph of the analysis would be similar to the one in Figure 1, where a sudden increase in the correlation could indicate collusion or front running.

(25)

25 4. Screening for manipulation and collusion

This chapter focuses on the implementation of the screens for manipulation and collusion. In section 1 data is generated to simulate the screen for eligible transaction rigging. The data generating process is based on banks’ behavior from the theoretical model. The results of the simulations are presented in section 2. Section 3 explains how screening for front running works in a similar way. Section 4 discusses the challenges of screening in practice and how banks could evade the screen.

4.1 Theoretical evidence and the data generating process

Since the data needed to screen for manipulation and collusion is not publicly available, data is generated according to the theoretical model from Boot (2015). Banks have a certain overall exposure position towards the rate when it becomes effective ( ), which includes many types of transactions that cannot be used to underpin quotes for Euribor or Libor. Therefore, banks have an incentive to manipulate their quote ( ), instead of quoting the true borrowing costs ( ) which they are asked to quote. From previous work (Boot) it follows that the optimal quote for a bank is given by:

(2)

The total number of banks in the panel is given by n and is a cost term for giving a

manipulated quote. It is optimal for panel banks to manipulate the rate by . The degree of manipulation is decreasing in the number of banks in a panel, because with a larger panel a bank’s individual effect on the rate is smaller. Manipulation is also decreasing in the cost term . These costs are for example the risk of getting caught for manipulating the rate. Further details on the derivation can be found in Appendix A.

This result also serves as the intuition for how banks would manipulate their eligible transactions, as shown in Figures 2a and 2b. After the reforms, the weighted average of the eligible transactions essentially is a bank’s quote. Under this type of manipulation, a bank would choose its eligible transactions such that the resulting rate favors its overall exposure position to the rate, by choosing its optimal quote .

To simulate the screen, data is generated according to the following data generating process. A bank i’s quote is given by:

(26)

26 Where is the degree of manipulation that followed from the theoretical model. is a dummy variable which takes on value 0 in periods of no manipulation and 1 in periods of manipulation. The error terms are drawn from a distribution (i.e. they are identically independently normally distributed with a variance of ).

When screening for eligible transaction rigging, the variable of interest is the correlation between the exposure position ( and the extent or direction of manipulation, which is given by:

(4)

Absent manipulation, a panel bank would “manipulate” the rate by , this can be seen as a bank’s error when assessing what het quote should be. In the simulations in the next chapter, the true borrowing costs are normalized to zero for simplification purposes. This does not affect the main intuition of the screen in this chapter. In reality, these costs move over time and the extent of manipulation is the difference between the quote and the true borrowing costs. Banks that are net lenders and would like the quote to be higher, submit a quote that is higher than the true borrowing costs, while those that would like a lower rate submit a quote that is lower than the true borrowing costs.

The bank’s exposure position to the rate, , is modeled as an AR(1) process, an

autoregressive model of order 1, as shown in (5). Banks have financial contracts that are tied to Euribor and Libor, including ones with durations of multiple trading days. Although exposure positions can change from one trading day to another, it is likely that there is some correlation over time.

(5)

The error terms are identically independently normally distributed with zero mean and variance . Simulations are carried for different values of coefficient , all with (assuming stationarity).

Subsequently, the variance of is given by:

(6)

A dummy variable is generated to distinguish between periods of manipulation and periods of no manipulation:

(27)

27

(7)

Where is the first day of collusion and is the last day of collusion.

The property of interest in this screen is the correlation between the degree of manipulation ( or simply after the simplification in these simulations) and the exposure position ( ). From the properties of the AR(1) model it follows that under manipulation (with the dummy variable being 1) the covariance of and is given by:

(8)

The variance of the quote reduces to the variance of when no manipulation takes place. The variance of quote is given by:

(9) Therefore under manipulation the correlation between the degree of manipulation and the exposure position is given by:

(10)

A short routine in Stata to generate this data as well as the graphs in the next section is shown in Appendix C.

4.2 Simulations for eligible transaction rigging

In this section, the screen for eligible transaction rigging is simulated based on the data generated in the previous section. Data is simulated for 600 trading days, which is a period of a bit less than 2.5 years. The rolling correlations between the direction of manipulation ( and the exposure position are simulated for different parameter values. For these simulations and are both normalized to 1. The simulations are carried out for three different panel sizes: 10, 25 and 50 banks. The panel size of 10 banks is close to the current smallest Libor panel, which comprises 11 banks. The largest panel size chosen is 50, which is close to the number of banks the Euribor panel used to comprise before it came to light the rate had been manipulated. In this section, simulations are shown for a panel of size

(28)

28 25, since this is reasonably close to the number of banks that are currently in the Euribor and the USD Libor panels.24

Figures 4a-4i illustrate the simulations when carried out once for each of the different parameter values, where the correlation are calculated over a rolling window of 50 trading days. The red line is what the correlation between exposure position and the extent of manipulation would be absent eligible transaction rigging. The blue line is the same correlation when manipulation takes place between day t=200 and t=400. It follows that manipulation is easier to detect when is larger, which is when banks’ exposure positions are more correlated with the exposure position in the previous period. This is because it reduces the variance in exposure positions . Manipulation is also easier to detect when the costs for manipulation ( ) is smaller. If manipulation is less costly, panel banks do it more and therefore it is easier to observe. This can also be seen more clearly in Figures 5a-5i which show the average correlations for the same parameter values after 200 simulations.

The simulations for panels of 10 and 50 banks can be found in Appendix B1 and B2 respectively. Essentially, the larger the number of banks in the panel, the harder it becomes to detect manipulation. This is because with a larger number of banks in the panel, the individual impact on the rate is smaller and therefore banks are less willing to manipulate. In the optimal quote resulting from the theoretical model, the number of banks and the cost of manipulating gamma ( ) have the same mitigating effect on the extent of manipulation.

24

(29)

29 Figur es 4 a –4 i. S im u lat ions w it h a n w it hou t m ani pul a ti on. N um be r of pane l ban ks : n=25 =0.1 =0.05 =0.1 =0.3 =0. 5 =0. 9

(30)

30 -1 -. 8 -. 6 -. 4 -. 2 0 .2 .4 .6 .8 1 co rr e la ti o n 0 50 100 150 200 250 300 350 400 450 500 550 600 50 trading day window

with manipulation without manipulation gamma=.05; rhov=.1; n=25 -1 -. 8 -. 6 -. 4 -. 2 0 .2 .4 .6 .8 1 co rr e la ti o n 0 50 100 150 200 250 300 350 400 450 500 550 600 50 trading day window

with manipulation without manipulation

gamma=.3; rhov=.9; n=25 _Figur

es 5 a –5 i. 2 00 s im ul a ti ons ave rage c orr el a ti on w it h a n w it hou t m ani pul a ti on . n=25 =0.05 =0.1 =0.3 =0. 1 =0. 5 =0. 9

(31)

31 Table 2 reports the average correlations for the different parameter values, after simulating 200 times. Note that this is the correlation over the entire period of 600 days and therefore includes days on which no manipulation took place. These correlations are therefore generally smaller than the rolling correlations in times of manipulation as shown in Figures 5a-5i.

Table 2. Average correlations of extent of manipulation and exposure position from 200 simulations. Panel size n=25.

For the simulations shown until now and were both normalized to 1. Figures 6a -6c illustrate what happens for larger variances of these two error terms. Figure 6a is a simulation with the variances of the error terms normalized to 1 and manipulation costs =0.05,

coefficient =0.5 and a panel comprising n=25 banks. In Figure 6b the same simulation is shown with a higher variance of the error term . This higher variance increases the variance of the exposure position and therefore manipulation is more extreme compared to the smaller variance of the error term , the other component of the quote. A simulation with a larger is shown in Figure 6c. A larger variance for error term leads to a higher variance in the extent of manipulation ( ) that is unrelated to the portfolio exposure position . This leads to a lower variance between the two and therefore makes it harder to detect the

(32)

32

Figure 6a. =1 & [ =0.05, =0.5 and n=25]

Figure 6b. =2 & [ =0.05, =0.5 and n=25]

(33)

33 4.3 Screening for front running

Similar simulations as the ones shown for eligible transaction rigging can also be carried out for front running. The statistic of interest in this type of manipulation is the correlation between the eligible input transactions (i.e. the Libor or Euribor) and the change in exposure position due to transactions conducted in the window of time between when the quotes have been submitted and when the rate published.

Banks face costs for adapting their exposure position. A bank would not adapt its exposure position too much, fearing to attract unwanted attention from other parties in the market or regulators. Moreover, a number of financial contracts that affect exposure positions do not only affect the exposure of one trading day, but of multiple days. Creating a too

extreme exposure position to the rate on one day could therefore lead to losses in the future if the quotes of other banks change.

Similar to the data generating process in section 4.1, the intuition behind the data generating process could be an adapted version of the model in Boot (2015). This model is explained in more detail in Appendix A. Panel banks maximize their profits (10) by adapting their exposure position from to :

(11) The first component of the profit function are the gains from the adapted exposure position, the second part consists of costs for adapting the exposure position, where is a cost measure for adapting the exposure position. Maximizing with respect to yields the following optimal adapted exposure position:

(12)

It follows that it is optimal for a bank to adapt its exposure position in the direction of the rate. The extent of manipulation is decreasing in cost term . Similar to section 4.1, data could be generated for different values of the parameter . When screening for front running the correlation of interest is the one between the change in exposure position and the rate. Banks whose exposure position becomes more positive when the rate is higher or more negative when the rate is lower could be suspected of manipulation. Similar to the case of eligible transaction rigging, absent manipulation this correlation is expected to be close to zero.

(34)

34 4.4 Screening in practice

One of the main challenges of conducting these screens in practice is that the characteristics of the required input data are unknown. In practice, the true borrowing costs ( ) are not observed, although regulators could have an idea by looking at similar rates. The extent of manipulation, , could be measured by looking at the difference between the quote and an alternative proxy for the rate or the difference between the quote and the average quote in the panel. It would probably optimal to look at a combination of the two. As previous research showed finding a proxy for rates such as Euribor and Libor is not that straightforward. On the other hand, looking at the difference between the quote of an individual bank and the

difference with the average in the panel would make detection more difficult if a large number of banks are participating in the manipulation. With a large number of manipulating banks the norm would be rigged.

Moreover, very little is known about the portfolio exposure positions of banks. Futures, options and derivatives markets are often said to be zero-sum games, as for every winning contract there is a correlative losing one (Lynch, 2011). If only panel banks would trade in these products, the total exposure position of the panel would have to be zero, but in reality there are many more players in the market. The only paper that used some portfolio exposure position data is from Snider and Youle (2010). They use quarterly data that American bank holding companies have to publish on their interest rate derivatives and net derivatives revenue. This data includes the notional value of interest rate swaps held by banks, but misses the level of detail needed to carry out a proper analysis. For the simulations in this chapter it was assumed that absent manipulation, the correlation between the portfolio

exposure position and the quote is zero. However, in practice reasons may exist for these correlations not to be zero.

The distribution of this data also has implications for which statistical tests could be used to detect manipulation. If the correlation between the exposure position and the extent of manipulation is not normally distributed, tests that assume normality cannot be applied. Under normality, one could apply a one-sample t-test to rolling window s of correlations to see when the average of all values becomes statistically different from zero. A similar non-parametric test exists that does not require the assumption of normal distributions. This is the Mann-Whitney U test, which tests the null hypothesis that two samples come from the same population.

Screening is about detecting suspicious patterns in industries that may need further investigations, such as surprise inspections. It is not about creating evidence or proving

(35)

35 manipulation, but rather about detecting strong suspicion which can help regulators in setting priorities and directing resources. The screens in this thesis rely on a control group, which is the period before manipulation started taking place. In order to be able to catch sophisticated conspiracies or manipulation it is essential to keep screens up to date and for them to be costly to evade by panel banks.

Though potentially costly, panel banks could evade the eligible transaction rigging screen to some extent. Screening for eligible transaction rigging is about the correlation between the exposure position and the manipulation of the quote. Instead of choosing eligible transactions that maximize the profits given the portfolio exposure position at the start of the window, banks could also choose to maximize its profits given the portfolio exposure position halfway the eligible transaction window. This approach is illustrated in Figures 7a and 7b. However, this could only work if bank’s exposure positions change often enough during a trading day. If the bank for example desires a high rate at the start of the window because it is a net lender and still desires a high rate halfway the eligible transaction window, the screen would still detect it and the bank will only have lost time to create the eligible transactions.

Figure 7a. Timeline eligible transaction rigging screen

Figure 7b. Timeline evading the eligible transaction rigging screen

In the case of screening for front running it is assumed that banks communicate at the end of the eligible transaction window and have profitable transactions after the window, when the other players in the market do not yet know what the future rate is going to be. In

Screening for manipulation and collusion after interbank rate reforms

Screening for manipulation and collusion after interbank rate reforms

Contents