Navigating Liquidity 5

Transcription

1 Global Research December 2010 Navigating Liquidity 5 Regulatory adjustments: a new hope The current market configuration (high correlations with low volatility) has been picked up by the microstructure High-frequency arbitrageurs and market-makers find it easier to hedge their liquidity-providing positions and as a result correlations increase and volatility decreases, so long as no real shock occurs. The 6 May Flash Crash is a typical example of the potential consequences of such shocks. The European market microstructure is more fragmented than the US This is due to Europe's country-based structure and the concentration that prevailed before MiFID. On the one hand, the quality of a market design based on a pre-trade consolidated tape has been called into question by the 6 May event. On the other, European markets need a more robust way to convey information on the price formation process than the current situation, where most prices are pegged to primary markets. A comprehensive post-trade consolidated tape, including trades in dark pools and Broker Crossing Networks, could be a solution to protect Europe's microstructure from primary market outages such as that of NYSE Euronext on 13 October. Unless the quality of primary markets improves, Europe will probably end up with fragmentation similar to that in the US: fewer venues with more than 5% market share each. European small/mid caps not spared by fragmentation UK-listed small and mid caps are more fragmented than large caps listed on other European exchanges. We believe that there is potential for further expansion in BATS and Turquoise's market shares on small and mid caps. Tick size: a crucial element of Ultra High Frequency market design The current tables are not well balanced enough, in our view: tick size is often too small and its increments too large. Using the example of stocks quoting close to the thresholds defined in the FESE tables, we look at the effect of intraday tick size changes: the number of order book updates increases dramatically, and without any structural rationale, when the tick decreases. We propose more reasonable tables in the appendix to this publication. Disclosures available on

2 Introduction Three years after MiFID, an issue of Navigating Liquidity every six months: here is our fifth in-depth study of market microstructure. During this period, the expertise of CA Cheuvreux's Quant Research team dedicated to execution has increased, and is now internationally recognised. Cheuvreux has participated in public hearings organised by the European Commission, the French parliament, and regulators. We have also contributed to the most recent academic conferences and seminars in this field, more specifically via academic publications on the following topics: Modelling of the dynamics of the order book ("High Frequency Simulations of an Order Book: a Two-Scales Approach", by Charles-Albert Lehalle, Olivier Guéant, Julien Razafinimanana, in Econophysics of Order-Driven Markets; 2010); Optimal scheduling of large orders ("Optimal control of trading algorithms: a general impulse control approach", by Bruno Bouchard, Ngoc-Minh Dang, Charles-Albert Lehalle, in SIAM Journal of Finance, 2010); Optimal liquidity seeking ("Optimal split of orders across liquidity pools: a stochastic algorithm approach", by Gilles Pagès, Sophie Laruelle, Charles-Albert Lehalle, submitted to SIAM Journal of Finance); and Post trade performance analysis ("Rigorous post-trade market impact measurement and the price formation process", by Charles-Albert Lehalle, Ngoc M. Dang, Institutional Investors' Liquidity Guide 2010). Moreover, CA Cheuvreux's "Research Initiative" on "Trading and market microstructure" in partnership with the Collège de France, sponsors seminars and international conferences on the price formation process and optimal trading. We strongly believe that as a broker, CA Cheuvreux's role is to convey neutral information not only on traded stocks, but also on the price formation process in itself. The optimality and the efficiency of the markets relies theoretically and practically on the level of information shared by players, Navigating Liquidity is our biannual contribution to this. Because our team of Quant researchers is global, split between New York, Paris and London, it is time to provide an in-depth analysis of the microstructure of equity markets globally, and this fifth issue of Navi gating Liquidity thus provides a comparison of market designs in the US and Europe. Furthermore, as the price formation process lies at the root of price variations, we have studied how the specific conjunction of correlations and volatility could be enlightened by a microscopic approach. About the authors Charles-Albert Lehalle Global Head of Quant Research Charles-Albert Lehalle Global Head of Quantitative Research Contacts clehalle@cheuvreux.com Romain Burgot Statistician, Quantitative Group, Paris rburgot@cheuvreux.com Matthieu Lasnier Statistician, Quantitative Group, New York mlasnier-ext@cheuvreux.com Stéphanie Pelin Engineer, Quantitative Group, Paris spelin-ext@cheuvreux.com CA Cheuvreux's Quantitative Research Group Our 15-strong research and development Quantitative Group is located in Paris, New York and London. It is dedicated to the optimisation of trading techniques. Our work includes the design and prototyping of trading algorithms, performance analysis tools, models for pre-trade analysis, and on-line monitoring and analytics. The members of this group are statisticians, probabilists, econophysicists and computer scientists and all have a strong economic background. This group collaborates with quantitative finance laboratories across the world, to continuously improve CA Cheuvreux's execution capabilities. Execution Services Contacts GENERAL HOTLINES: Paris: London: New York: Ian Peacock Global Head of Execution Services / ipeacock@cheuvreux.com SELL SIDE BUY SIDE Jonathan Carp Mark Freeman Head of Alternative Execution Sales Europe Head of Alternative Execution Sales to Buy Side / jcarp@cheuvreux.com / mfreeman@cheuvreux.com 2

3 CONTENTS Introduction 2 I Cross-stock correlations: why uncertainty is not driving the current surge 4 The facts: high correlation, low volatility 4 Historical fact: When volatility rises, correlation tends to rise 6 Increasing intraday correlation 7 Overnight versus overday correlation 8 II Comparing the US and Europe 11 Two different forms of fragmentation? 11 Trade-through rule vs. execution policies: the original sin? 11 Europe is more fragmented than the US, will it follow the US example? 14 Reliability of the exchanges will determine the outcome 15 6 May 2010 in NY: how far are we from systemic risk? 16 Example of the Flash Crash impact: Procter & Gamble 17 The measures taken since 6 May 18 Mini flash crashes 19 Outages in Europe: bad news for the price formation process 21 III Tracking market fragmentation 23 Fragmentation of main indices: monthly figures indicate shocks 23 Small & mid-caps: catching up main indices 25 IV Tick size and ultra high-frequency market design 29 What happens when tick size changes? 30 A tick size decrease allows a smaller spread if tick size is a constraint 32 A small tick size makes posted liquidity more unstable 33 What is the next step for tick size? 34 Another way to fill order books with vanishing liquidity 36 From a regulatory standpoint 38 V Appendices 40 Appendix 1: Glossary 40 Appendix 2: Stock distribution among destinations according to their mean turnover 43 Appendix 3: Tables of tick sizes 44 Other publications

4 I Cross-stock correlations: why uncertainty is not driving the current surge FIGURE 1: CLES 60 AVERAGE PAIRWISE CORRELATION AND AVERAGE VOLATILITY FIGURE 2: NASDAQ 100+DJIA AVERAGE PAIRWISE CORRELATION AND AVERAGE VOLATILITY 50 volatility correlations volatility correlations /09/ /06/ /11/ /04/ /08/ /01/ /06/ /11/ /03/ /08/ /01/ /10/ /03/ /04/ /09/ /08/ /01/ /06/ /11/ /03/ /08/ /01/ /06/ /11/ /10/ /03/ The facts: high correlation, low volatility Correlations between stocks of a given universe are a convenient proxy to assess the degree of idiosyncrasy among these stocks. The more preponderant the idiosyncratic resultant of each stock, the more rewarding the stock-picking strategies. FIGURE 11 and FIGURE 12 on page 10 show that from July this year, correlations between components of the major world indices (DAX, CAC 40, CLES 60, IBEX 35, TOPIX top 150, NASDAQ 100+ DJIA) began to soar, while volatility remained low after the recovery from the 2009 spike. FIGURE 1 and FIGURE 2 above show correlation and volatility trends on both the CLES 60 (Cheuvreux Large UK Stocks) and NASDAQ 100+DJIA (Dow Jones Industrial Average) indices since September Note that for the purpose of our analysis we have built several in-house indices consisting of a selection of major country indices or the combination of two indices. Thus the CLES 60 includes the UK stocks of the DJ Stoxx ex Euro Large, the TOPIX top 150 comprises the top 150 highest market capitalisations of the TOPIX, and the NASDAQ 100+DJIA is made up of the combination of both indices' components. We designed two metrics to assess the degree of correlation and the volatility of any given universe. By volatility, we mean the average volatility across all stocks and, by correlation, the average of correlations between each possible pair of the considered universe within a six-month window. The phenomena observed in FIGURE 1 and FIGURE 2 above raised questions among a number of specialists in the finance industry. The usual pattern observed in a high correlation period is one of rising volatility. 4

5 The main explanations put forward by the experts were those of: 1) a combined effect of increased trading on index derivatives and on Exchange-Traded Funds (ETFs); and 2) of the rise in High-Frequency Trading (HFT). In fact, in a market where investors are more exposed to indices by holding positions on index derivatives, everybody's positions tend to be like everybody else's. As a result of this behaviour, links between the returns of each stock increase as does, subsequently, the correlation across the stocks. FIGURE 3: NASDAQ 100+DJIA VOLATILITY VS. CORRELATIONS FIGURE 4: CLES 60 VOLATILITY VS. CORRELATIONS Volatility Correlations Volatility Correlations The second explanation concerns High-Frequency Trading (HFT). HFT players try to make a maximum gain out of liquidity arbitrages. Put a different way, they try to make money out of the mispricing of an instrument due to the liquidity needs of other players on the market. If, for example, a mutual fund comes to the market and decides to sell a large quantity of XYZ stock in a fairly aggressive manner, this sell order will push down the share price of XYZ driving it to a lower level compared to its peers, leading to an arbitrage opportunity. The action of the HFT player, by taking on the role of the counterparty of the deal will increase the link between the stock and its peers. There are two main reasons for this: Avoid information leakage by using smartposting algos The HFT player will limit the market impact incurred by XYZ by increasing the liquidity available; and Simultaneously it will exert selling pressure similar to that incurred by XYZ on peers, leading to a price adjustment of those stocks. Note that being detected as a trader selling a large amount of shares is information leakage for which the mutual fund may have to pay the price. One way for the seller to avoid being detected could be to trade less aggressively and to try to capture the maximum liquidity by using smart algorithms, such as Crossfire and SmartX, tracking liquidity wherever it is. In a concentrated market, "natural" sellers and buyers (i.e. investors who will keep their positions for more than few days) are each other's counterparts. In the current fragmented markets, a natural seller's counterpart is often a high-frequency trader, which will sell the position back to a natural buyer a few minutes or seconds later. The high-frequency trader will assume a risk proportional to the product of the volatility by the square root of the time between its buy and sell orders. In return he will be rewarded by a fraction of the bidask spread and potentially by rebates if he succeeds in being a liquidity provider during the trades. This is the way high-frequency traders and market makers provide "liquidity bridges" between the available trading pools. They are said to be counterparts in around 70% of the US deals and 40% of the European ones. 5

6 In an highly-correlated and low-volatility market, the high-frequency trader will find it easy to provide such liquidity bridges: firstly because the risk (proportional to the volatility) is low, and also because the trader will be able to hedge his market risk using indices and correlated stocks. Each of his deals will be doubled by a buy or sell of an index for hedging purposes. This will automatically increase the correlations in the market. Therefore: the more correlated the stocks are, the more high-frequency trading will increase the correlations. This section of Navigating Liquidity 5 studies indicators that should capture these effects: intraday correlations and intraday volatility. We will see that their behaviour seems to confirm that such an intraday hedging activity exists. High correlation in the market often goes hand in hand with high volatility Historical fact: When volatility rises, correlation tends to rise FIGURE 3 and FIGURE 4 above represent the joint distribution of the correlations and volatility of the NASDAQ 100+DJIA, and CLES 60, respectively. The two solid grey profiles beside and below the scatter plot show the marginal distributions of correlation and volatility. The current value of the correlation and volatility is indicated by a green diamond inside the scatter plot (at the junction of the dashed lines). Moreover, we used these green dashed lines to show the positions of the correlation and the volatility on their respective marginal distributions. As can be seen on both representations, the green line inside the volatility distribution profile corresponds approximately to the median distribution, showing that current volatility is at a medium level. Conversely, the green line crossing the correlation distribution shows that the current value of the correlation is fairly high compared to its historical level. The green clusters show the trend in correlation and volatility since August. If we take a closer look at FIGURE 3 and FIGURE 4, we see that the previous correlation spike, in March 2009 (inside the circle), occurred during a period of high volatility. Indeed, when uncertainty spreads in the market, volatility tends to increase, reflecting the fact that the market players lack a clear vision. Reacting to the augmentation of market risk, investors become keener to hold portfolios tracking the performances of the market. In this way they hedge themselves against the risk of being wrong compared to everybody else. Such a reaction results in correlation increasing in line with volatility. The current surge of correlation across stocks is clearly not the result of a surge in volatility. In the next few sections we will endeavour to assess the other reasons that could lie behind these growing correlations. 6

7 Increasing intraday correlation FIGURE 5: CLES 60 CORRELATION TO VOLATILITY RATIO, INTRADAY VS. EXTRADAY FIGURE 6: CLES 60, RATIO OF INTRADAY TO EXTRADAY CORRELATION ratio intraday ratio extraday /10/ /09/ /07/ /05/ /03/ /01/ /12/ /10/ /08/ /06/ /04/ /03/ /01/ /01/ /10/ /09/ /07/ /05/ /03/ /01/ /12/ /10/ /08/ /06/ /04/ /03/2009 Intraday correlation on the CLES 60 has been rising since April 2009 The intraday correlation is currently rising faster than its extraday counterpart FIGURE 5 shows how intraday correlation divided by volatility on CLES 60 stocks follows the same trend as their daily values. For the sake of clarity, we will distinguish extraday correlation and volatility from their intraday counterparts. The intraday correlation and volatility have been computed based on an historical dataset going back to January For each day we took the average intraday correlation and volatility over the previous six months. We then compared the intraday values to the extraday ones. FIGURE 5 shows the trend of the intraday ratio of correlation to volatility compared to its extraday value. For comparison purposes these ratios have been rescaled. We can see that the trend of intraday correlation compared to intraday volatility is roughly the same at the extraday scale. In other words, the intraday correlations grew to the same extent compared to volatility as their extraday equivalents. Figure 6 represents the ratio of intraday correlation to extraday correlation. It gives a more precise idea of the trend in intraday correlation compared to extraday correlation. In fact extraday correlation grew more rapidly than intraday values as of August 2009, but intraday values caught up this past summer, as the intraday correlation continued to grow while its extraday equivalent began slowing down. The preceding observations, which relate the appearance of a growing intraday correlation could be interpreted in two different ways. But before we look at these interpretations in more depth, let us go off at a tangent for a minute and talk about a few facts from signal theory. We have two series of observed data, in this case the price time series of two stocks. We can model these series as the sum of a fundamental component, which bears the information (called signal), and a noise component (denoted noise). The following formula sums up this relation. Price t =Signal t +Noise t 7

8 The values of the noise are not related to the signals or to each other but to the nature of the system (as it happens, the market), which drives the information going to the observer. The noises of the two price series are therefore not correlated. In fact, the noisier the data, the less correlation there is. Conversely, the stronger the signal is relative to the noise, the stronger the correlation. For the time series of a share price the signal and the noise are hidden. The only accessible variable is the combination of both (P in the formula.) In market data the noise comes from the microstructure of the market. This means that the price of a stock depends on both the Walrasian equilibrium between the demand and supply, and the way the stock is traded (is it a continuous double auction or auction fixing?) One stylised fact of high-frequency finance is the smaller the sampling-time scale, the noisier the data. Thus when we look at intraday correlation, we should find that the correlation is lower than extraday values. Let us go back to the interpretation of growing intraday correlation. Based on the explanation above, we can view the increase in intraday correlation as the result of two possible phenomena: The Signal components of the share prices are more correlated; The Noise components of the share prices are diminishing. Of course both phenomena could happen at the same time and, in fact, this is what we believe is happening. In the next section we will break down the extraday correlation into overday and overnight correlations, in an attempt to justify the above assertion. Overnight correlations are higher than overday ones Overnight versus overday correlation FIGURE 7, FIGURE 8, FIGURE 9 and FIGURE 10 show the average pairwise correlation on CLES 60, DAX, CAC 40 and IBEX 35 components on both overnight and overday data. Calculation of the overnight correlation is much the same as that for extraday values. Instead of looking at returns between the close of two consecutive days, we take the returns between the close of the previous day and the opening of the subsequent day. As in the case for extraday correlation, we computed these metrics on a six-month sliding window. Our calculation of overday correlation follows the same methodology as its overnight version. The overday correlation currently seems to be lower than the overnight one. This remark has been valid since the start of the surge in June We have the impression when looking at FIGURE 7, FIGURE 8, FIGURE 9 and FIGURE 10 that the overday correlation plays the role of a baseline for the overnight correlation, which keeps bouncing above it. On the CLES 60, DAX, CAC 40 and IBEX 35, the increase in correlation is mainly driven by the overnight side. This is clearly the case looking at FIGURE 7, FIGURE 8, FIGURE 9 and FIGURE 10. What is also noticeable from these figures is that the overnight correlation tends to be higher than the overday one in periods of high extraday correlation. Further examination of these four figures reveals that for three indices (the CAC 40, CLES 60 and IBEX 35) their overday correlation is now at its highest level since In fact, the last overnight correlation spikes in 2008 and 2009 were pretty much as high as the current ones. This is not the case for the overday correlation. This observation can be related to the above-mentioned comment about growing intraday correlation and the impact of HFT on this correlation. 8

9 FIGURE 7: CLES 60 OVERNIGHT VS OVERDAY CORRELATION FIGURE 8: DAX OVERNIGHT VS OVERDAY CORRELATION overday overnight overday overnight /08/ /01/ /06/ /11/ /03/ /08/ /01/ /06/ /11/ /04/ /09/ /10/ /03/ /10/ /03/ /08/ /01/ /06/ /11/ /03/ /08/ /01/ /06/ /11/ /04/ /09/ FIGURE 9: CAC 40 OVERNIGHT VS OVERDAY CORRELATION FIGURE 10: IBEX 35 OVERNIGHT VS OVERDAY CORRELATION overday overnight overday overnight /10/ /03/ /08/ /01/ /06/ /11/ /03/ /08/ /01/ /06/ /11/ /04/ /09/ /10/ /03/ /08/ /01/ /06/ /11/ /03/ /08/ /01/ /06/ /11/ /04/ /09/

10 In fact these discrepancies result from two different microstructures. We wrote above that the value of the price was largely dependent on a microstructural parameter. FIGURE 7 to FIGURE 10 present a good example of that assertion. During the night the market is organised as an auction fixing and during the day it is a continuous double auction. In fact the opening fixing is less liable to become contaminated by loud microstructure noise and thus tends to better reflect the investors' consensus on the price of an asset than the closing fixing does. As a matter of fact, the overnight correlations are higher. Based on the above explanation of the discrepancies between overnight and overday correlations, it can be said that the current spike of extraday correlation is in part due to increasing overday correlation, which is itself the consequence of a less noisy price formation process (PFP). The growing level of HFT activity could be one explanation for this. FIGURE 11: AVERAGE PAIRWISE CORRELATIONS ACROSS COMPONENTS OF ALL THE MAJOR INDICES FIGURE 12: AVERAGE VOLATILITY OF THE COMPONENTS OF ALL THE MAJOR COUNTRYWIDE INDICES CAC CLES 60 INDU+NASDAQ IBEX DAX TOPIX CAC CLES 60 INDU+NASDAQ IBEX DAX TOPIX /10/ /05/ /12/ /07/ /02/ /09/ /04/ /11/ /06/ /01/ /08/ /03/ /10/ /10/ /05/ /12/ /07/ /02/ /09/ /04/ /11/ /06/ /01/ /08/ /03/ /10/

11 II Comparing the US and Europe Two different forms of fragmentation? European and US market microstructures clearly share common elements as shown, for instance, in the previous section on the current relationship between correlation and volatility. Moreover, the Regulation National Market System (Reg NMS) in the US and Markets in Financial Instruments Directive (MiFID) in Europe have points in common, as they both open up competition between trading platforms for equity markets. Nevertheless, the nature and consequences of liquidity fragmentation on each side of the Atlantic are not the same. It is said, for instance, that high-frequency trading is involved in 70% of trades in the US but "only" involved in 40% of European trades. The nature of the markets is similar enough that the US Security and Exchange Commission (SEC) issued its Concept Release "seeking comment on the structure of equity markets" in January 2010 (i.e., prior to the 6 May event) at around the same time as hearings and reports got underway to prepare the review of MiFID that will be included in "MiFID 2" (to be released during the first quarter of 2011). It is clear that legislators on both sides of the Atlantic simultaneously identified issues regarding the price formation process. Owing to CA Cheuvreux's positioning in the electronic trading space in both North America and Europe, we have enough data and experience to provide a comparative study of the discrepancies of microstructures across the "global fragmented equity market" consisting of European and North American trading venues. US competition between trading pools is "structurally" organised, while the European vision is more "selection" oriented Trade-through rule vs. execution policies: the original sin? The most obvious difference between US and European regulations is the way access to competing venues is organised. In the US, the market is designed "structurally" in such a way as to provide investors with access to "best execution", which is defined as the most attractive price for a given quantity. Conversely, MiFID in Europe not only organises competition between the trading platforms, but also between access to these platforms: each broker can define its "execution policy" (i.e. the methodology to be used to access trading pools). As each investor is free to choose its broker, competitive pressure between brokers' execution policies is said to result in selection of the best policies, and consequently optimised execution on a market-wide scale. Regarding this point, US competition can be seen as "structurally" organised, while the European vision is more "selection" oriented. In brief, the US market design relies on the "trade-through rule" (rule 611 of Reg NMS) stating that if a trading venue is aware that a better price is made available by one of its competitors, it has to route any order to this other venue. To allow for application of this rule, a "consolidated quote" is publicly available; it acts as the reference to check if there is a better price available elsewhere. As quotes provided by all the venues are blended in the consolidated tape, this process "neutralises" differences between the nature or quality of a quote without further inspection. Implicitly, any investor looking at the consolidated tape and relying on the trade-through rule will miss any information on the nature of the liquidity that is available on each pool (see the focus below on "the market for lemons"). 11

12 FOCUS: "THE MARKET FOR LEMONS: QUALITY UNCERTAINTY AND THE MARKET MECHANISM" In 1970, George Akerlof proved in "The Market for 'Lemons': Quality Uncertainty and the Market Mechanism" that two effects should happen in markets where goods from different sources were indistinguishable: a secondary market with a lower price will develop, and niche markets will emerge. He took the example of car markets, where bad cars are known as "lemons" and good ones as "cherries". In the case of liquidity pools, predatory flows are our "lemons" because it is not possible, in an order book, to distinguish orders from investors with a long-term view from those that come from short-term predators. As a result, the "secondary market" (with more lemons than natural liquidity) has to be cheaper: this is the case of MTFs. In addition, nontoxic flows thus have little interest in mixing with lemons, as this would lead to an increase in the average quality of the global flow. Liquidity consumers will thus agree to pay a better price to every provider, predatory ones included. To leverage good flows in such markets, providers have to create their own niches, where consumers will agree to pay them a price for this. Broker Crossing Networks are such niche markets, where providers of non-toxic flow want to protect its value from being averaged with predatory flows. This is also why lemon providers would like to prevent good flows from leaving mixing pools. In the US, the consolidated tape gives the erroneous feeling that all quotes provide the same quality of liquidity With no consolidated tape in Europe, MTFs decided to yoke a large part of their process to the BBO of the primary markets Having a single reference to look at gives a feeling of equivalence between all the liquidity available in the aggregated pools (exchanges and ECNs). In contrast, deals occurring in pools that are not visible in the consolidated quote are usually perceived as "different". This is typically the case of dark pools and Broker Crossing Networks (BCN). The liquidity offered by pools not contributing to the consolidated tape is often considered to be "not as equivalent" as liquidity found in consolidated pools. In practice, this is not the case; it is obvious that liquidity provided by a co-hosted high-frequency trader does not have the same nature as buy-side originated passive orders posted in a Broker Crossing Network (BCN), even if the latter is not visible in the pre-trade consolidated tape. In Europe, each broker's execution policy defines a way to aggregate liquidity. There would thus be as many "consolidated quotes" as there are brokers' execution policies. As the execution policies describe a mechanism that is updated periodically, each corresponding consolidated quote would also have to change at each update. An interesting point is that despite this absence of an official reference, MTFs like Chi-X decided very early on to peg their "market" orders to the best opposite of the relevant primary market (Euronext Paris for Paris-listed stocks, the London Stock Exchange for London-listed stocks, etc.). This means that if you send an order "at any price" to Chi-X, it will never consume liquidity at a worse price than the first best price on the primary market. This mechanism can be seen as a "limited one-sided synthetic trade-through rule". The limitations are that if a better price is available on the primary market, the order will stay on Chi-X at very close to the best opposite (at the price of the best opposite on the primary market) and will not be routed (see FIGURE 13 and FIGURE 14 for illustrations). Other European MTFs, such as NASDAQ OMX Europe and BATS Europe, proposed a Smart Order Router (SOR) embedded in the upper layers of their matching engine with behaviour similar to the trade-through rule. As we underlined in Navigating Liquidity 3, the issues about trade-through and best execution rules are easy to answer and analyse on the paper for aggressive orders (i.e. liquidity consuming orders) but not that easy for passive ones (i.e. liquidity providers). Furthermore, we explained in Navigating Liquidity 4 that even for aggressive orders it is not that easy in practice: the high frequency of updates in the order books prevents (for large caps at least) certainty that the "best split across venues" calculated on the basis of a snapshot of the available limit order book (LOB) will be optimal when the orders actually reach the targeted pools. This is acutely linked with latency issues. 12

13 FIGURE 13: TRADE THROUGH RULE FIGURE 14: CHI-X RULE FOR MARKET ORDERS Consolidated tape - C@10.99 Market data Order Book A $11.00 Order Book B $10.98 Order Book C $10.99 Other MTF E11.00 Primary E10.98 Chi-X E10.99 After looking at the consolidated tape, trading venue C knows that a better price is available at B, and it routes the order to this venue Buy order After looking at the primary market price, Chi-X decides to not generate a deal at a worst price but does not route the order Buy order European liquidity seekers focus on balancing orders across "lit" venues; US ones optimise liquidity capture outside of the tape In Europe, most large hidden orders potentially interact with the visible flows European mid-points emerged to protect orders from highfrequency trading and its threat of adverse selection This structural difference between US and European market designs, i.e. the tradethrough rule vs. execution policies, can be seen as the root of the differences in their respective microstructures. In the US, the pre-trade consolidated tape maintains an apparent equivalence between the contributing sources of liquidity. In Europe, competition among the brokers' execution policies and market access acts as a crucial element between the investors and the market. As a result, brokers' Smart Order Routers (SOR) and liquidity-seeking algorithms play a dominant role in obtaining efficient execution in Europe, even on lit venues, and in the US, liquidity-seeking methodologies are focused more on capturing liquidity outside the consolidated tape (i.e. dark pools), without suffering from adverse selection. Another consequence is the emergence in Europe of two kinds of "dark pools": The "mid-points", a type of dark pool offering execution at an "imported price" (often defined as the mid point of the primary market) for orders smaller than the "Large in Scale" (LIS) size defined by MiFID; The "LIS dark pools" dedicated to orders larger than the LIS size. A specificity of LIS dark pools is that most of them are part of an "integrated book"; this means that the hidden liquidity posted in one of these venues interacts with the associated "lit" book (for instance large hidden orders posted on Chi-X's dark pool are in fact inserted into Chi-X's visible order book as "fully hidden orders"). Most dark pools provide a parameter called "minimum quantity", which specifies that orders smaller than this minimum quantity cannot match the large hidden order. This can be used to prevent a large hidden order building resistance (i.e. via adverse selection) in the market, matching small aggressive orders on the visible book. On the other hand, most European market operators (Chi-X, Turquoise, Euronext, BATS, and Xetra) offer "mid-points" whose books are completely independent from their visible counterparts. These mid-points clearly emerged to protect orders from highfrequency trading and its threat of adverse selection. In November 2010, Turquoise announced the introduction of an optional random periodic auction in its mid-point process, claiming that this will benefit investors who fear adverse selection and gaming. 13

14 In Europe, the best choice for optimal execution is therefore not to choose between lit and dark, but to split an order across visible books, mid-points and large dark pools, taking into account that the price on the primary market (to which most mid-prices are pegged) is of specific importance. By way of illustration, CA Cheuvreux' liquidity-seeking algo CrossFire has specific features in the US and in Europe: In the US, it clearly embeds specific dynamics adapted to the choice between the different dark pools, according to their usual adverse selection capability; In Europe, CrossFire's specific features are focused on interacting with visible pools and mid-points. Europe is more fragmented than the US, will it follow the US example? The previous section identified microstructural differences between US and European equity markets. In this part we will focus on more macroscopic differences such as market shares. Referring to the report issued by the SEC to document their "Concept Release on the Equity Market Structure" (Jan. 2010), market shares on the visible US equity space stand at around: 25% for NASDAQ, 19% for NYSE, 16% for NYSE Arca, 12% for BATS, 12% for Direct Edge, 5% for NASDAQ OMX BX and the remaining 10% spread among smaller exchanges, ECNs, dark pools and brokers (see FIGURE 15 below). If we compare these figures to European ones (source: BATS Trading Europe, FIGURE 16), a typical split of market shares in the European space is: 20% for the LSE group, 15% for Chi-X, 15% for Euronext, 15% for Deutsche Börse, 6% for BATS, 6% for Bolsa de Madrid, 6% for SIX Swiss Exchange, 6% for NASDAQ OMX, and the remaining 11% distributed over small exchanges and MTFs (including dark pools). FIGURE 15: TYPICAL SPLIT OF MARKET SHARE IN THE US FIGURE 16: TYPICAL SPLIT OF MARKET SHARE IN EUROPE NYSE (inc Arca) NASDAQ (inc OMX) BATS Direct Edge Remaining (smaller than 5%) LSE Chi-X EuroNext Xetra BATS Madrid SIX NASDAQ OMX Turquoise Remaining (smaller than 4%) Source: SEC report/crems (Jan. 2010), BATS Trading Europe (Nov. 2010) 14

15 At a macroscopic scale, Europe is far more fragmented than the US Adding together the market shares of NYSE and NYSE Arca, NASDAQ and NASDAQ OMX, LSE London and LSE Milan, NYSE Euronext Paris, Amsterdam, Brussels and Lisbon, and Arca, etc., and keeping only those with more than 5% market share builds a view of the fragmentation in terms of key operators, Europe is far more fragmented than the US. The major implication of such a situation is that in terms of fee structure and of IT specificities (connections, protocols, etc.), execution in Europe is far more complex than in the US. Not to mention the clearing and settlement structure in Europe, which is also more fragmented than in the US. The root of this difference is the regional split of listings across different European countries, and must be understood in view of the fact that Europe is regulated on a percountry basis (FSA for London-listed stocks, AMF for Paris-listed ones, etc.), while in the US the SEC is in overall charge of the regulation. While the existence of the Committee of European Security Regulators (CESR) has helped harmonise the European landscape, the more harmonisation inside Europe, the better. We hope that the future European Securities and Markets Authority (ESMA), which should be given more powers, especially with regard to developing proposals for technical standards, will improve the harmonisation of capability in Europe. Moreover, fragmentation is as hard to define as market depth: the definition we use here is the complexity generated in terms of execution. The more communication protocols and channels that can be used, the more venues to interact with to capture large bunches of liquidity, the greater the fragmentation. The interesting point to note here is that Europe has gone from a concentrated market to a very fragmented one, given the country-based specificities. In the US, the trade-through rule and the existence of a consolidated tape implies a single global price formation process that has clearly accelerated a better spread of market shares among the main players. At a stock-based scale, US microstructure hosts more small venues and has a better spread of market shares Reliability of the exchanges will determine the outcome This raises the question of whether the European landscape is more fragmented because Reg NMS was put in place two years before MiFID: will European fragmentation reach the level of the US situation in a few months' time? Looking carefully at fragmentation for typical US stocks, we see that because Nasdaq and NYSE do not, in most cases, quote the same stocks, typical fragmentation (on visible pools) is: less than 50% on the first exchange (NYSE or Nasdaq), between 10% and 20% on the next one, 10% on BATS, 10% on Direct Edge and the remainder on other venues. In Europe the main venue usually has a ca. 65% market share, and the next three are split ca. 22%, 8% and 5%. At this scale, the difference between the US and Europe is that the largest US market share is lower than its European equivalent, as more small venues exist in the US. In Navigating Liquidity 3 (NL3, dated October 2009), we showed how the future of European fragmentation will follow one of the two following "schematic" patterns: One large venue with a ca. 75% market share, the second largest with 15% and the remaining 10% market share spread across smaller venues; Two main venues, one with around 50% market share and the other around 40%, with the remainder split across smaller venues. 15

16 In NL3, we showed that the main variable that will influence the final scenario is the reliability of major exchanges. This stemmed from the fact that to obtain best execution, traders have to penalise the trading pools according to their gross variance. In order to suffer less penalisation than a newcomer, an incumbent exchange has to provide an exemplary level of quality. The recent outages of European markets (commented on in another section of this issue of Navigating Liquidity) is further proof that historical exchanges do not provide this level of quality, even if the quality of the service they provide has improved in the past three years. As a result, we expect that Europe will most probably end up with fragmentation similar to that now established in the US: the main venues will provide around 50% of the visible liquidity, with a challenger close behind (Chi-X is currently in a good position for this role) and two or three "minor" providers. Of course the potential change of rules that could emerge from MiFID 2 in 2011 could change this scenario. 6 May 2010 in NY: how far are we from systemic risk? On 6 May 2010, the US stock market suffered a sudden and rapid drop followed by a rapid recovery. The S&P 500, which was already on a downtrend since the beginning of the day, saw its slide start to accentuate at 2.30pm. Within minutes, the equity and futures markets suffered a decline of 5%. The high volatility on the markets at the beginning of the day suggests that this Flash Crash might have been caused by a temporary and significant drop in liquidity. FIGURE 17: S&P 500 ON 6 MAY 2010 FIGURE 18: AVERAGE VOLATILITY ACROSS ALL S&P 500 STOCKS ON 6 MAY 2010 On 6 May 2010, the S&P 500 opened not far from its last closing price. The economic background in Europe was causing some concern. The debt crisis seemed to be worsening as sovereign CDS spreads were widening. In particular, there were mounting concerns about the sustainability of the Greek economy. In this context of unsettling political and economic news, the market was already down at 11:00am and volatility (FIGURE 17 and FIGURE 18) was reaching higher levels than in previous days. At noon the S&P 500 had lost 1%. At 1:55pm the S&P 500's slide was accelerating and volatility increased dramatically. The Flash Crash itself happened at 2:46pm when the S&P 500 was showing a 7% loss compared to its level 50 minutes previously. It quickly went back up to 1, This rapid recovery saw the S&P 500 rebuild 5.85% of its value compared to the low of the day. 16

17 Example of the Flash Crash impact: Procter & Gamble This decline impacted all stocks on the major US indices. Stocks such as Procter & Gamble lost 36% in seven minutes. At 2:40pm Procter & Gamble was trading at USD61.50 (FIGURE 20). At 2:47pm it was worth USD FIGURE 19 to FIGURE 22 show various metrics which provide a better understanding of what happened to the Procter & Gamble (PG) share on 6 May. FIGURE 19 and FIGURE 21 show the volumes and number of trades on PG during one-minute intervals throughout the day. FIGURE 22 presents the trend in the average bid-ask spread weighted by volumes during one-minute time intervals. Volumes traded on Procter & Gamble began to increase significantly at 1:30pm (FIGURE 19). This movement continued until 2:44pm, which is when the price of PG stock was dislocated. In line with the trend observed for volumes, the number of trades increased dramatically at around 1:50pm to reach 600 trades per minute at 2:22pm. This barrier was not crossed downward until the end of the day. In FIGURE 22 we can see the volume-weighted average spread, which spiked at above USD2 at 2:40pm. After this shock the spread declined rapidly back to 30 cents and then gradually reverted to a typical level of around 1.5 cents, until twenty minutes before the close. FIGURE 19: VOLUMES ON PROCTER & GAMBLE, DURING ONE- MINUTE INTERVALS, 6 MAY FIGURE 20: PROCTER & GAMBLE SHARE PRICE, 6 MAY (USD) FIGURE 21: NUMBER OF DEALS ON PROCTER & GAMBLE, ONE-MINUTE INTERVALS ON 6 MAY FIGURE 22: BID-ASK SPREAD WEIGHTED BY VOLUME ON PROCTER & GAMBLE, 6 MAY 17

18 The sale of a large amount of US index futures blindly using a trading algorithm dedicated to following market volume may have been at the origin of the Flash Crash HFTs that were counterparts of the trade on futures hedged their positions on the equity market; HFT that were counterparts of the trade on equity hedged their positions on the futures market The measures taken since 6 May In a joint report ("Findings Regarding the Market Events of May 6, 2010") the SEC and the CFTC suggested several reasons that could explain these rapid movements on futures and stock markets on 6 May The sale of a large amount of US index futures (E-mini contracts) initiated by a fundamental trader at 2:32pm may have been the event that triggered the Flash Crash. The trader was trading using an automated execution algorithm aimed at tracking the market volume; he did not put a limit price on the algorithmic order. At the moment of the sale, volatility was already high and liquidity was low on the futures market. The counterparties of the trade, which have been identified as HFT players, then moved rapidly to the equity market to hedge themselves in this context of high market risk. The excess of volume thus brought to the equity market then, for the same reasons, brought players in these markets to hedge their positions on the futures market. This game of hot potato, which was transmitted step by step until it returned to its starting point, caused the escalation of volumes exchanged while the liquidity offered remained unchanged. This artificially increased volume caused the trader to increase his flow sent to the market and thus fed the vicious circle that was starting. The jump in volume incurred by the market was followed by a large number of HFT players liquidating their positions, further increasing the imbalance between the demand and supply of liquidity. The e-mini and SPY, which is the ETF tracking the performances of the S&P 500, were both declining slightly ahead of the stock market. In fact the e-mini was leading the general fall. The fact that the e-mini and SPY were showing worrisome desynchronisation, plus the high volatility on both derivatives, caused an interruption in the number of orders. This is the mechanism that transmitted the shock to the stock market. While an increasing number of orders were being paused, the aggressive pressure was rising. The stock market quickly followed the derivatives market. Furthermore, as most of the brokers paused their orders, the retail flow that is usually executed OTC by internal matching engines was routed to the market, thus increasing pressure on the aggressive side of the order books. Another mechanism may have played a role that might have worsened the situation. This is the Liquidity Replenishment Point (LRP). LRP is a NYSE-specific system that slackens the market for a period of time in order to let liquidity return. It is possible that such a mechanism could have exacerbated the imbalance between the supply and demand of liquidity. When an LRP is triggered all aggressive orders are rerouted to other exchanges. Such a mechanism has the immediate effect of removing from the market the passive liquidity present inside the NYSE while retaining the aggressive liquidity. In addition to all the reasons highlighted above, the SEC also mentioned the fact that the absence of a stock-specific circuit breaker mechanism may have been one of the aggravating factors in the Flash Crash. On 10 June the SEC decided to remedy this problem by introducing a mechanism that places a five-minute hold on stocks with a movement of more than 10% during the previous five minutes. Following the Flash Crash, the SEC had a large amount of data to process. In May the SEC renewed its proposal to create a consolidated audit trail. The audit trails are currently collected at the level of each SOR (NYSE, Nasdaq, FINRA, etc.). Unfortunately, there are as many formats as there are SORs, which makes it fairly difficult for the regulator to analyse the data. Consequently the SEC wants these audit trails to be standardised and collected in real time (milliseconds) at a single data centre for all equities on the market. Such a measure might be difficult to implement and prove rather expensive. 18

19 The SEC recently approved a new proposal aimed at preventing market-makers from quoting too far from the actual price. This practice makes it possible for market-makers, bound by an obligation to quote, not to execute. It turns out that at the time of the crash on 6 May, a large number of trades had been made against these stub quotes. The last decision made by the SEC following the events on 6 May was approval of the ban on naked access to the market. All market participants will now be asked to fulfil a minimum level of risk control before being granted access to the market. Every layer of the market contributed to the crash This event has shown that the concerns raised by the SEC in January 2010, when it issued its Concept Release "Seeking Comment on the Structure of Equity Markets" were well-founded. The components of this Flash Crash are enlightening: an investor selling without taking into account the liquidity risk, a trader blindly using an algorithm, highfrequency market makers hedging their positions among themselves (with one leg on the futures market and the other on the equity market), the pause in internalising retail orders increasing the phenomenon, and trading destinations doing nothing to preserve the integrity of the price formation process. Every layer of the market contributed to the crash. The intricate ecosystem of the market microstructure now comprises a subtle collection of agents: high frequency market making is needed to support competition between trading pools, this competition pushes platforms to improve their robustness, trading algorithms are needed to execute large orders on a fragmented market, liquidity costs act as a reward for investors who spend time and money in gathering information on the liquidity of their stocks. The legislators and regulators need to monitor all of them to avoid one category of agent having insufficient reward compared to the service he provides to the collective. For the US and European price formation processes, adding friction to the market design is a solution Of course, in such a context, the regulator's response cannot be to try to lock the price formation process into a smaller box, lowering the limits of the circuit breakers, tightening the constraints of market-makers: this would only end up in an over-constrained dynamic that will freeze up too frequently. Regulators should change the dynamic of this ecosystem by modifying some of the forces at work. The easiest way to change the dynamics of an ecosystem is to adjust its speed, and for the US and European price formation processes, adding friction to the market design is a possible solution. In this issue of Navigating Liquidity, we will show how the tick size controls the frequency of updates of the order book, and can consequently be used as an adjustment variable to control the velocity of the price formation process. This can be done on a stock-by-stock basis, and updated regularly, with respect to the monitoring of the market activity. Mini flash crashes In June 2010, the SEC launched a test phase for the stock specific circuit breaker. Since then, a series of flash crashes, which triggered these circuit breakers, have been reported. FIGURE 23 to FIGURE 26 provides four example of these: Washington Post on 16 June (FIGURE 23), Citigroup on 29 June (FIGURE 24), Nucor Corp on 14 September (FIGURE 25) and Progress Energy Inc. on 27 September (FIGURE 26) FIGURE 23 to FIGURE 26 show the price changes (first sub-chart) and the volumes executed (second sub-chart) inside a one-minute sliding window for each stock during the day on which the mini-crashes occurred. The price values at the moment of the crashes are not displayed because they were too far from the usual values of the stocks. The case of Washington Post is particularly interesting, as it is an inverted crash. Erroneous trades on NYSE Arca were said to have caused the jump. Washington Post was thus the first stock to experience a circuit breaker. 19

20 Progress Energy's share lost 90% of its value in a matter of milliseconds If we take a look at FIGURE 26, which represents the last occurrence of a flash crash (on Progress Energy Inc. on 27 September), we see that the decline is very steep and rapid. In less than one-tenth of a second the stock was already down by almost 90%. NYSE reported that it correctly triggered its circuit breaker. This was not the case of NASDAQ, which had to void the trades that occurred during the mini-flash crash. These events raise the question of the efficiency of the circuit breakers. If just one trading venue is not able to trigger its circuit breaker on a stock in trouble (as happened for Progress Energy) the risk of a mini-flash crash occurring increases. This possibility rises to the same extent as the speed of the crash. The point is that the crash on Progress Energy was fairly fast. FIGURE 23: MINI-FLASH CRASH ON WASHINGTON POST ON 16 JUNE (PRICE AND VOLUMES) FIGURE 24: MINI-FLASH CRASH ON CITIGROUP ON 29 JUNE (PRICE AND VOLUMES) :36 10:48 12:00 13:12 14:24 15: :36 10:48 12:00 13:12 14:24 15: x :36 10:48 12:00 13:12 14:24 15: :36 10:48 12:00 13:12 14:24 15:36 FIGURE 25: MINI-FLASH CRASH ON NUCOR CORP ON FIGURE 26: MINI-FLASH CRASH ON PROGRESS ENERGY ON 14 SEPTEMBER (PRICE AND VOLUMES) 27 SEPTEMBER (PRICE AND VOLUMES) :36 10:48 12:00 13:12 14:24 15: :36 10:48 12:00 13:12 14:24 15: x :36 10:48 12:00 13:12 14:24 15: x :36 10:48 12:00 13:12 14:24 15:

21 European primary markets, claiming that they primarily host the price formation process, continue to suffer from outages Outages in Europe: bad news for the price formation process At 3.42pm CET on 13 October 2010, the NYSE Euronext cash markets (consisting of the Paris, Amsterdam, Brussels and Lisbon exchanges) were halted due to human error. No more trades occurred on the venue until 4.20pm, when NYSE Euronext reported that quotation resumed, but without any market data being distributed. For 40 minutes, Euronext offered trading conditions that were very similar to those of a dark pool (no pre-trade visibility). In the end, members were able to request cancellation of their trades ex post. As a preliminary remark, concerns arise about the fact that European primary markets, claiming that they primarily host the price formation process, continue to suffer from outages and similar "bugs" or technical issues. Euronext is not an isolated case. The LSE announced on 4 November its decision to postpone, for several months, the upgrade of its trading engine following an outage of around 2 hours on Turquoise. The fragmentation can also be viewed as a solution to avoid relying on them providing a microstructure for equity trading in Europe. Even if Chi-X and BATS have been putting competitive pressure on historical exchanges for years now, the result is not as good as could have been expected. These kinds of events (i.e. markets not functioning as expected) can be placed alongside the 6 May event in the US. Nevertheless, these events are rare market phases during which it is possible to compare the price formation process without a "subjective reference price". Bear in mind that without a trade-through rule, most MTFs decided to import the price of primary exchanges rather than that of a "consolidated European Best Bid and Offer" (BBO). During such periods, mid-points stop trading and all orders previously "pegged" to the primary BBO are cancelled or paused (depending on the trading rules of each platform). FIGURE 27: BEHAVIOUR OF A COMPONENT OF THE FIGURE 28: BEHAVIOUR OF A COMPONENT OF THE CAC 40 EUROSTOXX 50 DURING THE OUTAGE DURING THE OUTAGE MTFs Euronext VIVENDI MTFs Euronext PUBLICIS GROUPE SA :29:40 17:21:23 17:13 :07 17:04 :50 16:56:34 16:48 :17 16:40 :01 16:31 :44 16:23:28 16:15 :11 16:06 :55 15:58:38 15:50: :29:06 17:19:42 17:10:19 17:00:55 16:51:31 16:42:08 16:32:44 16:23:20 16:13:57 16:04:33 15:55:09 15:45:46 15:36:22 0 Source: CA Cheuvreux Quantitative Research 21

22 As CA Cheuvreux records data for post-trade performance analysis in an order-by-order database, we are able to look at the behaviour of most European stocks during the above-mentioned outages. For a period of 20 minutes, two price formation processes took place in parallel The price formation process is not owned by a specific type of trading platform; it is built on an information-sharing scheme that is currently wanting in Europe The amount traded during the outage on most stocks was close to that usually traded off Euronext: some mid caps stopped trading, but not all of them, and most large caps continued trading on MTFs. For most of the latter, the trading price was close to the last traded price on Euronext. When the VWAP from the opening to the outage was close to the last traded price on Euronext, prices barely moved. Nevertheless, we observed deviations of around 2% on some stocks during the outage. As shown in FIGURE 27 and FIGURE 28, even stocks on the EuroStoxx 50 and the CAC 40 were subject to such price changes. Moreover, the post publication of market data by Euronext shows the differences between the price on MTFs and Euronext. For such stocks, it was as if two price formation processes took place in parallel. We can never know exactly the number of trades but we can nevertheless comment on what happened in real-time. On Vivendi, for instance (FIGURE 27), we can clearly see trading on BATS and Chi-X at higher prices than on Euronext, even when "anonymous trades" occurred on Euronext. One could ask why Euronext took the decision to trade in such conditions since it claimed at the public hearing for the MiFID review by the European Commission (20 September 2010) that "small trades" had to occur on visible pools. Moreover, this disjunction of prices proves that the primary markets do not inherently own the price formation process. Without any consolidated tape, and as MTFs decided to peg most of their trading activity to primary markets, they decided to index exploration of prices on their pools to the primary BBOs. The fact that the price on the primary markets leads that of MTFs stems from the fact that most agents (brokers, investors and MTFs themselves) decided that most of their decisions should be tied to the state of the primary markets. We hope that if a consolidated post-trade tape is asked for by regulators, it will suffice for information-sharing among pools and traders so that the kind of instability of the price formation process, as seen on Euronext on 13 October, will not happen again. To conclude with regard to this outage, we can say that the price formation process is not specifically in the hands of one trading destination or another, but that it comes out of the mix of information about traded prices and bids and offers. A consolidated post-trade tape in Europe, where deals under a given size would have to be reported almost immediately, including dark orders, would be the easiest and most robust means of guaranteeing the emergence of a European price formation process, rather than adding liquidity around prices displayed by historical markets. 22

23 III Tracking market fragmentation Fragmentation of main indices: monthly figures indicate shocks A useful way to track fragmentation is to study CA Cheuvreux's Fragmentation Index (CFI), presented in Navigating Liquidity 4. This index, based on the physical definition of entropy, summarises how concentrated volume is between all the venues available. For example, if four trading venues are available, a CFI score of 100% indicates trading volumes breaking down as follows: 25/25/25/25. Fragmentation on the European market can be tracked by looking at six main indices: AEX, BEL 20, CAC 40, DAX, FTSE 100 and SLI (30). Note that the Eurostoxx 50 is not included in the present study because it is intrinsically fragmented across countries. However, it can be used as a benchmark; its actual level of fragmentation is plotted in green in the chart below (the CFI is calculated here on segments of 20 days, every 5 days). As shown in FIGURE 29, fragmentation in Europe, as measured by the CFI, increased sharply until end 2009; since then, it has increased at a slower pace. The SLI (Swiss Leader Index) appears late in the fragmentation race, which is explained by the fact that MiFID does not apply to Switzerland. FIGURE 29: CHEUVREUX FRAGMENTATION INDEX FOR MAIN EUROPEAN INDICES 23

24 April and May saw disruptions on almost all main indices Witching days almost systematically reduce fragmentation More volatility, higher spreads and less % of time spent at the EBBO are factors for lower fragmentation What is interesting here are the simultaneous local shocks in fragmentation curves. Since the beginning of 2010, four such shocks have occurred. FTSE 100 faced a local sharp increase of its fragmentation, peaking at the end of April. Inversely, DAX, CAC 40 and SLI saw an abrupt decrease in fragmentation at the beginning of May. The increase in fragmentation on the FTSE 100 is confirmed by a significant decline in the LSE's market share for these stocks from March to April 2010 (from 59.7% to 56.2% of trading volumes). Chi-X clearly captured the share lost by the LSE that month. Indeed, Chi-X showed better prices, with a better percentage of time spent at the EBBO (% time at EBBO). This advantage lasted until June 2010, when the LSE recovered almost all its market share, with a better % time at EBBO. This phenomenon is emphasised by the artificially high market share for the LSE in March due to the triple witching day on 19 March. Many traders thus went to the LSE in order to hedge their equity options positions. This prompts the conclusion that this small upward shock is only a one-off deflection from the underlying trend. For more detailed charts on the impact of the options expiry, see FIGURE 34 to FIGURE 37. DAX lost fragmentation in April, mainly due to Xetra's recovery in market share. The main market attracted trading from Chi-X (which lost 2%) thanks to its new fee schedule in March. It came with a significant increase in the average daily number of trades in comparison with MTFs. Xetra was also the only venue to see its % time at EBBO improve from March to May. Conversely, the % time at EBBO for MTFs increased in June, as did their market shares, which explains the recovery in fragmentation. As for fragmentation losses for the CAC 40 and SLI, the same phenomenon is seen from March to April for the former, and from April to May for the latter: the main market and Chi-X captured market share from other MTFs. Volumes were thus more concentrated on two venues instead of four usually, leading to a decrease in the fragmentation score. One phenomenon to be highlighted: the declines in fragmentation for the DAX, CAC 40 and SLI came hand in hand with a decrease in the % time at EBBO for all venues, high volatility and high bid-ask spreads. Once the % time at EBBO rose again, fragmentation was back up. This was not the case for the FTSE 100, but only due to its unusual behaviour, as explained above. Note that in a context where fragmentation is complete (CFI equal to 100%), the % time at EBBO should be equal to 100 for all trading destinations, and the % time at EBBO with greatest size equal to zero. In fact, if the market is perpetually perfectly fragmented, this would mean that there is no advantage of being on one specific market rather than another: every market thus proposes, at all times, the best price available and spends 100% of the day at the EBBO. Be it a recovery of one venue in particular (e.g. Chi-X for the DAX or the LSE for the FTSE) or a shift of volumes to two venues, fragmentation movements cannot be attributed to a single parameter. This shows that countries react differently to a change in market design. The microstructure is still in an adjustment phase. We hope that MiFID 2 will aid the microstructure in transitioning to a more stable and homogeneous phase. 24

25 Small & mid-caps: catching up main indices Below, we concentrate on the turnover (in euros) of European stocks. Looking at the average daily turnover in September 2010, we can define four groups of stocks among the usual indices: Blue chips (stocks whose daily turnover is above EUR100m); Large caps (stocks with turnover between EUR20m and EUR100m); Mid-caps (stocks with turnover between EUR5m and EUR20m); Small caps (stocks with turnover between EUR1m and EUR5m). A table presented in the appendix gives the number of stocks belonging to each category on each exchange. Moreover, it gives the number of stocks belonging to each category for the local main indices (for example, DAX components are blue chips or large caps only). First, we split European segments according to the correlation between the market share of the primary market and the turnover of their components. Figure 30: MARKET SHARE ON PRIMARY MARKET WITH RESPECT TO TURNOVER FOR LSE, EURONEXT AND XETRA STOCKS, SEPT EURONEXT 90% XETRA Pimary Market Market share 80% 70% 60% LSE 50% 1 M euro 5 M euro 20 M euro 100 M euro Figure 31: EURONEXT, NASDAQ OMX (H&S) AND SIX Figure 32: EURONEXT, OSLO, COPENHAGEN AND LSE ITALY 90% 90% Pimary Market Market share 80% 70% EURONEXT Pimary Market Market share 80% 70% EURONEXT LSE ITALY NASDAQ OMX (H&S) 60% OSLO SIX COPENHAGEN 1 M euro 5 M euro 20 M euro 100 M euro 1 M euro 5 M euro 20 M euro 100 M euro 25

26 The large rectangular blocks encompass 50% of the stocks of each segment. Dots and other markers indicate outliers. In short, the majority of the stocks on each segment are located in its associated shape. In this visual representation, the closer the shapes are, the more similar the components of the segment are in terms of the correlation between the market share of the primary market and the daily turnover. Figure 30 shows that LSE-listed stocks have the same pattern as those listed on Euronext and Xetra, but their fragmentation is shifted downwards. This means that for similar daily turnover, LSE-listed stocks are more fragmented than those listed on Euronext or Xetra. As shown in Figure 31, stocks on the Euronext, Nasdaq OMX (H&S) and SIX segments have very similar behaviour. The stocks on these three marketplaces will be studied together. The third group of segments (in Figure 32) emphasis the specificities of the Oslo, Copenhagen and Milan segments: their fragmentation is very heterogeneous. This makes them difficult to use in our study. Note that Spain, Portugal and Austria are not included either because of the special case they represent in terms of fragmentation. To conclude, this issue of Navigating Liquidity will describe the fragmentation on mid and small caps on the LSE, Euronext, Xetra, Nasdaq OMX (H&S) and SIX segments. The LSE segment will be isolated in most charts due to its high level of fragmentation. FIGURE 33 shows the market share of each category (blue chip, large, mid, small, and UK stocks by category) and their trend since November FIGURE 33: HISTORICAL TURNOVER PROPORTION FOR EACH CATEGORY OF STOCKS OVER ALL MARKETS English Blue-chip English Large English Mid English Small Blue-Chip Large Mid Small 29/10/ /08/ /07/ /05/ /03/ /01/ /11/ /09/ /07/ /05/ /03/ /01/ /11/ FIGURE 33 shows that the proportion of turnover of our sample of stocks represented by blue chips, large, mid and small caps has remained relatively stable, despite a slight increasing trend for non-uk large cap stocks, and a slowly decreasing trend for non-uk blue chip stocks. FIGURE 34 to FIGURE 37 track the market share of the main European trading platforms for each category of stock. 26

27 FIGURE 34: MARKET SHARE OF MAIN MARKETS FIGURE 35: CHI X'S MARKET SHARE /10/10 31/08/10 04/07/10 07/05/10 09/03/10 10/01/10 13/11/09 15/09/09 19/07/09 22/05/09 24/03/09 25/01/09 28/11/08 English Blue-chip English Large English Mid English Small Blue-Chip Large Mid Small English Blue-chip English Large English Mid English Small Blue-Chip Large Mid Small 07/05/10 09/03/10 10/01/10 13/11/09 15/09/09 19/07/09 22/05/09 24/03/09 25/01/09 28/11/08 29/10/10 31/08/10 04/07/10 FIGURE 36: BATS' MARKET SHARE FIGURE 37: TURQUOISE'S MARKET SHARE English Blue-chip English Large English Mid English Small Blue-Chip Large Mid Small 29/10/10 31/08/10 04/07/10 07/05/10 09/03/10 10/01/10 13/11/09 15/09/09 19/07/09 22/05/09 24/03/09 25/01/09 28/11/ English Blue-chip English Large English Mid English Small Blue-Chip Large Mid Small 31/08/10 04/07/10 07/05/10 09/03/10 10/01/10 13/11/09 15/09/09 19/07/09 22/05/09 24/03/09 25/01/09 28/11/08 29/10/10 As expected, primary markets have a lower market share on large caps and blue chips As expected, primary markets have a lower market share on large caps and blue chips than for small and mid caps (FIGURE 34). These recent changes also enable us to conclude that almost all London-listed stock types, even small caps, are more fragmented than large caps on other markets. The fragmentation definitely affects London-issued stocks, regardless of their size. Turquoise's market The x-axis on these charts are the expiry dates for derivatives. We can see how they share on large, mid impact most segments significantly. This means that some investors focused on the and small caps is derivative space do not have access to MTFs. It is partly because MTFs do not implement greater than on more the dedicated fixing auctions that exist in primary markets. liquid stocks On FIGURE 37, we see that outside London, there is a reversal around March 2010: Turquoise's market share on large, mid and small caps is greater than on more liquid stocks. This is not true for LSE-listed stocks. Another specific behaviour for Turquoise is the slump in its market share just prior to April Turquoise had market-making agreements among its founding members, which aimed at supplying liquidity. However, these agreements expired in mid-march 2009 and this led to the sharp drop in activity, as shown in FIGURE

28 On most markets, the fragmentation of small and mid-caps currently stands at the level of blue chips and large caps in August 2009 On most markets, the fragmentation of small and mid caps currently stands at the level of blue chips and large caps in August According to the figures above, BATS and Turquoise can be seen as a potential source of the increase in fragmentation on small and mid caps. FIGURE 38: CHEUVREUX FRAGMENTATION INDEX FOR THE CAC 40 AND THE SBF 80 Small and mid-caps are joining larger ones in terms of fragmentation Until June 2009, small and mid-cap stocks composing the SBF 80 were clearly late in terms of fragmentation compared to large and blue chip stocks on the CAC 40 ( FIGURE 38). However, since that date, CAC 40 stock fragmentation has slowed, unlike for SBF 80 stocks. The gap between the two indices has steadily narrowed over the past year. Only the last two months show different behaviour, with the fragmentation of large stocks increasing and that of small ones decreasing. Only time will tell if this is a benign shock or a long-term trend. In any case, small and mid caps are becoming more fragmented, hence the use of a SOR to trade these stocks appears increasingly appropriate. 28

29 IV Tick size and ultra high-frequency market design For a given security, in a given range of prices, the tick size is the smallest increment between two limit prices. A tick size table gives a description of the tick size regime; it defines price ranges (with a lower and an upper threshold) as well as the tick to be used for each of these ranges of price. By way of example, we present one of the four FESE tables used by European markets and MTFs below. These tables establish a concordance to use the same tick sizes for the various markets trading a given stock. TABLE 1: FESE TABLE 1 USED FOR HIGHLY LIQUID ITALIAN AND UK STOCKS Lower bound for Upper bound for Tick size for this Minimum tick size Maximum tick size this range this range range of prices relative to the relative to the price price bp 100% bp 5bp bp 2bp bp 5bp bp 2bp bp 5bp bp 2bp 1,000 4, bp 5bp 5,000 9, bp 2bp 10, % 5bp As these increments are determined according to the range of prices considered, different tick sizes on the same trading day can be observed for a stock. It is even possible to observe an order book for which some prices are on a finer grid than others. Trade prices will therefore also have a minimum increment. There is nevertheless an exception to this rule: midpoints. As the traded price on a midpoint is the average of the best ask price and the best bid price on a reference market, the tick size of a midpoint is two times smaller than that of the reference market. Tick size is therefore the degree of precision to which we agree to work for the trading of a given security. Stated this way, it would appear that the smallest possible tick size would be best, but there are in fact many drawbacks in the use of a small tick size: It generates a huge amount of meaningless data, making the regulators surveillance task more difficult and quote stuffing strategies more powerful. It focuses trading on tactics rather than strategy, which can make the posted liquidity thinner and more evanescent, and can therefore hurt market depth. It makes some front-running strategies easier or requires more sophisticated execution strategies that would have to spray their liquidity more widely to avoid being spotted. 29

30 The only benefits of a tick size decrease are: A potential decrease in the spread, which, thanks to a virtuous circle of cheaper liquidity, would lead to an increase in turnover traded. It is important to note that when it is either not profitable enough or too risky to post liquidity then no one does so. The best example of this is the Flash Crash. Reducing the discretization effect of traded prices and spread. This is a relevant remark as long as the precision provided by tick size is not sufficient, as there is indeed no gain in having smaller increments when price differences become insignificant. In such a situation, the discretization effect does not hide any information and there are thus no concerns about increasing this theoretical precision. The rules agreed by FESE members have to be enforced What happens when tick size changes? In order to illustrate the effects of a change in tick size, we use an example. We will consider Xstrata, a very liquid stock, traded for more than GBP100m per day. This stock belongs to the highly liquid UK segment. This is a group of very liquid stocks created in July 2009 by the FESE members in order to choose a specific tick size table for this group. The cause for this specificity was a move by Turquoise and BATS in June These venues saw a need for a smaller tick size for some UK and Italian stocks (which at that time were traded with a relative tick size of 10 basis points). This move was viewed as a tick size war, as it gave the three main MTFs a significant market share gain for the short period of time when they had a different tick size from the primary market. Indeed, it enabled trading on MTFs with a smaller spread than the 10 basis point minimum imposed at the time by the rules of the LSE Group. After this move, the FESE members found a consensus, realising that tick size should not be used as a way to increase the market share of one venue to the detriment of overall liquidity. They therefore agreed on a set of tick size tables to be used for all European trading. The rules are quite simple: for a given stock, you should choose one of the tables, and this table should be used by any trading venue that enables trading on this stock. Four tables were created, one of which was dedicated to the trading of "highly liquid stocks", using a tick size as small as one basis point (relative to the price) for some price ranges, whereas these stocks were previously using a 10 basis point relative tick size until the aforementioned move by the MTFs. This is an important step in the history of tick size regulation, as markets themselves agreed that they should not leave this choice up to market forces, but had instead to find a rational way to determine the tick size for the sake of liquidity and not the interest of a select few. Nevertheless, it is time for the regulator to take on its role in this process set in place by the financial industry. The rules agreed to by FESE members now have to be enforced: there is no need for a dozen tick size tables in Europe, and there is no need to allow a market to use its own tick size table if the standard ones are well designed. As explained above, trading on Xstrata currently uses the FESE table 1. Therefore, the tick size for Xstrata for prices below 1,000 pence is 0.1, or 0.5 for prices higher than 1,000p but below 5,000p. This makes for a significant tick size change when prices reach this 1,000p threshold, as the tick size relative to the price is 1 basis point just under 1,000p and 5 basis points just over 1,000p. This enables us to observe the changes occurring in the price formation process when the tick size changes, but also calls into question the correctness of the design of this table, as it allows for jumps in tick size relative to price as high as 5x. For illustration purposes, we will use trading data for Xstrata on 31 August 2010, as it offers an example of such a shift in tick size. 30

31 The 5x jump in relative tick size allowed by current tick size tables are useful for experiments and studies, but are not desirable In FIGURE 39, we have plotted the traded prices for Xstrata on 31 August. The discretization effect of prices over 1,000p is clear. Apart from this trade data, we have limit order book updates for the five best limits on each side of the book. This enabled us to add a dotted line to indicate the number of limits (among these 10 limits) that used the bigger tick size of 0.5. This computation is made with snapshots of this limit order book every second; for each of these snapshots, we compute the number of limits that are strictly greater than 1,000p, which therefore ranges from 0 to 10. In FIGURE 39, we have represented these values with a different scale from prices so that a value of ten for this indicator is placed on the very top of the figure and obviously a value of 0 on the bottom of the figure. FIGURE 39: TRADED PRICES FOR XSTRATA ON 31 AUGUST The figures we give in the rest of this section on the subject of tick size are based on the data shown here between 8:10am and 2:20pm UK time. 14:20 13:49 13:18 12:47 12:16 11:45 11:15 10:44 10:13 09:42 09:11 08:40 08:10 We have removed the first ten minutes of continuous trading, as they are known to have special behaviour in terms of spread and market depth: this is the period when the order book is filling up with orders, so the spread always shows a decreasing trend. The ten minutes chosen are entirely arbitrary, so one should be careful when trying to draw conclusions about the trend in the trading process at the beginning of the sample shown here. We have also excluded the end of the day because of the shift in trading activity that occurs every day at 2:30pm UK time (or 1:30pm a few weeks a year) due to the opening of US markets. The same remark as above about the possible heterogeneity of the beginning of the sample applies here to the end of the sample: the choice of this limit for the sample is arbitrary. The influence of the interaction with US markets is often felt even before 1:30pm UK time, as this is when macroeconomic statistics are released, and therefore an important time of the day for trading all around the world. In this way, we have endeavoured to create a sample as uniform as possible regarding the type and number of investors participating in the price formation process. 31

32 A tick size decrease allows a smaller spread if tick size is a constraint In FIGURE 40, we have represented the bid-ask spread for Xstrata for the period of time we have chosen. The solid black line represents the quoted spread: at the end of every second, we compute the spread and calculate the average over rolling one-minute intervals. We have added green crosses to show the median quoted spread over the last minute in order to make the constraint imposed by tick size more obvious in this chart. The scale of the figure is in basis points, except for the proportion of orders in the order book that are on the big tick size grid, which is in its own scale, as on the previous figure. In FIGURE 41, we also represented the quoted bid-ask spread, but with grouped boxplots by the proportion of limit orders lying on the coarse grid among our ten limit snapshots. Consequently, the boxplots on the very left of the chart represent the boxplot of the quoted spread when the order book is entirely on the small tick size grid, whereas the one on the right is the quoted spread when the order book is entirely on the big tick size grid. FIGURE 40: BID-ASKQUOTED SPREAD ON THE LONDON STOCK EXCHANGE FIGURE 41: BID-ASK QUOTED SPREAD BOXPLOTS ON THE LONDON STOCK EXCHANGE :47 12:16 11:46 11:15 10:44 10:13 09:43 09:12 08:41 08:11 14:19 13:49 13: % 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% The spread is lowered when the tick size decreases if and only if the tick size is a constraint As we can see in FIGURE 40, when competition for providing liquidity is intense, a small tick allows for a lower spread. Spreads can be smaller with a smaller tick size, lowering the liquidity premium paid to execute small orders. 32

33 A small tick size makes posted liquidity more unstable For purposes of our next illustration, we have chosen to use a synthetic order book of the following five trading venues: LSE, Chi-X, BATS, Turquoise and NYSE ARCA. We have therefore used the five separate order books, arranged the different prices in order, and summed the volume available at the limit existing on more than one destination. The size available plotted in FIGURE 42 is that of this synthetic first limit. It can therefore come from different destinations. FIGURE 43 is a representation of the same data, but instead of dedicating the x-axis to time, we have used the variable proportion of limits in the order book that are on a coarse tick size grid in order to group our observations and make boxplots of the volume available on the first limit. FIGURE 42: VOLUME AVAILABLE ON THE BEST LIMIT (USING ORDER BOOKS OF THE FIVE TRADING VENUES) x FIGURE 43: BOXPLOTS OF THE VOLUME AVAILABLE ON THE BEST LIMITS GROUPED BY THE PROPORTION OF LIMITS IN THE ORDER BOOK ON A COARSE TICK SIZE GRID x :19 13:49 13:18 12:47 12:16 11:46 11:15 10:44 10:13 09:43 09:12 08:41 08:11 0 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% The figures are explicit enough: the smaller the tick size, the less volume is found on a given limit. This is not a surprising result: the more limits available for posting liquidity, the less liquidity you will find on each of them. Nevertheless, one cannot help but recognise that a sound and deeper order book is desirable. Furthermore, this thin liquidity on each limit has another cost, illustrated in the following charts. FIGURE 44 represents the average update frequency (in Hz) of the order book in one of the first five limits. In FIGURE 45, we have also plotted this number of updates but in boxplots grouped by the proportion of orders in our snapshot that lie on the coarse tick size grid. It is worth noting that the frequencies displayed are computed on 1-minute intervals, so to reach 250 Hz for this measure (the maximum reached), it is necessary to observe 15,000 order book updates in one minute, but the peaks of frequency can obviously be far greater than this average. 33

34 FIGURE 44: AVERAGE FREQUENCY OF ORDER BOOK UPDATES (FIVE TRADING VENUES) FIGURE 45: BOXPLOTS OF MEAN ORDER BOOK UPDATES GROUPED BY THE PROPORTION OF LIMITS IN THE ORDER BOOK ON A COARSE TICK SIZE GRID :18 12:47 12:16 11:46 11:15 10:44 10:13 09:43 09:12 08:41 08:11 14:19 13:49 0 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% A small tick size makes liquidity more unstable It is obvious from these figures that with a smaller tick size, the posted liquidity is more evanescent, it jumps from limit to limit, even for very small price improvements with the aim of attracting the liquidity-consuming flow of orders without having to wait for time priority. We have done our best to show what we believe to be two entirely different ways for the price formation process to take place: one in which orders have to wait for execution, sometimes in a very long queue, and one in which the ever-moving liquidity jumps according to very small price increments, sometimes offering a better spread; but at the price of uncertainty as to the order being executed, as this liquidity might no longer be there when the order reaches the market. Thus, the questions to be answered are: which of these regimes make us feel more confident, and do we still want to have an order book that changes at an ever faster pace following incidents such as the Flash Crash? What is the next step for tick size? The choice of an optimal tick size table for a given security is a very tricky problem that has been studied, and there are therefore some stylised facts that have emerged from academic literature, which aid in designing a pragmatic way to choose the right tick size: There is no gain in lowering the tick size past the point that enables a decrease in the spread. There are some side effects, as shown not only in our example but also in academic studies that have demonstrated the potential negative effects of a tick size decrease. Therefore, any decrease should always be implemented reasonably and smoothly. Despite the trend towards decreasing tick sizes in Europe, there is still room for a further decline. However, in our view, in cases in which a small tick size is useless, it should be increased. Currently, there are also a lot of stocks that do not need their current small tick size, and a larger one would help markets be more efficient. 34

35 Harmonisation of tick size regimes must be finalised A tick discretized joint distribution of spread, price, interest of the market and volatility is the quantitative key for determining optimal tick size There is no need for tick size regimes specific to a market; all markets should use the same tables The necessary range of tick size tables should be used for stocks with specific features Nevertheless, the major initiative of FESE members should be followed up, and we think that in order for optimal tick sizes to be set in a regulated fashion, true standard tick size tables must emerge and be used by all markets. There is no reason for a market to have its own tick size regime: stocks all have the main characteristics that determine optimal tick size: price, level of interest from investors (often measured as turnover traded) and volatility. Other reasons mentioned by markets cannot be accepted for long; they can slow the process, in order to allow time for their members to adapt, but there is no reason for this to take years. Our advice for regulators would therefore be to mandate a set of tables that are the only ones allowed, and then at least at the beginning, allowing for each stock, that markets together choose the relevant table for its trades in. A tick discretized joint distribution of spread, price, interest of the market and volatility is the quantitative key for determining optimal tick size. Nevertheless, we believe that an easier computation, based on the ratio of the bid-ask spread and the tick size, could easily be enforced by the regulator. This dynamic rule would enable a small tick size for stocks that need it without penalising other stocks with an overly small tick size. However, the designed process would have to make it easier for a stock to increase its tick size than to decrease it because of the fastest-moving component of the optimal relative tick size: volatility increases at a faster pace than it declines. As a recent example, Borsa Italiana (a market in which there is still potential for a further decrease in tick size) has proposed a new tick size table to its members. In our view, the design of this tick size table contains a very good idea for an improvement in the FESE tables: trying to make small jumps in relative tick size when crossing a price range bound. Conversely, designing this table just for Borsa Italiana, with only the price ranges observed for Italian stocks, is a major mistake. Another bad example of a table is the FESE table 4, which is in fact the same as the FESE table 1 but restricted to the prices mostly used by the markets that choose it. This is an example of an unacceptable way to avoid the standardisation of tick sizes: there is no good reason not to use the same tables as other European markets unless these are not well designed, in which case they need to evolve. There might be a concern about the fact that when they are too small compared to the trading currency, the increments have no meaning as the difference cannot be paid in the currency; however, if this is a legitimate reason, then it should be a concern for all markets and rules should be enforced apart from tick size in order to adapt to the specificities of the currency used for quotation. Our proposal would therefore be to apply this idea to any of the tick size tables: avoid jumps in relative tick size significantly greater than 1x. In a purely abstract world, without the probable high cost it would create, we would even propose a quotation method based on this principle: prices would be ( 1 + a) n with n a positive or negative integer. With such a quotation process, the relative increment between two prices would always be a. Avoiding the complexity of current tick size tables, we would then just have to define a as the precision of the quotation, on a stock-by-stock basis. This would set a uniform relative tick size for all prices, which is a desirable property for a tick size table if we do not want to have some of the problems of large tick sizes for some stocks and the problems of small ones for others. 35

36 More pragmatically, we propose a set of tick size tables that can be found in the appendix These tables offer the following advantages: They cover the wide range of prices in Europe: from the Italian cents, to the thousands for UK or Nordic stocks; They make the jumps in relative tick size as smooth as possible; They allow for a choice of either a very small (1bp) tick size relative to price, or large ones (up to 100bp); hence this set could be used for any level of liquidity and volatility specific to a stock. The choice of a very small relative tick size compared to a large one (i.e. from table 1 to table 6) should be made according to the volatility so that when the volatility is high, the tick size has to be large. This will have a stabilising effect concentrating the liquidity on fewer price limits in volatile contexts. In terms of liquidity: the less liquid a stock, the larger tick size. This will also increase the liquidity available on each price level. Are strategies requiring an order-totrade ratio greater than 100 useful for the quality of liquidity? Another way to fill order books with vanishing liquidity On 20 April 2010, the LSE announced a change in its pricing, effective on 4 May: A maximum charge of 0.29bp for aggressive trading if the total value traded is high enough; Free-of-charge trading for market makers that have a passive/aggressive ratio greater than 75%; The threshold for the high usage surcharge for FTSE 350 securities would increase from an order-to-trade ratio of 100/1 to a ratio of 500/1. In our view, the first two changes do not need further explanation. For the third change, this is what it means: posting a limit order, modifying it, or cancelling it is usually free of charge. However, some participants make so much use of messages sent to the market that a limit needs to be enforced on the ratio of the number of orders divided by the number of trades. If a participant goes beyond this limit, then orders are no longer free of charge. Obviously, a thousand orders could be sent without any trade, followed by a small trade in order to lower this ratio. So participants that would find such a change useful have a long-term need to send a lot of orders that are not executed. We are not claiming that, as previously on the Turkish market, it should be forbidden to cancel an order, lower its size, or change its price to make it further from the best limit. Conversely, we would be very interested to know what the benefits are of being able to have such a ratio. What would a usual participant do with this? It is forbidden to send signals to the markets for the purpose of confusing others. Therefore, such a capability should only be used because a participant had in mind that a price was the right one for its order, but a short time later, this no longer appears to be the case. However, the participant would have to think very fast for this ratio to be useful perhaps too fast. Market makers need a higher ratio than others, as their earnings are made from posting orders and being able to move them when the market moves against them. Many MTFs have even chosen to have totally free-of-charge orders in order to be more attractive to this kind of player; and the order-to-trade ratio allowed on the LSE would seem very low compared to that of an MTF. We insist that the rapid vanishing of orders resulting from this feature is a source of concern. 36

37 In FIGURE 46, we have plotted the number of order book updates for the FTSE 100 index (in five first limits of the order book), divided by the number of trades. A vertical black line is positioned on 4 May 2010, when this new pricing came into effect. FIGURE 46: NUMBER OF ORDER BOOK UPDATES DIVIDED BY THE NUMBER OF TRADES ON THE LSE /08/ /18/ /27/ /06/ /16/ /26/ /05/ /14/ /24/ /03/ /12/ /22/ /02/2010 The rapid disappearance of liquidity made possible by this feature is a concern In order to verify that the reason for this surge in order book updates was really this pricing promotion, we looked at other European markets to find a similar pattern, and found none. This event looks like an inconclusive experiment by ultra high-frequency traders on the LSE, as the order-to-trade ratio has since returned to its previous levels, but we are left to wonder what halted this experiment: were the strategies implemented not profitable enough, or did the LSE fail to acknowledge orders fast enough? If the latter is the real cause, high-frequency traders are probably glad to know that the time to acknowledge orders is set to decrease soon, with an average time of just 126 microseconds. 37

38 From a regulatory standpoint Without a doubt, substantial changes in the market microstructure have occurred in the past three years in the US and Europe. The symptoms are not the same, but there are some shared roots: the price formation process was affected by fragmentation following regulatory changes, and by a lack of liquidity caused by the crisis. A new type of agent, namely high-frequency traders (HFT), acting as market makers but in most cases without obligations, blurred the usual roles of each layer of the market structure (see Figure 47). Liquidity is now provided by high-frequency traders acting as investors. Note that this role was previously held by intermediaries. Figure 47: DIAGRAM OF THE MICROSTRUCTURE Investors HFT OTC Intermediaries SOR Crossing engine Dark Pools Visible book Dark Pool Mid point Visible book Dark Pool Mid point Hub Primary MTF MTF MTF MTF The consequences of this change in the microstructure are different in the US and Europe, mainly because of local regulations. In the US, the Flash Crash on 6 May showed that a market organised around a pre-trade consolidated tape can also have its weak points. In Europe, outages have shown that without shared information among agents, it is very difficult to obtain a robust price formation process. Some facts seem to be undeniable: HFT is the price to pay for fragmentation; it is not possible to put trading venues in competition with one another without agents building high-frequency liquidity bridges across them. In parallel, note that in Navigating Liquidity 3 we showed that there is no proof of a link between the activity of high-frequency traders and a potential decrease in market volatility. 38

39 The tick size is the easiest parameter to adjust the dynamics of the price formation process in a modern fragmented market The main question is: how much will market agents agree to pay to support this kind of bridge? Once this threshold is fixed, plenty of ways can be used to adjust the level of HFT activity, one of these being the tick size. As we have shown in this issue of Navigating Liquidity, the larger the tick, the lower the update frequency of the order book. The impact of market design is not limited to intraday trading. Undoubtedly, the price formation process, and the way that available liquidity allows it to go in one direction or another, play a role in the recent increase in realised correlations and the decrease in realised volatilities. In the US, the existence of the consolidated tape organises how information is shared among agents; it allows them to make decisions and strengthens the price formation process. Europe needs a way to share information without relying on primary markets alone. A consolidated post-trade tape is a good option that could leverage on fragmentation to improve the robustness of the price formation process. 39

40 V Appendices Appendix 1: Glossary Price Formation Process The price formation process covers all events occurring during trading that result in a constantly evolving market price. Such events are, for instance: insertion of new orders or cancellation of orders, or matching of opposite orders. Walrasian equilibrium An equilibrium between consumers and producers to set a perfect match between supply and demand. Regulation The regulation organises the market design, to ensure the efficiency of the price formation process. Reg NMS Regulation National Market System is a regulation promulgated by the SEC in 2005 that seeks to encourage competition among individual markets and among individual orders. It contains the trade-through rule, the access rule (addressing access to market data), the sub-penny rule (establishing minimum pricing increments) and market data rules. Consolidated tape Trade-through rule MiFID Large in Scale (LIS) ESMA CESR SEC This is the electronic service that provides last sale and trade data for issues admitted on US exchanges, consolidating all markets and driving the tradethrough rule. The trade through rule mandates that when a stock is traded in more than one market, transactions may not occur in one market if a better price is offered on another market. There is a mandatory re-routing of the order to other markets. The Markets in Financial Instruments Directive, applied since November 2007, is part of the European Commission s drive to improve competition among European financial markets. This directive is set to be revised after a review of its consequences, leading to "MiFID 2" in the first quarter of An order is considered to be large in scale compared with the normal market size if it is equal to or larger than the minimum order size specified in Table 2 in Annex II of the MiFID Implementing Regulation. The European Securities and Markets Authority has been touted by the European Commission as a replacement for the CESR, with additional responsibilities. The ESMA will be responsible for safeguarding the integrity and stability of the financial system, the transparency of markets and financial products, and the protection of investors, as well as preventing regulatory arbitrage and guaranteeing a level playing field. The Committee of European Securities Regulators. Its role is to improve coordination among securities regulators, act as an advisory group to assist the European Commission, and work to ensure more consistent and timely day-to-day implementation of community legislation in the Member States. The US Securities and Exchange Commission s mission is to protect investors, maintain fair, orderly and efficient markets, and facilitate capital formation. 40

41 Trading destinations MiFID removes domestic exchange concentration rules and recognises three trading destinations: Regulated Markets (RMs), Multilateral Trading Facilities (MTFs) and Systematic Internalisers (SIs). Everything else is over-the-counter (OTC). Primary market Multilateral trading facility (MTF) Electronic communication network (ECN) Lit pool Dark pool (mid-points) Dark pools (integrated books) Smart order router Broker crossing network (BCN) Matching engine Co-hosting Also called new issue market. Securities are issued for the first time in this market. In this study, Euronext Paris, the London Stock Exchange and Xetra are examples of primary markets. A multilateral system operated by an investment firm or a market operator which brings together multiple third-party buying and selling interests in financial instruments in a way that results in a contract. Chi-X, Turquoise and BATS are the MTFs studied here. This is a type of computer system that facilitates trading of financial products outside of stock exchanges. ECNs are the equivalent of MTFs in the US. A trading destination with a visible order book. A trading destination that does not disclose its order book. Trades are often reported with a delay. This book accepts hidden orders smaller than the Large in Scale (LIS), and matches only at the mid-point of the Primary Best Bid and Offer. A trading destination that does not disclose its order book. Trades are often reported with a delay. These integrated books accept all visible orders and hidden orders larger than the Large In Scale (LIS). These venues are said to be adapted to trade large quantities without leaving a footprint in the market. A device routing an order across a given set of trading destinations according to a disclosed execution policy. It can split an order into smaller ones to spray all available destinations if needed. This is an alternative trading system that matches buy and sell orders electronically for execution without first routing the order to an exchange or other displayed market. The order is either anonymously placed in a black box or flagged to other participants of the crossing network. Its advantage is to enable large block order execution without impacting the public quote. This is the software device holding all pending trades for every listed stock on the trading destination and matching orders to compute possible transactions. Once the match is made, information about the completed transaction flows out of the matching engine. Serving as a join host for several trading institutions' computers in order to reduce latency. 41

42 Execution costs and market depth measurements Execution costs are a mixture of fees, bid-ask spread, price impact, market impact, opportunity risk and market risk. Market risk Price impact Market impact Opportunity risk Tick Size Measurement of the uncertainty in the trend of the price. Impact of the volume of an aggressive order on the obtained average price. Possibly persistent impact of the volume of an aggressive order on the market price. Price of missing a transaction or obtaining a deteriorated price by not having placed an order on the adequate trading destination. The minimal difference allowed between two different prices. It is defined by the trading rules of each trading destination. VWAS: Volume-weighted average spread The mean of the bid-ask spread at each trade, weighted by the volume of the trade. Market Depth Limit Order Book (LOB) Best Bid and Offer (BBO) Market share Average daily number of trades Average trade size (ATS) This is the size of an order needed to move the market a given amount. If the market is deep, a large order is needed to change the price. Market Depth closely relates to the notion of liquidity. This is a record of unexecuted limit orders maintained by trading destinations. The highest bid or lowest offer price available in a market at a specific time. For local main indices, this is the ratio of the venue's turnover to the sum of the turnover of all four venues considered (main market, Chi-X, BATS and Turquoise). The higher it is, the more chance an investor has to be present when a trade occurs. Contrary to block venues, the granularity of a venue is a must to attract different order flows. Mean trading size in euros. This can be considered the "natural size" of orders on the venue. % Time at EBBO This is the proportion of the day during which the venue offers a spread equal to the European Best Bid and Offer (EBBO). % Time at EBBO with Greatest Size This is the proportion of the day during which the venue offers the greatest sizes at a spread equal to the EBBO. Adverse selection In general, adverse selection is used to describe an insurance phenomenon in which people that want to have health insurance are more likely to have health problems, and so are typically the kind of people you do not want to insure because of the risk. In this context, this term refers to a market process in which buyers and sellers have asymmetric information. Adverse selection can occur in dark pools, for example. If your order is completely filled, this implies that the counterparty had more liquidity than you. It can be assumed that the other side, being even larger, will be likely to cause market impact and thus push the price against you. The fact that your order was filled is an indicator that you actually did not want it to be filled (it would have been better to wait until the price had been pushed and then to cross). 42

43 Appendix 2: Stock distribution among destinations according to their mean turnover Looking at the average daily turnover in September 2010, four groups of stocks among the usual indices are defined: Blue chips (stocks whose daily turnover is above EUR100m); Large caps (stocks with turnover between EUR20m and EUR100m); Mid-caps (stocks with turnover between EUR5m and EUR20m); Small caps (stocks with turnover between EUR1m and EUR5m). The table below specifies, for each trading venue, the number of stocks belonging to each of these categories. Under each place, stocks belonging to some indices are marked with a cross. This means that, for example, stocks belonging to AEX quoted on Euronext Amsterdam are situated in "Blue Chip", "Large" and "Mid" category, but not in the "Small" category. TABLE OF STOCK DISTRIBUTION AMONG DESTINATIONS AND ACCORDING TO THE MEAN TURNOVER OF THE STOCK Blue chip (> EUR100m) Large (> EUR20m) Mid (> EUR5m) Small (> EUR1m) Euronext Brussels DJ STOXX 50 X DJ STOXX Large and Mid X X X BEL 20 X X X X Euronext Amsterdam DJ STOXX 50 X AEX X X X DJ STOXX Large, Mid, Small X X X Euronext Paris DJ STOXX 50 X CAC 40 X X X DJ STOXX Large X X X DJ STOXX Large, Mid, Small X X X X Xetra DJ STOXX 50 X DAX X X DJ STOXX Large X X DJ STOXX Large, Mid, Small X X X Nasdaq OMX Helsinki DJ STOXX 50 X DJ STOXX Large X X DJ STOXX Large and Mid X X X OMX HELSINKI 25 X X X X DJ STOXX Large, Mid, Small X X X X Nasdaq OMX Stockholm DJ STOXX 50 X DJ STOXX Large X X OMX STOCKHOLM 30 X X X DJ STOXX Large, Mid, Small X X X X LSE DJ STOXX 50 X X DJ STOXX Large and Mid X X X FTSE 100 X X X X DJ STOXX Large, Mid, Small X X X X SIX DJ STOXX 50 X SWISS LEADER PR INDEX X X X SMI X X X DJ STOXX Large X X X DJ STOXX Large, Mid, Small X X X 43

44 Appendix 3: Tables of tick sizes TABLE 2: OUR TICK SIZE TABLE PROPOSAL Bounds for the price of the order entered Lower bound Strict upper bound Tick size Table 1 Table 2 Table 3 Min relative tick size bp Max relative tick size bp Tick size Min relative tick size bp Max relative tick size bp Tick size Min relative tick size bp Max relative tick size bp E E E E E E E E E E E E , ,000 2, ,000 5, ,000 10, ,000 20, ,000 50, , , ,

45 TABLE 2: OUR TICK SIZE TABLE PROPOSAL (CONT.) Bounds for the price of the order entered Table 4 Table 5 Table 6 Lower bound Strict upper bound Tick size Min relative tick size bp Max relative tick size bp Tick size Min relative tick size bp Max relative tick size bp Tick size Min relative tick size bp Max relative tick size bp E E E , ,000 2, ,000 5, ,000 10, ,000 20, ,000 50, , , ,

46 Other publications Market Indicators provides liquidity benchmarks that identify the most appropriate trading venues for best execution. It analyses the performance of Chi-X, Turquoise, BATS and the primary markets for the following indices: AEX, BEL20, CAC40, DAX, FTSE 100 SMI and EUROSTOXX50. The venues are assessed using a number of different criteria including intraday market share, average daily number of trades, average trading size, volume weighted average spread, and auction liquidity. Execution Services is a comprehensive guide to our wide ranging execution offer: Sales Trading, Direct Market Access, Algorithmic Trading, Global Portfolio Trading, Contracts for Difference. Algorithmic Trading provides an in-depth look at our 13 customisable algorithmic strategies, to meet any investment parameters. Global Portfolio Trading presents an overview of this CA Cheuvreux and CLSA service, which offers clients quality execution with two of the world s leading agency brokers in 50 countries worldwide. Commission Sharing Agreements outlines the key benefits and logistics of signing a Commission Sharing Agreement with CA Cheuvreux. Note: This document is also available in French. Markets Trading Guide helps navigate the liquidity maze by detailing the trading conditions of the execution venues available through CA Cheuvreux. The Pocket Guide to Corporate Liquidity informs listed companies of the impact MiFID may have on them by providing simple and precise answers to 15 key questions on MiFID and liquidity fragmentation. 46