Navigating Liquidity 6
A global menu for optimal trading
January 2012
Global Research Navigating Liquidity 6
Q Due to electronic competition, global exchanges and regulators face the same issues: they have a limited set of solutions to design their markets Q Competition pressure is pushing for mergers around the world, without decreasing the number of trading pools. Deutsche Börse with NYSE Euronext, the Tokyo and Osaka stock exchanges, and BATS and Chi-X Europe belong to a long list of examples. Q The ongoing uniformisation of trading hours in Asia, following the way paved by the US and Europe a few years ago, also aims at increasing the competitiveness of exchanges. Q Technology is to converge within combined operators, whose breathing space would be in tick sizes and auction type choices. Q Inventory-driven clients are attracted by fixing auctions, whose specific patterns are disclosed in this publication. There are as many tick size tables as countries, and these need to be adjusted to each market microstructure: European ones are too small, while the US set-up could be improved by using more than a single tick size.
Q
Investors' timing and information flow at different scales drive execution costs
Each investor chooses his own timing based on the information he wishes to value. The interconnectedness of his timing with that of other market participants, investors or high-frequency traders is the key factor that drives execution costs (market impact and slippage included). Q Market data analysis on a small scale shows that high-frequency traders reduce the bid-ask spread. However, this reduction does not imply that trading costs shrink correlatively. To be paid, high-frequency traders must build their timing in such a way as to force other market participants to hit liquidity aggressively or – worse – to be adversely selected. The consequences are an increase in execution costs for other investors. Q
Q The demand for algorithmic trading is mainly oriented towards liquidity-seeking optimisation in the US, execution efficiency in a cross-market comparison in Asia, and routing and implementation shortfall in Europe Q Each benchmark provides access to a main trading feature. Nevertheless, traders’ real needs are still their timing and their demand for liquidity. Q CA Cheuvreux's offer is based on a suite of algorithms built on the "trading envelopes" model. The dynamics of the algorithms are therefore highly customisable and can be adapted to any kind of market design or market condition, while emphasising the liquidity-seeking dimension.
About the authors Charles-Albert Lehalle
Global Head of Quant Research
Romain Burgot
Statistician, Quantitative Group, Paris
Matthieu Lasnier
Statistician, Quantitative Group, New York
Stéphanie Pelin
Engineer, Quantitative Group, Paris
CA Cheuvreux 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: +33 1 41 89 80 88 New York: +1 212 492 8850
London: +44 207 621 52 00
Ian Peacock Global Head of Execution Services +44 207 621 5144 / ipeacock@cheuvreux.com SELL SIDE Jonathan Carp Head of Alternative Execution Sales Europe +44 207 621 5244 / jcarp@cheuvreux.com
2
www.cheuvreux.com
BUY SIDE Mark Freeman Head of Alternative Execution Sales to Buy Side +44 207 621 5285 / mfreeman@cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
CONTENTS
3
I— A global menu for market design: the choices of exchanges and regulators
4
Q
Towards similar auction mechanisms
4
Q
Around the world in four tick regimes: one size does not fit all
7
Q
Worldwide consolidation
8
Q
Inventory-driven investors need fixing auctions
16
II— Timing is money: investors need to trade accordingly
23
Q
Market design and information flow timing imply liquidity patterns
23
Q
Market impact depends on investment style
29
Q
High-frequency traders do not impact all investors equally
36
III— Algorithmic trading: adapting trading style to investors' needs
43
Q
Each trading feature has its own benchmark
43
Q
Customisation offers multi-feature trading styles
45
Q
Liquidity seeking: fine-tuning between price and speed
47
Appendices
51
Q
Appendix 1: Glossary
51
Q
Appendix 2: Other publications
55
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
I— A global menu for market design: the choices of exchanges and regulators Q
Towards similar auction mechanisms
Working versions of European regulations and directives (MiFID, MiFIR, EMIR, MAD) were issued in November, Chi-X began to offer trading in Australia this October, BATS is going forward to offer listing in the US: it is more than time to have a global analysis of the market micro-structure.
On paper, a microstructure built around competing electronic venues provides traceability of the transactions leveraging on the usual benefits of competition
Regulators and policymakers are comfortable in demanding to record and store information on the behaviour of market participants; this mood favours a market design organised around competing electronic markets. On paper this type of market design provides traceability of the transactions leveraging on the usual benefits of competition (price pressure and run for quality). Again on paper, the two other archetypal models: a highly concentrated model (typically the French one five years ago) or an intricate and high latency network of bilateral counterparts (think about the UK markets a few years ago), would probably be respectively too expensive or too "dark". In reality, transparent information on the price formation process (not only reporting transactions as soon as they occur, but also the full depth of the order-books at pre trade) and the appetite of competing trading venues for liquidity providers opened the door to liquidity arbitrageurs, mainly known as "High Frequency Traders" (HFT). First because "liquidity bridges" have to be established between available trading venues to ensure that a bid price somewhere is not greater than an ask one elsewhere (or an available ask being lower than an available bid). The arbitrageur will take half of the differences between the two prices and "improve" the level of information of other participants having access to fewer venues. The less such cross-trading venue arbitrages exist, the more blindly a market participant can send an order to any venue: he somehow delegates the information search to arbitrageurs (HFTs), and agrees to pay for this "service". CrĂŠdit Agricole Cheuvreux built a "Fragmentation Efficiency Index" (FEI) a few months ago, available on our website, to monitor this level of "neutrality" of a set of trading venues to liquidity searches, inspired from the concept of entropy used in physics to measure the level of heterogeneity of an environment. In short: the higher the index (maximum is 100%), the more blindly an order can be sent to a venue, while the lower its value (minimum is 0%), the more carefully an order has to be split. FIGURE 1 shows the recent trend in the FEI in Europe.
In reality, the appetite of competing trading venues for liquidity providers opened the door to HFT
4
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
FIGURE 1: RECENT TREND IN CA CHEUVREUX'S FRAGMENTATION EFFICIENCY INDEX (FEI) ON EUROPEAN MARKETS
100 DAX CAC40 FTSE 100
90 80 70 60 50 40 30 20 10 0 20
20 20 20 20 20 20 20 20 20 20 20 20 11 11 11 10 10 10 09 09 09 08 08 08 07 /11 /07 /03 /11 /07 /03 /11 /07 /03 /11 /07 /03 /11 /25 /24 /23 /20 /20 /19 /16 /15 /14 /11 /11 /10 /08 Source: Crédit Agricole Cheuvreux Quantitative Research
It can be seen that the liquidity offer on the components of the three main European indices shows a two-slope trend: the first one (fast) during 2008, the other one (slower) from the beginning of 2009. Just as a reminder, even if it is very difficult to monitor the activity of HFTs in the order book, they are said to be a party to around 70% of the transactions in the US, more than 40% in Europe, and 35% in Japan. The activity of one large player has been extensively studied by Prof. Albert Menkveld in his academic paper "High-Frequency Trading and the New Market Makers". Using data from the exchanges (Chi-X and NYSEEuronext) on Dutch stocks, he computed (these are not estimates but real recordings) that this large HF player drove Chi-X market share on these stocks in 2007 and 2008, being part of around 80% of Chi-X’s activity (FIGURE 2).
During the first two years of Chi-X on the Dutch market, one large HF player contributed to 80% of Chi-X’s activity
5
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
FIGURE 2: MONITORING OF THE TRADES OF ONE LARGE HF PLAYER ON CHI-X
Source: "High-Frequency Trading and The New-Market Makers", by Prof Albert Menkveld, SSRN Aug. 2011, p39
Arbitrageurs are not working for the sake of market efficiency: they are earning money. Injecting technology into the usual market-making practices (providing liquidity at the bid and ask on many venues, maintaining an inventory exposed to market risk – i.e. intra-day volatility – and adjusting the quotes to maintain the risk between given bounds), they succeeded in lowering their risk to a level that is largely covered by the bid-ask spread. Part three of section two (p36) of this issue of Navigating Liquidity will uncover part of this mechanism. It can be added here that comparing intra-day scale to daily scale, the price formation process exhibits more auto-correlation features at the former. Probably because no one can mount a large position in one second, but can do so in one day. Being able to identify the start of a large trade at intra-day allows for time to take profit on it, posting better-adapted quotes. This could be one of the sources of the qualitative advantage of high-frequency vs. low-frequency market making. We must bear in mind that the higher the frequency, the less resiliency is provided to order books by market makers. As underlined by the authors of the SEC-CFTC report on the Flash Crash: high-frequency market makers provided liquidity to the market during the event, trading a lot, without any positive impact on the liquidity shortage: no market maker was there to buy low and sell high, maintaining an inventory over a ten-minute time interval.
Injecting technology into the usual marketmaking practices, HFT succeed in lowering their exposure to market risk
Now that the technology is available and the trading techniques have been used for several years on the US and European markets, there is no doubt that attempts to use them will continue on Asian markets, as long as regulations allow it. Moreover, Europe has seen that competition between trading venues has implied consolidation (London Stock Exchange + Turquoise, BATS Europe + Chi-X, NYSEEuronext + Deutsche Börse) to leverage on lower technology costs, but maintaining as many different order books as possible. Since tick size has long been considered a key element of the market design, we continue to believe that it should be chosen carefully and kept in the hands of regulators. The next section (p7) is a global analysis of the actual situations in terms of tick size. The sequencing of different auction mechanisms (fixing, continuous, trading at last, etc.) is also an important component of the market design, influencing the way market participants provide and consume liquidity. While the fourth part of this section (p16) is dedicated to an in-depth study of fixing auction mechanisms, one has to remark here that Asian markets are gradually converging to common trading hours when possible (for instance, Tokyo and Singapore in 2011). Competition to capture flows of investors
Competition between Asian markets can drive convergence in terms of auction types and sequencing
6
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
wishing to invest on Asian-driven market factors can be enough to drive convergence in terms of auction types and sequencing.
Q
Around the world in four tick regimes: one size does not fit all
Different approaches to tick size Tick size is the minimal increment between the prices of two limit price orders. Depending on the market, it can be applicable to any stock whatever its price (as in the US) or it can be dependent on the price range and the quotation group that the stocks belong to (as is the case for UK stocks). The latter case is often referred to as a tick size regime and theoretically groups stocks according to their liquidity.
This lower bound for the spread should not be set as low as possible…
The effects of different tick sizes have been illustrated in previous issues of Navigating Liquidity, and include: Q
The tick size as a minimal bound for the spread;
An optimal tick size enhances market depth, enabling liquidity to concentrate on fewer, but larger, quotes. Q
In FIGURE 3 to FIGURE 6 we used green crosses to plot the volume-weighted average spread (over a day for one stock) of the 20 most traded stocks (in number of trades) on four different markets, against the daily VWAP of this stock. As the spread computed here is already relative to the price (bid-ask spread divided by mid-price) there is no reason for a relation to appear, except when tick size becomes an active lower bound for the spread (which is not necessarily an undesired feature when it improves market depth). The dark line on these figures represents the tick size which is either a line (with the log scale we used here) in the case of a single tick size as in FIGURE 3 and FIGURE 4, or a piecewise affine function of the price in the case of a tick size regime as in FIGURE 5 and FIGURE 6.
…as it helps concentrate liquidity and creates the market depth
FIGURE 3: US STOCK MARKET SPREADS AND TICK SIZE
FIGURE 4: INDIA'S STOCK MARKET SPREADS AND TICK SIZE
FIGURE 5: JAPANESE STOCK MARKET SPREADS AND TICK SIZE
FIGURE 6: GERMAN STOCK MARKET SPREADS AND TICK SIZE
Source: Crédit Agricole Cheuvreux Quantitative Research
7
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
As explained in previous issues of Navigating Liquidity, we strongly believe that there is an optimal tick size, depending on the liquidity of a stock (the less liquid, the higher the tick size), its price (the higher the price, the higher the tick size), and its current volatility (the more volatile, the higher the tick size). Despite the apparent complexity of these rules, they would enable markets to have the same behaviour amongst stocks and reduce the effects of different tick sizes.
Tick size regimes are preferable
Using the same tick size across all stocks, as in the US, puts a very strong constraint on stocks priced below USD10 and leaves stocks priced above USD200 with nearly empty order books that have less meaningful quotes (at least at the first limit shown on the consolidated tape). This is a kind of "one size fits none" policy. In such a market, a company aware of the importance of tick size on trading would have to use splits or reverse splits to reach a 'good' tick size. Unfortunately, there are often other considerations that would make the decision more complicated. In the case of the US market, trading fees are based on the number of shares exchanged, therefore the fee will be relatively less important for a higher priced stock, but so will the rebate for providing liquidity and probably also the appeal to retail investors. In such conditions, tick size will probably not even be considered when taking such a decision. Among markets using a tick size regime, there is no consensus on the optimal level of tick size relative to price, and Europe appears to be an outlier with its small tick sizes relative to price. This is a consequence of the tick size war that took place in 2009, when some MTFs engaged in a race to decrease tick size in order to gain market share. Up to now, only a gentlemen's agreement sets a common tick size for all the markets trading the same stock in Europe, but it also has the effect of preventing markets making any changes in tick size to search of optimality. As already stated, our hope is that the regulator decides to enforce rules on the subject and does not leave it to market forces.
Europe is at odds with the rest of the world on this subject, by choosing a tiny tick size
Q
Worldwide consolidation Q
Electronic trading global timeline
Markets have faced significant changes in the last forty years, both in terms of regulation and technology. Key steps of these evolutions are indicated in the following timeline. A colour code mentions the type of changes that met financial markets: an increase in fragmentation with new entrants (orange), a recent worldwide consolidation phase (blue) and related technological topics (pink). Vertical figures indicate the corresponding speed of processors over the years, and their strength (MIPS: million instructions per second) which increases according to Moore's law.
16-40 MHz 5 MIPS
2 MHz 0.64 MIPS
Note that the more recent the years, the more events to occur, and the more markets to merge.
8
ƒ
1970s:
US / Introduction of the NYSE's "designated order turnaround" system (DOT) that routed orders electronically to the floor.
ƒ
1980s:
Automation of index arbitrage and development of program trading.
ƒ
1986:
France / Introduction of the CAC trading system (fullycomputerised matching engine).
www.cheuvreux.com
Global Research Navigating Liquidity 6
3 GHz 2*22 000 MIPS
2.4 GHz 22 000 MIPS
2.6-3.6 GHz1.3-3.8 GHz 9000 MIPS 1700 MIPS
233 MHz 100 MIPS
January 2012
1990s:
US / Introduction of the Electronic Communication Networks. Cheuvreux' first DMA and algorithmic orders.
1992:
Initiation of the FIX protocol: international real-time exchange of information.
2000s:
US / Decimalisation of prices.
June 2001:
Creation of the Committee of European Securities Regulators (CESR), which will be replaced by ESMA under MiFID II.
April 2004:
Adoption of the MiFID directive at the EU level.
2005:
US / Rules promoting national market system are consolidated into RegNMS.
June 2005:
US / BATS Trading is formed.
February 2006: Archipelago becomes NYSE Arca after the NYSE's buyout.
June 2006:
NYSE (New York Stock Exchange) and Euronext merge to become NYSE Euronext.
July 2006:
Publication of implementing measures for MiFID.
July 2007:
Launch of Chi-X Amsterdam and Chi-X London.
August 2007:
One participant makes 80% of Chi-X Amsterdam's trades*.
September 2007: The same participant reaches 15% of all Dutch trades and drives Chi-X's increase in market share in Amsterdam.
November 2007: Deadline for the industry to apply the directive. Launch of Chi-X Paris.
August 2008:
Launch of Turquoise in Europe.
October 2008:
US / Knight Link is the No. 1 dark pool in share volume.
November 2008: Launch of BATS Trading Europe and Xetra Mid-Point.
March 2009:
Market-making agreements on Turquoise expire. Its market share drops.
May 2009:
Launch of Nasdaq OMX Europe.
June 2009:
Tick war: BATS and Turquoise activate a gradual reduction in tick sizes, leading to a tacit agreement between exchanges under the FESE.
July 2009:
A former Goldman Sachs programmer is arrested by the FBI with a USB key containing algorithmic trading codes.
August 2009:
US / Flash orders appear on the Nasdaq Stock Exchange.
October 2009:
Launch of NYSE ARCA Europe.
November 2009: Launch of Knight Link, first Systematic Internaliser in Europe.
Technological topics
2.93 GHz 76 383 MIPS
Consolidation or slowing unstructured fragmentation
3.2 GHz 2*24 000 MIPS
Fragmentation begins
9
www.cheuvreux.com
Global Research Navigating Liquidity 6
3.33 GHz 147 600 MIPS
January 2012
January 2010:
Turquoise offers futures and options on Norwegian stocks and an index on Turquoise derivatives. US / The SEC sends a request for comments on market microstructure.
April 2010:
The CESR also sends questions to participants on market microstructure.
May 2010:
May 6 : Flash Crash.
July 2010:
Closure of Nasdaq OMX Europe.
September 2010: Findings on the May 6th events are reported by the CFTC and the SEC.
December 2010: The European Commission launches a public consultation for the revision of MiFID (MiFID II). BATS Global and Chi-X Europe enter exclusive merger talks.
February 2011: End of the EC consultation. Regulatory change in Spain for Chi-X to get clearing and settlement. Feb. 18th: London Stock Exchange and Turquoise merge. Deutsche Börse and NYSE Euronext confirm advanced merger discussions. nd Feb. 22 : Outage on the Italian Stock Exchange. th Feb. 25 : Outage on the London Stock Exchange.
May 2011:
Markus Ferber appointed as rapporteur for MiFID II. His conclusions are to be presented to the ECON (European Parliament's Economic and Monetary Affairs Committee).
June 2011:
June 13 and 15 : Outages on Chi-X. th st th June 20 , 21 and 27 : Outages on NYSE Euronext.
August 2011:
India / first regulatory approval for Smart Order Routing.
September 2011: Turquoise offers futures and options on the FTSE 100 index.
October 2011:
November 2011: Final decision on the BATS-Chi-X merger from the UK Competition Commission. BATS Global and Chi-X Europe successfully closed their deal.
April 2012:
ESMA guidelines issued for MiFID II. Both BATS and Chi-X Europe to launch fully interoperable clearing.
May 2012:
Amendment deadline for MiFID II.
July 2012:
ECON committee vote on MiFID II.
End of 2012:
Trialogue process to reach agreement on MiFID II.
January 2014:
Earliest indicated implementation of MiFID II.
Fragmentation begins Consolidation or slowing unstructured fragmentation Technological topics
th
th
th
Proposal for MiFID II is unveiled by the European Commission. It contains one directive, and one regulation.
*Source: [High-Frequency Trading and The New Market Makers – Menkveld, 2011]
10
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
Q
European liquidity fragmentation: current situation
The Markets in Financial Instruments Directive (MiFID), which came into force on 1 November 2007, ushered in competition between traditional stock exchanges and alternative trading venues, and set up a legal framework to govern this competition. This step ranks among the milestone of worldwide fragmentation, as seen in the timeline above.
Chi-X, Turquoise and BATS Europe are the three main rivals for the incumbent exchanges: Deutsche Bรถrse, London Stock Exchange and NYSE Euronext
The European equity market is currently dominated by three securities and derivatives exchanges: Deutsche Bรถrse, the London Stock Exchange Group and NYSE Euronext. These three groups operate the most active regulated markets and also their own MTFs. The regulated market activities are rivalled by a handful of pan-European MTFs that started operations in 2007 and 2008: Chi-X, owned by brokerage house Instinet; Turquoise, launched by an investment banking consortium, and subsequently acquired by the LSE; and BATS Europe, a subsidiary of US stock exchange BATS. In order to visualise the degree of activity of these various trading platforms in Europe, let us consider data on European Local Main Indices (LMIs: AEX, BEL 20, CAC 40, DAX 30, FTSE 100 and SLI 30) from 1 January 2011 to 14 August 2011. European market shares are calculated on this dataset. Venues deemed too small have been excluded from our study: this is the case for NYSE ARCA and EQUIDUCT, for example.
Market shares on European LMIs indicate that Chi-X and NYSE Euronext predominate
Aggregating turnover on our dataset, we present in FIGURE 7 the market share in July 2011 of the seven venues available for trading on stocks from European LMIs. NYSE Euronext and Chi-X are the two major destinations, together representing almost half of turnover on European LMIs. They are followed by Deutsche Bรถrse and the London Stock Exchange. Please note that these figures are biased by the number of stocks covered by each venue (for example, Chi-X allows trading on all stocks while Deutsche Bรถrse is only available for DAX 30 stocks in our sample). FIGURE 8 is a view of the market shares on a weekly basis since May 2009. Note that NYSE Euronext and the London Stock Exchange have seen their market share decrease constantly since that date, unlike Chi-X, which drastically increased its share during the first month, and then remained stable until finally overtaking NYSE Euronext in the last few weeks. Contrary to other primary markets, Deutsche Bรถrse and VIRTX Stock Exchange remained relatively stable during the two years of our dataset. Finally, BATS's market share increased steadily during the first few years, until recently, when Turquoise overtook it. However, the upward trend for Turquoise and downward trend for BATS are relatively recent and need to be confirmed in the next few months.
11
www.cheuvreux.com
Global Research Navigating Liquidity 6
January 2012
FIGURE 7: MARKET SHARES ON EUROPEAN LMIs (OCTOBER 2011)
FIGURE 8: MARKET SHARE TRENDS ON EUROPEAN LMIs SINCE MAY 2009
5.8% BATS
60
7.4% TURQUOISE
22% EURONEXT
EURONEXT DEUTSCHE BORSE LONDON SE VIRTX SE CHI-X TURQUOISE BATS LSE-Turquoise merger
50
40
26% CHI-X
30
20
16% DEUTSCHE BORSE
10
8% VIRTX SE
15% LONDON SE
0 01
/20 1
1
02 02 06 03 07 05 04 07 11 08 10 09 /20 /20 /20 /20 /20 /20 /20 /20 /20 /20 /20 /20 11 11 11 11 11 11 11 11 11 11 11 11
Source: Crédit Agricole Cheuvreux Quantitative Research
Q
From division to union
We have seen that MiFID enabled the creation and development of multiple new trading platforms, bringing competition to markets. After this huge and rapid evolution, things appear to have stabilised, and a new trend towards venue consolidation has emerged. The merged entities seek to survive the fierce competition and make gains in technology, and the future of markets seems to lie in fewer but bigger competitors.
A consolidation trend is obvious among the exchanges. The London Stock Exchange and Turquoise were the first in line BATS/Chi-X, as well as Deutsche Börse/NYSE Euronext are the two main mergers pending in Europe, each seeking regulatory approval
Europe already saw the London Stock Exchange merge with Turquoise in February 2011. The same month, the London Stock Exchange migrated to the Millennium IT Technology platform for the main UK order book. Turquoise had migrated to this system five months previously. This new technology brought a positive combination of a reduction in latency to 126 microseconds for order acknowledgment and a potential threefold increase in capacity of volume handled. However, the move came at a cost with Turquoise suffering an outage on Monday 4 October 2010. With their merger, LSE and Turquoise were only the first movers in a wave of potential consolidation. Two other major mergers in European markets are currently in the spotlight. The first would link up BATS and Chi-X Europe. Negotiations went exclusive in December 2010, and the UK's Office of Fair Trading referred BATS and Chi-X Europe's proposed merger to the Competition Commission in June 2011. The Commission gave its approval and BATS acquired Chi-X in November 2011. Note that they may face the same problems as the LSE and Turquoise in terms of technology migration; shortly after the two venues declared that BATS technology would be chosen for both platforms, Chi-X suffered a wave of outages following more than four years without any major technical glitches. Both exchanges have experienced technical issues in December 2011 and January 2012. The second merger concerns Deutsche Börse and NYSE Euronext. Advanced talks were confirmed in February 2011. Nasdaq OMX Group and Intercontinental Exchange (ICE) made an alternative offer to NYSE Euronext in April 2011, which was rejected. Deutsche Börse shareholders approved the proposed combination in mid-July 2011. However, as for Chi-X Europe and BATS, regulatory authorities in both the United States and Europe still have to rule on the proposal. As we will see in the next section (p14), competition levels are at stake: the two firms could reach almost 40% market share on European LMIs under the most favourable scenario. This is even greater on derivative products, where the new group would have a virtual monopoly. The European 12
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
Commission gave both exchanges a questionnaire in October for them to justify the legitimacy of their merger. They revised their merger filing at the end of 2011 in order to gain approval. Among other things, they promised they would cap trading fees and offer open access to clearing. They also committed to divesting their single equity derivatives business. The US Department of Justice already granted them a positive response, on the condition that Deutsche Börse disposes of its stakes in US company Direct Edge. As well as the aforementioned two European examples, the merger wave has also affected exchanges around the world. We give a few examples below. The LSE and Canada's TMX were to merge at the end of June 2011. However this plan fell through when the required two-thirds majority of shareholders was not reached. A second offer for TMX from Maple Group, a consortium of domestic banks, had been refused twice, but finally received the backing of the TMX board at the end of October. The various offers for international groups show how important size is in today's worldwide fragmentation landscape. In an attempt to compete with the more developed Brazilian Stock Exchange, South American markets (Chile, Colombia and Peru) launched a single access point for investors on 30 May 2011. This represents the first example of cooperation for these markets. In response to the increasing presence of alternative venues in the Japanese market (Chi-X increased its market share from 0.8% in January 2011 to 1.7% in June 2011), the Tokyo Stock Exchange (91.2% in June) and the Osaka Securities Exchange (5.3% in June) reached agreement on a merger in November 2011. An offer from the Singapore Exchange to buy the Australian Stock Exchange was aborted after it was rejected by the Australian Treasurer in April 2011. Some interpreted this refusal as a risk and damaging for the country's appeal to foreign investors. Exchanges are not the only ones to dream of becoming bigger, investment firms and clearing houses are also part of trend. Two US proprietary trading firms have announced their merger, namely Madison Tyler and Virtu Financial, confirming the increasing challenge of technological investment in order to make profits and stay in competition. In addition, US group Getco acquired the British proprietary firm Automat in July 2011. After several months of rumours, in early September 2011 the London Stock Exchange confirmed preliminary talks with LCH.Clearnet.
Growing appears vital for exchanges in the stabilised fragmented and competitive market, but also for financial firms and clearing and settlement institutions
The trend towards getting bigger to become stronger thus extends to primary markets, alternative venues, investment firms and clearing houses, underlying the mainly technological challenges that markets participants face.
Q
Possible distribution of participants in tomorrow's markets
What impact could the merger trend have on European markets? Let us have another look at current market shares among European LMIs. FIGURE 9 describes the distribution of turnover in October 2011. FIGURE 9: TABLE OF MARKET SHARES ON EUROPEAN LMIs (OCTOBER 2011)
Market 1. CHI-X 2. EURONEXT 3. DEUTSCHE BÖRSE 4. LONDON STOCK EXCHANGE 5. VIRTX STOCK EXCHANGE 6. TURQUOISE 7. BATS
Market share on European LMIs 25.6% 22.3% 15.8% 15.2% 8.0% 7.4% 5.8% Source: Crédit Agricole Cheuvreux Quantitative Research
13
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
A more equitable repartition of European activity if the two mergers go through
Assuming all mergers are completed, three scenarios have to be taken into account. The first one, called the "additive scenario", imagines that both venues would keep their clients and that activity would remain unchanged. The combined market share would be the sum of both market shares. This is more likely to be the case for NYSE Euronext and Deutsche Börse given that they allow trading on different stocks. The second scenario imagines that the two venues share the same clients, who would thus consider the combined group as one venue, and that the turnover of the smallest venue would be lost. This is called the "loss scenario". Obviously the latter scenario is not credible for BATS and Chi-X, since both venues would, according to the official press, keep separate order books, but it gives a lower benchmark for the combined market share. We also consider a third scenario, which summarises Cheuvreux's point of view on the forecast market shares. In fact, as we mentioned in our October Market Liquidity Indicators, Chi-X-BATS may be forced to close BATS Europe's order book given Chi-X's long-term success in terms of gaining market share (unlike BATS, which is losing market share). Its turnover would then only be that of Chi-X (low case). Conversely, NYSE Euronext – Deutsche Börse are likely to keep their former respective clients, and their turnover would be the sum of both turnovers (high case). These three scenarios are presented in FIGURE 10.
FIGURE 10: SCENARIOS OF MARKET SHARES ON EUROPEAN LMIs FOR POTENTIALLY COMPLETED MERGERS (BASED ON OCTOBER FIGURES)
Market 1. EURONEXT – DEUTSCHE BÖRSE 2. CHI-X – BATS 3. LONDON SE – TURQUOISE 4. VIRTX STOCK EXCHANGE
Additive scenario 38.0% 31.4% 22.6% 8.0%
Loss scenario 28.4% 32.7% 28.8% 10.1%
Cheuvreux' s point of view 40.4% (high) 27.2% (low) 23.9% 8.4%
Source: Crédit Agricole Cheuvreux Quantitative Research
In order to have an idea as to which scenario is most likely to happen, we took a look at the impact of the LSE/Turquoise merger in February 2011 (FIGURE 11). No specific change was noticeable during the first few weeks after the merger. Only the past two months have shown a particular change, with the London Stock Exchange quickly decreasing and Turquoise increasing correspondingly. FIGURE 11: SUM OF LSE AND TURQUOISE MARKET SHARES SINCE MAY 2009 40 GROUP LONDON SE TURQUOISE MERGER
35 30 25 20 15 10 5 0 01
11 10 07 06 07 09 03 05 08 02 04 02 /20 /20 /20 /20 /20 /20 /20 /20 /20 /20 /20 /20 /20 11 11 11 11 11 11 11 11 11 11 11 11 11
Source: Crédit Agricole Cheuvreux Quantitative Research
14
www.cheuvreux.com
Global Research Navigating Liquidity 6
January 2012
Bear in mind that market shares do not mean absolute activity, but relative activity with regard to other platforms. Turnover is the parameter to look at when considering real activity: a loss in market share does not always represent a loss in turnover, but sometimes a smaller increase than for others. Q
Overall activity does not represent fragmentation
In previous sections, we have looked at total activity on European LMIs, in order to estimate the importance of each venue in the European market. However, these figures do not attest to the degree of fragmentation available for investors in the markets. In fact, fragmentation should relate to the venues available to an investor on a specific stock. Aggregating all indices is a problem, as the LSE is not a possible trading platform when trading France Telecom for example. In order to measure fragmentation, we should then look at European LMIs separately. However, indices can be aggregated when they are traded on the same venues. This is the case for the AEX, the BEL20, and the CAC40 for example. FIGURE 12: TABLE OF MARKET SHARES ON EUROPEAN LMIs (OCTOBER 2011)
Zone Corresponding indices NYSE EURONEXT AEX, BEL20, CAC40 DEUTSCHE BÖRSE DAX30 LONDON STOCK EXCHANGE FTSE 100 AVERAGE
Market shares MTFs
Primary market
Chi-X
Turquoise
BATS
CA Cheuvreux Fragmentation Efficiency Index
63.9%
23.8%
7.1%
5.2%
70.0%
64.9%
24.2%
6.7%
4.2%
67.7%
52.1%
31.0%
8.6%
8.2%
80.8%
60.3%
26.3%
7.5%
5.9%
73.4%
Source: Crédit Agricole Cheuvreux Quantitative Research
It thus makes sense to calculate the CA Cheuvreux Fragmentation Efficiency Index (FEI, formerly CFI), which we introduced in Navigating Liquidity 4. We present this in the last column of FIGURE 12 above and FIGURE 13 below for each group of indices. Average market shares give an idea of the average fragmentation available when trading on indices' stocks in Europe.
Our FEI represents the level of real fragmentation in the markets, relative to the hypothetical perfect fragmentation (with regard to available destinations)
As a reminder, the CA Cheuvreux Fragmentation Efficiency Index uses the physical formula of entropy (i.e. disorder in a system) to measure the level of fragmentation that a stock or an index faces. According to the formula defined by Kolmogorov and Shannon, if the market share of the th n venue is Mn, then the entropy is:
Entropy = −∑n M n × log(M n ) Entropy is often renormalised by a constant; we therefore propose to use the maximum entropy possible for a market with N different venues (i.e. log(N)). This gives us the following formula for normalised entropy:
H=
−1 ∑ M n × log(M n ) log( N ) n
The FEI ranges from zero to one, and can be followed historically and compared among stocks and indices. TABLE 13 presents the market shares of each trading venue, aggregating indices that can be aggregated, with the hypothesis that the two mergers will 15
www.cheuvreux.com
Global Research Navigating Liquidity 6
January 2012
succeed and that each group will benefit from the summed turnover of both merging venues. FIGURE 13: TABLE OF MARKET SHARES ON EUROPEAN LMIs (OCTOBER 2011)
Zone Corresponding indices
Primary market
DEUTSCHE BĂ–RSE - NYSE EURONEXT AEX, BEL20, CAC40, DAX30 LONDON STOCK EXCHANGE FTSE 100
Market shares MTFs Chi-X BATS
Turquoise
CA Cheuvreux Fragmentation Efficiency Index
64.3%
28.8%
7.0%
75.4%
52.1%
39.3%
8.6%
83.6%
Source: CrĂŠdit Agricole Cheuvreux Quantitative Research
The FEI appears to be higher thanks to the two mergers if we average (79% vs. 71%). Its interpretation in terms of fragmentation is subject to discussion. In fact, due to the renormalisation by the number of available venues, the FEI almost mechanically increases when eliminating a venue (the denominator is lower). Furthermore, it would increase to 100% if there was only one trading destination.
The FEI is higher when considering merged exchanges, indicating more efficient trading on more venues
One could argue that a decreasing number of venues implies that fragmentation is weaker. This increase in the FEI actually means that the distribution of turnover among available venues would be more equitable in the case of mergers. In economic terms, this implies a lower number of competitors, but stronger competition among them, as the competitors have more weight. Moreover, competitors with higher turnover should enable more efficient trading. The FEI would then represent the level of "efficient" fragmentation for the investor. In fact, let us imagine that a hundred venues are available for trading a stock, but only one of them has the majority of trading. The quality of liquidity on the other ninety-nine venues would be called into question. The FEI indicates how many destinations are available for trading, subject to their quality. It thus measures the ability to trade on several efficient venues, without having to choose between them. A high FEI indicates a good breakdown of venues, meaning that all offers with the same quality are at the same price: the investor cannot make a mistake and doesn't miss out on anything when he trades on a specific venue. In Navigating Liquidity 5, we mentioned the surprising fact that Europe was more fragmented than the US. Ongoing events tend to confirm the fact that both market structures are becoming more alike, as fewer but more similar exchanges will stay in competition. Q
Inventory-driven investors need fixing auctions
Worldwide equity trading contains two types of trading sequences: a continuous phase and a fixing auction phase. The number of fixing auctions depends on the market, going from none to four (on the Japanese primary market). While all types of platforms ushered in by MiFID are available during the continuous phase of trading, call auctions only take place on primary markets as yet.
16
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
Q
Participation in fixing auctions, relative to overall turnover, has been on an uptrend since early 2010
Fixing auctions: what's at stake
FIGURE 14 shows the proportion of fixing turnover relative to total turnover (primary market and MTFs) on FTSE 100 stocks, on a 20-day moving median, having eliminated witching days from our dataset since fixing volumes are unusually high on these days, as we will see. It appears that this proportion has increased since March 2010. The upward trend is significant, although the proportions show considerable surrounding fluctuations.
FIGURE 14: PROPORTION OF FIXING TURNOVER VS. TOTAL TURNOVER ON FTSE 100 STOCKS 14
13
12
11
10
9
8 30
11 26 08 21 03 09 20 30 13 25 08 19 /11 /09 /08 /0 /05 /0 /01 /11 /10 /0 /07 /0 /03 /11 /11 /11 /11 6 /11 /11 3 /11 /10 /10 /10 8 /10 /10 5 /10 Source: CrĂŠdit Agricole Cheuvreux Quantitative Research
Turnover in fixing auctions is higher when large positions are handled, such as on witching days
This increase in participation in fixing auctions suggests a change in the behaviour of participants. The presence of high-frequency traders (HFT) during the continuous phase, on alternative venues or on primary markets, may have shifted the volume of some players to fixing auctions. To understand why these fixing auctions are attractive and continue to play a major role in the price formation process, fixing mechanisms need to be studied more deeply. Daily fixing turnover shows another specificity of fixing auctions: they help unwind large positions. FIGURE 15 represents the daily turnover (in euros) on CAC40 stocks during fixing auctions. Black lines stand for turnover on triple witching days, while the light green ones stand for derivatives monthly expiry witching days. In fact, fixing turnover appears extremely high during witching days, especially triple ones, as would be expected. Positions in combination with derivative products that expire during these days are taken or closed during these auctions, benefiting from the same price. The London Stock Exchange has specifically created an intraday fixing auction phase on the third Friday of each month, where high turnover is also noted, in the same manner as in FIGURE 15. This suggests that fixing auctions are needed to deal with this type of specific volume days.
17
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
FIGURE 15: DAILY FIXING TURNOVER ON CAC40 STOCKS
Source: CrĂŠdit Agricole Cheuvreux Quantitative Research
Fixing auctions allow for a largely consensual price and make a significant contribution to the price formation process
Whether they happen for opening auctions, intraday auctions, closing auctions, or a volatility interruption auction, fixing auctions serve one main purpose: concentrating more agents, hence forming a more 'consensual' price with more volume. This concentration is useful for building order books before opening the market, but also when the price is used as a reference for many valuations, as closing prices often do. Another illustration that order concentration during fixing auctions allows for a more thorough price formation process can be made with trading halts triggered when a stock's volatility is too high. Academic studies have been carried out on the subject, such as in [Kandel et al. 2008 ("The Effect of a Closing Call Auction on Market Quality and Trading Strategies")]. According to the authors, "In the call auction, consolidation of order flows may reduce the price impact of a trade. Furthermore, the enhancement of information revelation could improve the price discovery process and, by reducing intraday volatility, result in increased price stability." Q
Basic matching rules during call auctions
Fixing rules differ slightly from one market to another. However, they basically follow the same principles. On most markets the fixing price is determined by a four-step approach based on conditional decision rules. If a rule does not lead to a clear auction price from the overlapping buy and sell orders, the next rule is applied, and so on. If orders are not executable against one another, no fixing price can be determined.
18
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
Rule 1 – Rule 2 –
Rule 3 –
Rule 4 –
Maximum Executable Volume: Establish the price(s) at which the highest quantity will be executed. Minimum Surplus: Determine the eligible auction price(s) at which the quantity left in the market at the fixing price is minimum. The word surplus will be used several times in this part and will always refer to the quantity not matched and thus left in the market, at the fixing price. Market Pressure: If the surplus for all potential auction prices is on the same side, it is used to determine the auction price. If the whole surplus lies on the buy side, the highest potential auction price is designated as the fixing price. If the whole surplus lies on the sell side, the lowest potential auction price becomes the fixing price. If the surplus is zero for all potential prices, or if there exist at least two potential prices with a surplus on opposite sides, the next rule is applied. Reference Price: If none of the preceding rules leads to a decision, a reference price is consulted. The reference price is generally the last price traded (either on the same day, or on the day before if no trade has yet occurred). Potential prices determined by the minimum surplus rule are examined according to their relative position to the reference price.
Differences in methodology are noticeable, notably the end time for the fixing auction. Some markets use a random time after the theoretical end during which orders can still be sent. FIGURE 16 is a recap of these maximum random durations (in seconds) for the trading destinations considered. Note that these durations can apply randomly on brackets of stocks. FIGURE 16: MAXIMUM RANDOM DURATIONS AFTER THE THEORETICAL END OF FIXING AUCTIONS (IN SECONDS)
Trading destination
Opening auction
Closing auction
London Stock Exchange Borsa Italiana Euronext Paris Deutsche Börse Nasdaq OMX Stockholm
0 60 0 30 0
0 60 0 30 30
Index/stocks considered FTSE100 FTSE MIB SBF120 DAX OMXS30
US Markets Australian Stock Exchange
300 30
300 30
NYSE stocks All Source: Crédit Agricole Cheuvreux Quantitative Research
Exchanges have established specific rules during fixing auctions, notably the end time
While the London Stock Exchange and NYSE Euronext do not apply any random durations after the theoretical end of the fixing auction, Deutsche Börse, for example, enables orders to be sent from 0 to 30 seconds after the theoretical time. This additional time can vary by day and by stock. Moreover, Xetra fixing auctions can be subject to an extended time before the price determination, in the case of volatility interruption or market order interruptions (if market orders or market-to-limit orders could not be executed at the end of the fixing auction). As we will see later on in this section, these are not the only specific features of German auctions. Some countries outside Europe also apply very specific rules. In the United States, for example, trading in stocks quoted on the NYSE is opened by specialists on the floor that are in charge of each particular stock, somewhere between 9:30am and 9:35am. However, orders cannot be sent after 9:30am. Closing auctions work the same way. In Asia as well, the Australian Stock Exchange indicates a theoretical open time for five 19
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
separate groups of stocks (depending on their code). The open time can happen in a range of 15 seconds around the indicated time. The closing auction starts at 16:00 for all stocks and matches somewhere between 16:10 and 16:11. Japan does not have any randomness in its fixing auctions but applies a specific calculation method - the Itayose method – and the opening of a stock happens as soon as this calculation is over. Other Asian exchanges use fixed times for fixing auctions.
Q
Pre-fixing dynamics unravelled •
Theoretical matching curves
The composition of traders is different at the opening auction and at the closing auction. This translates into two different approaches to the fixing volume. We have studied market feeds during auctions for stocks on the FTSE100, FTSE MIB, SBF120, DAX and OMXS30 from 1 January 2011 to 30 March 2011. The data used are sent by the exchanges and indicate for each given instant the volume that would be matched if the fixing auction had ended, the related price and potentially the surplus (defined above) and some limits of the order book. For the rest of this section, we will refer to the relative difference between the volume that would be matched at time t if the auction ended at time t with the final volume matched at the real fixing, for each time t: these will be called "theoretical matching curves". Thus, if a point of the matching curve equals zero, this means that total end volume has been sent to the market. FIGURE 17 presents the median matching curve during the opening auction, with the inter-quartile range (green) and a sample day (black). The figure has been built on CAGR.PA during our dataset period. The x-axis represents time passed since the beginning of the auction call (in minutes). FIGURE 18 presents the same data for the closing period.
FIGURE 17: MATCHING CURVE DURING THE OPENING AUCTION PREFIXING PERIOD
FIGURE 18: MATCHING CURVE DURING THE CLOSING AUCTION PREFIXING PERIOD
Source: Crédit Agricole Cheuvreux Quantitative Research
20
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
Traders arrive slowly during opening auctions, but participate more quickly in closing ones
During the pre-opening period, book building accelerates as time goes on, and matched volume only reaches its final level in the last few moments: the curve is convex. During the closing call, matched volume rapidly increases to its final level, as agents are already present on the market and many must participate in the fixing in order to unwind their positions: the curve is concave. Participants take longer to intervene in the opening session than they do in the closing one. This reveals less uncertainty but also less patience at the end of the day. In fact, participants face more constraints in the closing session, as it is the last time they can get or close a position. •
The German opaque order book is tested
Interestingly, one fixing stands out from the crowd, as it is the only one where a significant portion of the volume curve reaches a bigger value than the final fixing volume and then decreases towards it. For closing call volume curves on other markets, more than 90% of the distribution remains below the final value. This phenomenon can be seen in FIGURE 19 and FIGURE 20. They show the UK and German volume curves, respectively. FIGURE 19 shows that only the max volume curves (less probable situations) go above the zero threshold. FIGURE 20 shows that the median curve already reaches the zero threshold. This means that, 50% of the time, the final volume will be reached before the end of the fixing auction. Moreover, the upper quartile goes higher than this zero threshold. Since one fixing rule is the maximisation of volume, going over the zero threshold implies that a certain amount of orders have to be cancelled or modified before the end, and that this occurs in 25% of situations studied.
FIGURE 19: MATCHING CURVE ON CLOSING AUCTION
FIGURE 20: MATCHING CURVE ON CLOSING AUCTION
ON FTSE 100 STOCKS
ON DAX30 STOCKS
Source: CrĂŠdit Agricole Cheuvreux Quantitative Research
The German fixing order book is tested with orders that will be cancelled or modified before the end
One specific feature of German auctions is that the order book is not revealed to participants during the fixing auctions. This phenomenon may be a sign that players are trying to get information on volumes and price by sending orders that they would then cancel, thus testing the order book for a lack of transparency.
21
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
•
Surprisingly, participation in fixings takes place at each full second, indicating peaks of information
Informational peaks
Delving more into the dynamics of pre-fixing periods, one can see that, although the global approach to final volumes differs for opening and closing auctions, the pace at which the information arrives follows a similar pattern. Indeed, there are informational peaks that are common across all stocks and European markets. Almost every stock every day has an important number of updates around a particular time. These updates tend to make the matched volume closer to its final value, therefore one can see these peaks as informational peaks. To visualise these peaks, FIGURE 21 presents the probability that an update during the closing auction happens in a 5-second range. The stems stand for informational peaks. The x-axis is once again time passed since the beginning. DAX30 opening auction data are used. Peaks are seen at 10min, 5min, 3min and 2min30s, 1min, 30s and 10s. Although our figure represents German data, these times are fairly common across all opening auction data.
FIGURE 21: PROBABILITY THAT AN UPDATE DURING THE CLOSING AUCTION HAPPENS ON A 5-SEC RANGE
FIGURE 22: MATCHING CURVE ON DAX30 STOCKS DURING THE CLOSING AUCTION
Source: CrĂŠdit Agricole Cheuvreux Quantitative Research
Times differ when focusing on opening or closing auctions. When concentrating on the last minute before theoretical end t, peaks are located around t-10s for opening auctions, while they are around t-30s and t-5s for closing auctions. Fixing auctions have demonstrated their usefulness, particularly when dealing with large positions. They are thus dedicated to inventory-driven investors, as opposed to relativevaluation investors, whose price formation process needs continuous auctions. Opening auctions and closing auctions have shown their own specifics in terms of participants' behaviour, suggesting that they offer different services and attract different types of traders. In the next section, we will take a step back to look at the timespan of the day and study the overall behaviour of participants, to finally concentrate on specific players in the markets.
22
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
II— Timing is money: investors need to trade accordingly Market design and information flow timing imply liquidity patterns
Q
The concentration of trading often reflects how informative traded prices are
Intraday volume patterns are important, not only because they are critical for execution to a benchmark such as VWAP, but also because the concentration of trading in a specific time span during the day often reflects how informative the traded prices during this interval are. More than 20 years ago, Admati and Pfleiderer ("A Theory of Intraday Patterns: Volume and Price Variability") demonstrated the very intuitive result that there is an incentive, for both informed traders and liquidity traders, to time their trades at the very same moments. Furthermore, there are meeting points in terms of information, such as the opening or closing of some markets that influence trading on other markets, or the release of important macroeconomic news always at the same time of the day. Such meeting points create specific patterns. FIGURE 23: INTRADAY VOLUME PATTERNS ACROSS THE GLOBE
Source: CrĂŠdit Agricole Cheuvreux Quantitative Research
In FIGURE 23, we have represented the intraday volume patterns of the most important markets worldwide. More specifically, we have used 15-minute bins in order to accumulate the traded turnover on a specific market, and plotted these accumulated volumes with green bars. The fixing auctions have been plotted with a dark line ending with a point, in order to highlight these specific market phases. The timings represented
23
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
Information can come from a news release or from prices traded on another market
here in Coordinated Universal Time are obviously dependent on the period of the year, as many of these countries are subject to daylight savings time, which will change their position on this axis. Even the relative position would be different, as not all the countries use daylight savings time, and the ones that do don't enact the change on the same day. Finally, some markets have variable trading times, even in their own time zone, depending on the day of the week (Indonesia), during Ramadan (Pakistan), or to have closer trading times to a linked market that is subject to daylight savings time (the Brazilian market is an example of such a market; it has summer and winter trading times). The classical U-shaped or J-shaped patterns observed for US stocks can become a much more complex pattern when several effects are mixed. These specific patterns are often induced by one of the following causes: 1) The opening of another linked market; 2) The release of news at the same time of the day on a regular basis (e.g. macroeconomic news or earnings announcements, which are often released before the opening);
Many investors are benchmarked to a reference price, and this causes huge volume in order to target this price
3) The way reference prices are computed, and whether a specific fixing phase is used especially for this purpose. Before going into an in-depth analysis of these effects, let's clarify the third driver. The reference price in Europe for many funds’ NAV computations or equity derivatives' margin computations is the price of the closing auction, which makes it a very important source of liquidity. The effect of the increased importance of the closing auction when its price is used as a reference has been shown by Kandel et al. ("The effect of a closing auction on market quality and trading strategies") by comparing the reallocation effect of the introduction of a closing call auction to the markets of Paris (1998, with the reference price set to the price of this call auction) and Milan (2001, when the reference price remained computed as a VWAP over the continuous phase). Conversely, India has no closing fixing, and the reference price is computed as the VWAP over the last half-hour of trading, hence the two huge green bars at the end of India's trading day (see FIGURE 23). We focused on this in FIGURE 24 by plotting the volume pattern during the last 40 minutes of trading on a specific stock. This shows that there is a substantial shift in the level of volume during the last 30 minutes.
FIGURE 24: RELIANCE INDUSTRIES LTD'S VOLUME PATTERN IN INDIA DURING THE LAST 40 MINUTES OF TRADING (1 MINUTE BINS)
FIGURE 25: CNOOC LTD'S VOLUME PATTERN ON HONG KONG DURING THE LAST TWO MINUTES OF TRADING (1 SECOND BINS)
0.25%
0.2%
0.15%
0.1%
0.05%
0% 15
:58
:15
15 :58 :30
15 :58 :45
15 :59 :00
15 :59
:15
15 :59 :30
15 :59 :45
16 :00 :00
Source: CrĂŠdit Agricole Cheuvreux Quantitative Research
24
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
A striking example of the reference price effect in FIGURE 23 is Hong Kong, with 17% of the traded volume falling in the last 15-minute bucket and a quarter of this volume being concentrated in the very last minute. The reference price for this market is indeed the median of five prices observed every 15 seconds during the last minute of trading, which undoubtedly explains the very pronounced J-shaped pattern for this market. FIGURE 25 focuses on this effect and shows that inside the very last minute of trading in Hong Kong, the volume pattern is clearly driven by the specific way the reference price is computed on this market. As we can see, there is an increase in the level of volume traded for the last minute, but also four spikes in volume at 15:59:00, 15:59:15, 15:59:30 and 15:59:45, the exact times of the four first snapshot prices used for reference price computation. The absence of the fifth spike is probably due to the risk of not being able to complete the order if trying to get the price at the exact time of the close of the market. The Hong Kong stock exchange made an attempt to change this closing price determination by introducing a closing call auction in May 2008, but suspicions of manipulation of the closing price made it reverse the change on 23 March 2009.
The dependence of one market on another can be strong enough to delay the beginning of the price formation process
Regarding the effect of the opening of one market on another one, an example can be seen in FIGURE 23, on Brazil's volume pattern. The influence of the US market even causes the volume to increase at the very beginning of the day, creating a very special shape compared to all the other markets across the world. It has the effect of delaying the "real opening" of the Brazilian market. That is to say that it is effectively possible to trade at the opening of the market, but the real price formation process will only take place half an hour later, when the US market opens. This strong dependence on the US market also explains why Brazil has summer and winter trading times, with the aim of being more closely matched to US business hours. Another clear illustration of such a phenomenon is given on FIGURE 26, which represents the volume pattern in South Africa. As can be seen in FIGURE 23, this country is not subject to daylight savings time, so during the summer, the South African market opens at the same time as most European ones, but opens one hour earlier than European ones during the winter. This leads to a change in the shape that can be seen at the beginning of the curves. As is the case for Brazil, the dependence is strong enough to make the volume increase at the very beginning of the day. FIGURE 26: SOUTH AFRICA VOLUME PATTERN EXCLUDING FIXING (10-MINUTE BUCKETS) 4.5% Winter Summer
4% 3.5% 3% 2.5% 2% 1.5% 1% 0.5% 09
1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 :00 9:30 0:00 0:30 1:00 1:30 2:00 2:30 3:00 3:30 4:00 4:30 5:00 5:30 6:00 6:30
Source: CrĂŠdit Agricole Cheuvreux Quantitative Research
25
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
Q
Blue chip stocks’ volume patterns are severely impacted by all these factors
An example of mixed effects
We will now take a more in-depth look into the effect of the aforementioned causes of volume patterns. Consider the case of Total SA (the largest market cap on the CAC40) on FIGURE 27, which is a clear example of a mix of all these effects. In this chart, we represented the traded turnover in 10-minute buckets. This is made with data from 1st January 2006 to July 2011. We added in light green the same computation made on the days when the US stock market is closed, i.e. 36 days in our sample. Unlike the previous charts, we represented here the mean traded turnover (in 10-minute buckets) and not the proportion of volume that is traded in the bucket. This will therefore enable us to observe the effect on activity during the whole day, and not only the distribution of volume in a day.
FIGURE 27: TOTAL SA'S VOLUME PATTERN ON THE FRENCH MARKET DURING CONTINUOUS AUCTION
FIGURE 28: TOTAL SA'S VOLUME PATTERN INDUCED BY US MARKET BEING OPEN 15 Eur mil. All days NY "shift" NY closed
10 Eur mil.
5 Eur mil.
09 09 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 :00 :30 :00 :30 :00 :30 :00 :30 :00 :30 :00 :30 :00 :30 :00 :30 :00 :30
FIGURE 29: TOTAL SA'S VOLUME PATTERN INDUCED BY US MACROECONOMIC NEWS 25 Eur mil.
20 Eur mil.
FIGURE 30: TOTAL SA'S VOLUME PATTERN INDUCED BY DERIVATIVES EXPIRY 100 Eur mil.
All days US Employment Situation US ISM Manufacturing
90 Eur mil. 80 Eur mil.
All days Quarterly Expiry Monthly Expiry
70 Eur mil.
15 Eur mil.
60 Eur mil. 50 Eur mil. 40 Eur mil.
10 Eur mil.
30 Eur mil. 20 Eur mil.
5 Eur mil.
10 Eur mil. 09 09 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 :00 :30 :00 :30 :00 :30 :00 :30 :00 :30 :00 :30 :00 :30 :00 :30 :00 :30
09 09 10 10 11 11 12 12 13 13 14 14 15 15 16 16 17 17 :00 :30 :00 :30 :00 :30 :00 :30 :00 :30 :00 :30 :00 :30 :00 :30 :00 :30
Source: Crédit Agricole Cheuvreux Quantitative Research
26
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
1) Opening of the US market On FIGURE 28, we have plotted the volume pattern over our full sample (which appears on all four figures from FIGURE 27 to FIGURE 30 in order to facilitate comparisons of the different patterns), as well as the NY "shift" volume pattern and the NY closed volume pattern. The NY "shift" volume pattern has been computed using the days when there is only a five-hour difference between French time and US time. This usually happens during two weeks in March and one week in November, and is due to the fact that daylight savings time starts earlier in the US than in Europe, and lasts longer. The effect during this particular period of time is fairly clear, and the change in the shape of the volume pattern matches what intuition suggests, i.e. the increase in volume occurring at the time the US markets open is one hour earlier than usual. The spike due to macroeconomic news releases (which will be discussed later) is naturally also one hour earlier (13:30 CET instead of 14:30 CET usually). It is also worth noting the exaggerated spike in volume at 12pm is a consequence of the computation period: the time period includes March, and therefore the third Friday of March and the expiry of futures and options on the Eurostoxx 50. These expiry days are the cause of this spike (as we will see later), the consequence being that in this NY "shift" sample, the proportion of witching days is higher than in the full sample, hence the exaggerated spike. On a regular day of this sample, volume is no higher at 12pm than on the full sample days. The NY closed volume pattern has been computed on the days when the US market is closed but the French market is open. As we can see, there is not only a change in the volume pattern, but also in activity over the full day. And this is true even in the morning when the US market would not be open anyway. This shows the reluctance of investors to trade on such days when no information will come from the leading US market. The opening of a linked market can influence another in three ways: Investors trading on the markets that have just opened decide to invest in the other market, thus generating extra volume traded; Q
Q Information from the opening prices leads to an increase in the volume traded, as new information is digested and leads to a correction in the evaluation of the fundamental price of the stocks; Q
Arbitrages can be made between the two markets, generating extra volume.
The third reason is not fundamentally different from the first two but is not caused by the same pool of investors. We are thinking more of systematic arbitragers in the third case, as opposed to asset managers in the previous two. 2) US macroeconomic news On FIGURE 29, we have plotted the volume pattern over our full sample, as well as the "US employment situation" and "ISM manufacturing", which are the volume patterns computed on the days when these economic reports are published. The Bureau of Labor Statistics releases a report on the employment situation in the United States usually on the first Friday of the month at 8:30am (US eastern time, which converts into 14:30 on the chart). The "US employment situation" volume pattern has been computed with days on which this report has been released. The most popular indicator in this survey is non-farm payroll employment, which is meant to represent the number of jobs added or lost in the economy over the past month, not including jobs related to the farming sector. As shown in FIGURE 29, the volume traded just after the release of the report can be impressive, reflecting the information content of this report on the general health of the worldwide economy. 27
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
The other pattern, "ISM manufacturing", refers to the report usually released at 10am (US eastern time, which converts into 16:00 on the chart) on the first business day of the month. The ISM manufacturing composite index indicates whether manufacturing and the overall economy are growing or declining. We can see on FIGURE 29 that the effect of this indicator is more mitigated than the previous one, but this is not surprising as only a portion of the economy is concerned. Obviously, the immediate effect of the release of any economic indicator on the volume traded will depend on how surprising it is for the market (generally measured by its distance from the consensus range). But we can also see that even without any further information the volume pattern is different on these days, especially in the case of major releases such as the employment situation. 3) Equity derivatives expiries In FIGURE 30, we have plotted the volume pattern of Total SA on 'witching' days. These days include equity derivatives expiries. The name "witching" comes from the historically erratic behaviour shown by cash markets on these days, due to the unwinding of hedge positions or the need to deliver physical settlement. Nowadays, this still generates considerable activity on the cash market, as shown in FIGURE 30, but the effect on price is somewhat mitigated.
Knowledge to exclude statistical artefacts…
…pattern detection and robust quantitative methodologies...
In the case illustrated here, the expiries happen on the third Friday. The derivatives with an expiry here are futures and options on the Eurostoxx 50 as well as the CAC40. The EDSP ("Exchange Delivery Settlement Price") for derivatives on the Eurostoxx 50 is the arithmetic mean of the index price disseminated between 11:50 and 12:00 CET. So if you have a position on this future, hedged by stocks, and do not want to roll the position over, then you have to unwind your cash position, preferably at the same value your position in the future will have, that is to say the one computed on the EDSP. This is why you would target an unwinding of your cash position between 11:50 and 12:00, generating the huge volume as shown in FIGURE 30. The smaller effect seen in FIGURE 27 is just the consequence of averaging; the impact of monthly expiries is divided by a factor greater than 20 and the one of quarterly expiries by more than 60. It obviously makes little sense to incorporate this into a trading profile used for VWAP execution during a non expiry day, so both a more robust statistical methodology and outlier filtering are used for our execution algorithms. The extra volume between 15:40 and 16:00 that can be observed in FIGURE 30 is due to CAC40 derivatives, whose EDSP are computed on prices observed in this specific span of time. We have already shown in the previous section (p18) that equity derivatives expiries also have an effect on the volume traded at the closing fixing. We have not shown it here, as it would have made the plot even harder to read as this effect is very high on a stock such as Total SA. The cause of this effect is that the EDSP for single-stock futures and options is the closing call auction price.
…are key elements to provide a meaningful trading profile
Such knowledge about the causes of these surges in volume is important in order to differentiate between statistical outliers and 'true' volume patterns. It is also important for an investor to understand that a VWAP or a PVOL that includes the period of time encompassing these expiry times (especially 11:50-12:00) will really be reduced to getting the price between 11:50 and 12:00, as most of the volume will be traded in this time. Therefore, systematic statistical and quantitative measures are needed to capture these patterns and exclude them from the usual trading profile on the global universe of stocks that our algorithms allow you to trade.
28
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
Q
Different market impacts for different investor behaviour patterns
Market impact depends on investment style
This section is devoted to an in-depth analysis of what market impact is and of how the behaviour of investors influences market impact. This analysis leads CA Cheuvreux Quant Team to recommend bearing this link in mind when designing an execution strategy. We study the market impact on two different scales: intraday and extraday. On the intraday scale, we see that the impact of an order being executed increases over time. It also appears that at the end of the execution the price of the traded stock tends to revert back to its level before the execution. On the extraday scale, we study the evolution of prices over a period of several days before and after the execution day. We adapt the CAPM framework, which breaks down the price of assets into two components, a market component (also called systematic) and an idiosyncratic component, to the exploration of market impact. We can then separate a "market" market impact from an idiosyncratic market impact. The analysis of these two types of impacts enables us to emphasise different types of investor behaviour. There are investors who have a higher impact on the market component and those who mainly affect the idiosyncratic component. Both types of investors are not expected to trade the same way. Better understanding what impacts the price
In a CAPM perspective, the price of a stock is mainly driven by two forces: a market or systematic component (Syst. Comp.), and an idiosyncratic component (Idio. Comp.) based on information specific to the stock. Price = Syst. Comp. + Idio. Comp.
Market impact measures how the price of a stock moves when it is being traded
Market impact measures how the price of a stock moved away from its past level due to the presence of an order being executed. As in the CAPM framework, this movement could be a market-wide one or could be specific to the stock traded. In order to disentangle this combination of two different impacts, we developed a specific framework inspired from the CAPM. Market Impact = Idio. Market Impact + Syst. Market Impact When one combines the price of the stock and the market impact, one still needs to add the costs of trading to get the real price of an execution. We call these costs slippage. The costs of trading are made up of the cost of market access plus the cost of capturing liquidity. The cost of accessing the market is a fixed cost (fees, rebates, etc.) and the cost of liquidity capture is variable (price of crossing the spread, adverse selection, etc.). In the next section, we will set the fixed cost aside and consider that it is nil. Trading costs will thus be referred to as slippage: Execution Price = Price + Market Impact + Slippage
Market impact over the trading period The major benefit of an intraday analysis of market impact is to quantify the changes in the market impact as an order is being executed and the dilution of this impact after the end of an execution. However, this requires a huge amount of execution data and the ability to process it.
Market Impact = Price – Arrival Price
The difference between the price at any given time and the arrival price is a good and often used measure of market impact. FIGURE 31 and FIGURE 32 both show the trend in the return of a stock according to the arrival price in units of spread over time. For this study, we used Cheuvreux executions during 2010 on components of the French CAC40 stock index.
29
www.cheuvreux.com
Global Research Navigating Liquidity 6
January 2012
FIGURE 31: INTRADAY MARKET IMPACT IN SPREADS
FIGURE 32: PRICE REVERSION AFTER EXECUTION IN SPREADS
1.4
1.2
1.2
1
1
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0
0
0
10
20 30 40 50 60 70 80 Time Since Beginning of the Order in pct of Duration
90
100
110
120 130 140 150 160 170 180 190 Time Since Beginning of the Order in pct of Duration
200
Source: CrĂŠdit Agricole Cheuvreux Quantitative Research
Market impact is concave over execution time
Market impact reverts back after execution
FIGURE 31 shows the changes of the price of a stock while an order is being executed. We chose to look at the returns of the stock from the perspective of a buy order. This means that if the price change is above 0 at any given time, the investor will pay more to purchase a share or will get paid less to sell a share at this time rather than at the beginning of the execution. The x-axis is graduated in percentage of execution time. That is to say each execution starts at 0 and terminates at 100. The black curve shows the average of stock returns according to arrival price in units of spread across all the orders executed. The grey area is the 95% confidence interval of that average. The shape of the impact curve is concave over time, or to put it another way, a slice executed at the beginning of the order has more impact than one executed at the end of an order. This kind of result is exactly the same as those found in academic studies. See for example Market impact and trading profile of hidden orders in stock markets E. Moro, J. Vicente, L. G. Moyano, A. Gerig, J. D. Farmer, G. Vaglica, F. Lillo and R. N. Mantegna. The chart represented in FIGURE 32 shows the stock returns after the end of the execution. The units and the axis are the same as in FIGURE 31. The price reverts back after the execution to a level lower than that reached at the end of execution. Market impact may depend on several parameters. There are stock-specific characteristics: liquidity of the stock, volatility, spread size; and there are order-specific ones: aggressiveness of the execution, quantity to execute, expected duration. The better we understand how these parameters drive market impact, the better the pre-trade analysis and thus the execution quality. In what comes next, we focus on the structure of the dependency between market impact and different parameters, namely the participation rate (or participation liquidity ratio, PLR) and the duration of an order. We observe that the market impact increases when participation rate and duration increase, even independently.
30
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
FIGURE 33: INTRADAY MARKET IMPACT IN SPREADS DEPENDING ON THE PARTICIPATION RATE 2.5 24.9 2
1.5
18.3 15.0
1
11.6 0.5
Participation Rate in pct
21.5
8.3 0
-0.5
4.9
0
20 40 60 80 Time Since Beginning of the Order in pct of Duration
100
1.0
Source: Crédit Agricole Cheuvreux Quantitative Research
In FIGURE 33, we have plotted curves of market impact for different values of the participation rate. These are the same kind of curve as in FIGURE 31. The colour bar on the right of the chart indicates for each colour the corresponding value of the participation rate. The lowest participation rate stays near the bottom. It increases as the colour of the curve goes from blue to green. It is quite obvious that the market impact curves get steeper as the participation rate increases.
FIGURE 34: MARKET IMPACT VS. PARTICIPATION RATE FOR SHORT ORDER DURATIONS 2
4.5
1.5
18.5 0.5
13.5
Participation Rate in pct
23.5
8.5 0 3.4
0
20 40 60 80 Time Since Beginning of the Order in pct of Duration
100
33.0
3.5
28.3
1
25 < Duration (min) < 50 Duration (min) < 25
4
0.4
28.2
3 23.5
2.5
18.5
2 1.5
13.4
1
Participation Rate in pct
33.3
-0.5
FIGURE 35: MARKET IMPACT VS. PARTICIPATION RATE FOR LONG AND SHORT ORDER DURATIONS
8.5 0.5 3.2
0 -0.5
0
20 40 60 80 Time Since Beginning of the Order in pct of Duration
100
0.4
Source: Crédit Agricole Cheuvreux Quantitative Research
Market impact increases as the duration or participation rate increases
In FIGURE 34 and FIGURE 35, we have indicated impact curves for different participation rates and duration values. The components of both figures are the same as those in FIGURE 33, the colour indicates the value of the participation. FIGURE 34 shows the market impact when order durations are shorter than 25 minutes. FIGURE 35 shows market impact when the order duration is between 25 and 50 minutes. For comparison purposes, we have plotted the impact curves from FIGURE 34 too (dashed curves). If we compare two curves of the same colour, i.e. two impact curves obtained from orders of 31
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
the same participation rate but of different duration, we see that the dashed curve (the short duration) is less steep than the other one (the long duration), thus demonstrating that the longer the duration, the greater the market impact. At CA Cheuvreux, we have built a model based on the observations made above. A thorough analysis of our order flow leads us to design a model which looks like: Market Impact = Duration α × Participation γ FIGURE 36: MODELLING OF THE MARKET IMPACT. DEPENDENCE OF MARKET IMPACT ON DURATION AND PLR
FIGURE 37: SAME CHART SEEN FROM ANOTHER ANGLE
Source: Crédit Agricole Cheuvreux Quantitative Research
The two 3D charts above (FIGURE 36 and FIGURE 37) show how market impact depends on duration and on the participation rate. The blue dots are the average values of the market impact across buckets of duration and participation rate. The surface shows our model's prediction for the corresponding value. The duration reads on the lower left axis in FIGURE 36 and on the lower right axis in FIGURE 37. The duration unit is a number of 5-minute intervals. The participation rate reads on the lower bottom axis in FIGURE 36 and the lower right axis in FIGURE 37. The two different angles on these two figures show how well our model fits the surface of market impact. FIGURE 36 and FIGURE 37 show that the dependence of market impact on the two parameters separately is concave. This means that the impact of one additional percent of participation rate for an order of a given duration is lower for high participation rate orders than for low participation rate orders. Moreover, the calibration of the model gave us a gamma (the scale parameter of market impact according to participation rate) approximately equal to 0.5.
Market impact on a longer horizon: Different patterns for different investment styles The previous section allowed us to highlight different things about the intraday market impact. In this section, we will consider the problem of market impact from a slightly wider perspective. We will zoom out from a 5-minute sample data to daily data.
The analytical toolbox of the CAPM allows us to distinguish different types of investor behaviour
Data analysis at this scale, using the analytical toolbox of the CAPM, will allow us to distinguish different types of investor behaviour. For instance, some investors are more keen to trade baskets of index components, thus impacting the market widely, while others trade only a few stocks: some are trend followers, while others bet on mean reversion. The following analysis will allow us to rank the behaviour and especially to show that for each of these types, there is a best type of execution. Taking into account these
32
www.cheuvreux.com
Global Research Navigating Liquidity 6
January 2012
different types of behaviour and using the recommended execution is the best way to minimise its market impact.
The average market impact between the execution day and the previous day's closes is two or three spreads wide
From now on, we will study the changes of the daily closing price of a stock over a period of several days before and after the execution date. FIGURE 38 shows the average of price moves each day. We decided to look at price moves according to the price on the day before the execution date (date -1 in FIGURE 38). Each point of the curve is the average of these price moves across all Cheuvreux orders posted in 2010 depending on the distance from the execution day. Thus the point on date 0 is the average of closing price moves on the execution day and the point on date -10 the average of closing price moves ten days before the execution day. These moves are measured in number of spreads. We took the perspective of a buyer. This means that if the curve is above 0 on any given date d, the cost to acquire one share on d (or the amount received to sell one share) will be higher (or lower, if this is a sell order) than on 0. We see that the average execution day closing price is (in the case of a buy order) two or three spreads higher than the previous day's closing price. To apply a CAPM-like approach to market impact, we broke down stock prices into systematic and idiosyncratic components. FIGURE 39 shows the average of these two components. We can see in this chart that the jump present on the execution day (date 0 in FIGURE 38) is almost completely integrated into the idiosyncratic component, indicating that, on average, investors are more likely to impact the stock price than the market.
FIGURE 38: EXTRADAY MARKET IMPACT (IN SPREADS)
4
FIGURE 39: BREAKDOWN OF EXTRADAY MARKET IMPACT INTO SYSTEMATIC + IDIOSYNCRATIC COMPONENTS (IN SPREADS) 1.5
1
3
0.5
2
0
1 -0.5
0
-1
-1
-2 -20
-1.5
-2
-15
-10 -5 0 5 10 Number of Days Since Execution Day
15
20
-15
-10
-5 0 5 10 Number of Days Since Execution Day
15
20
Source: CrĂŠdit Agricole Cheuvreux Quantitative Research
Dependence between investment style and market impact on a monthly horizon Let us extend the above methodological framework for the purposes of analysing the specifics of extraday market impact according to investor behaviour. The same interpretative framework can be usefully applied here to link the shape of the extraday market impact curve to investor behaviour. FIGURE 40 and FIGURE 42 show curves of extraday market impact (similar to FIGURE 38) for two types of investor behaviour. Note that in the first case (FIGURE 40), the price after the execution day increases (or decreases if it is a sell order) for several days. Those investors are interested in stocks whose prices follow a trend. We call them trendfollowers. In the second case (FIGURE 42), the price of the stock after having taken a 33
www.cheuvreux.com
Global Research Navigating Liquidity 6
January 2012
plunge recovers quickly on execution day back to its level a few days earlier. This phenomenon is a sign that we are in the presence of investors whose preferred strategy is mean reversion.
A trend follower impacts the market more than a mean reverter
FIGURE 41 and FIGURE 43 both show the breakdown of the market impact curve of FIGURE 40 and FIGURE 42 into systematic and idiosyncratic components. One can observe that the impact of the market component in FIGURE 41 is greater than in FIGURE 43. So investors who are more trend followers are more likely to impact the market than investors who are more mean reverters. The two types of behaviour presented here are very normal. The first type characterises the trend followers, who buy exposure to the market and play long-term trends. The second type encompasses all the investors betting on mean reversion and who practise stock picking. Their market impact on the overall market is weaker than those of the first type.
FIGURE 40: MARKET IMPACT EXTRADAY FOR TREND-FOLLOWER INVESTORS (IN SPREADS)
40
FIGURE 41: BREAKDOWN OF MARKET IMPACT EXTRADAY INTO SYSTEMATIC + IDIOSYNCRATIC COMPONENTS FOR TRENDFOLLOWER INVESTORS (IN SPREADS) 40
30
30
20
20
10
10 0
0
-10
-10
-20 -20
-20
-15
-10 -5 0 5 10 Number of Days Since Execution Day
15
20
-30 -20
-15
-10 -5 0 5 10 Number of Days Since Execution Day
15
20
Source: CrĂŠdit Agricole Cheuvreux Quantitative Research
34
www.cheuvreux.com
Global Research Navigating Liquidity 6
January 2012
FIGURE 42: MARKET IMPACT EXTRADAY FOR INVESTORS PLAYING MEAN REVERSION (IN SPREADS)
8
FIGURE 43: BREAKDOWN OF MARKET IMPACT EXTRADAY INTO SYSTEMATIC + IDIOSYNCRATIC COMPONENTS FOR INVESTORS PLAYING MEAN REVERSION (IN SPREADS) 4
7
3.5
6
3
5
2.5
4
2
3
1.5
2
1
1
0.5
0
0
-1 -20
-15
-10 -5 0 5 10 Number of Days Since Execution Day
15
20
-0.5 -20
-15
-10 -5 0 5 10 Number of Days Since Execution Day
15
20
Source: Crédit Agricole Cheuvreux Quantitative Research
In this section, we have not addressed the problem of execution costs. Market impact is a measure of how the market price of one or several stocks changes when an order is under way. A measure of the execution cost should include, in addition to the market impact, a measure of the cost of liquidity capturing, the so-called slippage (see beginning of the section, p29). The slippage depends on the dynamic and interactions with the order book. A trading algorithm is made up of two main elements. The first is the strategy segment that aims to optimise a benchmark. This could, for instance, be the market impact in the case of an IS algorithm. The second element is the so-called tactic. Its role is to play with the different order books and liquidity pools to obtain optimal execution prices. More precisely, it optimises the slippage. As stated previously, the measure of impact is random. FIGURE 44 shows that total impact is the sum of idiosyncratic impact plus systematic impact. The uncertainty area (diagonal hatching) shows the possible values of total impact.
FIGURE 44: TOTAL IMPACT IS RANDOM AND EQUALS THE SUM OF SYSTEMATIC IMPACT + IDIOSYNCRATIC IMPACT
Source: Crédit Agricole Cheuvreux Quantitative Research
35
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
Q
High-frequency traders do not impact all investors equally
With the advent of competition between trading platforms, some arbitrageurs began to specialise in arbitraging liquidity imbalances between the same stock quoted on several pools. Thanks to the maker/taker fee schedule on MTFs, platforms may reward such trading strategies in order to provide liquidity. The intensive use of technology allows them to lower their inventory risk by trading very small orders, very frequently. These arbitrageurs are called "high-frequency traders". They are said to be a counterparty in 70% of the deals in the US, 40% in Europe. And 35% in Japan. High-frequency trading (HFT) is defined by the US financial regulator (SEC) as follows: 1) The use of extraordinarily high-speed and sophisticated computer programs for generating, routing and executing orders; 2) Use of co-location services and individual data feeds offered by exchanges and MTFs to minimise network and other types of latencies; 3) Very short timeframes for establishing and liquidating positions; 4) Submission of numerous orders that are cancelled shortly after submission; 5) Ending the trading day in as close to a flat position as possible (that is, not carrying significant, unhedged positions overnight). This type of trading can be specific to certain traders of funds or embedded in larger strategies or management of client orders. HFT is an important component of a market that seeks competition between pools. In brief, it is the "price to pay" to support competition. The question that has increasingly been raised and which remains unanswered is: "Is the value they extract from the market worth the service they provide?" This led the European Commission to try to address HFT in its review of MiFID, but things are extremely complex. The first element is to define exactly what service HFTs are providing to the markets.
Q
How could HFT lower the spread without changing execution costs?
One major argument given by HFT proponents is that they have decreased bid-ask spreads. In order to visualise how HFT can intervene in the market and decrease spreads, one strategy is detailed here. We will see that the effective spread is reduced on an overall basis, but that costs for other participants are unchanged or slightly increased, contrary to popular belief. Let us consider a market with two participants interacting with each other. These are player A (green in FIGURE 45, FIGURE 46 and FIGURE 47) and player B (orange in the three figures). We will bring a HFT into this market (violet in FIGURE 47), and see its impact. When the first two players are alone, they are present in the first limit of the order book, and waiting to be able to trade. In FIGURE 45, player A becomes impatient and hits player B’s price in order to trade. His cost is the effective bid-ask spread, noted ψ. The next time (FIGURE 46), since player A has already hit in order to trade, it is player B's turn to become impatient. Player B thus hits player A’s price in order to trade. His cost is, as for player A, the effective bid-ask spread: ψ. If the situation remains unchanged, both player A and player B will bear the same cost on average: ψ/2.
36
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
FIGURE 45: ORDER BOOK WITHOUT HFT
FIGURE 46: ORDER BOOK WITHOUT HFT
FIGURE 47: ORDER BOOK WITH ONE HFT
Source: Crédit Agricole Cheuvreux Quantitative Research
Let us concentrate on FIGURE 47, where a HFT intervenes on the market. He (or she) takes a position just on the inside of player B and becomes the first limit against player A. Player A becomes impatient and hits the HFT’s price. His cost is slightly more than half the former effective bid-ask spread: ψ/2+ε. For the moment, it seems that both the bidask spread and the cost for player A have been lowered.
The final effective bidask spread is genuinely reduced…
…but the average cost for other investors is higher than before
The HFT takes a new position very rapidly in front of player B. Since player B’s order has not been matched, he needs to hit the HFT’s new price. His cost is the same as player A's: ψ/2+ε. The final effective bid-ask spread is genuinely reduced, since it represents slightly more than half the former bid-ask spread. However, the average execution cost for player A and player B is still ψ/2+ε, since they are forced to become aggressive for each trade. On the other hand, the execution cost for the HFT player is nil. Add to this fact the smaller trade sizes that HFT prefer, which prompt other actors to fraction their trades and thus to pay more and report more often to clients or authorities. These reporting costs, which only non-proprietary trading firms face, are on the rise, in addition to the technological cost. These fixed losses are unevenly distributed and lead to a deterioration in the trading conditions for final investors. Q
Bid-ask spread: cost and uncertainty for investors
Let us consider that bid-ask spreads for investors would have been unchanged since 2007 if high-frequency traders had not appeared among financial players. This bid-ask spread is denoted by ψB, with B as "Before". As we saw in the previous subpart, HFT players lower the effective bid-ask spreads. However, the spread paid by other investors when hitting a HFT’s price is a little wider: ψB/2+ε. ε is the additional part due to HFT, and corresponds to half the new tighter spread: ψHFT/2. B
B
If we consider that HFTs are involved in q% of trades, the new bid-ask spread for investors is expected to be ψN (N as "new"), with: ψN = qψHFT + (1-q)ψB B
In practice, if we assume that the proportion of HFT among trades is 40% (q = 40% in our previous formula), ψB and ψN take the following values: B
ψB = 5.41bp B
ψN = 4.73bp
37
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
These were obtained by considering a "before" period from April to November 2007, and a "new" period from April to November 2011, on CAC40 stocks. Note that only primary market data (Euronext Paris) are taken into account. Now let's have a look at the real cost this implies for investors. We will consider one investor looking at posting one order. Let's say that he has a probability r to meet a HFT. His cost C will depend on the situation that occurs: Q If he meets a HFT, he has to become aggressive (with regard to our previous subpart), and pays C = ψHFT/2 + ψB/2 B
This case corresponds to the upper line of FIGURE 48. Q
If he does not but is aggressive for other reasons: he pays C = ψB B
This case corresponds to the middle line of FIGURE 48. Q
If he does not and is passive: he pays C = 0.
This case corresponds to the lower line of FIGURE 48.
FIGURE 48: EXECUTION COST FOR INVESTORS
Source: Crédit Agricole Cheuvreux Quantitative Research
FIGURE 49 represents the cost C for an investor and the uncertainty that relates to this cost C. The y-axis is thus the expected cost E(C) mentioned in FIGURE 48, and the x-axis is the uncertainty seen as the standard deviation of the cost C (the square root of variance V(C) in FIGURE 48).
38
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
FIGURE 49: EXECUTION COST AND ITS UNCERTAINTY
Source: Crédit Agricole Cheuvreux Quantitative Research
Costs increase with the number of HFT, but uncertainty is reduced
It highlights the fact that the smaller the cost the higher the uncertainty about it, just as in an efficient frontier for portfolio management: one cannot increase one’s expected gain without increasing the risk one takes. Secondly, the points of the line vary depending on the probability of meeting a HFT. As shown by the orange arrow, increasing the probability to meet a HFT increases the cost (by ε = ψHFT/2), but reduces the uncertainty of the cost. Above, we studied expectation and uncertainty for an investor involved in one trade in the market. However, the situation is different for investors that are involved in a large number of trades. From now one, the number of trades will be denoted by N and C1 ,..., C N the cost of its trades. The Central Limit Theorem indicates that the variance we are looking at decreases in 1/N.
⎛1 E⎜ ⎝N ⎛1 V⎜ ⎝N
⎞
N
∑ C ⎟⎠ = E (C ) i =1
i
⎞
N
∑ C ⎟⎠ = i =1
i
V (C ) N
Consequently, the standard deviation that is seen in various figures decreases in 1/√N. The uncertainty is already lower for any investor that has a larger number of trades, all else being equal. When increasing the number of trades, the line plotted in FIGURE 49 is shifted to the left by homothety: for the same probability of meeting a HFT, the expectation of the cost remains stable, but the uncertainty already decreases. This is highlighted in FIGURE 50, where the curves for three types of investors are presented.
39
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
FIGURE 50: WHICH UNCERTAINTY FOR WHICH INVESTOR?
Source: CrĂŠdit Agricole Cheuvreux Quantitative Research
The decrease in uncertainty thanks to HFT works for retail investors, but not for larger ones
To conclude, increasing the number of HFT (and thus the probability of meeting one) by 20% (from 30% to 50%) increases the expected cost by approximately 0.5bp, for any type of investor (from the pink star to the orange star in FIGURE 50). It also decreases uncertainty, but not for everyone: retail investors benefit from a large decrease of uncertainty on their cost, while the uncertainty on the cost of a pension fund cannot be reduced much further (it was already minimised by the large number of trades). To summarise, bid-ask spread is a source of cost for investors in the market. Both the expected value of this cost, and the uncertainty it is subject to are key figures for them. Depending on the number of trades and the probability he has of meeting a HFT actor, expectations will not vary, but uncertainty will, and significantly. This uncertainty area in which the total cost is located in terms of bid-ask spread is shown in FIGURE 51.
40
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
FIGURE 51: TOTAL SPREAD COST HAS UNCERTAINTY
Source: CrĂŠdit Agricole Cheuvreux Quantitative Research
Q
The slippage of a trading algorithm is comprised of the combination of impact costs and spread costs
Market impact and bid-ask spread: total execution cost of trading
The execution cost of a trading algorithm depends on several factors. An algorithm is made up of two major layers: the strategic layer, which controls the risk with respect to a benchmark (VWAP, arrival price), and the tactical layer, which seeks liquidity inside order books and across liquidity pools. Each of these layers is sensitive to one type of cost. For instance, a VWAP algorithm tracks market VWAP. It is specifically designed to reproduce the profile of market volumes over the day. At a lower level, a VWAP algorithm uses tactics to optimise access to liquidity. In this example, there are two different costs. The first is the cost of miss-tracking market volume and the second is the cost of crossing the spread or of trading while an HFT is present.
41
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
FIGURE 52: TOTAL EXECUTION COSTS: BID-ASK SPREAD VS. MARKET IMPACT DECOMPOSITION
Source: Crédit Agricole Cheuvreux Quantitative Research
Reducing a trading algorithm’s execution cost requires both types of cost to be reduced. These two costs are not deterministic. They are random but more or less independent. The total average cost is the sum of the two average costs, and the total standard deviation (see FIGURE 52) is equal, insofar as they are independent, to the square root of the sum of the variances of both costs.
42
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
III— Algorithmic trading: adapting trading style to investors' needs Q
Each trading feature has its own benchmark
Comparing the main needs in electronic execution from investors in the three main regions, it can been said that: In the US, where HF traders are almost everywhere, stealth trading is the main point to focus on. Q
Q In Asia, the comparison between the efficiency of trading in each local market is very important, since it is crucial to take into account the discrepancies from one market to the next.
In Europe, the need is a mix between avoiding being detected by HF traders, and understanding the efficiency of interacting with a set of order books, to adjust order routing policies as fast as possible. Q
On paper, European investors should target IS benchmarks, Asian ones VWAP or PoV, and US ones liquidity seekers
In terms of benchmarks, this implies that on paper European investors should target Implementation Shortfall (IS) benchmarks, Asian ones VWAP or Percentage of Volume (PoV), and US ones liquidity seekers. Of course, the size of the order and market conditions have to be taken into account to adjust these drastic rules: liquidity seekers are more suitable to trading orders without minimum participation rate constraints, IS are for small orders, VWAP for larger ones and PoV for very large ones.
Optimal execution is a combination of risk control and price and liquidity opportunity capture
On the next page, TABLE 1 summarises these remarks, adding information on typical market conditions naturally associated with each benchmark. The “type of hedged risk” column of this table is of importance, because a trading algorithm is always performing a “risk-reward” arbitrage. As soon as the risk it has to take care of is under control, it has the freedom to adjust the prices and quantities of orders sent to obtain good prices. Optimal execution is therefore the proper combination of risk control (via the following of an “optimal” trading rate) and price and liquidity opportunity capture. This table should not be read as “all conditions in the columns have to be filled to justify the choice of the associated benchmark”, but it gives information on all potential good reasons to use a given benchmark.
Each stock and each investment strategy have their specificities
In terms of trading style, other characteristics have to be linked with each benchmark: the liquidity of the stock traded and the type of investment style that motivated the trade. As we have underlined in previous sections of this issue of Navigating Liquidity: each stock has its specificities because of market design, and each investment style conditions the type of market impact that the trade will be exposed to (hedging trades are more exposed to “execution idiosyncratic impact” than long-term position-building trades that are more exposed to “systemic impact” against which trading algos cannot fight that much). TABLE 2 is a summary of the typical stock features that are suited for such or such benchmark; the typical investment reason related to the trade; and the main expected feature of each benchmark.
43
www.cheuvreux.com
44
www.cheuvreux.com Any size
US
Liquidity Seeker
Type of trade * Long duration position
* Hedging order * Long duration position * Unwind tracking error (delta hedging of a fast evolving inventory) * Alpha extraction * Hedge of a non-linear position (typically Gamma hedging) * Inventory-driven trade * Alpha extraction * Opportunistic position mounting * Already split / scheduled order
Type of stock
Medium to large market depth
Any market depth
Medium liquidity depth
Poor a fragmented market depth
PoV
VWAP / TWAP
Implementation Shortfall (IS)
Liquidity Seeker
Source: Crédit Agricole Cheuvreux Quantitative Research
Do not miss a liquidity burst or a relative price move on the stock
Do not miss an unexpected price move in the stock
Do not miss the rapid propagation of an unexpected news (especially if I have the information) Do not miss the slow propagation of information in the market
Type of hedged risk
Source: Crédit Agricole Cheuvreux Quantitative Research
* Relative price oriented (from one liquidity pool to another, or from one security to another) * Capture liquidity everywhere * Stealth (minimum information leakage using fragmentation)
* Will finish very fast if the price is good and enough liquidity is available * Will “cut losses” if the price goes too far away
* Follows current market flow * Very reactive, can be very aggressive * More price opportunity driven if the range between the max percent and min percent is large * Follows the “usual” market flow * Good if market moves with unexpected volumes in the same direction than the order (up for a buy order) * Can be passive
Main feature
The stock is expected to "oscillate" around its "fair value"
Possible price opportunities
Any "unusual" volume is negligible
Possible negative news
Market context
Benchmark
TABLE 2
Small size (0 to 6% of ADV)
Europe US
Implementation Shortfall (IS)
Medium size (from 5 to 15% of ADV)
PoV
Asia Europe
Large order size (more than 10% of ADV : Average daily consolidated volume)
Asia
VWAP / TWAP
Order characteristics
Region of preference
Benchmark
TABLE 1
January 2012
Global Research Navigating Liquidity 6
January 2012
Global Research Navigating Liquidity 6
Q
The real need is to combine different features of different benchmarks together
Customisation offers multi-feature trading styles
From a practical standpoint, the benchmarks can be combined: you can add minimum or maximum participation rates to an algo to make it behave like a PoV in case of really high or low market volume during the trading; you can use a “limit price” and a “would level” to make an algo behave like an IS. You can also plug a Smart Order Router (SOR) into an algo to tweak its behaviour to resemble that of a liquidity seeker. Thanks to this you can match sophisticated benchmarks like “follow market flows if there is high volume, finish fast if the price is really low, and use as much liquidity as possible” using a VWAP or TWAP backbone with a minimum participation rate, a limit price and a SOR to send orders to trading pools (including Dark Pools). FIGURE 53 gives an example of the combination of a VWAP and a minimum participation rate.
FIGURE 53: INTRA-DAY BEHAVIOUR OF A VWAP WITH A MIN PCT (ACTIVATED AT THE END OF THE TRADING PERIOD). TOP CHART IS FOR THE MARKET PRICE (GREY AND DARK LINES ARE FOR THE AVG. PRICE AND THE VWAP, DOTS FOR ALGO TRADES), BOTTOM CHART: THE TRADING ENVELOPES (UPPER IN RED, LOWER IN GREEN) AND THE TRADING CURVE (IN GREY) VS. THE MARKET CURVE (IN DARK).
Source: Crédit Agricole Cheuvreux Quantitative Research
45
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
Building trading algorithms around trading envelopes opens the door to any combination of features
Such “hard” combinations are what people in stochastic control call “bang-bang control”: 100% of a zero or one policy is applied when a threshold is crossed. It can be gamed (if the threshold is detected, for instance) or be adversely selected (finish too early because of the constraints such as participation rates and limit prices) if the execution conditions are not chosen with enough accuracy, and are not sufficiently adapted to market conditions. A better solution to be able to merge the desired features of different benchmarks in the same trading algorithm is to base the trading logics on trading envelopes. It is first worthwhile to remark that the market data needed to build a benchmark-based policy are noisy: because of high-frequency activity in the order book and uncertainty on some important parameters (of a market impact model or a volume curve, for instance). From a practical standpoint, the “optimal trading curve” to follow is expressed as a trade rate that has to be respected by the algorithm: if it obtains trades too fast with respect to this curve, it has to stop posting; if it does not obtain enough, it needs to obtain trades aggressively, paying the spread. Because of the uncertainty and noise around the analytics used, the “curve” is not a thin line at all; it is a thick one, thick enough to define in fact a “trading envelope” (see FIGURE 54 for such envelopes for a VWAP and an IS). Inside such an envelope, any trading trajectory cannot a priori be said to be more efficient than another from a risk-control perspective.
FIGURE 54: TRADING ENVELOPES AT DIFFERENT RISK LEVELS FOR A VWAP (LEFT) OR AN IS (RIGHT)
Source: Crédit Agricole Cheuvreux Quantitative Research
The next generation of trading algorithms is made up of trading envelopes, execution constraints and liquidity-seeking tactics
Inside a trading envelope, any combination of liquidity capturing tactics can be plugged in without harming the optimality of the risk control layer (that will be called a “strategic” layer). Now that equations to build optimal trading envelopes are widely known (see, for instance, “Optimal control of trading algorithms: a general impulse control approach” by Bruno Bouchard, Ngoc-Minh Dang and Charles-Albert Lehalle in SIAM J. Financial Mathematics [2011] for more technical details), it is also possible to inject any volume or price-driven constraints, if needed, into the way they are built. Customisation of a trading algorithm thus follows a two-step process: 1. Understanding the desired risk profile and injecting the needed constraints into the trading envelope building process. 2. Understanding the mix of liquidity adapted to the investment style and plugging the associated tactics into the envelopes with respect to properly-defined market conditions (to implement behaviours such as “go into Dark Pools if volatility increases”).
46
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
At Crédit Agricole Cheuvreux, we have a generic way to obtain optimal trading envelopes thanks to our contributions to academic advances in the process of solving optimal trading equations (“Optimal Execution with Limit Orders” by O. Guéant, C. A. Lehalle and J. Fernandez-Tapia [2011] and “Rigorous Strategic Trading: Balanced Portfolio and MeanReversion” by Charles-Albert Lehalle in The Journal of Trading, Vol. 4, No. 3. [2009], pp. 40-46). In terms of liquidity capture, we have customisation capabilities built around a pre-determined set of liquidity-capturing tactics, one of them being versatile enough to be offered as a generic liquidity seeking algorithm called CrossFire™.
CrossFire™ is CA Cheuvreux’s flagship liquidity- seeking algorithm
Q
Liquidity seeking: fine-tuning between price and speed
Liquidity seeking is a delicate process, especially when it involves using liquidity provided by Dark Pools. Pools that provide on average a small fraction of the liquidity traded on the whole market (e.g. Dark Pools) often provide liquidity in bursts. FIGURE 55 shows how trades are clustered on a large cap French stock on Chi-Delta (the dark mid-point operated by Chi-X in Europe) during one day. Even if on average it is possible to find 1% of the traded liquidity in this dark book, it is not the available on a regular basis, 10 shares are not traded on Chi-Delta as soon as 1,000 are traded on the Lit markets, 0 shares are traded for long, and sometimes in 5 to 20 trades, a lot more than 1% of what is traded on Lit books is done on Chi-Delta. Because of this, the statistics to use must be real-time ones: static heat maps are of no use for such smart routing tasks.
"Small" pools do not provide liquidity on a regular basis; static heat maps are of no use to liquidity seekers in such a context
Volume traded on Chi-Δ
Mid-Price on Euronext Paris
FIGURE 55: CLUSTERING OF ORDER ARRIVALS IN A DARK POOL
France Telecom, 22/09/2011. 11.6 11.55 11.5 11.45 11.4 11.35 11.3 09 :00
09 :42
10 :25
11 :08
11 :51
12 :34
13 :17
14 :00
14 :43
15 :26
16 :09
16 :52
17 :35
09 :42
10 :25
11 :08
11 :51
12 :34
13 :17
14 :00
14 :43
15 :26
16 :09
16 :52
17 :35
4000 3500 3000 2500 2000 1500 1000 500 0 09
:00
Source: Crédit Agricole Cheuvreux Quantitative Research
In such a context, it is crucial to build on-the-fly estimates of an "expected fill rate" that each venue can provide. Statistical learning is a field of applied mathematics dedicated to building such self-adapting estimates: use recent data to infer what amount of shares are currently offered for trading in a pool, and adjust a "forget rate" to let this estimate decrease over time when time lasts. FIGURE 56 compares the CA Cheuvreux estimate with the real data on one day. 47
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
FIGURE 56: PERFORMANCE OF THE INTENSITY ESTIMATE 4
6
x 10
Estimated Intensity vs Real Flow (France Telecom on Chi-Δ, 23/09/2011)
Realised (cumulated) Flow
5
4
3
2
1
0 0
1
2
3
4
Estimated Intensity
5
6
7 4
x 10
Source: Crédit Agricole Cheuvreux Quantitative Research
Using such estimates, it is possible to allocate a quantity to trade across different trading venues. FIGURE 57 gives typical "fill rate curves": on the x-axis, the distance to the best opposite (zero means that the order is aggressive, obtaining a fill immediately), and on the y-axis the "expected fill rate" (named in applied maths the "intensity of the trading point process"). Each line decreases when the distance to the best opposite (best bid for a sell order, best ask for a buy order) increases (to the right of the x-axis) and each line is different because at a given time, the liquidity provided by each pool at different depths varies. It is therefore possible to allocate fractions of an order on these lines: Q To obtain price improvement (horizontal line on the figure), allocate more quantity to the pool, providing the same fill rate as others but at a better price;
To obtain fast completion (vertical line on the figure), allocate more quantity to the pool, providing more fill rates at the same price. Q
Using on-the-fly estimates of fill rates, optimal allocations for an order from “price improvement” to “fast completion” can be obtained
From a practical standpoint, the fill rates have to be updated on the fly, taking into account the fill rates obtained and any publicly-available market data: the lines are therefore moving in real time and the allocation has to be reassessed continuously. To preserve the rank of the existing orders in the queue, the reassessment rate must not be too high. Any criterion more complex than "price improvement only" or "fast completion only" can be optimised following, for instance, the stochastic algorithmic scheme for Dark Pool splitting studied from an academic viewpoint in “Optimal split of orders across liquidity pools: a stochastic algorithm approach” by Gilles Pagès, Sophie Laruelle and CharlesAlbert Lehalle (2010), to be published in SIAM Journal on Financial Mathematics. CA Cheuvreux’s CrossFire™ implements such an optimised scheme to capture Dark Pool liquidity optimally. Moreover, such estimates are available on traders' trading stations, to give them more visibility on what is available in the Dark.
48
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
FIGURE 57: OPTIMAL ALLOCATION USING ESTIMATED FILL RATES
Source: Crédit Agricole Cheuvreux Quantitative Research
49
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
50
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
Appendices Q
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
This is the electronic service that provides last sale and trade data for issues admitted on US exchanges, consolidating all markets and driving the trade-through rule.
Trade-through 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.
MiFID
The Markets in Financial Instruments Directive, applied since November 2007, is part of the European Commissionâ&#x20AC;&#x2122;s drive to improve competition among European financial markets. This Directive was revised during 2011 and led to a draft of "MiFID 2", which will come up for a vote in summer 2012.
Large in Scale (LIS)
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.
ESMA
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.
CESR
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.
SEC
The US Securities and Exchange Commissionâ&#x20AC;&#x2122;s mission is to protect investors, maintain fair, orderly and efficient markets, and facilitate capital formation.
51
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
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
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.
Multilateral trading facility (MTF)
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.
Electronic communication network (ECN)
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.
Lit pool
A trading destination with a visible order book.
Dark pool (mid-points)
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.
Dark pools (integrated books)
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.
Smart order router
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.
Broker crossing network (BCN)
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.
Matching engine
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.
Co-hosting
Serving as a join host for several trading institutions' computers in order to reduce latency.
52
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
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
Measurement of the uncertainty in the trend of the price.
Price impact
Impact of the volume of an aggressive order on the obtained average price.
Market impact
Possibly persistent impact of the volume of an aggressive order on the market price.
Opportunity risk
Price of missing a transaction or obtaining a deteriorated price by not having placed an order on the adequate trading destination.
Tick Size
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
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.
Limit Order Book (LOB)
This is a record of unexecuted limit orders maintained by trading destinations.
Best Bid and Offer (BBO)
The highest bid or lowest offer price available in a market at a specific time.
Market share
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).
Average daily number of trades
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.
Average trade size (ATS)
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 Adverse selection
This is the proportion of the day during which the venue offers the greatest sizes at a spread equal to the EBBO. 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).
53
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
54
www.cheuvreux.com
January 2012
Global Research Navigating Liquidity 6
Q
Appendix 2: 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, that offers clients quality execution with two of the worldâ&#x20AC;&#x2122;s leading agency brokers on 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.
55
www.cheuvreux.com
RESEARCH & DISTRIBUTION CENTRES
DISTRIBUTION CENTRES
BENELUX
SPAIN
JAPAN
CRÉDIT AGRICOLE CHEUVREUX – AMSTERDAM BRANCH JOHANNES VERMEERSTRAAT 9 1071 DK AMSTERDAM TEL. : +31 20 573 06 66 — FAX : +31 20 573 06 90
CRÉDIT AGRICOLE CHEUVREUX ESPAÑA S.V. S.A. PASEO DE LA CASTELLANA 1 28046 MADRID TEL: +34 91 495 16 48 — FAX: +34 91 495 16 60
FRANCE
SWEDEN
CRÉDIT AGRICOLE CHEUVREUX S.A. 9, QUAI PAUL DOUMER 92400 COURBEVOIE TEL: +33 1 41 89 70 00 — FAX: +33 1 41 89 70 05
CRÉDIT AGRICOLE CHEUVREUX NORDIC AB REGERINGSGATAN 38 10393 STOCKHOLM TEL: +468 723 5100 — FAX: +468 723 5101
CHEUVREUX CRÉDIT AGRICOLE SECURITIES ASIA B.V., TOKYO BRANCH SHIODOME SUMITOMO BUILDING, 15TH FLOOR 1-9-2 HIGASHI-SHIMBASHI MINATO-KU TOKYO 105-0021 TEL: +81 3 4580 8522 — FAX: +81 3 4580 5534
GERMANY
SWITZERLAND
CRÉDIT AGRICOLE CHEUVREUX – FRANKFURT BRANCH TAUNUSANLAGE 14 D-60325 FRANKFURT AM MAIN TEL: +49 69 47 897 100 — FAX: +49 69 47 897 530
CRÉDIT AGRICOLE CHEUVREUX – ZURICH BRANCH BAHNHOFSTRASSE 18 8001 ZURICH TEL: +41 44 218 17 17 — FAX: +41 44 218 17 87
GREECE
TURKEY
CRÉDIT AGRICOLE CHEUVREUX - ATHENS BRANCH 1 KORAI STREET (3RD FLOOR) 10564 ATHENS TEL : +30 210 373 4000 — FAX: +30 210 373 4001
CRÉDIT AGRICOLE CHEUVREUX MENKUL DEGERLER A.S. BUYUKDERE CAD. YAPI KREDI PLAZA C BLOK KAT:15 LEVENT 34330 - ISTANBUL TEL: +90 212 371 19 00 — FAX: +90 212 371 19 01
ITALY
UNITED KINGDOM
CRÉDIT AGRICOLE CHEUVREUX – MILAN BRANCH VIA BRERA 21 20121 MILAN TEL: +39 02 80 62 83 00 — FAX: +39 02 86 46 15 70
CRÉDIT AGRICOLE CHEUVREUX INTERNATIONAL LIMITED 12TH FLOOR MOORHOUSE - 120 LONDON WALL LONDON EC2Y 5ET TEL: +44 207 621 5100 — FAX: +44 207 621 5101
UNITED STATES CRÉDIT AGRICOLE CHEUVREUX NORTH AMERICA, INC. BOSTON
99 SUMMER STREET, SUITE 220 BOSTON, MA 02110 TEL: +1 (617) 295 01009 NEW YORK
1301 AVENUE OF THE AMERICAS 15TH FLOOR NEW YORK, NY 10019 TEL: +1 (212) 492 8800 — FAX: +1 (212) 492 8801 SAN FRANCISCO
50 CALIFORNIA STREET, SUITE 860 SAN FRANCISCO, CA 94111 TEL: +1 (415) 255 9802 — FAX: +1 (415) 956 9940
Copyright © Crédit Agricole Cheuvreux, 2012. All rights reserved All prices are those current at the end of the previous trading session unless otherwise indicated. Prices are sourced from local exchanges via ThomsonReuters or Bloomberg unless otherwise indicated. Data is sourced from CA Cheuvreux and subject companies. This research report or summary ("Research") has been prepared by Crédit Agricole Cheuvreux or one of its affiliates or branches (collectively “CA Cheuvreux”) from information believed to be reliable. The opinions and projections expressed in this document are entirely those of CA Cheuvreux and are given as part of its normal research activity. CA Cheuvreux has not independently verified the information given in this document. Accordingly, no representation, guarantee or warranty, express or implied, is made as to the accuracy, completeness, correctness or fairness of this information and opinions contained in this document or the research or analysis upon which such information and opinions are based. Any opinions or estimates expressed herein reflect the judgment of CA Cheuvreux as of the date the Research was prepared and are subject to change at any time without notice. Unless otherwise stated, the information or opinions presented, or the research or analysis upon which they are based, are updated as necessary and at least annually. Not all investment strategies are appropriate at all times, and past performance is not necessarily a guide to future performance. CA Cheuvreux recommends that independent advice should be sought, and that investors should make their own independent decisions as to whether an investment or instrument is proper or appropriate based on their own individual judgment, their risk-tolerance, and after consulting their own investment advisers. CA Cheuvreux, its parent companies or its affiliates may effect transactions in the securities described herein for their own account or for the account of others, may have positions with the issuer thereof, or any of its affiliates, or may perform or seek to perform securities, investment banking or other services for such issuer or its affiliates. The organisational and administrative arrangements established by CA Cheuvreux for the management of conflicts of interest with respect to investment research are consistent with rules, regulations or codes applicable to the securities industry. These arrangements can be found in CA Cheuvreux’s policy for managing conflicts of interest, available at www.cheuvreux.com. Current research disclosures regarding companies mentioned in this Research are also available at www.cheuvreux.com. This Research is provided for information purposes only. It is not intended as an offer, invitation or solicitation to buy or sell any of the securities described or discussed herein and is intended for use only by those Professional Clients to whom it is made available by CA Cheuvreux. This Research is not for distribution to Retail Clients. The recipient acknowledges that, to the extent permitted by applicable securities laws and regulations, CA Cheuvreux disclaims all liability for providing this Research, and accepts no liability whatsoever for any direct, indirect or consequential loss arising from the use of this document or its contents. CA CHEUVREUX RESEARCH AND DISTRIBUTION 1. IN THE UNITED STATES, CREDIT AGRICOLE CHEUVREUX NORTH AMERICA, INC. (“CAC NORTH AMERICA”) SPECIFICALLY ADVISES THAT IT DID NOT PREPARE, HAS NOT CONTRIBUTED TO, HAS NOT ANALYZED, AND DOES NOT ENDORSE THIS RESEARCH. CAC North America is a SEC-registered broker-dealer, and is not an investment adviser. CAC North America does not manage assets of other entities, CAC North America does not provide investment advice, and CAC North America neither issues nor promulgates reports or analyses within the meaning of Section 202(a)(11) of the Investment Company Act of 1940, as amended. CAC North America is unable to provide any additional information of any sort regarding this Research, and can neither support or refute any of the content, opinions, estimates, or conclusions contained in the Research. CAC North America further advises that this Research is solely intended to be delivered to customers of CAC North America who qualify as “Major U.S. institutional investors” as defined in Rule 15a-6 of the Securities and Exchange Act of 1934, as amended (“CAC North America Customers”). CAC North America Customers are restricted from re-delivering the Research to any other entity, and shall be held strictly liable for any and all costs, legal fees, damages, fines, or penalties resulting from any re-delivery of the Research to persons or entities other than CAC North America Customers. The existence of this Research, or CAC North America’s forwarding the Research to certain of its customers, shall not be deemed a recommendation or endorsement by CAC North America of the Research, a recommendation to effect any transactions in the securities discussed herein, or an endorsement of any opinion expressed in the Research. 2. In the United Kingdom, this report is approved and/or distributed by Crédit Agricole Cheuvreux International Ltd – which is authorised by the Financial Services Authority (“FSA”). 3. In Italy, this Research is approved and/or distributed by Crédit Agricole Cheuvreux SA and is not intended for circulation or distribution either to the public at large or to any other parties other than professional clients and/or qualified counterparties, as defined in Legislative Decree No.58 of 24 February 1998 as amended, and implementing Consob regulations. 4. In Turkey, this document is based on information available to the public. CA Cheuvreux Menkul Degerler A.S. has spent reasonable care in verifying the accuracy and completeness of the information presented within the document and cannot be held responsible for any errors, omissions or for consequences arising from the use of this information. The information should not be construed, implicitly or explicitly, as constituting investment advice. This document is not an offer to buy or sell any of the securities discussed herein and has been prepared for the qualified representatives of our institutional investors. 5. In Germany this report is approved and/or distributed by Crédit Agricole Cheuvreux S.A. Niederlassung Deutschland (authorised and regulated by Bundesanstalt für Finanzdienstleistungsaufsicht (BaFin)). THE DISTRIBUTION OF THIS DOCUMENT BY CA CHEUVREUX IN OTHER JURISDICTIONS MAY BE RESTRICTED BY LAW, AND PERSONS INTO WHOSE POSSESSION THIS DOCUMENT COMES SHOULD INFORM THEMSELVES ABOUT, AND OBSERVE, ANY SUCH RESTRICTIONS. BY ACCEPTING THIS REPORT YOU AGREE TO BE BOUND BY THE FOREGOING INSTRUCTIONS. CLSA DISTRIBUTION 1. In the United States this report is distributed by CLSA solely to persons who qualify as “Major Institutional Investors” (as defined in Rule 15a-6 under the U.S. Exchange Act of 1934 as may be amended from time to time) and who deal with Credit Agricole Corporate & Investment Bank. 2. In the United Kingdom, this report is distributed by CLSA (UK) (which is authorised by the Financial Services Authority (“FSA”). 3. In Australia this publication/ communication is distributed by CLSA Australia Pty Ltd; 4. In Hong Kong this publication/ communication is distributed by CLSA Research Ltd; 5. In India this publication/ communication is distributed by CLSA Ltd. (Address: 8/F, Dalamal House, Nairman Point, Mumbai 400021. Tel No: +91-22-66505050. SEBI Registration No: BSE Capital Market Segment: INBO10826432; BSE F&O Segment: INFO10826432; NSE Capital Market Segment: IN230826436; NSE F&O Segment: INF230826436); 6. In Indonesia this publication/ communication is distributed by PT CLSA Indonesia; 7. In Japan this publication/ communication is distributed by Credit Agricole Securities Asia B.V. Tokyo Branch, a member of the JSDA licensed to use the “CLSA” logo in Japan; 8. In Korea this publication/ communication is distributed by CLSA Securities Korea Ltd; 9. In Malaysia this publication/ communication is distributed by CLSA Securities Sdn Bhd; 10. In the Philippines this publication/ communication is distributed by CLSA Philippines Inc. (a member of Philippine Stock Exchange and Securities Investors Protection Fund); 11. In Thailand this publication/ communication is distributed by CLSA Securities (Thailand) Limited; 12. In Taiwan this publication/ communication is distributed by CLSA Limited, Taipei Branch. 13. In Singapore this publication/ communication is distributed through CLSA Singapore Pte Ltd solely to persons who qualify as Institutional, Accredited and Expert Investors only, as defined in s.4A(1) of the Securities and Futures Act, Pursuant to Paragraphs 33, 34, 35 and 36 of the Financial Advisers (Amendment) Regulations 2005 with regards to an Accredited Investor, Expert Investor or Overseas Investor, sections 25, 27 and 36 of the Financial Adviser Act shall not apply to CLSA Singapore Pte Ltd. MICA (P) 168/12/2009 THE DISTRIBUTION OF THIS DOCUMENT BY CLSA IN OTHER JURISDICTIONS MAY BE RESTRICTED BY LAW, AND PERSONS INTO WHOSE POSSESSION THIS DOCUMENT COMES SHOULD INFORM THEMSELVES ABOUT, AND OBSERVE, ANY SUCH RESTRICTIONS. BY ACCEPTING THIS REPORT YOU AGREE TO BE BOUND BY THE FOREGOING INSTRUCTIONS. The analysts/contributors to this publication/ communication may be employed by a Credit Agricole/ CA Cheuvreux company which is different from the entity that distributes the publication/ communication in the respective jurisdictions. CLSA has not provided any contribution nor made any modifications to this publication. The MSCI indexes are the exclusive property of MSCI Inc. (“MSCI”). MSCI and the MSCI index names are service mark(s) of MSCI or its affiliates and have been licensed for use for certain purposes by CA CHEUVREUX. The financial securities referred to herein are not sponsored, endorsed, or promoted by MSCI, and MSCI bears no liability with respect to any such financial securities. The Prospectus contains a more detailed description of the limited relationship MSCI has with CA CHEUVREUX and any related financial securities. No purchaser, seller or holder of this product, or any other person or entity, should use or refer to any MSCI trade name, trademark or service mark to sponsor, endorse, market or promote this product without first contacting MSCI to determine whether MSCI’s permission is required. Under no circumstances may any person or entity claim any affiliation with MSCI without the prior written permission of MSCI. The Dow Jones GCC IndexSM, or other applicable index, are calculated, distributed and marketed by Dow Jones & Company, Inc. , a licensed trademark of CME Group Index Services LLC ("CME"), and have been licensed for use. All content of the Dow Jones IndexesSM © 2012 Dow Jones & Company, Inc. No part of this report may be reproduced in any manner or redistributed without the prior written permission of CA Cheuvreux.
Signatory of the Principles for Responsible Investment
www.cheuvreux.com