Valdi White PAPER SERIES Algorithmic trading: a complex map By Benjamin Becar, Product Manager, Valdi Algorithms, SunGard s global trading business
table of contents 3 Introduction 3 Automated trading worlds 3 Automated and algorithmic trading 5 Where is algo trading today, and where is it going? 8 More trends... 9 Who is playing? 9 The technology 10 SunGard s response: embrace the automation of trading 11 Glossary
Algorithmic Trading 3 Introduction Algorithmic trading is a global phenomenon, but the subject is a complex one, and there are major differences of market maturity in different parts of the world. In this White Paper we try to bring some context and clarity to a description of the state of the art. We consider first the issue of definition, and describe the various types of algorithm that exist: decision-making, execution and execution plus. We then look in more detail at execution algorithms, and how their usage is developing across different regions and asset classes. Finally, we discuss the various technical approaches that can be taken by a brokerage firm to establish or extend its algorithmic trading capabilities. Automated trading worlds We all try to understand the big picture when getting to grips with a financial market topic, but with automated trading the picture s multi-faceted shape presents us with a challenge. And within that big picture, defining precisely what algorithmic trading is (and is not) involves further complexity: understanding of this term depends heavily on who you are talking to, and on where in the world they are located. For big-picture purposes, automated trading can be defined as any automated action on an order, which may take place at any stage of the trade execution process: order creation, sending, modification or matching. There are then a number of different automated trading worlds to be considered. The challenges that participants face in different areas of the markets are widely varied, and deeply impacted by a range of factors. Automated and algorithmic trading Few terms in financial markets cause more confusion than the apparently simple label algorithmic trading. It may be applied by different people to quite diverse areas: anything from VWAP algorithms to index arbitrage, taking in smart routing, matching engines, basket trading and more along the way. Names often don t help much: market Hide & Pounce, VWAP, statistical arbitrage, ticker tape trading Fortunately, while there are many imaginatively titled algorithmic strategies, these can be grouped into a small number of categories, with more or less well defined boundaries. We consider the major ones in the two following sections. Note: You can also find a short glossary of the main relevant terms at the end of this document. Decision-making algorithms Let s look first at the major types of automated trading that, in the strictly purist view, are not truly algorithmic trading. All share the basic characteristic that they apply some combination of market data and algorithmically coded judgment to make fundamental decisions about what to buy or sell, in what quantity and at what price: hence decision-making algorithms. They are also sometimes referred to as systematic algorithms. The decisions are often triggered by prices in the market data stream breaking a limit, or a combination of limits. The algorithms applied can be highly complex and individual, and have high potential return on investment. They are usually built and/or customized by the institution that uses them, and their details are always highly confidential. The quantitative analysts ( quants ) who create these algos usually have strong math/econ academic backgrounds, and their winning strategies may be deployed and refined over significant periods of time.
4 SunGard Valdi Major categories of decision-making algorithms include: Market Making By constantly buying and selling at appropriate model-determined Bid and Offer prices, market makers make money in both rising and falling markets by repeatedly earning the Bid/Offer spread. Index Arbitrage Traders profit from (usually short-lived) discrepancies between index prices and the prices of the underlying constituents. Pairs Trading In this case, profit arises from negative correlation between the prices of two related instruments over a period. The technique is most often applied to equities, where (regardless of overall market direction) the trader expects one of the pair to outperform the other. Statistical Arbitrage Strategies here can be very complex. Again, a correlation is identified, but now it may involve hundreds of instruments, and profit is derived from discrepancies, usually within a short period. Index Arbitrage and Pairs Trading are both simple examples of this broader technique. Technical Analysis Another form of statistical analysis of the market, applied to patterns on time series price and volume charts. Moving Averages and Bollinger Bands are typical mathematical tools that can help identify patterns and trends. Automated gaming Frowned on in some quarters, gaming strategies try to recognize other market participants strategies and then game them by anticipating their further behavior. A simple strategy might look for a broker repeatedly (at equal intervals) sending the same quantity on the same instrument, indicating that a large quantity is being worked by an execution algorithm. The gamer can now put pressure on the price on that venue, while looking for a better price on another venue (perhaps a dark pool) and aiming to take the difference. High Frequency Trading While several of the strategies listed above can be used by High Frequency Traders, there are also strategies which are more HFT-specific, such as Ticker Tape Trading (monitoring a flow of quotes to identify information not yet published in the news, and trading based on this advance knowledge). Execution algorithms This category of algorithms is less controversial in its definition everyone can agree that this is the heart of algo trading. Basic execution algorithms, by contrast with decision-making algos, are often fairly commoditized, and most brokers offer customized variants to their clients. Trading via execution algos can be defined as working a large order so as to minimize market impact, for instance when an institutional investor asks a broker to buy 1,000,000 HSBC shares at the best price over the day. Where once a broker would use personal knowledge of the market to decide how to distribute that order over the appropriate time period, today an algorithm will often make the decisions and automatically release the child (component) orders. Execution algorithms are used by almost all brokers, and by many of their clients who have direct market access. There are three main provider groups: the major global brokers, software vendors (some being more specialized than others) and exchanges (usually for the simpler algos strictly speaking, at this level we are closer to synthetic order types than algorithms offered directly on the exchange trading platform). A typical mid-scale agency brokerage firm may use all three in different parts of its business. There is intense competition between and within each of the provider groups for client business, which has sharpened further as markets have matured in the major financial centers. The complexity of execution algos varies greatly. At the simple end one of the most frequently used is the Iceberg, which slices an order into several smaller components: Icebergs are offered by many exchanges, the arriving order being held in a buffer and the components released progressively onto the exchange s order book. Other simple execution strategies, such as Stop orders (for covering positions), are frequently used by online traders. At the next level we find benchmark algorithms such as VWAP (Volume Weighted Average Price), which use historical market volume profiles to determine more sophisticated sending patterns for component orders. This approach can of course further reduce market impact, and the quality of the resulting executions can be measured against the market profile (the benchmark) for the current day. Clearly, the larger the order, the more worthwhile it becomes to employ techniques of this type. Beyond VWAP and other participation algos such as Target Percentage of Volume (% Vol), the sky is almost literally the limit. The concept of Implementation Shortfall, defined as the difference between the decision (or
Algorithmic Trading 5 arrival) price and the final execution price for a trade, has given its name to a range of algos whose purpose is to minimize this shortfall (i.e. the market impact of working the trade). Techniques may include dynamic decisions on whether to make or take liquidity, considered at each stage during the working of an order. This is the ground on which the major brokers compete with one another to design complex customizations (often from a VWAP, Percentage of Volume or Implementation Shortfall base) that aim to beat one or more benchmarks. And still there is more. Execution plus High frequency trading, considered above with other decision-making algos, can be defined as using advanced technology to trade quickly and repeatedly for a profit : a typical HFT strategy might involve sending millions of quotes to an exchange, but getting executions only on 1% or less. The multiple order sending frequently causes HFT to be seen as the new algorithmic trading : worryingly, regulators are particularly prone to this. But in reality, while HFT has certain characteristics in common with conventional execution algorithmic trading, it is important to recognize that the algorithms in use have no connection whatsoever to those that are used to work a traditional fund manager s million-share order over time. One might add that it is particularly important to recognize this when seeking to regulate HFT and algorithmic trading. Smart routing, on the other hand, which is often considered as a trading technology in its own right, can quite properly be considered as a special category of execution algorithm, intended to obtain the best price(s) for an order by selecting the best trading venue, from among a competing set of such venues, on which to execute it. The execution will usually take place in multiple tranches (which may therefore of course be executed on several different venues): smart routing could indeed be defined as trading algorithmically across multiple venues. Here, the rules that drive venue selection and order sending constitute the execution algorithm, and they may take into account a range of factors, such as times of day, prices, available sizes and fee costs. Smart routing is taken a stage further in complexity when dark pools are among the potential trading venues. Prices and sizes being by definition invisible in this case, the routing algorithm has to include liquidity seeking as part of its strategy sending small orders to selected dark pools in order to determine where larger trades may be executed at favorable prices. Finally, combined algorithmic trading strategies are increasingly frequently deployed: the million-share order may be divided according to a simple Time-Weighted (TWAP) or more complex Implementation Shortfall algorithm, and the child orders then passed to the smart router for execution according to its rule set. Each of those children may then be further treated, and potentially subdivided, by creating from them synthetic orders, such as Icebergs. Where is algo trading today, and where is it going? As with the definition question, there are a variety of answers here; but the level of maturity achieved in algorithmic trading depends much less on who you are than on where you are geographically. Looking first at equities, the asset class in which algo trading is most widely used, let s consider some regional perspectives. Algo trading in equities USA With about 50 trading venues exchanges, displayed Alternative Trading Systems and dark pools and relying on high-speed networks for effective implementation of the find best price obligations imposed by Regulation NMS, US markets are the world s most competitive and most advanced. With the advantage also of a unified clearing network, complex trading strategies can readily be applied. It is very common (almost mandatory now) for market players to employ smart routing algorithms that take into account many parameters: discrepancy between venues in terms of latency, price, level of grayness of various dark pools, and the price maker/price taker model employed by many ATSs. Upstream of the smart routers, execution algorithms are also widely used: Aite Group estimates that 60% of US equity market volumes in 2010 emanated from execution algos [Fig. 1], and this percentage is still growing steadily (a short-lived dip appeared in 2010 following the Flash Crash, but algo trading is now on the rise again). Among the most popular algorithms in US equity markets are
6 SunGard Valdi implementation shortfall and benchmark strategies (mainly VWAP and Participation) [Fig. 2]. A substantial number of investors use liquidity seeking strategies across multiple dark pools, often launched from portfolio trading strategies. The high frequency trading (HFT) phenomenon is also most firmly established in the US equity markets. TABB Group has estimated (April 2011) that HFT accounts for some 54% of trades 1 (with growth at last slowing). Clearly, the US equity market can be seen as an already largely algorithmic game. 70% 60% 50% 40% 30% 20% 10% 0% 2004 2005 2006 2007 2008 2009 e2010 Figure 1: Estimated Algorithmic Trading Adoption Source: The US Electronic Trading Market 2010, Aite Group, November 2010 48.8% 41.7% VWAP TWAP POV 9.7% Figure 2: Participation algorithms in use Source: THE TRADE Algorithmic Trading Survey, March 2011 Algo trading in equities Europe In Europe, four years after the first MiFID Directive the clearing landscape is complex, with interoperability still only at an early stage of development. Trading costs have fallen, due to the competitive pressure placed on the incumbent exchanges by the new Multilateral Trading Facilities (MTFs), but clearing and market data costs remain high. One consequence is that high frequency trading is less prevalent than in the US, and there is also impact on the volume of algo trading (~35% of the market in 2011, as estimated by TABB Group [Fig. 3]. 1 Larry Tabb, TradeTech London 12 April 2011
Algorithmic Trading 7 Sales Desk 35% 39% 2010 2011e Program Desk *Direct To Exchange 10% 8% 9% 9% Algorithms 29% 35% Crossing Network 12% 13% Figure 3: Shares Traded, by Execution Type Source: US Equity trading 2010: low touch trend, TABB Group, July 2010 But market structure has been changing quite rapidly, and continues to do so. The major MTFs have taken significant market shares, Chi-X being now the #1 European trading venue in terms of volume traded [Fig. 4], and broker dark pools and crossing networks are also well established. A good smart routing strategy is therefore essential to achievement of true Best Execution, as required under the MiFID rules. Smart routing is still not used by all market players, however: many smaller brokers continue to trade only on the main exchanges i.e. using only one venue for each listed instrument. As competitive and regulatory pressure to deliver true Best Execution increases, this will promote the further growth of execution algos in general, and of smart routing in particular. 180000 160000 140000 120000 100000 80000 60000 40000 20000 0 BME Spanish Exchanges Deutsche Börse NYSE Euronext (Europe) London SE NASDAQ OMX Nordix Exchange BATS Europe Chi-X Europe Turquoise Oslo Børs SIX Swiss Exchange Figure 4: Major European Venues: Total Equities Volumes, January-June 2011 Source: SunGard- WFE- FESE
8 SunGard Valdi Algo trading in equities Asia-Pacific APAC is a complex world in itself. There are many markets and many currencies, and numerous exchanges, including those of Hong Kong and Singapore, still hold monopoly positions in their respective territories. Meanwhile, following the trend set earlier in the US and Europe, there is currently a high pace of growth for automated trading. All of this creates a challenging environment for market participants: new entrants must cope with multiple local regulations and trading practices, while the local players have to enhance their systems to compete with the Tier-1 firms that are growing their regional franchises. The relative immaturity of electronic markets in the APAC region means that misunderstandings about the definition of algorithmic trading are especially likely here. In some markets (China, India, and also Singapore, for instance) brokers and buy-side firms still ask Why would I need algorithms?. This has resulted in good opportunities across the region for the major Western brokerage firms, who brought in their algorithms to profit from market inefficiencies, and now offer those algos to local brokers, as well as to investors. The main challenge for these global players resides in the variety of market specificities they have to handle: order types, trading hours, trading patterns, market regulations, and constant change in the landscape: Smart Order Routing in India became possible in 2010, alternative venues are on the rise in Japan and Australia, and financial derivatives markets are taking off in China. The level of adoption of algorithms varies widely across the region, as illustrated in the chart below from Celent [Fig. 5]. Some locally based experts question these figures: the major part of algo volume currently comes from global tier-1 brokers, with most local players trading still being manual. The survey sample may have been biased towards the tier-1 firms. According to the Singapore Exchange, algorithmic trading in 2010 accounted for some 25% of volume in derivatives, and only about 5% in the cash equity market. 70% 60% 50% 40% 30% 20% 10% 0 Singapore Hong Kong Japan Australia India 2008 2009 2010 2011 2012 Figure 5: Levels of Algorithmic Trading in Asia s Leading Markets Source: Electronic Trading in Asia-Pacific: A Market by Market Update, Celent, October 2010 More trends Despite varying levels of algo trading usage globally and all the local specificities discussed above, some trends are very clear. The continued growth and global spread of high frequency trading is one. Usage of alternative venues and dark pools is also rising globally, so smart routing and liquidity seeking algos are becoming progressively more important. Also, given the increasingly complex range of available possibilities across technology, algorithms, venues and regulatory regimes and the increased volatility of the markets, pre-trade analytics (Transaction Cost Research) on the algo strategy options is an increasingly important field. The spread of the use of algorithmic trading from its equity-market roots is also very marked as noted above, on the Singapore Exchange the use of algos on the derivatives market far outstrips that on equities. TABB Group estimates, as another example, that over 60% of order flow in the US options market is now routed through DMA and execution algos. Increasingly, brokers differentiate themselves in their marketing via the breadth and innovation offered in their algo suites for derivatives: these often include traditional equity-market staples TWAP, VWAP etc. as well as derivatives-specific exotics.
Algorithmic Trading 9 In parallel, as high-volume foreign exchange trading becomes increasingly concentrated on the major electronic trading platforms and ECNs, the use of execution algos is also growing, with the leading banks typically offering FX algo suites that include both time slicing and liquidity seeking capabilities. Who is playing? Across the algo trading field, competition is fierce: the tier-1 brokers and electronic execution specialist firms (such as SunGard s Fox River and Valdi Liquidity Solutions) continue to invest heavily in their algorithmic trading platforms, which is making it progressively harder for others to compete. And the struggle for differentiation is fiercer than ever. This drives frequent launches of new and aggressive-sounding algorithms, with references to warfare, carnivores and birds of prey being the usual naming conventions! But it is often hard to get the message across about the subtle (even if significant) differences that these new introductions can make. We often hear the question What does this algo do anyway? There is a strong market tendency to rely on the algorithms of the major tier-1 players: it s easy and (relatively) cheap. But brokers who go down this route have to ask themselves what impact it may have over time on their differentiation, as algos become a more and more important part of what they offer to their clients as executing brokers. Some advertise the fact that they offer access to Broker X s algos, alongside their own, but others are coyer about this, worrying about the impact on relationships should clients become fully aware that they have outsourced their algo offers. Building algorithms in house is an option that is considered by many, but in practice realized by relatively few. The investment in terms of knowledge, training and technology is significant: as such, most projects currently are at prop desks and hedge funds, where strategy differentiation is often key to survival. And with some players changing algorithms daily, it requires significant investment to stay ahead of the curve. The stakes can be high, as was illustrated in July 2009 when a former Goldman Sachs employee was accused of stealing an algo worth millions of dollars. It s important to recognize that, as with any innovative technology, the pace of adoption of algorithmic trading is far from uniform. Innovators and early adopters have recognized the potential very rapidly, beginning largely in the US and with the tier-1 brokers then spreading the word internationally. For the laggards, the benefits are still unclear. A frequent challenge faced by technology vendors and leading brokers in less mature markets is the simple question Why do I need algorithmic trading?. We are a long way here from the arms race that s the daily reality in the US markets! The technology There are two main challenges involved in the implementation of algo trading technology. Firstly, integration of the software with existing systems can be complicated. A user firm may have to consider integration with a legacy order management system and trading gateways from provider X, with market data from provider Y, and/or specialized low-latency or time series market data from provider Z. The cutting-edge technology build can quickly turn into an implementation nightmare. Innovative technology providers have addressed this issue: for example StreamBase, the complex event processing platform provider, has built a broad range of adapters (the building blocks that interface with third-party components, including SunGard s algo trading products), hence diminishing the adoption barrier. The second area of challenge is of course performance. And there are numerous parameters to consider: latency (how fast can your order reach the market), throughput (how many orders can exit your engine in a given time period) and capacity (how much data can you process simultaneously). 64-bit systems, multithreading, high-speed networks and FPGAs may all come into play: another nightmare for the CTO!
10 SunGard Valdi SunGard s response: embrace the automation of trading In September 2011, a UK Government Foresight panel forecast that The number of human traders employed in the financial markets is set to fall dramatically over the next ten years as banks and brokers become increasingly reliant on computer-based algorithms to run their trading operations. Clearly, the panel s research led it to conclude that the trends described in this paper are set to continue, and could even accelerate: so there will be more automated trading across the board, and with increasing complexity. Our objective at SunGard has been to provide a modular toolkit a set of building blocks to help with the management of this complexity. Our Valdi Automated Trading suite provides a range of solutions including support for real-time Excel spreadsheet trading from the desktop, program trading, strategy and pairs trading and index arbitrage. We also offer a range of support for algorithmic trading at the execution level, the fit depending on the problems our customers are trying to solve, and on where they want us to assist with the order flow. Brief details follow below. Off-the shelf At the execution level, customers can start with our suite of synthetic order types: Valdi Tactics, which covers 15 strategies (from standard Stop orders or validities, to more complex Trailing Stop, One Cancels the Other, Linked Peg, etc.). For some markets, especially in Asia, customers also use a set of market-specific strategies (automated short selling, simulated order types, etc.). At the next level, we provide a range of trading algorithms: Valdi Algo Trading, including TWAP, VWAP and Percentage of Volume strategies. All of these algorithms can also be used in conjunction with the Valdi Smart Router, for improved execution in European and Asian equity markets. The Smart Router provides a range of routing strategies, including multiple dark pool access, and sub-millisecond routing, Customized Then there are customers for whom out of the box is not enough, but they lack the resources to carry out their own developments. Valdi Algo Services have proved to be a practical solution in many such cases: SunGard s specialist teams, located in financial centers worldwide, can deliver new algo strategy developments as turnkey projects, working to specifications that we develop with the customers concerned. Broker algorithms The SunGard Global Network interconnects more than 2000 buy-side firms with 530+ brokers for order routing to global markets. As part of the network service, the algorithmic trading suites of leading brokers are delivered to buy-side trading workstations via these links. This is achieved using standard FIX containers, meaning that new client connections can be rapidly established and the algos maintained easily over time, as they develop. Partnership with StreamBase Where requirements in terms of volume and latency are critical, SunGard partners with StreamBase, the leader in Complex Event Processing (CEP) technology. With visual development that really works, built-in multithreading and a wide range of adapters, StreamBase is used as a core infrastructure element for many leadingedge trading applications. And with SunGard s range of algos and market connectivity, the underlying set of building blocks necessary for implementation of a fully competitive algo strategy is complete.
Algorithmic Trading 11 GLOSSARY Automated Trading: any automated action on an order Algorithmic Trading: for some people, effectively means the same as Automated Trading; others consider it to mean, much more narrowly, the working of a large order so as to minimize market impact TWAP (Time Weighted Average Price): Algorithm for sending an order in several waves over a designated time period, potentially with a range of parameters designed to ensure as smooth a spread as possible of executions over the period VWAP (Volume Weighted Average Price): Related to TWAP, but each order wave is weighted in size according to a prediction of market volume at the specific time of day for each instrument involved, based on historical volume profile data % Vol (Target Percentage of Volume): Algorithm used to execute an order at a rate corresponding to a target percentage of the overall market volume Implementation Shortfall: the difference between the decision (arrival) price for an order and the final executed price; the term is also applied to describe algorithms that seek to minimize this shortfall High Frequency Trading: using advanced technology to trade quickly and repeatedly for a profit Smart Routing: Executing an order by sending waves to multiple trading venues, based on a defined rule set Liquidity Seeking: Associated with smart routing, where Dark Pools (see below) are involved the smart router sends exploratory orders to these pools in order to determine where sufficient liquidity may be located ATS (Alternative Trading System): A US trading venue which is not a traditional stock exchange. Other names for venues of this type, varying according to the applicable regulations, include: ECN Electronic Communication Network MTF Multilateral Trading Facility (Europe) PTS Proprietary Trading System (Japan) Dark Pool: a trading venue that does not publish bid and offer prices, but only (according to regulations) reports executed trades, potentially after a delay period; may be a regulated trading venue (ATS or other), or may be operated by a broker and regulated as part of his service to clients TCR (Transaction Cost Research): analysis applied to a proposed trading strategy to predict its market impact and hence its overall transaction cost
www.sungard.com/valdi About SunGard s Valdi and Valdi Automated Trading SunGard s Valdi provides equities, futures, fixed income and options traders with multi-asset trading solutions on 170+ markets worldwide. Valdi global trading solutions support the entire trade lifecycle, including integrated trade and order management systems, execution services, market data, risk management, compliance, and clearing and settlement services. Also offering global connectivity via the SunGard Global Network (SGN), SunGard s Valdi helps customers achieve increased performance and low latency execution across multiple platforms, instruments and geographies. Valdi Automated Trading provides a wide range of solutions to support automated and algorithmic trading. In the specific area of execution algos it offers a suite of synthetic orders, standard algorithms, and a set of services to support the development of more complex algos. SunGard also supports delivery of algorithms from brokers to their clients, via the SunGard Global Network. For more information about SunGard s solutions for global trading, please visit www.sungard.com/globaltrading or contact info.globaltrading@sungard.com. About SunGard SunGard is one of the world s leading software and technology services companies. SunGard has more than 20,000 employees and serves 25,000 customers in 70 countries. SunGard provides software and processing solutions for financial services, higher education and the public sector. SunGard also provides disaster recovery services, managed IT services, information availability consulting services and business continuity management software. With annual revenue of about $5 billion, SunGard is ranked 434 on the Fortune 500 and is the largest privately held business software and IT services company. Look for us wherever the mission is critical. For more information, please visit SunGard at www.sungard.com. 2011 SunGard. Trademark information: SunGard, the SunGard logo and Valdi are trademarks or registered trademarks of SunGard Data Systems Inc. or its subsidiaries in the U.S. and other countries. All other trade names are trademarks or registered trademarks of their respective holders.