Title: Algorithimic Trading: The State of Algo Trading
Feature: Trading Technology
Date: 1 March 2006
Algorithmic trading has left the drawing board and has become a valuable tool for traders. Waters asks leading players to sound off on how investment firms can use algorithms to optimize their trading strategies.
Should OMSes time-stamp trades?
Ary Khatchikian, Portware: As the industry becomes more sensitive to analytics and the regulators require more detailed reporting, OMSes and EMSes need to be able to time-stamp all events in a trade's lifecycle. Because a number of execution management systems were originally built to be event-driven, they generally have the ability to time-stamp any piece of data needed by the end user, from order generation to receipt to execution. This leads to a clean audit trail and allows the buy side to view P&L against multiple benchmarks simultaneously. These timestamps and audit trails are the core of the buy side's ability to track algorithmic execution over the long term.
What is the best way for the buy side to differentiate the available algorithmic tools from sell-side brokers?
Khatchikian: The first and most important differentiator is execution quality. Of course, this may also be the most difficult to gauge. The first step toward doing this is for buy-side firms to define their trading objectives and set in place a process that captures all relevant information to that objective—i.e., time-stamps of alpha generation, order generation and so forth. Over a period of time, it then becomes possible to see which algorithms are performing best for your specific objectives. It is distinctly possible that an algorithm that provides one firm excellent execution quality for its objective is not the right choice for another firm. Other factors to consider when evaluating algorithmic execution include the depth of knowledge and experience on a broker's algorithmic teams, as well as pre-trade analytic capabilities, which should be tightly coupled with a firm's EMS.
Besides the OTC equity markets, what is the next asset class that is ripe for the adoption of algo trading?
Khatchikian: Algorithmic trading largely depends on the electronic accessibility of liquidity pools for a given asset class. As these new assets go electronic, there will be opportunity to utilize the backbone of current equities algorithms to execute FX, futures and options orders, as well as fixed income in the not-too-distant future. FX markets have been moving very rapidly in this direction, which is why Portware has introduced a dedicated FX platform. Another thing we're beginning to see a lot more of is applying algorithms not just to a single asset class, but also across assets. For example, a trader may be unwinding his or her futures position, buying the cash but wanting to remain currency-neutral. An algorithm, or conditional trade, is required to make this happen. It's pretty exciting how fast things are moving.
Besides VWAP and arrival price, what other algorithms are being adopted by traders? Why?
Khatchikian: A number of sell-side firms have begun offering "liquidity seeking" algorithms that access a variety of liquidity pools, including ATSes and crossing networks. These algorithms, unlike their predecessors, trade with a less predictable pattern of execution and are being widely adopted as a tool to help in the execution of large blocks of illiquid stock. As these algorithms become more widely accessible, a number of users are starting to use them on large-cap, liquid stocks as well. Non-benchmark-specific algorithms—e.g., participation algorithms, percent of volume and so on—are also being adopted as alternative and conditional, or correlation, algorithms.
Will algorithms be strictly sell-side offerings or will we see OMS providers and execution management providers offering trading algorithms?
Khatchikian: Algorithms are already being offered through multiple channels, not just via the sell side. For instance, in addition to providing access to all broker-supplied algorithms, Portware delivers a suite of out-of-the-box execution algorithms as well as template strategies, which traders can work with to develop their own customized strategies. Clients can also write their own algorithms from scratch. That said, the brokerage community still has the capital to invest and expertise to produce best-of-breed algorithms to offer the buy-side community. One of the more interesting developments we've seen are requests from clients for the sell side to deploy their algorithms to actually live on third-party, broker-neutral systems like Portware's. The advantage to this is that the buy-side trader can utilize the algorithms developed by the experts on the sell side and route them to any destination they'd like, including brokers who didn't originally develop the algorithm. This gives buy-side traders unprecedented control over their orders and execution, as well as the information given up.
What is the best way for the buy side to differentiate the available algorithmic tools from sell-side brokers?
Jarrod Yuster, Merrill Lynch: At Merrill Lynch, we focus on three areas that we believe differentiate our offering: execution performance, service and customization. Ultimately, the reason for using algorithms is to reduce explicit and implicit costs, enhance performance and increase efficiency. Service is an important point of differentiation. Our service desk, comprised of experienced traders, provides education and guidance for our clients as to which algorithms are appropriate based on order characteristics, client preferences and market conditions. The service desk works in partnership with our clients to develop customized algorithms and analytics to cater to our clients' specific trading needs.
Besides VWAP and arrival price, what other algorithms are being adopted by traders? Why?
Yuster: Traders are looking for ways to navigate efficiently between strategies, including VWAP and arrival price, as well as to combine multiple algorithms and aggressiveness levels with both high-touch and no-touch execution methods. As algorithmic usage grows in sophistication, traders are realizing the value of using algorithms in tandem with ECNs, crossing engines and high-touch venues including blocks, IOIs and capital, as market conditions change. Technology developments fueling this diversified approach to algorithms include: strategy amends, or switching between strategies and parameters instantaneously, and pre-trade algorithmic criteria checks and algorithmic choosers that utilize real-time information to automate decision-making among various algorithms and trading venues.
When should a human trader be used instead of an algorithm?
Yuster: Through our research, algorithms are using statistical methods for modeling stock behavior. We use algorithms when they improve the mean and reduce the variance of execution performance versus a given benchmark, as well as when the stock has a relatively high predictability pattern. Comparatively, we will tend to use algorithms less often when there is a lack of predictability, a low level of liquidity, and/or a high amount of information content for the given order.
How are users dealing with the increased need for market data to use algorithms efficiently?
Yuster: At Merrill Lynch, we have a vast amount of data to leverage for execution modeling and analysis. The changes to market structure, as well as the onset of algorithmic trading, have changed the fundamentals of market information, such as increasing trade and quote intensity. These changes will continue to increase the amount of market data and message traffic. As a result, we have created a highly sophisticated, real-time analytical tool that cleanses, aggregates, and transforms raw market data into trading signals for our clients by leveraging our extensive statistical infrastructure.
What are the drawbacks to using algorithms?
Yuster: At Merrill Lynch, our pre-trade algorithmic criteria checks as well as our service desk are geared toward helping clients understand when to use algorithms, how to use them, and when not to use them. Algorithms are not the optimal trading solution for every order. Utilizing pre-trade tools combined with extensive post-trade order flow and performance analysis provides a statistical method for determining the optimal trading approach—an algorithm, sales trader, capital commitment, or DMA.
What is the best way for the buy side to differentiate the available algorithmic tools from sell-side brokers?
Joseph Wald, EdgeTrade: A buy-side trader needs to first understand the business model of the broker-dealer. Is the broker-dealer agency-only, or does it engage in forms of proprietary trading that would compromise the buy-side trader's interest? The buy-side trader should also ask whether algorithms can achieve trading objectives. If using an algorithm seems right, then it's important to understand what a particular algorithmic offering comprises to determine if it will meet the trading objective. EdgeTrade's buy-side clients rely on our consultative services. They look for strategic and un-conflicted guidance from an independent, agency-only broker and software developer that can assist them in applying the appropriate algorithm to a trading objective.
Besides VWAP and arrival price, what other algorithms are being adopted by traders? Why?
Wald: Algorithms that help a trader solve a particular problem rather than just meet a benchmark are in demand. For example, EdgeTrade recently launched Sumo, a category breakthrough, smart order execution algorithm that offers traders the ease of issuing single orders for aggressive, timely execution. Sumo is the antidote to the inherent drawbacks of market and limit orders. Sumo electronically micromanages the process of determining how much of an order to send out at any one time for execution, and the destinations for these pieces.
Do algorithms threaten traders' job security?
Wald: Algorithms do not pose a danger to the job security of buy-side traders. In fact, algorithms are accelerating the migration of the trading role from the sell side to the buy side. Think of algorithms as a tool set for helping to create a "super" buy-side trader. Algorithms are like night vision goggles for a sniper: The best buy-side traders will arm themselves with these precision tools to sharpen their execution skills.
When should a human trader be used instead of an algorithm?
Wald: A trader can never fully be supplanted by an algorithm. They're the ones to choose how a trade will be executed, whether by sending an order to an algorithm or via DMA. Certain types of trades lend themselves to an algorithm. However, the trader needs to be in a strong position to address unexpected circumstances that cannot be pre-programmed, such as extraordinary news developments, significant external events and other extenuating circumstances. For all their benefits, algorithms can't think or exercise trading judgment. The trader is always vital.
What are the drawbacks of using algorithms?
Wald: Unrealistic expectations of what algorithms can do are a key drawback. Also, traders should not look to build algorithms when it isn't their job. Constructing robust algorithms is hard work; it is best to outsource the tricky programming and tedious minutiae. Traders would be better off working with an execution specialist they can trust, and one that does not engage in proprietary trading, to recommend an appropriate algorithm to achieve a given objective. EdgeTrade develops its own algorithms and often customizes these strategies for trading clients as part of our consultative process. We also build new strategies for clients, enabling them to concentrate on their core business of making sound and effective trading decisions.
Should order management systems time-stamp trades?
Matthew Celebuski, Bear Stearns: In addition to regulatory time stamp requirements, the entire trade cycle should be time-stamped because post-trade analysis involves more than executions. When the portfolio manager decides to buy or sell an asset the trade cycle begins. In order to improve the trade cycle, buy-side firms must look at the effects of time delays, alpha and market impact. An optimal trading style is predictable and matches the assumptions of the portfolio manager. The portfolio manager's decision process can be enhanced significantly with realistic execution cost estimates and feedback from the trading desk. Every execution and every decision should be time-stamped.
What is the best way for the buy side to differentiate the available algorithmic tools from sell-side brokers?
Celebuski: The best way for the buy side to differentiate available algorithmic tools is to compare performance and risk by algorithm. We suggest implementation shortfall as the benchmark and standard deviation of performance as the risk measure. Factors affecting algorithmic performance such as liquidity, capitalization, and index membership should be taken into account when comparing algorithms across brokers. After thousands of executions, patterns should start to emerge. Clients without their own history of algorithmic trades can start by requesting consistent statistics from their brokers.
Will algorithms be strictly a sell-side offering, or will we see OMS providers and execution management providers offering trading algorithms?
Celebuski: Algorithms are a high-fixed-cost, low-variable-cost method of trading. The high cost of development and testing of algorithmic strategies will keep most buy-side development of algorithmic trading strategies to a minimum. In addition to development, participating in an arms race to enhance and upgrade the algorithms would be a significant resources drain on buy-side firms in a market where, due to the low variable cost, brokers offer the service at less than premium commission rates. Also, implementation of the algorithm into a high-throughput, fast-trading infrastructure is as important as the algorithm itself.
Are algorithms a danger to the job security of traders?
Celebuski: Algorithms access a specific source of liquidity in the markets, while traders continue to access liquidity from other areas. Algorithmic trading has become a meeting place for buyers and sellers of small size over time. As such, algorithms are one of many tools available to traders. Market fragmentation has forced the trader to become the quarterback of each trade with the ability to engage exchanges, ECNs, block desks, internal liquidity and a myriad of crossing networks. Algorithms increase a trader's efficiency in handling increased order flow to multiple venues.
When should a human trader be used instead of an algorithm?
Celebuski: We've identified seven criteria that individually or in combination can be used to gauge the appropriateness of a trade for algorithmic execution. One of these seven variables measures the stability of the trading profile of a stock. Because algorithms tend to trade continuously and in small size, the asset may have a more predictable trading pattern. As algorithmic volume increases, the liquidity available in small sizes over time increases. From this and six other criteria, our systems suggest a strategy from over 100 combinations. Frequently the suggestion is "don't use an algorithm."
What is the best way for the buy side to differentiate the available algorithmic tools from sell-side brokers?
Tony Huck and Christoper Amorello, ITG: Performance measurement is the best and easiest way to differentiate the offering. Predictability and consistency in algorithmic performance, a product suite that is easy to understand, and algorithms that deliver what is promised are three keys to understanding how well your current trading strategies are performing. Traders need to rely on algorithms and know what to expect when they implement them. Over-engineered and complex algorithms tend to be less predictable and consistent in performance. ITG Algorithms can deliver predictable and consistent performance, and together with ITG pre- and post-trade tools, they help a trader measure performance.
How can trade data, such as order-routing information, be improved to better the pre- and post-trade analysis reporting?
Huck and Amorello: Traders want to see not only benchmark performance, but they need to understand executions relative to market conditions over the life of an order. ITG is working on producing these types of reports that enhance pre- and post-trade analysis, taking the feedback to the next level. Incorporating the knowledge from extensive pre- and post-trade feedback can ultimately help improve trading performance. A better understanding of this total process can help traders choose the right algorithm and modify processes and strategies as they go.
Besides VWAP and arrival price, what other algorithms are being adopted by traders? Why?
Huck and Amorello: ITG Dark Algorithms, which aggregate trades and send them to various ATSes, have seen significant growth recently. Clients looking to maximize exposure to alternative and hidden liquidity have found the ITG Dark Algorithms to perform particularly well. These algorithms provide one way to get to multiple ATSes and sources of liquidity and, hence, serve to improve productivity. ITG List-Based Algorithms allow traders to manage the trading and execution of an entire basket or portfolio, not just single stocks. These algorithms enable traders to monitor executions while controlling sector and dollar neutrality, tracking error, and total risk.
When should a human trader be used instead of an algorithm?
Huck and Amorello: For optimal results, a trader working together with the right algorithms usually proves to be the best use of automated strategies. Human traders will always provide insight and value to the trading process that no strategy can. The trader can best determine where he or she can add the most value to the trading process. For example, there is typically a portion of a transaction that is less volatile, where an automated strategy works well. By contrast, some percentage of the trade will involve price movements, volatility, or news that a strategy cannot process as effectively. The challenge is to detect which parts of the trade are best handled automatically and which should be handled by the trader.
What are the drawbacks to using algorithms?
Huck and Amorello: The algorithm's user interface does not always give enough transparency or control. At ITG, we are working on various tools to help reduce the black-box lack of transparency so that a trader has more control over the order and the ability to monitor how well the trade is progressing. Of course, an algorithm's performance must be measured as well to ensure consistency. ITG's Transaction Cost Analysis (TCA) is key to demonstrating the performance of algorithms so that adjustments can be made to the process and improve performance.
Ary Khatchikian Jarrod Yuster Joseph Wald Matthew Celebuski Tony Huck Christopher Amorello |
Source:
© Incisive Media Ltd. 2009
Incisive Media Limited, , is a company registered in the United Kingdom with company registration number