Algorithmic Trading: 4 Perspectives
by Editorial Staff
futuresindustry.org
Keith Fishe is the chief executive officer of TradeForecaster, a technology services firm providing solutions for automated derivatives trading.
Tom Frank is the chief information officer of Interactive Brokers, a broker-dealer specializing in electronic trading and direct access to electronic exchanges and financial markets worldwide. He is also a vice president at Timber Hill, a market maker affiliate of Interactive Brokers.
Eric Goldberg is the chief executive officer and co-founder of Portware, a vendor of algorithmic trading systems applicable to equity, futures and forex markets.
Larry Tabb is the founder and chief executive officer of The Tabb Group, an advisory firm focused on technology issues in the financial services sector.
Scott Johnston (moderator) was the chief information officer of Chicago Mercantile Exchange from 2000 to 2004. During his career he has held a number of trading and technology management positions at such institutions as Swiss Bank Corp. and O’Connor & Associates.
Some say it started in the U.S. equity markets. Others say it started at the Deutsche TerminBoerse, in the early days of electronic futures trading in Europe. Either way, algorithmic trading—also known as automated or black box trading—is one of the hottest trends in financial markets today.
There is nothing new about using computers to execute trades, of course. What is new is the extent to which computers are replacing individual traders in the trade execution process, and the degree to which computers are being programmed to generate their own trading signals. These computer models—often referred to as black boxes—are able to spot trading opportunities faster than any human being, and their increasing activity is one of the main reasons for the explosion in the number of order messages sent to exchanges.
How far will these black boxes spread? What risks do they pose for the markets? Who should bear the costs of the increased bandwidth, and what role will individual traders have in markets where these black boxes dominate?
To guide our discussion, we asked Scott Johnston, the former chief information officer of Chicago Mercantile Exchange, to arrange a panel of experts with different perspectives on algorithmic trading. Drawing from their experiences in U.S. equity markets as well as electronic futures markets, they agreed that algorithmic trading improves market liquidity and reduces volatility, but there was no clear consensus on whether algorithmic trading improves transparency or on who should bear the market infrastructure costs. Regarding risk, our panelists agreed that clearing firms will have to move to real-time risk management, especially as black box traders demand direct access to the markets.
Scott Johnston: Let’s start this discussion by seeing if we can define algorithmic trading for our audience. How does it differ from electronic trading in general?
Eric Goldberg: I would consider anything where the computer is making the trade decision and executing on it as algorithmic trading. That can cover everything from very basic rules-based trading or smart order routing all the way up to fully automated pairs trading, statistical arbitrage, benchmark executions such as VWAP, and market making types of applications.
Larry Tabb: The problem with such a wide definition is that when people use the same term for so many different things it can get very confusing. I like to think about algorithmic trading as having six components. One is high-speed market data, which is the platform that everything else depends on. The next component is the decision as to what assets to buy or sell to achieve the investment goal. We are now seeing a growing number of buyside firms employing model-driving quantitative strategies in their investment decision process. The decision comes out of computers that have been programmed to look for certain measures within the market data and then make decisions as to what assets to buy or sell. This is what I would call “black box” trading. The third component is the trade execution, which determines what algorithm should be used to actually carry out the trade. That’s what I typically think of as algorithmic trading at the large institutions in the equity space. From there it goes to how the order is routed. In markets where you have multiple pools of liquidity, such as U.S. equities, many firms have developed smart order routing systems that use a set of rules to automate the search for best price. The fifth component is the actual matching process. Traditionally that was straightforward—the order went to an exchange or ECN, but now there is a lot of internalization, so there’s variability in that model. The last step is transaction cost analysis, which looks at the trading model and the execution to see how well the trading process worked. Too often these components are grouped together, and it becomes impossible to figure out what part of the puzzle you are talking about.
Keith Fishe: For the purpose of our discussion, I think that we have to narrow this topic down. If you go to the basic definition of the term, algorithmic trading is where you follow a rule set for taking or mitigating risk, that rule set has been coded, and the trading processes go from being manual to being computerized or programmed.
Tom Frank: I agree with Keith. That broad picture Larry described is indeed the complete cycle that a market professional would go through, but I think the algorithmic part of the cycle is in the steps of the trade decision and execution. The true meaning of the word algorithmic is a series of steps that you can write down and execute.
Scott Johnston: How does the current state of algorithmic trading in equities compare to other markets, such as index futures, interest rate products and foreign exchange? One measure that I have is the number of vended products that you can buy to engage in algorithmic trading. I haven’t seen that many in futures, but I certainly have seen a lot in equities.
Larry Tabb: I think algorithmic trading is probably the most advanced in equities, but we’re starting to see hedge funds and other alternative investment vehicles looking more aggressively outside the U.S. equity market. There are a lot of models working that market and it’s becoming harder to find opportunities quickly. Our expectation is that they will migrate into other markets such as futures, fixed income, and options, and also foreign markets, where the market structures are different and there is more opportunity.
Keith Fishe: Just looking at the interest rate futures arena, we are seeing more and more venues for electronic trading and more and more access through APIs. That’s also happening in the foreign exchange markets—the platforms for trading forex used to be closed systems and now they are opening up. I think you are going to see an increasing number of algorithms and programs in these markets, especially as people try to find ways to spread and arbitrage the different products on different platforms.
Eric Goldberg: One of the key factors that determines how quickly algorithmic trading spreads is the adoption of a standardized communication protocol. When everyone has a different protocol, the cost to translate to all those protocols really limits access for the typical trader. One of the reasons why algorithmic trading is so advanced in equities is because that marketplace very quickly standardized on the FIX protocol. Now standardized protocols are coming into place in many different asset classes and that barrier to access is really coming down.
Tom Frank: Standards certainly do open up the field to more participants, but there is a more fundamental issue. The marketplace has to become electronic before you can use FIX for order management. That conversion is well underway in areas such as equities and interest rate futures, but there are other parts of the marketplace such as bonds where it is very difficult to transact electronically. So we still have a way to go before algorithmic trading can be applied across all asset classes.
Scott Johnston: Does anyone have a thought on evolution of the user base as these markets have gone electronic?
Eric Goldberg: Going back to what Larry said about the hedge funds coming out of equities and hunting for opportunities in other markets, we definitely see a lot of people exploring assets that they never traded before because the access is so much easier. So in that sense, yes, I would expect the user base to expand significantly as education and accessibility increase and the cost of entry and execution decreases.
Keith Fishe: I think you get both expansion and attrition. The people who trade another asset class electronically are attracted to a new electronic platform that they haven’t traded before, but some trading strategies just can’t make the transition from the pits to upstairs. Hopefully the traders using those strategies can find new approaches.
Larry Tabb: There’s another aspect to this process. In the past, the broker-dealers and to a certain extent the institutional investors were very product-focused because they needed that particular product understanding and knowledge of the market microstructure in order to trade well. But these new crews of people doing electronic arbitrage across the markets are just looking at models and correlations and reversion to the mean. For them the market microstructure becomes less important as it is abstracted away and the access becomes more standardized.
Scott Johnston: I have heard from Eurodollar traders that in the good old days of open outcry, an eight-point move was a big move and did not happen very often. Now they claim that electronic trading has dramatically increased instantaneous volatility to the point where a 20-point move is not uncommon. What liquidity trends do you see driven by algorithmic trading, and does algorithmic trading have any effects—good or bad—on liquidity or volatility?
Keith Fishe: I’m not sure I would agree that algorithmic trading is causing short-term volatility spikes. If people are willing to assume risk by making markets, then what drives the price path is not necessarily the method they are using to get into the trade. You have to look very carefully at that 20-point move, when it happened and what the drivers were, before you can conclude that electronic trading is causing that volatility. It’s quite possible that the traders using very fast automated trading tools were able to get out of the way of a sharp move like that, and the traders who weren’t using those tools, who were still using point-and-click trading tactics, got hosed by the move.
Tom Frank: In terms of liquidity, the simple answer is: more, more, more. Algorithmic trading, and electronic trading in general, allows much broader access, and that brings more people to the market. Secondly, algorithmic trading allows dealers to manage their risk with greater efficiency and participate in markets in such an immediate way that they can afford to maintain very narrow markets on a broad spectrum of instruments. So over the years, and certainly in equity derivatives and in the stocks themselves, spreads have been going down and down, and trading volumes and liquidity up and up, as automation of the trading process takes out certain components of risk and slippage. As mentioned earlier, post-trade analysis is a very important part of effective trading. Anyone who builds an algorithm where they pay 20 pips over isn’t going to make a living. Those algorithms will naturally fall away.
Larry Tabb: If you look at the equity markets over the past several years, volatility has declined a lot, and I think one factor has been the dampening effect that algorithmic trading has had on volatility. Many of these models are looking for mean reversion, so as things move astray, the models move it back in the right direction. Maybe the issue is not that algorithmic trading creates volatility, it’s that the uneven implementation of models creates volatility, and once more models get into the market, the volatility will tend to go down.
Eric Goldberg: I agree with Larry that many arbitrage models dampen volatility. On the other hand, I think we have seen a loss of transparency. We don’t see the same depth of the market that we used to in the U.S. equity market because we are not seeing blocks traded as much. The large orders are broken into slices, and there are trading tools like icebergs where the real size is hidden. We end up with the same liquidity as before, but it’s hidden behind tighter spreads, so you have to dig a little bit to find out where the liquidity points are.
Tom Frank: But there are many trading venues where the full depth of the market is displayed. I understand what you are saying about icebergs and things like that, but the way that these markets function now, you know the average price you can get at any instant for a very large size, and in that regard there is greater transparency, not less.
Larry Tabb: I don’t agree. There are more and more desktop technologies that hold the orders back from the market until either the algorithm or the trader determines that the conditions are right. So the liquidity is held within a firm, or in a broker’s algorithmic staging area, rather than submitted to the exchange and resting in the order book, where it is exposed to the whole market. When we go out and talk to institutional investors, they say there is greater fragmentation in the equity market because of this phenomenon. So I think there is actually less transparency.
Keith Fishe: I would add one thing. In the equity market, there is basically a FIFO process—first in first out—in terms of order allocation. In other markets where you have a non-FIFO order book, where the matching engine uses a complicated algorithm to allocate orders, you could have higher volatility. In that kind of market, participants may put up larger quantities than they want to trade, and that exposure could cause them to respond in ways that increase volatility.
Scott Johnston: Undoubtedly we as traders like volatility. Unfortunately most of the time with volatility comes infrastructure issues, such as much higher peak message rates.
Larry Tabb: I think that is going to be one of the bigger issues that we grapple with in the next five years. How do we handle the phenomenal increase in market data? A few months ago, we determined that the peak rate in the U.S. equity market—including equity options—was 55,000 messages per second. A market data guy told me a few days ago that we have already passed that, we’re at 70,000 ticks per second. It’s doubling almost by the year. To handle that volume of market data, to manage it, store it, analyze it, look for correlations—it’s going to be an overwhelming task.
Tom Frank: But I have to ask, who really bears the cost here? Most consumers—unless they are proprietary trading shops—are only taking a very small fraction of all the data.
Eric Goldberg: It starts with the exchanges. As the exchanges invest in more message capacity, everyone who wants to take advantage of that will have to invest in their market data infrastructure. And the cost is going to be borne by everybody.
Larry Tabb: The problem is that algorithmic trading can have a dramatic effect on the ratio of order messages to actual trades. I have heard anecdotes about a trading firm that sent six million orders into one exchange but only one was actually executed. This kind of activity generates a tremendous tax on the whole infrastructure. The issue becomes, are the people doing normal trading bearing the brunt of the cost versus the people who are just out there testing or prodding the market in certain directions?
Keith Fishe: The one thing you don’t want to do, though, is scare away the liquidity providers, who are spreading related products. They may be exposing themselves in another market and just trying to lay off two assets against each other. The fact that the market is moving causes them to have to pull the order and submit another. If you penalize them, you are taking a liquidity provider out of the market. There are certainly some who are extremely inefficient in their spread capability and those guys probably should be penalized, but you have to be careful about putting them all in one group and saying they are all bad.
Tom Frank: It depends on whether they are narrowing the spread or not. I think that’s the important question.
Scott Johnston: Is exchange support for algorithmic trading and electronic trading similar to what they offer in support of traditional trading, and what are they getting right or wrong for liquidity providers and customers?
Eric Goldberg: One of the side-effects of algorithmic trading, as we discussed before, is that a much larger volume of orders can be submitted at any one moment in time. We have seen instances where execution venues have tried to throttle the flow because they weren’t able to handle that volume of order messages. They are experiencing serious issues with people who can suddenly pound in hundreds of orders per second and constantly move those around. Everyone—exchanges, data vendors and traders—is having bandwidth issues and working on how to resolve those issues.
Tom Frank: I’m not sure that this is a universal problem. I think it depends on the business plan. Island came onto the scene not too many years ago and showed the equity markets that one can operate at astronomically high rates and do it successfully. With respect to the futures industry, there has been a broad evolution over the last several years with the European exchanges leading the way. A decade ago their volume was very low. Eurex came in and created a real electronic system, and within three years that completely changed the face of the European market. Only in the last few years have the U.S. exchanges become serious about electronic trading, in fact one of them is built on a European platform. That evolution is not complete, however. From our experience of market-making, the U.S. exchanges haven’t done enough yet in terms of providing an ability to manage quotes so that market makers can maintain a narrow market on a large number of options. The capacity is absolutely there in terms of the premier futures products, but the options that surround those don’t really have a mature market yet, because not enough technical resources have been devoted to supporting highly liquid markets.
Keith Fishe: I agree with Tom. In order for the futures options to become more electronic and deeply liquid, the futures exchanges need to attract more liquidity providers to their options. Because the options are so correlated and there are just so many of them, the exchanges have to install some protections in their systems so the liquidity providers don’t blow up.
Scott Johnston: How does the role of the individual trader evolve? Will the click-trading systems of today be relevant in the future? What major changes will they have to make to stay competitive?
Tom Frank: One trend that has been going on for a few years is that the management of trading is moving further out into the offices of the buyside. That’s been enabled and fueled by both the transition to electronic trading and the adoption of algorithmic trading systems. Risk management and execution management are moving outside the broker’s trading room and into the client’s, and the brokers and independent vendors are providing the technology to enable that. Regarding click trading, my experience is that electronic access has tremendously improved the ability of individual investors of varying degrees of experience to cross over from more traditional equity products and participate in the futures products. I expect that to have a significant role in the future. And even individual traders use algorithmic capabilities and electronically route their orders to brokers.
Keith Fishe: With respect to click-based trading in the U.S. futures market, if the trader is pursuing a strategy that can be easily automated and electronically implemented, they are going to have trouble using click trading software to compete against the algorithms. A lot of the rule sets that I see in the futures markets can be coded and done more efficiently by a computer.
Eric Goldberg: Yes, but that only affects click traders who are doing strategies that can be automated. If you go back to the question, how are people going to compete, I think the answer depends on what kind of trader we are talking about. The value of algorithmic trading is that it increases your capacity to trade, it allows you to trade strategies that you never did before, and it gives you more control over how those strategies work. We have found that one of the driving forces in the acceptance of our products is the demand for cross-asset capability on one platform. This goes well beyond the quantitative traders doing arbitrage trades between correlated assets in different markets or different countries. The demand also comes from hedge funds trying to auto-hedge an options volatility book and asset managers auto-hedging the forex risk on their global equity portfolios in real-time. So I think we end up with a very different type of trader, not necessarily fewer traders.
Scott Johnston: From a credit provider perspective, what risks does algorithmic trading introduce or increase? How does the broker or FCM manage the risk, and who bears the downside?
Keith Fishe: At first glance you might think that an automated trading client would introduce more risk. If they code up wrong and trade more than their capital can accommodate, that obviously puts the FCM at risk. But really algorithmic trading is a benefit to FCMs because it allows them to control the risk. If those controls are in place—even if the client has direct market access—a lot of the risk is mitigated.
Eric Goldberg: In one sense I think you’re right. There is the theoretical danger of a runaway algorithm that is doing something unexpected, but that is the same issue we had before with someone leaning on a keyboard. But it’s a very different situation when the broker offers direct market access. The broker is allowing the client to go directly to the exchange without supervising the flow. He doesn’t have any way to interact with the customer flow, and that can be risky.
Keith Fishe: I’m not sure I agree. I have worked at a firm with direct market access. Our clearing firm required access to the servers that had direct market access. So even though we may have hosted those servers, the clearing firm still had the controls in place and they could put limits in before our trades went through.
Eric Goldberg: There is big negative to that, however. Yes, the advantage of electronic trading is that it gives you real-time compliance with risk limits, but everyone is looking to shave milliseconds off the execution time. As soon as you hold up orders to run them through compliance, the process slows down and that creates a disadvantage compared to other proprietary traders. Speaking more generally, with the increasing use of algorithmic trading tools, it becomes very important for the broker to understand the client’s underlying strategy and have the tools in place to analyze the risk of that strategy on a real-time basis.
Tom Frank: Well, it only takes milliseconds to perform the margin checks before passing an order into the market. It takes my firm, on average, 15 milliseconds to calculate the margin for an account with the assumption that the current order will trade and become part of the portfolio. Our entire round trip time to the exchange and back is considerably less than 200 milliseconds.
Larry Tabb: I pretty much agree. Algorithmic trading doesn’t necessarily lead to larger positions and bigger risk exposures, but they can accumulate faster and do damage more quickly. That forces the brokers and FCMs to have better technology to capture the risk more quickly and analyze it faster. That’s the big issue. It’s not like you can just look at the end-of-day statement, see where your customers are, and then call up someone and ask for a margin call for the next day. Real-time risk management is becoming more and more essential.
Scott Johnston: Last question, what are the difficulties faced by new entrants into algorithmic trading, and what advice would you give them in order to succeed?
Keith Fishe: I think the cost of the technology needed to do algorithmic trading has really been going down. When it started, it was a very capital-intensive activity and it was limited to major banks and other large trading firms. As the cost has dropped and the level of knowledge needed to do it has become more readily available, the participation has expanded all the way down to the retail trader level.
Eric Goldberg: I certainly think it is easier than ever before to get involved in algorithmic trading. The tools are there to do it. Usually the biggest challenge is to bring in the market data with the lowest possible latency, and the hurdles to that end up being higher if you are a small shop.
Larry Tabb: I agree with Eric and Keith that the infrastructure for doing this has become more approachable for more people. On the converse side, the bar is rising on the amount of intellectual property and distribution capabilities needed to play in this game. There are so many people with PhDs coming into this space that it is becoming harder to develop sophisticated models and differentiate them from all the other models out there. This also affects the brokers that are trying to develop a suite of tools to take to their clients. That’s one of the reasons why we expect the use of these models to spread across the markets very quickly in the coming years. |
