Debate: The pros and cons of algorithmic trading

COPYING AND DISTRIBUTING ARE PROHIBITED WITHOUT PERMISSION OF THE PUBLISHER: SContreras@Euromoney.com

By:
Published on:

It´s the new Big Bang in equity trading. As research and execution are unbundled, and as clients increasingly access markets directly, brokers need to find new ways to keep their institutional clients. Algorithmic trading is one way. It cuts execution costs, adds alpha and gives the creative brokers – those able to design flexible, customized algorithms – a new way to keep in the execution business, for now.

Algorithmic strategies explained

Richard Balarkas (RB) is a managing director of Credit Suisse in the investment banking division, based in London. He has responsibility for the Advanced Execution Services sales product in the equities department.
Daemon Bear (DB) is head of equity trading (Europe) at JPMorgan Asset Management. Daemon has been with the firm since1998. His previous trading roles were with Worldinvest Investment Managers (now New Star Asset Management) and GT Management Asset Management (now Invesco).
Alasdair Haynes (AH) is the CEO of ITG Europe and is responsible for the strategic development of ITG in Europe. He joined after a 20-year career in investment banking, having held senior positions with Bankers Trust, UBS and HSBC.
Tom Middleton (TM) is head of the algorithmic trading team, covering more than 17 markets across Europe and Africa for Citigroup Global Markets in London. Tom has been in the team since its inception, developing algorithmic trading strategies that adapt to the idiosyncracies of a diverse set of markets.
Peter Urry (PU) is head of dealing at JO Hambro Capital Management and is a 30-year veteran of the securities industry. Previously, he was a proprietary trader for one of London’s first hedge fund managers, IFM Trading, and a sales trader in risk arbitrage products for Barclays Capital.
Tim Wildenberg (TW) runs the European direct execution services team at UBS Investment Bank. He has been responsible for establishing UBS’s equity electronic trading service and recently has been working on rolling out UBS’s leading edge quantitative trading tools for client use as part of the Direct Strategy Access Service.

Stephen Wilson (SW) is head of exchange-traded products at Reuters.
Steven Wood (SW) is an executive director and global head of trading investment at Schroder Investment Management Limited. He joined Schroders in 2002 from JPMorgan. He is a member of the Euronext advisory board, the Liquidnet advisory board and the Deutsche Börse advisory board.


SW, Reuters
It is estimated that algorithmic trading – the automated execution of orders according to a predefined strategy to meet a specific benchmark – is used by an estimated 60% of US buy-side firms. This percentage is set to grow. The take-up in Europe is currently thought to be about half of that in the US but is again expected to rise dramatically. Before we examine the issues, can you just explain exactly what you all mean by the term?

AH, ITG Europe Yes – and for the purpose of this discussion the focus will be on equities only. Algorithmic trading strategies are mathematical models that automate trading processes using predefined parameters or benchmarks. Algorithms achieve alpha in an environment where speed and the prevention of information leakage are paramount. These types of algorithms were first developed for this purpose approximately 10 years ago. Algorithms can be classified into two broad groups: structured algorithms and opportunistic algorithms. Structured algorithms – for example, VWAP or implementation shortfall – make greater use of cost and risk modelling; opportunistic algorithms – such as pairs or volume participation – make greater use of heuristics.

TW, UBS We define algorithmic trading as any form of automated rule-based trading where decision-making is delegated to a computer model. That includes hedge funds running prop strategies, the sell side executing its normal order flow via a rule-based engine or the buy side using algorithms to execute their business without going through a normal sales-trader. The key development is the increasing use by the buy side of algorithms to execute their business as an alternative to the traditional route. Typically this type of algorithm is designed to target a trading benchmark: for instance VWAP, TWAP, implementation shortfall or market-on-close [see Algorithmic strategies explained ]. Although algorithmic trading has been around in some form for a while, decimalization in the US – which triggered a decrease in average trade size – and the increase in US market venues a few years ago, has recently prompted an explosion in it as traders have increasingly found it impossible to achieve high-quality execution without using the technology. More recently it is the use by the buy side of broker-built strategy servers that has become the recognized and most widely used definition of algorithmic trading.

AH, ITG Europe There are a number of different tools and uses. Even within the trading space there are clearly different types. There are heuristic-type algorithmic tools, which are used in opportunistic methods – for example, volume participation strategies. And then there are more quantitative and analytical tools, like implementation shortfall (the difference between the decision price and the price actually achieved). It’s a very wide subject. Algorithmic trading clearly is not the same product as direct market access (DMA), nor is it a completely separate tool – whereas algorithmic trading is a strategy for trading, DMA is a destination and routing issue, not a strategy.

TW, UBS Algorithmic trading services are very different to pure DMA, which is a much simpler product.

RB, Credit Suisse I think we’ve moved through three phases of algorithm development. Volume, time-based strategies – TWAPs, VWAPs – that was the first phase. The second phase was price-based, shortfall-based and reduced impact. And now we have idiosyncratic, personalized or customized strategies. But all three are still perfectly valid, because fundamentally as a trader you’re thinking, ‘How fast have I got to trade versus how much impact is it going to have?’, and if the price is moving in your favour one day, the simplest way to make sure you trade for six hours is to TWAP it.

TM, Citigroup It is very clear what the first type of algorithm is going to do: all of them can be broken down pretty simply into two parts. First is a scheduling part – how much do I want to trade in the next 30 seconds or in the next 10 minutes? Second is how I’m going to achieve that capture of liquidity. For instance, VWAP will trade according to a historical volume profile, possibly with dynamic adjustment, TWAP will trade similarly to VWAP, but with a flat volume profile, and so on.

SW, Reuters And the second group of algorithms?

TM, Citigroup Well, implementation shortfall is the first of the next generation, where perhaps the execution strategy is less clear. The aim of implementation shortfall is simple – to minimize alpha loss by trading at an optimal rate. Trade too quickly, and alpha is lost through higher transaction costs due to impact. Trade too slowly and the alpha has been realized by the time the order is completed. How you achieve that, though, is less clear much of the time. The most systematic way to do it is to use a pre-trade model to give an optimal time horizon and execution profile over the interval. But this is not the method selected by a significant proportion of brokers. Implementation shortfall strategies can add more value than a straightforward VWAP strategy – the difference between ‘good’ and ‘bad’ VWAP strategies might be a couple of basis points, whereas implementation shortfall strategies could vary by tens of basis points.

RB, Credit Suisse The next generation are the customized algorithms that we are doing more and more of. These are completely different to the dancing-round-the-spread TWAP and VWAP-type strategies. Investors are realizing that to add value they must act more like traders, and trading is all about taking a trader’s view. As traders on the buy side get more comfortable with taking responsibility, and develop their own styles, we don’t always know what their objective is. They might ask something like: “When spreads look like X and the depth on the other side looks like Y, and relative to my full order size it looks like Z, and if momentum signals look like this, then cross the spread, buy everything you can, be 100% aggressive.” In this scenario it is difficult to judge whether or not that’s the right strategy and to measure what is happening.

Benefits to the buy side

SW, Reuters What has attracted the buy side to algorithmic trading?

SW, Schroders Algorithmic trading is one of the tools we use to reduce our transaction costs – this is one of our major briefs. Rather than outsource trading to the sell-side, we undertake trading ourselves via DMA and algorithmic trading. One method is basket or program trading. There we get a list of stocks in and evaluate how we’re going to trade that position – whether to go for risk capital or go agency. In the past, obviously if the risk pricing was correct, we’d go risk. If the risk price was too high versus our estimate for the total impact of the agency program, we’d probably give the program to a broker to work and monitor the quality of execution. But now we trade it ourselves, using algorithmic trading and DMA. We can reduce the volatility quite dramatically against where the brokers would give us execution, mainly because, unlike the sell side, we can focus specifically on our own trade. We tend to use DMA and algorithmic trading a lot more for single stocks now, especially where we could be both illiquid and liquid. These tools have to be anonymous, so we can passively react to the marketplace, soak up any pools of liquidity that might be around on the exchange, as well as search for that large block from our usual high-touch, full-service counterparties. If there’s nothing natural in the market, we can place an algorithm on trade and access the market dynamically and with anonymity. So there are various ways you can use algorithmic trading, but it has to be within a DMA environment and you have to be able to dynamically react to market circumstances, instantaneously. That’s where you can add the alpha to your process by reducing transaction costs.

PU, JO Hambro Algorithms have been a natural follow-on for us from DMA trading. We started to use DMA to give us speed, control, cost savings and anonymity. Algorithms extend those features by grafting the processing capability of the computer onto a strategy that the trader can select which captures his view of how to handle the specific order.

DB, JPAM Our drive towards algorithmic trading has been led by two things: first, regulatory changes – the increased transparency of transaction costs through commission unbundling. Second, there has been a dramatic increase in the emphasis on performance over the past five years. That has led us to want more ownership and control of our execution, because all algorithmic trading really is, is smart ordering systems like TWAP that achieve better execution costs. So we’ve spent just over 18 months trying to in-source it back in. The trading desk mentality has moved towards full ownership. Our remit in all fields of execution has been to add value wherever we could. But now, our profit and loss is on a far more aggressive benchmark. In the past we’ve looked back over our execution and tried to analyse our performance, whereas now our benchmark for trading is to estimate a trade cost before we go to market, and then that decision price and any shortfall around that implementation is what we are compared against.

PU, JO Hambro We are a much smaller firm so the size of our orders allows us to be more nimble. If I choose the algo route, I pick a strategy in the same way as I would if I were giving the order to a sales trader to work. But you need to decide in advance what you’re trying to achieve and then be flexible in tweaking the strategy if you are not getting the outcome you want. The algorithm is a processing tool, the trader still has to take the responsibility for the execution.

RB, Credit Suisse One reason given by the buy side for using algorithmic trading is that those institutions see themselves taking increasing responsibility for results. But in Europe, for example, there are still a lot of institutions that don’t use algorithms, possibly because they believe it’s an abdication of responsibility. In fact the complete opposite is true.

DB, JPAM Two or three years ago, if heads of buy-side desks in London started talking about algorithms and DMA, the immediate response from the guys on the buy-side desk was: ‘Well that’s me out of a job!’ It couldn’t be further from the truth. Traders are now happy to embrace such tools and appreciate that they enable them to utilize capacity to a greater effect.

TW, UBS The buy side has become more market-facing rather than broker-facing. Previously the buy-side role was to talk to a bunch of brokers. Even in the early days of the FIX protocol, suddenly when people were exposed to real-time fills – execution reports – or suddenly understood that every tick on the screen was a fill, we were finding clients getting hit with system issues because their system was receiving so many fills. They were not set up to be market-facing rather than broker-facing. Now the buyside has definitely taken ownership of the execution, even on the block-trading side.

PU, JO Hambro One of the other advantages for me is that I see the fills coming back as the order’s working, and if the strategy isn’t delivering what you want or the market circumstances change, you can make changes. I don’t have to explain what I want to a sales-trader who in turn relays what he thinks I said to a trader. I get speed and accuracy.

SW, Schroders Especially if you’re based off a very aggressive benchmark. As soon as you get the order the trader owns it. If it takes us 10 seconds to get to the market we can lose a lot of performance over that time period. With the DMA-algorithmic route we can get instant access, formulate a trading strategy to get into the market there and then, as well as to start looking at other sources of liquidity if it’s a fast-moving market.

TW, UBS For the buy side, algorithms provide a tool for taking more control over their trading without having to trade the entire order themselves.

SW, Reuters And benefits for the sell side?

AH, ITG Europe Many people that come into this today are white-labelling current products because they want an algorithm in order to collect commission – to get paid through some form of unbundled service. But they’re not adding value with those algorithms. They’re using white-labelled older products from pretty well-established firms.

PU, JO Hambro But I think unbundling is really the crucial thing here. Unbundling will change the way the buy side trades and the way the sell side relates to buy-side desks. Both sides will focus on execution. For the buy side, execution cost will become transparent. For the sell side, the profitability of both their overall trading departments and of individual accounts will be measurable. The sell side will respond to this by reallocating resources towards profitable activities and relationships.

TM, Citigroup That will drive demand for the algos that are flexible and customized to the client that a white-labelled product will find it harder to provide. A broker who has outsourced algorithm development to a third party may find it harder to react to client demands in a timely fashion, compared with one that has an in-house development team.

Computers are best

SW, Reuters What sort of order flow is best for an algorithmic tool?

DB, JPAM Initially it was our low-touch volume – the small orders in liquid stocks – that went through. But now it’s being used a lot more to take part of an order, so that we can have something working in the background. Rather than giving our sales-traders the orders in line over the next day just to get something on the board, they can use their time looking for natural business – the other side of our trade in the market. So it’s taking smaller pieces of bigger pies as well as just the small pies on their own.

TW, UBS We’re seeing people starting to use algorithmic trading for specific types of business. Initially they use it to execute smaller orders but then, as they get more comfortable, they increase their order size and then that grows and grows. There is a point at which it is uneconomical for certain types of trades. For example, once an order gets very small, there is no point in doing anything other than placing it directly on the order book as the costs of trading it in more than one shape outweigh the value derived from splitting it.

SW, Schroders I think you’re starting to see that in the US. If you’re buying 200 shares of IBM, why does that trade have to go to the buy-side sales/trader? If you can place certain parameters around an algorithm and you can route the trade straight into the market or exchange and have the trade reported straight back again, then why not do that?

PU, JO Hambro We’re all interested in keeping our cost base down and doing more business without hiring more people.

RB, Credit Suisse You’ve got to remember that it is in part an efficiency tool. If there’s liquidity, then an algorithm can usually execute reasonably well dancing around the spread. If there isn’t, it can just watch the market for you. It might not get you 20 days’ volume done, but you might get a couple of hours’ worth done while you’re busy talking to a broker.

TM, Citigroup Look at where computers have a competitive advantage. They can monitor thousands of data feeds and massive amounts of market data simultaneously and they don’t need lunch. This applies to both so-called ‘enhanced’ DMA and quantitative-type algorithms. In both cases, computers can react to opportunities that might exist for just fractions of a second that human beings would miss, and they can do this simultaneously over thousands of orders. With smart order routing in the US, we now see the situation where there is no upside in the trader not using them if he wants to achieve best execution across multiple execution venues. For quantitative algorithms, if the algorithm has been researched and built properly, the execution strategy will be based on an objective, robust statistical model that will help to improve execution and reduce execution volatility still further.

SW, Reuters What tools and strategies are suitable for algorithmic trading?

TM, Citigroup Essentially any stock that is electronically traded is suitable for algorithmic trading. The difficulties occur when trades are particularly large, or particularly urgent. These, of course, are trades that are difficult anyway, but are more feasible through traditional means. Contingent or opportunistic trades are highly appropriate for algorithms – the trade can be entered in the system and the computer is left to monitor the order book until the conditions are satisfied, when it can execute instantly.

SW, Schroders It also depends on how the asset management firm is monitoring their execution costs – what parameters they have for execution and what the benchmark is. We’re focusing on implementation shortfall so we’re obviously looking at ways of developing that implementation shortfall algorithm – and it’s not as easy as a VWAP algorithm.

AH, ITG Europe Not all trades are suitable for algorithmic trading, for example, complex trades, and not all trades are suitable for DMA either, so they have to work together.

How to choose

SW, Reuters Can you explain the processes involved in deciding to move to algorithmic trading and what is involved in actually moving? What issues do you have to think about internally? How do you choose a provider and what levels of provider are there?

TW, UBS When choosing algorithmic trading, you first need to analyse the type of business that you are doing on the desk. Do you have a lot of small orders or a few very big orders? It is unlikely that you will find you have only big or only small. Most people have a number of different styles of flow. Once you have done this you need to consider what you are trying to achieve as a benchmark for your trading. You need to review the technology you have on your desktop and then you need to consider which provider you will use.

TM, Citigroup Our clients use algorithmic trading in a variety of ways. Obviously you need connectivity to your broker. This could be as simple as using a Bloomberg terminal, or as complex as having a global order management system (OMS) or execution management system (EMS). We also service our clients in a variety of ways – we have all the common strategies that suit most of our clients. We have also designed our algorithms to be, to a large degree, pre-customizable. There is a large parameter superset that allows us to put together thousands of execution strategies from pre-designed building blocks. Finally, there is a small group of clients with whom we work in partnership to develop unique, customized strategies.

PU, JO Hambro We use algos supplied by some of our existing counterparties which we access via Bloomberg. The operational issues are minimal.

Measuring performance

SW, Reuters How do you go about evaluating the performance of an algorithm versus the traditional method of doing whatever the algorithm is doing and then how do you measure the relative performance of one algorithm or provider against another?

AH, ITG Europe Not everyone’s algorithms are effective and providers need to be compared and evaluated. Our latest study evaluated more than 2.5 million orders, consisting of about 10 billion shares (valued at more than $80 billion) traded, from more than 40 institutions. It looked not only at the performance of algorithmic trading versus non-algorithmic trading but also compared the performance of different algorithmic systems offered by brokers. The performance of six brokers was analysed but their identity has not been disclosed. This study shows that algorithmic trading can play an important part in helping firms achieve best execution but that users have to be discerning as to which product they choose and under which circumstances it is appropriate for them. The key findings were that algorithms can reduce the cost of trading; average performance differences across providers for very small orders are few, but gaps between providers grow as order size grows; all algorithms are not created equal – there are indications that certain algorithms perform better than others at higher volume levels; and certainty of outcome – as important as quality of outcome – varies widely between brokers.

DB, JPAM I use the same tools for measuring my traders’ performance as I do to analyse the performance of the various algorithms. There’s a lot of testing and analysing. It’s not perfect but we’ve created a model and use that internally. We’ve never used an external survey provider to evaluate the performance solely of the trading desk, because there isn’t one we’re comfortable with. We will use our internal process to analyse the performance of the individual broker’s execution capabilities and of the algos as part and parcel of the whole study.

RB, Credit Suisse There might be a conflict of interest issue if you don’t use a third party. Is the issue that the products or services in the marketplace don’t do what you need them to?

DB, JPAM No, they don’t really. Naturally we need to back up our findings with a more widely used product. For analysing the processes, the offerings and our tools, we use the internal model. But we subscribe to an external one as well, just to give the client something to compare us against. We use this to compare performance against a competitive universe rather than against a preferred benchmark.

SW, Schroders We used a third-party vendor to analyse our trade data. This ranges from a global overview down to the trader level. We don’t use it for brokers, as most of those orders are limit orders, so they can’t be attributed to the impact costs as we don’t outsource the responsibility for the trade to the broker – we retain this ourselves. We do analyse DMA very aggressively but on a more dynamic level. It’s the traders’ responsibility to pick the right algorithm, and to change between algorithms and DMA as they work that order in the marketplace, so that they can optimizes the outcome of algorithmic trading. Our traders manage the passiveness of the DMA or algorithm and the aggressiveness of their order flow according to the market conditions and according to the stock’s performance in the market. It’s very hard to prove that an algorithm is doing a good job, because we use multiple algorithms within the same trade. So it’s really whether the trader is using his expertise correctly to minimize the costs by using that toolset within the algorithm, as well as using the toolset among different types of trading facilities.

TW, UBS Thus on the buy side your focus is to judge your trader. I think on the broking side, nevertheless, we are obliged clearly to be able to quantify the quality of the execution we provide. So we need – as we are doing – to be able to quantify the quality of the execution we have delivered.

DB, JPAM Yes, I think you still need to compare the performance of each algorithm along the same strategy line.

TW, UBS The beauty of algorithms is that they are easily measurable. But as soon as there’s a human element in there, trapping and measuring is really hard. If you give a broker an algorithm, you are able to specify exactly how you expect an order to trade, if a human gets involved the data is never so clean. I think over time people are going to expect to be able to justify having to pay a premium for a human worked order, and that’s much harder. I don’t know about algorithms to choose algorithms but it seems very clear that over time more orders will be executed with machines.

TM, Citigroup There are two issues here – whether we’re evaluating the performance of individual traders who use algorithms as one of several execution tools, and how we compare the performance of algorithmic trading providers. In both cases, the difficulties are that you need a huge amount of data to make any statistically significant deduction – and the only way that you can get this is through a product that offers full, broker-neutral peer comparison

PU, JO Hambro This gets back to the key issue, which is that DMA and algorithms have filled out the buy-side traders’ toolkit so that they can take control of their orders. They are a means to an end, not an end in themselves. Their use should be seen within the broader context of unbundling.

Dangerous decisions

SW, Reuters Let’s look at current and future developments. The topic of pre-trade algorithms is hot right now. Why?

TM, Citigroup Pre-trade is useful in that it will give you an idea of what a typical trade of that form has cost in the past at a particular trading rate. If you find that the alpha expectation over a given time horizon will be totally eroded simply by the impact of executing the order, you can review whether this is a trade worth doing.

AH, ITG Europe Pre-trade is one of two things. One, facts about average daily volume and so on; two, the analytics, where you start using a statistical model. It’s like using a speedometer in a car. It’s not precise to 100%, but it gives you good standard deviations of return to give you reliability numbers. People get some idea of what form of alpha erosion is going to take place as a result of the trade given the size, the market, and the change in that market, so while you’re trading, you can look at what the theoretical numbers ought to be. Everybody who drives here looks at the speedometer continuously. You use pre-trade before you trade, and you should continue to monitor throughout. They’re incredibly useful numbers for traders to have, but you’ve got to understand the reliability.

SW, Reuters By alpha erosion you mean adverse price moves?

AH, ITG Europe The market moving against you is one aspect. Overall, though, it is when market impact is greater than anticipated when the trade is placed.

TW, UBS I think that is a bad analogy, because I think a speedo is not pre-trade. Speedo is all intra-trade. What we’re talking about here is before you’ve started the car and driven.

AH, ITG Europe Well it’s like satellite navigation. Before you get into your car you have a pretty good idea where you’re going to go, how long the journey’s going to take, how you’re going to get there, and it gives you all the details. So to flesh out the analogy in full: pre-trade is satellite navigation, it gives a suggested route and estimated time of arrival; intra-trade is the speedometer – it shows what you are doing while you are doing it; and post-trade is the journey report.

RB, Credit Suisse It’s an area where people have to be careful, though. Certain brokers offer algorithms to help you pick an algorithm – pre-trade models that will give you a different answer about which algorithm to use depending on whether you want to be aggressive or not. But there’s some confusion here. A pre-trade model might work as a benchmark for a risk price in something you haven’t traded for six months, where you don’t know the price you should be getting. But it is in danger of becoming one of those tools that encourage people to abdicate responsibility. I haven’t seen a machine yet tell you where a price is going or what the trend is but in this area some people think these systems can do this – that they are a crystal ball which, married with an algorithm, guarantees them success.

SW, Schroders It depends on your definition of pre-trade. I see pre-trade as analytics, used mainly for programme trading, to define our bid sheet for risk programs, and also for studying the momentum over the last three or four days, just as a guide to how you want access the market. I agree that if you start trying to set a benchmark to beat, it could be very dangerous. That said, the sell side is utilizing pre-trade analytics on its single-stock risk-pricing with some success as they would not be running any loss ratios on capital commitment – the pre-trade analytics would predict the correct price to make to the buy-side trader. However, more work needs to be done in this area before it become robust enough to be valuable.

TW, UBS There is definitely a trend for people to look at these tools to know whether a stock is going up or down. Pre-trade analytics will definitely not tell you whether a stock is going to go up or down. They will give you an idea about how you should expect an order in a stock to trade based on a number of factors. The danger is that people will see them as a panacea.

PU, JO Hambro If I get an order, there’s one things that’s changed about the stock as far as I’m concerned, and that’s that I’ve got an order in it. It’s likely that the fund manager is making a judgement on a piece of fundamental news that might mean that the stock will not trade in line with its historical patterns

SW, Schroders A lot of it does come down to the fund manager’s perceived alpha return on the order. If the portfolio manager’s expectation is that the stock is going to go up 20% plus and they’ve got five days’ volume to buy, no matter what the pre-trade analytics say, they’re going to want to undertake the trade to lock in the potential alpha return.

AH, ITG Europe Pre-trade should not be just the tool used for a trader. The expected cost should also be seen by the fund manager, so he knows how much alpha erosion is expected on the trading desk, and the trading desk, if they are good, will improve on that number by looking at ways in which the net alpha is not reduced, and that’s absolutely critical. The fund manager should also be working off the same pre-trade analytics tools and building a portfolio for the creation of alpha.

RB, Credit Suisse That’s how pre-trade can start to be mis-sold. A fund manager says: “I think this stock is 250 basis points lower than where it should be.” Then you give the fund manager a tool that says: “By the time you´ve sold that and bought this, it will cost 300 basis points, so don´t do the trade.” Meanwhile, if the manager had looked at the market, he’d have seen that the trade was feasible.

AH, ITG Europe He’s getting a statistical number to give him a better chance when he’s building a portfolio. He’s looking at hundreds of different things. If it’s one stock, a human can generally beat a computer, but when you’re looking at thousands of stocks it’s impossible. A fund manager or a fund wants some idea what that alpha erosion is going to be on an academic basis and in a quantitative way, and that’s what pre-trade is providing for. This is done by using historical and statistical data to determine what future behaviour might be.

TM, Citigroup Pre-trade analysis is like any statistical model. Your “average” impact cost on a single trade might be n basis points, but the variance will be much bigger. I don’t think anyone’s selling pre-trade as a panacea for an event-driven fund manager who trades eight times a year. But at the other end of the spectrum if you’re a quant fund manager who trades once a week in 500 stocks, then it will show you which 10 stocks are particularly hard to get done. Pre-trade models and algorithms are trading tools, which for most buy-side clients will be used in conjunction with more traditional execution methods.

Unbundling to drive growth

SW, Reuters Will the choice of provider keep growing? Or will there be consolidation as companies start to go out of business?

SW, Schroders On the sell side we’re seeing more and more competitors trying to get into this arena, but I think they may be too late. With a lot of the algorithms being commoditized there has to be a differentiating factor for a new DMA/algorithmic provider to succeed.

TW, UBS I’m sure more people will arrive, but I think this links to the unbundling question. The economics of this business will become pretty evident, and it’s not cheap to build a quality algorithmic trading infrastructure. Providing global trading connectivity with network redundancy is difficult on its own – you´re not just talking about cash management-like systems, you´re talking about real-time trading in financial markets in professional size where you have market exposure – before you overlay it with some pretty bright people doing quantum maths. When you start looking at the volume of business and the value of the flow going through any of our systems, not just in the algorithmic space, the investment you’ve got to make keeping the thing alive is non-trivial. Sometimes you get a client who says: “I’m just going to write the algos myself.” Well I think there’ll be a finite number on the sell side who can do that, and even fewer on the buy side, because of the costs.

SW, Schroders We wouldn’t want to write them ourselves.

TW, UBS People will get recognized for specific services – quality execution, quality research or whatever. That’s unbundling. But just because that will happen does not mean that people will be able to knock up an algorithmic trading business just by spending a few million bucks. It’s not just the rocket science. There’s the whole back end of settling cheaply and cleanly. So there will be consolidation in the execution space – no question.

PU, JO Hambro Well you say, Steve, that you’re already doing a lot more business than you were on commission sharing arrangements. Have you shortened the list of brokers that you execute with?

SW, Schroders No, we haven’t shortened the list, even though we’ve gone unbundled, because we’re very active in the small-cap and mid-cap market, so we’re still seeing flows from the smaller brokers. I can see that there might be consolidation but this has not happened yet. The amount of DMA we are trading has gone up, as the traders do not follow commission target lists

DB, JPAM Absolutely – when you start out to satisfy best execution, traders have to have total autonomy – over every transaction, every execution.

PU, JO Hambro Before the buy side had DMA and decent algos it wasn’t really that feasible. I think the buy-side desk taking more control over its executions means that it’s much easier to argue internally that the trading desk can ensure good execution and still meet research targets via CSAs.

PU, JO Hambro Unbundling focuses people’s minds – not necessarily on agency broking but on research and risk trading. I think once you’ve lost this fuzziness of: “You’ve got to make me a bid on these, because we’ve got a great relationship”, and it comes down to: “You’ve got to make me a bid on these because I want a bid and you’re a risk house so what is it?” People’s willingness to put up capital when that’s all costed out just on that book is going to change.

RB, Credit Suisse If you’ve got a good pre-trade model, then you can allocate perfectly. I’ve got to get a risk price for this and get it done now, because someone out there’s probably not pricing it as well as he should. But I can spend all day trading this in a TWAP or VWAP.

TW, UBS Or I’ve got such a strong view about my alpha that I’ll pay the premium to get it done. I do think the cost of capital will go up.

PU, JO Hambro But I think the net cost of execution, which at the moment includes a loss ratio, probably should go down.

SW, Schroders It will be interesting to see whether the cost of capital does go up because, in our experience, if anything it’s gone down recently. And if you look at the program trading arena, where cost of capital is very cheap, every time somebody blows up and leaves the market, everybody says the prices will get wider, but somebody else comes in and fills the space.

DB, JPAM Those satellite outfits are appearing less, but in the long term I do believe the cost of capital will go up.

RB, Credit Suisse I think the reason it’s changing is because, as the buy side shifts more into electronic trading, it tilts the balance. And as you get smarter and momentum signals are built into your pre-trade tools and so on, things get more difficult for the sell side.

Every call for a risk trade will be there for a good reason, and that’s because the stock’s moving like a rocket and you want the broker to take the loss onto his balance sheet, and the economics just won’t add up.

TW, UBS I think unbundling is forcing people to understand their clients’ profit and losses as they never have before. New investment is going into analysing the profitability of clients. We’ve never seen it before, and this is why the cost of capital is going to rise. Unbundling is going to mean that you look at your research relationship, at your execution relationship, and price them accordingly.

SW, Reuters What other new trends will we see?

AH, ITG Europe You’re going to see direct strategy access rather than direct market access. I’m sure that is going to be the buzzword. Instead of going to the market directly, you go via different strategies that then route you to liquidity source(s), which may be a market or a crossing system, etc.

TW, UBS Certainly that’s our brand. The overall brand is direct execution of services. The direct market access is obviously, “you drive the car to market”, and the direct strategy access is the algo.

AH, ITG Europe I don’t think anybody knows where it’s going to stop. Over the past 18 months there have been all sorts of developments, going to different destinations, improving price the whole time. People used to think liquidity was too fragmented for an algorithmic trading tool to work – but it does. In today’s environment it’s thought that algorithmic trading will probably never be suitable for small-cap-type stocks. But who knows if in the future they will be appropriate. “Never” is the wrong word. It’s just what’s appropriate in the market at the time and how quickly it develops.

SW, Reuters Well, that’s all we have time for today. Thank you all very much.

You can have your questions answered by prominent market players.
Submit your question or topicand it could be brought to the table for discussion by some of the the leading minds in the industry.

Have a look at upcoming debates on the 'Debates' page