This month's new feature is the Sellside Round Table. While our buysiders continue to discuss trading ideas and model development (page 32), a select group of sellside figures has joined us at the editorial round table to debate the key challenges traders face in using algo execution in the FX markets - questions are posed by senior members of the Automated Trader team. Appropriately for such a large subject, this will be a two-part feature, with the distinct characteristics of FX algo usage covered in this issue, coupled with an assessment of the critical differences between FX and other asset classes in this context . In Q3, the debate moves on to FX algo development and the prospects for FX algo functionality that accommodates multi-asset strategies.
Joining us at the SELLSIDE ROUND TABLE
Cameron Mouat - Head of foreign exchange algorithmic execution, Deutsche Bank
Paul Beatty - Global Head of FX, Bloomberg Tradebook
Gary Stone - Chief Strategy Officer, Bloomberg Tradebook
James Dalton - Director of FX Algorithmic Execution, Citi
Jonathan Wykes - Head of AES FX Sales EMEA, Credit Suisse
Automated Trader: What are the key differences in the design and application of FX and equity execution algos? How does buyside demand growth for each category compare?
Cameron Mouat: One of the key differences is that FX execution venues do not publish transaction volumes to the same level as equity markets. This results in less transparency, and quantitative methods required to study the dynamics of market microstructure will have a higher degree of uncertainty. Execution algorithms need to factor this into their decision process.
Another major difference is that a significant amount of FX volume is transacted on execution venues without a central counterparty. Where possible an execution algo should utilise anonymous venues and natural liquidity within a bank to minimise signalling risk. Algorithmic execution is still relatively new in FX whereas it has been around for a number of years in equities and is a common execution service for the buyside.
Deutsche Bank recognises that some clients are looking for more transparency and control of their execution, which is why our algorithmic and transaction cost analysis products complement traditional services.
Paul Beatty & Gary Stone: We see two differences in the market structure and the types of algorithms that clients employ in seeking to achieve their investment goals. Relative to the equity market where regulators have imposed rules to make the market more transparent so that everyone in the market is dealt with in the same manner with equal access to the same displayed price. By contrast FX is an opaque OTC market with no central price and where prices may be tiered based on credit and size considerations. Additionally, there is no consolidated trade recap and little time, sales and volume transparency.
FX is still a market place that is in transition. Some market participants are "last look"/"order delivery" liquidity providers or market venues - where you have to send orders and wait for a notice of execution that you are filled. This is similar to how the U.S. equity market operated, from after the changes to the order handling rules of 1996 that spawned the ECNs, to the July 2001 launch of the final phase of the NASDAQ's SuperSoes Automated Order Execution System. Some FX market participants place firm auto-executable liquidity in a marketplace. So, with the liquidity fragmented in different pools and accessible under different conditions, true smart order routing is essential for seeking maximum liquidity. And, this isn't simply being smart in taking liquidity, but also distributing/representing passive orders in several marketplaces to reduce the potential of being traded around.
From an investment strategy implementation perspective, the FX market structure makes certain algorithms difficult to design and employ. With little volume information, it becomes challenging to design participation (go-along/volume-in-price) and volume-weighted-average (VWAP) algorithms. There is pent up demand for low-touch algorithms such as time-weighted-average-price (TWAP) and wave algorithms that send waves of small orders into the marketplace.
We are starting to see some of the traditional investment advisors beginning to use TWAP, Block and Average Price (soft limit) Algorithms that work orders to get a price and move in and out of the market without really pushing it around.
On the hedge fund side, we see behavior that is akin to the strategies in futures. Many futures traders are technical in nature and look at different price levels for strategy entry and exit. They leverage Reserve-Scaleback to ladder in and ladder out of positions.
We developed OCO (One Cancels Other), trailing stops, stop limits and algorithms like that, which are relatively commonplace in futures and customised them to operate in the FX market structure. Additionally, target orders, where a currency order is released when the price of a related security reaches a certainprice (e.g. buy/sell EUR when Oil drops below $90) and 'Eco' orders, where an order will be launched into the FX market to buy or sell a currency pair based on some economic statistic (e.g. if U.S. unemployment is greater than 9.0 percent, buy/sell EUR) are very popular with hedge funds.
James Dalton: If you start from the ground up and look at the actual structure of the foreign exchange market, particularly in comparison to the US equity market, there are nowhere near as many destinations we need to connect to, to construct a currency-execution algorithm. When you go further afield and look at less liquid equity markets the number of venues may match what we have to access in foreign exchange. Despite the differences in market structure, algorithms are typically constructed with similar intent across both asset classes. They are designed to source liquidity in an opportunistic fashion or to capture bid/offer spread via a passive execution either against a pre-determined benchmark or participation rate.
Many of the equities execution algorithmic strategies were initially targeted at achieving a volume or time-weighted average over the course of the day. Certainly the growth of index portfolios through the late 90s and early 00s had a strong influence on the nature of client algorithms being offered. As their market changed and the hedge fund segment grew, more aggressive or opportunistic algos were added to the equity suite, assisting those in the hunt for short term alpha.
The spread compression in FX off the back of competition in electronic risk pricing initially mitigated the need to build algorithms. When clients began to show an interest in algorithms it did not make sense to import logic directly from the equities business. FX is a 24-hour market, there are fewer individual assets to trade, no official book of record or tape where volume-dealt-at-price is printed - and the personality of each currency throughout the day could be viewed as psychotic to an outsider!
We set about ensuring that whatever we created for clients could adapt to the variability of liquidity conditions, while observing the primary goal of executing orders of any size without disrupting the market.
Jonathan Wykes: There are key differences between equity and the FX algorithms. There are also key differences between FX algorithms offered by the sellside; this is important to point out. Electronic trading, traditionally, in the FX space has always been trading on a risk price as opposed to trading in an agency fashion. It's been click-and-deal rather than the routing of orders to exchanges or electronic matching systems. When you think about electronic trading in equities it has always been order-driven and in a more agency fashion. In terms of key similarities from our perspective, both markets are extremely fragmented so they lend themselves to the use of Smart Order Routing technology.
One thing you don't get much of, in the FX market is data. Therefore you have a lack of transparency. From that perspective, when you look at doing things like participation-based algorithms, it is much more difficult to get the required data and have that information at your disposal.
At CS we do more of the agency-style execution algos because our interest is to give our clients the best execution and be able to benchmark that against the market. This allows us to be transparent and helps clients better understand the true executions they are receiving from a broad pool of liquidity.
Automated Trader: How widely are those key differences appreciated by the buyside when it comes to day to day use of algorithms?
Cameron Mouat: Buyside traders that execute large orders are generally quite familiar with the differences between equities and FX microstructure. For those traders that aren't, it is part of our job to educate them so they know the positives and negatives of the way they trade.
Paul Beatty & Gary Stone: Tradebook has a dedicated group of Execution Consultants that work with our clients, clarify the strategy and the anticipated results and advise them on the best way to engage the market place. From that interaction, we are able to guide the buyside with the appropriate strategy to achieve their desired result and if need be further customise algorithms for the FX market place. The Execution Consultants also conduct post trade learning reviews and talk about how strategies performed, how they interacted with the market place and how we can improve them.
James Dalton: That comes down to the nature of the underlying client. If you have an organisation on the buyside that has a very well developed execution desk, with an in depth understanding of how markets work and how liquidity shifts throughout the day, they get it and they appreciate the amount of research you put in to construct something that is appropriate to the time of day you trade. For other clients out there it's almost a case of encouraging them to try algorithms alongside existing execution tools. We then help them get a feel for their execution performance by providing them with detailed post-trade metrics.
Jonathan Wykes: If you are starting from a position where you are heading up a centralised dealing desk, you are probably already executing using equities algorithms which are all very similar in terms of their market behaviour. When you look at the way we trade, our FX algorithms are very similar so it's much easier to make that transformation; we successfully managed to decouple the asset class from the algorithm. However, if you are a person who has never traded using execution algorithms in equities or any other asset class then it's a bit more of a learning curve. This is because the traditional way to look at execution would always have been on risks, whether it was over the phone or electronically. To then move to an agency method you are taking on the risk yourself as opposed to paying someone else to take on that risk for you. That is one of the primary challenges that we have faced.
However, with the advent of being able to do sophisticated TCA, it's very easy to show people that if they start using these algorithms they can truly improve their execution quality. If they embrace them they can outperform risk and trade close to mid, participating in price improvement, rather than crossing the spread all the time.
Automated Trader: Which structural characteristics of the FX markets throw up the greatest challenges for FX algo providers and their clients?
Cameron Mouat: A major structural characteristic is that FX is still primarily a quote-driven market (versus equities, for example, which is an order-driven market). In a quote-driven market, dealing on the best prices from multiple quotes can often be a very poor overall execution strategy, with unnecessary market impact caused by the series of executions. One solution is to execute using only one quote-driven liquidity source split over several tranches. Another solution is to execute on orders from order-driven execution sources. Finding non-toxic liquidity is an important part of trading FX.
Another challenge is to come up with meaningful transaction cost analytics in a market which is relatively opaque. Benchmarks common in equities cannot be applied directly to FX. At DB we are looking at a number of ways to do this and provide clients with relevant benchmarks for execution.
Paul Beatty & Gary Stone: Liquidity. I think it's hard to tell how deep the market actually is. Many clients are seeking to develop their own liquidity aggregators. One of the challenges of the market is there can be a lot of phantom liquidity where one client may have their delivery orders in several places at the same time. So algorithms many be unsuccessful because they are using vanilla order routers that do not take subtlety into account.
Tradebook has worked very hard at developing our FX eco system to have sources of deep, firm liquidity. Additionally, to supplement, we interact with some sources of "last look." You have to score your liquidity sources and know a lot about how they behave and what actually is there.
To maximise liquidity capture, our smart order router keeps statistics of success and liquidity availability so that we can better determine what is out there and maximise opportunities both during passive phases to gain the spread and aggressive phases to minimise market impact.
James Dalton: We have to invest resources continually into the process of adapting to the market as it evolves. New liquidity pools arrive and clients shift between platforms, changing the nature of flow that passes through different channels. Making sure you are connected to all the relevant electronic markets is a given. Understanding who is behind the liquidity in those pools and taking that into account with your order-placement strategy is a crucial part of any algorithms construct.
Jonathan Wykes: As I've mentioned already, one is risk based and one isn't. I think one of the greatest challenges is the lack of transparency, the lack of being able to see depth of market and what liquidity is really out there. If you can't see what liquidity is out there, clients will require far more education to know which is the best strategy to choose. Should you use an opportunistic or aggressive strategy? Is the level of liquidity good on the other side or not?
We are fortunate that we have a 24-hour support desk that will give clients colour on the market and provide recommendations on strategy choice.
Automated Trader: How are those challenges best addressed?
James Dalton: By having a technology team that can respond to required changes within a short time frame. It is also a matter of developing relationships within the range of market venues you use. Staying abreast of planned changes and getting involved in the decision-making process is important.
Jonathan Wykes: Our clients can log into various pages and see top-of-book liquidity and the aggregate of the liquidity that we have. We provide very detailed TCA so when your order arrives in the market you can see exactly what the bid/offer spread is. When you look at your final execution, we take the average price and compare it to the mid at arrival to see what your shortfall was.
You can find out whether you are trading on your side of the mid-point or on the other side, and you can find out how many venues you traded over, how the algo accessed the liquidity, and if it was performing well for you or not. We can also look - when using a strategy that can play both sides of the spread - at how much of this strategy was executed on a passive basis versus you aggressing the other side. There are lots of ways in which you can do that and once you have that data it's much easier to then go and make recommendations about strategy choice in different time zones and currencies.
Automated Trader: What's the relative value add between providing effective algos and connecting them to sufficient liquidity? What's the higher priority, an algo's functionality or the number of venues to which it is connected?
Cameron Mouat: I think both are important. Simply adding liquidity sources doesn't necessarily improve execution performance. There is a lot of non-natural liquidity on venues in FX and a smart order router needs to consider how clean the liquidity is, how anonymous the venue is and the signals being given to the market. The algorithms contain the trading intelligence to maximise the benefit of a well-chosen pool of liquidity.
Paul Beatty & Gary Stone: The reality is, your algos won't work if you don't have deep enough liquidity. You are always looking to make sure you have best quality liquidity rather than quantity of liquidity (which can contain toxic or phantom liquidity making execution results suffer). They go hand in hand because the greatest algo in the world without liquidity is just a great algo and it doesn't do anything for anyone.
James Dalton: This is kind of a trick question. If you peel it back and look at every single currency pair, let's say the top ten most liquid currency pairs on the planet, the breakdown of liquidity from the first to the tenth is extreme in terms of dollar volume traded per day. You may only have one or two valid destinations to trade a less liquid currency pair, whereas a more liquid currency pair like the Euro you can trade anywhere.
The question doesn't really cover what we need to do to
offer FX algorithms across a reasonable spectrum of currencies. You need to tune your algos according to the pair.
Take Euro/Dollar as an extremely liquid example; you could have a great algo for it but if you are only running it against one ECN that turns over say less than 20 billion EUR/USD a day you are not going to get the same quality of execution that you would if you were connected to EBS plus CME or whoever else.
Its also important to bear in mind that internalisation strategies provided by a single bank can at times offer more liquidity, in a friendlier fashion, than full access to the street.
Jonathan Wykes: I think you are always going to have a lot of venues to trade G7. If you trade the Euro you are going to find that on every single venue. Therefore if you are trading on every single venue then you can be really aggressive and you can pay up for the liquidity and have minimal slippage.
I think a lot of people assume that because you don't find much liquidity in emerging-market currencies, algorithms simply don't work there. Actually it's completely the opposite and what you tend to find is that using a strategy that interacts with the liquidity is more important than having loads of different liquidity providers.
We find people get very good execution results in less liquid currency pairs at less liquid times of the day. If you have the right strategy you can actually provide liquidity into the market and essentially get the market to come onto your side as opposed to constantly having to pay up for it.
If you are looking for a risk price, there may not be any counterparties willing to make a competitive risk price so with an algo you are adding liquidity into the market rather than taking from the market. There are always people out there looking for a good price, so if you use a strategy wisely then you will execute.