My Machine: The AT Interview - Algo Reality
Investment management firm T. Rowe Price, headquartered in Baltimore, Maryland, has more than USD300bn under management. The company has a global presence and in common with other major asset managers uses execution algorithms in its various locations. AT talks to Clive Williams, head of European Equity Trading at T. Rowe Price's London office about his desk's experience with and expectations of algorithmic trading.
Clive Williams, head of European Equity
How long have you been using execution algorithms?
We've been using them for about two years in the UK, and our US operations were using them for some time before that.
What algorithms are you using?
We use Bloomberg's G-Trade for DMA and algorithms (for reasons of functionality and broker neutrality) as well as several broker provided algos. From a cynic's perspective, it is apparent that every trade sent electronically to a broker represents content and information, and brokers are increasingly driven by proprietary trading. While they obviously have procedures in place to prevent misuse of that information, there is always the concern that things might go awry.
What do you regard as the main differentiator for algorithm providers?
I think it is training; you have to be taught how to use algorithms properly and to understand their various nuances. It is all very well a provider claiming that they will give you all the algorithms you could possibly ever need, but if they don't explain how they work and the most effective way to use them, then the algorithms are worthless. The question of training is particularly important for us as we trade by sectors on the desk, with each of us having different responsibilities. For example, we have someone who specialises in small and mid-cap stocks, while somebody else specialises in financials and oils. As a result, very different tools and algorithms are required to trade these very different categories,and each trader must be able to understand which algorithm is suited for which purpose and how best to deploy it. Of the various providers we use, Bloomberg have been very good as regards training and have provided excellent support. They are also keen to know what we are looking for in terms of functionality and future capabilities and then actually responding to this. More recently we have noticed Merrill Lynch in particular adopting a similarly positive approach.
How does the feedback loop for algorithm improvements work in practice?
Brokers will come back to us with feedback as to how well we have done relative to benchmarks, and as they give us more comprehensive feedback on a timely basis you gain a better insight into your trading effectiveness. For example, Merrill will come in here every month and we will go through all the trades for the previous month, highlighting and discussing good and bad trades. We then analyse each of these trades to see what we were doing and how we were doing it - e.g. which particular algorithm or technique we were using at that particular time for that particular trade. That is useful as a learning and development process for both us and the algorithm providers.
What's your take on the competitive position of the various algorithm providers?
From our viewpoint the gap between the early leaders in algorithms and the rest of the pack has closed up. I think that has been because those trying to catch up have generally been stronger on the training than the early leaders. As I said earlier, if you want somebody to use your algorithms, you have to make sure they are comfortable using them.
One of the TRP team: Jeremy Ellis, European stock trader
Have you considered taking "off the shelf " algorithms?
The problem for us with that approach is the obstacle of integration with our order management system, which is not particularly extensible. We want everything on one platform - we don't want technological fragmentation with lots of proprietary links, as we regard that as an unacceptable operational risk. As a result, we prefer to deal with those providers who have the resources to integrate their algorithms with our OMS.
In a perfect world, an EMS sits on top of our OMS to give us access to a wide range of DMA, algorithms and various other tools. In a perfect world, FIX will come up with a solution for the integration of algorithms into OMSs, but that may still be some way off.
Is your intention that DMA and algorithms will drive more of your order execution back in house?
Yes, we want to put tools on the traders' desks so they can do more trading internally. That change is already well underway, so rather than outsourcing execution risk to brokers we are increasingly using the various algorithms and DMA tools at our disposal to do the trades ourselves. I would say that our low touch activity, including DMA and algorithms, now probably represents somewhere between 20% and 30% of our activity here in London, compared with virtually zero two years ago.
Is that shift towards bringing order execution in house something that would have happened anyway, regardless of the growth of DMA and algorithms?
No, I think there's definitely a strong link, especially when you consider that all that brokers do with many of the orders that are passed to them is put them in an algorithm anyway - and we can do that ourselves! This is one reason why I think our use of DMA and algorithms will continue to increase along with growth in in-house order execution.
And the old "information leakage" question?
Obviously information slippage is always the one thing we don't want. The more order execution you put out to brokers, the more chance you have that this information slippage will occur. You can be dealing with a great sales trader, but you can't know who's looking over his shoulder and who that person might in turn be speaking to.
An example of this is emerging markets, where you may talk to a sales trader who has to pass the order to an execution trader who in turn may also have to speak to a local trader about the order. As the number of people who legitimately know about the order increases, so does the risk of the information being more widely disseminated. Plus there is the additional risk of somebody somewhere along the line misinterpreting what was required.
Therefore, if we can do the same thing in-house using an algorithm or DMA, then we will take responsibility for our own trades directly.
Which markets are you covering from London? And are there still many markets you trade where algorithms are not viable?
We deal with the main European markets and EMEA from London. Of these, we are currently using algorithms for all of Western Europe, Poland, Hungary, the Czech Republic, and South Africa. However, many of the emerging markets cannot be traded with algorithms. In some cases this is because they are still open outcry. In others there is still a lot of block trading, with not very much liquidity actually going through the market. However, I think this off-exchange activity will reduce in time, thus making DMA and algorithms viable.
Nevertheless, given the progress that you have seen algorithms make in the past two years, do you think they might become relatively commonplace in emerging markets in the next two?
I think that's possible, though initially only in the more developed emerging markets. We are already starting to see this sort of activity in South Africa and we may well see growth in places like Russia. However, you have to bear in mind that a lot of the stocks in these emerging markets trade in London as well, so the need may not necessarily be there. For example, a lot of Russian names will trade as ADRs. As I mentioned earlier, for some exchanges it simply isn't feasible - obviously an integral part of any take-up in emerging markets is the replacement of floor trading with electronic trading. I'm sure this will happen, but perhaps not as fast as we would like.
From the perspective of algorithmic trading, what do you regard as the greatest difference between US and European equity markets?
Fragmentation - at present, I don't think European markets are as fragmented as those in the US, so there is less need to sniff liquidity. However, in the light of recent events on the exchange landscape that could all be about to change.
Do you think algorithms have become smarter?
Yes, I think they have, and as part of that they are also becoming more customisable. So instead of just getting a straightforward VWAP algorithm you now get some additional flexibility around how it operates. That makes sense in practical terms, because while you might initially be quite happy working just vanilla VWAP, if you suddenly see a block of liquidity then you need to be able to react to that. This sort of customisation means that you can trade more intelligently than just taking two hundred shares every two minutes.
But is there the risk that the flexibility and degree of user configuration available on many algorithms is almost pushing you back to where you once were as a purely manual trader?
I think we are still a very long way from that - I certainly don't regard the current level of flexibility as excessive.
Nevertheless there is now more need for the trader to interact with the algorithm?
Yes; for example, I will put something into an algorithm and then sit and monitor its progress. If it isn't doing what I need, I might switch it from being straight VWAP to using a more aggressive strategy with a price limit. In practice we find that we are changing the way we use algorithms throughout the day according to circumstances.
But are there nevertheless some stocks where you find you can leave the algorithm to get on with it on its own?
To some extent yes, but it really depends on what the individual fund manager wants to do. Some managers might be quite happy just to say follow VWAP throughout the day, but you aren't really adding very much value if you do that. On the other hand some managers will have a liquidity benchmark because they have given you a trade that is maybe four times the average daily volume for that stock. Therefore they are not so interested in benchmarking it to a VWAP - they just want to get their fills. As one of our portfolio managers is always reminding us: "Don't be a prick for a tick".
Do you find that your fund managers are taking an active interest in algorithms?
There is no question about that. They are very involved and there is a lot of discussion between us about how we can use DMA and algorithms to further enhance performance. This ties in with the generally close working relationship we have with the managers. (For example, one of our traders has been trading small companies for twenty years and she and the small-company fund manager know exactly how each other thinks.) We have a daily morning meeting with the managers when we discuss current trades, but there is also a constant flow of information and updates among us throughout the day.
Would you say that the improvement in the intelligence of the algorithms has meant you are able to use them for a wider range of stocks than previously?
Definitely - there are stocks that we would originally have had to execute manually or through a broker that we can now do through algorithms.
Are any of T. Rowe Price's funds completely automated, with an automated model deciding whether to trade and then passing the consequent order to an algorithm for execution?
No; while we have one manager who takes a fairly quantitative approach, we don't have any black box funds.
Are any of the algorithms you use incorporating data from algorithmic news feeds?
We were aware that these news feeds were being launched. However, we haven't yet been approached by any of our providers about this, though I can see some hedge funds might well be interested.
T Rowe Price, London
Your desk has a lot of experience in executing trades in
difficult markets that have challenges ranging from illiquidity
to outright abuse. In view of that, do you have many providers
looking for feedback or additional information on these markets?
One or two are starting to cotton on, but others simply haven't picked up on it. Ultimately if you accept that an algorithm is to some extent an attempt to synthesise and automate the skill of a trader then some insight into specialist trader skills is probably relevant.
I think where trader input is particularly valuable isn't so much in the development of the algorithm as a whole, but in its various configurations settings. Those can make a huge difference to the level of an algorithm's acceptance in the market place and also its performance.
Has your use of algorithms had an effect on your back office? For example, have small trade slices caused a problem in terms of throughput?
Our increasing use of algorithms hasn't caused any capacity problems with either the order management system or back office systems. The use of FIX has meant that we really haven't noticed any change in practical terms - if the FIX message from the broker matches our internal ticket in terms of price/size then it will just go straight to settlement regardless. In fact FIX has meant that multiple sourcing has become transparent - as part of the same overall trade we might work some with an algorithm, pick up a block from a broker and another block on Liquidnet. From a back office perspective that doesn't cause any problems.
What do you see as the future for algorithmic trading?
I think the increasing flexibility of the settings for algorithms is an indication of the direction in which things are going. To my mind, further customization of algorithms seems likely. At the same time, I think buyside adoption of algorithms will continue to increase. When you consider that one of the largest US broker dealers was doing 15% of its US customer order flow with algorithms in 2000 and it is now doing 80%, you get an inkling of the way things are going. If the brokers can do that with algorithms, why can't the buyside? It may not reach 80%, but it could nevertheless be a significant figure.
Do you think you will always take your algorithms from an external provider?
I think there is certainly the possibility that we may develop our own algorithms in the future. Whether we do or not depends upon the availability of the sort of toolkit that will facilitate this without also requiring a huge internal quantitative research program. Given the way that we often see today's expensive custom solutions morph into tomorrow's run of the mill commodities, it is always a possibility.
Do you get the feeling that there is still resistance in some quarters of the UK market to the use of algorithms?
Unfortunately yes, I think there is. I have come across at least one instance where a trader was hardly using algorithms because his fund managers didn't want him to. They were also blocking his attempts to acquire the tools for algorithmic trading.
Regrettably it appears that some fund managers still seem stuck in the world of buying stock from brokers who provide a good lunch - though hopefully that is becoming a thing of the past.
Are there any areas of algorithmic trading of particular interest for you going forward?
We'll be very interested to see how the concept of dark pools takes off outside the US. We think Liquidnet is excellent and have been using it since day one. However, in a UK context it is slightly limited because the UK operations of some very large investment managers who could contribute a lot of liquidity are not allowed to use it.
Ultimately we are very keen to find hidden liquidity, so any sort of algorithm or venue that allows us to do so is always most welcome.