The Gateway to Algorithmic and Automated Trading

What do the traders say?

Published in Automated Trader Magazine Issue 20 Q1 2011

We present the first instalment of a two-part opener to our new regular feature, ‘Buyside Beat’. In the weeks running up to the publication of the Q1 issue, David Dungay spoke to a number of traders and other buysiders about the key challenges they face in the quest for alpha. Discussions ranged over model development and deployment, HFT, IT investment, and the practicalities of trading in today’s markets. Those discussions continue, and we’ll bring you more in future issues, but for now – here’s what the traders are saying in Q1.

On the BUYSIDE BEAT this month

Dmitry Bourtov, CEO, Unimarket Corp Dmitry Bourtov, CEO, Unimarket Corp
John Reeve, CTO and head of trading, BlackCat Capital John Reeve, CTO and head of trading, BlackCat Capital
Miles Kumaresan, Principal and Head of Trading, Algonetix Miles Kumaresan, Principal and Head of Trading, Algonetix
Fred Pederson, Business Development Manager, Vincorex AG Fred Pederson, Business Development Manager, Vincorex AG
Thomas Parry, FX Trader, Algotecture Thomas Parry, FX Trader, Algotecture
Dr Peter Wiesing, Founder and CEO, Global Arbitrage Group Dr Peter Wiesing, Founder and CEO, Global Arbitrage Group

David Dungay: Do you feel that the profitable life expectancy of trading models in general is shortening?

Dmitry Bourtov: I wouldn't agree too much with that. In reality, every 7 to 9 months the volume of trading activity and messaging volume on the typical exchange is doubling. If you look at what has happened with CMEGlobex or Liffe Exchange over the last 5 years, and particularly in the last 2 to 3 years, this pattern occured every 7 to 9 months. To me, it's just a sign of the times moving faster. A lot of the things that were developed, the short-term things, might have a short life expectancy based on that increase because a lot of things change because of volumes. At the same time I still use systems that were used a long time ago and use fairly long time frames. They are still profitable even though the original concept was developed in the mid-nineties.

I don't think time is critical for everything because so many things are still valid today with a few tweaks. In general many things still continue to work today that were developed years ago.

John Reeve: I think it depends on whether the model captures some underlying characteristic that's fundamental to the way the market operates. It's easy to find patterns that exist for a few months or a year but are transient in nature. Models that capture some fundamental behaviour tend to keep working. For, example all the models we started trading in April 2008 are still working and have been consistently profitable.

illustration 0

Miles Kumaresan: Compared to the late nineties and early part of the last decade, I would say it is shorter but has been relatively constant over the past 5 years.

Fred Pederson: Yes, I would say so. The profitable situations in the market change faster nowadays so the simple things that used to work nicely two or three years ago don't work anymore. You have to constantly adapt your models and adjust strategies. I think this is due to increased competition and increased interest in high frequency trading.

I think about 5 years ago you could have put together one configuration that trades the DAX against the EURO-STOXX. A Eurex single exchange trader could keep a single correlation there for a year and make a lot of money without adjusting it. These days I think it's hard to make any money without doing inter-exchange trading. And even for that you will probably end up adapting your model on a weekly or monthly basis.

Thomas Parry: Yes, definitely. I think it's due to a combination of more participants in the market chasing the same opportunities, as well as better technology that enables them to identify and implement strategies faster to take advantage of various market inefficiencies. I think the biggest differentiating factor that all of the most successful market participants possess is the ability to quickly and efficiently implement new models while these opportunities are still profitable.

Peter Wiesing: I strongly agree that this is valid for all trading approaches for which popularity has increased. These include all hedge fund strategies together with high frequency trading market-maker strategies. All these markets have become extremely competitive in the last decade. As a result, profit possibilities decrease faster. Hedge fund managers and prop trading shops are forced to develop new ideas constantly and always reconsider their existing trading approaches.

David Dungay: If so, do you think this is because of wider availability and advances in available technology such as data mining and back testing - or some other factors?

Dmitry Bourtov: It's because of some of those reasons, yes. There is an illusion that exists that information is easily available because you can readily get a daily chart and it will look like the charts of 10 to 20 years ago. Then you deal with something really short term which is millisecond or less scale and you start getting different problems.

Data is available but the volume of information which you need to analyse is enormous. On top of that you put a lot of requirements on trading-system design in terms of speed to be able to trade quickly changing information. It's a huge amount of data and a huge amount of information that needs to be processed.

On the one side it's much easier to get access to this data but the volume of information is so much higher these days. It's not so easy to deal with this information from a software perspective compared with 10 years ago.

John Reeve: There is no doubt that approaches such as data-mining are being more widely used and all of the low-hanging fruit went some time ago. Of more concern is the unprecedented change in the markets over the last 5 to 10 years as a result of electronic trading and changing regulation.This has resulted in significant shifts in market behaviour.

Miles Kumaresan: Not really. I doubt very much that data-mining tools and back-testers would have contributed to reduced model lifespan. Basic commercial back-testers have existed for decades and so have sophisticated analytical software packages. At the competitive end of quant trading, building one's own back-tester is not a major effort.

Increases in hardware speed, in-memory databases, etc., have allowed more complex models to be explored, and as a result, these advances may have extended the lifespan by allowing quant models to differentiate themselves.

Fred Pederson: I think the biggest thing that has changed the market is competition, but not by those people that buy proprietary software off the shelf. Investors and traders with a big capital background are entering the market and because of that competition has increased at a professional level. It has increased the number of people who build their own software and servers and put a lot of money into development.

There is definitely an increase in the number of off-the-shelf products in this space. I think it's the big companies, which have founded their own HFT branch or other high frequency prop traders that are trading their own proprietary software, that have made the biggest difference. Off-the-shelf products do lower the barrier for entry into this market but that doesn't mean you will be successful. I doubt that you can be competitive for a long time with these products and so people will stick around for a while and then leave the market. They have to compete with the big prop shops that have their own HF software optimised to their specific needs.

illustration 1

Thomas Parry: Yes. Correct.

Peter Wiesing: Yes, technology is key for shortening the profitable life expectancy of strategies. Data on all temporal granularities and computer power are extremely cheap in comparison to potential trading profits. In addition, smart computer scientists and quant analysts are attracted by the financial industry. These conditions are ideal for creating and back testing new investment ideas. However, without a clear and focused investment idea, data mining itself will not create valuable trading strategies.

David Dungay: Do you have the sense that the marketplace for certain types of trading strategy - high-frequency market-making, for example - is becoming overcrowded and that this may be reducing the profitable half-life of that sort of strategy?

Dmitry Bourtov: It is hard to tell. Ten years ago it was just a few companies that were trying to do that. You can only really speculate about what people are doing and what their bottom lines are. So you don't really know what they achieve from this particular strategy. I wouldn't say it has lessened in the last few years, there is still a lot of interest in this area in the futures and FX markets and this interest is still growing. There are many more participants but I couldn't tell you if bottom lines are much less than a few years ago.

The markets are much bigger and if you look at the futures market 5 years ago, even before the crisis, the volumes are significantly higher now and liquidity has increased.

John Reeve: High-frequency trading in general is past its early innovation and growth phase. It is dominated by a relatively small number of major participants who will likely continue to do well from it. However, people coming late to this will find it much harder to make money. I'm not sure to what extent, if any, that influences model life.

Miles Kumaresan: In general, yes. In the last 10 years there has been significant growth in quant trading. It started with an explosive growth in the mid-frequency domain, this having a much lower entrance threshold. At one time every quant had a stat arb model, so to speak, and as a result this space went through an evolutionary phase to reach its current maturity. High-frequency trading on the other hand, requiring a major entry ticket, did not have as much of the noisy overcrowding as the mid-frequency. Nevertheless, high frequency trading is a very crowded area and competition does squeeze the profit margin.

Fred Pederson: Yes absolutely. I think market making is the best example. The spreads are getting narrower and narrower; if the spreads are getting tighter, then the margins are getting smaller for the high frequency traders. For this reason you can only be competitive if you increase the number of trades that will be executed. You can only do this now by investing a lot of money in having the best and fastest infrastructure.

The low hanging fruit has been gone now for a while, two or three years perhaps. A good example is the one I mentioned earlier with DAX/STOXX. Two or three years ago you could trade this pair only with one configuration and live comfortably from just that.

illustration 2

Thomas Parry: I've definitely seen the high-frequency space in the FX market become more crowded over the last couple of years. For example, almost all of the 'simple' inter-market scalps and the vast majority of the triangular arbitrage opportunities have been eliminated from the market. Additionally, liquidity providing strategies have also become more competitive due to spread compression and increased market transparency. However, I still firmly believe that there are plenty of opportunities to capture alpha if one understands how to correctly price liquidity and efficiently hedge flow. I think a lot of HF guys are also starting to incorporate more intelligent pricing that enables them to look further out (i.e., 3 to 5 ticks as opposed to the next tick) in order to make directional bets or skew their quotes in accordance with their market view.

Peter Wiesing: The problem with high-frequency trading as a liquidity providing strategy is special, because its success is mainly driven by optimising the speed of algorithms and hardware. High-frequency traders do this rather than identifying anomalies or market inefficiencies by using sophisticated quantitative analysis where speed is not as critical. High-frequency opportunities have a clear theoretical and practical limit.

High-frequency strategies have become extremely popular in the last couple of years. We run short-term trading strategies where we benefit from trends, which last minutes or hours, but do not employ high-frequency trading strategies. Therefore, I cannot ultimately assess if the profitable half-life of that sort of strategy decreased but I can observe that these strategies have lost a great portion of their potential profit opportunities.

David Dungay: If so, how directly does that influence the type of trading model you deploy? For example, do you seek to have models in timeframes, markets and with business logic types that are less popular?

Dmitry Bourtov: If you aren't trying to develop something for a particular time frame and you are looking at the bigger picture, to build something that is profitable and robust, then you have quite a wide spectrum. It depends on what you are able to achieve. If it's working on a daily bars basis that's fine or if it's working on a tick-by-tick basis then that's fine. If you are dealing with millisecond to microsecond delays in your system and it's really high frequency then great. If it makes money and it's robust enough then that's great. If it achieves its risk and reward goals then it works.

John Reeve: For BlackCat's fund portfolio we diversify as widely as possible in market, strategy type and time frame. All the strategies have elements that are unconventional and that has the benefit that the portfolio has low correlation with almost everything else. Although we trade short-term models we don't use any that are really high frequency. It's an area we have avoided so far.

Miles Kumaresan: Yes absolutely. I've been a quant trader for 15 years or so and have first-hand experience of a whole range of market conditions affecting models in a big way (positively and negatively). One thing I have learned from all this is that if you are planning to do this for a while, there will be enough surprises around the corner. As such, you build models that are smart enough to adapt to varying market conditions.

One has to go that extra mile to differentiate oneself. There are so many quant traders out there who are focusing on similar opportunities and similar models. They constantly tweak and struggle to squeeze alpha from these opportunities. When structural changes happen they often pay a high price. There is a detrimental bias in most model developers' brains that their model is going to be fine and continue to work for years to come, I am no exception to this. I too was susceptible to the same misconception once a long time ago.

illustration 3

Fred Pederson: We do try and find less popular markets on the one hand. On the other, we try to model correlations that are more and more complex. We are not interested in the low-hanging fruit because that is not where our expertise lies. We have a lot of PhDs of mathematics in the company and we are better at modelling correlations and time series analysis than actually trying to squeeze the last millisecond out of our system. We aren't interested in that, so we put more effort into forecasting based on multivariate correlations. We are definitely not the fastest where we trade. It is important to be quick but it's not important to be doing things down to the last microsecond if you have good strategies. Of course you decrease the slippage if you are faster but it comes at so much cost these days that it's not worth the effort for us. Perhaps one day it will be, but for now we try to focus on the mathematically interesting side of things.

Thomas Parry: I think it goes back to my previous answer. Having more of a view on outright structural movement of the market is vital and then using models more along those lines. It's about how much return on risk we are getting at a given point and also combining views about volatility with that as well.

Peter Wiesing: Yes, that is correct. We seek to trade markets in timeframes and in a flexible manner that is less popular among institutional investors, large hedge funds or prop trading firms. Due to the fact that we are relatively small, we are currently able to make the best allocations to all traded markets, while rarely hitting daily volume limits.

We also benefit from the fact that not all market participants are free of constraints. There are a significant numbers of financial players whose permissions lie outside absolute return optimisation. These players are bound by more rigid investment guidelines. The forces produced by such market participants often influence short-term pricing patterns which force relationships between securities to be inconsistent with overall market prices. These inconsistencies result in short term mispricings between securities which can be exploited.