Cantab Capital

Disciplined processes take Cantab to the top


For many in the hedge fund industry the last couple of years have been tough. The risk-on/risk-off markets have been difficult for managers in most strategies to make good returns from. In particular, CTAs have struggled to an extent not seen in the last 30 years. Having not previously had two back-to-back down years at the index level, returns in three of the last four years have been negative for CTAs (Barclay CTA Index). To this point flows to the strategy have not been impacted – Aspect Capital has gone from $4.2 billion to $6.7 billion, and new daddy of the strategy Winton has gone from $17 billion to $26 billion in AUM since the end of 2010.

Over the same period Cantab Capital of Cambridge in England has grown from a $1 billion systematic manager to running $4.5 billion and has closed. In the last year in particular Cantab Capital has taken a disproportionate amount of inflows to the sector for a good reason; it has made money in each of the last three years. The Hedge Fund Journal saw a recent manager letter which showed the returns given in Table 1.

It is notoriously difficult to nail down what it is a CTA does that is different from the rest. Knowing the capacity of the database or make and model of the computers used does not get an observer very far. There are 38 employees at Cantab Capital Partners, up from 23 two years ago, and, of the 38, 25 are employed in Systems, Research and Trading (up from 16). Is this the right number for a firm of this size? There are relatively few staff in Research and Trading in absolute terms compared to Winton (which has 110) or AHL (which has 88). Does this put Cantab at a disadvantage? The returns say not, even taking account of critical mass, tenure and the constraints of larger size for the competition. It is clear that one of the strengths of the Cantab way of doing things, which will be explored here, is the efficiency of the research process.

Maintaining the culture
There are four individual Partners in Cantab Capital Partners LLP (though the firm’s long-term incentive and alignment vehicle also has an interest in the partnership). They are Dr. Ewan Kirk (CEO and CIO), Erich Schlaikjer (chief technology officer), Chris Pugh (COO), and Dr. Tom Howat (senior scientist).  According to CEO Ewan Kirk, the current challenges of running the business are maintaining momentum and keeping the culture of the firm. “We have to make sure no hubris creeps into anyone’s view of the world. We have done well – beaten the competition, and grown assets. But we have to remember that there is such a thing as a luck factor, and our results need not necessarily have turned out as they have. So part of keeping the existing culture has been to say to people here, to borrow a line from Star Wars, ‘Don’t get cocky, kid!’”

It may be a surprise to some, but having a closed fund gives some management challenges to the Cantab CEO. “We may have to return some money to investors. There has to be careful communication with existing investors. We have to be open and transparent and show that nobody is being disadvantaged as we allocate the available capacity. However, at the same time I recognise that if we are down 15% next year there will be lots of capacity,” quips Kirk.

The CEO and CIO maintains that their investment culture is always the same. “We will always be using a rigorous scientific approach. It is often forgotten that the basis of the scientific method is scepticism. When we have a new model or a new trading system, we are asking what’s wrong with it, how can we break it, and when doesn’t it work rather than saying to ourselves, ‘This is great, isn’t it?’ In essence under this scientific approach, we are coming up with things that we can’t prove are wrong,” he says.

Kirk explains that the second element of the investment culture is the investment that Cantab Capital makes in technology. “We live and die by the technology that we have, the software that we build and our development processes. There is a focus on making what is already a great system better and better and better. It is a constant development process, and refining and improving what we have already got takes up half of our research time.”

Research architecture
Ewan Kirk continues, “The secret sauce of this type of investing is not in the models – what makes a good systematic firm is a great research architecture.” The emphasis from the CEO on the research architecture points to the significance of Dr. Tom Howat to the firm. Now he heads up the team of scientists responsible for the firm’s infrastructural development, and he was responsible for much of the design of the framework for back-testing, trading and real-time risk management of the investment strategies. He was the person who solved the problem of allocation to managed accounts referred to later. These tasks, responsibilities and achievements may be sufficient in their own right to justify being invited to join the partnership that owns the firm, but Tom Howat’s partners quickly cite his personal qualities when his name comes up, and it is his personal attributes as much as anything that made him partnership material.

Erich Schlaikjer, the CTO, describes meeting him for the first time. “Only a few times in my life I have interviewed people and thought, ‘This is a slam dunk’. Tom is really into computers, but he is really into lots of things, and we like curiosity in our people. Smartness is a necessity at our firm, but getting things done is very important. We want people who can churn stuff out and want to see their work used, and Tom has written more code than anyone else at the firm. It was a good fit from the start – he believes in our approach to working. His attitude to a rising issue or project was always, ‘Yeah, we can do that.’ Even when it involved a lot of work he could get it done.”

He continues, “Tom is unusual in a couple of ways for someone operating in the overlap of science, maths and technology. For one, he is happy to read other peoples’ code. More importantly he inspires others, and is good with other people himself. He doesn’t get sick, not even a cold, and he is so cheerful we want to clone him,” says Schlaikjer with a smile. “Tom drank the Kool-Aid, and it will be challenging to a next partner to match his contribution,” he summarises.

Each of the senior managers concurs that the culture of management of the firm is of a very flat hierarchy. Ewan Kirk explains how that works in practice: “After we have taken care of our specific duties (as CIO and CTO etcetera) we are all doing the same work here. I was just doing some programming – I was hands on because I had to get something fixed. I spend as much of my time as possible doing programming –  writing scripts, coming up with models, and testing models – doing the same job as everyone else here.” Kirk is happy to declare that the firm was custom designed by himself and CTO Erich Schlaikjer in a way that allows them to spend their time doing things that they enjoy doing. In doing so they have achieved something that most people struggle hard, and for a long time, to vainly try to achieve. “I like programming, I like maths and I like doing the things that we do here,” he discloses contentedly.

One of the structural decisions for large and medium-sized firms running systematic approaches to markets is the degree of separation and integration of the parts of the process. It has to be a management challenge for firms that have research teams in different countries and time zones, and there are some advantages to being all in the same place as opposed to spread across the world.

Click to enlarge image

Cantab Capital has a much more integrated and flexible way of operating within its one, mostly open-plan, office. “We have a collaborative and collegiate approach to our processes. We try to avoid people specialising too much,” explains CIO and head of research Ewan Kirk. “Our staff members do take individual responsibility for things: they might take responsibility for shepherding a model through the process. They may not have come up with the trading idea themselves. It may happen that they have an interest or an aptitude in that area. It will be up to them to prove it wrong (to go back to the scientific method).”

A key component of the Cantab Capital way of doing things is openness. For example the strategy presentations and code reviews (‘Big Meetings’ in Cantab-speak) which take place at least twice a week may be attended by anyone. In these meetings underlying code or the models are projected onto a big screen in the meeting room for all to see and critique. Everyone is involved in the processes, and the same way of working is applied on the quantitative side and on the programming side. The open meeting philosophy also applies to the weekly run-through of risk reports, though far fewer staff feel the need to get involved with that.

Openness applies in other ways. All the models, all the data, all the software within the firm are completely open to everyone at Cantab. The CIO says that the IP in the models is not as large a proportion of the models as many people think. “Alot of systematic firms consider the buy and sell levels to be proprietary, but that level of detail is not the edge for them. Turning a model into a trading system is a lot more difficult than having basic software and an ability to programme. The big trading firms between them have spent billions of dollars on technology and that is what makes a difference. It is about infrastructure.”

Research productivity is driven by discipline
Erich Schlaikjer, chief technology officer, warms to Ewan Kirk’s theme, putting an emphasis on software, discipline and the creation of a robust framework. He says “investors usually ask about the models Cantab employs – that’s the easy part. The hard part for a systematic manager is the technology, particularly the software. That is where they can be differentiated.” The CTO acknowledges that there is some hard signal processing in the models, “but the smarts, the core of what we do, is in the software. It is important to understand that we have built a framework here. We have created a platform for doing the creative stuff, and in doing that we have put software at the centre of what we are doing.”

This has a precedent in the work life of the Cantab CEO and CTO. “The dirty secret of what Ewan and I were doing at Goldman Sachs (in the Quantitative Strategies Group) was that we were coding all the time,” says Schlaikjer. “At the time, writing code was what the less well paid programmers did, so we did not make it widely known what we were doing minute by minute. The significance of technological systems to outcomes is not generally appreciated. I picked up from lecturing new graduates at GS that there was good awareness of the problems of ill-considered investments and hedging (such as the Orange County derivatives mess, and the problems of hedging at Metallgesellschaft) but not the failures in large technology projects (such as LSE’s Taurus system).” The lesson for the CTO’s work at Cantab Capital was to create a system architecture that was robust in a way not always seen at investment banks or CTAs.

Schlaikjer expands on his point: “A typical method in CTAs has been for a researcher to be given a disk full of data and a copy of MATLAB, and then, having come up with a trading idea, he has to hand it over to some IT guy and never sees it again. We set out to do the opposite of that. Cantab has a framework for research – once the idea has been run, all the back-testing, the correlation with other strategies and the relevant statistics come to the researcher as a step in the process.” The researcher does not generate the measures of effectiveness themselves, which is good for independence of the statistical feedback. Also, the feedback is generated as a routine from the software. Producing the results of trying out the investment idea does not consume the researcher’s time. This helps with the productivity of individual researchers.

Other methods help to achieve the same goal of researcher productivity. “We have a fanatical devotion to the discipline of doing something once (in a robust way) and moving on,” explains the Cantab chief technology officer. He says that computing departments are often unstructured enough that they solve the same problem over and over again. That cannot happen at Cantab Capital.

“A lot of the ways of being effective in research are about discipline,” he says. “Nobody re-writes a Monte Carlo engine – we have just one, and everyone must use it. It is a natural impulse of programmers to want to avoid reading other people’s code. They want to do it all again from scratch in a new language, (that is), in one they understand and are familiar with. Researchers may not re-write things here – first they must look for an existing copy of what they want.”

Part of the research process at Cantab Capital is a code review. There are always limits to what can be achieved with programming, and the CTO observes that sometimes programmers over-complicate things. The natural clarity of mathematics can be obscured by how the concept is encoded. This comes back to the intellectual property and open nature of how Cantab operates. If the code is written in a way that is readable then the IP remains with the firm wherever the programmer ends up. So at Cantab nobody is allowed to write routines in an obscure language that cannot be readily read by all.

At Cantab Capital the quants apply their smarts to activities in addition to the investment strategies – in execution, for example. All CTAs have a string of managed accounts – usually the capital in managed accounts is a multiple of that in the funds. Trades get executed in various lot sizes and prices and the trades have to be allocated across the fund and a dozen managed accounts in an equitable price (to ensure best execution). Erich Schlaikjer describes the issue: “The problem that arises is that the trades have to be allocated to the various accounts such that the trades per account aggregate to close to the average price across all of them. This is a famous hard maths problem that is an issue for operations, so that quants in many firms would not look at that kind of thing. Here we had a lot of fun solving that problem by automating it, and every day we look at reports which tell us the unfairness quotient of trading per account. It can be as low as $100 in cash terms across all the managed accounts.”

There is similar attention to detail in the change-test discipline applied to each new investment strategy. “For each set of recorded inputs for a strategy we have recorded outputs,” explains Schlaikjer. “This enables us to run it again every single day and see if there is any change to the outputs. If there is a change to the outputs we know that there has been a change in the millions of lines of code – it could be anything, an interpolation routine or a holiday somewhere. We also run these change-test disciplines on both the Linux and Microsoft platforms (32-bit and 64-bit versions) to make sure that the outputs are independent of the operating system environment.” It is these kinds of disciplines which accumulate into the rigorous framework for research at Cantab Capital.

Tom Howat, the senior scientist and newest partner, concurs that a very disciplined environment is needed. “We have an institutional development environment here. It is a good fit for me because I was used to source code control systems. These are the systems that tell you who has changed what and when.” Some of the technology-specific axioms that are applied at the firm are listed below.


• Don’t be tied in to a particular vendor, and minimise dependence on externally provided packages.
• No logic in SQL – that can get really messy.
• Do not mix original and computed data.
• All code must be written in a limited number of languages so that it is accessible to all 

Cantab Capital researchers try hard to avoid the common data-mining pitfalls. So for example, unless there is a good reason, the strategies have to work on every asset class, and on every time-frame. In testing investment strategies certain time-frames and asset classes are embargoed to ensure that there is minimal data-fitting. The researchers are not allowed to use those embargoed time-frames and asset classes until the end of the testing process. The management view is that, say, a trending strategy should work in every asset class and on every time-frame. There have to be some exceptions, like yield curve trades which canonly apply in fixed income (an asset class which has had a big risk allocation and delivered successful results for the funds). The generality is that researchers who discover that a strategy doesn’t seem to work on gold and silver will not be able to carry on with that project.

The development of a new strategy has to be implemented in a standard way. The results of the strategy have to be analysed in the standard reports. The strategy has to be presented in the Big Meeting, including the genesis of the trading idea. It has to be demonstrated that the strategy is complementary to the existing strategies. The person or people working on it have to have thought about what can go wrong with the strategy. That cycle of revising the strategy and presenting it has to be repeated a number of times before the last couple of stages in the process can be done. A new strategy has to go through the code review process and then paper trading before it can be run with real money. In paper trading the new strategy is run alongside the existing ones to see its impact at the portfolio level. There are tests on scalability too: “We want to make sure we don’t need to buy all the Swedish krona in the world to implement the strategy in our size,” laughs Erich Schlaikjer.

One team with fuzzy boundaries
An element of the all-one-team approach is that as a group they try to get consensus on how things should work. So, for example, if someone comes up with a better cost model, there is a desire to fit it seamlessly into each component of Cantab’s systems so that everyone benefits. The CTO’s take is that “if everyone is using the same back-test and this better cost model then everyone benefits, not just the originator.”

Staff are encouraged not be territorial or to stay in their box. Tom Howat gives an example: “We have a weekly meeting in which we update each other on what we are doing. There is a meeting which is mostly programming and there is a meeting which is mostly strategy development, but they are not exclusive. On the implementation side it could be that a programmer was reading about a new database design that might speed up a particular sub-routine – he is allowed to contribute that.” So there is cross fertilisation between quants and programmers. Howat is a ready example himself – nominally he is in programming, but his deep mathematical background (a PhD in mathematical biology from Trinity College Cambridge) allows him to get involved to help out the quants at times.

The strategies are tested to find out how robust they are. Erich Schlaikjer explains how: “All software has knobs, and sometimes we deliberately turn the dial until the software explodes. We do that by forcing the strategy to trade more and more, or feeding the model increasing noise rather than real data.” As a consequence the scientist might run out of memory, or data limitations arise. At that point more senior colleagues can point out that the firm has an object for testing short-run moving averages, and that the researcher should use the tool that is available. “The key to our efficiency in research is discipline in working with and writing software,” states Schlaikjer.

With all the discipline and framework at Cantab Capital there is still space for individuality, and for pursuing an interest. Google famously encourage their staff to set aside their day-to-day work every Friday to explore new ideas and new technologies. In the same spirit Cantab has acquired a 3D printer and what was described as a multi-superprocessor for uses that are as yet unclear. An equivalent of a Google Friday project at Cantab is that one of the programmers has downloaded a huge database of text and news and is mining it. Neither Ewan Kirk nor Erich Schlaikjer have any expectation of this turning into a strategy, but as the CTO acknowledges, “You never know.”

The senior management of Cantab Capital are proud of their great team spirit and that it is a cheerful place to work. Whilst that is easier to achieve when all the front office staff are within 20ft of one another, that is not the end of the ways the firm tries to build camaraderie. The staff interacts on a social level too. There is the classic techies’ play room (which has a piano as well as the more typical fussball table); the staff take jaunts together to do things like bowling and dragon boat racing, and they also jog together by the river at lunch time.

A recent development is the appointment of a “Data Tsar” at Cantab Capital. According to Erich Schlaikjer, some of the quants have become more focused on getting the data absolutely correct. For example, new issues have arisen from moving to more frequent trading. Schlaikjer explains, “When we traded once a day we could take closing prices. But now we trade once an hour. So there is a 9 o’clock portfolio and a 10 o’clock portfolio, and some markets are not open at 9.00, or perhaps they are open at 9.00 now, but weren’t until two years ago. So there is an effort to put together a 10-year time series. We have a lot for the Data Tsar to do.” Cantab has bought databases of raw data from the likes of Reuters and the exchanges, and also gets (traded) volatility surfaces provided by various brokers. The new appointee will get the databases organised so they can be accessed and read in a straightforward way. “He will have care for the storage of the data and the meaning of it,” according to Schlaikjer.

The CTO looks forward a little and raises a couple of the big discussion points at Cantab Capital. There have long been discussions about how big they want the firm to be – in one dimension that has been fixed for now with the hard closing of the funds. In a second sense, of staff numbers, the partners have agreed that they don’t want to get bigger than 50 people in investments (against 27 currently). A second long-standing discussion point has been what do they want to trade? Schlaikjer gives the latest thinking: “We started out trading futures and FX because they were the easiest and most liquid markets to trade. But our systems can trade anything, and could from inception. We have the technology to trade cash equities, CDOs or whatever. We have limited ourselves because we are wary of how our clients will react to us adding dissimilar assets to the mix. To a degree there has always been a desire to try different flavours of funds; for example we occasionally flirt with hiring an equities specialist.”

Differentiators and chicken
At this point we come back to the differentiators. In the eyes of the management the technological framework for research they have facilitates efficiency in that area. As important is the way of working which is collaborative, and Cantab Capital research staff are not constrained to staying put in a silo. As a consequence of the structure, technology and human capital the research work is very productive. Supporting evidence comes from the growing number of underlying sub-models: there are now over 800 of them. Those models are applied to 120 markets now, up from 80 two years ago.

A second differentiator is the risk allocation process and resultant allocations to markets or time-frames. The allocation of the risk budget at Cantab Capital seems to takes place with more degrees of freedom than seen elsewhere. The allocation process is responsive to markets, and gives greater scope for wider bands to the risk buckets than at most systematic firms. The bias in the risk budget to futures on short-term bonds and rates has served Cantab well from inception and particularly in the last two years. There has been a good payoff in returns for investors from having stuck with allocations of capital to the short-term strategies bucket – which were a drag on returnsfor the middle years of Cantab’s trading history.

The weekly lunch order is indicative of the mindset at Cantab Capital. Once a week orders are gathered across the staff. Ewan Kirk always orders chicken, and always gets out-voted. Rather than allow two choices across the firm, the CEO and CIO of Cantab Capital suggests that they should have a lunch algorithm that means that those that don’t get their preference one week get a bigger weighted vote in the next week. This is an ongoing debate within the firm – the discussion of development of an algorithm for ordered-in lunch is as regular as the Code Review Meetings. This debate says a lot about the firm – the flat hierarchy, the discussion, the collaboration, the mutual respect, and the use of coding to solve a problem by algorithm.