The Gateway to Algorithmic and Automated Trading

My Machine: The AT Interview - Amplitude Capital: Precision Automation

Published in Automated Trader Magazine Issue 03 October 2006

Automation, both in terms of trading and back office processes, is an intrinsic part of Amplitude Capital’s day to day operation. AT talks to two of Amplitude’s principals, CEO Karsten Schroeder and CFO John Harrison about the firm’s technology, methods and ambitions.

Amplitude Capital’s CEO Karsten Schroeder and CFO John Harrison
How did Amplitude start?

Karsten Schroeder: Originally I was working at McKinsey, but had a long standing interest in trading methodologies. For some five years prior to starting Amplitude I was researching ideas. During this period I also worked with Peter Voss, who is now Amplitude's COO and Steffen Bendel, who is our CTO in developing a trading platform that was the precursor to our current production platform.
Shamil Chandaria, Amplitude's Chairman who has an extensive background in the hedge fund industry, subsequently became involved in evaluating the strategy we had evolved. He took the view that this could form the basis for a commercially successful fund management company, which was duly launched in September 2004. Initially we were trading proprietary capital as a means of further validating the strategy in real time, with the first Amplitude Dynamic Trading Fund being launched in June 2005.

John, how did you become involved?

John Harrison: The catalyst was Shamil, whom I have known for many years. I don't have a trading background, being a corporate financier by training, so my role at Amplitude is very much concerned with finance, legal and operations rather than investment management. After discussions with Karsten and Shamil, I became involved in March of last year which was after the launch of our corporate businesses but prior to the launch the fund. As a result, there was a need at that time for more emphasis on the areas that I now cover, so the timing was rather felicitous.

How many employees do you currently have and how are they distributed across the various functions?

KS: We have ten full time employees, with six of them involved in research & development, programming and trading. These six include our CTO, Steffen Bendel, who oversees these activities in conjunction with me. The running of the trading platform and systems are supervised by Peter Voss, who also acts as compliance officer. Marketing, business development and investor relations duties are covered by John, Shamil and me. Shamil and I are the two Portfolio Managers.

How have assets under management grown since the foundation of Amplitude?

JH: At the end of 2004 we were trading with USD2m of proprietary funds. This had risen to USD19m (including the proprietary funds) by the end of 2005, following the launch of our fund in June 2005. Assets under management have risen steadily throughout 2006 and today we trade a limit of over USD95m.

How would you describe your trading methodology?

KS: Our trading strategies attempt to capture short market movements and are applied in parallel across each instrument. There is no discretionary element whatsoever - everything from signal generation to order placement to reconciliation is automated and done electronically. Individual strategies are based upon a range of methods. For example, what would traditionally be categorised as classical trend following represents some 60% of our methods and includes techniques such as moving averages, oscillators and breakout systems. We have a countertrend mean reversion element of approximately 40%. All of our indicators are proprietary.
Amplitude's offices: outside...

Is automation a key element in Amplitude's operations?
JH: Yes - most definitely. There was never any question that we would function in any other way from the outset. Even when trading proprietary funds prior to the launch of the first fund everything was automated.

How long do you keep positions open? Do you vary the time slicing of data you input to your models?

KS: It usually varies from as little as perhaps two hours up to maybe a week, but the average holding period across all trades is approximately one and a half days. And no, we don't vary the density of data we feed to our indicators.
In terms of trade frequency, we typically have an average of eight to ten trades per instrument per day. That trade frequency tends to increase during unprofitable periods, but decline in the presence of stronger trends, as positions stay open for longer. While the win loss ratio of our momentum/trend related strategies may be similar to many "traditional" trend followers, our higher trade frequency and relatively short average trade holding period are two of the factors that make our overall performance less dependent on just a handful of large winning trades. Our mean reversion strategies, with a much higher win loss ratio (approximately 80/20) provide an additional source of smoothing for our equity curve.

But you would still describe Amplitude's general approach as essentially trend following in nature?

KS: Well yes, but perhaps we are not necessarily trying to follow the same trends as everybody else. Also bear in mind that we have a reversionary element among our strategies. We would nevertheless tend to benefit from periods where trends were present and volatility higher. However, we do not regard those two conditions as synonymous - up to a certain threshold increased trending may go hand in hand with volatility, but at very high levels of volatility we find that this relationship no longer holds.

In view of the fact that this relationship breaks down under those circumstances, do you make any attempt to predict periods of very high volatility?

KS: It would be very nice if this were possible! Unfortunately we have not as yet discovered a robust way of accomplishing this.

Does the higher frequency of your trading strategies and the fact that execution is automated dictate the markets you trade?

JH: Yes, to a considerable extent. Only extremely liquid markets that have electronic access are acceptable. At present that translates into some twenty six instruments, the majority of which are exchange traded futures. The only exceptions are a handful of highly liquid FX pairs that we trade spot on Deutsche Bank's Autobahn system. The liquidity in both the spot and futures markets we trade is more than ample; we have estimated that as much as USD750m could be applied to these markets using our existing strategies without adversely affecting performance.

Nevertheless, do you intend to expand the range of markets traded?

KS: This is the subject of ongoing research, so in addition to the markets currently traded we are monitoring an additional twenty or so instruments for possible future inclusion.

Do you use a fixed capital allocation per market? Or does this vary, for example, during drawdowns?

KS: We currently allocate an equal weight to all the instruments we trade. However, there is a money management layer that may vary exposure to certain markets where conditions are deemed unattractive. In terms of risk management, we attempt to target the same risk level on all instruments on the basis of their 20 day historical volatility. There is no discretionary element as regards varying exposure during drawdowns or at any other time.
...and inside

Do you apply the same models to all markets or do you have market specific models?
KS: No - we have an inventory of strategies that can be applied to each instrument, but only the most appropriate of these strategies are used for each instrument.

Is further research a large part of Amplitude's day to day activities?

JH: Very much so. All research is done in-house and can be categorised into three main areas: new markets, new strategies, and high level issues such as money management. Substantial operational funding means that we are able to maintain our research efforts independently of external fee revenues.

How is the research process structured?

KS: The development of models follows a strict regime, where they are all tested in isolation to avoid over fitting. The actual system design process uses all available tick data in conjunction with a monthly rolling optimisation cycle, which re-optimises all the free parameters of each formula.
Research is not conducted on an ad hoc basis, but discussed, planned, prioritised and assigned to individuals at research meetings, which take place at least once a week. We take the view that an unstructured research program ultimately leads nowhere.

Is there such a thing as a typical development cycle?

KS: No, it varies considerably. For example, developing and backtesting an indicator or idea to see whether it is viable might only take a couple of weeks. On the other we have had one major research project relating to the parameterisation of models that took nine months of development time. One tends to find that projects that are not dedicated to just a single model take appreciably longer than those that are.

Is the in house optimisation technology you use particularly unusual? Is it a critical element in your research?

KS: There are rather limited options in terms of optimisation techniques (e.g. neural networks, genetic algorithms, Monte Carlo or exhaustive) but it certainly requires a substantial infrastructure. We made a substantial upgrade to our internal technology recently and we are now able to do optimisation in a way that simply would not have been possible one or two years ago. I would say that the way we do optimisation now is on the edge of what is practically possible from a technology point of view and that it is an integral part of our research process.
Amplitude Capital’s work force
The Amplitude team

How many different areas of research might you have running concurrently?
KS: When we have a major project underway, then the majority of the team will be dedicated to that. The allocation of research resources will depend upon how promising we think an approach might be and its urgency. We also tend to break things down so different people will work on different aspects of a particular problem or project.

How frequently are changes made to the production trading models?

KS: New ideas are only introduced to the trading program where we feel they will add value and are not strongly correlated to the existing set, so there is no regular cycle of replacement. Any upgrades to the model or new markets will be traded on our proprietary account. If that is successful then (and only then) might they be incorporated into live trading of client money. The crucial point is that we will only introduce new ideas or make refinements if they add to the overall return and reduce risk.

Which do you see as the most valuable form of diversification -- by timeframe, by market or by model?

KS: I really don't believe you can answer that question in general terms because timeframes, markets and models can have such a broad range of internal correlation. For example, two trend following models on the same market will provide minimal diversification, while mixing a mean reversion and a trend following model will give a high level of diversification. When you have a model, you have a hypothesis as to whether it makes sense or not and you test that. It is very hard to say how correlated or otherwise a potential new strategy will be with your existing models before you have completed all the testing. We try to prioritise and pursue routes that are the most promising and we are of course looking for a diversification element, but we do not integrate models that are not profitable in their own right just for the sake of diversification.

All the software technology you use in the development, trading and settlement processes was built in house. Were there no off-the-shelf systems that had the necessary functionality in any of these areas?

JH: It was immediately apparent to us that no vendor solutions offered the functionality we need for research and trading. In view of that, we felt that we might just as well develop our own middle and back office software as well. We had also observed that integrating third party software sometimes took as long as to write software in house. A further consideration was support and maintenance - with software built in house, problem resolution is immediately available and one is not at the mercy of a third party as regards maintenance and the addition of necessary functionality.
Where the numbers are crunched: Part of Amplitude's machine room

Does this in house preference even apply to commonly-used statistical and quantitative packages such as MATLAB or S-PLUS?
KS: Yes, we don't use any vendor software for calculations/optimisation in our R & D process, though we do use programs such as Excel for plotting charts etc.

Does Amplitude trade on a 24 hour basis?

JH: Yes, we trade nearly 24 hours from our offices on a shift basis. During each shift one or more authorised staff members controls and supervises the automatic trading, so there is at least one dedicated person monitoring trading at all times. Designated personnel are authorised to execute trades by phone in the event of line failure.

How do you handle the opening of large positions?

KS: Our automated execution strategy is responsive to current market conditions. It uses market order book data in order to finesse the placement of trades that make up a large position.

How has your automated trading affected the arrangement of middle and back office functions?

JH: The middle and back office functions have also been automated. They are capable of performing reconciliation with brokers every fifteen minutes, though in practice this is actually done every three hours or so. As we only trade liquid instruments with excellent price transparency we don't have any issues around marking to market etc.
We obviously have a high daily trade count so this is an important area for us, as we don't wish to risk accumulating a backlog of un-reconciled trades. If that happens, one is left with a lot of clearing up to do at month end before audited NAVs can be produced.

What about disaster recovery?

JH: We mirror our open trading positions but we don't mirror the entire infrastructure. The cost of replicating our trading servers and Radianz and Reuters connections would be considerable. The only time you would use that duplicate infrastructure would be if the office was physically destroyed and then there would obviously be a number of other issues to attend to. Therefore we feel it would be simpler just to liquidate the open positions and reopen them in a few days when other matters have been resolved.

Co-location has become popular among some automated trading shops. Is this something you have considered?

KS: No, not really. At present we don't consider this a worthwhile option in terms of latency advantage. Having our trading servers located near the exchange might at best shave off a few milliseconds, but in the context of other sources of latency and also the timeframe of our trading this is insignificant.
Amplitude Capital’s CEO Karsten Schroeder
Schroeder: "Automation is obviously essential to what we do in terms of the range of markets, models and timeframes we trade."

You appear to have the infrastructure that is capable of handling a far higher frequency of trades than you currently execute. Do you intend to trade at higher frequency?
KS: At present, no. While our infrastructure is indeed capable of handling a very large number of trades, our trading activity is not driven by that. There is obviously no point in operating at a frequency that is unprofitable just because it is possible to do so. The timeframe and frequency that we currently trade is optimal for us in terms of performance, slippage, costs etc.

What would you say distinguishes Amplitude from other managers?

KS: As to exact trading methodology, one can't really say. We don't know the detail of what our competitors do and they don't know what we do. In more general terms, I would say that our research process is rigorous and that we therefore can have a good degree of confidence that the robustness of our simulated results will be replicated in live trading.
I think the degree of automation we employ is also significant. Automation is obviously essential to what we do in terms of the range of markets, models and timeframes we trade. However, it is similarly important beyond the trading function in terms of STP etc. The net effect of our automation is improved efficiency in terms of breadth of trading, reduced frictional costs and minimal error rates. This in turn is very much connected with our original decision to build our automated technology in house, which has given us greater freedom in terms of flexibility, extensibility and maintenance.

Plans for the future?

JH: We are very optimistic about the future. We have successfully navigated our way through many of the pitfalls that can catch new businesses and we believe the product and platform we have built is strong and very scalable. From a corporate perspective, we shall continue to develop the tailoring of our product range to large institutional investors, especially by promoting our managed accounts facility, and we shall consider the virtues of a US feeder fund during the next year.
KS: More research! Our commitment to this is absolute. It is our capacity to stay ahead of the game that will drive the business in future years. We shall focus on developing and exploiting market opportunities that add value through continued application of innovation and technology.