 Philippe Bonnefoy: Eleuthera Capital AG |
When the blue blood Wall Street firm Kidder, Peabody & Co. set up a global fund management business in the mid-1980s, it tapped one of its new young stars, Philippe Bonnefoy, to look after international equities. Overseeing such a portfolio meant managing foreign exchange exposures and for Bonnefoy it was love at first sight. Within a few years he joined a proprietary trading firm that capitalised on a volatile and fast-changing macro trading environment for the next decade. In the mid-2000s, Bonnefoy started a venture using algorithms to seize short-term FX opportunities, a trading technique he has been honing ever since and which is central to the success of his firm Eleuthera Capital AG. Throughout the years, one thing has been constant: an appreciation for how technology and trading can go hand in hand. Adam Cox catches up with the Zurich-based fund manager.
AC: Let’s start by talking about how Eleuthera uses algos. Could you describe the main objective and how the algos you use go about achieving it?
PB: We have traded currencies for 30 years. The development of the use of computers to assist, analyse, back-test and now manage the trading process has been extraordinary. When we started, there was basically a Reuters terminal, a T-1 phone cable or satellite dish, we had telex machines in the trading room and a computer screen connected into a chilled computer room somewhere in the building. Now every price tick of the 30 developed market currency pairs we follow is consumed by our co-located servers in London and NY where it is analysed and filtered.
The objective of our trading system is to produce attractive, uncorrelated risk-adjusted returns that are not overly factor dependent and generate these positive returns through the cycle. Unlike much of the peer group we do not want to rely on carry and we do not want to rely on trend. In order to make these returns more consistent we mainly specialise and focus on intraday trading. In that way we avoid the large sudden drawdowns that are common in trend following funds where trends reverse and the losses that are generated in periods of range trading. We are not correlated to any other asset class or any other currency manager.
AC: Your website says Eleuthera targets an average annualised return of 15-25%, with a target annualised volatility of 10-15%. What are the main strategies you deploy via algos to achieve that? What sorts of timeframes and trade frequencies are involved?
PB: We have four main model groups: mean reversion, short-term trend, short-term breakout and statistical arbitrage. In the deployment of these model groups our objective is to be able to generate positive returns in almost any market environment. The worst market conditions for us are periods of low volatility with low intraday price dispersion. The best market conditions are when markets are noisy, boisterous and volatile. It is interesting to note that this is almost inversely correlated to equities which typically thrive in low volatility and suffer in periods of high volatility – so, by design, we make an excellent portfolio diversifier for many investors who have large equity portfolios.
As you can imagine, central bank sponsored quantitative easing made our life difficult over the last couple of years. G7 FX volatility hit an all-time low on 3rd July 2014. In the second half of the year the US began to exit QE and as policy between central banks diverged, volatility rose towards its long-term average. As such our performance also lifted back into our expected target range. We are up around 14% in the first couple of months of 2015. We do need price movement to generate trade opportunities and thankfully price dispersion amongst the developed market FX pairs has returned. During this period of artificially suppressed trading daily trading ranges we created several new models to cope with this new paradigm. Having traded through many market crises and abnormal periods we live by the motto of the US Special Forces: “Improvise, Adapt, Overcome.”
 |
AC: A lot of firms had to improvise in January when the Swiss National Bank dropped its bombshell. Whether they were able to adapt and overcome is another matter. How did you fare?
PB: The SNB, with whom we’ve met before, seemed to violate the one common tenet that all central banks have, that is they view their role as to provide orderly markets for the market participants. This surprise announcement, three days after giving a speech saying that this currency peg was a core part of their policy, was really bizarre. First, they created an artificially constrained price environment then suddenly abandoned it, creating another artificially distorted price environment. This is one of the more respected institutions in the world. As such, it was a big surprise, hence the significant market overreaction. And it’s interesting to see that many of the Swiss franc crosses are today back to similar levels to January 14th. We owned some positions before that announcement. We had found sterling weak and Swiss franc strong going into the overnight session, so we were long CHF and short GBP into the move. All told, the models did what they’re supposed to do and we made a relatively outsized gain.
AC: Now that volatility seems to be coming back, what does that mean for your trading?
PB: Markets are in an almost perfect place for us to make money. Volatility is like Goldilock’s porridge in the sense that you don’t want it too hot and you don’t want it too cold, you want it just right. We are now around the multi-year volatility averages for the developed G10 markets. That is really a very good place to be because it means that the major liquidity providers to the market – basically the top 10 to 12 banks and a couple of HFT firms – are happy to price us competitively in market conditions which allow them to effectively capture the price spread. Most of the big banks have an algo business themselves, which is systematically auto-hedging their market positions. They accommodate client risk transfer business then they do one of three things. They add to their own proprietary trading book, they cross the positions to other clients (and pocket the spread) or they sell out the position to the market often across multiple trading venues. This can also be done by changing their pricing of currency pairs to become the best bid or offer. If nothing is going on in the market it’s challenging for them both to take that client business on board at tight pricing and to exit these positions profitably. So, unusually low volatility makes it harder for them to trade profitably and to have an appetite for this risk transfer function. As a client that is bad for us. High volatility forces their algo traders and their risk management to reduce their risk appetite and potentially widen spreads, which is further trade friction for us. But at least in periods of high volatility we have many more potentially interesting trading opportunities.
AC: How long have you been using algos to trade the currency markets? How has the usage evolved over time? How often do you tweak the algos you use?
PB: We began deploying algos in 2008, initially as tools to assist our discretionary trading. We quickly realised that using algos was a much better way to manage short-term trading. Typically “tweaking” of the models happens if we find new and better ways to make our automated risk management functions more effective or if we can add further filtering to improve the trade selection process. We do introduce new models if we find strategies that can further diversify our trading strategies such as in the recent period of low market volatility and narrow intraday price dispersion.
AC: What are some of the issues, either positive or negative, that a fund such as yours faces now that weren’t factors in the past with respect to using algos?
PB: The positive effect is that using algos can reduce our trading costs and enhance our performance because we can implement trades much more quickly than if a human was involved. Indeed, one of the issues facing the sell side is how to stream prices that do not put their human trading clients at a disadvantage. We have been told of conversations where traditional traders have complained that they cannot push the button to trade fast enough because by the time they have reacted to a price movement that price has moved away. As the trading latency with a number of our liquidity providers is under 10 milliseconds, that is not a problem for us.
AC: What are some of the unique factors involved in trading FX that one doesn’t encounter in other markets and how do your algos address those factors?
PB: A unique feature of the spot FX market is that as an OTC market you are reliant on your trading parties streaming consistent tight pricing 24 hours a day, 5 1/2 days a week. That means it is incumbent on us to have a healthy symbiotic relationship with our liquidity providers. We spend a lot of time discussing how our algo strategies work with those firms that stream prices to us. If our trading behaviour was viewed to be predatory or parasitic it would be a lot harder for us to generate profitability as the pricing we received from our counterparties would become more defensive. In the past we did remove an algo because we felt the flow was too “hard” for our counterparties. It was an interesting situation as this algo was executing a mean reversion strategy which added liquidity to our counterparties in times of market stress. The unexpected outcome was that our liquidity providers were auto-hedging their own risk exposures so quickly that our liquidity that should have been welcome by our counterparties wasn’t helpful because they had already transferred the risk out to the market. This algo was profitable for us – but it was not profitable enough for us to want to negatively impact our trading relationships. We don’t have a high frequency trading strategy which some liquidity providers may find difficult to price. Our trading flow is generally positive for our counterparties which is how we want it to be. All participants in this relationship need to be profitable.
AC: Are you focused only on spot markets or do you algos to trade in other markets such as forwards or currency futures?
PB: We only want to trade markets where we know there will be consistent liquidity. As such we limit our investment universe to spot trading G-10 developed market currency pairs and cash gold.
AC: What kind of backtesting do you do?
 |
PB: Our backtesting is so rigorous that we have several servers dedicated to backtesting 24 hours a day all year. We had one test running recently that took almost three weeks to be completed running 24 hours a day non-stop. We are testing every market tick for 10 years. Let me put this in context, to be conservative let us say that there is a tick change every five seconds (it can be multiples of this in a busy period). That means for a 24 hour, 5-1/2 day week the data set could be around 50,000,000
price changes over a 10-year period. If we were a long-term trend follower that only analyses price activity on a daily basis One of our tests would be the equivalent of a doing a test of 50 million trading days, or about 200,000 years.
Our models are typically highly filtered, multi-factor models containing a number of proprietary non-technical inputs so there is a lot of complexity in the filtering of how the models will look at each price change. Each trade we do has pre-determined volatility adjusted risk limits and profit take objectives. Each tick will be sampled to see whether a position should be entered, stopped out, profit taken, exited, or added to. That is the reason that this backtest process can take so long. We only look back 10 years because the market micro-structure has changed so much that we do not believe that price action pre-2005 is relevant to today’s market. As an example, in 2005 spot trading was mainly done by voice. Now it is almost entirely electronic. As intraday price behaviour has changed enormously older price information is less helpful to analyse. The last 10 years also encompasses almost every single type of market – financial crisis, equity flash crash, strong trend, no trend, record high volatility and record low volatility and now the SNB event – so it is a great data set with which to work.
AC: How do algos assist with risk management activities?
PB: All risk management is embedded into a trading position prior to entry. Each trade we execute has its own distinct risk return parameters. Position sizing is determined by the risk budget given to each of the models, which in turn is then input into any trade which that model executes. There is a predetermined volatility-adjusted risk limit for each trade. This “stop” can be moved closer to protect open profits and reduce risk exposure but it can never be moved to allow for a position to have greater risk.
AC: How are your algos developed? Did you source any algos externally and customise them or develop everything in house?
PB: My partner, Riadh Fessi, and I have been developing trading strategies and coding algos for eight years. They are all designed in house. They are reviewed regularly and we will update the models if we believe that we can find ways to further reduce risk or screen out sub-optimal trades. We are always thinking of new trading ideas and new algos with which to express them.
AC: How much is automated? Do you ever override the algos with human decision-making?
PB: The process is completely automated. There is no manual override unless we believe that there is an issue with infrastructure which negatively impacts the performance of the trading models. If the models are operating normally we don’t second guess them.
AC: Looking ahead, what are the main technological changes that are likely to be important for the FX market and for firms such as yours?
PB: The high frequency trading world suggests that we should always push for reduced trading and data transition times. This is an area of diminishing returns where one will always have to scale up the technology to get the last millisecond or microsecond out of it. We’re not in the HFT business but we’re very sensitive to not wanting to be at the long end of a queue of people who are reacting to news or getting stopped out of a trade. Our servers are located in the major data centres in the world and hardwired to the banks servers. This considerably speeds up price transmission and trading execution. The extraordinarily leap in computational power each year means that our servers are able to handle more models, more factors, more inputs, more risk filters. That is a huge advantage. In terms of how we manage the business, every day we are pushing ourselves to become better than we were yesterday. And as they say in the US Navy SEALS: “The only easy day was yesterday.”
