E-mail:    

FX MANAGERS

Interview with Douglas Garistina

CEO of Sequoia Capital Fund Management

Douglas Garistina

Douglas Garistina shares his view about what makes the FX space unique, and explains why, after researching and developing an algorithmic approach to currency trading for several years - using various techniques such as genetic programming, factor modeling and several proprietary concepts - the company chose to apply systematic models in their currency program.

He talks about SCFM’s short-term, purely quantitative and systematic investment strategy, and explains how the company’s risk system has been embedded into the trading models.

 

FX Manager

Sequoia Capital Fund Management LLP

Strategy

Sequoia Systematic Fund

Location

London

Style

Multivariate Factor Model

Type

Short-term Systematic

Instruments

FX Spot

Traded Currencies

G10

FXTM

How long have you been trading foreign exchange for and what first attracted you to this industry? 

DG

I have been trading across a broad array of asset classes including FX since 1988, although the focus on applying systematic models in the FX space is really the result of research that my quant team and I started in 2006.  Prior to that, my focus had mainly been options market-making, running teams of traders and developing proprietary risk tools for Sequoia Capital LLP, a firm I founded in 1996.  Many of the technical concepts of illuminating dislocations or mispricings as well as methods for measuring risk, transfer very well from my earlier career work over to the systematic approach particularly in a closed universe of select currencies.

FXTM

In which way is trading currencies different from trading other financial instruments? 

DG

With all financial instruments, you have information in the pricing about two instruments, typically the product you are trading and the currency it is priced in.  For example, when trading crude oil, you think about the price of crude oil, but you are also receiving information about the dollar, although this information about the dollar is not generally utilized.  When trading currencies, this information is much more obvious as each cross has a different quote currency and base currency.  What makes the FX space unique is the triangulation that exists between a closed universe of currencies, giving rise to more exploitable opportunities.

FXTM

What do you particularly like about your job? 

DG

I find the moment of discovery to be particularly rewarding; when the team and I have been questioning and researching something, whether it is boosting signal-to-noise ratios in our datasets or enhancing our transaction cost modeling, there is a moment when you know you are onto something when the results of the work make a material improvement in your performance statistics.  It’s intellectually satisfying and it translates into winning, which is also always satisfying.  We’ve had a few of those moments this year, which have helped to contribute to our strong performance.

FXTM

When and how was the company born?

DG

SCFM was born as a spin-out from Sequoia Capital LLP in early 2011.  Our quant team and I had, at that point, been researching and developing an algorithmic approach to trading using various techniques such as genetic programming, factor modeling and several proprietary concepts while under the umbrella of the parent firm for around five years and believed we had enough robust strategies to become an independent investment management firm, separate from our parent firm.  From there, we have invested heavily in our infrastructure and given members of the team equity and performance stakes to ensure the business has what it needs to grow and weather the ups and downs of the performance cycle.

FXTM

How is the company structured today in terms of headcounts and offices?

DG

SCFM has its office in Central London, with a core team of eight personnel; four of whom are quants, a COO, two software engineers and myself.  We also have access to a couple of network and hardware engineers who provide the necessary expertise that we require in these areas as well as maintaining our off-site infrastructure at a professional data centre.

FXTM

What is the biggest strength of your team?

DG

The diversity of the team. 

Our quant team consists of an experienced trader with an engineering background, a research scientist with a hedge fund background, a research engineer with an electrical engineering background and an applied mathematician.  With their diverse experience, they each bring a different approach to our problem-solving and research and we end up with some very innovative solutions based on techniques used outside of our industry.  We use a scientific approach to our research and have a lot of peer review of the experiments we conduct.  That, combined with the fact that the team has been working together for several years means we all know how to get the best out of what each person brings to the table.

FXTM

What do you consider as being the key positions in an FX Management company?

DG

Because all of our trading and risk management is quantitative and systematic, the role of PM is less of a key role than in a firm where the trading is discretionary.  The entire quant team is able to operate the systems and the trading engines, which have the pre-trade risk layer embedded in them.  Therefore, they all share in a key role.  One of my roles as Head of Risk Management is shared with our COO, Hakan Malmros, so if there is ever anything to be decided regarding risk exposures it can be covered even while I am not there.  I would have to say that each of these areas are key (quant/trading team, COO, Risk Management) but due to the systematic approach the functions of these roles can be shared to a significant degree, providing redundancy that you can’t get with discretionary trading.

FXTM

Which authorities regulate the company?

DG

We are regulated in the United Kingdom by the FSA.

FXTM

How do you describe your currency investment strategy?

DG

Our investment strategy is short-term, purely quantitative and systematic.  Models are scaled to target a 15% volatility.  All of our models are designed to identify short-term dislocations in the markets that we apply them to, using technical or pricing signals from economically related markets.  The model we are running in the FX space is, broadly speaking, a type of stat-arb; it is neither mean-reverting nor momentum based, but able to take advantage of opportunities in both of these categories.  I think what is most important for investors to understand about our investment strategy is that it is in no way a ‘black box’.  We understand the economic logic behind the relationships that drive our signal generation and can see the inputs driving our performance.  We go to great lengths to work transparently with our investors so that they too can understand clearly what we are doing so that their expectations of us are in line with what we can deliver.

FXTM

How did you create and develop your FX trading model, and did it change over time?

DG

Our FX strategy is a theory-driven strategy using some commonly understood elements and inputs combined with various proprietary techniques and methodologies that developed over the past six years that ensure the models are robust, generalized and sustainable.  Research is continuous, into this model as well as upcoming strategies that we will run.  Part of our research focus is always devoted to performance of our live strategies.  However, we see it as vital that we don’t allow ‘style-drift’ and adhere to the core principles that we have based our models on.  Therefore, while we have not made any wholesale changes to our models, we have made minor enhancements that our research indicates will have a continuing positive effect on our outcomes, whether that is in the area of signal processing and accuracy, drawdown management or transaction cost efficiency.  We don’t subscribe to the idea that a model should stay static forever nor that it should be altered at the first sign of trouble.  Only if the results of rigorous research indicate a high probability of improvement do we proceed with an enhancement.  This requires a disciplined and objective approach such as the scientific method that forms the basis of our research approach.

FXTM

How do you manage risk? 

DG

Risk is managed at different levels by the entire team.  Because our trading is all model driven, nobody is putting on trades of their own volition that could involve them emotionally so everyone has a stake in ensuring that the positions we put on are within our risk boundaries and as specified by our systems and there is no reward for deviation from this.  Our risk system is embedded into our models so that when a new target position is generated, it must pass through the pre-trade risk layer, which consists of several tests measuring our exposures in different ways.  If the proposed positions violate any of these tests, the portfolio is scaled down by the system until there are no limit breaches.  Only then can the trades be sent to the trading engines.  The quants who run the models are at the front line and flag any issues to the rest of us if the risk layer throws up a warning.  We then all have the ability to view the risk reports wherever we are and if there are any matters to discuss, we do so.  Ultimately, I have the final authority on risk management and I have discretion only to reduce risk and only in defined circumstances.  We have used this discretion a handful of times over the past 18 months during events that could lead to a non-linear outcome that our models are not designed to identify.  For example, during the Greek election last year and the Euro Summits, we took the decision to target 7.5% volatility on those days, thus scaling our portfolio in half.  Once the event passed and the outcome was not disaster, we scaled back up to target 15% vol.  We also introduced risk reduction rules into our code to handle the Swiss National Bank’s decision to cap the Swiss Franc against the Euro.  This is an example of why having trading experience in a systematic team is vital.  From a traders view, certain positions are asymmetrical in outcome and can be dealt with by using rules to avoid certain positions in certain circumstances.  An understanding of market mechanics and the observational skills developed over years as a trader also serve to reduce risks by ensuring that processes that are implemented have a market awareness embedded in their logic.

FXTM

Tell us about a lesson you learnt from past trading decisions. 

DG

In the early days of running our model before we did much pre-processing of our data (and before we took on investors), we would notice the model occasionally wanting to put on large AUD/NZD trades and our investigations led us to conclude that the sparcity of some of the input data was driving these decisions.  This led us to come up with some new techniques for monitoring and filtering the input data.  This served to further clarify for us the importance of market knowledge and expertise as data, although accurate according to ones data provider, is not necessarily useful unless it is meaningfully complete or at least filtered to reduce unwanted noise.

FXTM

Do you use a blend of strategies or one only?

DG

Firm-wide, we have a group of strategies that can be blended for diversification across asset classes.  In the FX space, we are using one type of model but it is structured to generate many simultaneous signals in several crosses, each in themselves somewhat weak classifiers but aggregating into a robust prediction at the currency level.

FXTM

What are the market conditions that you consider ideal, and which ones are the most challenging, for the performance of your strategy? 

DG

As with most systematic strategies that rely on historical data, a free-moving market with little government intervention is ideal.  Generally, government intervention tends to go against the prevailing wisdom or logic of what should be happening in the markets and models are usually not particularly good at identifying that.  That said, we have been having a great year and there have been several instances of government intervention and risk-on/risk-off swings in sentiment, so it’s possible that with enough continuous government intervention in the historical data, our models have been able to handle this environment better.

FXTM

Can you give us an example of a memorable winning trading decision? 

DG

In the short term systematic space where we, as individuals, are not making the trading decisions, it’s difficult to identify particularly memorable winning trades.  It is more about memorable tweaks to the model or particularly memorable trading periods where we win on 80% of the trading days in a month.  The way our models are built, they don’t aim to do trades that should stand out in memory.  Our models perform several transactions daily each trying to exploit small opportunities that we expect to unfold over the coming day. It’s more of a steady chipping away process rather than swinging for ‘home runs’.

FXTM

Do you use Emerging Markets currencies?

DG

We don’t use EM currencies as our models rely on quality input data, which often doesn’t exist in the EM and our execution requires high liquidity to handle our high daily turnover.  For traders who do not have the same concerns, I believe the EM currencies offer very viable and exciting opportunities that may not exist in the developed currencies.

FXTM

When developing a strategy, do you give a higher priority to building entry signals, exit signals or money management rules? 

DG

Money management rules.  Without these, you are nowhere.  No system, regardless of how good it is at timing entries and exits, can be right all of the time so you must have prudent risk rules in place to keep these losses in check.

FXTM

Do you think that every strategy loses its accuracy sooner or later, or do you believe in long lasting market rules?

DG

If a model is left static, it could very well find itself losing accuracy over time although this may come back at some point if the market theory it is based on is robust or long-lasting.  To ensure the sustainability of our strategies, we have incorporated an objective, rules-based, model selection technique that we run periodically to select the best parameters to use going forward until the next selection.  This entire process, the timing and the scoring techniques we use are routinely validated to ensure that any updating we are making to the parameters is sensible and most likely to be positive in the live environment.

FXTM

Do you use any form of optimization, and how do you cope with curve fitting? 

DG

Our model selection technique is a form of optimization.  Some of the key strengths our team developed during our time creating genetically-evolved algorithms are methods for minimizing any over-fitting and identifying the most robust and truly predictive models.  Being a data-driven strategy, genetic evolution of algorithms is a recipe for overfitting if strict procedures are not followed.  As a result, we have developed a suite of tests or methods in this area that we have adapted for other models such as our FX strategy.  Some of these are classic, such as using Monte Carlo simulations, but most are proprietary and specific to our needs.  I would say, at a general level, that ensuring you have objectively defined methods that separate true productiveness from randomness, ways of testing for consistency over several out-of-sample periods and keeping an eye out for avoiding local optima, are key criteria for building robust, sustainable models.

FXTM

Do you favor any particular time frame in your strategies?

DG

We like the short-term space with position holding periods around the 1-2 day mark although we can hold positions for up to a week.  In particular, we find that generating new signals on a daily horizon or slightly less keeps our models nimble and not married to a particular theme for too long in an ever-changing environment.  The flip-side of this is capacity restriction, particularly with a high turnover model like ours.

FXTM

What should an inexperienced trader watch when choosing a time frame? 

DG

There are advantages and disadvantages to both long and short time frames.  Some of the best trades in my career were the result of analyzing weekly and monthly commodity charts where the patterns tend to be clearer.  You need tremendous patience watching a long-term pattern develop and not rushing into the trade too soon.  In the short term space, a major advantage is that you have so many more entry and exit opportunities that it makes testing an algorithmic approach much faster and easier.  Also, with more trades happening in a shorter space of time, you gather statistical information about your strategy much faster.  It is, however, a noisier space to work in.

FXTM

What is the leverage that you normally use?

 DG

On average, our FX strategy uses 3-5 times leverage and we limit this to 7 times max.

FXTM

How many execution brokers do you use? 

DG

We limit the number of liquidity providers we use as we would rather be a good client to a few than a small client to many.  We believe this helps us see tighter spreads.  All of our trading is electronic.

FXTM

Which historical data do you use when developing your strategies?

DG

All of our futures data is from Tick Data and our FX spots data is from Bloomberg and our liquidity providers.  Data quality is of the utmost importance as ‘rubbish in equals rubbish out’.  Even with the best data, we find that we are able to boost our signal/noise ratios with careful, unbiased filtering and other pre-processing techniques.

FXTM

Which software do you use in the research, risk and reconciliation functions? 

DG

We use Matlab for some research but most of the software we use for all of our functions has been built in-house and all of our production code is in C#.

FXTM

Which opportunities and risks do you see in ultra-high frequency trading for FX managers? 

DG

I think there are still opportunities for market-sniffing types to spot execution patterns and try to take advantage of them.  At the same time, execution algorithms and liquidity pool sweeping is getting more sophisticated to reduce the execution footprint. From my experience in the ultra-high frequency space in equity index and fixed income option markets, I believe it requires a commitment to an arms race on both the software and hardware fronts.  There is already a lot of exploitation of the mispricing opportunities that exist reducing the possible returns and thus requiring ever-faster code running on ever-faster graphics chips or proprietary chips.  None of this comes cheap and the larger, more established players can outspend newcomers on both software and hardware.

FXTM

Tell us about the efficiency and capacity for your program. 

DG

Our strategies have a high turnover on a daily basis so we have to focus on the most liquid markets.  We then need to constantly work on making the most of the available liquidity in these markets.  Only then can we be sure that the alpha that we squeeze out of the markets each day will amount to a net profit.  Our FX strategy is capacity constrained to somewhere north of $250m but how much more is an unknown as we continually work on improving what we can move through the markets each day.

FXTM

Can you give us your feeling about the move of the EurUsd in the next 6/12 months?

 DG

Personally, I think the Euro belongs lower still.  Professionally, I don’t care as long as our models keep getting it right.  As a former discretionary trader, it took some time getting used to this, but that’s the beauty of running quantitative models – hard-coded discipline and no emotion.

FXTM

What’s the best advice you would give to traders who want to enter the FX fund management industry?

DG

The competition in this industry is fierce, so be certain before you proceed down this path that what you are doing is both sufficiently differentiated from your peers (if not unique) and sustainable long-term. 

6 Dec 2013