Many systematic and quantitative managers are diversifying and growing their investor bases by launching lower cost strategies, but the challenge is to do so without cannibalising existing investor bases or eating into the capacity of other investment strategies. Crabel’s three strategies are lowly correlated, potentially complementary and have hardly any overlap in terms of trade types. 

Crabel’s flagship strategy commands relatively high fees due to the uniqueness of the strategy and because its trading time frames, which average one day and include intra-day trading, are capacity-constrained. Gemini excludes strategies with holding periods below two days and averages a 10-day medium-frequency holding period, which Crabel views as a less crowded trading time horizon.

Gemini also avoids those models that are sensitive to entry and exit or slippage costs. Crabel tentatively suggests that the Gemini strategy, which runs around $300 million as of June 2018, might have capacity in the region of $2-3 billion, though it is very difficult to pinpoint any exact figure. Crabel view capacity as something of a moving target because slippage and performance statistics need to be closely monitored. Gemini also excludes trend-inspired strategies, because Crabel and its clients have exposure to those in other vehicles.

Crabel’s first move into flat fee products was its Advanced Trend strategy, run by Grant Jaffarian, which charges a 1% flat fee and has raised nearly $1 billion, partly from US public pension plans. The fact that many institutions already have allocations to trend strategies has led to a search for new sources of diversification, and Gemini fits this bucket. It naturally needs to show low correlations to the conventional asset classes of stocks and bonds. Additionally, it is not only lowly correlated to trend strategies and other hedge fund strategies, but also very lowly correlated to Crabel’s flagship which has a sub-allocation of only around 3% to the current basket of Gemini strategies. Yet the Gemini strategy basket was developed within Crabel’s flagship, which has run all of the strategies on a live basis for many years. They include volatility breakout, mean reversion, factor-timing, and opportunistic strategies, which can include idiosyncratic ideas, spread trading, seasonality factors, volatility capture and others that are not easily categorised. 

The fees on Gemini, 1.25% for the institutional share class with minimum investment of $10 million, are comparable to many, and lower than some, alternative risk premia (ARP) or style premia strategies, but it appears very different from most ARP strategies. In contrast to some ARP strategies pursuing classic and generic factors and risk premia, Gemini has developed 40 proprietary models, which also evolve quite fast. “Our models are all fairly unique and uncorrelated to each other. Some models are designed to be hedges for each other,” says Pomada. 

One family of models can be grouped as volatility breakout or factor timing, which target specific areas of trend-following that are lowly correlated with traditional trend following. The other family of models can be categorized as mean reversion and opportunistic, which seek to exploit over-extended situations as well as some idiosyncratic market moves.

The return of volatility and dispersion, including divergence between developed and emerging markets, has helped Gemini to return 4.45% in the first half of 2018, after negative net performance for its first 17 months since launch in July 2016. Volatility breakout and opportunistic models made the largest contributions. “Our highest allocation is to volatility breakout models. We do not anticipate underperformance in normal, low volatility periods, but the launch period saw one of the lowest global cross market volatility and dispersion periods ever. This was particularly onerous. Societe Generale’s SG Short Term Traders’ Index saw a negative Sharpe of 2.5 over that period so we are pleased to have outperformed this,” says Pomada. The return pattern is also desirable in exhibiting positive skewness: the best month for the strategy, up 6.54%, is nearly twice the size of the worst month, down 3.38%.

The guiding principles for Crabel’s research process dictate that trade signals need to make intuitive sense, and this is partly informed by the team’s extensive trading experience. Pomada and his co-manager on Gemini, Steve Wisdom, both worked at Victor Niederhoeffer’s Manchester Trading. Wisdom had also been a proprietary trader at Societe Generale. Crabel’s scientific process is “observation and hypothesis-based. We look at what is happening in the marketplace, and form a hypothesis about why, how and how to capture it if it is repeated,” explains Pomada. 

Crabel uses terabytes of tick level data to test models, takes care to avoid ‘curve-fitting’ or ‘data-mining’, and seeks universally applicable models that work across multiple asset classes, and through different market regimes, such as high or low volatility, as a sign of robustness. That said, Crabel is also alert to alpha decay: “We are constantly improving models to make them smarter, better and faster,” says Pomada. Model turnover is around 25% per year but this is more about modifying models than making wholesale changes to them.

The chance to change models every year is one lure of working at Crabel. The firm throws researchers in at the deep end. “Our PMs can make a real impact in their first few years, much more so than a big bank employee,” says Pomada. 

Crabel has five US offices and the research staff are spread across three of them (Los Angeles headquarters; Southport, Connecticut and Cambridge, Massachusetts) and have plenty of communication through emailing and instant messaging. The corporate culture has a casual dress code, free parking, meals and snacks and a good match for pension contributions. Crabel maintains strong interaction and dialogue with the alternative investment industry, actively participating in industry events such as those organised by the MFA, Context and CME Group, as well as conferences arranged by prime brokers and investment banks.

Investment universe

Crabel Gemini trades around 200 markets, in equity indices, fixed income, currencies and commodities, selected partly for their liquidity but also for their diversifying qualities. For instance, in currencies, Gemini trades the Russian Rouble and South African Rand, and in equities, Crabel trades the Taiwan stock index. The UCITS obtains commodity exposure through an SPV certificate and has historically shown minimal tracking error versus the reference fund. For instance, in 2018 to April, the UCITS was up 6.46% and the non-UCITS 6.54%.

Execution and co-location 

Trade execution has been fully automated since 2009, using low latency technology to keep pace with high frequency traders. “If you trade a lot, high frequency traders try to figure out an edge and pick you off. You cannot be the sucker at the table,” observes Pomada. It is not just about speed however. It is also about market microstructure and masking flow so that other traders cannot front-run. Pomada stresses that the ongoing evolution of proprietary execution models matters more than speed per se. “A change in an algorithm can be more useful than shaving 50 or 500 seconds off latency,” says Pomada.

Crabel has invested tens of millions of dollars developing its co-location infrastructure and continues to spend millions each year to maintain it. Crabel has co-location in four locations: Chicago and New York since 2011, with Hong Kong and Frankfurt added in 2013. Some 99% of order flow is transmitted via STP, direct to exchanges. The proprietary Crabel Trading System is programmed to handle precise parameters and constraints, though there is still human monitoring round the clock, with at least two people on each time zone shift. Crabel does not have co-location everywhere however. “We trade on every futures exchange in the world, including in Taiwan, Singapore, Australia, Korea, Japan, Mexico and Brazil, and carry out cost benefit analysis to determine how much saving there is from shaving off times, versus our size on the exchange,” says Pomada. Crabel are getting better at modelling best execution. “We have very good models today that can estimate latency costs from each millisecond and microsecond. We had rudimentary models 10 years ago,” says Pomada. MontLake handles MiFID II reporting on best execution for the UCITS.

“It is hard to tell exactly how much co-location may add, but we estimate that the flagship strategy might save as much as 0.75% or 1% per year, from co-location given its trading frequency,” says Pomada. He estimates savings as somewhat less for Advanced Trend and Gemini but points out they are getting the execution infrastructure for free. Away from exchanges, shopping around for mainly OTC currency execution could be a larger source of savings. Pomada reckons Gemini could save as much as 1% a year from trading directly with 30 FX liquidity providers, including banks, ECNs and exchanges. “One thing will always be true; markets change and we can’t afford to become complacent. We challenge all facets of our trading organisation. Nothing is sacred, execution techniques, trading models and all portfolios are under a constant state of improvement,” he says.