Its one of the strange truisms in todays investment world: Traders are drowning in data and yet they still want more. In this age of so-called big data-the term marketers and solution providers call the huge sets of information from various sources-buyside firms continue to look for ways to exploit this endless stream of data to help score the next big deal. At Flyberry Capital, a small quant shop opened for business in 2012 in Cambridge, Mass. by a trio of graduates from the Massachusetts Institute of Technology, big data is more than a buzzword-its the very foundation of how this ambitious hedge fund researches, analyzes and executes its trades.
And theyve been tested under fire. Shortly after opening their fund, the Flyberry Capital principals entered a hedge fund competition to test their big data strategies against 350 other small and aspiring quant strategy funds to win the $1 million seed money from contest sponsors BattleFin and Lions Path. From the start, the new MIT graduates-founder and CEO Michael Chang, co-founder and COO Zeid Barakat and chief technology officer Sean Chang (no relation to Michael)-impressed their sponsors at what was dubbed The Hunger Game for Hedge Funds.
They were the smartest guys in the world with a really great strategy, Tim Harrington, president of BattleFin, a recruitment firm for hedge funds, told Bloomberg News. And they couldnt get capital.
All that has changed. After beating out the other hedge fund start-ups, Michael Chang, Barakat and their team of 10 employees are running a quant shop that is on the hunt for big data. With Sean Chang, a former Google fellow and MIT graduate with more than 50,000 hours of programing experience (Hes like the Michael Jordan of programmers, says Barakat), Flyberry created the Flyberry Engine, a search engine that scours the Internet for market data, news alerts, social media, government reports and announcements, and reports from weather, earthquake and other geosensors. While other hedge funds use traditional market data feeds from Bloomberg, Reuters and assorted stock exchanges, Flyberry casts its net much wider.
Flyberry Capital is the definition of a small quant shop: It has 10 employees, most of whom are analysts who hail from MIT, Harvard and Stanford, and only two of them are authorized to execute deals as traditional traders. The fund started as a proprietary trading shop and now calls itself a sub-$10 million firm with plans to grow tenfold in the coming years. And despite its big-data and high-tech pedigree, Flyberry Capital is not a high-frequency trading firm looking to score the next sub-second trade and move the markets in the process. Its portfolio focuses on a narrow basket of 18 or so highly liquid instruments such as index futures, currency futures, energy futures and different commodities. We never want to be the majority participant in markets and instead focus on things that we can easily enter and exit fairly rapidly, said Barakat.
Recently, weve been testing and applying the same strategy to cash equities and large market capital cash equities, he added. We have a number of trade strategies that are in the pipeline, as well. We expect to launch our fund vehicle in Q1 of 2014, and that it would be comprised of a basket of highly liquid traded instruments both on the future and cash equities side.
Asked whether they are long/short, Barakat admits his firm is agnostic with regard to direction. Whatever the signal tells us to trade, well trade. Were as comfortable going long as we are short, he said.
The Big Data Secret Sauce
Where does Flyberry Capital get its big data? Anywhere and everywhere, it seems.
Powered by 10 or so servers from a third-party IT service provider, Michael Chang and his team are on the lookout for events that move the market. The real philosophy about what we do is, we believe all of the price movement is subject to information shocks. Basically, when you see that theres an event coming, thats when you see a huge price movement. The question is: How can we identify those things? We try to use some novel information to inform us and make the right decision, said Chang.
When asked how many sources of potential information the Flyberry Engine scours for new data, Barakat replied, More than 350, and Chang chimed in with a confident, Thousands.
Heres how the data is sorted: The Flyberry Engine finds the data from those thousands of sources starting with low-latency news sources, and then looks for the best news source in order to obtain and parse the information in a rapid fashion in what they call a bucket. A second large bucket consists of geosensory information that pertains to earthquakes, weather patterns, forecasts, etc. Thanks to open-source initiatives from the U.S. and other governments, a lot of these [data sites] have just recently become available that can be highly reliable datasets to explore, Barakat said. Third, the Flyberry engine looks at different social media sources like Twitter and blogs to see what global sentiment looks like and see if there are patterns that we can observe, he added.
Using different data sources gives you a much better sense and a more complete picture of the markets than any individual data source can, Barakat said.
Into the Future
The Flyberry Engine not only looks for events as they occur-in some cases it also looks at events, such as corporate earnings and national unemployment figures, before they are scheduled to be released. Big data helps us to identify some potential trends. When a government figure comes out, for example, we can use data to give us better direction. We see all the numbers that might be coming out by aggregating a lot of [past information], and then we try to identify as many events as possible. And then when the information is announced, we try to implement our model, said Chang.
For recurring events, the Flyberry Engine will search out the same event for the past five years and gather any and all information related to that event. Why five years? We have years of data, but its generally because the electronic market became efficient in the last five years, Chang said.
When it comes to a single, out-of-the-blue event-say, a fire at a factory or a devastating tsunami-the Flyberry Engine will have to work on the fly. If its a completely uncharacterized, unexpected event, its likely that we wouldnt have a model thats been developed to trade off of that event. But in the same fashion, we can use the same big-data techniques to identify if this is sort of a high-risk event, Barakat said. And that might signal to us when we want to not participate in the market. As a risk management hedge, thats another area where we use the same techniques.
According to Flyberry chief compliance and marketing officer David Nichtenhauser, even one-off events can be analyzed beforehand. Most, if not all, of our models have had events that are recurring. Even infrequent events, like an earthquake, had to occur enough times such that we could test the veracity and robustness of those models.
Nichtenhauser continued, So if its a one-time event thats catastrophic or if its an unusual event like a fire at a Ford plant, it is highly unlikely well pick up that event because its so idiosyncratic and so rare that there is no way to test whether and how the market might respond to it. Now, thats not to say we wont in the future, but we dont have that kind of culling capacity to understand the markets behavior with regard to something that unusual.
Different from CEP
This may sound like a new flavor of complex event processing (CEP) — -the method of trading stocks instantly based on events reported in low-latency news feeds. CEP strategies have been in place in trading firms since the middle of last decade and work like this: If the sole parts manufacturer for the truck division of the Ford Motor Co. loses one of its two factories in a tornado, for example, a CEP solution would read that breaking news flash and immediately get to work by either dumping the stock of that parts maker and/or Ford, or possibly buying shares from the rival parts supplier.
Chang and his team see a big difference between their strategy and CEP. CEP tries to identify those events the traders and portfolio managers already know. When the U.S. monthly jobless claims are announced, for example, a typical CEP trading desk enters in an internal code for jobless claims to look for every type of jobless claims that have occurred, said Chang. They ask for the impact and what would happen to the 10-year note. And thats all they do, he said.
With Flyberrys search engine, they search out various sources of information in relation to the monthly jobless claims. They would look, for example, to see if paycheck processor ADP announced its own private payroll numbers and other sources of information that could be very difficult for those traditional CEP solutions to gather. You have to know those things in order to link these two things, Chang said.
This data-intensive method of looking at the markets is still fairly new, said Chang. Further, he doubted that a larger, more established asset management firm-even one with a larger IT budget that Flyberrys-could come up with a system like theirs any time soon. This is the engine we have been building for more than three years, and right now its very powerful. I wouldnt be able to imagine some big company building all those things from scratch, because basically what they are relying on is something they had in the past, like high-frequency trading and other forms of infrastructure, he said.
Not a Silver Bullet
According to Barakat, the principals inside Flyberry Capital realize big data is no substitute for good judgment. The former bioscience executive added that theres no implicit suggestion that the skill sets of a traditional traders are going to be obsolete in the near future. What he and his colleagues do is vastly different.
Its a very different approach to what we view it as the future of quant trading, since its not dependent on price movements or not dependent on momentum. Its not dependent on volatility, but its about taking surprise events, whats hitting the markets and whats causing major market reaction, he said from his Cambridge conference room, not far from his alma mater. It means having a good understanding in advance of those events of how they are going to move the markets, what we should trade, when we should get in and when we should get out.
