Big Data Is a Big Mess for Hedge Funds Hunting Signalsby
Citadel, Winton Capital move to exploit new data sources
Humans created huge amount of ‘low-grade data,’ says one quant
It was big data’s big moment.
The most unfathomable U.S. presidential race in memory gave Predata, a startup firm backed by hedge fund manager Kyle Bass, a chance to cut through the vitriol and coolly calculate the winner by monitoring chat data. Turns out that Predata, which anointed Hillary Clinton the likely victor days before the Nov. 8 vote, was no smarter than the cable news pundits.
The bad call is a reminder of the pitfalls hedge funds face as they try to profit from a world awash in data. After years of sluggish returns, firms seeking fresh ideas are joining the data revolution, tapping the torrent of social media, satellite and personal information already used by airlines to forecast ticket sales and health experts to track illnesses. But so far few managers have mastered the art of spotting clear trading signals buried in all the noise.
“There are those who realize their industry is changing and their fund isn’t going to exist if they don’t adapt,” said Matei Zatreanu, who led data efforts at $19 billion hedge fund King Street Capital Management. “But not many of them know what they are doing, except a few.”
Two Sigma, which says it was exploiting big data before it was a buzzword, is one of the handful. The $37 billion hedge fund has outperformed rivals partly by using machine-learning algorithms to find trading signals in the data. Now traditional hedge funds, which have relied on information such as company filings and shopping surveys, are on the hunt.
Citadel’s Data Bet
London quant firm Winton Capital Management, which manages $33 billion, said in April it’s starting a data science center in San Francisco and plans to hire some 40 scientists. The $26 billion Citadel in October promoted its former chief risk officer, Alex Lurye, into a new role of chief data officer. This month, Third Point founder Dan Loeb told investors that hedge funds need to use data sets and quantitative techniques to remain competitive.
Problem is, a lot of the data is useless and even the good stuff needs to be laboriously cleaned of erroneous bits and duplicates. One fund received credit card data for restaurant chain Cracker Barrel Old Country Store Inc. only to find that it included transactions for the Nutcracker ballet, said Zatreanu, who left King Street in May after eight years. Even bespoke high-resolution satellite imagery, which costs as much as $15,000 an image, can be obscured by clouds and rendered worthless.
“Humans have created a huge amount of low-grade data recently, but have failed to realize the rule that quantity does not equal quality,” said James Holloway, founder of quant fund Piquant Technologies in London.
The data also can be legally suspect. Vendors may not have the right to sell data, such as location tracking from cell phones, if consumers didn’t consent. Were the camera-equipped drones flying high enough to avoid violating privacy regulations?
WorldQuant, which manages $4.5 billion in Old Greenwich, Connecticut, has a team of researchers vet data. They talk to thousands of vendors a year, test data from perhaps a thousand of them and buy sets from a few hundred, said Matt Ober, the hedge fund’s co-head of data strategy.
“There are a lot of vendors popping up and you’ve got to get a sense of where they obtained the information from, and if they have the right to what they are looking at and selling,” Ober said.
Social media in particular requires a good scrub, said Holger Knauer, whose Catana Capital in Frankfurt collects about 300,000 data sets spanning online news and bank research. He said thousands of fake Twitter accounts post misinformation that can affect stock prices, and so Knauer has set up standards to weed out the offenders.
“You have to filter information out from these false accounts otherwise you can make wrong decisions,” he said.
Managers accustomed to price-earnings ratios are also struggling to interpret new-media data. Last year, when Chipotle Mexican Grill Inc. was rocked by waves of customers with foodborne illnesses, some hedge funds misread the impact on sales. They relied heavily on foot traffic from location apps that showed a dramatic decline while overlooking credit card transactions that revealed customers had meals delivered during the colder months, Zatreanu said.
“You need to corroborate the information with other sources to get a holistic picture of a business,” Zatreanu said. “Big data is messier and requires more scrutiny.”
Big data firms do get it right sometimes. Predata said it flubbed the U.S. election partly because it excluded the “seamier, populist corners of the Internet” -- where Donald Trump’s supporters gathered -- in its analysis. But the company, which did call the U.K.’s Brexit vote correctly, says it has anticipated attacks on oil facilities in North Africa, military provocations by Russian President Vladimir Putin and large moves in currencies and rates.
Earlier this year, Zatreanu, 32, teamed up with recruiter Alexey Loganchuk to start Augvest, which has held events from New York to Hong Kong to help hedge funds solve the big data puzzle. In August, about eighty hedge funders met under the high exposed-beam ceilings of a fintech office in Manhattan’s flatiron district to share their trials.
Role of Scientists
Zatreanu, who also runs data analysis firm System2, offers this advice to hedge funds: give data scientists more gravitas.
He said firms treat data quants as back-office employees unworthy of making investment calls. Some funds have relied on tech-support staff or interns for data analysis. One firm recently advertised for a scientist to do a job that included clerical work, Zatreanu said. They need more influence -- even a seat on a management committee -- to smooth the transition to big data, he said.
Zatreanu also advises firms to break down the walls between managers and quants, who struggle to understand each other. Many analysts need a crash course in statistics, he said, to avoid the common misstep of trying to use data to support preconceived ideas.
“A lot of funds are just checking the box and saying they are doing data-driven investing,” he said. “For this to succeed, you need to have the buy-in from senior management who may necessarily not want change.”