FT: WHEN SILICON VALLEY CAME TO WALL STREET

Gilead Sciences recently reached an $11.9-billion agreement to acquire Kite Pharma for a 50% premium to its monthly average stock pri+ce. Five days before the US takeover was announced, an artificially intelligent computer predicted it.

The algorithm, developed by Dataminr, a New York- based technology group, had noticed unusual social media chatter and options activity and alerted its clients about a possible buyout of Kite by a large biotech company. The following week, shares in Kite jumped 28%.

The prediction was not the first time clever social media analysis had anticipated a market movement before investors, analysts and journalists. For years, boutique hedge funds have used computers to crunch alternative data sources to gain a competitive edge over their rivals.

However, as artificial intelligence makes predictions more sophisticated and active managers come under pressure to reduce costs and improve returns, mainstream investment companies such as Schroders, Axa Investment Managers, JPMorgan Asset Management, Goldman Sachs Asset Management, State Street, NN Investment Partners and Fidelity have also begun using big data or machine learning to aid investment decisions.

"There has been a blurring between traditional judgmental managers and quantitative managers," says Gideon Smith, Europe chief investment officer at AXA Rosenberg Equities, the quantitative investment arm of France- based Axa Investment Managers. AXA Rosenberg said earlier this month it would integrate machine learning and big data into its investment processes.

"Any role where you are doing something repetitive is ripe to be done faster, quicker, cheaper by machines and robots, and so it is beholden on all of us in the asset management industry to look at what we do . . . and [whether] it is something that a machine might be able to do better."

As cheaper, passively managed funds grow in popularity and attract assets away from active managers, investment houses have come under fierce pressure to cut fees and improve performance. Under the spotlight of investors, regulators and passive rivals, active managers hope that artificial intelligence can help traditional fund houses beat the market and rein costs in.

With this in mind, Schroders, the UK's largest listed asset manager, has embarked on ambitious plans to recruit data scientists. It has expanded its data team from two to 18 over the past two years and plans to hire more staff as portfolio managers get used to the technology.

"There are all these big data sets that could be useful to inform the investment decisions that stockpickers are making," says Mark Ainsworth, head of data insights at Schroders. "But they don't currently exist in a form that is useful for investment managers.

"You can get a computer to make some sort of sense of a large body of text."

According to Mr Ainsworth, Schroders' data team accurately predicted how many betting shops the UK's competition regulator would force Ladbrokes and Gala Coral to sell following a £2.2-billion deal to create the UK's largest gambling house.

"The alternative is to put an intern in front of Google Maps for six weeks," he says.

For years, small tech groups have attempted to woo traditional money managers with machine learning technology, arguing that it was a strategic imperative for investment companies to understand the 2.5 quintillion bytes of data that are created every day. But until now many mutual fund groups have struggled to make the transition because of a lack of in-house expertise and the high cost of valuable information.

In the past few years, breakthroughs in machine learning and so-called natural-language processing, where computers learn to interpret human language, have made it possible for artificial intelligence to decipher social media communication, patent filings, satellite images and even analyst calls.

These developments have piqued the interest of big investment groups. "Unstructured data have grown exponentially," says Javier Rodriguez-Alarcon, head of quantitative investment strategies for Europe, the Middle East and Africa at Goldman Sachs Asset Management.

"You have a gap between the amount of information and the ability to process it, and [ for a long time] the only way to close that gap was to hire lots of Javiers.

"Now you can capture that gap by using computers."

Almost a third of asset managers surveyed this year by Tabb Group, a capital markets consultancy, said they were using alternative data, with 64% saying they believed it could help them beat their benchmarks. The most popular information was social media data, supply-chain analysis, business performance information and Web traffic.

Their interest has led to an explosion in the number of start-up tech companies offering alternative data analysis to financial institutions. There are now at least 172 around the world, according to YipitData, a Web data intelligence service for institutional investors.

"There has definitely been a significant increase from traditional mutual funds in the past year," says Emmett Kilduff, founder of Eagle Alpha, a wellknown alternative data provider.

"There are some macro factors: a lot of money is going to passive and quantitative [ funds], so the discretionary guys are trying to become more quantamental [quantitative and fundamental]."

The effects of a shift towards quantitative investing using machine learning tools will be significant for financial markets.

If companies attempt to control data signals or the same few vendors provide underlying technology to multiple asset managers, there could be concerns about crowding and competition.

“Does the regulator know what the changes to the market will be if people don’t know what is happening on the margins?” says Daniel Tammas-Hastings, founder of RiskSave, a robo-advice start-up. “If we built a machine-learning algorithm now and asked it to trade individual accounts, it would take two, three years before we knew whether it was right.”

Obstacles remain for traditional fund managers, who are often trained in economics and finance rather than data science or technology.

“They are too big, they are too slow,” says Scott Borgerson, cofounder of CargoMetrics, a maritime data analytics group, which is now a hedge fund. “These are big data Silicon Valley things, not Wall Street things.”

Stockpickers usually do not buy and sell stocks more than once within a trading day and cannot always benefit from a data-driven signal that is only a few hours ahead of a company announcement.

“There are many challenges for someone who is not native in data science to start using these tools,” says Armando Gonzalez, chief executive of RavenPack, which provides social media and news analysis to investors. “I would say 90% of quantitative funds and 30% of fundamental [managers] are using these tools.”