As lawmakers in Brasilia debated a controversial pension overhaul for months, a robot more than 5,000 miles away in London kept a close eye on all 513 of them. The algorithm, designed by technology startup Arkera Inc., tracked their comments in Brazilian newspapers and government web pages each day to predict the likelihood the bill would pass.
Weeks before the legislation cleared its biggest obstacle in July, the machine’s data crunching allowed Arkera analysts to predict the result almost to the letter, giving hedge fund clients in New York and London the insight to buy the Brazilian real near eight-month lows in May. It’s since rallied more than 8%.
This is the kind of edge that a new generation of researchers are betting will upend the research marketplace. For Arkera’s clients on Wall Street and in the City of London, that means getting robots to filter through the noise in faraway lands.
“There’s too many people to follow on Twitter, too many websites, too many articles,” said Nav Gupta, the 48-year-old co-founder of Arkera, which says its software can process as much information as 1,000 human analysts. “That’s a very expensive problem and everybody faces it.”
The company raised 4 million pounds ($4.9 million) last year from investors including Alan Howard of hedge fund Brevan Howard Asset Management LLP.
Using so-called artificial intelligence to automate swathes of the research process is quickly gaining traction because cost-conscious investment banks are downsizing. In the U.K. alone, there was a 30% drop in research budgets last year, Financial Conduct Authority data show. At the 12 biggest banks, there’s been a 7% drop since 2015 in the number of front-office staff covering currencies, such as traders and researchers, according to London-based research analytics consultancy Coalition Development Ltd.
That means it’s even harder than it used to be to afford analysts on the ground in developing nations, about the only places in the world where investors can get yield right now.
Data-science companies like Arkera and New York-based Sigmoidal say they can solve this problem using machines that learn as they go to dredge through tens of thousands of news articles, government statements and social media accounts in languages as varied as Spanish, Arabic and Chinese.
After an initial investment of up to $100,000, banks can save $1 million over seven years using such systems because they don’t need to hire as many data analysts, said Marek Bardonski, who was chief executive officer of Sigmoidal when he spoke with Bloomberg in July. He has since left the company. Previously, Bardonski, 27, was a computer scientist at graphics chipmaker Nvidia.
Take this year’s protests in Hong Kong. Bardonski said Sigmoidal’s software was able to track developments in the Cantonese-language press and even identify the non-verified Twitter feeds of protest leaders to monitor the risk of further unrest. The technology is useful for far-away countries wracked by political turmoil, places where investors are keen to put money but don’t have easy access to information.
“The system can give an edge over traditional analysts working for financial institutions,” said Bardonski, who said typical reports will include charts on sentiment, key word statistics and short written summaries. “Instead of getting 100,000 news articles, clients can get all the insights on one page.”
Neither Sigmoidal nor Arkera would let Bloomberg see an example of an automated report to see how readable it is compared with one produced by a human, citing rules against sharing proprietary data.
In Europe, the way investors consume research has evolved fast since new rules last year forced investors for the first time to pay separately for the analysis they receive. The so-called MiFID II legislation stopped a widespread practice of having the cost of research built into the fees that the likes of Goldman Sachs Group Inc. or Morgan Stanley got paid to execute trades.
The irony is that a year and a half since the rules came into force, many investment banks still offer research for free because clients aren’t willing to pay for it, according to Sarah Jane Mahmud, a senior Bloomberg Intelligence analyst who specializes in regulation. They get around the rules by publishing research on their websites for public consumption.
But the quality has gone downhill because mid-level analysts have left or been pushed out, leaving junior analysts to do the work so their more senior colleagues can go to client meetings. This is giving investors even more impetus to seek out bespoke research, like paying cash to speak with experts in the field or investing in automated research to support their senior fund managers and strategists.
“Asset managers now need to assess the value of every single research service to assess it it’s worth paying for, how much they should pay for it, and trying to filter the good from the bad,” Mahmud said.
Under MiFID II, asset managers must be prepared to demonstrate they’ve done due diligence on all investments they make for their clients, something that’s always been tricky in developing countries.
It was that very problem that inspired Gupta and his business partner Vinit Sahni, whose careers spanned firms including Citadel LP, DE Shaw & Co. and Goldman Sachs, to set up Arkera in 2015. During their 20-year careers in investment banking, trying to find information to substantiate something felt “like pulling teeth,” Sahni, 50, said.
So the pair set up a team of data scientists and engineers to design a search engine that investors can use to give them an edge in places like Turkey, Mexico and Egypt. It works kind of like Google, only it’s programmed to choose the most relevant sources from tens of thousands of articles, social media feeds and government releases.
As good as robots are getting at deciphering market jargon, even their developers admit they’ll never fully replace humans. In the next decade, Sahni said smart machines will significantly enhance the capabilities of human analysts.
“We will see advancements in cognitive abilities, communication and the physical potential of humans as we collaborate closely with machines and algorithms,” he said.