Of course, such seemingly high-level analysis isn’t exactly new. Quant funds have used machine learning for years. But at a time when Wall Street research is increasingly commoditised, it’s not hard to see why Bank of America is trying to capitalise on one of the hottest buzzwords in finance. “It’s hard to learn from the historical data because of the nature of the FX market, so we try to really push the frontier” with alternative data and machine learning, said Alice Leng, the currency strategist who authored Bank of America’s AI-based research.
Among the three biggest US banks, Bank of America is the first to incorporate insights from machine-learning models into its published currency research. JPMorgan’s FX research team has explored machine-learning applications, but hasn’t put out any reports using them. Wells Fargo says it favours a fundamental economic approach to FX strategy, partly because that’s where it has expertise. For the team’s first study, Bank of America’s machine-learning algorithms sifted through fundamental and survey data, such as government spending and consumer confidence, to determine how the euro-dollar currency pair might perform.
The team used both supervised learning, when the machine receives training in how to process information, and unsupervised learning, when no classification guidelines are given. The bank’s models concluded that in the aftermath of the Italian election, in which euro-sceptic parties swept into power, the common currency would likely weaken. However, fears of a deep and sustained selloff against the dollar, like the one witnessed during the European debt crisis, were overblown. Despite all the hype surrounding AI, most banks are still scratching the surface. A vast majority of financial institutions in a Digital Banking Report survey last fall said they used some form of machine learning, but less than 20% went beyond “fraud, risk and compliance,” said publisher Jim Marous.
Last week, Morgan Stanley said it hired Michael Kearns, a computer-science professor at the University of Pennsylvania who has worked for Steve Cohen’s former hedge fund, to expand its use of AI across the company. Caio Natividade, Deutsche Bank’s head of cross-asset quantitative research, sees plenty of upside, particularly when it comes to currencies. His team has incorporated machine learning into its analysis, and he says AI can be useful in decoding the oft-confounding utterances of central bankers. Machine learning in research “might be a selling point,” said Greenwich Associates’ Richard Johnson. New regulations designed to unbundle research and trading are “going to really make research have to stand on its own.”
The foreign-exchange markets still present particular challenges, according to Vasant Dhar, a data-science professor at New York University and founder of SCT Capital Management, a hedge fund that’s relied on machine-learning applications for two decades. The complexity and variety of the macroeconomic factors that can sway any given currency relative to another can make FX markets distinctly challenging to analyse versus stocks or bonds. “The models that you build in other kinds of asset classes don’t apply easily, they don’t seem to carry over to FX,” he said.
Then there’s the age-old worry about trusting intelligent computers to come up with answers by looking at information in ways that humans can’t possibly understand. The fact that machines don’t need to weave coherent stories to support their predictions makes it hard for some sceptics to separate the reality of what’s going on from the marketing around it. “I find these buzzwords to be extremely vague,” said Marwan Younes, who runs Massar Capital Management. “I’m very uncomfortable with relationships where you don’t understand exactly the fundamental reason why it works.”Others say greater access to machine-learning tools could lead investors to develop their own AI analysis instead of relying on Wall Street research.
Sources and Photo-credits: Bloomberg, Gulf Times