J.P. Morgan equity analyst: "Why I'm better than the machines"

eFC logo

As quant funds outperform discretionary investors on the buy-side, human researchers on the sell-side should surely be worried. - After all, who needs their carefully considered advice when an algorithm can make better sense of the market in a moment? One senior J.P. Morgan analyst working out of San Francisco says he's not that concerned, yet.

"Whenever the question is changing, the machines fall apart," said Rod Hall, a senior J.P. Morgan analyst covering telco and networking equipment and IT hardware. "What machines won't be good at, is figuring out what the next big questions to ask are," added Hall. "For example, it's difficult for an algorithm to ask the right questions about the replacement for the iPhone at Apple... This is where humans come into the equation."

Hall was speaking at this week's Newsweek conference on artificial intelligence (AI) and data science in London. Also present was Sylvain Champonnois, a member of Blackrock's long established scientific active equities team which has been in the systematic space since 1985. Blackrock made a push into AI in 2015 when it hired Bill MacCartney, a natural language processing expert from Google.

Champonnois agreed that today's algorithms fail when the paradigm shifts. The changing political situation is a case in point. "You have events like Trump and Brexit and the French election and your algo is based on data from the past," said Champonnois, adding that contemporary algorithms failed to function well during the Eurozone crisis and that most struggle to deal with the abnormalities of data from Japan.

Machine learning is more than just a simple algorithm though. In theory it's a self-reinforcing system which - in the case of investments - learns from past mistakes to make better investing decisions in future, as at AI hedge fund Sentient Technologies. So, will AI fully displace human stock pickers as the time horizon of the data sets it's based on increases? - After all, now that Trump's happened once, the algos will know how to model market reactions to a Trump-like event in future. Yves-Laurent Kom Samo, a Google Scholar and former FX quant trader at J.P. Morgan and equities algo trading strat at Goldman Sachs, said it will and that time horizons aren't really the issue. "The more data we have, the better we will be. When you have the data, I see no reason why machines won't be able to come up with new trading ideas," Kamo claimed.

For the moment, however, human stock pickers like Hall have their place. For all their attempts to develop artificially intelligent trading systems which superannuate humans, most funds have had limited success. Sentient, for example, only runs its own money and has been slow to release meaningful results. MacCartney only stuck around at Blackrock for 14 months before leaving and joining Apple. "There's a model where you hire a PhD and put him into a room thinking he's going to do amazing things, but the reality is that the algorithm he's developed may only tell you things you already know," said Champonnois.

In January, J.P. Morgan hired Geoffrey Zweig, a natural language processing expert from Microsoft, to form a new global machine learning division. Zweig is undoubtedly in a room with some PhDs and big things are possible. From Hall's perspective, however, J.P. Morgan's machine learning aspirations sound pretty modest. "We've been working out how to use AI to reduce our lower value workflows, like reading through 10Q reports" said Hall. J.P. Morgan wants computers to read companies' 10Q reports and identify the most salient points for humans to process later, he added. "AI is a co-pilot to allow us to think faster and process more information," said Hall. - It's not a replacement for human beings, yet.

Contact: sbutcher@efinancialcareers.com

Related articles