Before the internet arrived, I used to be an equity analyst. I'd spend my days finding and arranging metaphorical jigsaw pieces to come up with a complete picture and an investment recommendation, which the sales guys would disseminate by telephoning their favorite clients first, and the ones they didn't like so much a bit later.
Today, at asset management companies and other financial institutions, there are still large teams of analysts and portfolio managers, sifting through data, developing investment theses and making asset allocation decisions. The difference is that we are deluged in data. Not only is the amount of data available to us accelerating, the nature of it is changing, with new sources of information being seen as potentially relevant for analysis. For example, the location-based data that comes from your mobile phone shows whether you are in a mall. Scaled over the whole of the country, this could allow us to see what is happening to footfall, which would help with understanding retail sales in real time. The footage from closed-circuit television that shows what's happening in transportation is another example. Or social media analysis of events happening in real time from people on the ground. Or weather data. The list goes on.
The reality is that there is just too much data for humans to be able to use. Opportunities go wasted because a team of humans just cannot create a sophisticated response to all of this. Of course, funds have been using complex algorithmic-driven trading strategies for years, but this is largely confined to market data. Imagine if we could take all the data that's coming from the real economy and use that to discern price, predict performance, understand risk and make better investment decisions. The only feasible way to do this is to use computing power to ingest the data, understand correlation and causation, and deduce rapidly enough how to respond. Artificial intelligence has now advanced to the stage where this is possible. While this gives rise to some interesting opportunities to radically transform investment management and capital markets, it will be very disruptive for people who work in the industry.
Let's assume that you use very sophisticated AI-driven models to scan data from not just the market but a whole plethora of other sources to define, implement, monitor, refine and adjust your trading strategies. What kind of people do you now need to employ? Performance will come down to how well your combination of engineers, scientists and market people can define and improve those models. The kinds of people employed in the industry will change; we will need people who can model data, and others who can validate the models and the results. And we will not need so many people. Much of the dialogue around the types of jobs that will disappear because of artificial intelligence has centered on relatively unskilled jobs, but in financial services it will be expensive, highly educated Wall Street types finding themselves out of work. An asset manager friend of mine agreed. He also said there's no way he'd go public about using this strategy; he'd prefer investors think how great his team was at stockpicking!
Change is already underway. One of the most high-profile companies in this space is Kensho, which is backed by Google, Goldman Sachs and S&P Global. Kensho uses AI to scan vast data sets much more quickly and accurately than analysts, and sells the information to banks and other financial institutions.
One competitor is Sentifi, which takes in data from thousands of sources, then filters it for accuracy. The founder got the idea after the nuclear disaster at Fukushima, when an investment manager friend griped that it would take three weeks of sifting through seemingly unconnected data to work out the implications of this event on his portfolio.
One hedge fund taking artificial intelligence to the next level is Numerai — which doesn't even employ the AI talent! This San Francisco fund encrypts its trading data and then crowdsources AI-based algorithms from anyone who wants to have a go. Contributors who develop algorithms that successfully improve performance get paid in bitcoin.
It's a long way from when I used to analyze company reports, scan articles in actual newspapers and use old-fashioned methods like the telephone and company visits to develop my investment ideas.
