Wall Street's Research Jobs Are the Most Likely To Be Upended By AI (qz.com)

An anonymous reader quotes a report from Quartz: Research analysts are the most likely employees on Wall Street to find themselves working with -- or being replaced by -- robots, according to a survey by Greenwich Associates. By next year, some 75% of banks and financial firms will either explore or implement artificial intelligence technologies, harnessing a variety of digital services to extract insights from mountains of data. While AI is probably near the peak of its hype cycle, several factors have helped it gain traction in recent years, according to Greenwich. Billions of images and documents are now available online for training computers to spot patterns and other high-level tasks. Advances in graphical processing units, which are adept at the kind of data crunching required by AI, are making sifting through daunting datasets much easier. The cloud has also made it cheaper for researchers and startups to boost their computing power to service sophisticated AI-enabled systems. AI makes sense for financial research, as machines can crunch reams of data more quickly than human analysts and, with the right data, identify obscure correlations and patterns.

Comments

Eh, maybe

[JBMcB]

Problem with using AI in these scenarios is that it's really good at finding correlations in what you tell it to look at. So maybe it finds correlations between interconnected stock prices, or maybe futures and trading volumes, or the consumer price index and stock prices of certain retail stocks, things like that.

When everyone has AI's doing this, the margins get eaten up pretty quick, since everyone is getting the same results and takes the same positions.

The areas you make money on are finding the niche correlations. A nationalistic dictator takes over some African country and shuts down rare mineral exports causing a spike in prices. A geothermal plant in Iceland goes down, shutting down it's aluminum smelters and aluminum prices rise. Those are the things AI sucks at.

Re: Eh, maybe

[KramberryKoncerto]

There are some one-in-a-lifetime trades where people pour in really high stakes, like the one made by Soros in the GBP crash last century. On the other hand, stuff like brexit and Trump's win were really good bets even if you predicted their odds to be 50:50 - because disproportionately few people bet money on the alternative outcome.

I agree with GP - a lot of these roles would benefit from the increased productivity aided by good statistical tools, where one equity researcher has to work with or become a so-called data scientist to produce better insights.

Re:Eh, maybe

[ShanghaiBill]

And they play off the general rise in the stock market as their ability to make people money.

That is one way, but another trick is to launch dozens of funds, and shutdown those that lose money. So if you start 32 funds, and purely by chance half beat the market after a year, so you shutdown the other 16. After two year, you have eight left, after three years, you have four, ... and finally after 5 years, you have a fund that beat the market five years in a row, which you can then promote as obvious proof that you are smart at picking stocks.

Re:Eh, maybe

[sheramil]

They don't have to win constantly to turn a profit. Those things you mentioned happened rarely, and all the bots need is a 60 to 70% win rate. Winning more than you lose is the only way forward.

Yeah, and when that level of winning pans out and they get greedier, they pay for the development of AI that understands a little more about the real world. And when THAT pans out and they get even greedier, they'll okay the development of AI that actively interferes in the real world to produce situations that profit can be made from.

Then it'll be too late.

p-hacking

[VeryFluffyBunny]

Ah, it looks like the financial sector are going to explore the limitations of automated p-hacking. With p-hacking, the larger the data set, the greater the probability of identifying background noise as significant patterns. Without knowing what specific, clearly defined questions you want to answer, you've got no idea of what kinds of data will hold the answers you're looking for and so you end up answering irrelevant questions but thinking that these answers are somehow significant.