machineByte hosted its inaugural US ThinkTank on 12 December 2018 at the Ritz-Carlton in Half Moon Bay, California. mB Head of Content, Chantal Hermans led a discussion with 12 practitioners across all disciplines.
During the spirited discussion in Half Moon Bay, participants raised concerns about finding the right talent, the dilution of proper machine-learning techniques by new arrivals to the scene, and issues surrounding overfitting, unconscious biases and new models.
The group had a wide range of opinions on the future of talent in the industry, with two camps emerging. Modernists were accepting of new machine-learning teaching tools including Coursera, while purists thought those students would wind up degrading machine learning’s reputation within the finance industry. “It doesn’t matter if you do AI the right way [because] it’s all painted with the same brush,” one purist said.
The next era of machine learning could be a period with “more heat than light” according to one speaker. “People aren’t going to necessarily use [the tools] the right way and they’re going to make a lot of mistakes. It’s very obvious to me that we’re producing some very fragile systems.”
“Democratization of certain technologies and certain quant technologies, in the long-run, it’s really good. In the short-run, it’s a danger.”
The modernist camp argued that getting tools in people’s hands was valuable and that practitioners who weren’t good enough would get weeded out. “It’s very annoying for some of you guys who have done a PhD and spent years on it. You really know the nuts and bolts of it, but I’m sitting here and I’ve built a business. Our time frames are all measured by time and money,” one panelist said, explaining young people are becoming exposed to machine-learning concepts through these programs and that is valuable in-and-of itself.
“Democratization of certain technologies and certain quant technologies, in the long-run, it’s really good. In the short-run, it’s a danger,” he added.
The group raised another related issue — rebranding of old tools as new tools for marketing purposes. “Particularly deep-learning stuff has become such an artificial wrapper,” one attendee said. “I’ve seen so many people with these crazy pitches, they’re all the ARP strategies they had before.”
Active Vs. Passive
The investing world has experienced a shift from active management toward passive management over the last eight years, a move that has hit some quants hard.
Decision-making at larger financial institutions is largely still done in a qualitative way, even though it’s backed up by data, according to one speaker. “In some sense, quant has been relegated to the back- or middle-office [for] mass-scalable kinds of solutions.”
But another speaker disagreed. “I’m seeing a lot more allocators demanding more process-driven approaches from the buy side,” he said.
The group also expressed frustration that, in general, upper-levels of management didn’t seem to understand exactly how their strategies could be useful in the face of irrational human behavior. One trader provided an example of a stock whose price dropped due to a missed-earnings rumor that circulated hedge fund trading desks. Several quant shops bought the dip and when earnings were announced, the stock rebounded.
The discussion then moved toward overfitting and a discussion around how individuals can introduce bias into models simply by setting up the parameters. One participant spoke about facial recognition technology, stating that researchers didn’t tell the algorithms to look for different nose shapes or mouth shapes. Instead, they fed the machines the data and let them decide what was important.
He extrapolated this out to the financial industry. “When we give stocks and returns and book values and all these kinds of things that we’ve created in our infrastructure to a machine-learning algorithm, are we kind of just giving it: this is a nose, this is a mouth?” he asked. “Are we giving it our predefined versions of what should matter, because we feel comfortable?
“If there’s no human element to do a trade at all, then perhaps you can have a world where all computers are just talking to each other [but] we’re not there yet.”
The reaction from the table was mixed. One trader talked about the importance of using human intelligence early on in the process to classify data points to make them intelligible for machines later on. “Domain knowledge is important,” he said.
Another participant imagined that a future where computers do all the trading could solve this issue of unconscious biases. “If there’s no human element to do a trade at all, then perhaps you can have a world where all computers are just talking to each other [but] we’re not there yet.”
No commitment 30 day free trial
We deliver education to quantitative professionals on the latest innovations and developments in machine learning in investment management and other industries, such as tech, healthcare, transportation and education.
Get the latest exclusive news, innovations and approaches at financial institutions, tech firms and companies across the whole spectrum of machine learning.