AI Pushes Quants in New Directions
For a discipline that began about 60 years ago, artificial intelligence aka machine learning is gaining a higher profile the trading desk.
Access to new data sets and greater computing power delivered via cloud computing and graphical processing units has fueled interest in machine-learning based strategies, accord to panelists who spoke at this week’s STAC Summit in New York City.
Until recently, most quantitative traders have relied on easily consumed structured market data, said Robert Passarella, a manager at Protégé Partners and one of the panelists.
However, the amount of unstructured quant data, plus structured and unstructured non-quant data, has grown almost in line with Moore’s Law, doubling roughly every 18 months, he noted.
Access to this new wealth of data has changed the mindset of many quantitative traders, eschewing the back-testing of potential trading signals for running various simulations instead.
Although AI-based asset manager Qplum has done quant trading for a while, it found that quant trading cared more for overfitting, which occurs when a statistical model describes random error or noise instead of the underlying relationship. according to Gaurav Chakravorty, head of strategy development at Qplum and who also participated in the panel discussion.
“Our main work would be less overfitting,” Chakravorty added. “We’re more concerned about getting better results tomorrow. I don’t care about the past.”
He believes the next step that will entrench AI into trading will be the reliance on backward-facing research.
“It’s just like when I go to Chicago and need to find a good restaurant,” Chakravorty said. “I’m not going to call up a local friend to ask for recommendations. I’m going to search on Google.”
With some AI-based investors running multiple market simulations concurrently and paying roughly $11 per hour for their cloud computing resources, Passarella predicts that such inexpensive computing costs will only whet the appetites of firms to run more and more simulations.
Yet, some of the new data sets that some AI-based strategies might use lack historical depth and only provide data for the past three to four years, cautioned fellow panelist Puneet Batra, director of research and trading at SCT Capital.
“We haven’t seen all of the economic cycles and how that data would look,” he said. “There are people using satellite imagery to figure out how many cars that Ford is shipping from its factories or figuring out how many people are signing up for Netflix via credit card receipts and things like that. That’s fantastic on shorter time scales, but once you get to the macro level you have to be careful. For example we haven’t seen a bond bear market since 2004.”
More on Artificial Intelligence:
Data extraction and integration is the second stage of a digitization process.
Increased electronification has created useable and accessible real-time and historic trade data.
The fintech uses data so institutions can assess the environment impact of their portfolios.
Working with Riskfuel will reduce reliance on slow and expensive financial models.
The platform uses AI to match experts and users and digitize the research process.