Keeping up with trading computers is getting even tougher, says Jitesh Thakkar, as the new algorithms have technology that allows the systems to practically "think" and "learn" on their own.
The future of high-frequency trading and automated trading. We’re here with Jitesh Thakkar, and he is talking to us about what’s happening in the world of high-frequency trading.
What I see happening is a trend called machine learning—algorithms that evolve. For the most part, the algorithms that are deployed today are static algorithms that are predefined on certain buy and sell signals, but they are static.
So right now, we’re working with just simplistic if/then statements, but as they’re developing, they’re becoming more fluid and anticipating instead of waiting for the binary answer from the first and previous decision?
Absolutely.
So this is transitioning out of AI, artificial intelligence, where it sounds like we’re getting to the point where the machines are becoming more and more human in the way that they’re looking at their possibilities.
Right. Machine learning algorithms really are algorithms that change their decisions based on past successes or past failures, whereas the majority of algorithms deployed today don’t do that.
And you’re right, artificial intelligence really started in the 70s, and the applications were deployed in defense and the anti-missile technology and other technologies.
But now, I’m seeing the top-tier hedge funds starting to use these in the last four or five years. Really, it’s in the nesting stage and it’s in its infancy—the whole industry of machine learning-based algorithms—but I do see that as a future of automated trading.
And I would assume that some of this has to do with processing power, too? In the 70s, AI was really more conceptual than it was realistic. But recently, you have computers that are winning on Jeopardy.
Even on Wall Street, they have newsletters that are built for computers that are actually algorithms that the trading platforms read without any human intervention to decide what they’re trading that day.
Right. It’s a combination of multiple things, and you hit it right on the head. IBM has the Watson Project, which is a perfect example of machine learning.
They built this machine that beat all-time Jeopardy champions Ken Jennings and Brad Rutter. That actually was aired on February 11, 2011.
And the whole idea behind the Watson project was that they built a system. Initially, when they gave it all of the databases of knowledge and all of the encyclopedias and taxonomies, it was only able to answer successfully 15% of the time.
They then went back to the drawing board and made one key change that took it from 15% success ratio to 95% plus. That one change was machine learning, so Watson was able to evolve its answers based on past Jeopardy questions and past correct answers.
Related Reading:
