Fawcett: Quantopian timestamps user strategies to eliminate overfit algorithms
John Fawcett · a16z Podcast | How Technology Is Changing Investing · Jan 2, 2019 · at 6:09
John Fawcett, CEO of Quantopian, explains how Quantopian detects and eliminates overfit quantitative trading algorithms.
0:00 / 0:27exact quote · 27.9s
▶ Watch the full episode on YouTube → 720p mp4 · rendered on demand · StarZero watermark
“So when someone comes to Quantopia, they actually are creating their strategy in the browser, you know, on our system. And so we're building this big database of strategies. And that one piece of information is incredibly important because it makes a bright line between the data that's in sample that the author could have known about when they were working. And then everything that happened afterwards. And so then you can do comparisons statistically between the out of sample and the in sample, and you can eliminate many overfit strategies.”
quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →
More from John Fawcett
Fawcett: Successful algorithmic quants spend 100x more time on research
“We've noticed that the people who have good outcomes out of sample, when we're testing after they've created an algo, spend about a hundred times the amount of time on the research step.”
Fawcett: Active asset management will always be required for institutional investors
“I think for institutional investors, there's a real need for actively managed portfolios because they have different needs, right? So they're not just trying to appreciate their assets. They have to be defensively posture. They have to be prepared for their ob…”
Fawcett: Quantopian aims for total coverage of real-economy datasets within 10 years
“And the thing that we want to accomplish over the next 10 years is if you think about the explosion in data from the real economy, we want to have total coverage. We want every single data set that exists to have, you know, several humans look at it, try to ex…”
Fawcett: Data cleaning and signal combination are ideal tasks for full automation
“Cleaning the data, tagging the data, getting it to our community to do the creative part of turning it into signals. And then taking those signals and combining them, those are the right jobs for full automation.”
Fawcett: Half of Quantopian's new traffic lands on technical research discussions
“About half of our new traffic land on discussions about things like Markov chain models, or common filters in trading, or different research topics that people are trying out, or papers they're trying to reproduce on the platform.”