We achieved quantum advantage in portfolio optimization (in R&D mode) using a DWave quantum annealer6/11/2020 Over the past 3-4 months, our portfolio team (Jeffrey Cohen, Alex Khan, Clark Alexander and associate Professor Terrill Frantz) has focused on moving from classical computation of portfolio optimization to running that same code on a quantum computer. We have been running on the DWave system for a few months, and finally feel comfortable with how to operate it, tune it, speak to it in its own language, and achieve results we can verify and feel confident in sharing.
I am proud to announce that last night we achieved that milestone. We were able to optimize a set of 40 assets and select a (group of) portfolios that are best to invest in for someone who cares about minimizing risk and maximizing return, or about balancing those metrics. We will be publishing our results. Three things to share: 1. We learned a great deal about how to formulate a problem for a quantum annealer. You see, quantum computers don't multiply nor divide. They are not Microsoft Excel. You have to think differently to use them. 2. Quantum computing time, and our own time, is precious. We used just a little each month, and learned how to simulate quantum annealers very well. We could run a 'quantum computing' problem in simulation mode, exactly matching the quantum energy levels, using Python for up to 24 assets in seconds, 28 assets (268m) in a 1-3 minutes, and 32 assets (4.2b) in just over a day on a 48GB RAM server. We could not achieve brute force classically with 40 assets...we would lose too much time. On a quantum computer, 32 assets takes just a few seconds. 40 Assets classically was beyond our reach. 3. We are having fun doing this. We have Google Hangouts meetings for hours, code Python and DWave into the late night, dig into the math, physics, and economics when needed to crack a problem, and I think this has given us a better understanding of the equities markets. If you check our company's Twitter handle, @chicago_quantum, you will see posts on money flows that are well beyond our comprehension when we started this effort. We will publish our findings in a technical way (arXiv or a publication - call us Physics Review, American Banker, McKinsey Quarterly, and Harvard Business Review), and in non-technical ways (e.g., YouTube videos, Medium Articles, LinkedIn, and Tweets). Thank you, Jeffrey Cohen.
6 Comments
6/12/2020 22:57:56
1. Quantopian does classical portfolio optimization
Reply
6/15/2020 12:08:44
Hi Robert, good to hear from you. Thank you for the comment.
Reply
6/15/2020 13:54:45
Thanks for the reply. 7/1/2020 11:10:47
Looks interesting however one should be careful in portfolio computational approach as the market is not behaving logical, not taking into account a possible recession or even depression! Hard to forecast based on historical data . Yet could be useful. Curious is this would work for bonds .
Reply
7/1/2020 11:55:57
Hi Fred, we built our model as the markets fell, and tuned it as markets rose (and learned how to adjust for this). When portfolios are selected based only on historical variance / covariance, it removes noise and builds confidence. We also saw significant reversion to mean @ 5 years.
Reply
7/6/2020 12:46:06
Here is the arXiv link. It went live late last night:
Reply
Leave a Reply. |
Jeff CohenStrategic IT Management Consultant with a strong interest in Quantum Computing. Consulting for 29 years and this looks as interesting as cloud computing was in 2010. Archives
February 2021
|