Suggestions
Peter Cotton
Chief Data Scientist at ExodusPoint Capital Management, LP
Peter Cotton is a highly experienced data scientist currently leading data science at Intech Investments. He specializes in portfolio construction and has made significant contributions to the field with the invention of Schur Portfolio allocation. Peter is the brains behind 'microprediction,' including packages, a book, and a live probability exchange.
Before his current role, Peter founded Benchmark Solutions, a data company acquired by BBG, and worked on innovative applications of control theory in OTC trading and privacy-preserving Machine Learning at J.P. Morgan. His career kicked off by pioneering closed-form CDO pricing at Morgan Stanley.
Peter holds a Ph.D. in Mathematics from Stanford University and a BSc Hons in Physics and Mathematics from UNSW. Previously, he has held key positions at ExodusPoint Capital Management, J.P. Morgan, Benchmark Solutions, and Morgan Stanley.
Peter is a valuable resource for those interested in data science, offering insightful advice to job applicants. He keeps himself engaged during the pandemic with activities like zwift, bullet chess, and amateur epidemiology.
For those seeking assistance, Peter provides free services like live model residual analysis, benchmarking of time-series algorithms, and predicting live measured quantities. He also recommends valuable resources such as the 'Quant Book of the Year' and offers ways to connect through Slack and office hours.
Peter Cotton's expertise spans a wide range of domains, showcasing his deep knowledge and innovative mindset in the world of data science and portfolio management.
Highlights
There's a new version of ChatGPT out today that claims reasoning and problem solving ability.
So, I asked it to place the 11 financial sectors in a sensible ordering, where those that are close to each other are similar, and conversely.
This is called seriation in statistics, and it sits at the heart of some robust portfolio techniques such as Hierarchical Risk Parity by Marcos Lopez de Prado (or my generalization https://lnkd.in/eQk42ibj).
Open the notebook to play the same game. See if you can do better than "hey seriation". Of course you can generalize to stocks.
I was about to say it is easier than manipulating the linkage diagram derived from a correlation metric, but with the improved coding ability that isn't really true, is it?
Notebook: https://lnkd.in/e9rJGkCH
Maybe you can use it here: https://lnkd.in/ew9PdSHr
#portfoliomanagement #investing #ai #statistics
In this book I asserted that data science would eventually splinter into a decentralized collection of self-navigating itinerant algorithms coordinated mostly by market-inspired statistical and economic games.
If I were to say this is now ten times more likely, that would put an upper bound on the original probability :) ... so perhaps I should merely make a few observations:
-
The market is speaking. However you choose to measure the money directed at decentralized AI, it's hard to come away with a view that this hasn't increased by one, or possibly three, orders of magnitude.
-
Algorithms can navigate. Language is almost API and the gap will close. One has to be a pretty grizzly technology permabear to deny the weak intelligence possessed by large language models (especially if we use navigation as part of the definition of weak intelligence as per https://lnkd.in/eHPw8aRT), or the profound implications this has for the core thesis.
-
Algorithms are more general and adaptable. Another part of the thesis is that cost is driven towards zero due to reuse of algorithms. For the first time in years I've been finding that general purpose algorithms match auto-arima and friends with just a little fine-tuning, just to pick one domain. Feel free to use my frameworks to convince yourself (https://lnkd.in/gsFKXN2).
-
More things can be digitized by large language models, greatly increasing the fuel for prediction algorithms.
One thing hasn't changed - although there's a little more evidence since the book was written. Markets are still indispensable for prediction and it takes hubris to believe otherwise. This has always been obvious to a small quant community, but after the forecasting, machine learning and data science communities got embarrassed by my trivial entry in the M6 Competition (see https://lnkd.in/egVqVdqd), maybe the message is spreading, just a little bit.
#artificialintelligence #ai #datascience #prediction
https://lnkd.in/eUXX7KCW
