Hedge Funds Q&A: Adrian de Valois-Franklin, Castle Ridge Asset Management
The explosion in the use of artificial intelligence since the unveiling of ChatGPT in the autumn of 2022 has businesses in all industries scrambling to try and figure out how best to incorporate the technology into their processes but the hedge fund industry has been using variations of the technology for a while. AlphaWeek’s Greg Winterton spoke to Adrian de Valois-Franklin, CEO at Castle Ridge Asset Management, to see how AI is working in the hedge fund space.
GW: Adrian, for those perhaps less familiar with Castle Ridge, tell us a little bit about your firm.
AdVF: Castle Ridge Asset Management launched in 2015, and we’ve been exclusively focused on self-evolving AI-powered investment strategies since our inception. We offer hedge fund investment strategies powered by a proprietary AI system we call WALLACE - it continuously learns and evolves by detecting sustainable behavioural patterns in vast market data (fundamental, technical and sentiment). WALLACE also uses two completely new approaches to machine learning that we have built; Ranking Inference Engine (RIE) and self-evolving Geno-Synthetic Algorithms (GSA), which allow WALLACE to uncover unique market patterns that go far deeper than typical quantitative strategies.
GW: How does Castle Ridge incorporate AI into its strategy?
AdVF: As I mentioned, WALLACE is our AI platform, and it’s the culmination of over 20 years of research into what works and, more importantly, what does not work when trying to apply AI to financial markets. We don’t see this as ‘incorporating’ – WALLACE is the strategy from idea generation to trade execution.
It’s important to mention here that WALLACE doesn’t use ‘off-the-shelf’ black box machine learning approaches such as traditional neural networks and deep learning. These are acceptable tools to train static large language models (LLMs) or generative pre-trained transformers (GPTs), for machine vision (the definition of a human face or cat doesn’t change), or to teach an AI to play chess or Go. But in financial markets, the rules change at every turn, so you need an AI that can learn and adapt to an ever-changing environment. This is the challenge that static machine learning techniques face when applied to the complex world of financial markets – ultimately, they work until they don’t. So, we developed an entirely novel AI approach called Geno-Synthetic Algorithms, which analyse thousands of securities, each from over 42 dimensions simultaneously, and monitors complex interrelationships between these securities over time. As a result, WALLACE uncovers market patterns two to three levels deeper than typical quant factors.
GW: You’ve now built a supercomputer. Why?
AdVF: The supercomputer project was necessary because traditional AI hardware was not designed for the way Geno-Synthetic Algorithms work. Most commercially available systems are ‘tuned’ for deep learning algorithms. Compared to mainstream neural networks, LLMs and GPTs, Castle Ridge’s GSAs are far more computationally complex.
The GSA approach sets itself apart by its ability to find optimal solutions in a problem space defined using various data types, including integer, Boolean, float, decimal, and complex (imaginary and real) numbers. When dealing with such a vast hyperdimensional search space (think cosmic scales), a single optimization technique run on traditional computing hardware could take many years to find an acceptable solution. However, constant market fluctuations demand investment decisions in a matter of hours and minutes, so to handle this computational complexity, we had to design custom hardware, chassis, cooling and software for the supercomputer.
GW: There will be investors out there that still get twitchy about allocating to a strategy where there’s almost no human intervention. How does an AI strategy hit the brakes?
AdVF: The main thing I’d point out here is that we are responsible for telling WALLACE what kind of mandate we seek and setting the boundaries for that – whether it’s an equity market neutral strategy, a long/short strategy or a global currencies strategy, we set the risk limits, and WALLACE works within those limits. But it can adjust risk depending on the client’s risk appetite, as well as rapidly create and test bespoke strategies for clients to fill specific gaps in their portfolio construction.
The sudden buzz around AI has been a blessing and a curse for us. Allocators are trying to play catch-up and realize they can benefit from the uncorrelated nature and unique signals of investment strategies designed by AI. At the same time, the hype also creates a lot of noise for allocators, with some managers using AI more for marketing purposes.
There are few ‘pure-play’ AI/machine learning hedge funds globally – those that are fully systematic investment strategies designed by an AI from the ground up. While more-and-more hedge funds have started to claim the use of AI for marketing purposes, it is often applied as an afterthought to strategies that human portfolio managers designed. Adding further confusion, many firms now quote a “quantamental” approach. This approach often translates to a glorified AI screening tool, with humans overriding the AI decisions due to a lack of confidence or transparency in the system.
GW: Lastly, Adrian, what else is in store for Castle Ridge in the coming 12-18 months? How do you even top a supercomputer?
AdVF: The supercomputer really enables us to use WALLACE’s full capabilities to run metaheuristic and hyperdimensional models, so we’re using it to develop next-gen strategies, including multi-strats across asset classes. The higher-octane versions of these will be capacity constrained with multi-year lockups. Think of it like Castle Ridge’s very own “Medallion”.
Our goal was to build a powerful supercomputer specifically designed for the types of algorithms we use (GSA). So, we now have the ability to run new experiments and to do things with GSA that we couldn’t before—for instance, using metaheuristics to run GSA models on top of GSA models. The platform continuously evolves through a survival-of-the-fittest mechanism. WALLACE generates tens-of-thousands of generations of virtual portfolio managers, all competing to improve performance. In under an hour, WALLACE achieves what would take millions of years of natural selection in the real world.
As a result, we are creating new products, and working with clients to tailor specific mandates depending on their needs. The supercomputer has opened up some quite incredible possibilities for us.
**********
Adrian de Valois-Franklin is CEO at Castle Ridge Asset Management
© The Sortino Group Ltd
All Rights Reserved. No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or scanning or otherwise, except under the terms of the Copyright, Designs and Patents Act 1988 or under the terms of a licence issued by the Copyright Licensing Agency or other Reprographic Rights Organisation, without the written permission of the publisher. For more information about reprints from AlphaWeek, click here.