Traditional quant strategies tend to be static. Smart portfolio managers or data scientists come up with a recipe to beat the markets, and then automate their thinking into a set of rules-based algorithms. These static quant strategies work until they don't. This problem becomes acute when markets change rapidly, as they are doing today, says Adrian de-Valois Franklin, CEO, Castle Ridge Asset Management.

But the natural world offers us a different approach, that of evolution through random mutation. This is Darwin's theory of natural selection, and it underpins our understanding of all life today. It explains how mammals that ventured into the sea millions of years ago could have evolved into the whales, or how the dinosaurs of yesterday became the birds of today.

Darwinian evolution applied to investment management is an attractive concept. Investment strategies would evolve as markets evolve thanks to millions of random mutations, just like life itself. Those mutations that conferred an advantage would be more likely to be retained by future generations, just like genetic mutations of DNA in the natural world. Trading strategies which survived would be those best adapted to markets, just as natural evolution rewards those species which adapt best to their habitats.

The issue with this approach has always been computing power, because to do this properly requires enormous raw technological capacity. But that has changed in recent years.

Castle Ridge Asset Management launched in 2015, and we've been exclusively focused on self-evolving AI-powered investment strategies since our inception.  We created a proprietary AI system we call WALLACE, named after Alfred Russell Wallace whose own theory of natural selection inspired Darwin to publish. WALLACE continuously learns and evolves by detecting sustainable behavioural patterns in vast market data (fundamental, technical and sentiment).

And we developed an entirely novel evolutionary 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.

We have now built a new supercomputer to handle this, because existing computing hardware wasn't designed for the way Geno-Synthetic Algorithms work. These days, most systems are "tuned" for Deep Learning types of algorithms. Chipmakers have a "blind spot" when it comes to Evolutionary Computing. Compared to Deep Learning, our Geno-Synthetic Algorithms are far more computationally complex.

Geno-Synthetic Algorithms sets itself apart by its ability to find an optimal solution in a problem space that is defined using various data types, including integers, Boolean, floats, decimals, and complex (imaginary and real) numbers. When dealing with a vast search space (think cosmic scales), a single optimization technique run on traditional hardware could take years to find an acceptable solution. We need to do it in a matter of hours, every day, as markets and information change.  

On top of that, our goal was to bring WALLACE to a new level with Metaheuristics. Instead of a singular model running millions of years of evolution across thousands of securities in the universe, we could now run Geno-Synthetic Algorithms models for every security, allowing for a much more refined optimization. It is similar to having a specialist portfolio manager for every stock.

Building a meta-model on top of it enables WALLACE to assemble the best strategy and select the best portfolio for a given universe of securities. You can think of it like trying to solve a Rubik's cube, where every small square is itself another Rubik's cube. The complexity explodes exponentially.

To handle this computational complexity, Castle Ridge had to design custom hardware, chassis, cooling and software for the supercomputer. The WALLACE supercomputer packs thousands of square feet of traditional hardware footprint and datacentre level cooling requirements into a proprietary 7.5 by 5.5 foot fluorinert liquid-cooled system.

The WALLACE AI platform is 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. What sets WALLACE apart is its transparency and ability to explain every decision it makes. WALLACE replaces an army of human portfolio managers and analysts, requiring no sleep, vacations, or pep-talks. It is relentless in its goal, spotting hidden market opportunities to generate independent Alpha and maximize risk-adjusted returns.

Overnight, WALLACE can generate tens of thousands of generations of virtual portfolio managers. Each virtual PM has different characteristics, with the most fit breeding to create an even stronger generation of PMs. In only a few hours, WALLACE achieves what would take millions of years of natural selection in the real world. Darwin himself would surely have approved of how technology has liberated his theory and put it into practice in new ways.

By Adrian de-Valois Franklin, CEO, Castle Ridge Asset Management.