HOW IT WORKS
What's the best trading strategy? Rather than articulating a series or hierarchy of trading rules based on statistics and derivatives, the non-deterministic design of the Coindex Labs AI is based on a simple premise: all possible market knowledge is contained within price action, and all market factors are discounted in price.
It follows that a true AI can use comprehensive, granular price action and order book data to derive any and all possible patterns from a market without being influenced by a codified human pretense or unintentional bias via intentional human design.
MULTI-DIMENSION SEARCH SPACE
Using technologies developed at Harvard and pioneered by Uber's AI team, Coindex Labs has created a finance neuroevolution platform.
A compositional pattern producing network (CPPN) is a neural network that composes typically 2-dimension patterns against genetic functions and assesses the given pattern's fitness. This fitness is "how well the AI trades".
The Coindex Labs proprietary multi-dimensional CPPN has capabilities into dim-3 and dim-4 search space. The result is a neural network whose learning and intelligence pathways are comparable to a physical brain.
The platform endures an ongoing evolutionary process in which trading "species" constantly evolve and mutate, attempting to trade their strategies competitively in a market.
The financial successes and failures of each species gets encoded into a financial genome that represents a comprehensive market knowledge and understanding of market mechanics.
This genome is deployed by the species within a real market through a 'survival of the fittest' protocol that measures its real-world capability, rewarding success and punishing failure. This wisdom is then inherited as a phenotype and evolved upon by subsequent species, as they develop their strategies.
NOVEL SEARCH WITH AI
In today's financial markets, successfully finding alpha requires far more than simply repeated trial and error of every possible solution. This has been the downfall of AI trading - such an approach isn't efficient and leads to countless dead ends and design ruts, or strategies that work until they don't.
That changes with novel search.
Coindex Labs uses a causal structure that not only rewards and punishes success and failure, but also provides context for how novel and unique each generative species is compared to the existing wisdom base.
This focus on novel search leads to a more efficient learning process with a wider base of exploration, and ultimately more successful strategies.
The platform gains one of its greatest strengths from its non-deterministic genetic design in trading fitness. This design unlocks the possibility of contextual intelligence.
Most AI is inflexible such that it is essentially a machine-trained static system that requires identical parameters and conditions to be able to perform with any consistency to backtesting.
The Coindex Labs AI has a flexible architecture that enables a trading species to rapidly adapt to changing, or suddenly different market conditions and parameters.
This means that an oil trading species can instantly adjust to a market where Oman is no longer part of OPEC, and that a SPY species is also capable of trading VIX.