Neurosymbolic AI Framework
Our approach to Artificial Intelligence, for investing in and modeling financial markets, is summarised below. This comes from our extensive experience in applying AI to finance, and in particular dealing with noisy and non-stationary time series.
Our approach to AI is one of using neurosymbolic techniques, centered around fuzzy time series based causal relationships, in order to create stable, performant, and explainable investment systems.
We use causal, regime based models to generate non-linear AI driven investment factors, which are used for investment decisions and risk management. These are then accessible via investment products, research, and insights-as-a-service based software systems.
Causal Models
Causal modelling provides several advantages in the context of designing investment strategies with stable and repeatable performance over long time periods.
Underlying our AI based investment system is a causal graph of relationships, which provide non-linear investment factors that help to better predict asset returns.
Multiple Techniques and Time Horizons
We make use of multiple techniques from the AI field. Broadly speaking, we use: unsupervised learning, supervised learning, and reinforcement learning across our various systems. We also target multiple time horizons - from days and weeks down to intra-day.
We have a deep interest in combining models and representations of information in order to create more generally applicable, and less brittle, AI based investment systems. We make use of both statistical and symbolic techniques, and so attempt to replicate some of the likely processing in human minds.
Regime Based
Regime based models can be very intuitive. Driving regimes and their relationship with asset returns result from the causal relationship graph described on this page.
Our system is event driven in the sense that regime changes result, sometimes, in prediction changes for asset returns. The regimes our AI system deal with are generally much more non-linear and granular than those a human being would consider.
Explainable
We have been working on eXplainable AI systems for many years (hence our name, XAI), and our internal, proprietary prediction engines make use of this knowledge and experience.
Being able to understand what is driving the system and its non-linear investment factors at any point in time is very useful, both for internal diagnostics and understanding, as well as communicating with investors.
Our Services
