Empower with A.I.

Computers are able to overperform human intelligence in many fields. This trend will continue and grow. We are here to help asset managers and leverage their decision process with A.I. and advanced learning.

Systematic Trading Platform

platform

We have developed a fully front to back platform, that is to say that it integrates all the stages of the management process, starting from:

  • data analysis
  • backtesting
  • optimization of strategies thanks to evolutionary optimization tools that avoid over-learning
  • machine learning tools to improve strategies using other data sources
  • and finally, the fully automatic execution of trades taking into account risk limits

This completely turnkey platform is an extremely powerful tool for asset managers. They are able to enter their own strategies, test them and improve their decision process using latest machine learning techniques.
The machine learning techniques are designed to detect a specific market environment, and adapt the strategies accordingly, rather than following explicit rules. Prediction is generated by analyzing large data sets from multiple sources, learning from previous trades and finding the best strategy depending on market conditions. Noise in the data is reduced by employing filtering and dimensionality reduction techniques. We have developed 3 state of the art machine learning methods that helps asset managers fine-tuning their investment decisions:
- evolutionary optimization methods (swarm intelligence and Bayesian CMA ES o optimization) to find the best parameters in investment decisions
- supervised learning methods based on decision-tree-based ensemble methods to classify the outcome of investment decisions
- deep reinforcement learning methods to decide the best allocation between various strategies
Read more

Strategy Description

We created an extension of the C# language to describe trading algorithms simply. We are able to write in almost plain English all types of algorithms to analyse, optimise and use machine learning techniques to make trading decisions.

algo

Cutting Edge Optimisation

opti

We have developed evolutionary optimisation methods to find the best parameters of the strategy while avoiding over-fitting. Our innovating technology handles problems in high dimension for non-convex and discrete optimization cases. Based on particle swarm, CMA ES and genetic algorithm, these new methods are based on work in neuroscience and metaheuristics.

Supervised Learning

We use supervised learning with more than 100 variables derived from price and alternative data (macro data such as non-farm pay-roll, volatility and market volume, sentiment index, FED meetings, etc.). The algorithm analyses the impact of these data to retain only the most explanatory variables. We apply gradient boosting methods in random forest algorithms to give a probability on the prediction.

sl

Reinforcement Learning

rl

Reinforcement learning methods have been used by Google DeepMind, the first computer program to defeat a world champion at the ancient Chinese game of Go. Unlike supervised learning where the machine is given right and wrong answers, reinforcement learning finds the right action to optimize its "Reward" according to the state of the market.
The optimal action is usually a complex function. Neural networks are a way to represent this type of function and to solve this type of complex problem.