Lecture: Autotrading with a synthetic market: AI-driven back-testing environment
Our most popular product is an automated energy trading solution: a bot, written in Python, trading 24/7 at various European power and gas exchanges. Furthermore, the solution allows our customers to develop their own trading algorithms. However, the lack of a realistic testing environment prevents the performance analysis of these trading algorithms. In particular, how can we know how the market would react to trades executed by our bot, without actually trading live on the real market?
One possible way to solve this problem is what we call a synthetic market, built with the help of artificial intelligence (AI). Recurrent neural networks (in particular Long short-term memory neural networks) are very powerful tools for recognizing and learning patterns evolved with time. In this talk we will show:
- how a suitable neural network can be created in Python using keras package with Tensorflow backend;
- how it can be trained to learn the behavior of a real energy exchange;
- how it generates market activities emulating the exchange.
This will enable us to test trading ideas in a more realistic simulated environment, and also to utilize AI for creating more sophisticated trading algorithms.