Financial Genetics

Producing Financial Software using Genetic Algorithms and Machine Learning

About Financial Genetics

Nobbis Ltd trading as Financial Genetics is a software consultancy specialising in financial software.

We are using genetic algorithms and machine learning to optimise algo trading.

Currently we have a exchange simulator allowing entering of equity orders, with AI agents providing trades to track real market movements .

Future developments will allow optimisation of algos, and sentiment analysis.

Please contact us if you would like access to the simulator via a FIX connection.

Design of a 2 Exchange Simulator

The diagram shows the proposed architecture for a 2 exchange setup, in reality there will be more, representing dark and lit pools. As shown agents will submit orders and carry out arbitrage between the servers. Clients will be able to input orders through either an app or a web interface. Algos will be sent to an algo runner and optionally optimised . A mixture of protocols will be used, where performance is critical a proprietary binary interface will be used, otherwise FIX 4.2 will be used.

Algo Optimization

This area is still under development, though much of the genetic algo work has been completed.

The aim is to produce a DSL for an algo which can then be implemented by a script runner. The algo will send orders to the exchange simulator and the genetic algorithm engine will optimise it

Machine Learning and Sentiment Analysis

The next stage of development will use NLP processing for sentiment analysis and a Deep Learning Convolutional network to analyse time series data from the markets.

Sentiment Analysis Design

Development News

September 2014

Development is proceeding in a number of directions

We are currently re training the Stanford NLP with our own data , this is a laborious process taking thousands of sentences and applying sentiment values to them

Once that is done there will be extensive testing against the corpus I have been collecting, then further analysis against various financial parameters

Development has also started on a machine learning framework to analyse both textual data from twitter and time series data from the markets and the simulator.

In addition to the FIX based messages we will be introducing messaging based on the Simple Binary Encoding Format Simple Binary Encoding Format to improve latency when translating messages.

June 2014

The matching engine is finished and the simulator is up and running on its new host.

March / April 2014

Coding of the matching engine is almost complete, enough to produce background orders. Once the various clients have been wired back into the application you will be able to send in orders and set up portfolios. The previous server has been retired and we are on a temporary host until a better hosting solution is found.

October 2013

A great refactoring of the matching engine (virtually a rewrite) to produce better performance, particularly with concurrency in mind.

June / July 2013

The first iteration of the genetic programming side has been completed there is a general framework with a couple of specific implementations of decision making features : - a simple virtual machine and a simple neural network.

Contact Financial Genetics

Write to:

Nobbis Ltd, Suite 206,

1 Royal Exchange Avenue,

London EC3V 3LT.


We are located in the heart of the city of London