Neuro-evolution of augmenting topologies (NEAT) is a recently developed neuro-evolutionary algorithm. This study uses NEAT to evolve dynamic trading agents for the German Bond Futures Market. High frequency data for three German Bond Futures is used to train and test the agents. Four fitness functions are tested and their out of sample performance is presented. The results suggest the methodology can outperform a random agent. However, while some structure was found in the data, the agents fail to yield positive returns when realistic transaction costs are included. A number of avenues of future work are indicated.
展开▼
