A Hybrid Genetic Programming-Particle Swarm Approach for Designing Trading Strategies in Software and Hardware

The aim of this project is to investigate the design and implementation of high frequency trading rules using genetic programming and swarm intelligence. The approach taken was to design, build and test two different versions of our basic approach: a genetic program which digs out relevant trading signal information from tick level Foreign Exchange market data and an adaption of this genetic program which contains a particle swarm optimisation on top of the original algorithm. Because of their nature, evolutionary algorithms are generally considered to be slow. After observing that the fitness evaluation part of the algorithm is where the majority of the computation time is absorbed, we reduced this CPU original time by implementing an alternative version making use of the Field Programmable Gate Arrays (FPGAs). Thus, we show in the thesis how we implemented the trading algorithms on different architectures, while also showing the interaction between the CPU and FPGAs. The application of the combination of genetic programming and swarm intelligence within a hybrid CPU/FPGA environment is as far as we know novel and has not appeared in the computational or financial literatures. We obtained good robustness results and found significant profitability after performing a range of statistical tests and we concluded that our approach is reliable under stable market conditions and it does not provide spurious signals in the trading rules where there does not to be relevant information in the market data. We have also shown that the degree of predictability and hence profitability has decreased from 2003/4 to 2008 which is consistent with the rapid growth in Algorithmic Trading in Foreign Exchange Markets.