Rule Based Forex Trading System by Zezheng Zhang

Technical analysis assisted forextrading requires lots of practice and experience that could potentially be learnt by machines. Professional traders make decisions based on prior knowledge with the assistance of technical indicators. In an effort to learn the trading rules using machines, a set of trading rules based on different technical indicators are created and optimized on the EUR/USD 5 minute data from previous year. Using genetic algorithm and designed loss function to minimize risks involved, the position and trading orders couldbe learned using a linear combination of trading signals from rules. The performance is tested on the same currency pair at later period and compared with “always in the market” strategies using always buy, always sell, random forest and SVM models. The results showed promising returns using genetic algorithms and trading rule based models which indicates the trading rules are capableof predictingthe trend to some extent. Genetic algorithmmaximising Sharpe ratio is the best performing model whichcould lead to less capital in the market and more return for the unit of risk bearing.