CREATING TRADING ALGORITHMS USING MACHINE MODEL BUILDING METHODS
Finding trading ideas and systems is no easy task. I know from experience that sooner than later even the most brilliant and diverse mind will run out of ideas. Not to mention the time-consuming task of back testing. The processes of system building not only solve the problem of idea generation, but also increase the speed of back testing by light years. After we have determined which financial instrument, we want to develop the system on, we define the data structure. Then we specify which indicators to consider in different combinations. The genetic algorithms test extremely large number of combinations based on the fitness functions and find the systems that are valuable. Once we have selected those with adequate past performance, we can test them on out-of-sample data. Systems that perform well both learning and testing periods are then subjected to validation procedures. At the end of the process, we receive the most robust systems, which we can also use during live trading. From the success point of view there is no difference between manual or generated strategies. A human built system can fail as much as a generated system. One important aspect of system building is diversification. It is a perfect way to generate multiple systems on different instruments to create an uncorrelated portfolio.
XI. évf. 2023. SPECIAL ISSUE1 57-61
DOI: 10.24387/CI.2023.SPECIAL ISSUE.9