Implementing and comparing various algorithmic trading strategies

Most financial firms use algorithms to buy and sell financial assets. It is possible for amateur investors with programming knowledge or vice-versa, to implement algorithms and improve their strategies. In the theoretical part, the history of stock exchanges, analytical methodologies, strategies and testing process are discussed. For the empirical part, a few algorithms based on different strategies are written. Though the algorithms’ results are analysed and compared, the main goal is to implement strategies, not find the one with the best results. A website called Quantopian is used to recuperate the data, write the algorithms, run the program and analyse the results. The financial assets that are traded are US stocks. The period used to test the algorithms goes from the 1st of January 2004 to the 1st of January 2018. This enables us to test long term strategies: at least 10 years and see the impact of an important crisis: the subprime crisis of 2007. Though the algorithms trade US stocks, they are stock-generic, which means that they will work with stocks from any stock market. The programming language used is Python, since it is the only language supported by Quantopian. This paper discusses basic strategies for different investment styles and how to improve them. The two algorithms and their different versions are compared to each other and the best algorithm is chosen based on the results and risks. Tools are used to try and find the best elements possible so that the second algorithm can beat the first. In this paper, we see that the first version of the first algorithm has the best returns. The focus is on the use of tools to implement and improve these algorithms. The two main algorithms beat the chosen benchmark. However, it was expected that the second algorithm would beat the first one, since a separate tool was used to try and improve it. The results do not show that the tool is useless. Demonstrating the use and results of the tool in question requires more tests and data