Using K-Nearest-Neighbor with valuation metrics to detect similarities between stock performances

Abstract: Algorithmic trading has increased in popularity since the publication of Agent-Human Interactions in the Continuous Double Auction by IBM researchers Das et al. (2001). Today many investors acquire algorithms that act on their behalf on the stock markets. Most of the algorithms have worked on predicting stock prices and making transactions when price thresholds are triggered. This project has a different objective and aims to construct a machine learning algorithm to cluster stocks with similar stock performances, and ultimately test the possibility if such stocks continue to perform similarly in the future. The KNN-model succeeds in its mission to cluster stocks with similar market performances. Statistical measurements highlighted a moderate correlation amongst stocks and their neighbors. Furthermore, some stocks did not continue to perform similarly in the short-term future, and the main reason has been of natural causes, such as management changes, and not meeting market expectations. Those factors impose a possibility for stocks to break their developing pattern at any time and move in a different direction than expected, which imposes a substantial limitation when clustering stocks that are expected to perform similarly in the future