Evaluation of an unsupervised learning approach for portfolio optimization

Abstract

Throughout this directed research, we aim to identify opportunities for machine learning to support portfolio optimization. Based on a thorough literature review we decide to pursue an unsupervised learning approach and test its performance by conducting benchmarking against classic portfolio optimization techniques. To ensure the validity of our findings we explore the model’s robustness by conducting an array of experiments. In summary, we deem our version of the clustering algorithm to provide a suitable investment framework for return-focused investors with lower risk aversion. We suggest further research towards mitigating the algorithm’s inconsistencies and exploring additional tuning methodologies.