Work detail

Application of Machine Learning in Portfolio Construction

Author: Bc. Ondřej Karlíček
Year: 2021 - summer
Leaders: Mgr. Jan Šíla MSc.
Work type: Bachelors
Language: English
Pages: 68
Awards and prizes:
Abstract: The thesis investigates the application of machine learning in portfolio construction. The analysis was conducted on a dataset consisting of 442 American stocks. Initially, we cluster stocks using Principal Component Analysis
and K-means algorithms. Then we select stock from each cluster based on
return/risk metrics. Where risk was estimated by Value at Risk, and return
was predicted using Random Forest and GARCH models. This leaves us with
11 stocks for every monthly period during 2020. The results indicate that
the portfolios constructed from the selected stocks were able to outperform
the market benchmark. However, the return predictions were not accurate
enough. Thus, the portfolio from selected stock using the 1/N approach
achieved better results than the portfolio optimized by the Mean-Variance




Patria Finance