Portfolio Management: A Machine Learning Perspective

Author

Chih-Jui Tsen

Date of Award

2020

First Advisor

Tai Young-Taft

Second Advisor

Harold Hastings

Abstract

Machine Learning (ML) has steadily been advancing at a respectable rate ever since the cost of computation decreased substantially and the majority of the world’s infrastructure shifted towards digitalization. Coupled with traditional statistical modelling processes and methods, ML revolutionized numerous industries and continues to be developed with improved applicability and variability to tackle specific domains. This thesis explores the possibility of formulating a profitable and goal-oriented portfolio with a combination of ML techniques and statistical tools. This thesis provides an exposition of methodologies as an elucidation of basic functions and capabilities of analysis processes and as an investigation of the methodologies’ applicability pertaining to the financial sector. The experiments carried out utilizes variant of the clustering algorithms, such as spectral and hierarchical clustering, for portfolio generation and asset pairing, followed by regressive and classifying tasks performed by neural networks with varying architectures. Although the experimental results did not immediately invoke production-grade confidence and simulate massive profits over the index which warrants an application adaptation, the results provides domain specific insights beneficial to the creation of more capable models that should bring the reality of mass-accessible automated portfolios closer to realization.

Recommended Citation

Tsen, Chih-Jui, "Portfolio Management: A Machine Learning Perspective" (2020). Senior Theses. 1463.
https://digitalcommons.bard.edu/sr-theses/1463