An Evaluation of Self-Supervised Learning for Portfolio Diversification

International Conference on Artificial Neural Networks (ICANN) 2023

12 Pages Posted: 15 Aug 2022 Last revised: 10 Jul 2023

See all articles by Yongxin Yang

Yongxin Yang

Queen Mary University of London

Timothy M. Hospedales

University of Edinburgh

Date Written: August 10, 2022

Abstract

Recently self-supervised learning (SSL) has achieved impressive performance in computer vision (CV) and natural language processing (NLP) tasks, and some early attempts are made in the area of finance. In this paper, we apply SSL to extract features from financial time series data, and use those features to measure the similarities between assets in the market. As similarity measurement is the key to portfolio diversification, we consider two portfolio optimisation problems: index tracking (IT) and minimum variance portfolio (MVP), with the additional diversification terms linked to different similarity measurements, which are sourced from different SSL algorithms. Both IT and MVP are both convex optimisation problems with deterministic solutions, therefore the performance difference is traced back to SSL algorithms, rather than other factors. Extensive experiments are conducted with eight SSL algorithms, and the analysis of the results of the experiments demonstrates the advantages of SSL over non-SSL alternatives.

Suggested Citation

Yang, Yongxin and Hospedales, Timothy M., An Evaluation of Self-Supervised Learning for Portfolio Diversification (August 10, 2022). International Conference on Artificial Neural Networks (ICANN) 2023, Available at SSRN: https://ssrn.com/abstract=4187326 or http://dx.doi.org/10.2139/ssrn.4187326

Yongxin Yang (Contact Author)

Queen Mary University of London ( email )

Mile End Road
London, E1 4NS
United Kingdom

Timothy M. Hospedales

University of Edinburgh

Old College
South Bridge
Edinburgh, Scotland EH8 9JY
United Kingdom

Do you have a job opening that you would like to promote on SSRN?

Paper statistics

Downloads
97
Abstract Views
310
Rank
509,053
PlumX Metrics