A comparative analysis of machine learning models for forecasting JSE Stock Returns

This study examines the application of machine learning models to predict the cross-section of Johannesburg Stock Exchange (JSE)- listed share returns. Four models are developed and compared using monthly data from 2005 to 2021: neural networks, random forest, long short- term memory (LSTM) networks, and conventional linear regression. The explanatory variables comprise nine firm-specific financial metrics, motivated by prior research. The sample is divided into a training period (2005–2016) and a testing period (2016–2021), further split into 1-year, 3-year, and 5-year testing intervals. The results show that the LSTM model performsbest across most evaluation metrics and investment scenarios, with the random forest model close behind, offering slightly better risk-adjusted returns.

Grantor dc:publisher.institution

Department of Finance and Tax

Year dc:date.issued

2025

Author dc:creator

• Muir, Cameron James

Advisor dc:contributor.advisor

• Van Rensburg, Paul

• JSE
• Stock returns

Language dc:language.iso

en

Handle dc:identifier.uri

http://hdl.handle.net/11427/42503

OAI identifier oai:identifier

oai:open.uct.ac.za:11427/42503