Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Artificial neural networks and the cross-section of equity returns: identifying nonlinear opportunities on the Johannesburg Stock Exchange
This study investigates the potential of Artificial Neural Networks (ANNs) to forecast stock returns on the Johannesburg Stock Exchange (JSE) using fundamental and technical factors. The optimal neural network architecture is explored, considering varying model depths and node counts. The activation...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English English |
| Published: |
Department of Finance and Tax
2026
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| Summary: | This study investigates the potential of Artificial Neural Networks (ANNs) to forecast stock returns on the Johannesburg Stock Exchange (JSE) using fundamental and technical factors. The optimal neural network architecture is explored, considering varying model depths and node counts. The activation function, training algorithm, learning rate, number of epochs, batch size, and loss function are kept constant across architectures. The findings suggest that portfolios constructed from ANN forecasts have the potential to outperform an equal-weighted benchmark. Model performance depends on network architecture, with a three hidden layer model with 64 nodes in the first hidden layer yielding the best results. Addition of further hidden layers or nodes is found to reduce model generalization, mainly due to overfitting, while less complex models are found to underfit. Models with a reduced variables set outperformed, confirming the importance of feature selection. While ANNs are found to underperform a linear model, the top performing ANN outperforms on risk-adjusted metrics over the test period, suggesting benefits to non-linear return forecasting on the JSE. However, with no clear relationship between in-sample and test period performance across architectures, this superior performance could be data specific, highlighting challenges in selecting an optimal model ex-ante. On the other hand, limitations in feature selection and training likely constrained model performance, with potential to improve generalization. This study provides a foundation for further research into return forecasting with ANNs on the JSE, contributing to the growing field of artificial intelligence and machine learning in finance. Future research could further optimize model hyperparameters, improve feature selection, and account for the time-varying nature of features. |
|---|