Predicting Systematic Risk Using Artificial Neural Networks
| dc.article.end-page | 57 | |
| dc.article.start-page | 1 | |
| dc.contributor.advisor | Alovokpinhou, Sedjro Aaron | |
| dc.contributor.author | Setloboko, Thabiso | |
| dc.date.accessioned | 2024-05-23T08:47:45Z | |
| dc.date.available | 2024-05-23T08:47:45Z | |
| dc.date.issued | 2023 | |
| dc.description | A research proposal submitted to the faculty of Commerce, Law, and Management, University of the Witwatersrand, in partial fulfilment of the requirements for the degree of Master of Management in Finance and Investments. | |
| dc.description.abstract | Financial institutions and investors are always investigating mathematical models that can enable them to make accurate predictions of time varying variables. For the longest time, statistical and autoregressive models have been at the forefront of forecasting. However, these are only accurate in short horizons; that is, these models are more accurate in daily, weekly, and monthly forecasts. This paper seeks to investigate long-horizon (yearly) forecasts using machine learning models called Artificial Neural Networks. The network uses neurons similar to biological neurons in living things, allowing them to study complex data patterns and retain pattern behaviors that allow them to make accurate predictions. The paper is based on the novel discovery that in forecasting long-horizon time series data, neural networks outperform statistical models significantly. The paper uses market data from the Johannesburg Stock Exchange and the New York Stock Exchange to represent the emerging and advanced markets, respectively. The forecasted data involves pre and post COVID-19. The shock introduced by the coronavirus is investigated to check if the forecasting ability of the model is affected. The empirical results demonstrate that the models accurately forecast systematic risk in the South African market more than in the American market. The accuracy of the model is measured by using root mean square error and mean absolute error. The model produced low error values for both markets, indicating their effectiveness in forecasting. It was expected that the error measures would consistently get lower as the time horizon increased; however, there were inconsistencies. For a portfolio manager, the results obtained in this research are interesting because the model handles large quantities of data and forecasts long-horizon systematic risk with little error. However, further investigation on this model still needs to be done | |
| dc.description.librarian | MM2024 | |
| dc.faculty | Faculty of Commerce, Law and Management | |
| dc.identifier.uri | https://hdl.handle.net/10539/38539 | |
| dc.language.iso | en | |
| dc.publisher | University of the Witswatersrand, Johannesburg | |
| dc.rights | © University of the Witswatersrand, Johannesburg | |
| dc.school | Wits Business School | |
| dc.subject | Systematic Risk | |
| dc.subject | Artificial Neural Networks | |
| dc.subject | Investors | |
| dc.subject | Financial institutions | |
| dc.subject | UCTD | |
| dc.subject.other | SDG-8: Decent work and economic growth | |
| dc.title | Predicting Systematic Risk Using Artificial Neural Networks | |
| dc.type | Dissertation |