In the weak form, the Efficient Market Hypothesis (EMH) states that it is not possible to forecast the future price of an asset based on the information contained in the historical prices of that same asset. Under this assumption, the market behaves as a random walk and as a result, price forecasting is impossible. Furthermore, financial forecasting is a difficult task due to the intrinsic complexities of any financial system. The purpose of this study is to examine the potential of developing an international investment strategy using future index price predictions and offsetting predicted price declines by investing in negatively correlated international markets. Therefore, the first objective of this study was to examine the feasibility and accuracy of using a machine learning technique to model and predict the future price of stock market indices of South Africa (All Share Index) and a variety of other developed and developing international markets, which included South Africa, Brazil, Russia, India and China of the BRIC countries and Italy, France, Netherlands, Switzerland, Germany, Nigeria, Australia, Hong Kong, Saudi Arabia, Japan, the U.S., Turkey and the U.K., which were identified as South Africa’s major trading partners. Secondly, an analysis of market correlation between each country’s equity index and South Africa’s ALSI was conducted to determine which of these international indices were positively and negatively correlated to the South African ALSI. This allowed an extrapolation of potential international diversification opportunities. By using machine learning to predict future price trends of the South African All Share Index (ALSI) within a specified time period, the market correlation aspect of this study was able to suggest possible negatively correlated safe haven markets to invest in to offset predicted losses in an expected declining local market. The study’s major limitations include a single method for regression analysis (GARCH(1, 1)) and a limited number of variables in the feature space when predicting future prices. Additional parameters could prove a more robust modelling technique. The data used was a series of past closing prices of each country’s major index. The data was split into five periods, where each period was assigned an overarching theme based on the prevailing market conditions at the time. The ALSI data set was subjected to a unit root test and found to be non-stationary. The analysis thereafter followed a two-step test, with the first being the determination of market correlation of the South African equity market with other markets, using a generalised autoregressive conditional heteroskedasticity (GARCH (1: 1)) approach given the non-stationary nature of the ALSI historic data. The results showed strong positive market correlations between South Africa and China, India, Nigeria, Russia and Saudi Arabia, and strong negative correlation between South Africa and Australia, Germany, the Netherlands, and the United Kingdom. Secondly, the specific area of machine learning employed in this study was support vector machines, as implemented using Python programming. The results compare the actual index price with those predicted by the model and showed that this technique has the ability to predict the future price of the Index within an acceptable accuracy. The accuracy measure used was the mean relative error which in most cases was calculated to be between 95 and 98 which is considered relatively high. However, the results of the investment approach described above was considered to be too inconsistent to consider this diversification strategy viable. From a South African perspective, this approach has not been documented previously.
| Identifer | oai:union.ndltd.org:netd.ac.za/oai:union.ndltd.org:uct/oai:localhost:11427/31830 |
| Date | 07 May 2020 |
| Creators | Phillip, Jarryd |
| Contributors | Toerien, Francois |
| Publisher | Faculty of Commerce, Department of Finance and Tax |
| Source Sets | South African National ETD Portal |
| Language | English |
| Detected Language | English |
| Type | Master Thesis, Masters, MCom |
| Format | application/pdf |
Page generated in 0.0019 seconds