Browsing by Author "Makgoale, Katlego"

Filter results by typing the first few letters
Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Mean-Variance Optimisation of A South African Index Based Portfolio Using Machine Learning
    (University of the Witwatersrand, Johannesburg, 2021) Makgoale, Katlego; Jakubose, Sibanda
    This study embarked on a comparison of the effectiveness of the Markowitz Mean- Variance Portfolio Optimisation against utilising a Machine Learning Technique to construct an optimal portfolio. The study aimed to: Construct an optimal portfolio using the Mean-Variance Analysis Framework, Construct an optimal portfolio using a Machine Learning Technique (Support Vector Regression), Contrast the results of the Minimum-Variance Portfolio and the Machine Learning Portfolio. The stocks of the FTSE JSE FIN15 index were chosen to construct the portfolio. The historical returns of the stocks in the index were used to trained (December 2014 to June 2019) and test the models(June 2019 to December 2020). The Mean-Variance Analysis and Minimum-Variance Portfolio were constructed using Python code that the author compiled. Similarly, the Support Vector Regression model was built in Python. The weights for the Machine Learning portfolio were calculated using the pseudo-inverse matrix and the predicted value of the Regression Model. It was found that the Minimum-Variance and Machine Learning portfolio produced different portfolios, but both containing fewer holdings than the original index. The performance of the Minimum-Variance Portfolio exceeded that of the index and the Machine Learning Portfolio with regards to relative(excess) returns and total returns in the out of sample period. It was found that the Machine Learning portfolio performs well at replicating the index returns but fails to exceed them and typically has a higher risk associated with it. It was concluded that the Minimum-Variance portfolio would be the most attractive to a risk-averse investor and the Machine Learning portfolio underperforms the Minimum variance and the index. Therefore confirming the effectiveness of Mean-variance Optimisation in a South African context against a Machine Learning Technique