Application of machine learning algorithms to predict the closing price of the Johannesburg Stock Exchange all-share index

Abstract

Stock markets are regarded as one of the most important indicators of the economy’s strength and development. Predicting stock prices is of critical importance for investors who wish to minimise the risks of investments. Stock price prediction is a difficult task since stock prices are influenced by factors such as the financial status of the company, socioeconomic conditions of the country, political atmospheres, and natural hazards. The Efficient Market Hypothesis (EMH) states that stock markets behave like a random walk and due to this reason, it is complex to forecast the stock market. Researchers use time series forecasting, technical, and fundamental analyses to predict the stock values while proving or disproving the EMH. In the past, researchers used traditional methods such as Autoregressive Integrated Moving Average (ARIMA) to predict stock prices. Currently, deep learning architectures are widely used to solve time-dependent problems and can provide a huge push to the problem of stock price prediction. The main objective of this study is to develop a framework that forecasts the daily closing price of All- Share index data based on deep learning techniques. To achieve this objective, Long Short Term Memory (LSTM) and Gated Recurrent Unit (GRU) are employed. A Vector Autoregressive (VAR) model is used to benchmark the deep learning techniques. The analysis is based on the Financial Times Stock Exchange (FTSE)/ Johannesburg Stock Exchange (JSE) All-Share (J203) data collected from Iress Expert. The results show that all the methods are able to predict the closing price of the index. GRU predicted the future closing price with an average Mean Absolute Percentage Error (MAPE) of 9.349% maximum while LSTM was able to predict with the maximum average error of 9.459%. A VAR model performed with the maximum average error of 2.152%.