Testing the predictability of the Botswana stock exchange: evidence from supervised machine learning

Abstract

Prediction of the stock market is a vital part of the economy especially for emerging markets in developing countries. There is significant literature on predicting the stock market particularly in developed countries like the US. However, there is need for more research in emerging markets such as developing countries like the Botswana Stock Exchange. This paper aims at evaluating the predictability of the Botswana Stock Exchange using supervised machine learning to specifically assess and test the null hypothesis of the Random Walk Theory. Machine learning is one of the upcoming trends of data mining; hence few machine learning algorithms have been used where their results have been compared using classification evaluation parameters such as Accuracy, Mean Average Error (MAE), Receiver Operating Characteristic Area (ROC), Kappa Statistic, Precision and Recall. Naïve Bayes have been considered the most effective model as it yielded the highest accuracy of 83.3% with the least error margin. The results reject the null hypothesis of the Random walk Theory for Botswana Stock Exchange for the period of January-December 2015, clearly indicating that the Botswana Stock market is predictable using machine learning techniques.