PREDICTIVE MACHINE LEARNING METHODS FOR STOCK RETURNS AMONG EMERGING ECONOMIES IN AFRICA

Abstract

The study examined the performance of competing models across the stock returns of three African countries, namely Nigeria, Tanzania, and Uganda, to determine whether a single predictive model dominates across emerging markets. The study utilised secondary data from the stock markets of Nigeria, Tanzania, and Uganda, sourced from the website (www.investing.com), spanning 11 years and comprising a total of 2772 observations for Nigeria, 2862 observations for Tanzania, and 2798 observations for Uganda. The descriptive statistics of the stock returns across Nigeria, Tanzania and Uganda revealed that the mean returns are lowest for Uganda and highest for Nigeria. This showed that Nigerian stock gained more during the period under review, followed by Tanzania stock returns. The standard deviation for Nigerian stock returns was highest, showing evidence of high volatility in Nigerian stock returns. The stock returns across the countries exhibited positive skewness. Among the stock returns, Tanzania and Uganda exhibited very high kurtosis. Lastly, across the countries, all the stock returns exhibited non-normality (p-values<0.05), which is mostly the case for stock returns. The findings from the training models revealed that the following: Artificial Neural Network dominated for Nigeria stock returns, no single model dominated for Tanzania stock returns, Autoregressive Integrated Moving Average with exogenous variables (ARIMAX) model dominated for Uganda stock returns. While the summary from the testing models revealed that the following: Random forest dominated for Nigeria stock returns, Support vector machine for Tanzania stock returns, and  Artificial Neural Network dominated for Uganda stock returns. This study concluded that the performance of the predicting models for training and testing performed differently, as no single predicting model dominated across the African countries considered. This study recommended that data scientists, machine learning experts and policy makers in the stock market should consider competing models in different scenarios across emerging markets in Africa to enhance reliable prediction and decision making.