Abstract
Making stock market returns predictions are regarded as an efficient tool for financial risk management and portfolio diversification. Various methods of forecasting for investment decision-making that provides accurate results have been examined. These methods have also been employed to examine various returns of stock market indices. This study provides predictions for BRICS countries' (Brazil, Russia, China, India, and South Africa) daily returns of stock market indices. Making use of machine learning models. This was done using various machine-learning models. The choice of the stock markets is because of the structural breaks that the developed and emerging markets are suffering from in numerous ways. Emerging markets tend to be more volatile and therefore are susceptible to suffering from structural breaks in comparison to their developed counterparts.
The goal of this investigation is to provide an in-depth analysis of various machine learning methods, in making predictions in a less structural break and more structural break condition. The main objective, therefore, is for us to figure out which of the suggested machine learning techniques is most accurate and reliable at predicting stock market returns. Selecting the best model with so many models involved is a challenge, and this was solved using the Model Confidence Set (MCS). MCS tries to solve this problem by selecting a group of models that are considered equally good. Moreover, we create an ensemble using the bagging technique to check if it will outperform individual models. To train and test the various machine learning techniques, the weekly stock market returns of selected indices were used. In the end, the results showed that no one specific technique of the five used could be applied uniformly and produce accurate results on all markets. However, what we noted is that the ensemble model outperformed all single models.