A Deep Learning Model for Emerging Stock Market Prediction

Abstract

Stock market prediction is a challenging task where the data is often noisy and volatile. The challenge is exacerbated in emerging stock markets where data volatility increases due to several endogenous and exogenous variables. Despite the advances in models for emerging stock market prediction, including Autoregressive Integrated Moving Average (ARIMA), Support Vector Machines (SVMs), and deep learning models, the field still requires more research. The complexity and non-linearity of emerging stock market data require models that can capture intricate relationships between variables, and the consideration of exogenous variables. This study aimed to present a deep learning model incorporating LSTM algorithm and XG-boost feature selection technique to predict the next-day closing price. The proposed model utilizes exogenous variables to improve the accuracy of predictions. Four datasets are constructed for emerging markets including Qatar, Saudi Arabia, Brazil, and China. The performance of the model is evaluated using: Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Normalized Root Mean Squared Error (NRMSE), and Adjusted R Squared. Experimental results showed that exogenous variables improve the model performance compared with a model using only endogenous variables. Exogenous variables reduced MAPE in varying proportions from 0.03 to 0.13. In addition, the results disclosed that LSTM algorithm is appropriate for predicting the next-day closing price in emerging stock markets.