Enhancing Dhaka Stock Exchange Index Prediction: Neural Networks Based Approach Incorporating Macroeconomic and Technical Indicators

Stock market prediction has been among the key interests of investors, economists and computer scientists. But it is hard to predict the market, as many variables can influence the market and disordered data can play a significant role to lower the prediction ability. There’s been a shortfall of studies which analyzes the broad index of Dhaka Stock Exchange. In this study a new approach to predict the broad index of the Dhaka Stock Exchange combining both macroeconomic and technical indicators is proposed. Over 10 years of historical data have been analyzed for this research which is rare for research on Dhaka Stock Exchange. Recent research shows us that Recurrent Neural Networks increase the prediction accuracy among the developing stock markets. This study incorporates Simple Recurrent Neural Networks, Long Short-term Memory networks, Gated Recurrent Units and the Transformer model. However, GRU and LSTM performed best among the models. The result shows that the R-squared score is above 98 percent, which indicates that there is a significant correlation between the test and the prediction data. Also, the mean absolute percentage error is below 0.6 percent, which is considered excellent in the case of financial markets. But for a very short period of trading these results may not be enough. More indicators can be considered in future research for better results.