Stock Price Prediction in the Banking Sector of Bangladesh: an Ensemble LSTM Framework

Predicting the values of stock prices remains one of the most challenging problems due to the complex, nonlinear, and volatile nature of financial markets. Many influencing factors dictate the nature of stock prices. Not only macro-economic factors like inflation rate and exchange rate, but also socio-political issues exert influence on the overall stock market behaviour of the banks in Bangladesh. This study proposes a data-driven framework for predicting the closing stock prices of banks in the Dhaka Stock Exchange (DSE). An exhaustive dataset of stock market data and relevant factors was also prepared for this study. The framework proposed consists of a two-stage ensemble learning technique. Separate Long Short-Term Memory (LSTM) sub-models are trained on historical data of individual banks, and the outputs are then integrated to develop a meta-level dataset. A meta-model was trained on the meta-level dataset. Model evaluation is performed through a pipeline of trained submodels and a metamodel on a merged test dataset. The training and testing processes have been performed on different dataset combinations to capture the relevance of the factors to the predictions.