Forecasting market movement in Dhaka Stock Exchange: LSTM vs. ARIMA

Abstract

This paper creates and examines the performance of LSTM models against ARIMA models, using Dhaka Stock Exchange data from beginning of 2000 to the beginning of 2019. An algorithm has been designed to simultaneously train and test the models using datasets of 340 companies of the market. The empirical result shows the absolute dominance of ARIMA over LSTM, which contradicts some previous works. At the end, we try to discuss the possible reasons behind the unsatisfactory performance of LSTM and explore some of the possible future expansions and extensions of the current work.