DEVELOPING A PREDICTIVE MODEL TO PREDICT THE EXCHANGE RATE OF BANGLADESHI TAKA (BDT) AGAINST US DOLLAR (USD)

Abstract

This study is concerned with applying machine learning to develop a predictive model to forecast the exchange rate of Bangladeshi Taka (BDT) against US dollar (USD). There are studies that showed that machine learning and deep learning methods fit well in predicting exchange rate and outperform traditional models. However, in Bangladesh, machine-learning based predictive modeling for the said purpose has remained an uncharted area so far and this paper attempts to fulfill this research gap. For this, the economic data of Bangladesh for 11 fiscal years (from 2009-10 to 2019-20) were collected from the central bank database and analyzed using the analytics tool-KNIME. The findings suggest that (1) decision tree makes a comparatively better predictive model with very high level of accuracy, (2) feature selection technique does not contribute to the improvement of accuracy, and (3) to get the optimal model, multiple experiments with different combinations of variables should be conducted. The findings of this paper can be of immense help to several stakeholders including the country’s central bank, policymakers, academicians, and researchers as it can be the base point to explore novel avenues of research and to advance the application of predictive modeling in new areas.