Predicting Exchange Rate Changes in Sri Lanka Using the LSTM-based Deep Neural Network Model

The exchange rate is a significant macroeconomic variable reflecting a country's economic position and directly impacts the economic condition through its fluctuations. Therefore, improved prediction of exchange rates would be immensely valuable for key economic stakeholders to make the right decisions on economic activities. Therefore, accurate exchange rate prediction is crucial for policymakers, especially for decision making based on economic forecasts.

Various macroeconomic and global market factors contribute to fluctuations in the exchange rate, while political, economic, and social sentiments can also influence its movements. Therefore, accurate prediction of exchange rates is widely acknowledged as a challenging task. This study identifies the most appropriate macroeconomic variables and global market factors for exchange rate prediction in Sri Lanka, including oil price, Dow Jones Industrial Average (DJIA) stock market index, interest rates, and secondary market treasury bill rates. In addition, this study considers public sentiment through news headlines to understand the impact of socioeconomic and political factors on exchange rate fluctuations. Input sequences for the deep learning model were generated from text features extracted from Tweets related to the foreign exchange market and the most appropriate macroeconomic and global market factors for exchange rate prediction in Sri Lanka.

This paper suggests a Long Short Term Memory (LSTM)-based Deep Neutral Network (DNN) model to predict the exchange rate, incorporating macroeconomic and global market factors with financial sentiments. The Mean Absolute Percentage Error (MAPE) value achieved for the model is 0.6693%.