Multi-Step-Ahead Prediction of Exchange Rates Using Artificial Neural Networks: A Study on Selected Sri Lankan Foreign Exchange Rates

Abstract

The exchange rate is a very important economic indicator for countries with market economies as its fluctuations heavily affect the most areas in the economy. Accordingly, predicting future values of foreign exchange rates is very important in policymaking. This study was conducted to perform multi-step ahead predictions on foreign exchange rates of Sri Lankan Rupee against three international currencies using Artificial Neural Network models, to measure the accuracies of these models and identify shortcomings if present. Multi-Layer Perceptron, Simple Recurrent Neural Network, Long Short-Term Memory, Gated Recurrent Unit and Convolutional Neural Network architectures were employed for this study. Most of the models except few Gated Recurrent Unit models were able to predict 10-days-ahead exchange rates with a higher level of accuracies (97%-99%). According to the findings Stateful Simple Recurrent Neural Networks with one input layer, a hidden layer, a flatten layer and an output layer performed as the best architecture to predict the three exchange rates selected.