Comparison of Time Series and Neural Network Models to Predict Coconut Fibre Exports in Sri Lanka

Abstract

Coconut is one of the major exports in Sri Lanka since the colonial era. Recently, Sri Lanka has become the number one exporter of coconut fibre products in the world. Therefore, predicting export volume of coconut fibre exhibits crucial importance for future investments. The objective of this research study is to develop a model to forecast monthly exports of coconut fibre in Sri Lanka. In this study two models: a time series model; Seasonal Auto-Regressive Integrated Moving Average (SARIMA) and Artificial Neural Network (ANN) model; Feedforward neural network with the Backpropagation algorithm model were considered. Past lagged observations of the data series and moving average of technical indicators were employed as the explanatory variables to build the ANN model. The predictive performance of ARIMA model and ANN model were evaluated using statistical indices, namely, Normalized Mean Square Error (NMSE), Mean Absolute Percentage Error (MAPE) and Directional Symmetry. Based on the results obtained, ANN model can be recommended as a better model to forecast exports of coconut fibre in Sri Lanka.