Modeling Annual Coffee Production in Sri Lanka

Abstract

Coffee production is a source of revenue for the economic sector in Sri Lanka. During 1988 to 2020, the mean annual coffee production is 7672 metric tons with a coefficient of variation is 30%. The advanced knowledge of annual coffee production has many advantages. However, past studies found that no model has been developed to model annual coffee production in Sri Lanka. In this study, an ARIMA (1,2,0) model was identified as the best fitted model to forecast the annual coffee production. The model was trained using data from 1988 to 2020 and validated using data in 2021. The best-fitted model was selected by comparing different statistical indicators such as Akaike Information Criteria, Schwarz Criteria, Log-likelihood Criteria, and volatility of the three parsimonious models. It was found that the errors of the best fitted model were white noise. The percentage errors for the forecast values for the training and validation data sets were within ± 10. The predicted annual production for 2022, 2023, 2024, and 2025 are 6987 MT, 6221 MT, 7209 MT, and 6664 MT, respectively. This is the first empirical study to develop a statistical model to predict annual coffee production in Sri Lanka. The model can be improved by using external variables as explanatory variables and considering dummy variables to capture the structural breaks.