Model Building and Forecasting of Nepal's Total Import Amount Using SARIMA Model

Abstract

This study aims to analyze and forecast Nepal's monthly total import trends using the Seasonal Autoregressive Integrated Moving Average (SARIMA) model. The research employs the Box- Jenkins methodology to develop and evaluate SARIMA models using historical import data from August 2006 to June 2025. A time-series cross-validation (TSCV) approach was used to split the data into training (90%) and testing (10%) sets for model building and evaluation. 240 probable SARIMA models were tested to identify the best-performing model, with the analysis conducted in R software utilizing packages such as forecast, ggplot2, and tseries. The study identified the SARIMA (2,1,1) (1,1,1) [12] model as the most suitable for capturing the seasonal and non-seasonal trends in Nepal's monthly total imports. The model achieved a Mean Absolute Percentage Error (MAPE) of 7.30% on test data, demonstrating high forecasting accuracy. Using the selected model, the study successfully forecasted Nepal's import trends for the next 12 months. This research contributes to the field by providing a rigorous time-series analysis of Nepal's import data using the SARIMA model and time series cross validation. The study's application of the SARIMA methodology to Nepal's trade data fills a gap in the literature and offers a robust framework for forecasting trade metrics in similar contexts.