Forecasting Moldova’s monthly exports using autoregressive models with seasonal dummies

Abstract

Exports significantly influence Moldova’s economic stability and regional integration, particularly due to its agriculture-driven economy characterized by pronounced seasonality. This study evaluates the forecasting accuracy of autoregressive integrated moving average (ARIMA) models, enhanced with seasonal dummy variables, for predicting Moldova’s monthly exports. It systematically compares traditional ARIMA methods, ARIMA with seasonal dummies, and modern techniques such as Gated Recurrent Unit (GRU) neural networks. Forecast performance is assessed using Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Theil’s U-statistic, and the Cost-Value Ratio (CVR). Results demonstrate that combining forecasts from multiple ARIMA models significantly improves predictive accuracy, with weighted combinations based on MAPE reducing forecast errors by approximately 10%. The findings highlight the effectiveness and adaptability of combined forecasting models, emphasizing their value for small, open economies like Moldova’s that are susceptible to external shocks and seasonal fluctuations. This research provides robust empirical evidence supporting the adoption of combined econometric forecasting methodologies in economic policymaking and strategic planning.