engleski

Comparative analysis of merchandise exports time series forecasting models in Croatia

This paper analyses appropriateness of models for forecasting merchandise exports in the gross domestic product in Croatia. The estimation of four trend models and the ARIMA model is conducted. The comparison of three selected trend models and the ARIMA (1, 1, 1) model is conducted. Firstly, information criteria are examined in order to select an adequate model. Moreover, following measures of predictive efficiency are used: Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Theil Inequality Coefficient, bias, variance and covariance proportion. Calculated measures have shown that it is not clear which model should be used, since mentioned indicators show different results regarding the most appropriate forecasting model. However, the analysis has shown that the trend models are superior to the ARIMA (1, 1, 1) model for forecasting merchandise exports in Croatia. Comparison of the actual and the forecasted values shows that the forecast conducted using the linear and the log-linear trend model is close to the actual value only for τ=1. After that, the actual values of merchandise exports decline sharply, while the trend models predict an increase of merchandise exports in the forecast period. The explanation of this result lies in the fact that due to the economic crisis, the assumption of stable economic conditions is violated.

forecasting; trend models; ARIMA models; measures of predictive efficiency

nije evidentirano