engleski

Seasonal Adjustment and Forecasting of Croatian GDP under Differebt Scenarios

Forecasting Croatian GDP has proved to be difficult task, not only in the past which includes war for independence, but also nowadays. A lot more outliers and irregularities were detected in the period 1990-1995 than later, when economic situation has been stabilised. In this paper we examine seasonal adjustment using two different methods: X-12-ARIMA which is said to be ad hoc method (empirically based) and TRAMO/SEATS which is model-based method. Even though empirically based methods are still dominating statistical agencies throughout the world, model based method TRAMO/SEATS has been considered as a very serious contender. We have applied both methods on Croatian GDP series and have compared obtained results. The period covered was form 1st quarter 1997 until 4th quarter 2004, fixed prices. The overall seasonal adjustment quality index for X-12-ARIMA was 2.169 and 2.085 for TRAMO/SEATS thus giving the advantage for model-based method. X-12-ARIMA used correction for trading days effect, and detected no outlier, while TRAMO/SEATS detected one outlier in 4th quarter 1997. This method has proved to be better in detection of outliers. Statistics on residuals for both methods were satisfactory, but some statistics have not been performed with X-12-ARIMA because the series was not long enough. Forecast errors were within the tolerant limits. Forecasting the future Croatian GDP is not only statistical task, but also it has some political weight. Since the Croatian approach towards EU has been slowed, GDP could be slightly lower in the first quarter of 2005 than expected. Altogether, both methods produces acceptable results since the GDP series was not difficult to adjust and to forecast, but we think TRAMO/SEATS method produced better results, both in the quality of seasonal adjustment and the quality of forecasting. TRAMO/SEATS has produced slightly higher forecast for the next three years.

seasonal adjustment; forecasting; GDP

nije evidentirano