چکیده:
In general gross domestic product (GDP) is a substantial element in macro-economic analysis. Policy makers of a country use variations of GDP for long run planning. Considering different economic conditions of a country, forecasting is a useful tool to identify the variations of GDP for planning. In this paper, quarterly GDP value during (1998-2003) is used as a base of analysis. The quarterly GDP values of the year (2004 -2005) are forecasted using Time series, Exponential smoothing and Neural network approaches. The results are compared with actual quarterly GDP value and error measurement are computed in each methods. Consequently statistical analyses are accomplished to show the best method of forecasting. We have shown that neural network approach method is the best alternative to forecast the GDP of Iran.
خلاصه ماشینی:
"In contrast, however, Batchelor (1982) [7] showed that although survey-based growth expectations in Belgium, France, Germany, and Italy produce lower root mean square errors (RMSE) than simple extrapolative predictors, they include no additional information in more complex autoregressive integrated moving average (ARIMA) forecasting models.
Although, it may seem intuitively plausible that business tendency survey data provide additional information in standard time series models of output growth, empirical studies provide somewhat different results.
Table 4: The computational results of three methods in forecasting period Quarter ACTUAL TIME SERIES EXPONENTIAL SMOOTHING GRNN 1 90489 75621 86137 82662,5949 2 116489 77328 86137 103659,4813 3 98644 79035,1 86137 88384,4791 4 91090 80742,1 86137 82190,2614 5 99703 82449,1 86137 85696,5792 6 121080 84156,2 86137 109746,1945 7 104358 85863,2 86137 92350,8042 8 93780 87570,3 86137 85304,2481 Figure 4.
The hypothesis-based statistical comparison 6- Conclusion In this paper time series, exponential smoothing, and neural network methods were considered to forecast quarterly gross domestic product of Iran.
Although, the statistical results show that no methods have different means significantly from actual GDP in the forecasting period, but hypothesis tests indicate that neural network approach (in comparison with the discussed models) is effective from MAPE view point to be used for forecasts."