Modelling and Forecasting GDP of Greece with a Modified Exponential Smoothing State Space Framework

The comparison between models and their predictions has been a big topic in the literature in recent years. There are two main methods used to compare the prediction quality; the average square error and the distance in time. In order to evaluate the forecast performance, as well as order the forecasts, researchers have developed several measures of accuracy. In the current paper to predict the GDP of Greece, we use non-linear dynamic models ETS, using state-space-based likelihood calculations in order to choose models and calculate the forecast standard errors. The estimation of the models is made with the function of maximum likelihood, while the choice between models with additive and multiplier errors is made with the use of the Akaike (AIC) criterion based on likelihood and not on one-step ahead forecasting. The process is completed by clearly defined methods for the assessment, the likelihood evaluation with the BFGS algorithm and the analytical derivation of the forecasted points and intervals under a Gaussian error assumption. The results of the work showed that the model with a multiplier error, with a multiplier tendency to depreciation and additional seasonality, is the most appropriate for the period under examination. Moreover, the results of the forecast showed a fall in GDP for the coming quarters with this decline to depreciate.