Absztrakt
This paper presents different methods for the forecasting of unemployment rates in two Romanian counties. The stationarity of the monthly unemployment rate time series between January 2000 and November 2016 was examined using the ADF and KPSS tests. Based on time series, a forecast was estimated using two approaches: the Box-Jenkins methodology and the Artificial Neural Network-based NAR model. Results showed a decreasing trend by the end of the forecasted period in all cases, except for the NAR model of Harghita County. Comparing the forecasted values with the officially registered unemployment rates from the same period, we observed that, by the end of the period, the differences between the real and predicted values became higher in the NAR model than in the ARMA model-based forecasting. These results indicate that, in these particular cases, NAR neuron network model-based forecasts fit well if values are estimated for a short-term period, while for medium-term forecasts the ARMA model-based forecasting is more precise.
Kulcsszavak: regional unemployment, time-series models, forecasting and prediction methods, Box-Jenkins methodology, Artificial Neural Network
JEL: C32, C53, R15