A Hybrid Approach versus Arima and Ann Model for TFR Time Series Forecasting

Abstract:

In this paper I have studied the time series of total fertility rate (TFR).  Based on the data for the years 1990-2013, with monthly frequency, obtained by INSTAT, we have done its modeling and forecasting using three types of methods: the autoregressive integrated moving average ARIMA, nonlinear autoregresive neural network (NAR) and the proposed hybrid method of ARIMA-ANN.  As TFR is influenced by many political, economic and psychological factors, it is difficult to identify a unique economic model which can yield stable forecasts. Using the technical analysis of the historical data is known the best model to forecast future rate.

The empirical analysis has shown very good results, mainly in the proposed hybrid model. The performance of the three methods was compared based on standard statistical measures. The ARIMA-ANN model generated the best model, with the lowest RMSE, MAE, MPE and MAPE measures.