Forecasting Fertility in Northern Ireland using a Time Varying Coefficients Model.

The project builds upon the earlier study NILS 022, Predicting Short Run Changes in Fertility in Northern Ireland. The project assembled a panel of women from 1997 – 2007 aged between 16 and 44 years in each year, together with their characteristics drawn from the 2001 census and the births they had each year from the GRO. The analysis was based on the logit model with time dummies employed as the basis of the forecast model. This was unsatisfactory due to the importance of the trend element in the forecast – essentially the projection of the trend into the future was arbitrary.

This project would also construct a fertility panel using a similar approach to the previous project with the exception that the data would cover more years as this becomes available. It is proposed that a more technically advanced model would be used where some of the coefficients would be allowed to vary according to an autoregressive scheme. The particular coefficients allowed to vary would be determined by visual inspection of the results of estimating the model annually backed up by hypothesis tests. Once the time varying coefficients were determined the statistical model can be expressed in state space form and estimated. The autoregressive progressive process would be incorporated into the forecast model, allowing the standard errors of the forecast to include the effect of the evolution of the coefficients over time.