Journal of Official Statistics, Vol.13, No.3, 1997. pp. 279–303
Comparing and Assessing Time Series Methods for Forecasting Age-Specific Fertility and Mortality Rates
William R. Bell
Abstract:Fertility and mortality rates exhibit strong age patterns, and various researchers have developed methods to capture this structure and use it in forecasting. Two general approaches have been developed. The curve fitting approach involves fitting parametric curves to the age-specific rates. The principal components approach involves computing principal components to obtain a linear transformation of the data with simplified structure. This article reviews and compares proposed alternative versions to these two approaches, and then evaluates the out-of-sample performance of the various alternatives in forecasting age-specific U.S. white male and female mortality rates. None of the approaches tried produced short-term forecasts more accurate than those obtained from a simple random walk with drift model applied to the rates for each age separately. Also, the curve fitting approach and low-dimensional principal components approaches clearly require a bias adjustment to avoid having approximation errors compromise short-term forecast accuracy.
Keywords:Population projection; demographic forecasting; principal components.
Copyright © Statistics Sweden 1996-2018. Open AccessISSN 0282-423XCreated and Maintained by OKS Group