Risk Analysis, Vol. 25, No. 5, 2005 DOI: 10.1111/j.1539-6924.2005.00671.x Forecasting Spanish Natural Life Expectancy Montserrat Guillen 1* and Antoni Vidiella-i-Anguera 1 Knowledge of trends in life expectancy is of major importance for policy planning. It is also a key indicator for assessing future development of life insurance products, substantiality of existing retirement schemes, and long-term care for the elderly. This article examines the feasi- bility of decomposing age-gender-specific accidental and natural mortality rates. We study this decomposition by using the Lee and Carter model. In particular, we fit the Poisson log-bilinear version of this model proposed by Wilmoth and Brouhns et al. to historical (1975–1998) Span- ish mortality rates. In addition, by using the model introduced by Wilmoth and Valkonen we analyze mortality-gender differentials for accidental and natural rates. We present aggregated life expectancy forecasts compared with those constructed using nondecomposed mortality rates. KEY WORDS: Age-sex-specific mortality; forecasting; Lee-Carter model; life expectancy; Poisson log-bilinear regression 1. INTRODUCTION One relevant issue in the study of mortality trends is the influence of changes related to deaths from ex- ternal causes such as accidents or fatalities. Especially in the male population, traffic- and work-related ac- cidents have had an enormous impact on the mortal- ity rates among young adults (see Felipe et al. (1) for the Spanish case). We argue that changes in mortality caused by accidents can have an influence on pro- jected life expectancy, and to our knowledge this sub- ject has not been much discussed in the literature. (2) The economic and social costs of accidents are well known. General concern about the costs of such losses seems to be growing, but little is known of their influence on the calculation of demographic indica- tors such as life expectancy. Governments introduce regulations aimed at reducing the risks of accidents. 1 Departament d’Econometria, Estadistica i Economia Espanyola, Universitat de Barcelona, Diagonal, 690 08034 Barcelona, Spain. ∗ Address correspondence to Montserrat Guillen, Departament d’Econometria, Estadistica i Economia Espanyola, Universitat de Barcelona, Diagonal, 690 08034 Barcelona, Spain; tel: +34 934037039; fax: +34 934021821; mguillen@ub.edu. Good examples are the strong penalties imposed for driving at a high speed or under the influence of alco- hol. When looking at work-related accidents, there are many incentives for firms to implement risk-reducing practices and improve working conditions. The elimi- nation of casualties is a top priority for welfare states. For example, traffic safety programs are aimed to diminish the number of victims. (3) In this article, we are interested in obtaining fore- casts of age-specific mortality by gender with confi- dence intervals in Spain. We also examine the trends of mortality due to natural causes and their impact on the calculation of life expectancy. Our main objective is to forecast life expectancy when mortality due to ex- ternal or accidental causes is not taken into account, which we call natural life expectancy. This means that external causes have been removed from all possible causes of death. We also study the difference between life expectancy at birth and natural life expectancy at birth in order to identify possible trends in the past and to predict future behavior. Besides, we compare gen- ders and examine the difference between men’s and women’s life expectancy if external causes of death are not taken into account. We believe that our findings 1161 0272-4332/05/0100-1161$22.00/1 C  2005 Society for Risk Analysis