1 Parametric mortality smoothing: deciding on the optimal method Klára Hulíková 1 , Boris Burcin 2 , Tereza Pachlová 3 , Dan Kašpar 4 Background and the main research question Traditionally, mortality smoothing is a common step in life table construction. Without application of any smoothing method the age-specific mortality rates express significant level of variability, above all for less populous countries or higher ages. Nowadays, a demographer or anyone constructing the life table could use some of specialized software applying some of different methods of mortality smoothing. However, there rises usually a crucial question – which of many possible functions is the optimal one for a particular population. Moreover, mortality smoothing is extremely important in spheres like life insurance or health and social systems forecasts, where as detailed and accurate life tables as possible are usually needed. Other field of usage of mortality smoothing methods could be also the mortality estimation for the highest ages (e.g. Gavrilov, Gavrilova, 1991; Burcin et al., 2010; Hulíková Tesárková, 2013). Among the simplest ways of mortality smoothing, there are the parametric functions, often called also mortality laws. These functions could be taken as an example of the oldest and most traditional demographic tools. In the paper, we focused solely on these parametric mortality functions with the aim to explore, whether there exist some rather universally suiting method of parametric mortality smoothing or not. Also in this paper we concentrated solely on period data. Goals of the paper Based on many studies describing the parametric smoothing in detail or dealing with definitions of different functions (among others e.g. Gompertz, 1825; Makeham, 1860; Thatcher, 1999; Wilmoth, 1995; Koschin et al., 1998; Burcin et al., 2010) we defined two goals of the paper as follows: 1) Using the AIC (Akaike Information Criterion), suitability of different selected parametric functions was evaluated when those functions were applied to many different populations (see data specification below). Based on the results of this empirical part of the study it is possible to calculate the ratios of cases when each function produced the most suitable results. 2) In the analytical part of the study, based on the values of the AIC, the most suitable function was selected for each analyzed population. Through the multinomial logistic regression analysis the probability of each model to be the optimal one was modelled. The explanatory variables were selected as calendar year, sex, geographical region and differently defined measures of overall mortality level (e.g. temporary life expectancy, etc.). These results lead to the main aim of the research, to propose and create any simple possibility of decision about the optimal model suitable for a particular population defined by various variables used in the analysis in the role of the explanatory variables. 1 Charles University in Prague, Faculty of Science, Department of Demography and Geodemography (klara.hulikova@gmail.com) 2 Charles University in Prague, Faculty of Science, Department of Demography and Geodemography (boris.burcin@gmail.com) 3 Charles University in Prague, Faculty of Science, Department of Demography and Geodemography (pachlovat@gmail.com) 4 Charles University in Prague, Faculty of Science, Department of Demography and Geodemography (kaspard@natur.cuni.cz)